Bayesian Personalized Ranking Model With Multi-type Implicit Feedback Confidence

With the development of the Internet and big data technology,the problem of information overload has become increasingly serious.The recommendation system is an effective tool to solve the problem of information overload.The current personalized recommendation technology has been widely used in e-commerce platforms and social networks.Among these recommendation technologies,scholars have proposed many classic collaborative filtering recommendation algorithms based on explicit feedback.However,the collaborative filtering recommendation algorithm based on explicit feedback has problems such as less data,limited acquisition mode and difficulty in obtaining,which makes the recommendation performance poor.Therefore,in order to make it easier to obtain a rich amount of data,a more diverse data based collaborative filtering recommendation algorithm with implicit feedback is becoming a hot topic.For the same user,the multiple types of implicit feedback generated have different effects on predicting user preferences.At present,most collaborative filtering recommendation algorithms based on implicit feedback only choose one or two types of feedback to implement recommendations.The problem of data sparseness is serious,which makes the recommendation result not accurate enough.If these multiple types of implicit feedback are applied to the recommendation algorithm,the data sparsity problem in the recommendation system can be alleviated.Therefore,this paper proposes Bayesian Personalized Ranking model with multi-type implicit feedback confidence(MTCBPR),which is aimed at solving the problem of data sparseness caused by users' single behavior mode of less interaction with existing items in the recommendation method with implicit feedback.Complement and fuse multiple behaviors of users and use confidence to measure their impact on the quality of recommendations.Through the logic regression and tree-based feature selection,the confidence of the data set is used to filter multiple types of implicit feedback to filter out more effective deterministic auxiliary feedback,and preference expressions of different types of deterministic auxiliary feedbacks are quantified according to confidence.Through the data processing and contrast experiments on the public data set sobazaar,the experimental results are obtained.The proposed MTCBPR algorithm is superior to the baseline algorithm in the evaluation indicators,which can further alleviate the data sparse problem and fully interpret the user's preferences and willingness to effectively improve recommendation performance.In addition,in order to verify the feasibility of the MTCBPR algorithm on other real data set,the third-party software obtains multiple types of implicit feedback from users on the Internet health service platform,including the user's appointment registration,online consultation of famous doctors,finding famous doctors and visiting famous doctors.These typical historical behaviors of users are organized into a dataset Topmd.The experimental results show that the MTCBPR algorithm still has a good recommendation effect in this scenario.