The values used to build graphs.

With the rapid development of the internet, information overload has become a prevalent issue. In order to tackle information overload, recommendation systems serve as an effective tool that can offer personalized recommendation services to users. The efficiency of recommendation systems is, however, hampered by the prevalent problems with data sparsity and cold start issues in collaborative filtering recommendations. Researchers typically address these issues by utilizing user social information clustering methods. Nevertheless, in practice, previous studies have shown that inaccurate similarity calculations and poor clustering results have led to a decrease in prediction accuracy. This paper suggests a collaborative filtering recommendation algorithm that incorporates several relationships in order to overcome these difficulties. This method first calculates user similarity based on implicit social relationships and trust relationships. After clustering users using the spectral clustering technique, it makes use of user-based collaborative filtering recommendations within the cluster containing the target person. The collaborative filtering recommendation system that integrates many relationships effectively decreases prediction errors and improves recommendation accuracy, as shown by the results of simulated studies.