Research On Hotel Review Text Mining Based On Sentiment Classification

The popularity of the Internet and the application of mobile terminals has promoted the e-commerce rapid development,and a large number of behavioral data is generated by consumers when they conduct transactions on major e-commerce platforms,online comment text is one type of the behavioral data.The online comment text is the feedback from the consumer to the e-commerce platform in the form of text after the consumer's personal experience,and is guided by the public opinion of the public.Effective emotional classification of such opinions can not only help consumers make decisions,but also help merchants improve the service.At present,when emotional classification of short texts is made,because the short texts have problems such as high feature dimension,domain difference,and expression impliedness,which will affect the performance of classification.This paper will improve the existing sentiment classification methods for the above problems,and used them for emotional classification of hotel review texts.The main research contents are as follows:(1)Aiming at the problem that the difference of domain sentiment words and the incompleteness of the basic sentiment dictionary cause the accuracy of sentiment classification is not high,a fusion method of emotional semantic expansion is proposed to construct the sentiment dictionary of hotel review text.This paper crawl the online hotel comment text as a corpus,combine the seed words with Word2 vec and SO-PMI to expand the emotional words,construct the intra-curricular sentiment dictionary of the hotel review text,and demonstrate the effectiveness of the constructed emotional dictionary in the emotional classification.(2)This paper construct a method based on the combination of dependency syntax analysis and LDA topic model for feature extraction.Firstly,using the dependency syntax analysis combined with the sentiment dictionary to extract the emotional elements of the review text.Secondly,using the LDA theme model to extract the feature items of the emotional elements,the feature extraction method not only directly considers the features related to the theme and emotion,but also indirectly considers contextual semantic information that affects emotions.Experiments show that thismethod is more advantageous than traditional feature extraction.(3)This paper improve the text sentiment classification method,and construct a deep learning sentiment classification method based on theme and emotional features.The dependent syntactic analysis and the vectorized representation of the feature items extracted by the LDA topic model are combined with the text vector to classify emotion as the input vector of the long short-term memory network classification model to improve the existing sentiment classification method without considering the text topic features and semantics emotional information on classification performance.The experimental results show that the method is better than the traditional method for emotion classification,and can effectively improve the performance of emotion classification.By crawling the online hotel review texts of Ctrip.com,Qunar.com and the public comment website as the experimental data,the simulation experiments show that the domain affective dictionary constructed in this paper and the deep learning affective classification method based on subject and affective features improve the performance of text affective classification.