Runs - Real-time Summarization 2018¶

Run ID: IR-N
Participant: UA-Gplsi
Track: Real-time Summarization
Year: 2018
Submission: 8/4/2018
Type: automatic
Task: b
MD5: 5a2656c34e0eebd3a1245f6e15df50a9
Run description: The Social Analytics system retrieves relevant tweets. The IRN information retrieval system sorts the tweets based on their relevance.

Participants | Proceedings | Appendix

Run ID: IR-NR1
Participant: UA-Gplsi
Track: Real-time Summarization
Year: 2018
Submission: 8/4/2018
Type: automatic
Task: b
MD5: b67f5799209fbe5e19c3a39ba7cdbc9b
Run description: The Social Analytics system retrieves relevant tweets. The IRN information retrieval system sorts the tweets based on their relevance. We detect duplicates tweets with 50% of similarty

Participants | Proceedings | Appendix

Run ID: IR-NR2
Participant: UA-Gplsi
Track: Real-time Summarization
Year: 2018
Submission: 8/4/2018
Type: automatic
Task: b
MD5: 55c38c2240901c6f231801197178943f
Run description: The Social Analytics system retrieves relevant tweets. The IRN information retrieval system sorts the tweets based on their relevance. We detect duplicates tweets with 50% of equal words

Participants | Proceedings | Appendix

Participants | Proceedings | Appendix

Participants | Proceedings | Appendix

Participants | Proceedings | Appendix

Run ID: IRIT-RunB1
Participant: IRIT
Track: Real-time Summarization
Year: 2018
Submission: 8/4/2018
Type: automatic
Task: b
MD5: 5624193bebd2ee15b02b4a9a5695bbc2
Run description: This run is composed of two components. The first component consists of two-stage filters. The first filter discards poor quality tweets and the second is a relevance filter based on a binary classifier. The classifier was trained on the TRECRTF 2015 data set using a XGboost Algorithme. This run makes use of the assessment feedback to re-train the binary classifier with the new instances. At the end of the day, candidates tweets are ranked according to their relevance score and TOPK-100 tweets are iteratively selected, with discarding redundant tweets.

Participants | Proceedings | Appendix

Run ID: IRIT-RunB2
Participant: IRIT
Track: Real-time Summarization
Year: 2018
Submission: 8/4/2018
Type: automatic
Task: b
MD5: a234d36e2d14bb1909f672f42edae979
Run description: This run is composed of two components. The first component consists of two-stage filters. The first filter discards poor quality tweets and the second is a relevance filter based on a binary classifier. The classifier was trained on the TRECRTF 2015 data set using a XGboost Algorithme. This run makes use of the assessment feedback to re-train the binary classifier with the new instances. The second component consists on the selection of the tweets for each day. It is formulated as an optimization problem using integer linear programming (ILP) to select a subset of tweets that maximizes the global summary relevance and fulfil constraints related to non-redundancy, coverage, temporal diversity and summary length. To resolve the ILP To we use the branch and bound algorithm

Participants | Proceedings | Appendix

Run ID: IRIT-RunB3
Participant: IRIT
Track: Real-time Summarization
Year: 2018
Submission: 8/4/2018
Type: automatic
Task: b
MD5: 4b65f48a712316a789d5354e9470c363
Run description: This run is composed of two components. The first component consists of two-stage filters. The first filter discards poor quality tweets and the second is a relevance filter based on a binary classifier. The classifier was trained on the TRECRTF 2015 data set using a Random Forest algorithm. At the end of the day, candidates tweets are ranked according to their relevance score and TOPK-100 tweets are iteratively selected, with discarding redundant tweets.