Construction Of Topic Model Based On Keyword Vector And Visualanalysis Of Comment Data

With the maturity of users’ social networks,the reference value of users’ consumption experience for merchants is increasing.However,the review data in review websites had problems such as large amount of data and long time span,which reduces the consumption experience of customers.Therefore,in view of the high dimensional and multiple characteristics of the review data,this paper studies the visual analysis method for the keywords of the review data from the perspective of the topic of the review text.Firstly,aim at the problems of incomplete evaluation data,excessive noise or outliers,inconsistent parts of speech and so on,a method of abstraction and processing of user evaluation data is proposed.Text cleaning of comment data set based on string matching method removed the content weakly related to the evaluation topic and fills the missing value.A user evaluation data stop word table is constructed,and based on this,unnecessary dimensions are reduced in topic extraction.At the same time,based on the method of morphology reduction,the deformation of tenses,single and plural words is eliminated,and different word morphology in single paragraph text is restored.Secondly,aim at the problem that the traditional topic model and word vector model cannot take into account the global prediction and the correlation between local words and words,a topic model construction method based on the keyword vector of review data is proposed.First,construct topic extraction and keyword representation methods: Extract the topic of comment data based on LDA model as the basis of topic model data.Then,the CBOW model in Word2 Vec is optimized based on the negative sampling method to transform the comment data topic into word vector.Second,a keyword vector topic model of review data is constructed,which integrates LDA and Word2 Vec.The distance between keyword vectors and the word vectors in user reviews is calculated based on cosine similarity,and Page Rank method is used for iteration,so as to improve the accuracy of the topic model and support text visual analysis.Third,verify the feasibility of the topic model construction method of keyword vector: compare with the traditional algorithms of LDA and TF-IDF,the accuracy of the model is evaluated from the three perspectives of accuracy,recall rate and F1 value,indicating that the method described in this paper can more accurately obtain the keywords of user comment text used in the visualization system.Then,combined with the topic model construction method of keyword vector,we design classification,recommendation and temporal graph representation methods,so that users can perform visual analysis from multiple perspectives and reduce the burden of visual analysis.Finally,evaluate the feasibility and practicality of visual analysis through case analysis,in-depth interviews,and user feedback.This system can be extended to visual analysis of comment texts in other fields,explore hidden patterns in comment texts.