| 英文摘要 |
Product reviews have become the important recourse in the online environment of Internet. However, the quality of the reviews is spotty and this influences the accuracy and the reliability for data mining. This paper focuses on how to excavate the helpful product reviews buried under the mass of data. The proposed method is as follows: filter words using the best first-search strategy, use latent Dirichlet allocation (LDA) to get the topic distribution, use the Kullback-Leibler (KL) divergence to calculate the similarity, extract the popular opinion of reviews, observe the difference between the popular opinion and the review, perform emotion detailing by getting the specific value of each attribute, consider the credibility as well as the metadata of the reviews, and finally train the weights of feature vectors according to the support vector machine (SVM). Experimental results demonstrate the ability of the proposed method to significantly improve the classification accuracy. |
本站僅提供期刊文獻檢索。
