School of Computer Science and Technology, Shandong University, Jinan, China

---

Abstract—To establish links between a large number of reviews and events, we propose a web reviews and events matching approach by event feature segments and semi-Markov conditional random fields (CRFs). We extract named entities and verb phrases from reviews as event feature segments. We use semi-Markov CRFs to label the reviews and to recognize event feature segments at the segment level. This approach uses event feature segments to match reviews and events. Therefore, it is more accurate than other approaches which use only named entities to match. We use several feature rules to recognize the variants of named entities, such as abbreviation and acronym. In addition, we use phrase dependency parsing tree to recognize verb phrases. A compositive similarity measurement function is presented to combine similarity results of event feature segments. Experimental results demonstrate that this method can accurately match reviews and events.

Index Terms—event feature segments, semi-Markov CRFs, feature rules, phrase dependency parsing tree

[PDF]

Cite: Yuanzi Xu, Qingzhong Li, Zhongmin Yan, Wei Wang, "Web Reviews and Events Matching Based on Event Feature Segments and Semi-Markov Conditional Random Fields," Journal of Software vol. 9, no. 9, pp. 2401-2408, 2014.