Design And Implementation Of Rating Prediction Algorithm Based On User Feedback Information

The era of intelligence has arrived,and the overwhelming information make people feel at a loss.How to accurately locate the information of interest is particularly important.In this context,the recommendation system came into being,which is very helpful to solve the problem of information overload.Rating prediction is an important and meaningful research area for the recommendation system,which recommends items to the user by predicting the user's rating to the item.However,the sparseness of the rating data has led to the unsatisfactory results of many prediction algorithms.Therefore,how to improve the accuracy of the rating prediction algorithm has become a challenge for the recommendation system.In this paper,we alleviate the sparsity of the rating data by establishing prediction models based on user attention and based on the importance of the user's rating data,and study the weighted model learning based on user rating data.Main contributions are summarized as follows:· Rating prediction based on user attention: In an e-commerce website,the user cannot pay attention to all the items but selectively attend the items interested in,so the user's attention to the items can implicitly feedback the user's interests.In this paper,we propose prediction algorithm based on user attention named UAD and UADplus which greatly alleviating the data sparse problem by transferring the user's attention to the target field based on collaborative filtering technology and matrix decomposition technology.· Rating prediction based on the importance of user rating data: The user's rating times can implicitly reflect the user's interest and learn degree.Therefore,in this paper,we distinguish the ratings from different user-item groups by weighting the user's rating value based on the user's rating times.Rating prediction algorithm based on the importance of user rating data named Record CF is proposed by collaborative filtering,which reduces the negative impact of the sparseness of rating data· Weighted model learning based on user rating data: Most classical prediction algorithms learn model parameters by minimizing the loss caused by the error between the predicted rating and the true rating.However,this method does not distinguish the importance of the rating from different user-item groups in the loss function.In this paper,we improve the effect of the model learning by studying the weighted model learning process which weights the error between the user's true rating value and the predicted value.