Research On Recommendation Algorithm Based On Multi-factor Implicit Feedback Information

With the rapid development of information technology,especially the rapid rise of Internet applications,generating a lot of data,people are facing more and more serious information overload.On the other hand,the development of technology also prompted the emergence and development of recommended systems,recommended system technology to a certain extent can solve the information overload problem.By studying the status quo of the personalized recommendation system,it is found that the traditional recommendation algorithm is recommended to the user by using explicit feedback information while ignoring the rich implicit feedback information,which leads to the low accuracy of the recommended system.In view of this situation,a comprehensive recommendation algorithm based on the combination of user behavior and behavior time,item popularity and user activity is presented.Firstly,the typical user behavior of hidden users in common websites is classified and analyzed,and the preprocessing method of converting user behavior data into scoring matrix is given.On this basis,in view of the relative user behavior,the recent user behavior is more able to express the user's interest and hobby.Put the time attenuation factor into the user score matrix and form a time weighted score matrix to solve the traditional recommendation system ignores the time factor in the calculation process.Time-weighted scoring matrix can increase the impact of recent user behavior on user's interest preferences,reduce the impact of long-standing user behavior on user's interest preferences,which reflect the different ratings of users at different times.Then,the matrix factorization method is used to realize the recommendation score prediction.For the vacancy values in the scoring matrix,according to the negative sample selection strategy in the one class collaborative filtering problem,the implicit information of user activity is defined on the basis of the existing objective function of using the implicit feedback information of the item popularity.And use this information to improve the objective function of matrix factorization.The parallelization of the matrix is performed on the implementation using the alternating least squares method to improve the recommended computational efficiency.Finally,this paper introduces the comprehensive recommendation process of the implicit feedback information,such as user behavior and behavior time,item popularity and user activity,which is proposed in the music data set last.fm.By comparing the traditional scoring matrix factorization recommendation method and the matrix decomposition recommendation method which only incorporates the item popularity,the experimental results show that the improved comprehensive recommendation method can approximate the user score matrix more accurately in the prediction and improve the Top-N recommendation efficiency.