Runs - Microblog 2012¶

AIrun1¶

Run ID: AIrun1
Participant: AI_ROMA3
Track: Microblog
Year: 2012
Submission: 7/10/2012
Type: automatic
Task: adhoc
MD5: 8794cdeb20af2067e1abc1185c3d1342
Run description: Wikipedia Miner for Entity Recognition and query expansion. Internal Dictionary of twitter words for query expansion. Twitter Feature like the presence of urls, number of retweet. Text Analysis for discovering the presence of noise in the tweet like alphanumeric word, repeated letters and emoticons.

aWekaModel¶

Run ID: aWekaModel
Participant: SCIAITeam
Track: Microblog
Year: 2012
Submission: 7/9/2012
Type: automatic
Task: adhoc
MD5: fad20e736ce84f205d4fe55b7a4b4c8d
Run description: Used a bayesNet learning model with weka. We used a list of emotes from wikipedia as part of our model data.

baseline¶

Run ID: baseline
Participant: SCIAITeam
Track: Microblog
Year: 2012
Submission: 7/9/2012
Type: automatic
Task: adhoc
MD5: b8a90bd622eb8d738d2b5d2c0d1963e5
Run description: Baseline Run using simple Lucene

basic¶

Run ID: basic
Participant: SCIAITeam
Track: Microblog
Year: 2012
Submission: 7/29/2012
Type: automatic
Task: filtering
MD5: c06989d60961e1c8a0185cb8174a001d
Run description: run using simple lucene

BAUdfi0f¶

Results | Participants | Input | Summary | Appendix

Run ID: BAUdfi0f
Participant: BAU
Track: Microblog
Year: 2012
Submission: 8/1/2012
Type: automatic
Task: filtering
MD5: 0b324f30761ea08d070408c82192c5a5
Run description: We don't use any external resources

BAUdfree¶

Run ID: BAUdfree
Participant: BAU
Track: Microblog
Year: 2012
Submission: 7/11/2012
Type: automatic
Task: adhoc
MD5: b6dffd231db1b22c76241539a1180053
Run description: We don't use any external resources.

BAUdfreef¶

Results | Participants | Input | Summary | Appendix

Run ID: BAUdfreef
Participant: BAU
Track: Microblog
Year: 2012
Submission: 8/1/2012
Type: automatic
Task: filtering
MD5: 193429a49b6cc75a92be03c7d338d1dc
Run description: We don't use any external resources

BAUdph¶

Run ID: BAUdph
Participant: BAU
Track: Microblog
Year: 2012
Submission: 7/11/2012
Type: automatic
Task: adhoc
MD5: 3dc71aadb0f452d7b977c5ccffaa5ee7
Run description: we don't use any external resources.

BAUdphf¶

Results | Participants | Input | Summary | Appendix

Run ID: BAUdphf
Participant: BAU
Track: Microblog
Year: 2012
Submission: 8/1/2012
Type: automatic
Task: filtering
MD5: 32a52000f9532236176aaeaede7e5a0e
Run description: We don't use any external resources

BAUjskls¶

Run ID: BAUjskls
Participant: BAU
Track: Microblog
Year: 2012
Submission: 7/11/2012
Type: automatic
Task: adhoc
MD5: d36f7edf28a237423a5d759d0581b642
Run description: we don't use any external resources.

BAUjsklsf¶

Results | Participants | Input | Summary | Appendix

Run ID: BAUjsklsf
Participant: BAU
Track: Microblog
Year: 2012
Submission: 8/1/2012
Type: automatic
Task: filtering
MD5: af0a5c15fc02fedb85a96137800ffc4f
Run description: We don't use any external resources

BAUtf¶

Run ID: BAUtf
Participant: BAU
Track: Microblog
Year: 2012
Submission: 7/11/2012
Type: automatic
Task: adhoc
MD5: 1d377ddd75cdf662b0583b4d3b975f3c
Run description: we don't use any external resources.

BL¶

Run ID: BL
Participant: QCRI
Track: Microblog
Year: 2012
Submission: 7/10/2012
Type: automatic
Task: adhoc
MD5: 68b8f7722eb2e3c3b9cb4ec456a8c4ca
Run description: This is our Baseline run. Queries and used directly to search the Tweets collection. The indexed tweets are those identified as English ones only using a language detector

BLFB¶

Run ID: BLFB
Participant: QCRI
Track: Microblog
Year: 2012
Submission: 7/10/2012
Type: automatic
Task: adhoc
MD5: 9635767ee7478ba9d209bb44252de829
Run description: The same as Baseline run + using pseudo relevance feedback

BM25¶

Run ID: BM25
Participant: qcri_twitsear
Track: Microblog
Year: 2012
Submission: 7/11/2012
Type: automatic
Task: adhoc
MD5: 6ab35a741e348ae09b2173cb6d084911
Run description: Using BM25 as the basic retrieval model.

BM25PRF¶

Run ID: BM25PRF
Participant: qcri_twitsear
Track: Microblog
Year: 2012
Submission: 7/11/2012
Type: automatic
Task: adhoc
MD5: c836e96edf38d5034532c4033f3f10ed
Run description: Using BM25 as the basic retrieval model. Then the pseudo relevance feedback is applied. No future and external evidence was used during the retrieval and feedback.

BM25TRF¶

Run ID: BM25TRF
Participant: qcri_twitsear
Track: Microblog
Year: 2012
Submission: 7/11/2012
Type: automatic
Task: adhoc
MD5: 5866cbc137a66cae43cca0c8e874f30a
Run description: Using BM25 as the basic retrieval model. Then the temporal based relevance feedback is applied. The bursty pattern of tweets are detected in the initial retrieval results, and then the tweets in the peaks are used for relevance feedback. No future and external evidence was used during the retrieval and feedback.

cmuPhrE¶

Run ID: cmuPhrE
Participant: CMU_Callan
Track: Microblog
Year: 2012
Submission: 7/6/2012
Type: automatic
Task: adhoc
MD5: 780fe7bba30e8583abe90cbc6bebed5a
Run description: Uses the Indri search engine. The query is structured with automatically extracted phrases. Stopwords are included in the query. The links in the tweets were crawled in late 2011 and document expansion was used to add terms from the web pages to the tweet it was linked from.

cmuPrfPhr¶

Run ID: cmuPrfPhr
Participant: CMU_Callan
Track: Microblog
Year: 2012
Submission: 7/6/2012
Type: automatic
Task: adhoc
MD5: 5d9e2a26745825e0777f8b1ecc8b2893
Run description: Uses the Indri search engine. Pseudo-relevance feedback query expansion was applied. The query is structured with automatically extracted phrases. Stopwords are included in both query expansion and the final query.

cmuPrfPhrE¶

Run ID: cmuPrfPhrE
Participant: CMU_Callan
Track: Microblog
Year: 2012
Submission: 7/6/2012
Type: automatic
Task: adhoc
MD5: 038c5802cccdf923e891b7ab83d8d8ba
Run description: Uses the Indri search engine. Pseudo-relevance feedback query expansion was applied. The query is structured with automatically extracted phrases. Stopwords are included in the query expansion terms and in the final query. Web pages linked from tweets were downloaded in late 2011 and document expansion was used to expand each tweet with terms from the web pages it linked to.

cmuPrfPhrENo¶

Run ID: cmuPrfPhrENo
Participant: CMU_Callan
Track: Microblog
Year: 2012
Submission: 7/6/2012
Type: automatic
Task: adhoc
MD5: 7638fbeb523d6f962f46749fa6daf2e3
Run description: Uses the Indri search engine. Pseudo-relevance feedback query expansion was applied. The query is structured with automatically extracted phrases. Stopwords are not included in the query expansion and but included in the final query. Web pages linked from tweets were downloaded in late 2011 and document expansion was used to expand each tweet with terms from the web pages it linked to.

csirolrhuq111¶

Run ID: csirolrhuq111
Participant: csiro
Track: Microblog
Year: 2012
Submission: 8/1/2012
Type: automatic
Task: filtering
MD5: 0ac042569df1cbdb8c4b563b211e5614
Run description: Uses hashtags, word unigrams and query text as features, and used logistic regression as the classifier. The score returned for each tweet is the probability of being relevant to a specific topic.

csiroNE112¶

Run ID: csiroNE112
Participant: csiro
Track: Microblog
Year: 2012
Submission: 7/10/2012
Type: automatic
Task: adhoc
MD5: 2c816101b0c3831b3a6cca9d1c3d99a7
Run description: Non-English tweets removed, short tweets and obvious retweets removed, remainder stopped with custom list. Some lexical normalisation. Named entities in queries are up-weighted.

csiroQEll112¶

Run ID: csiroQEll112
Participant: csiro
Track: Microblog
Year: 2012
Submission: 7/10/2012
Type: automatic
Task: adhoc
MD5: f3356eff38e4e6e608761a9e64220ff8
Run description: Non-English tweets removed, short tweets and obvious retweets removed, remainder stopped with custom list. Some lexical normalisation. Pseudo-relevance feedback expands queries with extra hashtags (only).

