A novel recommendation-based framework for reconnecting and selecting the efficient friendship path in the heterogeneous social IoT network

Abstract

Automating the selection process for the most suitable service in a dynamic Internet of Things (IoT) ecosystem to improve critical metrics such as resilience, throughput, delay, energy consumption, confidence level, and cost is considered an important challenge, and in this regard, the Social Internet of Things (SIoT) paradigm has greatly helped to deal with this challenge through the merging of Complex Network (CN) principles within the IoT domain. In this study, we combined metaheuristics and Deep Reinforcement Learning (DRL) to develop a new unsupervised group-driven recommender framework for predicting, reconnecting, and choosing the optimal friendship path between requester and service provider nodes in a SIoT environment. There are four main phases to the presented framework. We first suggested a new method to learn features associated with the heterogeneous social IoT structure and detect ever-changing semantically related clusters. In the second phase, we propose a novel optimization model that utilizes the Artificial Bee Colony (ABC) metaheuristics to accurately predict community-oriented social connections. We came up with a new strategy to select an efficient group-based friendship path in the third phase. It hybridized the techniques of metaheuristic-driven Ant Colony Optimization (ACO) and DRL-oriented Proximal Policy Optimization (PPO). In the final phase, we introduce an innovative ACO-centered recommender model to improve the framework's accuracy and speed while also providing socially aware, community-driven service recommendations. We conducted extensive experiments on four real-world datasets to assess the efficacy of the proposed framework, and the findings show that it outperforms leading baselines.