Research On Micro-summary Of Product Reviews Based On Unsupervised Learning

With the rapid development of Web 2.0 technologies, the product reviews information that user generated online is proliferating in an exploding way, which has the characteristics of large number, brief, targeted, highly subjective, normative low and so on. It is valuable to provide the concise summary of reviews to users, fastly and accurately. Opinion mining on product reviews have been extensively researched, with the rapid popularization of mobile internet and social network, the contents text generated by users tend to be short text. Traditional long text opinion mining methods cannot be applied to mining short text reviews because of its less information. Also users tend to get strong representation, readable short information messages. Considering of the special nature of Chinese, abroad studies about English opinion mining can not apply to Chinese product reviews study. More and more attention are forcing on short text messages about Chinese product reviews.In this paper, the definition of micro-summay and micropinon was came forward for the problem of large numbers of Chinese product short text reviews online. A new unsupervised approach to generate concise and readable unstructured representative micropinon sets of large numbers of short text reviews on the network was proposed. Formulate the problem of generating such a micropinion set as an optimization problem, measure representativeness based on a modified mutual information function, model readability with an n-gram language model and combine semantic similarity and string similarity to calculate word similarity; also made some heuristic algorithms to solve the optimization problem. Experiment results show that the proposed unsupervised abstract microsummary generated method of Chinsese short text reviews obtain excellent representation, readability and redundancy; achieved good precission, recall and F-value.Finally, the micro-summay prototype system of Chinese product reviews based on micropinions was designed, with product reviews corpus acquisition, data preprocessing, micropinion generating and generate micro-summary modules contined. It analysis and process massive short text reviews online so as to provied users with high quality, briefy, low redundancy and readable micro-summaries.