csiroQERF111¶

Run ID: csiroQERF111
Participant: csiro
Track: Microblog
Year: 2012
Submission: 8/1/2012
Type: automatic
Task: filtering
MD5: c5ea853472a6c3b36741483aaa872b1d
Run description: using Indri retrieves an initial ranked list then expands the queries using judgments of the top-5 tweets, then retrieves a set within the time limits. Only top-10 is considered "yes".

csiroQEt112¶

Run ID: csiroQEt112
Participant: csiro
Track: Microblog
Year: 2012
Submission: 7/10/2012
Type: automatic
Task: adhoc
MD5: fff1ddcbcdf62b334a8c23e46a1fca05
Run description: Non-English tweets removed, short tweets and obvious retweets removed, remainder stopped with custom list. Some lexical normalisation. Pseudo-relevance feedback to expand queries (no hashtags).

csiroR112¶

Run ID: csiroR112
Participant: csiro
Track: Microblog
Year: 2012
Submission: 7/11/2012
Type: automatic
Task: adhoc
MD5: 4f3e4c6747701d6a5a2c7dfba07bc151
Run description: Non-English tweets removed, short tweets and obvious retweets removed, remainder stopped with custom list. Some lexical normalisation. Vanilla query-likelihood ranking.

csiroshuq111¶

Run ID: csiroshuq111
Participant: csiro
Track: Microblog
Year: 2012
Submission: 8/1/2012
Type: automatic
Task: filtering
MD5: f8bd526391f6d577c167f729c1a12ee8
Run description: An initial ranked list was retrieved using indri, from which top-5 judgements were used as positive examples to train an SVM classifier. The probability of being classified to be relevant was returned as the score.

csiroSVMqe111¶

Run ID: csiroSVMqe111
Participant: csiro
Track: Microblog
Year: 2012
Submission: 8/1/2012
Type: automatic
Task: filtering
MD5: 115340a637239d66c40525908821c16e
Run description: An initial ranked list was retrieved using indri with pseudo relevance feedback from which top-5 judgements were used as positive examples to train an SVM classifier. The probability of being classified to be relevant was returned as the score.

expansion¶

Run ID: expansion
Participant: SCIAITeam
Track: Microblog
Year: 2012
Submission: 7/9/2012
Type: automatic
Task: adhoc
MD5: 3c9c3ec455ca0f252a9f897af43f3272
Run description: Internal Query Expansion. However we used a model for Part of Speech Tagging.

expansion2¶

Run ID: expansion2
Participant: SCIAITeam
Track: Microblog
Year: 2012
Submission: 7/29/2012
Type: automatic
Task: filtering
MD5: c60ab275983d4ce26bc97a9b463373c0
Run description: internal query expansion

expansionurl¶

Run ID: expansionurl
Participant: SCIAITeam
Track: Microblog
Year: 2012
Submission: 7/29/2012
Type: automatic
Task: filtering
MD5: 679dd9027a6e6d298461924beed19283
Run description: uses internal query expansion and urls to determine relevance

exttempws¶

Run ID: exttempws
Participant: UEdinburgh
Track: Microblog
Year: 2012
Submission: 7/10/2012
Type: automatic
Task: adhoc
MD5: b1cc79d04b46aacc1d331e13a1cf2ecb
Run description: Uses temporal distributions based on consecutive candidate events. Temporal evidence is used externally to topical relevance given by a retrieval system. Results are reranked according to the temporal features of the query during the time the tweet was emitted. All processing and related algorithms are stream-friendly - i.e. they can be supported on the fly, as tweets arrive. No query-expansion is used neither were external or future data.

exttempwsf¶

Run ID: exttempwsf
Participant: UEdinburgh
Track: Microblog
Year: 2012
Submission: 7/10/2012
Type: automatic
Task: adhoc
MD5: d9d19c094046176659b6b0bddaf5192f
Run description: Uses temporal distributions based on consecutive candidate events. Temporal evidence is used externally to topical relevance given by a retrieval system. Results are reranked according to the temporal features of the query during the time the tweet was emitted. All processing and related algorithms are stream-friendly - i.e. they can be supported on the fly, as tweets arrive. The number of followers of users was used; no query-expansion or other external or future data.

FASILKOM01¶

Run ID: FASILKOM01
Participant: FASILKOMUI
Track: Microblog
Year: 2012
Submission: 7/10/2012
Type: automatic
Task: adhoc
MD5: fbba4448139749d952dbdc32e03d2a99
Run description: we use internal dataset for query expansion on specific range time before querytweettime. we use external dataset for query expansion on specific range time before querytweettime using news from online news portal. we cluster the result based on time range to re-rank document. we use hashtag, %match, and link features to re-rank document.

FASILKOM02¶

Run ID: FASILKOM02
Participant: FASILKOMUI
Track: Microblog
Year: 2012
Submission: 7/10/2012
Type: automatic
Task: adhoc
MD5: 64ec62a9d42d2fe2b2c2a8b3b72c6ae5
Run description: we use external dataset for query expansion on specific range time before querytweettime using news data from online portal news. we cluster the result based on time range to re-rank document.

FASILKOM03¶

Run ID: FASILKOM03
Participant: FASILKOMUI
Track: Microblog
Year: 2012
Submission: 7/10/2012
Type: automatic
Task: adhoc
MD5: 2c68c5479ac9e2c1658ee3471f6829c1
Run description: we use internal dataset for query expansion on specific range time before querytweettime. we use external dataset for query expansion on specific range time before querytweettime using news data from online portal news.

FASILKOM04¶

Run ID: FASILKOM04
Participant: FASILKOMUI
Track: Microblog
Year: 2012
Submission: 7/10/2012
Type: automatic
Task: adhoc
MD5: 5f6ba943493f42671fd12088eadb8c56
Run description: we use internal query expansion on specific range time before querytweettime. we cluster the result based on time range to re-rank document. we use hashtag, %match, and link in tweet to adding some score and re-rank document.

FasilkomF1¶

Results | Participants | Input | Summary | Appendix

Run ID: FasilkomF1
Participant: FASILKOMUI
Track: Microblog
Year: 2012
Submission: 7/31/2012
Type: automatic
Task: filtering
MD5: 68b9526a158ba16f9765c9d82f5f3f2a
Run description: In this run, we use features such as the score of matching words, existence of link, mention, and hashtag. We also use feature adapted from Rocchio algorithm which use query terms. For external resources, we use news articles which are released before query tweettime.

FasilkomF2¶

Results | Participants | Input | Summary | Appendix

Run ID: FasilkomF2
Participant: FASILKOMUI
Track: Microblog
Year: 2012
Submission: 7/31/2012
Type: automatic
Task: filtering
MD5: 3936c732fc06d00f87061bfa4d84ebef
Run description: In this run, we use features such as the score of matching words, existence of link, mention, and hashtag. We also use feature adapted from Rocchio algorithm which use query and expansion terms. For external resources, we use news articles which are released before query tweettime.

FasilkomF3¶

Results | Participants | Input | Summary | Appendix

Run ID: FasilkomF3
Participant: FASILKOMUI
Track: Microblog
Year: 2012
Submission: 7/31/2012
Type: automatic
Task: filtering
MD5: c70b4b025630c01b3cb6eabd00fcf144
Run description: In this run, we use features such as the score of matching words, existence of link, mention, and hashtag. We also use feature adapted from Rocchio algorithm. We do not use any external resources for this run.

FasilkomF4¶

Results | Participants | Input | Summary | Appendix

Run ID: FasilkomF4
Participant: FASILKOMUI
Track: Microblog
Year: 2012
Submission: 7/31/2012
Type: automatic
Task: filtering
MD5: db6012906ae033b04f4f3b8b15c8b3f9
Run description: In this run, we use features such as the score of matching words, existence of link, mention, and hashtag. We also use external resources from news articles which are released before query tweettime.

FRUN1¶

Results | Participants | Input | Summary | Appendix

Run ID: FRUN1
Participant: uog_tw
Track: Microblog
Year: 2012
Submission: 7/31/2012
Type: automatic
Task: filtering
MD5: f366c83abee7642a43d5d3a7cfd5f5bb
Run description: Full real-time approach, using a novel term weighting scheme that exploits temporal evidence, called IBF.

FRUN2¶

Results | Participants | Input | Summary | Appendix

Run ID: FRUN2
Participant: uog_tw
Track: Microblog
Year: 2012
Submission: 7/31/2012
Type: automatic
Task: filtering
MD5: 1428c5c63f7bdc06816710485e445f59
Run description: Full real-time approach, using a novel term weighting scheme that exploits temporal evidence, called IBF. This time we make use of PRF for query expansion.

