Improved Smooth Neighborhood Recommendation Algorithm Combined With Implicit User Relationship Network

With the rapid development of the Internet and information technology,a series of data such as product information,transaction information,social information on the network have exploded,and it is difficult for users to obtain information that meets their needs within a limited time,so the “information overload” problem occurs.In order to deal with this kind of problem,the recommendation system came into being.Since its birth in the 1990s,the recommendation system has established a relatively complete theoretical system,and the types of recommendation algorithms are also diverse.Specifically,there are content-based recommendation algorithms,collaborative filtering-based recommendation algorithms,association rule-based recommendation algorithms,Knowledge-based recommendation algorithms and hybrid recommendation algorithms.Collaborative filtering is a popular research direction in recommendation algorithms,and it is recommended by mining score data reflecting users' interest preferences.The collaborative filtering algorithm can be subdivided into two branches: neighbor-based collaborative filtering and model-based collaborative filtering.Among them,the most mainstream method in model-based collaborative filtering is matrix decomposition,and this article is based on this.The collaborative filtering algorithm is combined to introduce a smooth neighborhood recommendation algorithm combined with an explicit relationship network.This algorithm uses kernel functions to obtain neighborhood information from covariates on specific users-item networks,which can effectively solve the cold start problem of rating data.At the same time,the algorithm uses a divide-and-conquer alternating least squares calculation strategy to improve the operation efficiency.Aiming at the problems of the sparseness of the explicit user relationship network and the insufficient description of the user relationship,this paper introduces a trust propagation model,hoping to construct a network structure reflecting the trust between users through trust inference,and use it as an implicit user relationship net-work to replace the original explicit user relationship network,and then combine it with the smooth neighborhood recommendation algorithm to achieve improvement.Finally,in this paper,several types of traditional collaborative filtering algorithms and two types of collaborative filtering algorithms that combine explicit and implicit network structures are tested on simulated data sets and actual data sets.Simulate different degrees of data sparsity and cold start problems by setting different missing rates and cold start rates on the simulated data set,and artificially construct a relationship network structure between users and projects for the algorithm to use;actual data set selects the online music data set from Last.fm,which faces serious sparsity and cold start problems,and provides information to construct a relationship network between users and projects,which meets the research needs of this article.The experimental results show that the introduction of the network structure will improve the performance of the collaborative filtering algorithm,effectively solve the cold start problem and perform more robustly;the performance of the improved smooth neighborhood recommendation algorithm in this paper is better than the comparable algorithms in most cases.