Chinese event place phrase recognition of emergency event using Maximum Entropy

Abstract

This paper provides a new method combining Maximum Entropy with rules for identify event place phrase. Firstly, all phrases which not include event trigger are extracted from event mention, and a rule base about event place phrases analyzes and filters these phrases for obtaining the phrase candidate set. Secondly, we explore some rich text features from three kinds of linguistics features that contain phrase, event trigger and context information. Thirdly, in order to establish a train set, we use some feature words representing these text features to build feature vector space. Then, a machine learning model to identify event place phrase is trained by using L-BFGS functions algorithm. At last, this predictive model is used to classify the test set. The result shows that the method is efficient. In open test, the recall, precision and F-measure reach 0.6296296, 0.8095238 and 0.7083333 respectively.