FRUN3¶

Results | Participants | Input | Summary | Appendix

Run ID: FRUN3
Participant: uog_tw
Track: Microblog
Year: 2012
Submission: 7/31/2012
Type: automatic
Task: filtering
MD5: 6abb91eb170dcf01a3dc8560db103e76
Run description: Term frequency normalized by query length, with query expansion using relevance feedback, as described by the guidelines.

gucasB¶

Run ID: gucasB
Participant: GUCAS
Track: Microblog
Year: 2012
Submission: 7/31/2012
Type: automatic
Task: filtering
MD5: 2d697f7996085ed79e6d9cd447a26e4d
Run description: A run using content-based model

gucasBasic¶

Run ID: gucasBasic
Participant: GUCAS
Track: Microblog
Year: 2012
Submission: 7/10/2012
Type: automatic
Task: adhoc
MD5: af14f7f76acb692f148125f9b24cccbd
Run description: A run using a content-based model

gucasGen¶

Run ID: gucasGen
Participant: GUCAS
Track: Microblog
Year: 2012
Submission: 7/10/2012
Type: automatic
Task: adhoc
MD5: ddc8962e7a307e59db0693bb9976acbe
Run description: A run using a learning method based on the tweet content

gucasGenReg¶

Run ID: gucasGenReg
Participant: GUCAS
Track: Microblog
Year: 2012
Submission: 7/10/2012
Type: automatic
Task: adhoc
MD5: ee3f4700dde1a658de1454e314d39a48
Run description: A run using a combined method

gucasL1¶

Run ID: gucasL1
Participant: GUCAS
Track: Microblog
Year: 2012
Submission: 7/31/2012
Type: automatic
Task: filtering
MD5: f1eba23b4a9e090aee85a21c10836a46
Run description: A run using LTR1

gucasL2¶

Run ID: gucasL2
Participant: GUCAS
Track: Microblog
Year: 2012
Submission: 7/31/2012
Type: automatic
Task: filtering
MD5: 1c6258f3c712f2ceabd4f4139b173a48
Run description: A run using LTR2

gucasQuery¶

Run ID: gucasQuery
Participant: GUCAS
Track: Microblog
Year: 2012
Submission: 7/10/2012
Type: automatic
Task: adhoc
MD5: 5087f6bbf89fb8c925af68c5cc4028fd
Run description: A run using a learning method which taking queries into account

hitDELMrun2¶

Run ID: hitDELMrun2
Participant: HIT_MTLAB
Track: Microblog
Year: 2012
Submission: 7/10/2012
Type: automatic
Task: adhoc
MD5: 11a3966283edbe760c27d958956eb082
Run description: Considering very short tweet messages carries low information , we develop a retrieval model which combine query expansion and document expansion simultaneous under a language modeling approach. The retrieval performance can be much improved by document enrichment which makes estimating document model in LM more accuracy.

hitLRrun1¶

Run ID: hitLRrun1
Participant: HIT_MTLAB
Track: Microblog
Year: 2012
Submission: 7/10/2012
Type: automatic
Task: adhoc
MD5: 93ef42272c3857ae7e15511a88927709
Run description: We employ and ultilize a simple learning-to-rank model (modified Logistic regression model), which uses not only not the content relevance of a tweet, but also some twitter specific features such as whether a URL link is included in the tweet, followers count, hashtag and so on.

hitQryFBrun4¶

Run ID: hitQryFBrun4
Participant: HIT_MTLAB
Track: Microblog
Year: 2012
Submission: 7/10/2012
Type: automatic
Task: adhoc
MD5: 799468513db2663578b39eb48573cedf
Run description: Twitter queries are also typically very short in length, for this task we used pseudo-relevance feedback based on Lavrenko-Croft models to try overcoming term miss match in retrieval. In this method, we simply preserved twitter specific symbol (mention, hashtag and url-link in tweet) to expand more effective terms by capturing tweet specific signal.

hitRSW¶

Run ID: hitRSW
Participant: HIT_MTLAB
Track: Microblog
Year: 2012
Submission: 8/06/2012
Type: automatic
Task: filtering
Run description: In this run, we also adopt windows measure with various number of relevance tweets.Compared with runn window2run, the number of tweet in windows is adaptive adjusted by the time instead of fixing two in window2run. And also the adjusted process is timely.

hitURLrun3¶

Run ID: hitURLrun3
Participant: HIT_MTLAB
Track: Microblog
Year: 2012
Submission: 7/10/2012
Type: automatic
Task: adhoc
MD5: 2270ba24507247dfc0c6faa6c4158fda
Run description: We get the last relevance score of a tweet with the query by combine two aspects of relevance score linearly weighting. One is the relevance score between tweet itself with query, and the second one is the relevance score between the query whit content directed by url-links included in tweet. If a tweet has no url-link, we give a default score to the second part. This is our only run which used external resources, but are timely with respect to queries.

hitUWT¶

Run ID: hitUWT
Participant: HIT_MTLAB
Track: Microblog
Year: 2012
Submission: 8/06/2012
Type: automatic
Task: filtering
Run description: KL divergence is used to find out the most relevance tweet . The ranking logistic regression model is the filter model for others. In this run, we used documents linked from tweets.

IBCN1¶

Run ID: IBCN1
Participant: UGENT_IBCN_SIS
Track: Microblog
Year: 2012
Submission: 7/10/2012
Type: automatic
Task: adhoc
MD5: 912fda15177a0cd4bebf807c5ef5ae12
Run description: - no external resources - basic features; ordered by probability of Logistic Regression classifier

IBCN2¶

Run ID: IBCN2
Participant: UGENT_IBCN_SIS
Track: Microblog
Year: 2012
Submission: 7/10/2012
Type: automatic
Task: adhoc
MD5: 8bb87450f6c3476b7f3b2e0297302ccd
Run description: - external resources--- only used titles from linked pages - used several title- and tweet-based features; ordered by probability of Logistic Regression classifier

IBCN3¶

Run ID: IBCN3
Participant: UGENT_IBCN_SIS
Track: Microblog
Year: 2012
Submission: 7/10/2012
Type: automatic
Task: adhoc
MD5: caf2dcab8bc75b8d6ca464ac122b69cd
Run description: - external resources--- only used titles from linked pages - used well-chosen title- and tweet-based features; ordered by probability of Logistic Regression classifier

IBCN4¶

Run ID: IBCN4
Participant: UGENT_IBCN_SIS
Track: Microblog
Year: 2012
Submission: 7/10/2012
Type: automatic
Task: adhoc
MD5: 956140b1f439c16025691cc44b3a4f1a
Run description: - external resources--- only used titles from linked pages - used well-chosen title- and tweet-based features; ordered by probability of Logistic Regression classifier, but partly re-ranked based on time

IBMBaseline¶

Run ID: IBMBaseline
Participant: IBM
Track: Microblog
Year: 2012
Submission: 7/11/2012
Type: automatic
Task: adhoc
MD5: 2cf91617a3c2df2523b49f38729962ba
Run description: Baseline Indri full-dependence MRF.

IBMLTR¶

Run ID: IBMLTR
Participant: IBM
Track: Microblog
Year: 2012
Submission: 7/11/2012
Type: automatic
Task: adhoc
MD5: fbe20b985539aa8fb35d92860b00e552
Run description: Baseline + LambdaMART learning-to-rank (JSON-based features)

IBMLTRFuture¶

Run ID: IBMLTRFuture
Participant: IBM
Track: Microblog
Year: 2012
Submission: 7/11/2012
Type: automatic
Task: adhoc
MD5: ff8eefebd99e3eb7b48998bfb977c884
Run description: Baseline + LambdaMART learning-to-rank (JSON-based features) + future features (num retweets, num followers, num statuses, etc.)

ICTNETFTRUN1¶

Run ID: ICTNETFTRUN1
Participant: ICTNET
Track: Microblog
Year: 2012
Submission: 7/31/2012
Type: automatic
Task: filtering
MD5: d2887c7100e10a05d064c516f1a80dbd
Run description: We applying an adaptive filtering method to this task. Tweets are handled one by one just like a stream. Topics are initialized with original queries and first relevant tweet. For each tweet, we judge its relevance using VSM and update topic description if tweets are judged positive. Besides, we take some static features of tweets into consideration to compute a "quality score" for each tweet. In the submission file, we omitted all the tweets with a score of 0.0 under VSM.

ICTNETFTRUN2¶

Run ID: ICTNETFTRUN2
Participant: ICTNET
Track: Microblog
Year: 2012
Submission: 8/1/2012
Type: automatic
Task: filtering
MD5: ba8f13a9ad8945714182c40acb1c182a
Run description: Similar method to ICTNETFTRUN1 is adopted. Besides, we enhanced the initial description of the topics with top retrieval results from Google.

ICTNETFTRUN3¶

Run ID: ICTNETFTRUN3
Participant: ICTNET
Track: Microblog
Year: 2012
Submission: 8/1/2012
Type: automatic
Task: filtering
MD5: 9c6ed944f388541596cbb0717d1e75f6
Run description: Adaptive filtering method is applied, and language model is used to calculate the similarity between tweets and topics. Positive tweets are used to update the description of topics, which is maintained using a queue structure with a limited size. "Quality score" is also included.

ICTNETFTRUN4¶

Run ID: ICTNETFTRUN4
Participant: ICTNET
Track: Microblog
Year: 2012
Submission: 8/1/2012
Type: automatic
Task: filtering
MD5: 1484f61e1472af2fcc4b0b96a578abe5
Run description: Similar method to ICTNETFTRUN3 is adopted, and top retrieval results from google are used to expand the topic.

ICTWDSERUN1¶

Run ID: ICTWDSERUN1
Participant: ICTNET
Track: Microblog
Year: 2012
Submission: 7/10/2012
Type: automatic
Task: adhoc
MD5: 56d5fa1cdb4e56f9b9610c6c14412cbc
Run description: we apply the learning to rank framework(svm rank) to rank the tweets retrieved by Indri. Pseudo relevance-feedback method is used to expand the query.

