Research And Application Of Personalized Recommendation System Based On Multidimensional Contexts

Personalized recommendation,as an effective information filtering method,has been successfully applied to E-commerce,music and movies fields.Traditional recommendation systems(such as content-based and collaborative filtering)tend to use relatively simple and efficient user models.However,in most application scenarios,there is a certain correlation between different context factors.The recommendation system needs to integrate multiple different types of contexts.Therefore,this paper makes study of score prediction and top-k recommendation problems from different stages of the recommendation system.The main work and contributions of this article are as follows.(1)A context recommendation algorithm based on community and individual influence is proposed.First,an algorithm of context-level community detection is proposed in combination with user,project,and multidimensional context information,which effectively accomplishes pre-filtering of contextual information.Then,the individual influence is incorporated into the recommendation calculation,and a context recommendation algorithm(IIBER)based on community and individual influence is proposed.The intrinsic relationship between the user and the project is effectively tapped,and the unstable recommendation caused by the dynamic change of the community is optimized.Finally,simulation results on three real data sets show that IIBER is superior to the benchmark methods in both F-measure and diversity.(2)A method of context-aware recommendation system based on matrix factorization and expansion is introduced,namely the context sparse linear method(CSLIM),and the SLIM algorithm based on context score deviation is discussed.Then,a context recommendation algorithm based on GCSLIM and context factor stacking bias,DE-GCSLIM,is proposed,which scoring is bias of implicit learning context stacking.Finally,the simulation experiment selected three types of real context datasets and compared the multiple indicators with the reference context recommendation algorithm.The results show that the overall performance of DE-GCSLIM is better than other algorithms.(3)From the point of view of e-government service recommendation,a context recommendation algorithm suitable for specific domain scenarios is proposed,and a framework with certain real-time calculation characteristics is designed to meet the practical application.First,investigation of the characteristics of e-government services is performed,feature modeling based on multi-dimensional contextual information of users and tasks is executed.Second,an e-government recommendation algorithm(CAS-UR)is proposed based on community and association sequence mining,and the performance and simulation results of the recommendation algorithm are discussed and analyzed.Finally,a recommendation framework that is suitable for practical application scenarios is designed,and the data development flow diagram and the recommended application result display are visually displayed.