Recommendation Algorithms Based on Implicit Feedback and Weighted User Preferences
In view of the unreasonable division of positive and negative samples,ignoring the frequency of user operations and failing to accurately model user preferences,we propose a recommendation algorithm based on implicit feedback and weighted user preferences(IFW-LFM).The algorithm considers the relationship between user operation frequency and positive and negative sample division,learns and improves wALS algorithm,and re-mines potential positive and negative samples from missing values according to user operation frequency.It sets the samples with user operation frequency greater than 1 as positive samples and those with user operation fre-quency of 1 or 0 as positive or negative samples,eliminating the need to artificially introduce negative samples;defines the confidence level according to the influence of user operation frequency on the degree of user preference,specifies the user preference and applies it to the framework of the hidden factor model;uses the user listening to song start and end time,listening duration and other implicit feedback data to improve the utilisation of implicit feedback samples.The results of the comparison experiments on two music datasets illustrate that the accuracy,recall and NDCG values of the proposed method have a maximum average improvement of 45.81%,83.83%and 60.33%respectively compared with the five classical implicit feedback algorithms of UserCF,ItemCF,LFM,BPR and SVD,which has better recommendation results.
recommendation algorithmsimplicit feedbackfrequency of operationuser preferencesmusic recommendation
万方数据
维普
