Runs - Clinical Decision Support 2015¶

artificial¶

Run ID: artificial
Participant: HITSJ
Track: Clinical Decision Support
Year: 2015
Submission: 7/26/2015
Type: manual
Task: a
MD5: 7f99dee42ad7c5da6b91ef2ca290f0d2
Run description: This is a manual result.According to the topics, we construct queries artificially. The main sources we used to extend the topics involves UMLS and wordnet. Search engine is Indri 5.8 and result not re-ranking.
Code: https://github.com/hitwilab/TREC2015-submission.git

artificialB¶

Run ID: artificialB
Participant: HITSJ
Track: Clinical Decision Support
Year: 2015
Submission: 7/29/2015
Type: manual
Task: b
MD5: 2de5340fe01bb9974004757f2d113d4c
Run description: This is a manual result of taskB. According to the topics, we construct queries artificially. The difference with taskA is we jiont the expansion of the diagnosis field to the query. The main sources we used to extend the topics involves UMLS and wordnet. Search engine is Lucene and result not re-ranking.
Code: https://github.com/hitwilab/TREC2015-submission.git

AUEBrun1B¶

Run ID: AUEBrun1B
Participant: DBNET_AUEB
Track: Clinical Decision Support
Year: 2015
Submission: 7/30/2015
Type: automatic
Task: b
MD5: a11f10ebf17718c4673b895c3a846212
Run description: For the text extraction we used BeautifulSoup(http://www.crummy.com/software/BeautifulSoup/). For the preprocessing we used Porter Stemmer and Bigram. Furthermore we expanded the queries by using Bigram and Metamap(http://metamap.nlm.nih.gov/).Furthermore we tried to extract additional information from the diagnosis field: we added terms from Metamap and also whenever it was feasible we found wiki article about the diagnossi and expanded the query by adding the first paragraph of the wiki article (for this we used a Java API (https://bitbucket.org/axelclk/info.bliki.wiki/wiki/Home). For the relevance evaluation we used our homemade Pagerank

AUEBrun2B¶

Run ID: AUEBrun2B
Participant: DBNET_AUEB
Track: Clinical Decision Support
Year: 2015
Submission: 7/30/2015
Type: automatic
Task: b
MD5: 39c5414ab0563ecc835c28b669c4f297
Run description: For the text extraction we used BeautifulSoup(http://www.crummy.com/software/BeautifulSoup/). For the preprocessing we used Porter Stemmer and Bigram. Furthermore we expanded the queries by using Bigram and Metamap(http://metamap.nlm.nih.gov/).Furthermore we tried to extract additional information from the diagnosis field: we added terms from Metamap and also whenever it was feasible we found wiki article about the diagnossi and expanded the query by adding the first paragraph of the wiki article (for this we used a Java API (https://bitbucket.org/axelclk/info.bliki.wiki/wiki/Home). For the relevance evaluation we used TFIDF

auto¶

Run ID: auto
Participant: OHSU
Track: Clinical Decision Support
Year: 2015
Submission: 7/28/2015
Type: automatic
Task: a
MD5: 2ebe5a6ffd7f29c642ae1819bbf6f5bd
Run description: vector space similarity model with dependencies and umls concept query expansion.

autob¶

Run ID: autob
Participant: OHSU
Track: Clinical Decision Support
Year: 2015
Submission: 7/31/2015
Type: automatic
Task: b
MD5: 2549e42a7dbd3d56be02372641e3a21b
Run description: OHSU automatic run for topiccb

BtBase1¶

Run ID: BtBase1
Participant: SIBtex2
Track: Clinical Decision Support
Year: 2015
Submission: 7/24/2015
Type: automatic
Task: a
MD5: 5694ed8747c59b04e12dd819f72991bd
Run description: baseline, stemming, edismax, all fields

BtBase3¶

Run ID: BtBase3
Participant: SIBtex2
Track: Clinical Decision Support
Year: 2015
Submission: 7/24/2015
Type: automatic
Task: a
MD5: d9f68056344f639a3a13044436e4eb4d
Run description: stemming, real edismax, all fields

BtCleanAll4¶

Run ID: BtCleanAll4
Participant: SIBtex2
Track: Clinical Decision Support
Year: 2015
Submission: 7/24/2015
Type: automatic
Task: a
MD5: aca92556e1a287ac01109fde72cd6a36
Run description: stemming, real edismax, all fields, strong cleaning of the queries

CAMspud1¶

Run ID: CAMspud1
Participant: CL_CAMB
Track: Clinical Decision Support
Year: 2015
Submission: 7/21/2015
Type: automatic
Task: a
MD5: 6a52ae3bf5905cec9f21abd54c6c9137
Run description: Uses a new language modelling approach with discriminative query modelling to weight salient terms of long queries. Query expansion (30 terms) is achieved using RM3 on the top 5 documents.

CAMspud3¶

Run ID: CAMspud3
Participant: CL_CAMB
Track: Clinical Decision Support
Year: 2015
Submission: 7/21/2015
Type: automatic
Task: a
MD5: 156924a5eb0d8a1037bc2612c9084cfd
Run description: Uses a new language modelling approach with discriminative query modelling to weight salient terms of long queries. Query expansion (30 terms) is achieved using a new query topic modelling technique on the top 20 documents.

CAMspud5¶

Run ID: CAMspud5
Participant: CL_CAMB
Track: Clinical Decision Support
Year: 2015
Submission: 7/21/2015
Type: automatic
Task: a
MD5: 7754d29a8f46dc0bf8ff4f38eb0e8282
Run description: Uses a new language modelling approach with discriminative query modelling to weight salient terms of long queries. Query expansion (30 terms) is achieved using RM3 on the top 5 documents.

CAMspud6¶

Run ID: CAMspud6
Participant: CL_CAMB
Track: Clinical Decision Support
Year: 2015
Submission: 7/30/2015
Type: automatic
Task: b
MD5: b9034ccf3049645545b3624049dedaf9
Run description: Language model based on Polya process. A new query model for longer queries is used with this model but no feedback was used.

CAMspud7¶

Run ID: CAMspud7
Participant: CL_CAMB
Track: Clinical Decision Support
Year: 2015
Submission: 7/30/2015
Type: automatic
Task: b
MD5: 94f8bb97788595aea3c82af736d1173a
Run description: Language model based on Polya process. A new query model for longer queries is used with this model and RM3 feedback was used.

CAMspud8¶

Run ID: CAMspud8
Participant: CL_CAMB
Track: Clinical Decision Support
Year: 2015
Submission: 7/30/2015
Type: automatic
Task: b
MD5: e25a5e6abae457b58f540076e2e98596
Run description: Language model based on Polya process. A new query model for longer queries is used with this model and a new query expansion technique was used.

cbnu0¶

Run ID: cbnu0
Participant: cbnu
Track: Clinical Decision Support
Year: 2015
Submission: 7/28/2015
Type: automatic
Task: a
MD5: c2594387ad23c3424a9735b4ecc3bf85
Run description: Baseline: language model

cbnu1¶

Run ID: cbnu1
Participant: cbnu
Track: Clinical Decision Support
Year: 2015
Submission: 7/28/2015
Type: automatic
Task: a
MD5: b3b36d7f00b7724d6bb0e5e1d56326e4
Run description: query expansion using medical terms based on UMLS and Wikipedia

cbnu2¶

Run ID: cbnu2
Participant: cbnu
Track: Clinical Decision Support
Year: 2015
Submission: 7/28/2015
Type: automatic
Task: a
MD5: 6f10377eefb9c08d3108f5c68624aedd
Run description: query expansion based on LDA model

cbnu3¶

Run ID: cbnu3
Participant: cbnu
Track: Clinical Decision Support
Year: 2015
Submission: 7/31/2015
Type: automatic
Task: b
MD5: 0cff2bf231570fe0cca394b3dddeb8ce
Run description: query expansion using medical terms based on UMLS and Wikipedia according to query types

cbnu4¶

Run ID: cbnu4
Participant: cbnu
Track: Clinical Decision Support
Year: 2015
Submission: 7/31/2015
Type: automatic
Task: b
MD5: e40a662cb9ab95e9392f6e74932d344b
Run description: query expansion using medical terms and synonym diseases based on UMLS and Wikipedia according to query types

cbnu5¶

Run ID: cbnu5
Participant: cbnu
Track: Clinical Decision Support
Year: 2015
Submission: 7/31/2015
Type: automatic
Task: b
MD5: ec03857b934fa0b1d5aef356341065d8
Run description: query expansion based on LDA model according to query types

DAIICTrun1¶

Run ID: DAIICTrun1
Participant: DA_IICT
Track: Clinical Decision Support
Year: 2015
Submission: 7/29/2015
Type: automatic
Task: a
MD5: 78a4905cbceb5f5d094f4f8ecae552d5
Run description: DAIICTrun1 uses summary as query and retrieves the documents using In_expC2 retrieval model and Bo1 Rocchio query expansion model available in terrier.

DAIICTrun2¶

Run ID: DAIICTrun2
Participant: DA_IICT
Track: Clinical Decision Support
Year: 2015
Submission: 7/29/2015
Type: automatic
Task: a
MD5: b06f5bf03a209af78f6f8e7fc9863c45
Run description: DAIICTrun2 uses description as query and retrieves the documents using In_expC2 retrieval model and Bo1 Rocchio query expansion model available in terrier.

DuthBaseF¶

Run ID: DuthBaseF
Participant: DUTH
Track: Clinical Decision Support
Year: 2015
Submission: 7/30/2015
Type: automatic
Task: b
MD5: 4c3ff813fd743c9e61c50cad80a0c2df
Run description: Automatic run using summaries in full collection.

DuthBaseS¶

Run ID: DuthBaseS
Participant: DUTH
Track: Clinical Decision Support
Year: 2015
Submission: 7/30/2015
Type: automatic
Task: b
MD5: aded89da4ed8f55d3c625dd4e70cc018
Run description: Automatic run using summaries in a automatic filtered collection.

DuthMmMt16f¶

Run ID: DuthMmMt16f
Participant: DUTH
Track: Clinical Decision Support
Year: 2015
Submission: 7/28/2015
Type: automatic
Task: a
MD5: 94e3cba17249b571ffece345c1eb1b9d
Run description: Automatic run using UMLS MetaMap and Metathesaurus in full collection.

DuthMmMt16s¶

Run ID: DuthMmMt16s
Participant: DUTH
Track: Clinical Decision Support
Year: 2015
Submission: 7/28/2015
Type: automatic
Task: a
MD5: 9ff0bba7058d510f76f1bffa99b045d1
Run description: Automatic run using UMLS MetaMap and Metathesaurus in a automatic filtered collection.

DuthMmMtB16f¶

Run ID: DuthMmMtB16f
Participant: DUTH
Track: Clinical Decision Support
Year: 2015
Submission: 7/30/2015
Type: automatic
Task: b
MD5: 2fb9d3a054ef78a90daf2e49ea259838
Run description: Automatic run using UMLS MetaMap and Metathesaurus (summaries and diagnosis) in full collection.

