Runs - Temporal Summarization 2015¶

Run ID: DMSL1NMF2
Participant: BJUT
Track: Temporal Summarization
Year: 2015
Submission: 8/28/2015
Task: ps
MD5: 4f7bec45f79352b0478e476eb7a3ed7d
Run description: In this runI adopt the improved Non-negative matrix factorization algorithm with two regularizations as the cluster method .One regularization considers both structure of manifold and structure of semantics the other regularization is L2-norm to control the complexity of model .Compared with APCLUSTER Affinity Propagation Clustering algorithm,in some topics ,our method is better or comparable in residual error.

DMSL1VSH3¶

Participants | Proceedings | Input | Summary

Run ID: DMSL1VSH3
Participant: BJUT
Track: Temporal Summarization
Year: 2015
Submission: 8/28/2015
Task: ps
MD5: df076bf1c94589886029593b62d2aeab
Run description: use VSHCLUSTER Vertex Substitution Heuristic method

DMSL2A1¶

Participants | Proceedings | Input | Summary

Run ID: DMSL2A1
Participant: BJUT
Track: Temporal Summarization
Year: 2015
Submission: 8/28/2015
Task: so
MD5: 1924ecee976980673ef86874a3639825
Run description: use Affinity Propagation (AP) method

DMSL2N2¶

Participants | Proceedings | Input | Summary

Run ID: DMSL2N2
Participant: BJUT
Track: Temporal Summarization
Year: 2015
Submission: 8/28/2015
Task: so
MD5: a631345e8b4d649c1f67e9a067247b94
Run description: In this runI adopt the improved Non-negative matrix factorization algorithm with two regularizations as the cluster method .One regularization considers both structure of manifold and structure of semantics the other regularization is L2-norm to control the complexity of model .Compared with APCLUSTER Affinity Propagation Clustering algorithm,in some topics ,our method is better or comparable in residual error.

DMSL2V3¶

Participants | Proceedings | Input | Summary

Run ID: DMSL2V3
Participant: BJUT
Track: Temporal Summarization
Year: 2015
Submission: 8/28/2015
Task: so
MD5: e25d2625d14a7e73df6fae51da00b0af
Run description: use VSHCLUSTER Vertex Substitution Heuristic method

docs¶

Participants | Proceedings | Input | Summary

Run ID: docs
Participant: CWI
Track: Temporal Summarization
Year: 2015
Submission: 9/3/2015
Task: fs
MD5: 4cf85fc4af5b4f909b758d9a7561face
Run description: clustering a stream of news articles with 3NN & cosine, matching clusters /w a embers containing all query terms

docsRecall¶

Participants | Proceedings | Input | Summary

Run ID: docsRecall
Participant: CWI
Track: Temporal Summarization
Year: 2015
Submission: 9/3/2015
Task: fs
MD5: f972746368a542cbbb7a5889feb471e0
Run description: clustering a stream of news articles with 3NN & cosine, length <= 30 && gain >= 0.3 to increase recall

FS1A¶

Participants | Proceedings | Input | Summary

Run ID: FS1A
Participant: IRIT
Track: Temporal Summarization
Year: 2015
Submission: 9/3/2015
Task: ps
MD5: 0029e8a8fc9b792cfeabda6718f66ef3
Run description: 3-steps based approach for filtering and summarization. Iteratively, in each hour, first we select top relevant documents using BM25 model provided by the orignal query, then we select relevant sentences based on the presence and the proximity of query terms in the sentence, and finally, we detect the novelty by combining two features : text divergence and the detection of new entites using an AND operator.

FS1B¶

Participants | Proceedings | Input | Summary

Run ID: FS1B
Participant: IRIT
Track: Temporal Summarization
Year: 2015
Submission: 9/3/2015
Task: ps
MD5: 2805723f9d386f1f8a995aef2c77bfd1
Run description: We used a method that is based on a fusion method that is applied at the term query level. This method assume that every query term is assumed as a query per se. Then, the scores of each ranking is fused with the other ranking from the same query.

FS2A¶

Participants | Proceedings | Input | Summary

Run ID: FS2A
Participant: IRIT
Track: Temporal Summarization
Year: 2015
Submission: 9/3/2015
Task: ps
MD5: e616df299b27dcdd618dfeee1225e1ef
Run description: 3-steps based approach for filtering and summarization. Iteratively, in each hour, first we select top relevant documents using BM25 model provided by the orignal query, then we select relevant sentences based on the presence and the proximity of query terms in the sentence, and finally, we detect the novelty by combining two features : text divergence and the detection of new entites using an OR operator.

FS2B¶

Participants | Proceedings | Input | Summary

Run ID: FS2B
Participant: IRIT
Track: Temporal Summarization
Year: 2015
Submission: 9/3/2015
Task: ps
MD5: f473008e45a4bc867e96ecffb27de7ac
Run description: This run is based on a temporal language model.

