Research On Sentiment Analysis With Product Reviews

With the rapid development of the Internet industry, more and more people are keen on posting comment to express their views and opinions on a certain things or events on the Internet. For example, the transaction of product contains a large number of such opinion commentary in the field of electronic commerce. From the perspective of consumers, they can express the level of their own preference by posting comment, while they also use the previous comment information of a certain product as decision-making; From the perspective of the producers and merchants, they can improve production and improve sales decisions by the feedback of these comments in order to meet consumers’demand. Current research on sentiment analysis with reviews is focus on text mining, natural language processing and other related areas, and it has a strong application value.However, there are many problems need to be solved in the field of sentiment analysis. On the one hand, the sentiment analysis of the traditional product reviews are essentially coarse-grained research that based on the level of chapter or sentence, it rarely involve more fine-grained research that based on the level of word or phrase; On the other hand, with the emergence of various types of new products, the issues of domain is more prominent in sentiment analysis with reviews, the classification model that is trained in a certain domain often can’t be directly used in other domain, so that each domain requires a lot of manual labeled training data set, it will consume enormous manpower, material and time.To solve above two problems, this paper mainly carried out the following several aspects of research.(1) The method which a fine-grained sentiment tendency analysis for product reviews is studied. By combining the theory of conditional random field with the category relations between the attributes of evaluate object and its evaluation, we built a sequence model of common recognition and extraction, and specially carried on the detailed analysis and the design of algorithm for the recognition and extraction of semantic feature based on semantic character, finally proposes a method that it can common recognize and extract for the attributes of evaluation object and its evaluation about product.(2) Research and design a fine-grained sentiment analysis system for product reviews. To make the technique of fine-grained sentiment analysis effectively applied to practical application, this paper construct a fine-grained sentiment analysis system for reviews of the purifier by the method that is proposed in the (1).(3) This paper make a preliminary exploration and related research for the cross-domain sentiment analysis with product reviews. To solve the prominent problem of domain in sentiment analysis research for product reviews, this paper focuses on verifying the existence of the otherness and relatedness between different product domain and researching the impact of the select of feature for the effect of classification in the cross-domain sentiment analysis.