Research Of Recommendation Algorithm Based On User Curiosity

The Recommender System has great significance and value in both the application field and the research field.The Recommender System according to whether or not to use side information for recommendation can be divided into a Recommender System that does not use side information and a side information based Recommender System.The former only uses the user's past historical behavior data to simulate the user's preferences and will faces the problem of data sparsity.Although the latter uses side information to alleviate the problem of data sparsity,it also has the problem of insufficient use of side information.Based on the Top K recommendation task in the Recommender System,this paper focuses on the flaws of incomplete utilization of social network side information in the social-based Recommender System.Existing social-based Recommender Systems using social network side information generally ignore the user's curiosity in the social environment and the dynamic diffusion of information in the social network.As the dynamic diffusion of information in social networks will strengthen the user's curiosity,at the same time,according to the curiosity-driven theory of psychologist Daniel Eills Berlyne,the user's curiosity can promote the user's exploratory behavior,thereby affecting the user's decision-making.Therefore,this paper will improve the existing social-based Recommender System from the perspective of user curiosity.In order to explore the impact of user curiosity on recommendation,this paper refers to the executable steps to quantify user curiosity proposed by psychologist Berlyne to simulate user curiosity.Specifically,this kind of user curiosity aroused by user's uncertainty.At the same time,this paper attempts to use psychology-inspired viewpoints to incorporate the curiosity induced by uncertainty stimuli into the existing Recommender System,and propose a Curiosity Enhanced Bayesian Personalized Ranking model(CBPR).This paper proposes two opposite user preference hypotheses to simulate the impact of user curiosity on recommendations,which also represents the impact of user curiosity on user choices.The two hypotheses correspond to the CBPR-1 model and the CBPR-2 model,respectively.The former considers the user curiosity has a positive effect on user choices,and the latter believes that user curiosity has a negative effect on user choices.This paper conducts experiments on two real social recommendation datasets,and compares them with a variety of related start-of-art recommendation methods.The accuracy indicators such as Precision,Recall and F1 Score are used to verify the experimental results.The experimental results show the CBPR-1 model is more in line with the reality,that is,user curiosity has a positive impact on user selection.At the same time,the experimental results show that the curiosity-based Recommender System proposed in this paper is better than the existing related models in Top K recommendation tasks.