Recommendation platform in Internet of Things leveraging on a self-organizing multiagent approach

Identifying user requirements and preferences on the basis of the current context, is one of main challenges of the Internet of Things (IoT) paradigm. Users, services and applications interact maintaining, often unreliable, relations which need of smart management systems to satisfy their demands. Traditional information handling approaches in distributed systems are most often unsuitable for modern Smart Environments due to the huge amount and the extreme dynamism of the entities involved. This paper proposes NARIoT platform that allows building recommendation systems in IoT environment. The approach relies on vector representations of IoT resources obtained by using of a word embedding tool, the Doc2Vec neural model, which, starting from text documents describing the resources, provides real-valued vectors mapping them. The vectors are handled through intelligent agents, which self-organize themselves creating an ordered virtual structure, so enabling informed mechanisms of information filtering. In particular, an ordered overlay network emerges from the autonomous, parallel and decentralized work of intelligent agents, thus enabling efficient recommendation operations. The experimental validation confirms the effectiveness of the approach and provides very encouraging results.