FS3A¶

Participants | Proceedings | Input | Summary

Run ID: FS3A
Participant: IRIT
Track: Temporal Summarization
Year: 2015
Submission: 9/3/2015
Task: ps
MD5: d3b9886349bce26048eb4416aa1e6af1
Run description: 3-steps based approach for filtering and summarization. Iteratively, in each hour, first we select top relevant documents using BM25 model provided by the orignal query, then we select relevant sentences based on the presence and the proximity of query terms in the sentence, and finally, we detect the novelty based only on the text divergence

FS3B¶

Participants | Proceedings | Input | Summary

Run ID: FS3B
Participant: IRIT
Track: Temporal Summarization
Year: 2015
Submission: 9/3/2015
Task: ps
MD5: 889fd26b7c69042426d319801b243d8b
Run description: We used a method that is based on a fusion method that is applied at the term query level. This method assume that every query term is assumed as a query per se. Then, the scores of each ranking is fused with the other ranking from the same query.

FS4A¶

Participants | Proceedings | Input | Summary

Run ID: FS4A
Participant: IRIT
Track: Temporal Summarization
Year: 2015
Submission: 9/3/2015
Task: ps
MD5: 6120c0fd13d9a81a2966ae06599ef398
Run description: FS4A: 3-steps based approach for filtering and summarization. Iteratively, in each hour, first we select top relevant documents using BM25 model provided by an extended query, then we select relevant sentences based on the presence and the proximity of query terms in the sentence, and finally, we detect the novelty by combining two features : text divergence and the detection of new entites using an AND operator.

FS4B¶

Participants | Proceedings | Input | Summary

Run ID: FS4B
Participant: IRIT
Track: Temporal Summarization
Year: 2015
Submission: 9/3/2015
Task: ps
MD5: 88f0f0dc5b6e9bb75546aefdbb1f30c8
Run description: This run is also based on a temporal language model.

FS5A¶

Participants | Proceedings | Input | Summary

Run ID: FS5A
Participant: IRIT
Track: Temporal Summarization
Year: 2015
Submission: 9/3/2015
Task: ps
MD5: 88b95270a99067e18933b722b130488b
Run description: 3-steps based approach for filtering and summarization. Iteratively, in each hour, first we select top relevant documents using BM25 model provided by an extended query, then we select relevant sentences based on the presence and the proximity of query terms in the sentence, and finally, we detect the novelty by combining two features : text divergence and the detection of new entites using an OR operator.

FS6A¶

Participants | Proceedings | Input | Summary

Run ID: FS6A
Participant: IRIT
Track: Temporal Summarization
Year: 2015
Submission: 9/3/2015
Task: ps
MD5: 70f7422933b052e4f1a64148004b7915
Run description: 3-steps based approach for filtering and summarization. Iteratively, in each hour, first we select top relevant documents using BM25 model provided by an extended query, then we select relevant sentences based on the presence and the proximity of query terms in the sentence, and finally, , we detect the novelty based only on the text divergence

IGn¶

Participants | Proceedings | Input | Summary

Run ID: IGn
Participant: CWI
Track: Temporal Summarization
Year: 2015
Submission: 9/3/2015
Task: fs
MD5: 07893107f5f2aebbd988a1b5f669c28d
Run description: clustering a stream of news articles with 3NN & normalized IG

IGnPrecision¶

Participants | Proceedings | Input | Summary

Run ID: IGnPrecision
Participant: CWI
Track: Temporal Summarization
Year: 2015
Submission: 9/3/2015
Task: fs
MD5: 6afa7c0ba7f71f620c9602488c498c9a
Run description: clustering a stream of news articles with 3NN & normalized IG, half hour window, gain .5 top-1.

IGnRecall¶

Participants | Proceedings | Input | Summary

Run ID: IGnRecall
Participant: CWI
Track: Temporal Summarization
Year: 2015
Submission: 9/3/2015
Task: fs
MD5: 80f97fae18c80d6271f41c213416d4f0
Run description: clustering a stream of news articles with 3NN & normalized IG, length <= 30 && gain >= 0.3 to increase recall

InL2DecrQE1ID1¶

Participants | Proceedings | Input | Summary

Run ID: InL2DecrQE1ID1
Participant: USI
Track: Temporal Summarization
Year: 2015
Submission: 9/3/2015
Task: ps
MD5: 88ad9165e181a910247b8207ab3feb9e
Run description: We combined relevance and novelty to calculate the score of each sentence. We used Divergence From Randomness (DFR) Framework and particular InL2 to calculate the relevance of a sentence given an event. The novelty score is based on the number of novel terms that each sentence has compared to the terms of the summary that is already produced. The number of sentences selected each hour decreases as time passes. The first sentences of a summary should contain all the query terms. On the rest of the blocks we used query expansion by expanding the query with the top 5 most appeared terms of the summary already produced.

InL2DecrQE2ID2¶

Participants | Proceedings | Input | Summary

Run ID: InL2DecrQE2ID2
Participant: USI
Track: Temporal Summarization
Year: 2015
Submission: 9/3/2015
Task: ps
MD5: 5493bba0dc3b39a3d5238d3bb75e3cc8
Run description: We combined relevance and novelty to calculate the score of each sentence. We used Divergence From Randomness (DFR) Framework and particular InL2 to calculate the relevance of a sentence given an event. The novelty score is based on the number of novel terms that each sentence has compared to the terms of the summary that is already produced. The number of sentences selected each hour decreases as time passes. The first sentences of a summary may contain any of the query terms. On the rest of the blocks we used query expansion by expanding the query with the top 5 most appeared terms of the summary already produced.

