Design And Implementation Of Toy And Game Recommendation System Based On English User Evaluation

With the rapid development of the competitive Internet industry,the Internet data has become more and more complicated,and the massive amount of information flooding into the public has caused the information overload problems.To solve this problem,recommendation systems have been created.In the field of e-commerce,current recommendation systems mainly user clicks,browsing,product ratings and other information to make recommendations,which ignores the importance of user evaluation.So,there are two main problems:firstly,users still face the problem of review overload and cannot quickly obtain value information to make purchase decisions when selecting products.Secondly,the current product recommendation system does not combine user reviews to make recommendations,which is lacking from the perspective of personalized recommendations.This thesis focuses on the recommendation system for toys and games,to improve the two existing problems.The main work is as follows.1)To address the evaluation overload problem,this thesis designs a user evaluation recommendation list.Firstly,introduces the topic model.Topic analysis is conducted to obtain the topic distribution of evaluations and topic distribution preference of users.Calculate the distance between topic distribution of each evaluation and topic distribution preference of users.2)To address the problem of missing dimensions in recommendations,this thesis adds user evaluations to the recommendation process.Firstly,based on the machine learning sentiment classification technique,classifies the sentiment of user evaluations.IN the sentiment classification experiments of this e-commerce recommendation system,the BERT model is introduced to compensate for the problem of multiple meanings that cannot be solved by machine learning algorithms in complex sentences,making the meaning of sentences more clearly represented.The experiments show that based on the BERT-LSTM model,the user evaluation classification effect is significantly improved.By implementing product recommendations based on user evaluations,finally optimizes the rating values to obtain new ratings according to the user rating classification results.Ultimately,based on the traditional collaborative filtering algorithm,the change in the effect of the algorithm after adding extended ratings is investigated.The distance between the user’s topic distribution preferences and the topic distribution characteristics of the product is calculated to optimize the recommendation list.The experimental results show that the recommendation method incorporating user evaluations outperforms the original collaborative filtering algorithm in both MAE and RMSE metrics,with a 38%decrease in MAE and a 34%decrease in RMSE,which verifying the effectiveness of the evaluationbased recommendation method.3)Based on the results of the above two parts of the study,the presentation of a toy and game recommendation system based on English user ratings was implemented,and the functional testing and front-end demonstration of the system was completed.