Co-Extracting Opinion Targets and Opinion Words From Online Reviews Based on The Partially Supervised Word Alignment Model

Abstract

Mining opinion targets and opinion words from online reviews are important tasks for fine-grained opinion mining, the key component of which involves detecting opinion relations among words. A novel approach based on the Partially Supervised Word Alignment model is proposed in a monolingual scenario to mine opinion relations in sentences and estimate the associations between opinion target candidates and potential opinion words by incorporating partial alignment links into the alignment process that is generated by the use of pos tagging, where a potential opinion relation is comprised of an opinion target candidate and its corresponding modified word. Next, an Opinion Relation Graph will be constructed to model all opinion target/word candidates and the opinion relations among them. Where all nouns/noun phrases in sentences are assumed to be an opinion target candidates, and all adjectives/verbs are regarded as potential opinion words, each candidate will be assigned a confidence. Finally, candidates with higher confidence than a threshold will be extracted as the opinion targets or opinion words. Moreover our model captures opinion relations more precisely, especially for long span relations. Our experimental results on the customer review datasets (CRD), which includes English reviews of five different products show that our approach provides better Precision (74.2%), Recall (64.4%) and F-Measure (65.3%). Our Partially Supervised Word Alignment model effectively alleviates the negative effects of parsing errors when dealing with informal online texts. In particular, this model not only inherits the advantages of the traditional methods for opinion relation identification, but it also has a more precise performance because of the use of partial supervision. Thus, it is reasonable to expect that the PSWAM is likely to yield better results for extracting opinion targets and opinion words.