Research On Emotion Analysis For Chinese Product Reviews

In the Web2.0 era,users are the main body of creating information.With the rapid development of social media and e-commerce platform,a large number of users participate in the comments about people,events and products every day,which effectively convey people's various views and emotional tendencies.The massive Chinese product review data provides the possibility to obtain the emotional satisfaction of users for products quickly,efficiently and scientifically.However,it is difficult to summarize,sort out and apply the massive text reviews scientifically and reasonably.How to extract potential value from these unstructured data has become an urgent need of the Internet industry.The text sentiment analysis technology based on natural language processing mainly studies and applies how to quickly realize the automatic classification,detection and induction of user comment data,and evaluate user satisfaction on the basis of obtaining and processing Chinese user comment data in the field of big data quickly,efficiently and scientifically.However,in specific subject areas,the traditional document level emotion analysis method still has some deficiencies in multi-dimensional cognition of data,so it is urgent to find some new technical methods from the perspective of attributes.For the problem of emotion analysis of massive Chinese product reviews in the era of big data,based on the research and analysis of emotion analysis technology and methods at home and abroad,the performance of traditional general emotion dictionary is tested by using the same data set,and the performance and characteristics of emotion analysis method based on emotion dictionary and machine learning are compared and analyzed through emotion analysis experiment,having thoroughly studied the structure,modeling idea and semantic ability of Word2 Vec model and Bert language representation model.Finally,a fine-grained sentiment analysis method at subject attribute level is proposed to realize the automatic classification,detection and induction of user comment data,which provides a possibility and choice for quick,efficient and scientific evaluation of user satisfaction.The main research contents and results are as follows:(1)Experiment and test of traditional emotion analysis methodThree kinds of general emotion dictionaries are selected,and experiments are designed to test the accuracy of tagging and the coverage of emotion words in the field.Emotion analysis experiments based on dictionaries are carried out using Chinese hotel comment data,and the emotion dictionary of Dalian University of Technology shows the best performance;a variety of discrete features are extracted to vectorize the text,and emotion analysis experiments are carried out on multiple classifiers,and then cited the method based on Chi square test information is used for feature selection.Results display that the accuracy of 0.852 is obtained by using a kind of hybrid feature of unigram and bigram with abundant information and polynomial naive Bayes as classifier model,which is much higher than that of 0.746 in the emotion dictionary of Dalian University of Technology.It shows that the method of discrete representation can capture more statistical and linguistic features,so as to obtain better emotion classification performance.(2)Research on emotion analysis method based on distributed representationTwo methods,word2 Vec and BERT,are used to vectorize the comment text and complete the downstream emotion analysis task.The accuracy of word2Vec-Xgboost model is 0.866,which is 0.014 higher than that of the method based on discrete representation,while BERT-Bilstm model is 0.890,which is 0.024 higher than the former.The results show that more semantic information and text order features can be extracted by using the distributed text representation method,and the characteristics of all levels of the BERT model can further deal with the problem of polysemy,capture more deep semantic and grammatical features,and improve the accuracy of emotional analysis.(3)Research on the expansion method of affective dictionary with Word2 Vec for learningTaking advantage of Word2Vec's distributed representation,this paper proposes a method based on representation learning to expand emotional lexicon,add more domain related emotional words,and optimize the performance of emotional lexicon.The results show that the overall tagging accuracy of the emotion dictionary using Word2 Vec word vector is 0.855,and the coverage of the emotion words in the field is 0.883.The performance of the emotion dictionary is far larger than that of the technology emotion dictionary.It shows that this method can be used to expand the emotion dictionary and build the emotion dictionary and rule base related to the field when performing the emotion analysis task.(4)Design and implementation of subject attribute level emotion analysis systemWe design a comment emotion analysis system for Ctrip hotel.By collecting comment data from web crawler,we integrate the Chinese Academy of Sciences,Harbin Institute of technology and Sichuan University's intelligent laboratory,and remove the emotion words,degree adverbs and negatives,and construct a new stop words list After cleaning the comment corpus after word segmentation,LDA topic modeling is carried out after the text is vectorized,the emotion classification model in snownlp is retrained using the marked corpus,and the emotion analysis is carried out for the text after vectorization,the theme emotion polarity distribution map is drawn,and the data on the web,such as the crawling travel mode and occupancy time,are combined Visualization.The results show that the system can help hotel businesses understand user evaluation from multiple dimensions,optimize management strategies and improve service quality.In summary,the emotion analysis method studied in this paper effectively solves the problems about automatic classification,detection,induction of Chinese user comment data and quick,efficient and scientific evaluation of user satisfaction.Based on the comparative analysis of various document level emotion analysis methods,this paper proposes an experimental fine-grained emotion analysis method based on subject level and attribute level,and discusses each method the advantages and problems of emotion analysis methods.In the actual application scenario,emotional analysis mainly focuses on e-commerce,hotel,catering,automobile and other industries.According to the characteristics of these fields,the fine-grained emotion analysis based on subject level can help businesses more intuitively understand user comment data,find problems in business management,optimize products and services,enhance industry competitiveness,and provide new technical reference for the text sentiment analysis technology field,which has certain application and promotion value.