Research On Sentiment Orientation Analysis Of User Reviews Based On Product Features

With the rapid development of the Internet and the rise of Web2.0, the network is no longer just a means of access to information, but also a platform to express their views and opinions. More and more online shopping users have a habit of posting reviews of products at e-commerce websites after the purchase and use of a product or service to express their experience. These reviews typically contain a wealth of valuable information. To view these reviews can not only guide consumers to buy correct products, but also help manufacturers to produce products that meet market demand. However, these reviews generally described in natural language are unstructured and to obtain information from a large number of reviews by artificial means is often very difficult. Therefore, the product review mining has become a hot research topic in recent years.This dissertation chooses the real users reviews crawled from e-commerce websites as the study object and makes deeply research on product features extraction, the construction of sentiment lexicons, the sentiment analysis of reviews. The main work of this dissertation includes the following aspects:(1) A product features extraction method which combines Apriori algorithm with text mode is proposed and three methods of feature pruning are used to filter the candidate product features.(2) Sentiment lexicons are constructed. We construct commendatory and derogatory lexicon, adverbs of degree lexicon, negative words lexicon and dynamic words lexicon by taking Hownet lexicon, NTUSD lexicon and reviews corpus as resources.(3) A method of extracting evaluation combination units and calculating its emotional tendency is proposed, further we analyze the emotional tendency of reviews based on features.In order to verify the feasibility of the proposed reviews mining method in the dissertation, taking reviews of three digital products from Jingdong Mall as an experimental corpus, we carries out several experiments on extraction of product features, recognition of opinion sentences, extraction of evaluation combination units, sentiment analysis of reviews. The results show that the methods proposed in this dissertation have comparatively high feasibility and effectiveness.