DuthStef¶

Run ID: DuthStef
Participant: DUTH
Track: Clinical Decision Support
Year: 2015
Submission: 7/28/2015
Type: manual
Task: a
MD5: b6576b371be81b2d466e4cbfdc3926aa
Run description: Manual run in a automatic filtered collection.

ecnu1¶

Run ID: ecnu1
Participant: ECNU
Track: Clinical Decision Support
Year: 2015
Submission: 7/28/2015
Type: automatic
Task: a
MD5: 8dae38f88e739d3f816d24c966b1dba9
Run description: combination of BM25, PL2 and BB2. BM25 is promoted by Google search expansion
Code: https://github.com/heyunh2015/ecnu_cds2015.git

ecnu2¶

Run ID: ecnu2
Participant: ECNU
Track: Clinical Decision Support
Year: 2015
Submission: 7/28/2015
Type: automatic
Task: a
MD5: 17133e1bfffe4c75bafdbe033a70fbf9
Run description: learning to rank based on randomforest, applying scikit-learn
Code: https://github.com/heyunh2015/ecnu_cds2015.git

ecnu3¶

Run ID: ecnu3
Participant: ECNU
Track: Clinical Decision Support
Year: 2015
Submission: 7/30/2015
Type: automatic
Task: b
MD5: e165a388c5c3de057fae0dceb3f4a2b4
Run description: Combination of BM25, PL2, BB2. Query is added with text of diagnosis. BM25 is promoted by Google expansion
Code: https://github.com/heyunh2015/ecnu_cds2015.git

ecnu4¶

Run ID: ecnu4
Participant: ECNU
Track: Clinical Decision Support
Year: 2015
Submission: 7/30/2015
Type: automatic
Task: b
MD5: 073a2e5eb472af43a94d1a58cbe51ac3
Run description: Combination of BM25, PL2, BB2. Query is added with text of diagnosis. Promoted by learning to rank based on SvmRank.
Code: https://github.com/heyunh2015/ecnu_cds2015.git

ECNUBP¶

Run ID: ECNUBP
Participant: ECNU
Track: Clinical Decision Support
Year: 2015
Submission: 7/30/2015
Type: automatic
Task: b
MD5: 01ce74230504bda9a7aa96b92554acfa
Run description: Combination of BM25, PL2, BB2, and position information. Query is added with text of diagnosis.
Code: https://github.com/heyunh2015/ecnu_cds2015.git

ECNUPB¶

Run ID: ECNUPB
Participant: ECNU
Track: Clinical Decision Support
Year: 2015
Submission: 7/28/2015
Type: automatic
Task: a
MD5: 2a926b2b3dba3ad2cf9025b543af4b30
Run description: Position based model. Combine kernel function and BM25 algorithm.
Code: https://github.com/heyunh2015/ecnu_cds2015.git

EMSEasmer¶

Run ID: EMSEasmer
Participant: EMSE
Track: Clinical Decision Support
Year: 2015
Submission: 7/20/2015
Type: automatic
Task: a
MD5: f698120ba3150f1628fa7f6cda79ccd6
Run description: Query reformulation in Indri using a mixed approach with both LSI feedback expansion and UMLS entity expansion using metamap for entity recognition.

EMSElsi¶

Run ID: EMSElsi
Participant: EMSE
Track: Clinical Decision Support
Year: 2015
Submission: 7/20/2015
Type: automatic
Task: a
MD5: e3e22d5552019ed11ac9784cb33b639f
Run description: Query reformulation in Indri using a LSI feedback expansion.

EMSErm3¶

Run ID: EMSErm3
Participant: EMSE
Track: Clinical Decision Support
Year: 2015
Submission: 7/20/2015
Type: automatic
Task: a
MD5: 297b344c4cf16dade73cc0a9e12f4816
Run description: Query expansion RM3 (pseudo relevance feedback) as implemented in Indri.

EPBRNBM25R2¶

Run ID: EPBRNBM25R2
Participant: EPBRN
Track: Clinical Decision Support
Year: 2015
Submission: 7/29/2015
Type: automatic
Task: a
MD5: c4e1b6151335c77a4d386c663671d5c9
Run description: We have implemented BM25 weighting as our second run to retrieve top 1000 documents.

EPBRNRSVMR3¶

Run ID: EPBRNRSVMR3
Participant: EPBRN
Track: Clinical Decision Support
Year: 2015
Submission: 7/29/2015
Type: automatic
Task: a
MD5: 504657399802cc076b6dbe2ac912111d
Run description: We used learning to rank approach using RankSVM as our third run to retrieve top 1000 documents. We trained our RankSVM model using the 2014 qrels and queries. We treated 2015 dataset as test set and predicted relevance scores on top 1000 documents which were initially retrieved based on BM25 weighting.

EPBRNTFIDFR1¶

Run ID: EPBRNTFIDFR1
Participant: EPBRN
Track: Clinical Decision Support
Year: 2015
Submission: 7/29/2015
Type: automatic
Task: a
MD5: ef823feef412865ad5392a13b57f8335
Run description: We have implemented TF-IDF weighting as baseline run to retrieve top 1000 documents.

FDUAuto¶

Run ID: FDUAuto
Participant: FDUDMIIP
Track: Clinical Decision Support
Year: 2015
Submission: 7/31/2015
Type: automatic
Task: b
MD5: 60a4f6fe680dee39773e56a3416dde37
Run description: diagnosis MeSH QE classification

FDUAuto1¶

Run ID: FDUAuto1
Participant: FDUDMIIP
Track: Clinical Decision Support
Year: 2015
Submission: 7/28/2015
Type: automatic
Task: a
MD5: 51977fa9e9d36a964c9c47f550f2d712
Run description: AnnoTator concpet keyword MeSH Query Expansion

FDUAuto2¶

Run ID: FDUAuto2
Participant: FDUDMIIP
Track: Clinical Decision Support
Year: 2015
Submission: 7/28/2015
Type: automatic
Task: a
MD5: 08a54c34dd5fcd3e8043996e5d3ca9db
Run description: AnnoTator concpet keyword MeSH Query Expansion

FDUManual¶

Run ID: FDUManual
Participant: FDUDMIIP
Track: Clinical Decision Support
Year: 2015
Submission: 7/28/2015
Type: manual
Task: a
MD5: 4be2f2ae2986ad119e4f52a945776499
Run description: Manual keyword MeSH Query Expansion

FDUManual1¶

Run ID: FDUManual1
Participant: FDUDMIIP
Track: Clinical Decision Support
Year: 2015
Submission: 7/31/2015
Type: manual
Task: b
MD5: 1cee7a20d1dffeedd77797ee34cdf138
Run description: manual key word MeSH QE classification

FDUManual2¶

Run ID: FDUManual2
Participant: FDUDMIIP
Track: Clinical Decision Support
Year: 2015
Submission: 7/31/2015
Type: manual
Task: b
MD5: 77bd641c7cd6ef003d62bdc347f09457
Run description: manual keyword/diagnosis MeSH QE classification

FORTHICSd0¶

Run ID: FORTHICSd0
Participant: FORTH_ICS_ISL
Track: Clinical Decision Support
Year: 2015
Submission: 7/27/2015
Type: automatic
Task: a
MD5: 6808781f423cba0f7c149c9888459bd8
Run description: Random Walk on a graph of documents and identified entities: Stohastic re-ranking of the top-250 results retrieved by Lucene based on named entities (diseases, drugs, proteins, and chemical substances coming from DBpedia) identified in the retrieved results. Setting: topic description as the query terms, decay factor = 0.0 for the random jumps.

FORTHICSd0e7¶

Run ID: FORTHICSd0e7
Participant: FORTH_ICS_ISL
Track: Clinical Decision Support
Year: 2015
Submission: 7/30/2015
Type: automatic
Task: b
MD5: 0301340cd6b94d4a08e07034b4e6d3c0
Run description: Random Walk on a graph of documents and enriched identified entities: Stohastic re-ranking of the top-250 results retrieved by Lucene based on named entities (diseases, drugs, proteins, and chemical substances coming from DBpedia) identified in the retrieved results and on information about the identified entities coming from DBpedia. Setting: topic description as the query terms, decay factor = 0.0 for the random jumps, probability p3 = 0.7 to move to a document-node (and p3 = 0.3 to move to a property-node) when being in an entity node.

FORTHICSd2¶

Run ID: FORTHICSd2
Participant: FORTH_ICS_ISL
Track: Clinical Decision Support
Year: 2015
Submission: 7/27/2015
Type: automatic
Task: a
MD5: 7e14f4c0f547022dfd66654b245b1659
Run description: Random Walk on a graph of documents and identified entities: Stohastic re-ranking of the top-250 results retrieved by Lucene based on named entities (diseases, drugs, proteins, and chemical substances coming from DBpedia) identified in the retrieved results. Setting: topic description as the query terms, decay factor = 0.2 for the random jumps.

FORTHICSdQE¶

Run ID: FORTHICSdQE
Participant: FORTH_ICS_ISL
Track: Clinical Decision Support
Year: 2015
Submission: 7/30/2015
Type: automatic
Task: b
MD5: 3e52f9d2b5f4ea4d117c768ddd61bf80
Run description: Random Walk on a graph of documents and entities and then query expansion based on the top-k scored entities. Setting: topic description as the query terms, decay factor = 0.0 for the random jumps, K = 10.

FORTHICSdQER¶

Run ID: FORTHICSdQER
Participant: FORTH_ICS_ISL
Track: Clinical Decision Support
Year: 2015
Submission: 7/30/2015
Type: automatic
Task: b
MD5: 7eacb2a545d29412a8e48b709e5cec58
Run description: Random Walk on a graph of documents and entities, query expansion based on the top-K scored entities, and then again random walk on the graph of the new documents and entities. Setting: topic description as the query terms, decay factor = 0.0 for the random jumps, K = 10.

FORTHICSs0¶

Run ID: FORTHICSs0
Participant: FORTH_ICS_ISL
Track: Clinical Decision Support
Year: 2015
Submission: 7/27/2015
Type: automatic
Task: a
MD5: 06d61fc0df9fa6f9ac0651551ca31871
Run description: Random Walk on a graph of documents and identified entities: Stohastic re-ranking of the top-250 results retrieved by Lucene based on named entities (diseases, drugs, proteins, and chemical substances coming from DBpedia) identified in the retrieved results. Setting: topic summary as the query terms, decay factor = 0.0 for the random jumps.

FrameAFinal¶

Run ID: FrameAFinal
Participant: SCIAITeam
Track: Clinical Decision Support
Year: 2015
Submission: 7/27/2015
Type: automatic
Task: a
MD5: d4220f4973de08616686d457827d2bbe
Run description: Our baseline Lucene run was expanded upon to frame the top 30 documents for selected entities of interest. Our framing technique was based on the HITIQA framing technique from Small et al. The top 10 documents with the highest scoring frames were then selected from the 30 for this submission.

FusionAdv¶

Run ID: FusionAdv
Participant: Foreseer
Track: Clinical Decision Support
Year: 2015
Submission: 7/31/2015
Type: manual
Task: b
MD5: 6d75ee6c2b0649db1b836604390ea3d8
Run description: This is a manual run based on the FusionManB. We create another two ranking lists, one with no topic type information and another with #syn(Diagnosis). Those three rankings are fused by simple linear rule.

FusionAuto¶

Run ID: FusionAuto
Participant: Foreseer
Track: Clinical Decision Support
Year: 2015
Submission: 7/29/2015
Type: automatic
Task: a
MD5: 7bb1808b7ff28cec1d28434d41d95d9c
Run description: This is a submission based on several ranking lists. We used a learning to rank package called RankLib and selected Random Forest algorithm to combine the results from different automatic runs, including the text PRF retrieval results and Metamap concepts PRF retrieval results.

FusionAutoB¶

Run ID: FusionAutoB
Participant: Foreseer
Track: Clinical Decision Support
Year: 2015
Submission: 7/30/2015
Type: automatic
Task: b
MD5: b68bbf22bfc4c215440d4330b9e267ba
Run description: This is an automatic run based on the FusionAuto submission for Task A. Add the words and CUIs of diagnosis and use the model trained for FusionAuto to do the fusion job.

FusionMAll¶

Run ID: FusionMAll
Participant: Foreseer
Track: Clinical Decision Support
Year: 2015
Submission: 7/29/2015
Type: manual
Task: a
MD5: 49dac101ccd23fbf83c2fd1ac75178f0
Run description: This is a submission generated from different ranking lists, including the free text retrieval after manual modification by domain expert, free text PRF retrieval and Metamap concepts retrieval.

FusionManB¶

Run ID: FusionManB
Participant: Foreseer
Track: Clinical Decision Support
Year: 2015
Submission: 7/30/2015
Type: manual
Task: b
MD5: 4f4656a3a76cef21ddd4eeec7c5685c4
Run description: This is an manual run based on the FusionManual submission for Task A. Add the words and CUIs of diagnosis and use the model trained for FusionManual to do the fusion job.

FusionManual¶

Run ID: FusionManual
Participant: Foreseer
Track: Clinical Decision Support
Year: 2015
Submission: 7/29/2015
Type: manual
Task: a
MD5: 01c828a20ade3bfc05b0714bb0b0d910
Run description: This is a submission based on several ranking lists. We used a learning to rank package called RankLib and selected Random Forest algorithm to combine the results from different automatic runs and manual runs, including the text PRF retrieval results, Metamap concepts PRF retrieval results and a text retrieval result based on manual query modifications by domain expert.

