Content And User Preference Learning Based Personalized Item Recommendation Model

With the development of information technology and Internet,the information overload problem is getting worse and it becomes more difficult for users to find information they really need,personalized recommender system has been proposed to solve this problem.The vigorous development of E-commerce makes the personalized recommender system become one of the biggest application fields.The academia and industry focus on how to provide users personalized recommend lists from the mass commodities efficiently.Because of the large amount of implicit feedback and easy to obtain,many scholars put forward a lot of models based on users’ implicit feedback rather than users’ explicit feedback.BPR(Bayesian Personalized Ranking)is the most popular model.The existing implicit feedback-based models can’t be directly applied in E-commerce item recommendation,because the implicit feedback in E-commerce is vary large and exists inherent noise data and has unequal amount of positive and negative samples,which will make the recommendation results low precision and efficiency.So in this article,we focus on the E-commerce implicit feedback,and propose a content-aware and item-set-based preference learning personalized item recommendation model in order to improve the accuracy and efficiency of recommendation,and the main contribution are listed as follows.1)Improving BPR via content-based hybrid sampling strategyExisting series of BPR improved models suffer from the noisy data and tends to select low quality,incomparable and low credibility item pairs for model learning and leads to slow convergence and log accuracy.To copy with these problems,we propose a content-based hybrid sampling strategy,which considers item-pair information value,item category information and user potential preference together,in order to select high quality,comparable and high credibility item pairs for model learning.Then we put the sample strategy into BPR model framework,and propose an improved BPR model via content-based hybrid sample strategy.The experimental results show that the proposed model could lead to speeded and high accuracy recommendation.2)A content-aware and item-set-based preference learning personalized item recommendation modelWith the large amount implicit data and abundant commodity content information and the user online shopping behavior characteristics on E-commerce,the definition and hypothesis of existing BPR improved models can’t describe user relative preference accurately.We propose a content-aware and item-set-based preference learning personalized item recommendation model.The model accomplish recommendation by using hybrid sampling strategy to select item-pairs,and then using heterogeneous implicit feedback,commodity information and user online shopping behavior characteristics to definite user relative preference more accurately,and then using user potential preference to learning the confidence of each user relative preference.The experimental results show that the proposed method could lead to speeded,stable and high accuracy recommendation model.