The Research Of Recommendation Algorithm Based On Multiple Auxiliary Information Fusion

Due to the effectiveness and simplicity of the collaborative filtering technology,the collaborative filtering technology has become the mainstream method in modern recommendation systems,but the collaborative filtering algorithms are vulnerable to the negative impact of the sparseness of rating matrix and the cold-start problems.Previous research has shown that the introduction of auxiliary information to the Item can alleviate the above problems faced by the collaborative filtering technology.An understanding of the items will also help to understand the user's own preferences,but most research often use a single type of auxiliary information,which will result in an inaccurate understanding of the user's own preferences.Effective use of multiple auxiliary information of the item can better understand the user's own preferences and has important research significance.Based on the existing research,this paper does the following work to address the shortcomings of previous research: 1)Propose the multi-information variational autoencoder to obtain the deep latent representation of multiple auxiliary information,through effective variational inference,multiple auxiliary information fusion tasks are transformed into inferred the latent random distributions that can generate multiple auxiliary information,enabling tightly coupling of multiple auxiliary information.2)Propose the collaborative multi-auxiliary information variational autoencoder that can use the various auxiliary information of the item to the recommendation tasks,and the multi-information variational autoencoder and the probability matrix factorization are combined in a tightly coupled manner,and the knowledge that multiple auxiliary information fusioned can be used to enhance the performance of recommended tasks.In the analysis of experimental results,the performance of the proposed algorithm is compared with the performance of previous algorithms on two real data sets.The results show that the recall rate of the recommendation algorithms based on multiple auxiliary information fusions has an average increase of 1.1%-12.6% in different experimental sets.