GRIUMenRun1¶

Run ID: GRIUMenRun1
Participant: GRIUM
Track: Clinical Decision Support
Year: 2015
Submission: 7/28/2015
Type: automatic
Task: a
MD5: 64a80eff949a9ff8620e7f720aef5ad3
Run description: Now there are 2 runs. Run1 is using BM25 model. Run2 is using language model with concept exaction from query. The external resources used are UMLS and Metamap. I plan to submit Run3. But it is not done. If I can finish it in several days. I will submit it later. thanks.
Code: http://github.rep

GRIUMenRun2¶

Run ID: GRIUMenRun2
Participant: GRIUM
Track: Clinical Decision Support
Year: 2015
Submission: 7/28/2015
Type: automatic
Task: a
MD5: 64e9a531b7927b35ef2b9d9e7967e96a
Run description: Now there are 2 runs. Run1 is using BM25 model. Run2 is using language model with concept exaction from query. The external resources used are UMLS and Metamap. I plan to submit Run3. But it is not done. If I can finish it in several days. I will submit it later. thanks.
Code: http://github.rep

HipocratAr1¶

Run ID: HipocratAr1
Participant: Hipocrates15
Track: Clinical Decision Support
Year: 2015
Submission: 7/15/2015
Type: automatic
Task: a
MD5: 727e6257b2f9f58808cfb919b960653e
Run description: HipocratAr1 is a run submitted to the A Task. All the processes involved to generate the final results are automatic. We developed different programs for pre-processing of the data and final formatting of the results. The index was created using the Terrier Platform (an open access information retrieval system for research). Retrieval was carried out on the description tag using the TF IDF weighting measure as designed in Lemur (another information retrieval open source software for research) in cluded in Terrier.

HipocratAr2¶

Run ID: HipocratAr2
Participant: Hipocrates15
Track: Clinical Decision Support
Year: 2015
Submission: 7/15/2015
Type: automatic
Task: a
MD5: bdbbb49e1b02d69f7f737ef7115c898e
Run description: HipocratAr2 is a run submitted to the A Task. All the processes involved to generate the final results are automatic. We developed different programs for pre-processing of the data and final formatting of the results. The index was created using the Terrier Platform (an open access information retrieval system for research). Retrieval was carried out on the description tag using a measure that weighs query terms with respect to their proximity, Divergence from Randomness based dependence model, as implemented in the Terrier Platform.

HipocratAr3¶

Run ID: HipocratAr3
Participant: Hipocrates15
Track: Clinical Decision Support
Year: 2015
Submission: 7/16/2015
Type: manual
Task: a
MD5: 82c064f687e6b7b5220722cad1acffdd
Run description: HipocratAr3: This run is submitted for Task A. We created an index using the Terrier Platform that is an open access retrieval platform for research. The queries were modified manually to include terms from the description and the summary eliminating duplicates. These terms were mainly nouns that we also searched in MESH (UMLS) to add, when applicable, other terms related to the topic. Retrieval was carried out with the Divergence from Randomness based dependence model as implemented in the Terrier platform.

hltcoe4sdrf¶

Run ID: hltcoe4sdrf
Participant: hltcoe
Track: Clinical Decision Support
Year: 2015
Submission: 7/31/2015
Type: automatic
Task: b
MD5: 0170f19fcfbd3e6551a2e0dde1bf27dc
Run description: Domain-independent run. No medical or for that matter any external resources used. JHU HAIRCUT retrieval engine with character 4-gram tokenization. Summaries. Diagnosis if available. Relevance feedback employed.

hltcoe4srf¶

Run ID: hltcoe4srf
Participant: hltcoe
Track: Clinical Decision Support
Year: 2015
Submission: 7/29/2015
Type: automatic
Task: a
MD5: f649b1a0fa036ffd0008b3d67878ffc8
Run description: Domain-independent run. No medical or for that matter any external resources used. JHU HAIRCUT retrieval engine with character 4-gram tokenization. Relevance feedback employed.

hltcoe5sdrf¶

Run ID: hltcoe5sdrf
Participant: hltcoe
Track: Clinical Decision Support
Year: 2015
Submission: 7/31/2015
Type: automatic
Task: b
MD5: a8ad8069d298da8f22c3139644edbe60
Run description: Domain-independent run. No medical or for that matter any external resources used. JHU HAIRCUT retrieval engine with character 5-gram tokenization. Summaries. Diagnosis if available. Relevance feedback employed.

hltcoe5srf¶

Run ID: hltcoe5srf
Participant: hltcoe
Track: Clinical Decision Support
Year: 2015
Submission: 7/29/2015
Type: automatic
Task: a
MD5: 73be8257c793895b493fb108c009c72b
Run description: Domain-independent run. No medical or for that matter any external resources used. JHU HAIRCUT retrieval engine with character 5-gram tokenization. Relevance feedback employed.

hltcoewsdrf¶

Run ID: hltcoewsdrf
Participant: hltcoe
Track: Clinical Decision Support
Year: 2015
Submission: 7/31/2015
Type: automatic
Task: b
MD5: 551132640de1158160f98eeac151a97e
Run description: Domain-independent run. No medical or for that matter any external resources used. JHU HAIRCUT retrieval engine with word-based tokenization. Summaries. Diagnosis if available. Relevance feedback employed.

hltcoewsrf¶

Run ID: hltcoewsrf
Participant: hltcoe
Track: Clinical Decision Support
Year: 2015
Submission: 7/29/2015
Type: automatic
Task: a
MD5: e4deea03b1fee494ebef6ba3876f0734
Run description: Domain-independent run. No medical or for that matter any external resources used. JHU HAIRCUT retrieval engine with plain word tokenization. Relevance feedback employed.

hybrid¶

Run ID: hybrid
Participant: OHSU
Track: Clinical Decision Support
Year: 2015
Submission: 7/28/2015
Type: manual
Task: a
MD5: 067a29de84806b6b5d484f3cddbb7e0f
Run description: A hybrid of the two OHSU approaches. reranks results by a linear combination of the manual and automatic score.

hybridb¶

Run ID: hybridb
Participant: OHSU
Track: Clinical Decision Support
Year: 2015
Submission: 7/31/2015
Type: manual
Task: b
MD5: 255f600815ae74f1839c1c0cf531ced0
Run description: ohsu hybrid run b.

KISTI001¶

Run ID: KISTI001
Participant: KISTI
Track: Clinical Decision Support
Year: 2015
Submission: 7/28/2015
Type: automatic
Task: a
MD5: db7f6c480846b00b68cb91fc826fe08c
Run description: query-likelihood with Diriclet smoothing

KISTI001B¶

Run ID: KISTI001B
Participant: KISTI
Track: Clinical Decision Support
Year: 2015
Submission: 7/29/2015
Type: automatic
Task: b
MD5: e588c5134111be925d1b3a7d8916c056
Run description: query-likelihood with Dirichlet smoothing

KISTI002¶

Run ID: KISTI002
Participant: KISTI
Track: Clinical Decision Support
Year: 2015
Submission: 7/28/2015
Type: automatic
Task: a
MD5: 821f1ea0b8133431de3b61a276ea2baf
Run description: cluster-based external expansion model

KISTI002B¶

Run ID: KISTI002B
Participant: KISTI
Track: Clinical Decision Support
Year: 2015
Submission: 7/29/2015
Type: automatic
Task: b
MD5: f6be730d6f0d1202c70011e744e672ac
Run description: cluster-based external expansion model

KISTI003¶

Run ID: KISTI003
Participant: KISTI
Track: Clinical Decision Support
Year: 2015
Submission: 7/28/2015
Type: automatic
Task: a
MD5: 3d6a5e9a9ac49b35d08433677e4f0e60
Run description: field-based feedback model

KISTI003B¶

Run ID: KISTI003B
Participant: KISTI
Track: Clinical Decision Support
Year: 2015
Submission: 7/29/2015
Type: automatic
Task: b
MD5: 9aca76351235900e5366413520fcc169
Run description: field-based feedback model

lamdarun01¶

Run ID: lamdarun01
Participant: LAMDA
Track: Clinical Decision Support
Year: 2015
Submission: 7/28/2015
Type: automatic
Task: a
MD5: 9db7b31e44804f52935223699ca83e66
Run description: First of all, we made NXML Parser by java. The Parser is parsing all of offered Nxml files by using java XML libraries and DTD documents and we extract contents of pmcid, title, abstract and body which is a XML tag that we want to extract by using XPath. These extracted information are stored in MongoDB based on Nosql and we constructed the database. Second, we used ElasticSearch Framework based on Lucene for Information retrieval. We found data from MongoDB by searching pmcid and indexed in Elasticsearch. We used standard tokenization which offered from Elasticsearch and lemmatized words. For search, Query are made by using both Summary and Query Expansion. In Query Expansion, we discovered Metathesaurus concepts of Description which offered in topic by using Metamap and used words which have specific Semantic types from the result of Metamap to Query expansion. The specific Semantic types are disease name, disease symptom, part of body, process of diagnosis and finding symptom.
Code: https://github.com/Lamda-TREC/TREC-CDSS

lamdarun02¶

Run ID: lamdarun02
Participant: LAMDA
Track: Clinical Decision Support
Year: 2015
Submission: 7/28/2015
Type: automatic
Task: a
MD5: 8d4dcda86ca4d8f36c0074322c6c415b
Run description: First of all, we made NXML Parser by java. The Parser is parsing all of offered Nxml files by using java XML libraries and DTD documents and we extract contents of pmcid, title, abstract and body which is a XML tag that we want to extract by using XPath. These extracted information are stored in MongoDB based on Nosql and we constructed the database. Second, we used ElasticSearch Framework based on Lucene for Information retrieval. We found data from MongoDB by searching pmcid and indexed in Elasticsearch. We used standard tokenization which offered from Elasticsearch and lemmatized words and applied Edge Ngram to each word. Also, we used Divergence From Randomness (DFR) as a Similarity model. For search, Query are made by using both Summary and Query Expansion. In Query Expansion, we discovered Metathesaurus concepts of Description which offered in topic by using Metamap and used words which have specific Semantic types from the result of Metamap to Query expansion. The specific Semantic types are disease name, disease symptom, part of body, process of diagnosis and finding symptom.
Code: https://github.com/Lamda-TREC/TREC-CDSS

lamdarun03¶

Run ID: lamdarun03
Participant: LAMDA
Track: Clinical Decision Support
Year: 2015
Submission: 7/28/2015
Type: automatic
Task: a
MD5: aa76629d0ab1e92417776c902b74c292
Run description: First of all, we made NXML Parser by java. The Parser is parsing all of offered Nxml files by using java XML libraries and DTD documents and extracts contents of pmcid, title, abstract and body which is a XML tag that we want to extract by using XPath. These extracted information are stored in MongoDB based on Nosql and we constructed the database. Second, we used ElasticSearch Framework based on Lucene for Information retrieval. We found data from MongoDB by searching pmcid and indexed in Elasticsearch. We used standard tokenization which offered from Elasticsearch and lemmatized words. For search, Query are made by using both Summary and Query Expansion. In Query Expansion, we discovered Metathesaurus concepts of Description which offered in topic by using Metamap and used words which have specific Semantic types from the result of Metamap to Query expansion. The specific Semantic types are disease name and disease symptom, part of body, process of diagnosis and finding symptom. In addition, we adapted the weight boosting to query filed. We found the optimization weight value with 2014 topics and applied the optimized weight value in title, abstract and body.
Code: https://github.com/Lamda-TREC/TREC-CDSS

lamdarun04¶

Run ID: lamdarun04
Participant: LAMDA
Track: Clinical Decision Support
Year: 2015
Submission: 7/30/2015
Type: automatic
Task: b
MD5: be00463a811c8c485ac766c367eb510a
Run description: First of all, we made NXML Parser by java. The Parser is parsing all of offered Nxml files by using java XML libraries and DTD documents and extracts contents of pmcid, title, abstract and body which is a XML tag that we want to extract by using XPath. These extracted information are stored in MongoDB based on Nosql and we constructed the database. Second, we used ElasticSearch Framework based on Lucene for Information retrieval. We found data from MongoDB by searching pmcid and indexed in Elasticsearch. We used standard tokenization which offered from Elasticsearch and lemmatized words. For search, Query are made by using both Summary and Query Expansion. In Query Expansion, we discovered Metathesaurus concepts of Description which offered in topic by using Metamap and used words which have specific Semantic types from the result of Metamap to Query expansion. The specific Semantic types are disease name and disease symptom, part of body, process of diagnosis and finding symptom. Also, we adapted the weight boosting to query filed. We found the optimization weight value with 2014 topics and applied the optimizated weight value in title, abstract and body. In addition, we make query expansion which expanded from the new diagnosis field to their synonyms by using Metamap.
Code: https://github.com/Lamda-TREC/TREC-CDSS

