The Topic Mining Of Product Reviews Based On Emotional Classification

With the development of Internet,more and more people start to buy products on Internet.They post a lot of reviews about the goods.Those reviews reflect the customers' objective description and subjective feeling about the product quality,price,appearance,performance and so on.These comments can provide a lot of useful information about goods.However,the large number of datas is often difficult for customer to read one by one.So,how to quickly extract them from the numerous reviews has become a problem to be solved.The paper takes the topic mining as the main task,putting a new model,named CL-LDA,which based on the topic mining and combined with emotional classification.In the process,the paper does a series of work to process the datas,such as,Chinese words segmentation with jieba,removing the stop words,text classification,topic mining,and so on.The body of the research mainly includes the following two aspects:(1)Text classification on the basis of emotion tendentiousness,the text is divided into positive and negative text sets.In the process,the paper is using the method based on emotional dictionary.First,build collocation phrases which consists of evaluation objects and emotional words.Then,create an emotional dictionary which consists of HowNet and high frequency emotional words in the articles.Last but the most important thing is matching the two groups of emotional words one by one.Positive words marked as 1 and negative words marked as-1,according to the mark to classify text.(2)The topic mining based on Latent Dirichlet Allocation.Estimate the numbers of topic using the likelihood function.Train the themes of the CL-LDA model using Gibbs sampling.Lastly,evaluate the results according to perplexity and F.The experimental results show that the model well solves the problem that results mix in positive and negative emotional words.And perplexity is the smaller and F is bigger than LDA model.Trough the verification,CL-LDA model is valid.