Efficient and Fine-grained Profile Matching with Privacy Preserving in Mobile Social Networks

The proliferation of the internet and smart devices has dramatically increased the user count of Mobile Social Networks (MSNs). Connecting with new users is the core feature of MSNs. The increased number of active users on MSNs highlights the need for efficient and effective profile-matching and recommendation mechanisms. This paper introduces a novel framework for dynamic friend matching and recommendation by using advanced classification and clustering techniques. To improve matching results in the user connection as per social characteristics and behavior, the suggested framework combines ideal clustering using the Outline Coefficient and Davies-Bouldin Record with K-Means clustering. We propose a hybrid fine-tune ensemble classifier that blends the best aspects of the Random Forest and XGBoost methods. The hybrid model of grouping users has improved the accuracy of friend recommendations. A 95.23% classification accuracy is obtained by the ensemble classifier by combining the strong feature selection and averaging of Random Forest with the gradient boosting of XGBoost. The Davies-Bouldin Index and Silhouette Coefficient are applied to the clustering step to find the ideal number of clusters to provide cohesive and well-separated groups. These well-defined groups are then used in the K-Means algorithm, providing a basic structure for future matching and recommendation processes. Experimental results have demonstrated the efficiency of the proposed hybrid model.