Smart object recommendation (SORec) architecture using representation learning in Smart objects-Based Social Network (SBSN)

Abstract

With the popularization of Internet of Things (IoT) and Social Internet of Things (SIoT), innumerable smart objects are allowed to establish associations among each other. Among numerous existing smart object recommendation methods, none have explored the failures of smart objects in smart social networks. As smart objects are subject to failure, we strongly believe that incorporation of smart objects’ failure in addition to other contextual factors (such as geographical, social, manufacturer, and economic influence) play a crucial role in smart object recommendation. However, heterogeneous nature of these contextual factors makes it challenging for estimating user and smart object representations in the Smart object-Based Social Network (SBSN). The key contribution of this work is to propose a novel smart object recommendation architecture called SORec for SBSN. The aforementioned contextual influences are incorporated into SORec by deriving transition probabilities among two nodes in SBSN which help in attaining more precise representation of users and objects. The latent representations of smart objects and users are estimated by employing Node2Vec graph embedding technique which help in improved training of the model and superior recommendation performance. Extensive experiments have been conducted on the real-world dataset to evaluate the performance of the SORec. Experimental results clearly indicate that recommendations generated by SORec are more precise, when compared to its variants and baseline methods such as PMF (41.03%), KGE (19.59%), DeepWalk (25.49%), and LINE (18.11%).