ICTWDSERUN2¶

Run ID: ICTWDSERUN2
Participant: ICTNET
Track: Microblog
Year: 2012
Submission: 7/10/2012
Type: automatic
Task: adhoc
MD5: 98c9fb265d0702dd4672df78246b2dca
Run description: we apply the learning to rank framework(listNet) to rank the tweets retrieved by Indri. Pseudo relevance-feedback method , which makes use of the word concurrency is used to expand the query.

ICTWDSERUN3¶

Run ID: ICTWDSERUN3
Participant: ICTNET
Track: Microblog
Year: 2012
Submission: 7/10/2012
Type: automatic
Task: adhoc
MD5: 3faafffbdd7dbdae8797d89ca8b9214c
Run description: we apply the learning to rank framework(SVM Rank) to rank the tweets retrieved by Indri. Top results retrieved by Google are used to expand the query.

ICTWDSERUN4¶

Run ID: ICTWDSERUN4
Participant: ICTNET
Track: Microblog
Year: 2012
Submission: 7/11/2012
Type: automatic
Task: adhoc
MD5: a7f518aecb3b41963c2acf7461b11770
Run description: we apply the learning to rank framework(ListNet) to rank the tweets retrieved by Indri. Top results from Google are used to expand the query. Besides, we adopt the K-Means algorithm to determine the number of tweets for submission.

IIEIR01¶

Run ID: IIEIR01
Participant: IIEIR
Track: Microblog
Year: 2012
Submission: 7/10/2012
Type: automatic
Task: adhoc
MD5: 66a0db4caf0090666892358aa2b335ec
Run description: this run use briefly Language Model algorithm and time information.

IIEIR02¶

Run ID: IIEIR02
Participant: IIEIR
Track: Microblog
Year: 2012
Submission: 7/10/2012
Type: automatic
Task: adhoc
MD5: 3f38ed1e482dce25143a43bb3bde82f9
Run description: this run use briefly Language Model algorithm and time information.

IIEIR03¶

Run ID: IIEIR03
Participant: IIEIR
Track: Microblog
Year: 2012
Submission: 7/10/2012
Type: automatic
Task: adhoc
MD5: d20187488e1b35b1ff7c199f89d0be8f
Run description: this run use briefly Language Model algorithm and time information.

IIEIR04¶

Run ID: IIEIR04
Participant: IIEIR
Track: Microblog
Year: 2012
Submission: 7/10/2012
Type: automatic
Task: adhoc
MD5: eedec4d614d21ff985990b9ca20e02cc
Run description: this run use briefly Language Model algorithm and time information.

indri¶

Run ID: indri
Participant: udel
Track: Microblog
Year: 2012
Submission: 7/11/2012
Type: automatic
Task: adhoc
MD5: c93572b6ad74334b3205c079ba7d1204
Run description: a baseline run using Indri's build-in language models with pseudo-relevance feedback

IRITbnetK¶

Run ID: IRITbnetK
Participant: IRIT
Track: Microblog
Year: 2012
Submission: 7/10/2012
Type: automatic
Task: adhoc
MD5: 9d326ddff8be1e544156c1d1ce2ad68a
Run description: This run is performed by a Bayesian Network Model that integrates the term frequency in tweet.

IRITbnetKSO¶

Run ID: IRITbnetKSO
Participant: IRIT
Track: Microblog
Year: 2012
Submission: 7/10/2012
Type: automatic
Task: adhoc
MD5: 7e61639f0e00e8dca6643b548cf4497c
Run description: This run is performed by a Bayesian Network Model that integrates the term frequency in the tweet, the social importance of microbloggers and the temporal relavance of tweet time.

IRITfdvsm¶

Run ID: IRITfdvsm
Participant: IRIT
Track: Microblog
Year: 2012
Submission: 7/10/2012
Type: automatic
Task: adhoc
MD5: 15c3db9ba61f8d132a5b9691ed7d8e18
Run description: The run is created using a machine learning appraoch. Trec microblog 2011 topics ans qrels were used for the learning. The model is based on a set features: lucene VSM score, tweet popularity score, tweet length, exact term matching, url presence, url frequency, hashtag frequency.

IRITfdvsmurl¶

Run ID: IRITfdvsmurl
Participant: IRIT
Track: Microblog
Year: 2012
Submission: 7/10/2012
Type: automatic
Task: adhoc
MD5: bfa3f82af649637044bd77ca99e0ea88
Run description: The run is created using a machine learning appraoch. A part from the tweet, the search was done on the web pages' titles of the links if they exist. Trec microblog 2011 topics ans qrels were used for the learning. The model is based on a set features: lucene VSM score, tweet popularity score, tweet length, exact term matching, url presence, url frequency, hashtag frequency.

irsicombsum¶

Results | Participants | Input | Summary | Appendix

Run ID: irsicombsum
Participant: IRSI
Track: Microblog
Year: 2012
Submission: 8/1/2012
Type: automatic
Task: filtering
MD5: 4b97f5748857d76da45f86cfdb24210c
Run description: We have generated initial run by InL2 (c=1.0) model. Then we treated the top 20 retrieved tweets as queries and generated another 20 runs. Then we use fusion algorithm to determine relevant tweets.

IRSIISI¶

Run ID: IRSIISI
Participant: IRSI
Track: Microblog
Year: 2012
Submission: 7/11/2012
Type: automatic
Task: adhoc
MD5: dd3378b818dee4f8cc858a9965854204
Run description: Top 30 full tweets were added to expand query. A new re-ranking algorithm also used.

IRSIISI1¶

Run ID: IRSIISI1
Participant: IRSI
Track: Microblog
Year: 2012
Submission: 7/11/2012
Type: automatic
Task: adhoc
MD5: ec308a1789118f5c01891b1b193f265f
Run description: Top 35 full tweets were added to expand query. A new re-ranking algorithm also used.

IRSIISI2¶

Run ID: IRSIISI2
Participant: IRSI
Track: Microblog
Year: 2012
Submission: 7/11/2012
Type: automatic
Task: adhoc
MD5: 4312a62f8d44ae65fd42769a82e02585
Run description: Top 40 full tweets were added to expand query. A new re-ranking algorithm also used.

IRSIISI3¶

Run ID: IRSIISI3
Participant: IRSI
Track: Microblog
Year: 2012
Submission: 7/11/2012
Type: automatic
Task: adhoc
MD5: e9cc2a3fbba79e9ed881f72b56f7601d
Run description: Top 45 full tweets were added to expand query. A new re-ranking algorithm also used.

irsivoting¶

Results | Participants | Input | Summary | Appendix

Run ID: irsivoting
Participant: IRSI
Track: Microblog
Year: 2012
Submission: 7/31/2012
Type: automatic
Task: filtering
MD5: c0a171cd841e95fa91741c95df7b6421
Run description: We have generated initial run by InL2 (c=1.0) model. Then we treated the top 20 retrieved tweets as queries and generated another 20 runs. Then we use voting algorithm to determine relevant tweets.

KLIM¶

Run ID: KLIM
Participant: FUB
Track: Microblog
Year: 2012
Submission: 7/10/2012
Type: automatic
Task: adhoc
MD5: f56818a16b272073f43f77ca2be247e8
Run description: Achieved by using the DFReeKLIM retrieval model togheter with the Bose-Einstain query expansion (30 documents and 10 terms)

KLIMLL¶

Run ID: KLIMLL
Participant: FUB
Track: Microblog
Year: 2012
Submission: 7/10/2012
Type: automatic
Task: adhoc
MD5: b2d5b217f52230f766fc808caf4fb285
Run description: We re-rank documents retrieved by the DFReeKLIM retrieval model togheter with the Bose-Einstain query expansion (30 documents and 10 terms) via a combination of relevance score with a smoothing temporal score obtained by applying the log-logistic hazard function on relevance ranks.

KLIMLP1¶

Run ID: KLIMLP1
Participant: FUB
Track: Microblog
Year: 2012
Submission: 7/10/2012
Type: automatic
Task: adhoc
MD5: d4f1f0b6263d1c960d1531e40624fd04
Run description: We retrieve tweets using the DFReeKLIM retrieval model and applying the Bose-Einstain query expansion algorithm (30 documents and 10 terms). We also filter out non-english tweets using LingPipe, trained using the standard sample collection provided togheter with the tool.

