Research On Algorithm Of Customer Reviews Information Extraction

With the rapid development of trade networks, in order to enhance customers'satisfaction, share customers'shopping experience, On-line merchants let customers that have been purchased goods to expression their opinion or suggestion has become a very common thing. As more and more users felt the convenience of online shopping and some other favorable to them, the number of reviews has been increasing. Therefore, the number of best-selling products in a large shopping site can be reached to hundreds of users'reviews. So many customers comment to manufacturers or potential purchaser that they are very difficult to track users'opinion or suggestion to their purchased products, which caused additional difficulties to them to make decision. A new research area - - Research on Customer Reviews Information Extraction Based on Product Feature produced under this kind of background. Now, more and more researchers join in this subject, Hu and Liu in 2004 published that customer comments and information extraction is one of the first and the most complete documentation.This paper first introduced the research background and current situation at home and abroad of Research on Customer Reviews Information Extraction Based on Product Feature, elaborated on the purpose and significance of the current network trading platform functional modules in detail, pointed out its insufficiency, and proposed that applying the module of extracting user'reviews information to the network trade platform. The relevant knowledge which this paper used is also studied.Then, this paper studied Hu, Liu'customer reviews extraction algorithm, the algorithm based on support to extract the product features and the technique of using WordNet to identify the orientation of opinion are all studied deeply. This paper finds that Hu, Liu's algorithm has its limitation.After studied the limitations of existing algorithms, this paper proposed based on the Bernoulli model to extract product features, elaborated the process that the knowledge of probability and statistics apply to extract product feature which the user involved in detail. And compared to the algorithm based on support to extract the product features, experimental results show that the terms from the algorithm based on probability to extract product feature are shorter and more effective to end-users to find appropriate product, so it can play a role of navigation truly.In addition,this paper proposed the technique of relaxation labeling apply to identify the orientation of opinion which the user involved, then using orientation of potential opinion to identify the opinion phrase. This chapter use recall rate and accuracy to evaluated this algorithm and the algorithm of using WordNet to identify the orientation of opinion (Hu and Liu, 2004), the experimental results were analyzed. In the experiment of identifying the orientation of opinion which the user involved, because this paper's algorithm can process the sensitive content public opinion word, The precision of this paper's algorithm is higher 0.03 than the algorithm of using the WordNet to identify semantics orientation of opinion. However, because this paper's algorithm can't distinguish the words of which has not appeared in WordNet, or the words have not enough classification information in WordNet, The recalling rate of this paper's algorithm is lower 0.03 than the algorithm of using the WordNet to identify semantics orientation of opinion. This needs to improve in the behind work.