| How to fully mine the information in the data to improve the quality of recommendation is the focus of recommendation task research.Reasonable data modeling and effective recommendation algorithm are the key ways to solve this problem.Early recommendation algorithms were limited to the study of homogeneous information,ignored the effective use of multi-source heterogeneous information and failed to fully mine various semantic connections in the data.Existing recommendation algorithms begin to use heterogeneous information to assist in improving recommendation quality,and have achieved certain results.However,the current methods still have problems such as inaccurate relationship modeling,unreasonable semantic fusion mode and high information loss degree in modeling multi-source heterogeneous data and extracting multiscale semantic information,so there is still much room for improvement in the recommendation results.The recommendation algorithm based on heterogeneous information network and multi-scale semantic fusion needs to solve several key problems:firstly,how to effectively mine and utilize heterogeneous information in multisource heterogeneous data.Secondly,how to integrate multi-scale semantic information effectively to overcome the problem of information loss during information aggregation.Finally,how to alleviate the problem of sparse data in recommendation tasks.Aiming at the above key scientific issues,the main research contents and innovations of this paper are as follows:(1)Aiming at the problem of utilizing heterogeneous information from multiple sources,a heterogeneous graph model based on meta-path fusion guided by relational attention is proposed.The model can flexibly mine different types of semantic information by designing the metapath structure effectively and aggregate them from the low and high order semantic levels by using the two-stage attention mechanism.Experimental results show that this model can effectively utilize the rich semantic relationships in heterogeneous information networks to improve recommendation results.(2)Aiming at the problem of information loss when the existing models aggregate neighborhood information,a recommendation mechanism based on multi-scale semantic fusion is proposed.By dividing neighbor groups,this mechanism constructs neighbor group interaction between nodes and extracts multi-scale information representation of nodes rich in semantic information.The experimental results show that this mechanism effectively alleviates the information loss in the process of aggregation and improves the recommendation results.(3)Aiming at the problem of sparse data in recommendation tasks,a recommendation algorithm based on contrast learning is proposed to enhance data.Based on the heterogeneous information network,this algorithm designs the classification mechanism and contrast pattern of positive and negative samples and realizes the contrast learning between low-order semantic representation and high-order semantic representation through cross-view comparison.Experimental results show that the proposed algorithm effectively alleviates the data sparsity problem and improves the recommendation results. |
