Research And Application On Implicit Product Feature Extration Method Based On The Opinion Words

Electronic commerce development is becoming more and more rapidly with a lot of information appeared every day including purchase records, product reviews and so on. Analyzing the reviews to get the product emotional tendencies of users will bring certain reference value to businesses and other users. However, without identifying the product features that the reviews are talking about, the businessman will not know how to improve the product and other users can’t contrast to choose. Therefore, in order to make the analysis more fine grained, it’s necessary to do research on product feature mining based on the opinion words. The product features can be divided into explicit features and implicit features, research results on explicit product feature extractions are a lot while the implicit feature extractions are few.Based on the above background, this paper considered the implicit product feature extractions as the research target and Chinese reviews on the network as the research object. The work of this paper content can be summarized as follows:(1) This paper put forward a comprehensive weighted method to establish opinion words and contexts lexicon. The existing methods about extracting effective words only consider term frequency while our method considered four aspects which affect the validity of word. And then a score can be calculated with their weights. Experiments show that this method can improve the accuracy of the establishment about opinion words and contexts lexicon.(2) In this paper we present a modified topic model joint topic-opinion model(JTO) for extracting implicit features of opinion words including special and general ones. Our model is based on an extension to standard LDA model by adding an opinion level. This model considers both topics and context of opinion words. Experiments proved that the presented method in feature extraction of implicit products has better performance on accuracy.(3) Researchers have been devoted to using context to extract implicit features. However, little concerns have been given to the situation that not all the contexts are meaningful. To solve this problem, in this paper we present a new method to evaluate the contribution of the contexts for extracting. We build an improved Co-occurrence matrix firstly. And then a LDA topic model is used to get the topic probability of the opinion word. The weight of context can be obtained by using cosine similarity in the improved Co-occurrence matrix and LDA topic model. Experiments have showed that our method provides higher accuracy in extracting the implicit features.(4) This paper designed an implicit product feature extraction system based on the above two models and applied the system into the project about RongHua company.