Research On Recommendation Algorithm Based On Graph Embedding Technology

In recent years,with the continuous expansion of the scale of the Internet and the explosion of various applications based on the Network,the information capacity in the network presents an explosive growth.In this case,how to quickly and accurately find the information needed by users from the mass data has become a big problem for Internet users.For this problem,the early solution is the search engine represented by Google,and recommendation system,as a more active,accurate and effective information filtering technology,has been widely concerned and studied in recent years.This paper conducts in-depth research on the current mainstream recommendation methods and related technologies,and the main work is as follows:(1)The item-based collaborative filtering(Item-CF)algorithm has a very high requirement on similarity between items.If the similarity calculation of items is inaccurate,the recommendation performance will be greatly affected.To solve this problem,we introduce mutual information and self-attention mechanism into the Item-CF algorithm.On this basis,we design a collaborative filtering recommendation algorithm--DWSA-CF,which combines deep walk and self-attention mechanism.The algorithm is mainly divided into two parts.First,the mutual information between items is used to carry out weighted random walk to generate a large number of item sequences,which are input into the Skip-Gram integrated with the attention mechanism to generate the embedded vector matrix of items.Second,the generated items are embedded into the vector matrix and item-CF algorithm is combined to generate the final Top-k recommendation list.We implemented DWSA-CF algorithm under PyCharm,and carried out simulation experiments on three public data sets.Experimental results show that DWSA-CF algorithm can calculate the similarity between items more accurately,and has a better recommendation effect for datasets with high data sparsity.(2)The classical graph embedding algorithm GraphSage has the following defects: First,the relational graph between items is constantly changing dynamically,especially when new item nodes are added,the algorithm needs to recalculate the embedding vector for each node,which greatly increases the computation amount;Second,the algorithm can not solve the cold start problem well.To solve the above problems,we introduce the classification algorithm and LSTM algorithm,and propose a GraphSage and LSTM recommendation algorithm based on classification--CLS＿GraphSage+LSTM.Firstly,the algorithm classifies the nodes,samples the neighbor nodes of the target node,and generates the embedding vector of the target node by learning aggregate representation of neighbor nodes.For the new node,the new node is added to the graph by directly connecting it with the existing node to ensure that the node can be covered during sampling and the embedding vector of the new node can be learned.Finally,the Top-k recommendation is realized by combining the time series information and LSTM algorithm.We carry out classification simulation experiments on two data sets,and the experimental results show that the node sampling performance of the improved GraphSage algorithm is improved.The results of recommendation simulation experiments on three data sets show that CLS＿GraphSAGE+LSTM algorithm can not only effectively alleviate the cold start problem,but also reduce the computation of node embedding vector and improve the overall efficiency of the recommendation algorithm.