Heuristic recommendation technique in Internet of Things featuring swarm intelligence approach

TLDR: In this paper, the authors proposed a heuristic method to build a recommender engine in IoT environment exploiting swarm intelligence techniques, where smart objects are represented using real-valued vectors obtained through the Doc2Vec model, a word embedding technique able to capture the semantic context representing documents and sentences with dense vectors.

Abstract: In smart environments, traditional information management approaches are often unsuitable to tackle with the needed elaborations due to the amount and the high dynamicity of entities involved. Smart objects (enhanced devices or IoT services belonging to a smart system) interact and maintain relations which need of effective and efficient selection/filtering mechanisms to better meet users’ requirements. Recommender systems provide useful and customized information, properly selected and filtered, for users and services. This paper proposes a heuristic method to build a recommender engine in IoT environment exploiting swarm intelligence techniques. Smart objects are represented using real-valued vectors obtained through the Doc2Vec model, a word embedding technique able to capture the semantic context representing documents and sentences with dense vectors. The vectors are associated to mobile agents that move in a virtual 2D space following a bio-inspired model - the flocking model - in which agents perform simple and local operations autonomously obtaining a global intelligent organization. A similarity rule, based on the assigned vectors, was designed so enabling agents to discriminate among them. A closer positioning (clustering) of only similar agents is achieved. The intelligent positioning allows easy identifying of similar smart objects, thus enabling a fast and effective selection operations. Experimental evaluations have allowed to demonstrate the validity of the approach, and on how the proposed methodology allows obtaining an increasing in performance of about 50%, in terms of clustering quality and relevance, compared to other existing approaches.