InL2DocsQE2ID5¶

Participants | Proceedings | Input | Summary

Run ID: InL2DocsQE2ID5
Participant: USI
Track: Temporal Summarization
Year: 2015
Submission: 9/3/2015
Task: ps
MD5: d783d4661edd36404ed686f7bc717e54
Run description: We combined relevance and novelty to calculate the score of each sentence. We used Divergence From Randomness (DFR) Framework and particular InL2 to calculate the relevance of a sentence given an event. The novelty score is based on the number of novel terms that each sentence has compared to the terms of the summary that is already produced. The number of sentences selected each hour is the same for all the blocks (5 sentences up to 1000). The first sentences of a summary may contain any of the query terms. On the rest of the blocks we used query expansion by expanding the query with the top 5 most appeared terms of the summary already produced.

InL2IncrQE1ID7¶

Participants | Proceedings | Input | Summary

Run ID: InL2IncrQE1ID7
Participant: USI
Track: Temporal Summarization
Year: 2015
Submission: 9/3/2015
Task: ps
MD5: 1f85f8e1fec33be3eac76baee91b4141
Run description: We combined relevance and novelty to calculate the score of each sentence. We used Divergence From Randomness (DFR) Framework and particular InL2 to calculate the relevance of a sentence given an event. The novelty score is based on the number of novel terms that each sentence has compared to the terms of the summary that is already produced. The number of sentences selected each hour increases as time passes. The first sentences of a summary should contain all the query terms. On the rest of the blocks we used query expansion by expanding the query with the top 5 most appeared terms of the summary already produced.

InL2IncrQE2ID4¶

Participants | Proceedings | Input | Summary

Run ID: InL2IncrQE2ID4
Participant: USI
Track: Temporal Summarization
Year: 2015
Submission: 9/3/2015
Task: ps
MD5: 7addecfb95e70dbec61f406fceffd1f6
Run description: We combined relevance and novelty to calculate the score of each sentence. We used Divergence From Randomness (DFR) Framework and particular InL2 to calculate the relevance of a sentence given an event. The novelty score is based on the number of novel terms that each sentence has compared to the terms of the summary that is already produced. The number of sentences selected each hour increases as time passes. The first sentences of a summary may contain any of the query terms. On the rest of the blocks we used query expansion by expanding the query with the top 5 most appeared terms of the summary already produced.

InL2StabQE1ID6¶

Participants | Proceedings | Input | Summary

Run ID: InL2StabQE1ID6
Participant: USI
Track: Temporal Summarization
Year: 2015
Submission: 9/3/2015
Task: ps
MD5: 723f0b9c8ce76f174c9f713a4d28c558
Run description: We combined relevance and novelty to calculate the score of each sentence. We used Divergence From Randomness (DFR) Framework and particular InL2 to calculate the relevance of a sentence given an event. The novelty score is based on the number of novel terms that each sentence has compared to the terms of the summary that is already produced. The number of sentences selected each hour is the same and depends on the number of time blocks per event. The first sentences of a summary should contain all the query terms. On the rest of the blocks we used query expansion by expanding the query with the top 5 most appeared terms of the summary already produced.

InL2StabQE2ID3¶

Participants | Proceedings | Input | Summary

Run ID: InL2StabQE2ID3
Participant: USI
Track: Temporal Summarization
Year: 2015
Submission: 9/3/2015
Task: ps
MD5: b8b0d3be772b32fa0ecee36e4ab64a40
Run description: We combined relevance and novelty to calculate the score of each sentence. We used Divergence From Randomness (DFR) Framework and particular InL2 to calculate the relevance of a sentence given an event. The novelty score is based on the number of novel terms that each sentence has compared to the terms of the summary that is already produced. The number of sentences selected each hour is the same for every block and depends on the number of time blocks per event. The first sentences of a summary may contain any of the query terms. On the rest of the blocks we used query expansion by expanding the query with the top 5 most appeared terms of the summary already produced.

l3sattrec15run1¶

Participants | Input | Summary

Run ID: l3sattrec15run1
Participant: l3sattrec15
Track: Temporal Summarization
Year: 2015
Submission: 9/2/2015
Task: ps
MD5: 422f94206b74e94f8136d6c5727726f2
Run description: To select update sentences for each hour of the event we adopt the following approach: First we retrieve the top-m documents for the query to account for prevalence, then we select the top-q summary-worthy sentences using a learning to rank model, next we remove redundancy using minhash based clustering and finally we filter by information gain to select novel updates. We used DUC 2007 corpus to train the document summarisation algorithm.

l3sattrec15run2¶

Participants | Input | Summary

Run ID: l3sattrec15run2
Participant: l3sattrec15
Track: Temporal Summarization
Year: 2015
Submission: 9/4/2015
Task: ps
MD5: edd4d6c1eded7375e57bb558eab59570
Run description: To select update sentences for each hour of the event we adopt the following approach: First we retrieve the top-m documents for the query to account for prevalence, then we select the top-q summary-worthy sentences using a learning to rank model, next we remove redundancy using minhash based clustering and finally we filter by information gain to select novel updates. We use DUC 2007 to train the multi-document summarizer.

l3sattrec15run3¶

Participants | Input | Summary

Run ID: l3sattrec15run3
Participant: l3sattrec15
Track: Temporal Summarization
Year: 2015
Submission: 9/4/2015
Task: ps
MD5: 08ffd7d3283e1d7d28afc25ba9be49f0
Run description: To select update sentences for each hour of the event we adopt the following approach: First we retrieve the top-m documents for the query to account for prevalence, then we select the top-q summary-worthy sentences using a learning to rank model, next we remove redundancy using minhash based clustering and finally we filter by information gain to select novel updates. We use DUC 2007 to train the multi-document summarizer.

