A combination method of CRF with Syntactic Rules to identify opinion_holder

摘要：

This paper presents another aspect of sentiment analysis: identifying opinion_holder in the opinionated sentences. To extract opinion_holder, we firstly explore Conditional Random Field(CRF) based on six features including contextual, opinionated_trigger words, POS tags, named entity, dependency and proposed sentence structure feature, and dependency is adjusted to be better helpful for containing contextual dependency information. Then we propose two novel syntactic rules with opinionated_trigger words to directly identify opinion_holder from the parse trees. The results show that the precision from CRF is much higher than that of syntactic rules, while the recall is lower than. So we combine CRF with syntactic rules used as additional three features including HolderNode, ChunkPosition and Paths for the CRF to train our model. The combination results of the system illustrate the higher recall and higher F-measure under the almost same high precision.

关键词： opinion_holder CRF features syntactic rules combination

作者: Yuan KUANG Yanquan ZHOU Huacan HE

作者单位: Beijing University of Posts　and Telecommunications　Beijing, China Northwestern Polytechnic University Xi’an, China

会议类型: 国际会议

会议名称: The 6th International Conference on Natural Language Processing and Knowledge Engineering(第六届IEEE自然语言处理与知识工程国际会议 NLP-KE 2010)

会议地点: 北京

会议语种:英文

页码: 1-8

在线出版日期: 2010-08-21（万方平台首次上网日期，不代表论文的发表时间）