lamdarun05¶

Run ID: lamdarun05
Participant: LAMDA
Track: Clinical Decision Support
Year: 2015
Submission: 7/30/2015
Type: automatic
Task: b
MD5: 4dad7b8b467c2f592211c6d6f435373d
Run description: First of all, we made NXML Parser by java. The Parser is parsing all of offered Nxml files by using java XML libraries and DTD documents and extracts contents of pmcid, title, abstract and body which is a XML tag that we want to extract by using XPath. These extracted information are stored in MongoDB based on Nosql and we constructed the database. Second, we used ElasticSearch Framework based on Lucene for Information retrieval. We found data from MongoDB by searching pmcid and indexed in Elasticsearch. We used standard tokenization which offered from Elasticsearch and lemmatized words. For search, Query are made by using both Summary and Query Expansion. In Query Expansion, we discovered Metathesaurus concepts of Description which offered in topic by using Metamap and used words which have specific Semantic types from the result of Metamap to Query expansion. The specific Semantic types are disease name and disease symptom, part of body, process of diagnosis and finding symptom. Also, for query expansion of the new diagnosis, we used Wikipedia. We brought information of each new diagnosis and extracted term which have specific Semantic types from the result of Metamap. The specific Semantic types are disease name and disease symptom, part of body and process of diagnosis. After that we applied LDA(Latent Dirichlet Allocation) to the terms and selected top 20 terms.
Code: https://github.com/Lamda-TREC/TREC-CDSS

lamdarun06¶

Run ID: lamdarun06
Participant: LAMDA
Track: Clinical Decision Support
Year: 2015
Submission: 7/30/2015
Type: automatic
Task: b
MD5: 0e77b4ec67ab4a258f35e983d88d0f4f
Run description: First of all, we made NXML Parser by java. The Parser is parsing all of offered Nxml files by using java XML libraries and DTD documents and extracts contents of pmcid, title, abstract and body which is a XML tag that we want to extract by using XPath. These extracted information are stored in MongoDB based on Nosql and we constructed the database. Second, we used ElasticSearch Framework based on Lucene for Information retrieval. We found data from MongoDB by searching pmcid and indexed in Elasticsearch. We used standard tokenization which offered from Elasticsearch and lemmatized words. For search, Query are made by using both Summary and Query Expansion. In Query Expansion, we discovered Metathesaurus concepts of Description which offered in topic by using Metamap and used words which have specific Semantic types from the result of Metamap to Query expansion. The specific Semantic types are disease name and disease symptom, part of body, process of diagnosis and finding symptom. In addition, we calculated new score with Borda Fuse scoring method. To be specific, we run two times. In first run, we made query only using new diagnosis and their synonyms which are preferred by Metamap. In the second run, we joined queries which are original expanded query and the first run query. After that we recalculated score by using the two sets of score and Borda Fuse scoring method.
Code: https://github.com/Lamda-TREC/TREC-CDSS

LIMSIrun1BoW¶

Run ID: LIMSIrun1BoW
Participant: LIMSI
Track: Clinical Decision Support
Year: 2015
Submission: 7/29/2015
Type: automatic
Task: a
MD5: ecb5096bec4ab787356870de1d85d118
Run description: The corpus was indexed using Solr (Lucene). In this run we first performed a query using all words from the topic (Summary) to search Title, Abstract and Text fields of documents in the corpus. We then performed Pseudo-Relevance Feedback using terms from the Title and Abstract fields from the retrieved documents. The adjusted query was used to search Title, Abstract and Text fields again. In a second step the retrieved document were reranked for Clinical Question Type and time (to give more preference to recent documents). We used the weighted Borda Fuse algorithm for this. For the additional reranking of retrieved results (according to Clinical Question Type), we used the Linguistic Classification System (LCS). The classification models were built by selecting documents that were most suited to (one of) respective Clinical Question Types. This selection was carried out using manually constructed MeSH queries on the PMC corpus.

LIMSIrun2MSH¶

Run ID: LIMSIrun2MSH
Participant: LIMSI
Track: Clinical Decision Support
Year: 2015
Submission: 7/29/2015
Type: automatic
Task: a
MD5: 9ec0bc5a6407b04cc3f22511a1d984d9
Run description: The corpus was indexed using Solr (Lucene). We run the MTI indexer on the topics and on the documents in the corpus that had no (indexed) MeSH terms. In this run we first performed a query using all words from the topic (Description) to search Title, Abstract and Text fields and MeSH terms of documents in the corpus. We replicated the MeSH term explosion during the indexing phase. We then performed Pseudo-Relevance Feedback using terms from the Title and Abstract fields from the retrieved documents. The adjusted query was used to search Title, Abstract and Text fields again. In a second step the retrieved document were reranked for Clinical Question Type and time (to give more preference to recent documents). We used the weighted Borda Fuse algorithm for this. For the additional reranking of retrieved results (according to Clinical Question Type), we used the Linguistic Classification System (LCS). The classification models were built by selecting documents that were most suited to (one of) respective Clinical Question Types. This selection was carried out using manually constructed MeSH queries on the PMC corpus.

LIMSIrun3SmF¶

Run ID: LIMSIrun3SmF
Participant: LIMSI
Track: Clinical Decision Support
Year: 2015
Submission: 7/29/2015
Type: automatic
Task: a
MD5: e19cc54615b05c8e2c0cac44700c1782
Run description: The corpus was indexed using Solr (Lucene). In this run we first performed a query using all words from the topic (Summary) to search Title, Abstract and Text fields of documents in the corpus. We then performed Pseudo-Relevance Feedback using terms from the Title and Abstract fields from the retrieved documents. The adjusted query was used to search Title, Abstract and Text fields again. In a second step we performed Pseudo-Relevance Feedback on the top50 returned documents from the adjusted query. The terms with highest TF/IDF counts were processed through MetaMap with restrictions on which semantic types were allowed (to account for different Clinical Question Types). Terms that corresponded to selected semantic Types were boosted in the final query (which comprises the adjusted query + the selected MetaMap terms). In this run we only reranked for time afterwards.

LIMSIrun4Syn¶

Run ID: LIMSIrun4Syn
Participant: LIMSI
Track: Clinical Decision Support
Year: 2015
Submission: 7/31/2015
Type: automatic
Task: b
MD5: 32cf186e49844ffb96efa49ab1a35b2b
Run description: In this run we used MetaMap to convert the diagnoses to CUI, which were then used to find synonyms in the UMLS. For topics 1-10 (diagnosis) we used the topics to query a corpus of 2100 disease pages that were extracted from Wikipedia en Medline+. The diagnosis synonyms were combined with the words from the original topic to query the Title, Abstract and Text fields. More weight was given to diagnosis terms found in the title and abstract fields.

LIMSIrun5MPF¶

Run ID: LIMSIrun5MPF
Participant: LIMSI
Track: Clinical Decision Support
Year: 2015
Submission: 7/31/2015
Type: automatic
Task: b
MD5: eb6305c1fca80a76605abba8ccee9c7e
Run description: In this run we used the MTI indexer to convert diagnoses to MeSH terms. (For topic 1-10 we searched diagnoses in a corpus of 2100 disease webpages extracted from Wikipedia and Medline+.) We combined these terms with a manually selected set of Main Heading and Subheadings (one set per Clinical Question Type) to form MeSH queries. These were then used to query our MeSH version of the corpus. In this version, we have extracted all MeSH terms from their respective documents in the PMC corpus. Those documents that did not have MeSH terms were assigned terms with the MTI indexer. In a second step we performed Pseudo-Relevance Feedback on the retrieved results (using text from the Title and Abstract fields) to increase recall. The final results were then reranked timewise to boost more recent documents.

LIMSIrun6Wik¶

Run ID: LIMSIrun6Wik
Participant: LIMSI
Track: Clinical Decision Support
Year: 2015
Submission: 7/31/2015
Type: automatic
Task: b
MD5: 29efa1972ad7389926db2e3b75f1ab6f
Run description: In this run we extracted Clinical Question Type-specific terms from Wikipedia to supplement the query. This was done by selecting part of the Wikipedia page that belonged to each diagnoses ('Signs and Symptoms' for diagnosis; 'Diagnosis' for test; and 'Treatment' or 'Managing' for treatment). In two cases we did not find a relevant paragraph in the Wikipedia article. As a back-off we used the introductory paragraph. These paragraphs were then processed using MetaMap, with severe restrictions on the semantic types of the concepts that should be found per Clinical Question Type. For each recognized concept, we looked up the semantic variants in the UMLS. The diagnoses were treated similarly: We processed them with MetaMap to identify their CUI. (For the ten first topics we provided our own diagnoses, retrieved from a corpus of 2100 disease webpages from Wikipedia and Medline+.) For each CUI we then extracted the semantic variants from the UMLS. Diagnosis terms and Clinical Question Type-specific terms were then combined in a query, which was used to search the Title, abstract and text fields. Terms that were found in the title or abstract received a boosted weigth of 4 and 3 respectively.

manual¶

Run ID: manual
Participant: OHSU
Track: Clinical Decision Support
Year: 2015
Submission: 7/28/2015
Type: manual
Task: a
MD5: 3bbc7c56229c187c288c5ffb9975a09e
Run description: manual queries with lucene elastic search and MeSH.

manualb¶

Run ID: manualb
Participant: OHSU
Track: Clinical Decision Support
Year: 2015
Submission: 7/30/2015
Type: manual
Task: b
MD5: e833703a4698675534a358166941de3b
Run description: Manual query against lucene index with mesh terms.

NOVASEARCH1¶

Run ID: NOVASEARCH1
Participant: NOVASEARCH
Track: Clinical Decision Support
Year: 2015
Submission: 7/29/2015
Type: automatic
Task: a
MD5: 8d3b2e44c1580c8b19bb5cc488595128
Run description: Fusion of multiple runs with RRF rank fusion. Individual runs with multiple similarity functions (BM25L, BM25+, Language Models and TF-IDF). All runs with MeSH expansion and PRF from top documents. External resources: MeSH

NOVASEARCH2¶

Run ID: NOVASEARCH2
Participant: NOVASEARCH
Track: Clinical Decision Support
Year: 2015
Submission: 7/29/2015
Type: automatic
Task: a
MD5: 3ba9614a3f125fe09a8b91ea7813b323
Run description: BM25L ret. function, MeSH expansion, PRF and custom reranking to improve precision at top results. External resources: MeSH

NOVASEARCH3¶

Run ID: NOVASEARCH3
Participant: NOVASEARCH
Track: Clinical Decision Support
Year: 2015
Submission: 7/29/2015
Type: automatic
Task: a
MD5: 631bfa5a9a351c9f2fae362a7bd65e55
Run description: Selection of best runs that apply MeSH and SNOMED CT term expansion, based on last year results. External resources: MeSH and SNOMED CT

NOVASEARCH4¶

Run ID: NOVASEARCH4
Participant: NOVASEARCH
Track: Clinical Decision Support
Year: 2015
Submission: 7/30/2015
Type: automatic
Task: b
MD5: ab2c40f39c234de2128fb235331c229e
Run description: BM25L ret. function, MeSH expansion, PRF and custom reranking to improve precision at top results. Similar to NOVASEARCH2 with diagnostic terms appended to the query. External resources: MeSH

NOVASEARCH5¶

Run ID: NOVASEARCH5
Participant: NOVASEARCH
Track: Clinical Decision Support
Year: 2015
Submission: 7/30/2015
Type: automatic
Task: b
MD5: e7db5a1287e2db10063042eedcb7f4fd
Run description: Selection of best runs that apply MeSH and SNOMED CT term expansion and PRF, based on last year results. Diagnostic terms appended to the query. External resources: MeSH and SNOMED CT

