Research On Personalized Recommendation System Of Ecommerce Website Based On Commentary

Data mining technology is a multidisciplinary subject which involves artificial intelligence,machine learning and other fields.With the rapid development of e-commerce websites,there is a large number of website user data generated every day.How to excavate potential valuable information from those data and provide decisions for people's life and work is a problem worth studying.In this thesis,the author creates a personalized recommendation system based on comments for e-commerce websites selling characteristic agricultural products by means of information.The system is able to recommend characteristic agricultural products to website users,improves the efficiency of users' purchase and reduces the selling cost of the website,which has strong practical value.Based on users' comments,personalized recommendation of characteristic agricultural products is realized.The main research contents in this thesis are as follow:(1)User comments collected from characteristic e-commerce websites are pre-processed and their features are extracted.Based on TF-IDF based weight calculation and vectoring representation,a structured set of comments on characteristics of featured agricultural products is obtained;(2)Combining advantages of K-means algorithm and BIRCH algorithm,traditional K-means algorithm is improved,and the robustness of the constructed core tree is enhanced.Then the improved algorithm is applied to the clustering of the set of comments on characteristics of featured agricultural products;(3)According to the clustered set of comments on characteristics of featured agricultural products and HowNet emotional dictionary,a polarity dictionary for characteristic agricultural products is created.Next,based on the polarity dictionary,an emotional analysis is carried out to analyzed the set of feature comments,so as to obtain users' ratings of characteristic agricultural products and overall forecast scores;(4)In line with HowNet emotional dictionary,similarity computing and collaborative filtering recommendation technology are adopted to build a personalized recommendation model based on user comments,so as to achieve personalized recommendation of characteristic agricultural products;Based on traditional collaborative filtering recommendation technology,the author applies data mining technology based on user reviews to personalized recommendation.Then through a combination of text preprocessing,featureextraction,weight calculation,K-means clustering,emotional analysis,collaborative filtering and other technologies,a personalized recommendation model for characteristic agricultural products is constructed,and the validity of comment-based collaborative filtering recommendation method for MAE and RMSE evaluation indicators is verified.Compared with traditional collaborative filtering methods such as SlopeOne and SVD++,comment-based SlopeOne decreases by 6.75% in MAE and0.57% in RMSE,while comment-based SVD++ decreases by 6.70% in MAE and0.63% in RMSE.