Online Reviews Clustering Analysis And Recommendation Based On LDA Topic Model

With the development of information technology,massive online commentary emerged in the online shopping platform,which contained lots of product information.It was a thorny issue for consumers to quickly refer to the information which they were interested in.The text-based clustering method is an important way to alleviate this problem and has a broad application prospect.The online reviews in e-commerce environment have the characteristics of uneven quality,sparse features of short texts,and weak ability to describe information.This causes the text to be vague and content confusing,which restricts the application scope of online reviews.The text clustering can reveal the consistency of the text content,so as to discover the common information contained in the same category,which helps the re-organization and secondary application of the online commentary.For this reason,the focus of this paper is on the cluster analysis of online review and the recommendation method of online review.We aim to find a suitable clustering model for online review short texts,and assist consumers to obtain information more effectively by selecting important comments..The main research content of this paper is as follows:1.Aiming at the problem of ignoring deep semantic information in the traditional spatial model,this paper presents a vectorized representation of online reviews using the LDA topic model and explores the most suitable short text clustering method under the LDA model.The experimental results show that the LDA model can accurately capture the topic information in the review,making the clustering results more reasonable.And in each clustering algorithm,spectral clustering algorithm based on graph segmentation performs best.2.In order to provide consumers with more important comments,we introduces the complex networks to compute the value of online reviews.Based on the LDA algorithm,an undirected complex network is constructed with each comment as a node and the similarity between comments as the weight of the edge.Community classification is based on the category information of spectral clustering results.Since each community contains different topic information,a new review set is formed by selecting the community's most moderated review as an important comment.The experimental results show that the method presented by this method has a more obvious theme of comments,which can help consumers find the content more faster.3.Considering that consumers will focus on the emotional tendencies of purchased users in browsing online reviews,this article uses supervised machine algorithms to classify emotions,and recommends positive and negative sentimental reviews.Through the introduction of multi-feature fusion and ensemble learning methods,the text is sentimentally calculated,and the content of the review is combined with the category information after spectral clustering.Experimental results show that the effect of text sentiment classification algorithm proposed in this paper surpasses the classic SVM algorithm and other unsupervised algorithms by more than 10%,and forms a new comment set with more emotions.