NOVASEARCH6¶

Run ID: NOVASEARCH6
Participant: NOVASEARCH
Track: Clinical Decision Support
Year: 2015
Submission: 7/30/2015
Type: automatic
Task: b
MD5: 9a7b460b98207210e8abc4f75706aff2
Run description: NOVASEARCH5 run with custom journal relevance filtering, to keep only articles from journals relevant for each query type. External resources: MeSH and SNOMED CT

nuuuuncDFML¶

Run ID: nuuuuncDFML
Participant: NU_UU_UNC
Track: Clinical Decision Support
Year: 2015
Submission: 7/29/2015
Type: manual
Task: a
MD5: b5d0a51fb7be1174f02136cf4cbda749
Run description: We have indexed the entire corpus in a lucene index which contains: document full text, document title, Semantic concepts form the document abstract. The application MedTagger, formerly cTAKES, is used to extract the semantic concepts of specific features from the Unified Medical Language Systems (UMLS). Our process includes: a. Manual: based on templates instantiated with UMLS concepts provided by a physician - A Physician expert produced diagnosis and specific pairings for each Case Topic. The concepts were extracted and mapped to UMLS using MedTagger. b. MedTagger for concept extraction both from case summaries and titles/abstracts - MedTagger (formerly known as cTAKES) is an open source Natural Language Processing tool which maps free text to UMLS concepts. c. Lucene for document indexing - Lucene is an open source tool for information retrieval maintained by Apache. We are using version 5.1.0 to index the features which we derive from the documents and to search the index with features derived from the Case Topics. d. Google for diagnoses - A python script uses the summary of the topics to start a Google Custom Search Engine (CSE) request and collecting all titles. The search is limited to wikimedz.com, nlm.nih.gov/medlineplus, MayoClinic.org, WebMD.com and Wikipedia.org. In a second step MedTagger is identifying concepts within the titles to reduce noise. e. Clinical Queries filter - Clinical Queries are a sub-collection of PubMed specific to a clinical topic such as Diagnosis, and which meet specific methodological criteria. These articles are those that are returned in response to a hand-curated PubMed search. DFML is an iterative approach to combine data fusion to rank documents and machine learning to classify degree of relevance. As the TREC Clinical Decision Support Track uses Normalized Discounted Cumulative Gain (NDCG) for measuring performance, the errors in the task are not only bounded by ranking errors, but also bounded by classification error. To normalize those errors, the linear combination method was used to select 1,000 relevant documents for the first step by fusing features, and the Ordinary Logistic Regression (OLR) was used to classify relevance of those selected documents for the second step. TF-IDF similarity scores between title, keywords, or full-text and summary were main features for the machine learning process, and UMLS concepts extracted from Medtagger were used by using original content, Google expansion, and manual expansion. Clinical queries filter was also utilized as one of features. The linear combination used the precision@100 by independent feature as its weight after conducting parameter sweeping. The class probabilities from OLR were converted into ranking scores by assuming expected relevance score as sum of the class probabilities multiplied by graded relevance scale. The final ranks was produced according to the estimated relevance score. This approach leverages classical ranking algorithm and machine learning for classification to suit two major dimensions of information retrieval: ranking and classification.

nuuuuncDFMLB¶

Run ID: nuuuuncDFMLB
Participant: NU_UU_UNC
Track: Clinical Decision Support
Year: 2015
Submission: 7/30/2015
Type: automatic
Task: b
MD5: c0886bd53ea79368263ea7dbeb8c4300
Run description: We have indexed the entire corpus in a lucene index which contains: document full text, document title, Semantic concepts form the document abstract. The application MedTagger, formerly cTAKES, is used to extract the semantic concepts of specific features from the Unified Medical Language Systems (UMLS). Our process includes: a. Manual: based on templates instantiated with UMLS concepts provided by a physician - A Physician expert produced diagnosis and specific pairings for each Case Topic. The concepts were extracted and mapped to UMLS using MedTagger. b. MedTagger for concept extraction both from case summaries and titles/abstracts - MedTagger (formerly known as cTAKES) is an open source Natural Language Processing tool which maps free text to UMLS concepts. c. Lucene for document indexing - Lucene is an open source tool for information retrieval maintained by Apache. We are using version 5.1.0 to index the features which we derive from the documents and to search the index with features derived from the Case Topics. d. Google for diagnoses - A python script uses the summary of the topics to start a Google Custom Search Engine (CSE) request and collecting all titles. The search is limited to wikimedz.com, nlm.nih.gov/medlineplus, MayoClinic.org, WebMD.com and Wikipedia.org. In a second step MedTagger is identifying concepts within the titles to reduce noise. e. Clinical Queries filter - Clinical Queries are a sub-collection of PubMed specific to a clinical topic such as Diagnosis, and which meet specific methodological criteria. These articles are those that are returned in response to a hand-curated PubMed search. DFMLB is an iterative approach to combine data fusion to rank documents and machine learning to classify degree of relevance. As the TREC Clinical Decision Support Track uses Normalized Discounted Cumulative Gain (NDCG) for measuring performance, the errors in the task are not only bounded by ranking errors, but also bounded by classification error. To normalize those errors, the relevance scores from each approach were converged into one ranking score after normalization. We used multi-classification for Task A, but we changed it to binary classification to keep our approach simple and to reduce some noises from multi-classification for Task B . TF-IDF similarity scores between document title, keywords, or full-text and topic summary were main features for the machine learning process, and UMLS concepts extracted by Medtagger were compared between document abstract, title, or summary and topic summary. For diagnosis topic, UMLS concepts were expanded by high ranked Google search results, and for test and treatment topic they were expanded by the given diagnosis. To make training data, one physician in our team manually produced diagnosis information for 2014 dataset. Diagnosis and therapy information from MeSH were also utilized. The linear combination used the precision@100 by independent feature as its weight after conducting parameter sweeping. The class probability for relevance by the logistic regression was used as ranking scores. The final ranks was produced by combining normalized two ranking scores. This approach leverages classical ranking algorithm and machine learning for classification to suit two major dimensions of information retrieval: ranking and classification.

nuuuuncHAKT¶

Run ID: nuuuuncHAKT
Participant: NU_UU_UNC
Track: Clinical Decision Support
Year: 2015
Submission: 7/29/2015
Type: automatic
Task: a
MD5: 6c58016f1bc94dac3e697b98121849d4
Run description: We have indexed the entire corpus in a lucene index which contains: document full text, document title, Semantic concepts form the document abstract. The application MedTagger, formerly cTAKES, is used to extract the semantic concepts of specific features from the Unified Medical Language Systems (UMLS). Our process includes: a. Manual: based on templates instantiated with UMLS concepts provided by a physician - A Physician expert produced diagnosis and specific pairings for each Case Topic. The concepts were extracted and mapped to UMLS using MedTagger. b. MedTagger for concept extraction both from case summaries and titles/abstracts - MedTagger (formerly known as cTAKES) is an open source Natural Language Processing tool which maps free text to UMLS concepts. c. Lucene for document indexing - Lucene is an open source tool for information retrieval maintained by Apache. We are using version 5.1.0 to index the features which we derive from the documents and to search the index with features derived from the Case Topics. d. Google for diagnoses - A python script uses the summary of the topics to start a Google Custom Search Engine (CSE) request and collecting all titles. The search is limited to wikimedz.com, nlm.nih.gov/medlineplus, MayoClinic.org, WebMD.com and Wikipedia.org. In a second step MedTagger is identifying concepts within the titles to reduce noise. e. Clinical Queries filter - Clinical Queries are a sub-collection of PubMed specific to a clinical topic such as Diagnosis, and which meet specific methodological criteria. These articles are those that are returned in response to a hand-curated PubMed search. 3. Hierarchal + keywords +title HAKT (pounced hack-it) HAKT is a semantic hierarchical iterative approach using heuristically derived filters for diagnosis and symptoms. The algorithm uses a nested logic to prioritize different search strategies and tuning of filtering. The searches are enhanced by using a boolean combination of the query from the hierarchy, a keyword search, a title search and a term based on the topic type. Two cycles of searching are performed till at least a thousand search results are found. The inner cycle begins with using the semantic concepts of the google feature created diagnosis to search the titles of all documents. Next in the hierarchy, the concepts found in the google search feature are used to search the semantic concepts found in the document abstract. This is followed by a feature containing a filtered subset of diagnosis concepts found in the topic summary. The next level of the hierarchy contains symptom concepts from the topic summary. The results of each traversal of the hierarchy are filtered to enrich for clinically relevant documents. Three passes through the hierarchy are made. In pass 1, the documents must be in the set of documents meeting the NCBI Clinical Query for the topic type. For example, diagnosis topics will only return documents meeting the diagnosis Clinical Query. The second pass uses a less stringent filter of documents meeting the Clinical query for Treatment or Diagnosis. The third pass removes the Clinical Queries filter to allow for maximum retrieval.

nuuuuncHMKTB¶

Run ID: nuuuuncHMKTB
Participant: NU_UU_UNC
Track: Clinical Decision Support
Year: 2015
Submission: 7/30/2015
Type: automatic
Task: b
MD5: ded8f26b2916eb086917d34382828ba4
Run description: We have indexed the entire corpus in a lucene index which contains: document full text, document title, Semantic concepts form the document abstract. The application MedTagger, formerly cTAKES, is used to extract the semantic concepts of specific features from the Unified Medical Language Systems (UMLS). Our process includes: a. Manual: based on templates instantiated with UMLS concepts provided by a physician - A Physician expert produced diagnosis and specific pairings for each Case Topic. The concepts were extracted and mapped to UMLS using MedTagger. b. MedTagger for concept extraction both from case summaries and titles/abstracts - MedTagger (formerly known as cTAKES) is an open source Natural Language Processing tool which maps free text to UMLS concepts. c. Lucene for document indexing - Lucene is an open source tool for information retrieval maintained by Apache. We are using version 5.1.0 to index the features which we derive from the documents and to search the index with features derived from the Case Topics. d. Google for diagnoses - A python script uses the summary of the topics to start a Google Custom Search Engine (CSE) request and collecting all titles. The search is limited to wikimedz.com, nlm.nih.gov/medlineplus, MayoClinic.org, WebMD.com and Wikipedia.org. In a second step MedTagger is identifying concepts within the titles to reduce noise. e. Clinical Queries filter - Clinical Queries are a sub-collection of PubMed specific to a clinical topic such as Diagnosis, and which meet specific methodological criteria. These articles are those that are returned in response to a hand-curated PubMed search. HMKTB is a semantic hierarchical iterative approach using heuristically derived filters for diagnosis and symptoms. The algorithm uses a nested logic to prioritize different search strategies and tuning of filtering. The searches are enhanced by using a boolean combination of the query from the hierarchy, a keyword search, a title search and a term based on the topic type. Two cycles of searching are performed till at least a thousand search results are found. The inner cycle begins with using the semantic concepts of the google feature created diagnosis for the topics 1 to 10 and the given diagnosis (concept feature for the topics 11 to 30 to search the titles of all documents. Next in the hierarchy, the concepts found in the google search feature/given diagnosis are used to search the semantic concepts found in the document abstract. This is followed by a feature containing a filtered subset of diagnosis concepts found in the topic summary. The next level of the hierarchy contains symptom concepts from the topic summary. The results of each traversal of the hierarchy are filtered to enrich for clinically relevant documents. Three passes through the hierarchy are made. In pass 1, the documents must be in the set of documents meeting the NCBI Clinical Query for the topic type. For example, diagnosis topics will only return documents meeting the diagnosis Clinical Query. The second pass uses a less stringent filter of documents meeting the Clinical query for Treatment or Diagnosis. The third pass removes the Clinical Queries filter to allow for maximum retrieval.

