Preliminary Results of Group Detection Technique Based on User to Vector Encoding

This paper presents a novel approach for detecting groups of users based on observations of co-occurrences of user behavior. A Deep Neural Network is trained to encode users into a vector representation using an innovative adaptation of the word embedding architecture used in Natural Language Processing, which has been recently applied and modified for various domains, including graph data, recommender systems, and DNA gene sequence embedding. Preliminary experiments show promising results for the proposed adaptation to group detection based on the user-to-vector encoding derived from behavior observations in a variety of scenarios.