Topic Evolution And Prediction Based On User Comments

In recent years,with the rapid development of e-commerce,people are accustomed to online shopping and post a lot of comments after receiving the goods.Studying online product reviews can obtain a lot of valuable information,and accurately grasp customer consumption needs and consumption preferences for merchants and enterprises habit etc.From the customer’s comments on the product,it is possible to determine the customer’s preferences,what are the advantages,selling points and defects of the product.This motivates text mining of online reviews and evolution and predictive trend analysis of review topics for enterprise reference applications.However,how to dig out the hot comment topics from the massive product reviews and analyze and predict the evolution path is an urgent problem to be solved.In order to solve the above problems,we use the LDA topic model,which is the most widely used in the field of topic identification and evolution analysis,and has a better effect.The main research contents of this paper are:(1)Review the domestic and foreign literatures on text review mining,the theoretical development and application status of topic models,topic evolution and sentiment analysis.Select topic models and corresponding research methods through literature analysis combined with the problems to be solved,and formulate the research technology roadmap of this paper.(2)Introduce the development of LDA topic model and some related knowledge of basic probability theory.Then,build an LDA model and train it,use perplexity,KL divergence,JS divergence,and average similarity to evaluate the LDA model,and improve the perplexity based on the word segmentation characteristics of this paper,so as to determine the number of topics more accurately.The evolution of the topic model is divided into topic strength evolution and topic content evolution.Using the topic strength model and JS divergence to judge the topic evolution path,etc.,the gray prediction model and its evaluation index posterior difference ratio are derived.(3)Use the Python programming language to call the third-party jieba library to segment the comment text,call other third-party libraries to perform word frequency statistics,get the LDA topic word distribution and word distribution frequency,and summarize the topics according to the topic words and semantics.In this paper,the frequency of the subject word is calculated as the weight as the parameter of the prediction model,and the average prediction value of the prediction result of the original prediction model is increased by 2.192%.This method provides a new idea and method for the subject prediction of the subject model.Finally,in order to verify the validity of the model used and the gray prediction model,the collector is used to collect the comment text of the computer products of Jingdong Mall,and the sentiment analysis is carried out on the comment text to obtain the word cloud map,and the LDA topic model is run to obtain the optimal Number of topics,topic visualization,model evaluation.Then,the subject word frequency obtained by the model is combined with the gray prediction model,and the Matlab tool is used to predict the subject evolution,and the posterior difference ratio is used to evaluate the prediction model.