LDA¶

Participants | Proceedings | Input | Summary

Run ID: LDA
Participant: UvA.ILPS
Track: Temporal Summarization
Year: 2015
Submission: 9/3/2015
Task: so
MD5: 7ee3fa04c199cc4508be0a5206222975
Run description: LDA

LDAv2¶

Participants | Proceedings | Input | Summary

Run ID: LDAv2
Participant: UvA.ILPS
Track: Temporal Summarization
Year: 2015
Submission: 9/4/2015
Task: so
MD5: 52a10fe5fa7fa2780c4806f4943e4ca7
Run description: Latent Dirichlet Allocation probabilistic ranking variation

LexRank¶

Participants | Proceedings | Input | Summary

Run ID: LexRank
Participant: UvA.ILPS
Track: Temporal Summarization
Year: 2015
Submission: 9/3/2015
Task: so
MD5: 35fde70fefc5072fab13a369342e3554
Run description: LexRank summarization

LLR¶

Participants | Proceedings | Input | Summary

Run ID: LLR
Participant: UvA.ILPS
Track: Temporal Summarization
Year: 2015
Submission: 9/3/2015
Task: so
MD5: 8956ac65e9fbdf29ec240fe2cfb5ce70
Run description: log likelihood ratio

LM¶

Participants | Proceedings | Input | Summary

Run ID: LM
Participant: UvA.ILPS
Track: Temporal Summarization
Year: 2015
Submission: 9/4/2015
Task: so
MD5: 092134788735ee9006fd77e69d93f9ce
Run description: Language Modeling

OS1A¶

Participants | Proceedings | Input | Summary

Run ID: OS1A
Participant: IRIT
Track: Temporal Summarization
Year: 2015
Submission: 9/3/2015
Task: so
MD5: 394e3a674d368a590720a72735a4a13c
Run description: An approach for summarization. Iteratively, in each hour, first we top documents annotated as relevant, then we select relevant sentences based on the presence and the proximity of query terms in the sentence, and finally, we detect the novelty by combining two features : text divergence and the detection of new entites using an AND operator.

OS1C¶

Participants | Proceedings | Input | Summary

Run ID: OS1C
Participant: IRIT
Track: Temporal Summarization
Year: 2015
Submission: 9/3/2015
Task: so
MD5: 9426568084fe9dc9283ac54345c21caf
Run description: This run is based on a real time selective summarization approach. The aim was to make a decision on whether the given sentence will be added to the summary or not on real time way without the need of buffering. For a given sentence, the novelty degree and redundancy score are assessed. Only sentences that get a novelty degree and redundancy score above a threshold are added to the summary. For this run the following parameters were adopted: 1- The thresholds were fixed to 0.27 and 3 for the novelty degree and redundancy score respectively. 2- Linear smoothing was used in the estimation of the redundancy score; 3- A decay function was taken into account when computing the novelty degree

OS2A¶

Participants | Proceedings | Input | Summary

Run ID: OS2A
Participant: IRIT
Track: Temporal Summarization
Year: 2015
Submission: 9/3/2015
Task: so
MD5: 572a0a6b30a8c909cfe84921a58c0490
Run description: An approach for summarization. Iteratively, in each hour, first we top documents annotated as relevant, then we select relevant sentences based on the presence and the proximity of query terms in the sentence, and finally, we detect the novelty by combining two features : text divergence and the detection of new entites using an OR operator.

OS2C¶

Participants | Proceedings | Input | Summary

Run ID: OS2C
Participant: IRIT
Track: Temporal Summarization
Year: 2015
Submission: 9/3/2015
Task: so
MD5: dac9ae33242537d44f31c713d0f96696
Run description: We use the same approach as our run OS1C expect that for this run the following parameters were adopted: 1- The thresholds of the novelty degree and redundancy score were set to the average of the values observed during last time window. 2- Linear smoothing was used in the estimation of the redundancy score; 3- A decay function was taken into account when computing the novelty degree

OS3A¶

Participants | Proceedings | Input | Summary

Run ID: OS3A
Participant: IRIT
Track: Temporal Summarization
Year: 2015
Submission: 9/3/2015
Task: so
MD5: 71bc481b2a457b958851d8ea2b4b9cfd
Run description: An approach for summarization. Iteratively, in each hour, first we top documents annotated as relevant, then we select relevant sentences based on the presence and the proximity of query terms in the sentence, and finally, we detect the novelty based only on the text divergence

OS3C¶

Participants | Proceedings | Input | Summary

Run ID: OS3C
Participant: IRIT
Track: Temporal Summarization
Year: 2015
Submission: 9/3/2015
Task: so
MD5: 4e4967dfb052deff18dd3cd5864676fe
Run description: We use the same approach as our run OS1C expect that for this run the following parameters were adopted: 1-The thresholds were fixed to 0.27 and 3 for the novelty degree and redundancy score respectively. 2- Dirichlet smoothing was used in the estimation of the redundancy score; 3- A decay function was taken into account when computing the novelty degree

