The Analysis And Application Of Relations Between Item Popularity And Recommendations

With the development of internet, people turn to network for some activities. But, user can’t select interested items easily from these large amount of information with the increase of internet application and service. So, personalized recommendation was necessary for business website to increase benefit. Recommendation efficiency was weighed by accuracy of interest prediction, clicked rates and sale quantity in business system.But, the existed recommendation algorithm made interest prediction based on user’s historical behavior, and it ignored the effect of environment and group behavior to user rating and selection. But, social psychology theory showed that user interest always changed and it was easy to be affected by environment and group opinion. For this problem, personalized recommendation algorithm based on popularity was proposed after analyzing the influence of popularity on user interest. It mainly included the following contents:(1) Study popularity effect on user behavior according to social psychological phenomenon such as conformity. Analyzed the relation of interest change and group opinion. The influence of popularity on recommendation based on user rating and behavior was researched by experiment, and this paper summarized the theory of popularity in recommendation.(2) Adjust rating prediction by popularity. Interest prediction rating was adjusted to improve recommendation accuracy by the changing relation of popularity and rating. Comparison experiment was given based on MovieLens dataset, and results showed that this method can improve prediction accuracy.(3) Propose rank algorithm of recommendation based on popularity. This article mixed general recommendation rank and popularity factor of item as comprehensive standard. Because, prediction error defect or non-personalized defect would be induced respectively by taking past behavior standard or popularity factor as the weighted standard of recommendation. Experiment results indicated that recommendation efficiency can be improved with this method.