Fine-grained Analysis Of Product Reviews Based On Hierarchical Topic Model

With the development of e-commerce,more users will post comments after shopping to express their attitude towards products.Much useful information is contained in the commentary text.Consumers can understand the quality information of products through the review text,and businesses can help optimize their products and improve their competitive advantages according to user reviews.At present,using text mining method to analyze user emotion has become an important research direction.However,most existing studies calculate the overall emotional polarity of the text based on the emotional words in user comments,and the analysis strength is relatively rough,and it is impossible to establish sentiment analysis for fine-grained attributes of products.This study proposes a fine-grained analysis method for product reviews based on h LDA,and takes the user reviews of JD.com as an example.The main work is as follows: First,use a crawler to obtain user reviews of Lenovo Xiaoxin series laptops from the JD website and preprocess them.Secondly,construct the product attribute hierarchy tree.For the crawled user comments,a text-level topic tree of comments is obtained based on h LDA.According to the PMI algorithm,a hierarchical tree of product attributes with weights is further refined.Then,the sentiment-related words corresponding to the product attributes are extracted by analyzing the sentence structure,and their sentiment scores are calculated according to the sentiment dictionary expanded in this study.Finally,a product evaluation system is constructed to help product developers to obtain demand and analyze product satisfaction.In this paper,the h LDA is used to grasp the product attributes that consumers focus on in product reviews,and the following conclusions are drawn: the emotional polarity value of the screen,appearance and performance is high,which brings a better user experience to users;The emotional scores of power portability and cooling and quietness are low,indicating that the needs of users are not well met.According to the above research results,this paper puts forward four suggestions: Keep abreast of market feedback and improve product attributes of consumer concern;pay attention to the unmet needs of users and improve product design;continue to maintain competition for product attributes with high user emotional tendencies;segment customer groups and tap niche needs.This will help product developer to grasp the merit and demerit of their own products and explore the direction of product development.