The Copenhagen Team Participation in the Check-Worthiness Task of the Competition of Automatic Identification and Verification of Claims in Political Debates of the CLEF-2018 CheckThat! Lab
Research output: Chapter in Book/Report/Conference proceeding › Article in proceedings › Research › peer-review
Standard
The Copenhagen Team Participation in the Check-Worthiness Task of the Competition of Automatic Identification and Verification of Claims in Political Debates of the CLEF-2018 CheckThat! Lab. / Hansen, Casper; Hansen, Christian; Simonsen, Jakob Grue; Lioma, Christina.
CLEF 2018 Working Notes. ed. / Linda Cappellato ; Nicola Ferro ; Jian-Yun Nie; Laure Soulier. 10. ed. CEUR-WS.org, 2018. 81 (CEUR Workshop Proceedings, Vol. 2125).Research output: Chapter in Book/Report/Conference proceeding › Article in proceedings › Research › peer-review
Harvard
APA
Vancouver
Author
Bibtex
}
RIS
TY - GEN
T1 - The Copenhagen Team Participation in the Check-Worthiness Task of the Competition of Automatic Identification and Verification of Claims in Political Debates of the CLEF-2018 CheckThat! Lab
AU - Hansen, Casper
AU - Hansen, Christian
AU - Simonsen, Jakob Grue
AU - Lioma, Christina
PY - 2018
Y1 - 2018
N2 - We predict which claim in a political debate should be prioritizedfor fact-checking. A particular challenge is, given a debate, how toproduce a ranked list of its sentences based on their worthiness for factchecking. We develop a Recurrent Neural Network (RNN) model thatlearns a sentence embedding, which is then used to predict the checkworthinessof a sentence. Our sentence embedding encodes both semanticand syntactic dependencies using pretrained word2vec word embeddingsas well as part-of-speech tagging and syntactic dependency parsing. Thisresults in a multi-representation of each word, which we use as input to aRNN with GRU memory units; the output from each word is aggregatedusing attention, followed by a fully connected layer, from which the outputis predicted using a sigmoid function. The overall performance of ourtechniques is successful, achieving the overall second best performing run(MAP: 0.1152) in the competition, as well as the highest overall performance(MAP: 0.1810) for our contrastive run with a 32% improvementover the second highest MAP score in the English language category. Inour primary run we combined our sentence embedding with state of theart check-worthy features, whereas in the contrastive run we consideredour sentence embedding alone
AB - We predict which claim in a political debate should be prioritizedfor fact-checking. A particular challenge is, given a debate, how toproduce a ranked list of its sentences based on their worthiness for factchecking. We develop a Recurrent Neural Network (RNN) model thatlearns a sentence embedding, which is then used to predict the checkworthinessof a sentence. Our sentence embedding encodes both semanticand syntactic dependencies using pretrained word2vec word embeddingsas well as part-of-speech tagging and syntactic dependency parsing. Thisresults in a multi-representation of each word, which we use as input to aRNN with GRU memory units; the output from each word is aggregatedusing attention, followed by a fully connected layer, from which the outputis predicted using a sigmoid function. The overall performance of ourtechniques is successful, achieving the overall second best performing run(MAP: 0.1152) in the competition, as well as the highest overall performance(MAP: 0.1810) for our contrastive run with a 32% improvementover the second highest MAP score in the English language category. Inour primary run we combined our sentence embedding with state of theart check-worthy features, whereas in the contrastive run we consideredour sentence embedding alone
KW - CNN
KW - Fact checking
KW - Political debates
KW - RNN
M3 - Article in proceedings
T3 - CEUR Workshop Proceedings
BT - CLEF 2018 Working Notes
A2 - Cappellato , Linda
A2 - Ferro , Nicola
A2 - Nie, Jian-Yun
A2 - Soulier, Laure
PB - CEUR-WS.org
T2 - 19th Working Notes of CLEF Conference and Labs of the Evaluation Forum, CLEF 2018
Y2 - 10 September 2018 through 14 September 2018
ER -
ID: 202539747