Research On Chinese Online Reviews Sentiment Classification

Due to the Big Data Time is coming, huge amounts of useful information with opinions and sentiments generated on the internet, such as user’s geographic location, browsing history, product reviews. We can analysis those data, mining user’s-implicit intention, and provide personalized service to users. This article will do some research about sentiment analysis for Chinese online reviews. We will do some experiments based on the theory of natural language processing (NLP) and machine learning. Our goal is to implement an algorithm that can automatically know exactly online reviews’sentiment. Sentiment classification algorithm will be applied in analysis and decision-making, public opinion analysis and monitoring, information forecasting, personalized recommendation and other fields.This thesis involves the research contents are mainly the following three points:Firstly, pretreatment process analysis for Chinese online reviews.We analyzed the different of standardized text and online reviews in the aspect of word segmentation, feature selection, etc. And we presented a method for Chinese online reviews sentiment analysis pretreatment.Secondly, feature selection for Chinese online reviews sentiment classification. We improved traditional feature selection algorithm (IG, DF, MI), and new algorithms can reflect text’s sentiment characteristics, easy to select fewer high quality feature items. We compared different classifiers to find out the optimal combination of classifier and feature selection algorithm, and explained the reason. Finally, realized a distributed system based on Spark. We successfully used Spark framework to realize a distributed system for large-scale data processing.