OS4C¶

Participants | Proceedings | Input | Summary

Run ID: OS4C
Participant: IRIT
Track: Temporal Summarization
Year: 2015
Submission: 9/3/2015
Task: so
MD5: 95f21952b3cc9026774999136a857aee
Run description: We use the same approach as our run OS1C expect that for this run the following parameters were adopted: 1-The thresholds of the novelty degree and redundancy score were set to the average of the values observed during last time window; 2- Dirichlet smoothing was used in the estimation of the redundancy score; 3- A decay function was taken into account when computing the novelty degree

OS5C¶

Participants | Proceedings | Input | Summary

Run ID: OS5C
Participant: IRIT
Track: Temporal Summarization
Year: 2015
Submission: 9/3/2015
Task: so
MD5: 00b8d560fc1ab60436352c67cd591948
Run description: We use the same approach as our run OS1C expect that for this run the following parameters were adopted: 1-The thresholds were fixed to 0.27 and 3 for the novelty degree and redundancy score respectively. 2- Dirichlet smoothing was used in the estimation of the redundancy score; 3- A decay function was not taken into account when computing the novelty degree

OS6C¶

Participants | Proceedings | Input | Summary

Run ID: OS6C
Participant: IRIT
Track: Temporal Summarization
Year: 2015
Submission: 9/3/2015
Task: so
MD5: dbdd7a441d2179eae3e2378b16d3592e
Run description: We use the same approach as our run OS1C expect that for this run the following parameters were adopted: 1-The thresholds of the novelty degree and redundancy score were set to the average of the values observed during last time window; 2-Dirichlet smoothing was used in the estimation of the redundancy score; 3-A decay function was not taken into account when computing the novelty degree

OS7C¶

Participants | Proceedings | Input | Summary

Run ID: OS7C
Participant: IRIT
Track: Temporal Summarization
Year: 2015
Submission: 9/3/2015
Task: so
MD5: ad098bd56c6a5e2ec082457d6925d488
Run description: We use the same approach as our run OS1C expect that for this run the following parameters were adopted: 1-The thresholds were fixed to 0.27 and 3 for the novelty degree and redundancy score respectively. 2- Linear smoothing was used in the estimation of the redundancy score; 3- A decay function was not taken into account when computing the novelty degree

OS8C¶

Participants | Proceedings | Input | Summary

Run ID: OS8C
Participant: IRIT
Track: Temporal Summarization
Year: 2015
Submission: 9/3/2015
Task: so
MD5: 915e72cfdce61028e85fb8eb7fc14981
Run description: We use the same approach as our run OS1C expect that for this run the following parameters were adopted: 1-The thresholds of the novelty degree and redundancy score were set to the average of the values observed during last time window. 2-Linear smoothing was used in the estimation of the redundancy score; 3-A decay function was taken into account when computing the novelty degree

ProfOnly3¶

Participants | Proceedings | Input | Summary

Run ID: ProfOnly3
Participant: udel_fang
Track: Temporal Summarization
Year: 2015
Submission: 9/4/2015
Task: so
MD5: 4ef4be7c9860d16b9cdbce184e9ad7c5
Run description: This run focuses on building a rich query representation by using external resources, such as auxiliary corpora that were generated before the query start time. Documents were processed in batch and decisions are made at the end of each hour. Sentences within the documents then are ranked and selected.

ProfOnlyFS3¶

Participants | Proceedings | Input | Summary

Run ID: ProfOnlyFS3
Participant: udel_fang
Track: Temporal Summarization
Year: 2015
Submission: 9/4/2015
Task: ps
MD5: c303f6fd2c8e531d88e86ed77c4c4218
Run description: This run focuses on building a rich query representation by using external resources, such as auxiliary corpora that were generated before the query start time. Documents were processed in batch and decisions are made at the end of each hour. Sentences within the documents then are ranked and selected.

QL¶

Participants | Proceedings | Input | Summary

Run ID: QL
Participant: UvA.ILPS
Track: Temporal Summarization
Year: 2015
Submission: 9/3/2015
Task: so
MD5: d90488807657784dc1d0d232432b6356
Run description: query likelihood

QLF¶

Participants | Proceedings | Input | Summary

Run ID: QLF
Participant: UvA.ILPS
Track: Temporal Summarization
Year: 2015
Submission: 9/4/2015
Task: so
MD5: edd46204f56775480f31b238f878ab45
Run description: Query likelihood with a higher threshold for sentence selection

QLLP¶

Participants | Proceedings | Input | Summary

Run ID: QLLP
Participant: UvA.ILPS
Track: Temporal Summarization
Year: 2015
Submission: 9/4/2015
Task: so
MD5: d63241cc40195f831ee0853d37b4637e
Run description: Query likelihood with smoothing

Run1¶

Participants | Proceedings | Input | Summary

Run ID: Run1
Participant: AIPHES
Track: Temporal Summarization
Year: 2015
Submission: 9/3/2015
Task: so
MD5: 1831f1cfd3cbe1743403f4ec0886589a
Run description: This run uses a sequential clustering approach with at most 1000 clusters, min cluster score of 3, 'full token discount', and strict boilerplate removal.

Run2¶

Participants | Proceedings | Input | Summary

Run ID: Run2
Participant: AIPHES
Track: Temporal Summarization
Year: 2015
Submission: 9/3/2015
Task: so
MD5: a5849698eddfdc23d6271cbb25aed853
Run description: This run uses a sequential clustering approach with at most 1000 clusters, min cluster score of 1, 'full token discount', and strict boilerplate removal.

