Review Anomaly Detection And Text Visual Analysis Based On Temporal And Emotional Features

With the development of the internet and e-commerce,the generation of user comments has greatly affected users’ shopping decisions.At the same time,facing the increasingly fierce competition in the e-commerce industry,the development of review website has a certain impact on the enterprise’s online marketing,so the content and the number of online reviews can often reflect the problems of the store,therefore,it is necessary to study the number of user comments and the sentiment analysis of the text.However,in the present research process,the content and quantity anomalies of the commentary text are neglected.Aiming at the above-mentioned problems,the following aspects are mainly studied:Firstly,aiming at the abnormal fluctuation of the time series curve of the number of comments,the Prophet model was introduced to identify the abnormal number of comments from the time dimension.Due to the influence of seasonal and holiday factors on the abnormal number of comments,the flexibility parameters of seasonal and holiday items in the model were adjusted to the optimal by means of cross validation.Considering the individual differences of holidays caused by regional factors,the holiday variables and the impact time factor of holidays were dynamically adjusted.Experiments were designed to evaluate the influence of parameter adjustment on model accuracy,solve the problem of poor identification of abnormal results caused by inaccurate model parameters,improve model fit,and reduce anomaly detection error.Secondly,the T-textblob model was proposed for aspect based sentiment analysis of the keywords in the review text.Considering the importance of keywords and their emotional intensity on the whole review text,the TF-IDF model and Textblob model were combined to construct the T-textblob model containing double influence factors of weight and emotional tendency to extract strong emotional keywords and their corresponding adjectives,and analyze the text emotional tendency.To solve the problem that the current review text analysis only focuses on the keywords themselves and ignores their emotional tendency and emotional intensity,key information and consumer attitudes during abnormal periods can be quickly obtained through the words extracted by the model.Finally,the number of abnormal reviews identified by the Prophet model and the emotional tendency classification results of the Textblob model were visualized on the comment line-bar chart,so that users could quickly obtain the abnormal time period and the overall trend of the number of reviews of the store.The words and their corresponding adjectives extracted by T-textblob model are depicted from the dimensions of sentiment tendency and sentiment intensity with the help of word cloud map,river map,matrix tree map and relationship graph,so as to effectively analyze the sentiment of the review text during abnormal periods.Add case studies and user feedback to validate and gradually refine the visual analytics view.