GraphSAGE与遗传算法：推荐系统中的聚类性能优化

GraphSAGE and Genetic Algorithms: Clustering Performance Optimization in Recommender Systems

摘要: 在信息过载的背景下，如何提高推荐系统的推荐质量和准确性成为了研究热点，而图神经网络作为一种新兴的技术，在这方面展现出了巨大的潜力。本研究首先将四种广泛使用的图神经网络模型GAT、GCN、GIN和GraphSAGE应用于推荐系统中，并通过比较发现GraphSAGE在多项指标上的表现最优。具体来说，GraphSAGE在推荐系统上实现了RMSE (0.6459)、MAE (0.4989)、AUC (0.8701)、Recall (0.6377)、Precision (0.8348)和F1-score (0.7231)的优异性能。进一步，本研究将GraphSAGE与传统的协同过滤算法在推荐系统上的表现进行比较，结果表明GraphSAGE在推荐系统上仍保持优势。在此基础上，本研究还探索了GraphSAGE与遗传算法优化K-means的结合算法——GraphKGA算法对于推荐系统性能的影响，实验证明，GraphKGA算法应用于推荐系统中的RMSE (0.6364)和MAE (0.4855)均取得优异结果，表明其能有效提高推荐系统的性能。

Abstract: In the context of information overload, how to improve the quality and accuracy of recommendation systems has become a research hotspot, and graph neural networks, as an emerging technology, have shown great potential in this regard. This study first applied four widely used graph neural network models, GAT, GCN, GIN, and GraphSAGE, to the recommendation system, and found through comparison that GraphSAGE performed best in multiple indicators. Specifically, GraphSAGE achieved excellent performance in RMSE (0.6459), MAE (0.4989), AUC (0.8701), Recall (0.6377), Precision (0.8348), and F1-score (0.7231) in the recommendation system. Furthermore, this study compared the performance of GraphSAGE with traditional collaborative filtering algorithms in the recommendation system, and the results showed that GraphSAGE still maintained its advantage in the recommendation system. On this basis, this study also explored the impact of the GraphKGA algorithm, which is a combination of GraphSAGE and genetic algorithm optimized K-means, on the performance of the recommendation system. Experiments have shown that the RMSE (0.6364) and MAE (0.4855) of the GraphKGA algorithm applied to the recommendation system have achieved excellent results, indicating that it can effectively improve the performance of the recommendation system.