Run3¶

Participants | Proceedings | Input | Summary

Run ID: Run3
Participant: AIPHES
Track: Temporal Summarization
Year: 2015
Submission: 9/4/2015
Task: so
MD5: 25d733b2a9287daa3b6eb9e59a944f65
Run description: This run uses a sequential clustering approach with at most 1000 clusters, min cluster score of 1, reduced token discount, and less strict boilerplate removal.

Run4¶

Participants | Proceedings | Input | Summary

Run ID: Run4
Participant: AIPHES
Track: Temporal Summarization
Year: 2015
Submission: 9/4/2015
Task: so
MD5: 65bf003f1e4b9be90a2eb38352afd359
Run description: This run uses a sequential clustering approach with at most 100 clusters, min cluster score of 1, reduced token discount, and less strict boilerplate removal.

runvec1¶

Participants | Proceedings | Input | Summary

Run ID: runvec1
Participant: ISCASIR
Track: Temporal Summarization
Year: 2015
Submission: 9/3/2015
Task: so
MD5: bd1d4d28ef15fabf3c8b060893bba0e6
Run description: This run is supported by the word2vec technique.

runvec2¶

Participants | Proceedings | Input | Summary

Run ID: runvec2
Participant: ISCASIR
Track: Temporal Summarization
Year: 2015
Submission: 9/4/2015
Task: so
MD5: 07734b934002007cf117c70b53663e37
Run description: No query expanding and supported by word2vec.

TF¶

Participants | Proceedings | Input | Summary

Run ID: TF
Participant: UvA.ILPS
Track: Temporal Summarization
Year: 2015
Submission: 9/3/2015
Task: so
MD5: a2483c055ddae080f16586e50ea13456
Run description: query term frequency

TFFilter¶

Participants | Proceedings | Input | Summary

Run ID: TFFilter
Participant: UvA.ILPS
Track: Temporal Summarization
Year: 2015
Submission: 9/4/2015
Task: ps
MD5: f3aa16e85ec9f4788b8b5ed3c8ef662b
Run description: Filtering documents by query term frequency

TFISF¶

Participants | Proceedings | Input | Summary

Run ID: TFISF
Participant: UvA.ILPS
Track: Temporal Summarization
Year: 2015
Submission: 9/3/2015
Task: so
MD5: 748b4ad97e4878bc2df63d1d8fd6bc41
Run description: tfisf sentence ranking

TFISFW¶

Participants | Proceedings | Input | Summary

Run ID: TFISFW
Participant: UvA.ILPS
Track: Temporal Summarization
Year: 2015
Submission: 9/4/2015
Task: so
MD5: c642e36826ee7b87254594350b2cacad
Run description: TFISF with Wordnet query expansion

TFISFW2V¶

Participants | Proceedings | Input | Summary

Run ID: TFISFW2V
Participant: UvA.ILPS
Track: Temporal Summarization
Year: 2015
Submission: 9/4/2015
Task: so
MD5: 13d8253051066ea6be3392bcb8ad868c
Run description: TF.ISF sentence ranking with Word2Vec query expansion

TFW¶

Participants | Proceedings | Input | Summary

Run ID: TFW
Participant: UvA.ILPS
Track: Temporal Summarization
Year: 2015
Submission: 9/4/2015
Task: so
MD5: 2e8473ab4a01aa7de642f3011c12f676
Run description: Term Frequency with Wordnet query expansion

TFW2V¶

Participants | Proceedings | Input | Summary

Run ID: TFW2V
Participant: UvA.ILPS
Track: Temporal Summarization
Year: 2015
Submission: 9/4/2015
Task: so
MD5: 28ebe70195ec92f7d24466c109e2cb32
Run description: Term Frequency Word2Vec

titles¶

Participants | Proceedings | Input | Summary

Run ID: titles
Participant: CWI
Track: Temporal Summarization
Year: 2015
Submission: 9/3/2015
Task: fs
MD5: edd5198dfc485da241505ba5cbcf47fb
Run description: clustering a stream of news articles with 3NN & cosine, matching clusters /w a member containing all query terms in the title

uogTrdEEQR3¶

Participants | Proceedings | Input | Summary

Run ID: uogTrdEEQR3
Participant: uogTr
Track: Temporal Summarization
Year: 2015
Submission: 9/4/2015
Task: so
MD5: 6d2ce79d679fd04e92d3ef01c706cfa3
Run description: R3 -- Entity-focused run, iterating over the corpus document-by-document, scoring sentences similar to query) using entity-entity interaction feature, selecting top-k updates.

uogTrdEQR1¶

Participants | Proceedings | Input | Summary

Run ID: uogTrdEQR1
Participant: uogTr
Track: Temporal Summarization
Year: 2015
Submission: 9/4/2015
Task: so
MD5: 221a9065212b63de46231e8e51dfef75
Run description: R1 -- Entity-focused run, iterating over the corpus document-by-document, scoring sentences (similar to query) using entity importance feature, selecting top-k updates.

