Bio-Inspired Agents for a Distributed NLP-Based Clustering in Smart Environments

Due to the high number and extreme dynamism of the entities involved in last-generation smart environments, often, traditional data management approaches aren't particularly appropriate. In order to better meet users and services requirements, the objects, belonging to an intelligent environment, have to be able to interact and maintain relationships that require efficient management mechanisms. Clustering approaches, for example, have proven to be a valuable technique to extract useful information in data. This paper proposes a multiagent algorithm to perform distributed clustering in Smart environments through Swarm Intelligence (SI) techniques that leverage Natural Language Processing (NLP) concepts. By exploiting the Doc2Vec model, a word embedding tool that allows representing text documents with high-dimensional vectors, also capturing the semantic context, it is possible to describe the intelligent objects - advanced devices or services - through real-valued vectors. These vectors are assigned to the agents that change their position in a virtual space according to a bio-inspired model. The agents follow simultaneously simple and local movement rules obtaining a global and intelligent organization that enables a proper and useful positioning (clustering). Preliminary experimental evaluations have shown the validity of the approach.