KLIMLPLL¶

Run ID: KLIMLPLL
Participant: FUB
Track: Microblog
Year: 2012
Submission: 7/10/2012
Type: automatic
Task: adhoc
MD5: e329e2a32ed32d0784821024ca09cc05
Run description: We retrieve tweets using the DFReeKLIM retrieval model and applying the Bose-Einstain query expansion algorithm (30 documents and 10 terms). We also filter out non-english tweets using LingPipe, trained using the standard sample collection provided togheter with the tool. We finally re-rank retrieved tweets by combining the relevance score with a temporal score obtained by applying the log-logistic hazard function on the rank of relevance.

kobeL2R¶

Run ID: kobeL2R
Participant: KobeU
Track: Microblog
Year: 2012
Submission: 7/5/2012
Type: automatic
Task: adhoc
MD5: 6dd1bb021ede19370ace9a59254356f0
Run description: This run uses learning to rank approach that combines some query expansion methods and text-features.

kobeMHC¶

Run ID: kobeMHC
Participant: KobeU
Track: Microblog
Year: 2012
Submission: 7/10/2012
Type: manual
Task: adhoc
MD5: 4bdf984f640298c5b8c5b0ec85b5d500
Run description: this run use a temporal profile of a manually selected tweet and be applied the query expansion based on the temporal profile similarity.

kobeMHC2¶

Run ID: kobeMHC2
Participant: KobeU
Track: Microblog
Year: 2012
Submission: 7/11/2012
Type: manual
Task: adhoc
MD5: 3c0857b94abd6acce9ee151ff9a6a4e4
Run description: this run uses a query combined with a tweet selected by a user and re-ranks tweets with another query expansion using temporal profile based on the time stamps of the retrieved tweets.

langluc¶

Run ID: langluc
Participant: udel
Track: Microblog
Year: 2012
Submission: 7/11/2012
Type: automatic
Task: adhoc
MD5: dbfbdf9505cef5f984362a357d4be352
Run description: Just Basic Language Model in Lucene

mergedRun¶

Run ID: mergedRun
Participant: qcri_twitsear
Track: Microblog
Year: 2012
Submission: 7/11/2012
Type: automatic
Task: adhoc
MD5: 7f2d866bf8d8922b241ff91c863304d8
Run description: Using BM25 as the basic retrieval model. Then five different feedback approaches are adopted that combine the temporal based reranking and PRF. Finally the 5 result lists are merged by a learned merge model that was trained to maximize MAP using 2011 relevance judgement as training corpus.

nemisExt¶

Run ID: nemisExt
Participant: NEMIS_ISTI_CNR
Track: Microblog
Year: 2012
Submission: 8/1/2012
Type: automatic
Task: filtering
MD5: 21d957ce0709d1919043d28dd02f103b
Run description: In this run we used an external corpus to train the named entity recognition system plus the title of pages linked from tweets, plus a subset of google ngrams to perform hashtag splitting. The filtering is performed by a Naive Bayes classifier

nemisNotExt¶

Run ID: nemisNotExt
Participant: NEMIS_ISTI_CNR
Track: Microblog
Year: 2012
Submission: 8/1/2012
Type: automatic
Task: filtering
MD5: 98010e87d4c00c89fa1758ba80ac5056
Run description: This run doesn't use any external resource, the filtering is performed by a Naive Bayes classifier

okapiv1¶

Run ID: okapiv1
Participant: unir_de
Track: Microblog
Year: 2012
Submission: 7/23/2012
Type: automatic
Task: filtering
MD5: 965487be7acf89e5c3e9c6abfdd0a593
Run description: Properties of a Tweet are mapped onto semantically distinct event types, each of which is feed into a separate event streams. IDF weights are calculated online by using sliding time windows. For scoring an adapted OkapBM25 is used. Instead of the TF in the formula the tf is derived from ranks of the token within the different event streams, i.e. the rank of a token in the RT, HT or Token are combined. Formula tf = [stream_weight]*(1/Log(rank+0.5). Due to stream and event based processing the scores are always real time. Due to its real time nature a run would in fact take 16 days, but the event rate was accelerated to 1000/s.

okapiv2rel¶

Run ID: okapiv2rel
Participant: unir_de
Track: Microblog
Year: 2012
Submission: 7/31/2012
Type: automatic
Task: filtering
MD5: 3f7e9a2220059df1ebc602b6c73b2719
Run description: Event based okapi bm25 like the first submitted run (okapiv1), but incorporates relevance feedback, i.e. each relevant tweet is feed into the search profile. the top 5 terms are then added to the query. relevance threshold is 0.5

otM12h¶

Run ID: otM12h
Participant: ot
Track: Microblog
Year: 2012
Submission: 7/9/2012
Type: automatic
Task: adhoc
MD5: 70c98c13d414290d7e89775b7bebb282
Run description: same as otM12ih except that IDF was not used

otM12i¶

Run ID: otM12i
Participant: ot
Track: Microblog
Year: 2012
Submission: 7/9/2012
Type: automatic
Task: adhoc
MD5: edb0871d0cf3fe39b93a29b3526aadd5
Run description: tf.idf vector run just based on query terms; English inflections were matched; English stopwords were not indexed; only future evidence was full-collection IDF

otM12ih¶

Run ID: otM12ih
Participant: ot
Track: Microblog
Year: 2012
Submission: 7/9/2012
Type: automatic
Task: adhoc
MD5: 1f6bbb6b33d5b17e0cfff6c1a8ad58d5
Run description: same as otM12i except that the word 'HTTP' was added to each query because it was a common term in relevant items last year; only future evidence was full-collection IDF

otM12ihe¶

Run ID: otM12ihe
Participant: ot
Track: Microblog
Year: 2012
Submission: 7/9/2012
Type: automatic
Task: adhoc
MD5: f7d294ad37a15bf56192dd011332547f
Run description: blind feedback run based on first 30 pre-querytweettime items retrieved by otM12ih before filtering out retweets or 301s; only future evidence was full-collection IDF

PKUICST1¶

Run ID: PKUICST1
Participant: PKUICST
Track: Microblog
Year: 2012
Submission: 7/9/2012
Type: automatic
Task: adhoc
MD5: e57a47afa8679dacbfbd591e58408270
Run description: Using Learning-to-rank algorithm, generating candidate tweets from KL2SLocCtxt Model, training of 49 topics from last year.

PKUICST2¶

Run ID: PKUICST2
Participant: PKUICST
Track: Microblog
Year: 2012
Submission: 7/9/2012
Type: automatic
Task: adhoc
MD5: 11a7526f8ffe550093cb145f7dc03f4e
Run description: Using Learning-to-rank algorithm, generating candidate tweets from KL2SLocCtxt Model, training of 33 highrel topics from last year.

PKUICST3¶

Run ID: PKUICST3
Participant: PKUICST
Track: Microblog
Year: 2012
Submission: 7/9/2012
Type: automatic
Task: adhoc
MD5: 4c67c0034aabda486c96ac98e2371d15
Run description: KL2SLocCtxt Model, using parameters as set in the Paper

PKUICST4¶

Run ID: PKUICST4
Participant: PKUICST
Track: Microblog
Year: 2012
Submission: 7/9/2012
Type: automatic
Task: adhoc
MD5: a9e9286bf957f3475936cca03363cc01
Run description: Using Learning-to-rank algorithm, generating candidate tweets from KL2SFB Model, training of 49 topics from last year.

PKUICSTF1¶

Run ID: PKUICSTF1
Participant: PKUICST
Track: Microblog
Year: 2012
Submission: 7/31/2012
Type: automatic
Task: filtering
MD5: f5e11ec95a7e9429735f48d67e89d5f3
Run description: Using a two-layer filtering model which combined rocchio model and Boolean retrieve model.

PKUICSTF2¶

Run ID: PKUICSTF2
Participant: PKUICST
Track: Microblog
Year: 2012
Submission: 7/31/2012
Type: automatic
Task: filtering
MD5: c2a5414a9659168776606c5ddca74c9f
Run description: using two-stage strategy, including SVM classifier(c=2,g=1) and a specific threshold(4.7)

PKUICSTF3¶

Run ID: PKUICSTF3
Participant: PKUICST
Track: Microblog
Year: 2012
Submission: 7/31/2012
Type: automatic
Task: filtering
MD5: 54e0ff7d26de513d15bc7f8fed1e345a
Run description: using language model only with a static threshold set as -6.02

PKUICSTF4¶

Run ID: PKUICSTF4
Participant: PKUICST
Track: Microblog
Year: 2012
Submission: 7/31/2012
Type: automatic
Task: filtering
MD5: 442ad59459c91d6351f3dbdf5bd5a86b
Run description: using two-stage strategy, including SVM classifier(c=2,g=1) and a specific threshold(4.4),using external link resources

PRISrun1¶

Run ID: PRISrun1
Participant: PRIS
Track: Microblog
Year: 2012
Submission: 7/31/2012
Type: automatic
Task: filtering
MD5: 38105472349b3b1042e22dc746650131
Run description: We didn't use external resources or future evidence. We use 'learning to rank' for scoring and real-time dynamic adjustment for threshold evolution.

prisRun1¶

Run ID: prisRun1
Participant: BUPT_WILDCAT
Track: Microblog
Year: 2012
Submission: 7/10/2012
Type: automatic
Task: adhoc
MD5: f946cec88980126f98d58e2f0a2d201d
Run description: We didn't use the future evidence. We use resistance network and LDA for query expansion and Learning to rank for re-scoring.

PRISrun2¶

Run ID: PRISrun2
Participant: PRIS
Track: Microblog
Year: 2012
Submission: 7/31/2012
Type: automatic
Task: filtering
MD5: 79993b40a0099131ee5c1cf3d727990d
Run description: We didn't use future evidence. We use 'learning to rank' for scoring and real-time dynamic adjustment for threshold evolution.

prisRun2¶

Run ID: prisRun2
Participant: BUPT_WILDCAT
Track: Microblog
Year: 2012
Submission: 7/10/2012
Type: automatic
Task: adhoc
MD5: b7dd122fbf6245c73b089c41999c246f
Run description: We didn't use the future evidence. We use resistance network, LDA and thesaurus for query expansion and Learning to rank for re-scoring.

