Research On Mining Techniques Of Product Reviews Based On Multi-Document Summarization

The research of mining techniques of product reviews is an important research direction of ustructured data processing under the internet environment,Through these techniques.we can get valuable information from the product reviews automatically.In this paper,a multi-document summarization algorithm based on product feature is proposed,then analyze the reviews based on the summary.The main work of this thesis is as follows:1) This paper extend the feature lexicon and sentiment lexicon by distributed representations, then use filtering rules to filter the results.By this way,we can get high-quality thesaurus of feature word and sentimental word.2) This paper presents a method to automatically generate sentiment lexicon for Chinese sentiment classification based on lucene and LTP.Use lucene to find sentences which contains a sentimental word with konwn polarity and has a relationship of COO with the unkonwn-polarity sentimental word.From these sentences.we can judge the polarity of the sentimental word.the accuracy of this method is very high.3) According to the characteristic of syntactic structure,we design a summarization algorithm based on product feature in order to get a representative and readability summary.The summary will be the basis of classification of reviews.The sentences of reviews will be classified into relevant category according to the feature word and sentimental word in the sentence. Experiments show that the classification results of product reviews achive good results.4)In order to realize automatic product reviews mining.we design an automatic summarization system based on the researchs mentions above and the requirements of customers.It can help customers get valuable information quickly.