| With the rapid development of B2C business model and online shopping, the number of network reviews grows fast, making information more and more complex, and making useful information difficult to obtain customer comments. Therefore, there is an urgent need to have an effective means to organize a variety of large amounts of data, and provide a glance show the statistical results of the data to the consumer in the form. Opinion mining technology of product views of network is developed rapidly in this application demand environment. Opinion Mining is an important research direction of data mining, which is based on data mining and text mining technology, and it has the ability to text understanding and tendentious analysis.Opinion target extraction and opinion tendentious analysis are two core tasks of opinion mining. In-depth study of these two basic tasks, and product reviews as an object of research, analysis tasks are divided into the opinion target extraction, identification of the opinion target associated pairs and emotional tendentious analysis three parts, the product of the various components commendatory and derogatory evaluation of the number of parts and each attribute as a characteristic, analysis and statistics of each feature of the product, and eventually presented to the users.This paper mainly consisted of three parts:(1) In the opinion target extraction, this paper proposed an extraction method based on pattern matching, this approach firstly gets seed rules set through a large number of sample statistics, to extract the effective evaluation sentences, and then extracts accurate opinion targets through the combination of syntactic structures and the POS-distance correlation algorithm. Seed rules and opinion targets are stored in the corresponding pattern libraries. Meanwhile, the training and expansion of the learning of rules and opinion targets is carried out by the semi-supervised learning methods and rules of dynamic replacement. This approach allows to extract the opinion target has greatly improved the precision and recall rate.(2) This paper proposed the generalization combination syntax tree and the similarity matching combination of patterns. We use pattern matching method to extract opinion target association. There is difficult to exact pattern matching used in this structure of the syntax tree, so that the recall rate of result is not high in the final extraction. Firstly, we organize the syntax tree node with generalization combination, and then use the similarity algorithm for pattern matching. The experiments show that the result is better than exact matching, and also better than the situation of syntax tree nodes with not generalized and combine.(3) In the emotional tendency analysis process, split distinguish introduced into the emotional dictionary matching method. Evaluation words by syntactic tree junction point generalization combined word the Evaluation of words not found in the evaluation Dictionary, the word segmentation as the smallest unit again in the evaluation of the dictionary to find, and the linear weighted the tendentious way the word of the entire evaluation discriminate. Experimental results show that this method can effectively reduce emotional tendencies misjudgment of the situation, the emotional tendency results with higher accuracy.(4) This paper designed and implemented a the generic comments opinion mining system, from the acquisition of Internet product reviews, product acquisition, emotional tendentious analysis to show the form of humane results in one integrated environment. The system consists of data processing platform and user visualization platform constitutes. Including the collection of network product reviews, the identification of opinion target associated pairs and emotional tendentious analysis and product characteristics tendentious show modules. |