PRISrun3¶

Run ID: PRISrun3
Participant: PRIS
Track: Microblog
Year: 2012
Submission: 7/31/2012
Type: automatic
Task: filtering
MD5: a50ff975dd707041349bd59724792e7b
Run description: We didn't use future evidence. We use 'learning to rank' for scoring and real-time dynamic adjustment for threshold evolution.

prisRun3¶

Run ID: prisRun3
Participant: BUPT_WILDCAT
Track: Microblog
Year: 2012
Submission: 7/10/2012
Type: automatic
Task: adhoc
MD5: 72db54644fc576455bfba4c58b0543d0
Run description: We didn't use the future evidence. We use resistance network and LDA for query expansion and Learning to rank for re-scoring.

PRISrun4¶

Run ID: PRISrun4
Participant: PRIS
Track: Microblog
Year: 2012
Submission: 7/31/2012
Type: automatic
Task: filtering
MD5: 798c9763b5ac99917d4e163530dedc97
Run description: We didn't use future or external evidence. We use 'learning to rank' for scoring and real-time dynamic adjustment for threshold evolution.

prisRun4¶

Run ID: prisRun4
Participant: BUPT_WILDCAT
Track: Microblog
Year: 2012
Submission: 7/10/2012
Type: automatic
Task: adhoc
MD5: 9991ab27ff979cd42694563787d24524
Run description: We didn't use the future evidence. We use resistance network, LDA and thesaurus for query expansion and Learning to rank for re-scoring.

QEWeb¶

Run ID: QEWeb
Participant: QCRI
Track: Microblog
Year: 2012
Submission: 7/10/2012
Type: automatic
Task: adhoc
MD5: 847665dd2560dd8295ecff7b1a7fd548
Run description: Queries are used to search the web for relevant web results on the same topic in the period of the tweets collection (with upper bound to the query date). then the top result from the web is used to expand the query and search the collection

QEWebFB¶

Run ID: QEWebFB
Participant: QCRI
Track: Microblog
Year: 2012
Submission: 7/10/2012
Type: automatic
Task: adhoc
MD5: 323769de877e1f380fbb3595d0de3ab9
Run description: Queries are used to search the web for relevant web results on the same topic in the period of the tweets collection (with upper bound to the query date). then the top result from the web is used to expand the query and search the collection + applying pseudo relevance feedback

QFilRun1¶

Run ID: QFilRun1
Participant: qcri_twitsear
Track: Microblog
Year: 2012
Submission: 8/1/2012
Type: automatic
Task: filtering
MD5: 5899b7a8c5f09bcbc200ba7526b081a6
Run description: This run is based on BM25 scoring function with Rocchio feedback. Using 22,125 tweets in the corpus that are out of the time range between the lowest querytweettime and the highest querynewesttweet of all queries as prior IDF values.

QFilRun2¶

Run ID: QFilRun2
Participant: qcri_twitsear
Track: Microblog
Year: 2012
Submission: 8/1/2012
Type: automatic
Task: filtering
MD5: da8a8a7ae2a27d21d4e59d692328c87d
Run description: This run is based on BM25 scoring function with Rocchio feedback. Using only tweets that are up to the current tweet for computing prior IDF values.

QFilRun3¶

Run ID: QFilRun3
Participant: qcri_twitsear
Track: Microblog
Year: 2012
Submission: 8/06/2012
Type: automatic
Task: filtering
Run description: This run is based on BM25 scoring function with Rocchio feedback using only tweets that are up to the current tweet for computing prior IDF values, and more strict threshold update rules were applied.

RetrievalThr¶

Run ID: RetrievalThr
Participant: QCRI
Track: Microblog
Year: 2012
Submission: 8/1/2012
Type: automatic
Task: filtering
MD5: 1fc0fb1fab71d194071a822102ec97ca
Run description: Just normalizing scores and taking a fixed threshold for all topics based on the training set

RUN1¶

Run ID: RUN1
Participant: uog_tw
Track: Microblog
Year: 2012
Submission: 7/10/2012
Type: automatic
Task: adhoc
MD5: a717fb77d35c0d7b7edd01bb922fab03
Run description: Base run. Filtered based upon language, number of hash tags, number of mentions, number of URLs. Uses TF only.

RUN2¶

Run ID: RUN2
Participant: uog_tw
Track: Microblog
Year: 2012
Submission: 7/11/2012
Type: automatic
Task: adhoc
MD5: 14432b2f75ea58105dc899b592b125b9
Run description: Base run. Filtered based upon language, number of hash tags, number of mentions, number of URLs. TF ranking, with query expansion using PRF.

RUN3¶

Run ID: RUN3
Participant: uog_tw
Track: Microblog
Year: 2012
Submission: 7/11/2012
Type: automatic
Task: adhoc
MD5: bcb92e119b72927ea43a622b0259c18a
Run description: Base run. Filtered based upon language, number of hash tags, number of mentions, number of URLs. TF ranking, with query expansion using PRF enhanced by burst detection.

timemexp¶

Run ID: timemexp
Participant: udel
Track: Microblog
Year: 2012
Submission: 7/11/2012
Type: automatic
Task: adhoc
MD5: 8499ce1811bd26ee69cbe7d8162e1ac6
Run description: Just Basic Language Model in Lucene plus weight-age given to terms which are in top documents in basic language model plus query expansion

timemodel¶

Run ID: timemodel
Participant: udel
Track: Microblog
Year: 2012
Submission: 7/11/2012
Type: automatic
Task: adhoc
MD5: 9aaa016792c77a51b4ec8d6e9ce27cfb
Run description: Just Basic Language Model in Lucene plus weight-age given to terms which are in top documents in basic language model

tsqe¶

Run ID: tsqe
Participant: KobeU
Track: Microblog
Year: 2012
Submission: 7/4/2012
Type: automatic
Task: adhoc
MD5: 4b2bb10fe7fbef4a509df53e519a14e7
Run description: This run uses a query expansion method based on the pseudo-relevance feedback considering temporal dynamics of words.

udelcosrun¶

Results | Participants | Input | Summary | Appendix

Run ID: udelcosrun
Participant: udel
Track: Microblog
Year: 2012
Submission: 7/31/2012
Type: automatic
Task: filtering
MD5: 9e0cea3b9250c0d4b86a029388a60ea4
Run description: Uses cosine similarity and thresholding

udellngth¶

Results | Participants | Input | Summary | Appendix

Run ID: udellngth
Participant: udel
Track: Microblog
Year: 2012
Submission: 8/1/2012
Type: automatic
Task: filtering
MD5: 108618bf34aaca98843bfde91e8ed595
Run description: Language Modeling threshold value from the relevance judgements of training queries

UDInfoMBCW¶

Run ID: UDInfoMBCW
Participant: udel_fang
Track: Microblog
Year: 2012
Submission: 7/10/2012
Type: automatic
Task: adhoc
MD5: 632f200dd4fb330be93548942c6dd7a9
Run description: Use DBPedia Sept. 2011 to detect concepts and query expansion based on each concept. For each concept, decrease additional gain similar to diversity search to avoid over-weighting some concepts.

UDInfoMBEx¶

Run ID: UDInfoMBEx
Participant: udel_fang
Track: Microblog
Year: 2012
Submission: 7/10/2012
Type: automatic
Task: adhoc
MD5: 42ce4911d264d2ce6ba9380113f0f705
Run description: Use DBpedia Sept. 2011 to detect concepts and query expansion based each concept.

UDInfoMBIDF¶

Run ID: UDInfoMBIDF
Participant: udel_fang
Track: Microblog
Year: 2012
Submission: 7/10/2012
Type: automatic
Task: adhoc
MD5: c374a18513e880f12fae1b6b181099e4
Run description: Ranked by IDF, if IDF scores are equal, document in hot-discussed period will rank higher.

UDInfoMBTp¶

Run ID: UDInfoMBTp
Participant: udel_fang
Track: Microblog
Year: 2012
Submission: 7/10/2012
Type: automatic
Task: adhoc
MD5: 6e38edc1d4c5b4187e82ebd52a312143
Run description: Only consider IDF, with strong temporal information.

uiucGSLIS01¶

Run ID: uiucGSLIS01
Participant: uiucGSLIS
Track: Microblog
Year: 2012
Submission: 7/9/2012
Type: automatic
Task: adhoc
MD5: 926bd3ccad82ac214a46640fb620e061
Run description: relies on a novel "attention economic" ranking model and feedback. external resource: relies on the TREC AP corpus for calculating % English words everything about this model is timely except for the fact that the learned weights of features were calculated against all 49 training queries, some of which are "newer" than some test queries.

uiucGSLIS02¶

Run ID: uiucGSLIS02
Participant: uiucGSLIS
Track: Microblog
Year: 2012
Submission: 7/9/2012
Type: automatic
Task: adhoc
MD5: 6b6cfb9035441d0e420c988fe48e3b39
Run description: simple run based on proposed "attention model" with feedback (20 docs, 20 terms, 0.5 interpolation).

uiucGSLIS03¶

Run ID: uiucGSLIS03
Participant: uiucGSLIS
Track: Microblog
Year: 2012
Submission: 7/9/2012
Type: automatic
Task: adhoc
MD5: d9a5050e1917d37c115b9f6e7e092f72
Run description: simple run based on proposed "attention model" with feedback (20 docs, 20 terms, 0.5 interpolation). additionally, documents are promoted for recency via a model based on survival analysis prediction (with censoring at 90%).

uiucGSLIS04¶

Run ID: uiucGSLIS04
Participant: uiucGSLIS
Track: Microblog
Year: 2012
Submission: 7/9/2012
Type: automatic
Task: adhoc
MD5: 1c368bdff06f490994a3274dda3bb205
Run description: simple run based on proposed "attention model". no feedback used.

