Recommendation algorithm performance optimization method based on knowledge graph

Abstract

The invention relates to a recommendation algorithm performance optimization method based on a knowledge graph and particularly relates to a method for solving interpretability of a recommendation system, relieving cold start and improving algorithm accuracy. Firstly, an interpretability problem of the algorithm being solved, preprocessing an original data set, constructing a knowledge graph, obtaining the path information by using a knowledge graph meta-path extraction technology, inputting the path information into a recurrent neural network GRU, adding an attention mechanism and an averagepooling layer, and carrying out importance distinguishing on different paths to obtain a model prediction vector; secondly, relieving a cold start problem of the algorithm: multiplying the node attribute information of the user and the article to obtain an attribute fusion vector; and combining the vector results of the two parts according to a ratio of 3: 7, and training the model by using a cross entropy loss function to obtain a prediction result. The method is advantaged in that experiments show that compared with a recommendation algorithm based on a GRU model, accuracy of the algorithm is improved by 2.3%, and compared with a traditional matrix decomposition model, accuracy is improved by 5.1%.

Description

technical field [0001] This patent relates to a method of fusing the path information of the knowledge map and the content information of the nodes, thereby improving the accuracy and recall rate of the recommendation algorithm, which belongs to the technical field of knowledge maps and recommendation systems. Background technique [0002] In recent years, with the continuous development of network information technology, people can easily obtain rich information from various channels. But at the same time, people are also facing the problem of information overload, and it is difficult to find the information they want accurately and quickly from the massive data. In this case, personalized recommendation algorithm came into being. Personalized recommendation can use efficient Internet tools to provide people with accurate services, so as to help users extract useful information from massive amounts of information, so as to meet the needs of users. The recommendation syste...

AI Technical Summary

Problems solved by technology

2. In terms of interpretability: most of the existing models use the model to improve the accuracy of prediction. The defect of the deep learning model is to learn from the data, but it cannot explain the prediction results of the model, and it is difficult to analyze the interpretability.

Images