uogTrdSqCR5¶

Participants | Proceedings | Input | Summary

Run ID: uogTrdSqCR5
Participant: uogTr
Track: Temporal Summarization
Year: 2015
Submission: 9/4/2015
Task: so
MD5: d4d885024f79c6c0b7e6704e72f6b053
Run description: R5 -- Baseline run, iterating over the corpus document-by-document, ranking sentences by cosine similarity to query, selecting updates by cosine similarity threshold.

uogTrhEEQR4¶

Participants | Proceedings | Input | Summary

Run ID: uogTrhEEQR4
Participant: uogTr
Track: Temporal Summarization
Year: 2015
Submission: 9/4/2015
Task: so
MD5: dbbb104b9b7ea47935b3e9f1e2b78fbe
Run description: R4 -- Entity-focused run, iterating over the corpus in hour-by-hour batches, scoring sentences similar to query) using entity-entity interaction feature, selecting top-k updates.

uogTrhEQR2¶

Participants | Proceedings | Input | Summary

Run ID: uogTrhEQR2
Participant: uogTr
Track: Temporal Summarization
Year: 2015
Submission: 9/4/2015
Task: so
MD5: 66be7981b85474a597351069971c80e4
Run description: R2 -- Entity-focused run, iterating over the corpus in hour-by-hour batches, scoring sentences (similar to query) using entity importance feature, selecting top-k updates.

uogTrhSqCR6¶

Participants | Proceedings | Input | Summary

Run ID: uogTrhSqCR6
Participant: uogTr
Track: Temporal Summarization
Year: 2015
Submission: 9/4/2015
Task: so
MD5: d6e328502492342e028de7c3796ad1ce
Run description: R6 -- Baseline run, iterating over the corpus in hour-by-hour batches, ranking sentences by cosine similarity to query, selecting updates by cosine similarity threshold.

uogTrT1MANU¶

Participants | Proceedings | Input | Summary

Run ID: uogTrT1MANU
Participant: uogTr
Track: Temporal Summarization
Year: 2015
Submission: 9/3/2015
Task: ps
MD5: 059f0be1b1a3364660bd651dcddfc312
Run description: This is a MANUAL run. Sentences were selected by a human assessor. Note that this is a valid run for all tasks (1/2/3).

uogTrT1X2cSCP¶

Participants | Proceedings | Input | Summary

Run ID: uogTrT1X2cSCP
Participant: uogTr
Track: Temporal Summarization
Year: 2015
Submission: 9/4/2015
Task: ps
MD5: e492915cba4b55b67c5ff897bab9985f
Run description: This run performs explicit diversification of the event updates based on a pre-defined taxonomy of important intents to cover. The intents were proposed by crowd workers given the event types. Score boosting based on proximity to likely relevant sentences is also applied.

uogTrT1X2iNCP¶

Participants | Proceedings | Input | Summary

Run ID: uogTrT1X2iNCP
Participant: uogTr
Track: Temporal Summarization
Year: 2015
Submission: 9/4/2015
Task: ps
MD5: 440ab380c30f78c83a3db76ce4b79c77
Run description: This run performs explicit diversification of the event updates based on a pre-defined taxonomy of important intents to cover. The intents were automatically extracted from Wikipedia infoboxes from past events of the same type. Intents were supplemented with topics identified over time from an aligned News stream. Score boosting based on proximity to likely relevant sentences is also applied. External Resources: Wikipedia pages for events predating the KBA corpus. (News Stream is the KBA corpus)

uogTrT1X2iSC¶

Participants | Proceedings | Input | Summary

Run ID: uogTrT1X2iSC
Participant: uogTr
Track: Temporal Summarization
Year: 2015
Submission: 9/4/2015
Task: ps
MD5: fd95ca78a5fa149a4c0b59300bd86033
Run description: This run performs explicit diversification of the event updates based on a pre-defined taxonomy of important intents to cover. The intents were automatically extracted from Wikipedia infoboxes from past events of the same type. External Resources: Wikipedia pages for events predating the KBA corpus.

uogTrT1X2iSCP¶

Participants | Proceedings | Input | Summary

Run ID: uogTrT1X2iSCP
Participant: uogTr
Track: Temporal Summarization
Year: 2015
Submission: 9/3/2015
Task: ps
MD5: 35ef3624abc14b0c94644a5d4f09b80f
Run description: This run performs explicit diversification of the event updates based on a pre-defined taxonomy of important intents to cover. The intents were automatically extracted from Wikipedia infoboxes from past events of the same type. Score boosting based on proximity to likely relevant sentences is also applied. Exzternal Resources: Wikipedia pages for events predating the KBA corpus.

uogTrT1X2iTCP¶

Participants | Proceedings | Input | Summary

Run ID: uogTrT1X2iTCP
Participant: uogTr
Track: Temporal Summarization
Year: 2015
Submission: 9/4/2015
Task: ps
MD5: 9e7f900efcc50b290491098712944ee2
Run description: This run performs explicit diversification of the event updates based on a pre-defined taxonomy of important intents to cover. The intents were automatically extracted from Wikipedia infoboxes from past events of the same type. Intents were supplemented with topics identified over time from an aligned Twitter stream. Score boosting based on proximity to likely relevant sentences is also applied. External Resources: Wikipedia pages for events predating the KBA corpus. Aligned Twitter Stream

uogTrT2EimpP¶

Participants | Proceedings | Input | Summary

Run ID: uogTrT2EimpP
Participant: uogTr
Track: Temporal Summarization
Year: 2015
Submission: 9/4/2015
Task: so
MD5: 660c8ddc9c286d81f93bde78378475c6
Run description: Relevance + Emtity importance scoring of sentences Entities are extracted from old news corpora.

uogTrT2EintP¶

Participants | Proceedings | Input | Summary

Run ID: uogTrT2EintP
Participant: uogTr
Track: Temporal Summarization
Year: 2015
Submission: 9/4/2015
Task: so
MD5: fbc2a27297a54884aa88259f49fac4d9
Run description: Relevance + Entity Interaction scoring of sentences Entities are extracted from old news corpora.

