seminars

Exploiting Deep Learning and Probabilisic Modeling for Behavior Analytics

Date
Tuesday, April 9, 2019

Time
2.30 p.m.

Location
ISI seminar room 1st floor

Speaker(s)
Dr. Giuseppe Manco ICAR-CNR, Rende (CS), Italy

ABSTRACT
=========
Behavior analytics is an important topic in different contexts including: consumer analytics, social computing, fraud detection, and group decision-making. Behavior refers to actions and reactions of any individual, in response to various stimuli or inputs. The recent advent of technologies for collecting and tracking behavioral data at large scale, has made it possible to devise new mathematical models that allow to analyze, understand, and predict actions. These include models for event streams, social network connections, purchasing habits and opinion formation. Understanding the structural, topical and temporal dynamics of user behavior cascades can provide insights on the complex patterns that govern the information propagation process and it can be used to forecast future events. This is crucial in several domains, including: Medical events (where the focus is on sequences of acute incidents, doctor’s visits, tests, diagnoses, and medications), Consumer behavior (purchasing patterns), “Quantified self” data (such as wearable devices and apps to record eating, traveling, working, sleeping, waking) , Social media actions (previous posts, shares, comments, messages, etc.), smart cities and mobility patterns (trajectories, taxi/car/public transportation adoptions, etc.).
Although behavior analytics can be devised in a broad range of domains, the focus of this talk towards a specific question: based on the observed sequence of events involving a population of individuals, can we predict what kind of event will take place at what time in the future? This issue has been gaining increasing attention and several approaches were proposed in the recent literature based on stochastic modeling. Within this field, latent variable models represent the state of the art, due to their capabilities in modeling the hidden causal relationships that ultimately influence preferences, choices and actions. Recently, new approaches based on deep learning architectures were also proposed, achieving competitive advantages with respect to the current state of the art.

In this talk, we review the current literature on the subject and study how new paradigms based on embedding deep learning into probabilistic latent variable modeling can be profitably exploited to behavior analytics. These approaches were proven quite successful in domains such as computer vision and speech processing. However, their adoption for modeling user preferences and behavior is still unexplored, although recently it is starting to gain attention. We show that merging latent factor modeling and neural functionalities. enables substantial advantages in capturing behavior history modeling both long-term preferences and short-term behavior.

BIO
=========
Giuseppe Manco is senior researcher at ICAR-CNR and contract professor at University of Calabria where he has been teaching Databases, Data Mining and Social Network Analysis. He was recipient of the 2014 YAHOO FREP (Faculty Research and Engagement Program) prize, for research activity on machine learning applied to social network analysis. His current research interests include machine learning, knowledge discovery and data mining, probabilistic modelling and recommender systems, social network analysis, workflow and process mining, cybersecurity. He is the scientific coordinator of the ICAR research lab “ADALab: Laboratory of Advanced Analytics on Complex Data”. He is the author of over 100 research papers on the subjects of knowledge discovery, data mining, logic and databases. He has been serving in the program committee of several international/national conferences, including: IJCAI, AAAI, IEEE ICDM, ECMLPKDD, SIAM SDM, PAKDD and he was program co-chair of ECML-PKDD 2016. He is serving as an associate editor for the Journal of Intelligent Information Systems, Knowledge and Information Systems and Machine Learning.