Research On Online Reviews Analysis Method Based On Topic Model

Online comment is the product of the rapid development of e-commerce in the Internet era,is an important part of online shopping.When people can not see specific objects,comments become an important reference channel that affects people's buying decisions.The comment also has practical value to the business,and the customer's feedback to the commodity can help the business to make the future product improvement or marketing strategy.The analysis of online comments has important practical significance.However,the number of online comments is huge,and the sentence is changeable,and the content is complex,and theme and emotion are concluded at the same time.one by one to browse all the comments are not realistic.Therefore,it is necessary to put forward an effective method of online comment analysis.This work focus on jumbled online reviews,and explore the related technical review of comment textual analysis,in order to dig the useful information in the online review text and to provide more effective analysis methods and tools for the analysis.Firstly,in order to solve the negative influence of implicit comments on the modeling,a new method to mining the attribute word-evaluation word relation database is proposed.Then,the structure of the implicit comment is complemented by the relational database,and the negative word detection is carried out.An improved sentiment topic model is used to analyze the comments.But the emotional topic model can only analyze the static global review,and can not get the dynamic change of the comments.Therefore,this paper puts forward a dynamic sentiment topic model,and analyzes the evolution of the topic emotion.The specific research methods and results are described as follows.(1)Mining the relationship between explicit attribute words and evaluation words.The writing of online comments is not restricted by any grammar and sentence patterns,which makes the structure of many comments incomplete.According to this feature,the comments can be divided into explicit and implicit comments.The explicit sentence is represented as a word vector by preprocessing,and the noisy word vector is removed by word filtering and mutual information filtering.Semantic similarity algorithm based How Net is used to calculate the similarity between word vectors,and the similar comments are clustered into clusters by AP clustering.Finally,the information gain algorithm is used to get the relationship between attribute words and evaluation words.The attribute word-evaluation word relation database is constructed.(2)Using the improved SSTM model to analyze the emotion of the online reviews.The lack of implicit comments attribute words and the influence of negative words on the emotional polarity of the evaluation words,both of which will reduce the effect of modeling.Therefore,the attribute words of the implicit comment sentence are identified according to the obtained relation database,and then the emotional polarity of the evaluation word modified by the negative word is reversed.After optimizing the online comment data,the SSTM model with the prior information is used to model the comments to improve the accuracy of sentiment classification.(3)Dynamic evolution analysis of online reviews.The topic and the emotion of the comments are changing over time.After the study of several dynamic topic models,the paper proposes the dynamic affective topic model DSTM,which is combined the emotional features and temporal features,and then parameter estimation.The comment set is divided into different subsets according to the time slice.The state of each subset depends only on the state of the previous time set.The DSTM model is used to analyze the change of the topic emotion of the comment over time.