Optimization Of Personalized Recommendation Algorithm Based On Score Prediction

In recent years,with the diversification of network development,the problem of information overload is becoming more and more serious.Classified directory and search engine are two kinds of mainstream solutions to the problem of information overload.However,with the continuous expansion of the Internet scale,the classified directory website can only cover a limited number of popular websites,while the search engine needs users to actively provide keywords to find information,which has great limitations when users do not know their own needs.At this time,personalized recommendation technology came into being.Personalized recommendation technology can analyze and mine users' potential preferences,which is an important means to solve the problem of information redundancy and improve the efficiency and quality of people's information acquisition in the era of big data.This paper studies the personalized recommendation algorithm based on score prediction,and optimizes the algorithm from two aspects: neighborhood based and learning based.The "score" here refers to the general score,that is,the user's preference score for items,not just the narrow user score.Experimental results show that compared with the algorithm before optimization,this paper achieves personalized recommendation with higher accuracy and robustness.The research contents and contributions of this paper include the following aspects:1.In the neighborhood based scoring prediction recommendation,aiming at the shortcomings of traditional user based recommendation algorithms,such as ignoring the necessity of user evaluation normalization and the weighted user attributes in the calculation of user similarity,resulting in inaccurate recommendation,this paper focuses on the improvement of user similarity calculation,normalizes user rating,improves the Jaccard similarity calculation method of user rating,adds user attribute similarity,and combines it with user rating similarity linearly.Finally,an optimized user based collaborative filtering recommendation system o-recommend is proposed.Through experiments,the necessity of normalization of user score,improvement of calculation method and weighted user attributes in thecalculation of user comprehensive similarity is verified,and it is proved that the improved algorithm improves the recommendation accuracy2.In the learning based score prediction recommendation,aiming at the shortcomings of the traditional recommendation algorithm,which is mainly based on the explicit feedback information,that is,the method of predicting the independent selection of goods by users,this paper proposes a commodity recommendation system based on multi-type implicit feedback(MIF).MIF takes commodity pair rather than single commodity as the basis of user interest modeling,and introduces the important implicit feedback information of commodity comparison information.Bayesian ranking model is used to model the comparison relationship among commodities and give different confidence to different types of implicit comparison.After that,the effectiveness of the improved algorithm is verified by experiments.