Behavioral Modeling and Scalable Analytics – BMSA

Research Activity

The group’s approach involves a fruitful synergy between applied research and basic research. Within this framework, the group’s research activities span the following three main directions, for which the reference technologies are probabilistic modeling and deep representation learning:

Use of latent factor generative models in social media marketing, social network analysis, information diffusion, and event forecasting. These models represent a challenge both from an application perspective (their adoption is not trivial for modeling information propagation or in contexts where data topology is complex) and for their theoretical implications (current learning techniques based on statistical inference are not adequate).
Enhanced information filtering through the integration of symbolic and sub-symbolic modeling. The main application reference concerns recommender systems, and the activity consists of studying how to extend current paradigms in two directions. On the one hand, the goal is to make recommendation systems capable of leveraging information structured in knowledge graphs. Knowledge graphs increase the amount of available information, thereby strengthening the connections between the entities involved and providing more robust support (also in terms of interpretability) for the recommendation results. On the other hand, the aim is to combine user modeling and domain constraint specification, in an approach that unifies reasoning and learning. Current mathematical models can be considered “static,” as they are adaptations of traditional supervised learning. Conversely, we are interested in expressing dynamic models, i.e., where predictions can be calibrated to contingent needs.
Behavioral analysis for monitoring complex systems such as computer networks, social networks, industrial processes, and sensor networks for environmental and energy monitoring. The objective is to detect and predict unexpected or anomalous events, in order to support the activation of effective countermeasures. The focus will be placed on two fundamental aspects. On one side, the necessity of guaranteeing adequate levels of efficiency in domains where the quantity of data to be processed is massive and distributed. On the other side, the models’ ability to adapt to changes and evolutions (concept drift) in the behavior of the entities operating within the system under analysis.

Goals

The research group focuses on Behavior Computing & Analytics (BCA): mathematical and computationally efficient models for the analysis of complex systems and entities (e.g., individuals, IoT/mobile devices, smart objects, etc.) that interact in complex environments. BCA represents an important research topic in various contexts: for example, consumer profiling, social computing, computational advertising and group decision-making, cybersecurity, opinion formation, smart industry, smart society, and smart IoT.

The term “Behavior” refers to an efficient mathematical abstraction that can summarize, describe, and predict the actions and reactions undertaken by an entity in response to various stimuli or inputs in its natural environment (pattern recognition), also highlighting possible deviations from typical and expected behavior (anomaly detection).

The research objective of the group is therefore to investigate mathematical and computational tools to understand the structural and evolutionary dynamics of behavioral flows, in order to capture/model the mechanisms that govern them and predict events and anomalies in the short and long term.

Application Fields

User Profiling & Computational Advertising (Recommender Systems), Industria 4.0 (manutenzione predittiva, ottimizzazione processi produttivi, safety), Cybersecurity