Research Of Social Network Information Recommendation Model Based On Multifactorial Coupling

With the rapid development of information technology and the Internet,social network resources are extremely rich,and information dissemination has become more convenient.However,the problem of “information overload” has surfaced,and users are facing serious network data redundancy difficulties.It is becoming more and more difficult to obtain valuable information from massive data.As an effective means to solve the problem of "information overload",the research of information recommendation has received extensive attention from academia and industry.In particular,the social network information recommendation model has many application scenarios and commercial value.However,due to the continuous expansion of the scale of social networks and the increasing complexity of network topology,the recommendation algorithms of previous studies have been difficult to adapt to the requirements of the times.Therefore,it is of great theoretical value and reality to study the information recommendation model to further improve the accuracy and efficiency of recommendation.significance.Several information recommendation modes commonly used are generally classified into content-based,association-based rules,and collaborative filtering.The collaborative filtering mode includes a memory-based collaborative filtering recommendation model and a matrix decomposition information recommendation model.These two recommended models have better precision and become a research hotspot.However,with more and more types of data and more and more complex application scenarios,both types of recommendations are faced with problems such as network "cold start" and data sparsity,which affect the accuracy of information recommendation.As the data samples become larger and larger,the information recommendation efficiency of newly added users is extremely low,and the recommendation model has scalability problems,which seriously affects the operational efficiency and user experience of social networks.In view of the difficulties and challenges in the above research,the main research contents of this paper are as follows:Firstly,a multi-factor coupling model of social network information dissemination is constructed.Combined with coupling theory,the influence of various factors on information dissemination is quantified,and the law of information dissemination is deeply analyzed.At present,the study considers the individual effects of each factor and does not consider the coupling effect of the mutual influence of each factor.By constructing a multi-factor coupling model of information dissemination,this paper analyzes that both user preference factors and user relationship factors have a greater impact on social network information dissemination.In the follow-up study,the memory-based collaborative filtering model,link prediction model and interest point recommendation model are improved by combining user preferences and user relationship factors.Secondly,the improved random forest algorithm is used to construct the hierarchical hybrid collaborative filtering model RMHCF,which solves the problem of data sparsity and scalability.Due to the traditional memory-based collaborative filtering mode,the similarity calculation is not comprehensive enough to reduce the accuracy of the prediction score.This paper combines time,multi-source heterogeneous geography,user behavior and user context to influence user interest preferences,and uses coupled metrics algorithm to improve user similarity calculation method and improve the accuracy of prediction score;and integrate random forest machine learning algorithm to improve recommendation.The model handles the efficiency of large sample data and solves the scalability problem.Experiments verify that the RMHCF model has better recommendation accuracy.The random-forest-based Meta-Level hybridization collaborative filtering model will improve the efficiency of the offline data samples of the restricted Bozeman machine training in the subsequent research,and improve the real-time recommendation effect of the recommendation model of interest points.Thirdly,the multi-path random walkaround prediction model LRWM based on information correlation and information negative feedback is constructed,which solves the problem of low node discrimination and negative feedback information interference,and improves the accuracy of link prediction.The fusion information association and its stability optimize the user similarity calculation index,which solves the problem of low node discrimination caused by attribute and topology index,and improves the accuracy of link prediction.At the same time,there is negative feedback information in the social network to interfere with link prediction.Previous studies have solved this problem by deleting the negative feedback link,but it has destroyed the integrity of the network structure.In this paper,by improving the link weight and preserving the complete network structure,the problem of negative feedback information interference is solved,and the multi-path random walk link prediction model LRWM is constructed by using the leapfrog algorithm,which further improves the link prediction accuracy.Using the LRWM model,more accurate and comprehensive user relationship data can be obtained,and it can be incorporated into the probability matrix decomposition information recommendation framework of subsequent research as an additional data source to further improve the recommendation accuracy.Finally,the LBSN user interest point recommendation model for fusion link prediction is constructed,which solves the problem of “cold start” and data redundancy,and improves the accuracy of interest point recommendation.Compared with the memory-based collaborative filtering model,the matrix decomposition model has better information recommendation accuracy and scalability,and is more suitable for user interest point recommendation,but is susceptible to “cold start”.This paper combines LRWM link prediction model and project attributes,mine additional data source supplement,incorporate probability matrix decomposition model,improve prediction scoring algorithm,and construct a probabilistic matrix decomposition model LPMF for fusion link prediction,which effectively solves the "cold start" problem.The information recommendation accuracy is improved.At the same time,the parameter unsupervised training process of the restricted Pozmann machine deep learning is optimized,and the hierarchical hybrid collaborative filtering model RMHCF is combined to improve the efficiency of extracting hidden factors and solve the data redundancy problem;A variety of hidden factors,improved LPMF model to build LBSN user interest point recommendation model,experimental results show that the model has better recommendation accuracy.