nuuuuncMDRUB¶

Run ID: nuuuuncMDRUB
Participant: NU_UU_UNC
Track: Clinical Decision Support
Year: 2015
Submission: 7/30/2015
Type: manual
Task: b
MD5: bdef163afdbb3357ec46e556580b5f9e
Run description: We have indexed the entire corpus in a lucene index which contains: document full text, document title, Semantic concepts form the document abstract. The application MedTagger, formerly cTAKES, is used to extract the semantic concepts of specific features from the Unified Medical Language Systems (UMLS). Our process includes: a. Manual: based on templates instantiated with UMLS concepts provided by a physician - A Physician expert produced diagnosis and specific pairings for each Case Topic. The concepts were extracted and mapped to UMLS using MedTagger. b. MedTagger for concept extraction both from case summaries and titles/abstracts - MedTagger (formerly known as cTAKES) is an open source Natural Language Processing tool which maps free text to UMLS concepts. c. Lucene for document indexing - Lucene is an open source tool for information retrieval maintained by Apache. We are using version 5.1.0 to index the features which we derive from the documents and to search the index with features derived from the Case Topics. d. Google for diagnoses - A python script uses the summary of the topics to start a Google Custom Search Engine (CSE) request and collecting all titles. The search is limited to wikimedz.com, nlm.nih.gov/medlineplus, MayoClinic.org, WebMD.com and Wikipedia.org. In a second step MedTagger is identifying concepts within the titles to reduce noise. e. Clinical Queries filter - Clinical Queries are a sub-collection of PubMed specific to a clinical topic such as Diagnosis, and which meet specific methodological criteria. These articles are those that are returned in response to a hand-curated PubMed search. The topics 1 to 10 will be generated according to a predefined template,[disease] + [type] + [any specific patient population constraints: age(child, adolescent, adult),], a physician has created manual queries based on topic and task. The query is parsed into MedTagger concepts then used to search our Lucene index. The cases 11 to 30 are using the given diagnosis and their concepts instead of the manual feature.

nuuuuncMDRUN¶

Run ID: nuuuuncMDRUN
Participant: NU_UU_UNC
Track: Clinical Decision Support
Year: 2015
Submission: 7/29/2015
Type: manual
Task: a
MD5: 064b0e78d5d74a72d3af9ca477ca0509
Run description: We have indexed the entire corpus in a lucene index which contains: document full text, document title, Semantic concepts form the document abstract. The application MedTagger, formerly cTAKES, is used to extract the semantic concepts of specific features from the Unified Medical Language Systems (UMLS). Our process includes: a. Manual: based on templates instantiated with UMLS concepts provided by a physician - A Physician expert produced diagnosis and specific pairings for each Case Topic. The concepts were extracted and mapped to UMLS using MedTagger. b. MedTagger for concept extraction both from case summaries and titles/abstracts - MedTagger (formerly known as cTAKES) is an open source Natural Language Processing tool which maps free text to UMLS concepts. c. Lucene for document indexing - Lucene is an open source tool for information retrieval maintained by Apache. We are using version 5.1.0 to index the features which we derive from the documents and to search the index with features derived from the Case Topics. d. Google for diagnoses - A python script uses the summary of the topics to start a Google Custom Search Engine (CSE) request and collecting all titles. The search is limited to wikimedz.com, nlm.nih.gov/medlineplus, MayoClinic.org, WebMD.com and Wikipedia.org. In a second step MedTagger is identifying concepts within the titles to reduce noise. e. Clinical Queries filter - Clinical Queries are a sub-collection of PubMed specific to a clinical topic such as Diagnosis, and which meet specific methodological criteria. These articles are those that are returned in response to a hand-curated PubMed search. Completely Manual - MDRUN According to a predefined template,[disease] + [type] + [any specific patient population constraints: age(child, adolescent, adult),], a physician has created manual queries based on topic and task. The query is parsed into MedTagger concepts then used to search our Lucene index.

PL2c10¶

Run ID: PL2c10
Participant: PKUICST
Track: Clinical Decision Support
Year: 2015
Submission: 7/28/2015
Type: automatic
Task: a
MD5: 2137171974b2beed75d2f791ee034780
Run description: Using NlmMESHhttp://www.nlm.nih.gov/pubs/factsheets/mesh.htmland TextTiling Methodhttp://www.aclweb.org/anthology/J97-1003
Code: https://github.com/Michealzzw/Trec2015PKUICST.git

PL2c28¶

Run ID: PL2c28
Participant: PKUICST
Track: Clinical Decision Support
Year: 2015
Submission: 7/28/2015
Type: automatic
Task: a
MD5: 2e1be2fbfcfd6a53c814351c2519f290
Run description: Using NlmMESHhttp://www.nlm.nih.gov/pubs/factsheets/mesh.htmland TextTiling Methodhttp://www.aclweb.org/anthology/J97-1003
Code: https://github.com/Michealzzw/Trec2015PKUICST.git

PL2c6¶

Run ID: PL2c6
Participant: PKUICST
Track: Clinical Decision Support
Year: 2015
Submission: 7/28/2015
Type: automatic
Task: a
MD5: c8ae8972ae8976fad163d5e9951f88b8
Run description: Using NlmMESHhttp://www.nlm.nih.gov/pubs/factsheets/mesh.htmland TextTiling Methodhttp://www.aclweb.org/anthology/J97-1003
Code: https://github.com/Michealzzw/Trec2015PKUICST.git

PPR¶

Run ID: PPR
Participant: CBIA_VT
Track: Clinical Decision Support
Year: 2015
Submission: 7/28/2015
Type: automatic
Task: a
MD5: b42aec487ef5683ec499eabe487e3c7c
Run description: Personalized PageRank of UMLS graph, UMLS concepts directly expanded, original summary, relevance feedback, BM25

PPRdiag¶

Run ID: PPRdiag
Participant: CBIA_VT
Track: Clinical Decision Support
Year: 2015
Submission: 7/29/2015
Type: automatic
Task: b
MD5: 18a800502435b5b8abe44bf16b633683
Run description: Personalized PageRank of UMLS graph, UMLS concepts directly expanded, original summary and diagnosis, relevance feedback, BM25

prna1¶

Run ID: prna1
Participant: prna
Track: Clinical Decision Support
Year: 2015
Submission: 7/29/2015
Type: automatic
Task: a
MD5: 993c2449d65637ee1129e69602543293
Run description: We used an automated multiple-steps driven method to extract relevant biomedical articles corresponding to each given topic. We performed clinical concepts extraction with ontology mapping for identifying important IDF-weighted topical keywords from the given topic descriptions, which were used to extract relevant diagnoses concepts from Wikipedia clinical medicine category articles. Ultimately, the Wiki concepts related to relevant diagnoses, tests, and treatments were used in mapping pertinent biomedical articles, which were further filtered by named entity information, and ordered by publication date and importance in relation to the extracted Wiki keywords.

prna2¶

Run ID: prna2
Participant: prna
Track: Clinical Decision Support
Year: 2015
Submission: 7/29/2015
Type: automatic
Task: a
MD5: 361ca933720007c909b09cc21db81c29
Run description: We used an automated multiple-steps driven method to extract relevant biomedical articles corresponding to each given topic. We performed clinical concepts extraction with ontology mapping for identifying important IDF-weighted topical keywords from the given topic summaries, which were used to extract relevant diagnoses concepts from Wikipedia clinical medicine category articles. Ultimately, the Wiki concepts related to relevant diagnoses, tests, and treatments were used in mapping pertinent biomedical articles, which were further filtered by named entity information, and ordered by publication date and importance in relation to the extracted Wiki keywords.

prna3¶

Run ID: prna3
Participant: prna
Track: Clinical Decision Support
Year: 2015
Submission: 7/29/2015
Type: automatic
Task: a
MD5: 8f9c3121180dd3606184f633a785701d
Run description: We used an automated multiple-steps driven method to extract relevant biomedical articles corresponding to each given topic. We performed clinical concepts extraction with ontology mapping for identifying important topical keywords from the given topic descriptions, which were then expanded using deep learning-based word/phrase embeddings and used to extract relevant diagnoses concepts from Wikipedia clinical medicine category articles. Ultimately, the Wiki concepts related to relevant diagnoses, tests, and treatments were used in mapping pertinent biomedical articles, which were further filtered by named entity information, and ordered by publication date and importance in relation to the extracted Wiki keywords.

prnaB1¶

Run ID: prnaB1
Participant: prna
Track: Clinical Decision Support
Year: 2015
Submission: 7/30/2015
Type: automatic
Task: b
MD5: d4b0bcb34228e9fa89474eba1b17c037
Run description: We used an automated multiple-steps driven method to extract relevant biomedical articles corresponding to each given topic. We performed clinical concepts extraction with ontology mapping for identifying important IDF-weighted topical keywords from the given topic descriptions, which were used to extract relevant diagnoses concepts from Wikipedia clinical medicine category articles. Ultimately, the given ground truth diagnoses and the Wiki concepts related to relevant diagnoses, tests, and treatments were used in mapping pertinent biomedical articles, which were further filtered by named entity information, and ordered by publication date and importance in relation to the extracted Wiki keywords.

prnaB2¶

Run ID: prnaB2
Participant: prna
Track: Clinical Decision Support
Year: 2015
Submission: 7/30/2015
Type: automatic
Task: b
MD5: cf2f9413795a82159589ca511803bb64
Run description: We used an automated multiple-steps driven method to extract relevant biomedical articles corresponding to each given topic. We performed clinical concepts extraction with ontology mapping for identifying important IDF-weighted topical keywords from the given topic summaries, which were used to extract relevant diagnoses concepts from Wikipedia clinical medicine category articles. Ultimately, the given ground truth diagnoses and the Wiki concepts related to relevant diagnoses, tests, and treatments were used in mapping pertinent biomedical articles, which were further filtered by named entity information, and ordered by publication date and importance in relation to the extracted Wiki keywords.

prnaB3¶

Run ID: prnaB3
Participant: prna
Track: Clinical Decision Support
Year: 2015
Submission: 7/30/2015
Type: automatic
Task: b
MD5: cb5177ae522a562a862868a71ec7cfd6
Run description: We used an automated multiple-steps driven method to extract relevant biomedical articles corresponding to each given topic. We performed clinical concepts extraction with ontology mapping for identifying important topical keywords from the given topic descriptions, which were then expanded using deep learning-based word/phrase embeddings and used to extract relevant diagnoses concepts from Wikipedia clinical medicine category articles. Ultimately, the given ground truth diagnoses and the Wiki concepts related to relevant diagnoses, tests, and treatments were used in mapping pertinent biomedical articles, which were further filtered by named entity information, and ordered by publication date and importance in relation to the extracted Wiki keywords.

QFB¶

Run ID: QFB
Participant: CBIA_VT
Track: Clinical Decision Support
Year: 2015
Submission: 7/28/2015
Type: automatic
Task: a
MD5: 561c4952ad7eb003e3d314732c65e6b4
Run description: original summary, relevant feedback, BM25

QFBdiag¶

Run ID: QFBdiag
Participant: CBIA_VT
Track: Clinical Decision Support
Year: 2015
Submission: 7/29/2015
Type: automatic
Task: b
MD5: e147675e6489dbd4053f2f32795cd586
Run description: original summary and diagnosis, relevance feedback, BM25

RRFfused¶

Run ID: RRFfused
Participant: SCIAITeam
Track: Clinical Decision Support
Year: 2015
Submission: 7/29/2015
Type: manual
Task: b
MD5: 40ae01dc28f0adc8699ea8fb8a6e5b73
Run description: The diagnosis field was manually added to the topics. Reciprocal Ranked Fusion was used to fuse the baseline lists generated for Tasks A and B.

RRFFused¶

Run ID: RRFFused
Participant: SCIAITeam
Track: Clinical Decision Support
Year: 2015
Submission: 7/24/2015
Type: automatic
Task: a
MD5: 64e292319a5e9127e8ee084657fbe30e
Run description: Reciprocal Ranked Fusion was used to fuse four lists generated from Lucene baselines and framed outputs.

run1¶

Run ID: run1
Participant: DBNET_AUEB
Track: Clinical Decision Support
Year: 2015
Submission: 7/28/2015
Type: automatic
Task: a
MD5: 9f985bd56a8978784a05dca56f15d6e4
Run description: For the text extraction we used BeautifulSoup(http://www.crummy.com/software/BeautifulSoup/). For the preprocessing we used Porter Stemmer and Bigram. Furthermore we expanded the queries by using Bigram and Metamap(http://metamap.nlm.nih.gov/). For the relevance evaluation we used Pagerank.

Run1DBpSimp¶

Run ID: Run1DBpSimp
Participant: LIST_LUX
Track: Clinical Decision Support
Year: 2015
Submission: 7/28/2015
Type: automatic
Task: a
MD5: 61ed844eff100812c04fb9e54466d523
Run description: IR model: TF_IDF. Query expansion using 30 expanded terms within top 20 documents. Semantic annotation of queries using DBpedia. Query simplification by deleting information about patients.

run2¶

Run ID: run2
Participant: DBNET_AUEB
Track: Clinical Decision Support
Year: 2015
Submission: 7/28/2015
Type: automatic
Task: a
MD5: f2ed9207375b64ed500a7a61c582f57f
Run description: For the text extraction we used BeautifulSoup(http://www.crummy.com/software/BeautifulSoup/). For the preprocessing we used Porter Stemmer and Bigram. Furthermore we expanded the queries by using Bigram and Metamap(http://metamap.nlm.nih.gov/). For the relevance evaluation we used TFIDF..

