Research On Comprehensive Sorting Algorithm Of Fresh Food E-commerce Products Based On User Evaluation

As the most popular shopping medium for users in recent years,e-commerce has served hundreds of millions of users and is accompanied by thousands of daily transactions and visits.In the current online shopping environment,in order to improve the user experience and collect user shopping opinions,all major e-commerce platforms use the historical review data of a product as an additional attribute of the product,which on the one hand can help consumers understand the real situation of these goods,simultaneously on the other hand can also help the e-commerce platform to obtain feedback from users.However,the vast amount of review data has uneven quality and mixed information,making it impossible for both the user and the platform to easily obtain effective information,so it is important to analyze the overall sentiment of the review text to obtain effective information.Therefore,it is important to analyze the overall sentiment of review texts to obtain effective information.Therefore,how to integrate the sentiment of review texts into the ranking of products in order to recommend better products for consumers is the focus of this thesis.To address the problems of current product ranking,this thesis proposes a comprehensive ranking method under sentiment analysis based on user reviews,and uses this ranking as the basis for product recommendation.For the review text sentiment analysis problem,the existing BERT-based sentiment analysis model is used as a benchmark,and the semantic weighting of the global text is achieved by introducing an attention mechanism in the semantic layer to avoid the problem of missing partial position representation of the text.Experiments on the constructed e-commerce platform text dataset show that the F1 value of the improved sentiment analysis model is 86.4%,an improvement of 1.6% compared to the benchmark model.Meanwhile,in order to capture the co-occurrence relationship between global words in the corpus and strengthen the connection between different text sentences,this thesis constructs the constructed corpus into a large heterogeneous text graph,and then learns each node representation in the graph based on graph neural network under the supervision of sample labels,so as to obtain the node features with structural information.Finally,the weighted semantic information is stitched with the global relational information to obtain the final review text representation.It was found through the experimental data that F1 value of the sentiment analysis model,which combines semantic and global relationship information,is 88.5%,an increase of 3.7% compared to the baseline model.The other factors of the product are then coded using feature engineering,combined with the sentiment value factor of the product evaluation information,and weighted according to both factor analysis and principal component analysis.This new ranking method not only considers the original product parameters,but also incorporates the user’s evaluation sentiment into the ranking decision process.Finally,the new ranking results are shown in graphs,the possible factors for generating the ranking are illustrated by visual analysis.The thesis proposes a sentiment analysis model that incorporates weighted semantics and global text relationships for the review text of e-commerce products,and then designs and implements a fresh food ranking method that combines the sentiment tendencies of user evaluation text,making it possible for users and the platform to quickly obtain effective information,improving the efficiency of user shopping while also enabling merchants to pay more attention to the quality of products and after-sales,providing consumers with high-quality products and more thoughtful It also enables merchants to pay more attention to product quality and after-sales service,providing consumers with high-quality products and more considerate service.