UNCQE¶

Run ID: UNCQE
Participant: UNC_SILS
Track: Microblog
Year: 2012
Submission: 7/10/2012
Type: automatic
Task: adhoc
MD5: e4e72e77365f2dcb2e146d306b7683ae
Run description: This (baseline) run uses Indri's default pseudo-relevance feedback facility to expand the original query with terms appearing in the original top results. The original query was converted to a full dependence model query and the expansion terms were added as a bag-of-words query. Equal weight was given to the original and expanded query. The Indri query template is: #weight(0.50 FD 0.50 #weight(w_1 e_1 w_2 e_2 ... w_10 e_10)) where w_i and e_i correspond to the weight given to the ith expansion term. The query-expansion parameters were set to fbDocs=100 and fbTerms=10.

UNCRQE¶

Run ID: UNCRQE
Participant: UNC_SILS
Track: Microblog
Year: 2012
Submission: 7/10/2012
Type: automatic
Task: adhoc
MD5: d079cbb8aced23b569f37d8e68338180
Run description: This run uses a modification of the method described in Massoudi et al. [1] in order to expand the original query with terms that co-occur with the original query-terms in recent tweets (relative to the query date/time). The original query was converted to a full dependence model query and the expansion terms were added as a bag-of-words query. Equal weight was given to the original and expanded query. The Indri query template is: #weight(0.50 FD 0.50 #weight(w_1 e_1 w_2 e_2 ... w_10 e_10)) where w_i and e_i correspond to the weight given to the ith expansion term. [1] Kamran Massoudi, Manos Tsagkias, Maarten de Rijke, and Wouter Weerkamp. 2011. Incorporating query expansion and quality indicators in searching microblog posts. In Proceedings of the 33rd European conference on Advances in information retrieval (ECIR'11), Paul Clough, Colum Foley, Cathal Gurrin, Hyowon Lee, and Gareth J. F. Jones (Eds.). Springer-Verlag, Berlin, Heidelberg, 362-367.

UNCTP¶

Run ID: UNCTP
Participant: UNC_SILS
Track: Microblog
Year: 2012
Submission: 7/10/2012
Type: automatic
Task: adhoc
MD5: 31d71badb4d78d9f6b5e615053cde088
Run description: This run uses the method described in Dakka et al. [1] in order to boost the retrieval score of tweets published during relevant time periods. At a high level, the relevant time periods for a given query are automatically found by observing the temporal distribution of the top-ranked tweets. [1] Wisam Dakka, Luis Gravano, and Panagiotis G. Ipeirotis. 2008. Answering general time sensitive queries. In Proceedings of the 17th ACM conference on Information and knowledge management (CIKM '08). ACM, New York, NY, USA, 1437-1438. DOI=10.1145/1458082.1458320 http://doi.acm.org/10.1145/1458082.1458320

UNCTQE¶

Run ID: UNCTQE
Participant: UNC_SILS
Track: Microblog
Year: 2012
Submission: 7/10/2012
Type: automatic
Task: adhoc
MD5: 7fc735bf09b3945d4514f3684c5294b9
Run description: This run combines the methods described in Dakka et al. [1] and Massoudi et al. [2] in order to expand the original query with terms that co-occur with the original query-terms in tweets published during relevant time periods. The relevant time periods were discovered automatically using the same binning approach described in Dakka et al. [1]. The original query was converted to a full dependence model query and the expansion terms were added as a bag-of-words query. Equal weight was given to the original and expanded query. The Indri query template is: #weight(0.50 FD 0.50 #weight(w_1 e_1 w_2 e_2 ... w_10 e_10)) where w_i and e_i correspond to the weight given to the ith expansion term. [1] Wisam Dakka, Luis Gravano, and Panagiotis G. Ipeirotis. 2008. Answering general time sensitive queries. In Proceedings of the 17th ACM conference on Information and knowledge management (CIKM '08). ACM, New York, NY, USA, 1437-1438. DOI=10.1145/1458082.1458320 http://doi.acm.org/10.1145/1458082.1458320. [2] Kamran Massoudi, Manos Tsagkias, Maarten de Rijke, and Wouter Weerkamp. 2011. Incorporating query expansion and quality indicators in searching microblog posts. In Proceedings of the 33rd European conference on Advances in information retrieval (ECIR'11), Paul Clough, Colum Foley, Cathal Gurrin, Hyowon Lee, and Gareth J. F. Jones (Eds.). Springer-Verlag, Berlin, Heidelberg, 362-367.

UnifiedThr¶

Run ID: UnifiedThr
Participant: QCRI
Track: Microblog
Year: 2012
Submission: 8/1/2012
Type: automatic
Task: filtering
MD5: 27d10f1a716386029cd9b4b499acc9fd
Run description: Applying a unified threshold on the score of each result after normalization

uogTrBsE¶

Run ID: uogTrBsE
Participant: uogTr
Track: Microblog
Year: 2012
Submission: 7/10/2012
Type: automatic
Task: adhoc
MD5: 9bb8af1d76968a335a0af062840e38d1
Run description: DFR Retrieval Model plus query expansion and sample expansion using linked documents

uogTrCIDE¶

Run ID: uogTrCIDE
Participant: uogTr
Track: Microblog
Year: 2012
Submission: 7/11/2012
Type: automatic
Task: adhoc
MD5: e351f5e4443c17b9051f92042222c640
Run description: DFR Retrieval Model plus Cross Index Document Expansion of the URLIndex

uogTrFADmI¶

Run ID: uogTrFADmI
Participant: uogTr
Track: Microblog
Year: 2012
Submission: 8/1/2012
Type: manual
Task: filtering
MD5: 4f8477929d1044d0902ef129875068de
Run description: This run uses the Rocchio's classifier and Dirichlet's language model to continuously update a profile over time. The classifier is initialised with an expanded centroid representation. The threshold was adapted using query independent features.

uogTrFADmN¶

Run ID: uogTrFADmN
Participant: uogTr
Track: Microblog
Year: 2012
Submission: 8/1/2012
Type: automatic
Task: filtering
MD5: f0f099ec57c715d3d17e61f5a510782e
Run description: This run uses the Rocchio's classifier and Dirichlet's language model to continuously update a profile over time. The classifier is initialised with an expanded centroid representation and the threshold used to make a decision on each incoming tweet in the stream was chosen and tuned using the 10 training topics.

uogTrFFDmN¶

Run ID: uogTrFFDmN
Participant: uogTr
Track: Microblog
Year: 2012
Submission: 8/1/2012
Type: automatic
Task: filtering
MD5: bf3ff4bcc080f283a1d75b9bba6214b6
Run description: This run uses the Rocchio's classifier and Dirichlet's language model to continuously update a profile over time. The threshold was chosen and tuned using the 10 training topics.

uogTrFFeDm¶

Run ID: uogTrFFeDm
Participant: uogTr
Track: Microblog
Year: 2012
Submission: 8/1/2012
Type: automatic
Task: filtering
MD5: 3eec5f1d3d97421599fb9f2847560c52
Run description: This run uses the Rocchio's classifier and Dirichlet's language model to continuously update a profile over time. An expanded repsentation of the centroid of the classifier is created using query expansion. The threshold was chosen and tuned using the 10 training topics.

uogTrLsE¶

Run ID: uogTrLsE
Participant: uogTr
Track: Microblog
Year: 2012
Submission: 7/11/2012
Type: automatic
Task: adhoc
MD5: 88b75489bf9ed70b4be84c681d9860d6
Run description: DFR Retrieval Model plus query expansion and sample expansion using linked documents with Learning to Rank

urlAllFB¶

Run ID: urlAllFB
Participant: HIT_MTLAB
Track: Microblog
Year: 2012
Submission: 8/06/2012
Type: automatic
Task: filtering
Run description: In this run, we combine various information to deal with adaptive filtering, including query expansion, document expansion and web resources linked from the tweets. Note that all resources we used are timely with respect to the queries

urlContent¶

Run ID: urlContent
Participant: SCIAITeam
Track: Microblog
Year: 2012
Submission: 7/10/2012
Type: automatic
Task: adhoc
MD5: 27c59a712a405bc5125a287902fcdd30
Run description: Used links and the url content on the page to find relevance

UvAfilter¶

Run ID: UvAfilter
Participant: UvA
Track: Microblog
Year: 2012
Submission: 7/11/2012
Type: automatic
Task: adhoc
MD5: dafb28f5cabcde1691d4cb0b1ad89a05
Run description: Baseline Indri retrieval using MRF queries. Keep only tweets with a link AND no mentions AND no pronouns.

