TY - GEN
T1 - A multiple instance learning framework for identifying key sentences and detecting events
AU - Wang, Wei
AU - Ning, Yue
AU - Rangwala, Huzefa
AU - Ramakrishnan, Naren
N1 - Publisher Copyright:
© 2016 ACM.
PY - 2016/10/24
Y1 - 2016/10/24
N2 - State-of-the-art event encoding approaches rely on senten or phrase level labeling, which are both time consuming a infeasible to extend to large scale text corpora and emergi domains. Using a multiple instance learning approach, take advantage of the fact that while labels at the senten level are difficult to obtain, they are relatively easy to gath at the document level. This enables us to view the proble of event detection and extraction in a unified manner. U ing distributed representations of text, we develop a multip instance formulation that simultaneously classifies news ticles and extracts sentences indicative of events without a engineered features. We evaluate our model in its ability detect news articles about civil unrest events (from Spani text) across ten Latin American countries and identify t key sentences pertaining to these events. Our model, trained without annotated sentence labels, yields performance that is competitive with selected state-of-the-art models for event detection and sentence identification. Additionally, qualitative experimental results show that the extracted event-related sentences are informative and enhance various downstream applications such as article summarization, visualization, and event encoding.
AB - State-of-the-art event encoding approaches rely on senten or phrase level labeling, which are both time consuming a infeasible to extend to large scale text corpora and emergi domains. Using a multiple instance learning approach, take advantage of the fact that while labels at the senten level are difficult to obtain, they are relatively easy to gath at the document level. This enables us to view the proble of event detection and extraction in a unified manner. U ing distributed representations of text, we develop a multip instance formulation that simultaneously classifies news ticles and extracts sentences indicative of events without a engineered features. We evaluate our model in its ability detect news articles about civil unrest events (from Spani text) across ten Latin American countries and identify t key sentences pertaining to these events. Our model, trained without annotated sentence labels, yields performance that is competitive with selected state-of-the-art models for event detection and sentence identification. Additionally, qualitative experimental results show that the extracted event-related sentences are informative and enhance various downstream applications such as article summarization, visualization, and event encoding.
KW - CNN
KW - Deep learning
KW - Event detection
KW - Information extraction
KW - MIL
UR - http://www.scopus.com/inward/record.url?scp=84996490731&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=84996490731&partnerID=8YFLogxK
U2 - 10.1145/2983323.2983821
DO - 10.1145/2983323.2983821
M3 - Conference contribution
AN - SCOPUS:84996490731
T3 - International Conference on Information and Knowledge Management, Proceedings
SP - 509
EP - 518
BT - CIKM 2016 - Proceedings of the 2016 ACM Conference on Information and Knowledge Management
T2 - 25th ACM International Conference on Information and Knowledge Management, CIKM 2016
Y2 - 24 October 2016 through 28 October 2016
ER -