基于ChunkCRF的观点持有者识别及其在观点摘要中的应用

Chunk-CRF-based Opinion Holder Identification and Application to Opinion Summarization

含有观点的文档中准确识别出观点的持有者是预处理步骤.通过建立ChunkCRF模型对观点表达句进行观点持有者的识别;对于同一个观点句中含有多个观点持有者的情况,借助语言学手段进行预处理,再利用模型进行观点持有者识别.在此基础上还进行了观点的摘要与倾向性分析的工作.实验结果表明,基于ChunkCRF的中文观点持有者识别方法达到了80%上以的准确率,并且能够更好的配合观点的摘要与倾向性分析工作.

Accurate opinion-holder-identification is an important preprocessing for opinion summarization based on opinion holder. In this paper, a ChunkCRF model was constructed to divide the opinionated sentence into particular chunks with the aim of effectively identify the opinion sources; at the same time, in multi-Opinion-Holder cases, syntactic analysis was made use of before applying CRF model. Moreover, opinion summarization and polarity analysis were also given. Experiment results showed the ChunkCRF-based method identified opinion holders with precision over 80%, moreover, it could also assist the opinion summarization and polarity analysis well.