Research On Recommendation System Based On Local Association Modeling

In recent years,with the rapid development of mobile Internet,people have stepped into the era of information explosion.As one of the effective means to alleviate the problem of "Information Overload",recommendation system can not only provide users with interesting information quickly,but also contains huge commercial value.Matrix Completion is one of the core techniques for constructing a recommendation system,and its aims is to fill in the Matrix containing missing values.In the background of high-dimensional and sparse scoring data,this paper mainly studies the limitations of matrix completion recommendation methods based on traditional matrix decomposition and deep graph neural network.Inspired by local association modeling,we propose two local association modeling based improved matrix completion recommendation models.Firstly,to solve the problem of the low prediction accuracy of the traditional matrix decomposition-based matrix completion models,this paper proposes a matrix completion recommendation model based on clustering information and cascading residuals.The model takes advantage of the similar properties of similar users and similar items and uses clustering method to cluster them respectively.Then it extracts and learns the local association information of similar users and similar items,namely clustering feature,and combines it with global content information(ID feature),so as to enhance the feature expression ability of independent users and independent items.In addition,the cascading residual learning mechanism was introduced to optimize the initial prediction results in order to better approximate the real score.A large number of experiments are carried out on real data sets,and the results verify the effectiveness and superiority of the proposed method.Secondly,in view of the limitation of the graph neural network based inductive matrix completion models that adopts a fixed averaging weight in information aggregation process in local graph structure.That is,the target node is equally dependent on all the neighbor nodes,this paper proposes a graph attention network recommendation system enchanted with edge information.In this model,the graph attention network is used in local subgraph structure,to implement the strategy of assigning weight to the features of neighbor nodes according to their importance during message aggregation process.Meanwhile,in view of the anonymous mark of nodes in local subgraphs,it cannot realize the matrix completion directly using graph attention networks.This paper proposes an explicit introduction of edge feature representation in the graph attention network layer to distinguish neighbor nodes with different edge types,so as to better aggregate them into the target node.Extensive experiments show that the model proposed in this paper is superior to the most advanced methods in the performance of model prediction accuracy on multiple real data sets.