Research Of PMI-Strapping Algorithm For Extracting Product Features From Chinese Online Reviews

Because of the rapid progress of information technology, we are stepping into a big data age which is experiencing an intelligent transformation. Massive data that come from a variety of information resources contains a huge wealth, also has become our huge burden. In recent years, the emergence of a spurt of development of electronic commerce has occurred, and mobile terminal gradually become popular, user participation in the Internet’s passion is more and more high. These have become important driving forces for big data. And the businessman is enjoying enormous benefits brought by the Internet. Meanwhile, they also have to face the fierce competition. Look for ways by using big data to further understand the users to provide them with more personalized and high-quality products and services, has become the key to win this fierce competition. The mass of the reviews accumulated online contains the truly emotion and experience of the users, reflects their deep love and habit, but also more easily impact on others’ purchase decision as word-of-mouth.The research of online reviews mining started in the early twenty-first Century, and product feature extraction is one of the key, which is to find users are concerned in the reviews is what aspects of product. But the existing research is still not completely resolved the difficulties such as machine semantic comprehension, noise, the poor transplantation of methods and so forth. Due to the higher complexity of the language of Chinese, Research on product feature extraction from Chinese online reviews is faced with more difficult challenges. This article innovates and improvements on the basis of previous research, according to the characteristics of Chinese, realize effectively extracting product features, and extend the algorithm in sentiment analysis.This paper first introduces the online review mining background, reviews the research and the relevant theoretical basis. In the model of product feature extraction, making use of linguistic theory, three principles are proposed to select the language rules, and obtained three language rules, which are used to get better candidate feature set. Then aiming at improving the defect of the original PMI algorithm, give the PMI-Strapping algorithm. Starting from a simple seed, iterative optimization, combined with dynamic auto setting threshold, realize product feature extraction. At the same time, propose the corpus cutting method and artificial supervision method, to improve the threshold setting so that the algorithm can adapt to different mining target. The products feature extraction model is applied to the actual set of online reviews, and the results show obviously better than the traditional PMI algorithm’s performance. Then, this paper summarizes the important ideas of PMI-Strapping feature extraction algorithm. And make a use of the ideas in sentiment analysis, then construct an algorithm, to solve two problems in the research of the sentiment analysis: sentiment analysis is divorced from product feature; sentiment analysis model is simply linearized. Finally, this paper summarizes the full text, and the follow-up work is prospected.