The Design And Implementation Of User-Behavior Time-aware Music Recommendation Algorithm

In the era of Internet information today,the amount of data growing,and the problem of information overload gradually prominent.The mission of music recommendation system is to recommend appropriate music to users,and help users find the music they really interested in more easily.In order to get a better recommendation,researchers gradually turned their attention to the impact of contextual information such as time,place and mood on the music recommendation system.However,in the application of time factors,most of the researches only consider the transfer of users’ long-term interest,and did not take the change of users’short-term preference into account.In fact,in a short time(such as:one day),the user’s listening preferences are also changing over time,compared to the transfer of long-term interests which are more obvious,the change of short-term preferences is usually more covert.Therefore,this paper focus on mining the behavior of the user’s listening music,and applying it to get better recommendation accuracy.In this article,we are committed to the study of the law of users listening music,that is,when will a user listen to music and what kinds of music will he listen to,and make judgments of the different music demands of users at different time,then generate fine-grained personalized recommendation.This paper presents a new User-Behavior Time-aware Music Recommendation(UBTMR)model which produces music recommendations that meet the needs of users at different times.In this paper,the recommendation system and the music recommendation system are briefly introduced at first,and carry out the works according to several major problems of music recommendation system currently:1)In the music recommendation system,the law of user’s listening preference changes in the short-term has been rarely studied and made full use.In this paper,we propose a new recommendation method which consider the law of users’ listening habits in the fine-grained time,and excavate the circulatory rules of users’ listening habits in the short time(such as one day)and apply them to the recommendation;2)The user’s rating records are usually far less than the user’s listening records in online music systems,and most recommendation systems are recommending based on users’ ratings records,which makes the system suffers from data sparsity challenges.Therefore,this article uses users’ listening records as a basis for judging the degree of the user preferring the song,then calculate and recommending according to these listening records;3)In view of the fact that the recommending intensities between music are differently in different periods,this paper proposes the concept of asymmetric recommendation degree,instead of the symmetry similarity which commonly used in the recommended operation.And take the users’ activities,songs’ popularities and other characteristics,which are lack of comprehensive consideration in existing music recommendation systems,into account to enhance the recommended accuracy;4)During the operation,the classical collaborative filtering algorithm finds the same amounts of nearest neighbors for each song,but does not considering whether this is the most suitable range of nearest neighbors for this song.In this paper,we consider the influence of songs’ popularities on the amount of nearest neighbors of each song,and utilize k-means clustering to gather similar songs together to find nearest neighbors.Experiments show that take both user’s long-term and short-term listening habits into account make the law of users listening music be excavated better,which can make the recommendations more detailed and accurate.At the same time,take the users’ activities and songs’ popularities into consideration makes the values of the Recall,Precision and F1 improved.The results of the non-uniform slice experiments in the last show that this method is more suitable for the actual time.