Run2DBpComb¶

Run ID: Run2DBpComb
Participant: LIST_LUX
Track: Clinical Decision Support
Year: 2015
Submission: 7/28/2015
Type: automatic
Task: a
MD5: c378276989101fe16641d61b8c180d2f
Run description: Combination of 2 IR models: Hiemstra LM and LGD. Query expansion using 30 expanded terms within top 20 documents. Semantic annotation of queries using DBpedia.

Run4HLM¶

Run ID: Run4HLM
Participant: LIST_LUX
Track: Clinical Decision Support
Year: 2015
Submission: 7/28/2015
Type: automatic
Task: a
MD5: 404fcb99781c6046fde31f407640187c
Run description: IR model: Hiemstra LM. Query expansion using 30 expanded terms within top 20 documents.

Run5DBpAbs¶

Run ID: Run5DBpAbs
Participant: LIST_LUX
Track: Clinical Decision Support
Year: 2015
Submission: 7/30/2015
Type: automatic
Task: b
MD5: 038c488ec65ebadab3411d8bf6bf1809
Run description: Run 5: Indexing only the titles and the abstracts. IR model: TF_IDF. Query expansion using 30 expanded terms within top 20 documents. Semantic annotation of queries using DBpedia. Query simplification by deleting information about patients.

runindri¶

Run ID: runindri
Participant: HITSJ
Track: Clinical Decision Support
Year: 2015
Submission: 7/27/2015
Type: automatic
Task: a
MD5: 78421f1df9368e745cb2a4429042846b
Run description: This is a auto result.We use "metamap" to extract concept of UMLS from the topics, and then extend the concepts in the UMLS corpus.Find the synonyms of the concepts. Combine the expansion of the type, we get the query. The search engine is Indri and resule no re-ranking.
Code: https://github.com/hitwilab/TREC2015-submission.git

runindriB¶

Run ID: runindriB
Participant: HITSJ
Track: Clinical Decision Support
Year: 2015
Submission: 7/29/2015
Type: automatic
Task: b
MD5: 2141a9ab633c6d8932eac7beb73c768e
Run description: This is a auto result.We use "metamap" to extract concepts of UMLS from the topics2015B.xml, and then extend the concepts in the UMLS corpus.Find the synonyms of the concepts. Combine the expansion of the type and diagnosis field,we get the query. The search engine is Indri and result no re-ranking.
Code: https://github.com/hitwilab/TREC2015-submission.git

runindriML¶

Run ID: runindriML
Participant: HITSJ
Track: Clinical Decision Support
Year: 2015
Submission: 7/28/2015
Type: automatic
Task: a
MD5: bced25e5b13694127e31cb234522eeb6
Run description: This is a auto result.We use "metamap" to extract concept of UMLS from the topics, and then extend the concepts in the UMLS corpus.Find the synonyms of the concepts. Combine the expansion of the type, we get the query. The search engine is Indri and result re-ranking with method of machine learning of SVM.
Code: https://github.com/hitwilab/TREC2015-submission.git

runnetwork¶

Run ID: runnetwork
Participant: HITSJ
Track: Clinical Decision Support
Year: 2015
Submission: 7/30/2015
Type: automatic
Task: b
MD5: 6184e59c4ee4216a79e69f7dfae41606
Run description: his is a auto result. We use "metamap" to extract concepts of UMLS from the topics2015B.xml, and then extend the concepts in the UMLS corpus, find the synonyms of the concepts. Combine the expansion of the type and diagnosis field, we get the query. The search engine is Indri and we get the initial results.We use networks to find the potential documents which would be related to the results. Thus we get the submission.
Code: https://github.com/hitwilab/TREC2015-submission.git

SCIAILuceneA¶

Run ID: SCIAILuceneA
Participant: SCIAITeam
Track: Clinical Decision Support
Year: 2015
Submission: 7/14/2015
Type: automatic
Task: a
MD5: 5d73b6894d022237a28c280135f23b52
Run description: Prior to the run, Lucene was used to index the corpus of pub med docs. Topic summaries were run as queries through Lucene, and the top 20 documents were returned for each topic.

SCIAILuceneB¶

Run ID: SCIAILuceneB
Participant: SCIAITeam
Track: Clinical Decision Support
Year: 2015
Submission: 7/29/2015
Type: manual
Task: b
MD5: 97c76444fc270c90620bf89bc8631d34
Run description: The diagnosis field was manually added to the topics. The diagnosis and summary were run as queries through Lucene which returned a ranked list of the top 20 documents for each topic.

SH1¶

Run ID: SH1
Participant: Sortinghat
Track: Clinical Decision Support
Year: 2015
Submission: 7/26/2015
Type: automatic
Task: a
MD5: ed28943bf6d4c396ffd9fd4702817286
Run description: We will provide all info on github page.
Code: https://github.com/wsliiitb/trec2015.git

SHB1¶

Run ID: SHB1
Participant: Sortinghat
Track: Clinical Decision Support
Year: 2015
Submission: 7/30/2015
Type: automatic
Task: b
MD5: 27e21bcd2339644d817343200a48e205
Run description: please look into github provided above for all description.
Code: https://github.com/wsliiitb/trec2015.git

SIBTEX2CITIN¶

Run ID: SIBTEX2CITIN
Participant: SIBtex
Track: Clinical Decision Support
Year: 2015
Submission: 7/28/2015
Type: automatic
Task: a
MD5: d0fc4b51d4f4451e83fab40c6c90bbde
Run description: run 2 : citations in. From run 1 (baseline), reranking was performed according to citations in. We first computed all available citations in PubMed Central. Then, each paper's RSV had a boosting according to the number of citations in.

SIBTEX3CTOUT¶

Run ID: SIBTEX3CTOUT
Participant: SIBtex
Track: Clinical Decision Support
Year: 2015
Submission: 7/28/2015
Type: automatic
Task: a
MD5: 4aa4b5e8b248529ec3bf2173d960e5a1
Run description: run 3 : citations out. From run 1 (baseline), reranking was performed according to citations out. For each paper in the ranking, a small part of its RSV was given to its citations out.

SIBTEX5COMBO¶

Run ID: SIBTEX5COMBO
Participant: SIBtex
Track: Clinical Decision Support
Year: 2015
Submission: 7/28/2015
Type: automatic
Task: a
MD5: 7b928f461d47a082ff9b6157001c758e
Run description: run 5 : combo. A linear combo was done between run 2, and (SIBTEX2 runs BtPf32.res and BtCleanAll4.res)

SNUMedinfo1¶

Run ID: SNUMedinfo1
Participant: SNUMedinfo
Track: Clinical Decision Support
Year: 2015
Submission: 7/29/2015
Type: automatic
Task: a
MD5: 84cdb1cb2f699e4d524869afd7e96091
Run description: External tagged knowledge based query expansion (MEDLINE) Task-specific classifier Borda-fuse

SNUMedinfo11¶

Run ID: SNUMedinfo11
Participant: SNUMedinfo
Track: Clinical Decision Support
Year: 2015
Submission: 7/29/2015
Type: automatic
Task: b
MD5: 0443be6c478dc0536bf48259b9ee77e1
Run description: External tagged knowledge based query expansion Quality ranking Borda-fuse

SNUMedinfo12¶

Run ID: SNUMedinfo12
Participant: SNUMedinfo
Track: Clinical Decision Support
Year: 2015
Submission: 7/29/2015
Type: automatic
Task: b
MD5: 63fed7b04129d29c7243a9d629217aa0
Run description: External tagged knowledge based query expansion Quality ranking Borda-fuse

SNUMedinfo13¶

Run ID: SNUMedinfo13
Participant: SNUMedinfo
Track: Clinical Decision Support
Year: 2015
Submission: 7/29/2015
Type: automatic
Task: b
MD5: e5ce3de2aa18db7397145cbc8a6aec13
Run description: External tagged knowledge based query expansion Quality ranking Borda-fuse

SNUMedinfo2¶

Run ID: SNUMedinfo2
Participant: SNUMedinfo
Track: Clinical Decision Support
Year: 2015
Submission: 7/29/2015
Type: automatic
Task: a
MD5: 8a503428a579f2ff763d5aaf26b21e0a
Run description: External tagged knowledge based query expansion (MEDLINE) Task-specific classifier Borda-fuse

SNUMedinfo3¶

Run ID: SNUMedinfo3
Participant: SNUMedinfo
Track: Clinical Decision Support
Year: 2015
Submission: 7/29/2015
Type: automatic
Task: a
MD5: 36a34abea257f7e2b6fcf32d89ddf613
Run description: External tagged knowledge based query expansion (MEDLINE) Task-specific classifier Borda-fuse

TUW1¶

Run ID: TUW1
Participant: TUW
Track: Clinical Decision Support
Year: 2015
Submission: 7/28/2015
Type: automatic
Task: a
MD5: c416248faad810946fffed0597744c0d
Run description: TUW1: baseline terrier implementation of BM25(k1=2.3, b=0.55) + query expansion (bo1, 3 docs, 10 terms) External resource: None

TUW2¶

Run ID: TUW2
Participant: TUW
Track: Clinical Decision Support
Year: 2015
Submission: 7/28/2015
Type: automatic
Task: a
MD5: 71252052dca2cfbe0d6aff33265ac440
Run description: TUW2: same as run 1, but I used Metamap to expand initial query External Resouces: Metamap

TUW3¶

Run ID: TUW3
Participant: TUW
Track: Clinical Decision Support
Year: 2015
Submission: 7/28/2015
Type: automatic
Task: a
MD5: e21bdfef9863495655d79f27da1ae53d
Run description: TUW3: same as run 2, but I assigned different weights to the expanded Metamap terms External Resouces: Metamap

TUW4¶

Run ID: TUW4
Participant: TUW
Track: Clinical Decision Support
Year: 2015
Submission: 7/30/2015
Type: automatic
Task: b
MD5: 8e1a18d4b5cdac36fa9744daf045211c
Run description: This run has the same configuration of run1, but weigthing 6 to the terms in the diagnosis field
Code: https://github.com/joaopalotti/tuw_at_trec_cds_2015

TUW5¶

Run ID: TUW5
Participant: TUW
Track: Clinical Decision Support
Year: 2015
Submission: 7/30/2015
Type: automatic
Task: b
MD5: fc74c9e2a7c7f111a2c99cd18ef3a8de
Run description: This run has the same configuration of run2, but weigthing 6 to the terms in the diagnosis field and +2 for the metamap expansions. External resource: Metamap
Code: https://github.com/joaopalotti/tuw_at_trec_cds_2015

TUW6¶

Run ID: TUW6
Participant: TUW
Track: Clinical Decision Support
Year: 2015
Submission: 7/30/2015
Type: automatic
Task: b
MD5: 477357077f8a9895e83c8f424e22ed17
Run description: This run has the same configuration of run3, but weigthing 6 to the terms in the diagnosis field and +3 for the metamap expansions. External resource: Metamap
Code: https://github.com/joaopalotti/tuw_at_trec_cds_2015

udelArun1¶

Run ID: udelArun1
Participant: udel
Track: Clinical Decision Support
Year: 2015
Submission: 7/28/2015
Type: automatic
Task: a
MD5: 89d04f33c524bcfe6d62b6d992117e46
Run description: udelArun1 is built on the top of Terrier IR framework and uses one DFR model, which is In_expB2, for retrieval. In addition, it adopts pseudo relevance feedback that is based on KL model for performing query expansion.

udelArun2¶

Run ID: udelArun2
Participant: udel
Track: Clinical Decision Support
Year: 2015
Submission: 7/28/2015
Type: automatic
Task: a
MD5: 02f50cc81ed3880f6bec73b83a079094
Run description: udelArun2 is built on the top Terrier IR platform and uses one DFR model, In_expB2, for retrieval. This run also uses query expansion mechanism that is based on pseudo relevance feedback. In addition, the feedback received from MetaMap is being used for performing query refinement.

