| Different recommendation algorithms have different requirements for the content or form of data.For example,content-based recommendation focuses on classifying users or products by label attributes.The more label attributes,the finer granularity of label,the higher accuracy of recommending;The Rule-based recommendation focuses on discovering the association rules between commodities,but the discovery of rules is difficult.Especially when the lack of commodity types and transaction volume,the predicament of generating rules is even obvious;the collaborative filter-based recommendation focuses on user social interaction Network or other connections,for relatively independent users in the system,this method is often powerless.Therefore,when there are few attribute tags between users and products,there is not enough interactive information between users and products,and the social relations between users are lacking or not public,it is difficult for the currently used recommendation systems to achieve effective recommendations.The Mass Diffusion and Heat Conduction recommendation algorithms both based on bipartite graphs,which are the results of cross-research between physics and computer science.The two do not depend heavily on user data,and only need to analysis the transaction between the user and the product.which can effectively bypass the cold start and data sparsity problems.Then according to the principle of diffusion,the purchase probability of each commodity is calculated.The MD recommendation algorithm is more prominent in accuracy,while the HC algorithm is superior to novelty.The two have the same theoretical basis,so this paper proposes a mixed material diffusion and heat conduction recommendation scheme.which studies the relationship between user activity and product accuracy and diversity,so the user activity is used as a mixed weight.parameters,and proposed a user activity calculation method suitable for this project,that is,calculating the user activity by analyzing the user’s consumption records.The scheme points out that when the user activity is high,the diversity of recommendations is emphasized,and the hybrid scheme is biased towards the results of heat conduction recommendations,and vice versa.The test results and consumption analysis results show that the proposal meets the expected requirements for completing product recommendation,and at the same time increases user stickiness,improves user activity,and accomplished the established achievement. |
