A Composite Kernel Approach for Detecting Interactive Segments in Chinese Topic Documents

Discovering the interactions between persons mentioned in a set of topic documents can help readers construct the background of a topic and facilitate comprehension. In this paper, we propose a rich interactive tree structure to represent syntactic, content, and semantic information in text. We also present a composite kernel classification method that integrates the tree structure with a bigram kernel to identify text segments that mention person interactions in topic documents. Empirical evaluations demonstrate that the proposed tree structure and bigram kernel are effective and the composite kernel approach outperforms well-known relation extraction and PPI methods.