[Adapt] [Semina]Improving Information Extraction by Acquiring External Evidence with Reinforcement Learning

[邢俊劼]

Hi Adapters,

       This afternoon I’ll give a talk about an interesting approach to information extraction.

       When facing an IE task, a common issue is that “ambiguous” expression occurs. For example, when extracting number of people from “A couple and their four children had a good vacation”, the extractor may return an incorrect answer because “couple” stands for two people. Even a large annotated model may not cover such cases.

       The paper gives an approach to solve the problem by using external information from web and reconcile to a more accurate output, not enlarging the label process or adding ambiguous words handling.

       Models in this paper are a basic extractor based on CRF and a simple reinforcement learning model. The overall process lays on a markov decision process.

       See you this afternoon!

Best wishes,

Gavin

 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://cs.sjtu.edu.cn/pipermail/adapt/attachments/20170426/73a8266f/attachment.html>