Research On Recommendation Algorithm Based On User Profile And Factorization Machine

The rapid developm ent of network techno logy has brought about the problem of information overload,and the em ergence of personalized recommendation system has effectively improved this problem.However,with the increasing dem and for personalized users,the deficiencies of the tra ditional personalized recommendation have g radually emerged.It mainly performed as the traditional personalized recommendation mainly depends on the u ser rating data,and treat the s coring prediction as the fi nal target of the recommendation which lead the p roblem of low recommendation quality,cold start and high data sparsity.It has been found that the real ization of accurate personalized recomm endation must be based on a profound understanding and characterization of user characteristics.User profile(UP)technology is a tagged user model based on the inform ation of users' social attributes and consum er behaviors,which is a labe l of user characteristics and information.The recommendation system can truly understand the user' s needs based on the information of user profile,thereby ach ieving high-accuracy recomm endations.The recommendation system can truly understand th e user's needs based on the inform ation of user profile,so as to achieve accurate recommendation.The factorization machine is a model developed by matrix decomposition to solve the problem of feature combination under sparse data and has high prediction quality for large sparse matrices.Therefore,the cor e work of this paper is to build user profile,and then combine the advantages of user profile and factorization model to propose a recommendation algorithm,so as to effectively solve the data sparse probl em and improve the recommendation quality.The main work and contributions of this paper are as follows:(1)In view of the fact that the traditional user profile method does not consider the strong correlation between user attributes and user interests,th is paper proposes a user profile modeling method combined with topic m odel and user attribute.Firstly,we extract information of user's hidden natural attribute and interest preference from the user's comment.Then,through the in-depth re search of topic model,we propose the LDA-JSD(Latent Dirichlet Allocation-Jensen Shannon Distance)m ethod to cluster users by user interest in semantic level.Then we extract the characteristics of user groups,and obtain the relationship between the natural attributes and in terests of each user group center users,so as to m ake the user profile model m ore accurate and personalized.Finally we get an improved user profile model,providing personalized recommendation and precision marketing basis.(2)According to the tr aditional regression model of personalized recommendation is to consider each feature independently,ignoring the interaction between a lar ge number of features,eventually led to poor recomme ndation.This paper proposes a recommendation algorithm based on user profile and factor decom position(UP-FM).In this paper,we introduce the m ultidimensional features extracted from user prof ile into FM m odeling,combined with the advantages of FM m odel can feature interaction under the sparse data and high quality prediction,and according to the best features to use the discrete feature when the FM model is linear decomposition.It also proposes the concept of trigger topic words,which effectively links the po tential factors between the user prof ile and the recomm ended items from the semantic aspect,effectively solves the problem of data sparsity.Finally,it is proved that the UP-FM recomm endation algorithm proposed in this paper is effective and reasonable,which can significantly im prove the recomm endation quality by doing experiments on real data sets.