uw¶

Run ID: uw
Participant: UWaterlooMDS
Track: Microblog
Year: 2012
Submission: 8/1/2012
Type: automatic
Task: filtering
MD5: f5a9b4545bf9d26fd120807e5eb13511
Run description: Logistic regression, with feedback for explicitly labeled docs in qrels -- unlabeled ignored.

uwatgclrbase¶

Run ID: uwatgclrbase
Participant: UWaterlooMDS
Track: Microblog
Year: 2012
Submission: 7/9/2012
Type: automatic
Task: adhoc
MD5: 74656467008a3662b461ada5ce43d47b
Run description: This run uses Gord Cormack's Logistic Regression classifier. For each topic, the classifier is trained on 1000 copies of the topic statement (as spam/positive/relevant) and 200 randomly chosen tweets before the query time (as ham/negative/non-relevant). The classifier than classifies everything prior to the query time and the 1000 highest scoring are returned.

uwatgclrman¶

Run ID: uwatgclrman
Participant: UWaterlooMDS
Track: Microblog
Year: 2012
Submission: 7/10/2012
Type: manual
Task: adhoc
MD5: c2b2c2cef753d7f26f83b3177655cbc5
Run description: This run first had a user manually query the corpus for tweets they would deem to be relevant to the topic (before the query time). Such tweets would then form the basis for the positive class training examples for Gord Cormack's logistic regression classifier. Where the negative training examples are selected at random from before the query time. For topics that the user could not find 20 tweets, 5 copies of the topic statement were trained upon for each tweet less than 20. Topics were trained in a 1:1 fashion of positive to negative tweets (except where the topic statement is used then it is 5:1).

uwatrrfall¶

Run ID: uwatrrfall
Participant: UWaterlooMDS
Track: Microblog
Year: 2012
Submission: 7/9/2012
Type: automatic
Task: adhoc
MD5: 16b1dbbba34766d9a6148078499fb621
Run description: Performs a set of baseline runs using Wumpus (5 different retrieval methods over 6 different sets of features) and Gord Cormack's Logistic Regression (trained with a 5:1 bias towards the topic statement). The top 20 from each run were then used as either training data or candidate documents for query expansion creating hundreds of sets of results. In addition, the default Okapi query expansion of Wumpus is performed (with 4 retrieval methods on 6 different sets of features) with two different language models based upon the corpus (e.g. all tweets before last year's oldest query time and all tweets before this year's oldest query time). All these runs are combined using Reciprocal Rank Fusion.

uwatrrflm¶

Run ID: uwatrrflm
Participant: UWaterlooMDS
Track: Microblog
Year: 2012
Submission: 7/9/2012
Type: automatic
Task: adhoc
MD5: be6ec10b83154317dbf35e6e03a2806e
Run description: The default Okapi query expansion of the Wumpus Search Engine is performed (with 4 retrieval methods on 6 different sets of features) with two different language models based upon the corpus (e.g. all tweets before last year's oldest query time and all tweets before this year's oldest query time). All these result sets are combined using Reciprocal Rank Fusion.

uwcmb12BL¶

Run ID: uwcmb12BL
Participant: waterloo
Track: Microblog
Year: 2012
Submission: 7/10/2012
Type: automatic
Task: adhoc
MD5: 68a9650141c8d1871388a5731f074b92
Run description: Base line using Okapi BM25 ranking.

uwcmb12CP¶

Run ID: uwcmb12CP
Participant: waterloo
Track: Microblog
Year: 2012
Submission: 7/10/2012
Type: automatic
Task: adhoc
MD5: 1b9d4d1ecd452971fef81ff0ba665572
Run description: Query expansion using terms from frequent patterns. Choosing terms by clustering patterns.

uwcmb12CT¶

Run ID: uwcmb12CT
Participant: waterloo
Track: Microblog
Year: 2012
Submission: 7/10/2012
Type: automatic
Task: adhoc
MD5: e87d515f86efbf3921e943a72516a8bc
Run description: Query expansion by terms picked by running the algorithm that picks terms from frequent patterns on the highest score results

uwcmb12NT¶

Run ID: uwcmb12NT
Participant: waterloo
Track: Microblog
Year: 2012
Submission: 7/10/2012
Type: automatic
Task: adhoc
MD5: e96f33a037f449c8d3444c8a32d253eb
Run description: Query expansion by terms from the most highly scored frequent patterns, consuming the patterns one by one.

uwn¶

Run ID: uwn
Participant: UWaterlooMDS
Track: Microblog
Year: 2012
Submission: 8/1/2012
Type: automatic
Task: filtering
MD5: b40f54269bb7dd17f49c4a43b2d68720
Run description: Logistic regression, with no feedback.

vsmv1¶

Run ID: vsmv1
Participant: unir_de
Track: Microblog
Year: 2012
Submission: 8/1/2012
Type: automatic
Task: filtering
MD5: 4d83bbaad46ea441ac5c4bf5d8cacb29
Run description: event based like e.g. okapiv1. comparable to the run vsmv2rel, but this is the baseline without relevance feedback. it used VSM. Threshold for relevance 0.5

vsmv2rel¶

Run ID: vsmv2rel
Participant: unir_de
Track: Microblog
Year: 2012
Submission: 7/31/2012
Type: automatic
Task: filtering
MD5: bfe86f2f58fe15f97129c113437ae2a7
Run description: Event based like e.g. run okapiv1, but used a vector space model with relevance feedback. relevance threshold 0.5

weka¶

Run ID: weka
Participant: SCIAITeam
Track: Microblog
Year: 2012
Submission: 7/29/2012
Type: automatic
Task: filtering
MD5: 30232820f360b1abe32726f02d910720
Run description: different types of emotes people use on twitter

window2run¶

Run ID: window2run
Participant: HIT_MTLAB
Track: Microblog
Year: 2012
Submission: 8/06/2012
Type: automatic
Task: filtering
Run description: In this run, we adopt windows measure with two most relevance tweets based on negative KL between query and tweet dynamically. In detail, during the filtering, the tweet will be judged to be true if tweet score is higher than any one tweet in our windows, otherwise to be judged false.Then we use online ranking logistic regression model to learn based on previous judge.

XMRUN1¶

Run ID: XMRUN1
Participant: XMU_PANCHAO
Track: Microblog
Year: 2012
Submission: 7/10/2012
Type: automatic
Task: adhoc
MD5: 0355efaf4916814e62883e0476b48e74
Run description: base run

XMRUN2¶

Run ID: XMRUN2
Participant: XMU_PANCHAO
Track: Microblog
Year: 2012
Submission: 7/10/2012
Type: automatic
Task: adhoc
MD5: 2c8ebf57317f1c0578b9996341699ac7
Run description: use a language model to improve the performance

XMRUN3¶

Run ID: XMRUN3
Participant: XMU_PANCHAO
Track: Microblog
Year: 2012
Submission: 7/10/2012
Type: automatic
Task: adhoc
MD5: 42f9cfee8b42d963d46277a20d5833b0
Run description: use a language model and query expansion to improve the performance

XMRUN4¶

Run ID: XMRUN4
Participant: XMU_PANCHAO
Track: Microblog
Year: 2012
Submission: 7/10/2012
Type: automatic
Task: adhoc
MD5: 02534398dec069c7ea01085ab81cd5f5
Run description: we focus on whether a tweet contains links, and whether the author is referenced in other tweets

YORK1¶

Run ID: YORK1
Participant: york
Track: Microblog
Year: 2012
Submission: 7/11/2012
Type: automatic
Task: adhoc
MD5: bb4ee800acda3befc9124fcaec60cab4
Run description: We use a weighted Rocchio's feedback model, in which the DFRee weighting model and the KL weighting model (doc=20 term=30 beta=1.4) for query expansion were used.

york12bd1i¶

Run ID: york12bd1i
Participant: york
Track: Microblog
Year: 2012
Submission: 8/1/2012
Type: automatic
Task: filtering
MD5: bc08e1c956e14c8645e07350cb6abec8
Run description: DFRee + KLFeeback20_30, top 50 are viewed as relevant "yes"

york12mb3¶

Run ID: york12mb3
Participant: york
Track: Microblog
Year: 2012
Submission: 7/11/2012
Type: automatic
Task: adhoc
MD5: 307528238daa95b631eb14da0c7fc0ec
Run description: We use a weighted Rocchio's feedback model, in which the DFRee weighting model and the KL weighting model (doc=20 term=30 beta=1) for query expansion were used. After that we conducted filtering according whether the tweet has links and hashtags.

york12mb4¶

Run ID: york12mb4
Participant: york
Track: Microblog
Year: 2012
Submission: 7/11/2012
Type: automatic
Task: adhoc
MD5: 4619a386df76ae68f5f0040750bf898f
Run description: We use an enhanced Rocchio's feedback model, in which the DFRee weighting model, the proximity model (weight=0.1 + FD + wins=8) and the KL weighting model (doc=20 term=30 beta=1) for query expansion were used.

YORK2¶

Run ID: YORK2
Participant: york
Track: Microblog
Year: 2012
Submission: 7/11/2012
Type: automatic
Task: adhoc
MD5: 53859269ab99a6dcf8ac3480e8c7912e
Run description: We use a weighted Rocchio's feedback model, in which the BM25 (b=0.3) weighting model and the KL weighting model (doc=20 term=30 beta=1.4) for query expansion were used.