UWCTSRun1¶

Participants | Proceedings | Input | Summary

Run ID: UWCTSRun1
Participant: WaterlooClarke
Track: Temporal Summarization
Year: 2015
Submission: 8/24/2015
Task: ps
MD5: a3550b746fbe198e9ffafd5158c374d5
Run description: UWCTSRun1 follows the strategy of pushing the first sentence found in each article as determined natively by python-goose.

UWCTSRun2¶

Participants | Proceedings | Input | Summary

Run ID: UWCTSRun2
Participant: WaterlooClarke
Track: Temporal Summarization
Year: 2015
Submission: 8/24/2015
Task: ps
MD5: 2c0ed563ba8a2d91ecea9d1b4a10d71e
Run description: UWCTSRun2 follows the strategy of pushing only the document headline for each of the selected documents.

UWCTSRun3¶

Participants | Proceedings | Input | Summary

Run ID: UWCTSRun3
Participant: WaterlooClarke
Track: Temporal Summarization
Year: 2015
Submission: 8/24/2015
Task: ps
MD5: 255c98ce193081a385f8935424751a64
Run description: UWCTSRun3 follows the strategy of pushing the first sentence found in each article determined using a combination of python-readability and python-goose. Python-readability parses the clean HTML and only keeps the HTML containing the main article. Then python-goose is used to extract sentences from the HTML.

UWCTSRun4¶

Participants | Proceedings | Input | Summary

Run ID: UWCTSRun4
Participant: WaterlooClarke
Track: Temporal Summarization
Year: 2015
Submission: 8/25/2015
Task: so
MD5: 09968fdd77c4be1a8778297b10297150
Run description: UWCTSRun4 follows the strategy of pushing the first sentence found in each article as determined natively by python-goose.

UWCTSRun5¶

Participants | Proceedings | Input | Summary

Run ID: UWCTSRun5
Participant: WaterlooClarke
Track: Temporal Summarization
Year: 2015
Submission: 8/25/2015
Task: so
MD5: ef67270e9c83c97d558dc79b168437a9
Run description: UWCTSRun5 follows the strategy of pushing only the document headline for each of the selected documents.

UWCTSRun6¶

Participants | Proceedings | Input | Summary

Run ID: UWCTSRun6
Participant: WaterlooClarke
Track: Temporal Summarization
Year: 2015
Submission: 8/25/2015
Task: so
MD5: bf5721eaaefe862eaf7e5d9ff11820e8
Run description: UWCTSRun6 follows the strategy of pushing the first sentence found in each article determined using a combination of python-readability and python-goose. Python-readability parses the clean HTML and only keeps the HTML containing the main article. Then python-goose is used to extract sentences from the HTML.

WikiOnly2¶

Participants | Proceedings | Input | Summary

Run ID: WikiOnly2
Participant: udel_fang
Track: Temporal Summarization
Year: 2015
Submission: 9/4/2015
Task: so
MD5: c3d0db158735e9349f97f9805c6e8114
Run description: This run focuses on building a rich query representation by using external resources, such as Wikipedia (revisions before query start time). Documents were processed in batch and decisions are made at the end of each hour. Sentences within the documents then are ranked and selected.

WikiOnlyFS2¶

Participants | Proceedings | Input | Summary

Run ID: WikiOnlyFS2
Participant: udel_fang
Track: Temporal Summarization
Year: 2015
Submission: 9/4/2015
Task: ps
MD5: ba16dd029d0e45efc8fce0a4f12a6849
Run description: This run focuses on building a rich query representation by using external resources, such as Wikipedia (revisions before query start time). Documents were processed in batch and decisions are made at the end of each hour. Sentences within the documents then are ranked and selected.

WikiProfMix1¶

Participants | Proceedings | Input | Summary

Run ID: WikiProfMix1
Participant: udel_fang
Track: Temporal Summarization
Year: 2015
Submission: 9/4/2015
Task: so
MD5: b3902f2b69e3bf0d7e84cf8e159d48b9
Run description: This run focuses on building a rich query representation by using external resources, such as Wikipedia and other auxiliary corpora that were generated before the query start time. Documents were processed in batch and decisions are made at the end of each hour. Sentences within the documents then are ranked and selected.

WikiProfMixFS1¶

Participants | Proceedings | Input | Summary

Run ID: WikiProfMixFS1
Participant: udel_fang
Track: Temporal Summarization
Year: 2015
Submission: 9/4/2015
Task: ps
MD5: 992d47f0ee0df0a9d469e61559273ecd
Run description: This run focuses on building a rich query representation by using external resources, such as Wikipedia and other auxiliary corpora that were generated before the query start time. Documents were processed in batch and decisions are made at the end of each hour. Sentences within the documents then are ranked and selected.