Bootstrapping both Product Properties and Opinion Words from Chinese Reviews with Cross-Training

Abstract

We consider the problem of 1 identifying product features and opinion words in a unified process from Chinese customer reviews when only a much small seed set of opinion words is available. In particular, we consider a problem setting motivated by the task of identifying product features with opinion words and learning opinion words through features alternately and iteratively. In customer reviews, opinion words usually have a close relationship with product features, and the association between them is measured by a revised formula of mutual information in this paper. A bootstrapping iterative learning strategy is proposed to alternately both of them. A linguistic rule is adopted to identify lowfrequent features and opinion words. Furthermore, a mapping function from opinion words to features is proposed to identify implicit features in sentence. Empirical results on three kinds of product reviews indicate the effectiveness of our method.

Sentiment Lexicon Expansion Based on Neural PU Learning, Double Dictionary Lookup, and Polarity Association

“…Qiu et al (2011) proposed a double propagation (DP) method that uses both sentiment and target relation and various connectives to extract sentiment words. (Wang and Wang, 2008) did similar works. Huang et al (2014) detected new sentiment words using lexical patterns.…”

Mining Opinion Targets and Opinion Words Using Word Alignment Model With Positive and Negative Reviews

“…Detecting opinion relations and calculating opinion associations among words are the key component of this type of method. To indicate opinion associations Wang and Wang [6] adopted the co-occurrence frequency of opinion targets and opinion words. To identify opinion relations among words Hu and Liu [5] exploited nearest-neighbor rules .…”

“…Through the word alignment, an opinion target can find its corresponding modifier, for example in figure 1, "colorful" and "big" are opinion words aligned with the target word "screen". The WAM does not constrain indentifying modified relations as compared to previous nearest-neighbor rules [5], [6], [7] to a limited window; so more complex relations can be capture, such as long-span modified relations.…”

A Review Study of Co-Extracting Opinion Targets and Opinion Words from Online Reviews

“…Connection Identification is utilized to recognize the connection between opinion words and targets words. [6] B. Wang and H. Wang (2011), this research work utilized the recipe of shared data to device the relationship since common data of a low reappearance word pair has a tendency to be high. In this framework shared data is utilized to quantify the affiliation.…”