udelArun3¶

Run ID: udelArun3
Participant: udel
Track: Clinical Decision Support
Year: 2015
Submission: 7/28/2015
Type: automatic
Task: a
MD5: 1d647deb0129fbfd387cc066dd2b0e5a
Run description: udelArun3 is built on the top Terrier IR platform and uses one DFR model, In_expB2, for retrieval. This run uses query expansion mechanism that is based on pseudo relevance feedback. In addition, the feedback received from MetaMap is being used for performing query refinement. Finally, we have added a feature for document classification based on the three query types. This feature is used to indicate the relevancy of the retrieved documents to the querying tasks that are being performed by the users.

udelBrun1¶

Run ID: udelBrun1
Participant: udel
Track: Clinical Decision Support
Year: 2015
Submission: 7/30/2015
Type: automatic
Task: b
MD5: a9bd59986f77f34e8fbdf670a1abc162
Run description: udelBrun1 is built on the top of Terrier IR framework and uses one DFR model, which is In_expB2, for retrieval. In addition, it uses pseudo relevance feedback that is based on KL model in order to perform query expansion. This run is equivalent to the previously submitted udelArun1; however, the latter one uses the diagnosis field as an extra information.

udelBrun2¶

Run ID: udelBrun2
Participant: udel
Track: Clinical Decision Support
Year: 2015
Submission: 7/30/2015
Type: automatic
Task: b
MD5: 086d93ef1eaef9967722a481c2240f2f
Run description: udelBrun2 is built on the top of Terrier IR framework and uses one DFR model, which is In_expB2, for retrieval. In addition, it adopts pseudo relevance feedback that is based on KL model in order to perform query expansion. This run uses the diagnosis field as an extra information to be added to each one of the queries. Furthermore, Medical Subject Headings (MeSH) have been used to expand each one of the diagnosis with more relevant terms.

udelBrun3¶

Run ID: udelBrun3
Participant: udel
Track: Clinical Decision Support
Year: 2015
Submission: 7/30/2015
Type: automatic
Task: b
MD5: f026cd064f6833b122a3b39dc284e6ac
Run description: udelBrun3 is built on the top Terrier IR platform and uses one DFR model, In_expB2, for retrieval. This run uses query expansion mechanism that is based on pseudo relevance feedback. In addition, we have added a feature for document classification based on the three query types. This feature is used to indicate the relevance of the retrieved documents to the querying tasks that are being performed by the users. Finally, this run uses the diagnosis field as an extra information to be added to queries with "test" and "treatment" types.

UMLS¶

Run ID: UMLS
Participant: CBIA_VT
Track: Clinical Decision Support
Year: 2015
Submission: 7/28/2015
Type: automatic
Task: a
MD5: 17fa638c79c48fae1567446a494cd0fe
Run description: UMLS concepts directly expanded, original summary, relevance feedback, BM25

UMLSdiag¶

Run ID: UMLSdiag
Participant: CBIA_VT
Track: Clinical Decision Support
Year: 2015
Submission: 7/29/2015
Type: automatic
Task: b
MD5: 92514f124b647a8b9bd9ef0d06a770fb
Run description: UMLS concepts directly expanded, original summary and diagnosis, relevance feedback, BM25

utdhltrikcv¶

Run ID: utdhltrikcv
Participant: UTDHLTRI
Track: Clinical Decision Support
Year: 2015
Submission: 7/29/2015
Type: automatic
Task: a
MD5: 0203b36e7fd3a61a9f89e4354742e16b
Run description: Cross-validated statistical weights

utdhltrikcvb¶

Run ID: utdhltrikcvb
Participant: UTDHLTRI
Track: Clinical Decision Support
Year: 2015
Submission: 7/30/2015
Type: automatic
Task: b
MD5: f0f4ec592d2aeb6ffc686f688aa65040
Run description: Cross-validated statistical weights using diagnosis as keyword

utdhltril2r¶

Run ID: utdhltril2r
Participant: UTDHLTRI
Track: Clinical Decision Support
Year: 2015
Submission: 7/28/2015
Type: automatic
Task: a
MD5: 16ca3447acc62145669ae5ae10268af1
Run description: Basic learning-to-rank using gradient ascent trained on the 2014 judgments.

utdhltripar¶

Run ID: utdhltripar
Participant: UTDHLTRI
Track: Clinical Decision Support
Year: 2015
Submission: 7/29/2015
Type: automatic
Task: a
MD5: 427740746b49e8e296ea86ae24bea79e
Run description: L2R using statistical relevance scores over all sections in each document.

utdhltriprfb¶

Run ID: utdhltriprfb
Participant: UTDHLTRI
Track: Clinical Decision Support
Year: 2015
Submission: 7/30/2015
Type: automatic
Task: b
MD5: bdcc78347c68a3ddb32fdbd9717b10d4
Run description: utdhltrikcvb with pseudo-relevance feedback

utdhltrisprf¶

Run ID: utdhltrisprf
Participant: UTDHLTRI
Track: Clinical Decision Support
Year: 2015
Submission: 7/30/2015
Type: automatic
Task: b
MD5: 1ccb4ad474fff0401b00fc5a643aecf1
Run description: utdhltrikcvb with semantic pseudo-relevance feedback using the expected answer type

UWCPL2¶

Run ID: UWCPL2
Participant: WaterlooClarke
Track: Clinical Decision Support
Year: 2015
Submission: 7/28/2015
Type: automatic
Task: a
MD5: 7cbaec94fd23b34a58d3fed6acdac994
Run description: This run uses the Terrier search engine and expands the query using Rocchio's algorithm. Additionally, PL2 is the retrieval function used with tuning the c parameter to be equal to 2.75.

UWCSolrBM25¶

Run ID: UWCSolrBM25
Participant: WaterlooClarke
Track: Clinical Decision Support
Year: 2015
Submission: 7/27/2015
Type: automatic
Task: a
MD5: 658858e0bebd5ad7806473e2f88bf46c
Run description: The run was generated using Solr search engine and BM25 method

UWCSolrTerr¶

Run ID: UWCSolrTerr
Participant: WaterlooClarke
Track: Clinical Decision Support
Year: 2015
Submission: 7/28/2015
Type: automatic
Task: a
MD5: c78dd80708e2bd01b3f7a84d06fbf23a
Run description: PL2 retrieval function was used for Terrier search engine and Rocchio's Algorithm was used as a query expansion technique. In Solr search engine, the original query was expanded, where the query is first filtered to keep medical terms and then expanded with one synonym for each medical term. BM25 was used as the retrieval function for Solr search engine. Reciprocal rank fusion score was finally used to fuse the results of Terrier and Solr by producing a new combined ranked list.

UWMUO1¶

Run ID: UWMUO1
Participant: UWM_UO
Track: Clinical Decision Support
Year: 2015
Submission: 7/26/2015
Type: automatic
Task: a
MD5: 0e019d950915bac5305622117cec2548
Run description: Bayesian smoothing with Dirichlet Prior had been set up as default for the retrieval in Terrier. PorterStemmer has been adopted.

UWMUO2¶

Run ID: UWMUO2
Participant: UWM_UO
Track: Clinical Decision Support
Year: 2015
Submission: 7/26/2015
Type: automatic
Task: a
MD5: 3f938ada7643a059dd58a016d9ef2d10
Run description: Query Expansion (QE) with MeSH Keywords The summary + QE (weighted keywords including MeSH keywords). Keywords included in at least two different documents based on retrieved top 20 results had been extracted for query expansion. The keywords including MeSH term had been selected by filtering the keywords using MeSH terms based on terms included in MH field (MESH Heading) of the 2015 MeSH descriptor file from NIH (National Library of Medicine - http://www.nlm.nih.gov/mesh/filelist.html).

UWMUO3¶

Run ID: UWMUO3
Participant: UWM_UO
Track: Clinical Decision Support
Year: 2015
Submission: 7/26/2015
Type: automatic
Task: a
MD5: 29ba80ff4bbbfeee1a7170fe29830c9e
Run description: Keywords included in at least two different documents based on retrieved top 20 results had been extracted for query expansion. It has not been considered whether each document includes MeSH terms as keywords.

UWMUO4¶

Run ID: UWMUO4
Participant: UWM_UO
Track: Clinical Decision Support
Year: 2015
Submission: 7/26/2015
Type: manual
Task: b
MD5: 29f4aa21e9eaedb935054e0c09c93db3
Run description: The original queries consisting of summaries has been reformulated manually using the meaningful keywords existing in the summary.

UWMUO5¶

Run ID: UWMUO5
Participant: UWM_UO
Track: Clinical Decision Support
Year: 2015
Submission: 7/26/2015
Type: automatic
Task: b
MD5: 3c0a650e8b5370e9ec3b8091243ee4f2
Run description: QE with MeSH (Keywords & Title) The summary + QE (including MeSH in keywords & title). The results from UWMUO2 have been revised by considering the title as well. The MeSH terms occurred in at least two different titles have been added for QE, too.

UWMUO6¶

Run ID: UWMUO6
Participant: UWM_UO
Track: Clinical Decision Support
Year: 2015
Submission: 7/29/2015
Type: automatic
Task: b
MD5: 36e0f830745cc83f3edd539744c3c06c
Run description: Query Expansion (QE) with MeSH Keywords: The summary + diagnosis + QE (weighted keywords including MeSH keywords). Keywords included in at least two different documents based on retrieved top 20 results for (summary + diagnosis) had been extracted for query expansion. The keywords including MeSH term had been selected by filtering the keywords using MeSH terms based on terms included in MH field (MESH Heading) of the 2015 MeSH descriptor file from NIH (National Library of Medicine - http://www.nlm.nih.gov/mesh/filelist.html).

wsuirdaa¶

Run ID: wsuirdaa
Participant: wsu_ir
Track: Clinical Decision Support
Year: 2015
Submission: 7/29/2015
Type: automatic
Task: a
MD5: 0cb1d7013d9953970fc4b682e06d6a1f
Run description: Markov Random Field model is adopted to expand the queries with the first-level concepts. Indri toolkit is used to conduct this research and obtain the submitted runs.
Code: https://github.com/teanalab/MRF-L

wsuirdma¶

Run ID: wsuirdma
Participant: wsu_ir
Track: Clinical Decision Support
Year: 2015
Submission: 7/29/2015
Type: manual
Task: a
MD5: 66acf91234814f6dc743ace7e0ef644f
Run description: Markov Random Field model is adopted to expand the queries with the first-level concepts. Indri toolkit is used to conduct this research and obtain the submitted runs. main resources used: UMLS, Wikepedia
Code: https://github.com/teanalab/MRF-L

wsuirdmb¶

Run ID: wsuirdmb
Participant: wsu_ir
Track: Clinical Decision Support
Year: 2015
Submission: 7/31/2015
Type: manual
Task: b
MD5: 99e764df217b4bd2d96f4950407400c1
Run description: A model based on Markov Random Field is implemented to achieve the submitted top ranked document. Metamap API, Indri toolkit, UMLS ontology and UMLS semantic network are exploited to obtain these submitted runs.
Code: https://github.com/teanalab/MRF-L

wsuirsaa¶

Run ID: wsuirsaa
Participant: wsu_ir
Track: Clinical Decision Support
Year: 2015
Submission: 7/29/2015
Type: automatic
Task: a
MD5: 042d5f5945dfee922b2c8ecb4a9fe94d
Run description: Markov Random Field model is adopted to expand the queries with the first-level concepts. Indri toolkit is used to conduct this research and obtain the submitted runs.
Code: https://github.com/teanalab/MRF-L

wsuirsab¶

Run ID: wsuirsab
Participant: wsu_ir
Track: Clinical Decision Support
Year: 2015
Submission: 7/31/2015
Type: automatic
Task: b
MD5: 152ce9240759a3bedf371fdeb6f8505b
Run description: A model based on Markov Random Field is implemented in achieving top ranked document. Metamap API, Indri toolkit, UMLS ontology and UMLS semantic network are exploited to obtain the submitted runs.
Code: https://github.com/teanalab/MRF-L

wsuirsmb¶

Run ID: wsuirsmb
Participant: wsu_ir
Track: Clinical Decision Support
Year: 2015
Submission: 7/31/2015
Type: manual
Task: b
MD5: f005c3a605be026d71104508ec68280c
Run description: A model based on Markov Random Field is implemented to achieve the submitted top ranked document. Metamap API, Indri toolkit, UMLS ontology and UMLS semantic network are exploited to obtain these submitted runs.
Code: https://github.com/teanalab/MRF-L