Research On Representation Learning Method And Recommendation Technology For Social Network

In recent years,the vigorous development of social networks has brought great convenience to people's communication and life.More and more people have estab-lished and maintained social relationships from social networks.While constructing the link relationship between network users,other behaviors of users,such as ratings,can also be recorded by the system.The collected data can be further analyzed and better promote various social network experiences.In order to build a more complete social network,in light of the sparseness of the network,it is necessary to first make a link prediction for network users.At present,most social networks have the function of recommending friends,which is the forward link prediction.However,few social networks consider negative link prediction.At the same time,common link prediction algorithms are vulnerable to network sparsity and the like,and the actual effect is not good.Therefore,this paper proposes a symbol network link prediction algorithm based on network representation learning.Through the modeling of forward link and negative link,the network user's low-dimensional representation vector is learned and used for link prediction.Furthermore,since users in social networks have behaviors such as purchases and ratings,and these behaviors are often affected by their linked users,recommendation technology based on social information has become an effective means to improve rec-ommendation performance.However,the existing social recommendation technology often only stays in the general induction of the impact on the linked users,and does not clearly classify and quantify its internal mechanism of action.In order to solve this problem,this paper researches the influencing mechanism of social users who treat rec-ommended users,and proposes a new probability matrix decomposition model based on user-to-user trust relationship fusion modeling.Taken together,the research content and contributions of this paper are as follows:1)Research on symbol network link prediction methods based on network repre-sentation learning.This paper firstly verifies the important role of second-order dis-tance in symbolic networks for link prediction through data analysis.Then,using the network representation learning technology to capture the first-order distance and the second-order distance,respectively,and design the corresponding optimization algo-rithm to solve,and then obtain the network corresponding low-dimensional representa-tion vector.Finally,extensive experiments on two real data sets verify the validity of the proposed model.2)A user rating prediction method based on probability matrix decomposition un-der the trust mechanism is studied.This paper analyzes the user trust relationship,and uses the probability matrix decomposition model to effectively fuse two different mech-anisms that trust users directly affect user ratings and indirectly affect user preferences.On this basis,aiming at the problem that different users are affected by the two mecha-nisms with different weights,they are clustered and mapped to corresponding weights by means of user score relevance,so as to achieve personalized user model parameter selection.Finally,a number of experiments in the public dataset verify the validity of the technical framework proposed in this paper.