| Nowadays, the electronic commerce plays a more and more important role in our daily life. Consumers always express opinions on the product via the Web after using the product. The automatic mining on these comments is important for the potential consumers and enterprises. We focus on Chinese product reviews. We analyzed the comments on two levels including document-level sentiment classification and feature-based product opinion mining. The main contents are as follows:We employ machine learning algorithm to perform the document-level sentiment classification of the product reviews. We collect corpus from online reviews; investigate the N-Gram based feature representation including Word-Based Uigram, Bigram and Chinese Character-Based Unigram, Bigram, trigram; analysis different feature weighting approaches(TF, BOOL, TFIDF), compare different classification algorithms (Naive Bayes, Maximum Entropy and Support Vector Machine). The SVM using Chinese Character Bigram-based feature extraction method and word frequency based text representation has the best performance, of which the accuracy was 94.74%. We researched suffix tree based structure algorithm extracting the Key Substring Group features. Experiments show that the Key Substring Group features have better description of the comments sentiment classification, lower dimension, and better accuracy than other text features represented in SVM.We investigated dependency parsing based algorithm and keyword matching based algorithm for feature-based opinion mining. We construct a product features library and a Chinese polarity Dictionary. Experiments show that the keyword based method is better than the dependency parsing based method.We designed and implemented a product review opinion mining system. The system can automatically crawl and extract specified comments on review pages, then analysis the reviews, save the result into the products opinion library. Users can get visualized result which will be helpful for decision making. |
PDF Full Text Request