Human Behavior Discovery via Multidimensional Latent Factor Modeling

Currently, the Latent Factor Modeling is one of the most used approaches in various research fields that aim at identifying interesting features that determine the evolution of a given phenomenon. Typically, the latent factors are used to relate individual atomic elements each other: for example, semantically similar words in documents of a textual corpus (text analysis), products to buy and users (recommendation) or news in a social network (information diffusion). In this paper, we define a new latent-factor-based approach aimed at discovering human behavioral profiles. The difference from the current literature is the relaxation of the atomicity constraint of the analyzed elements. We instantiate the proposed model within the context of Human Behavior Computing, where the elements in analysis are the human actions. The latter are characterized by multiple features defined over different domains, such as "what is being done", "where", "when", or "how". We performed a test on a real-life dataset to prove the validity of the proposed approach.