Research interests

My research activity is focused on unsupervised data mining and knowledge representation. I'm interested to study how to analyze streams of multidimensional dynamic data coming from heterogeneous sources (like online social networks) to create unified models, mostly based on graphs, of representation of entities, users, topics that can be efficiently queried to find latent relationships among elements in real-time or near-real-time scenarios.

I'm studying how multidimensional data analysis algorithms (like tensor decompositions, co-clustering) can be influenced and improved by the availability of additional information describing and linking input data.

I'm also interested in the analysis of collaborative networks, in particular applied to bibliometrics: I'm focusing my study on the structure, behaviour, and evolving dynamics of networks of autonomous researchers that collaborate to better achieve common or compatible scientific goals.

More details can be found below.

Social Computing and Civic Social Networks

Università degli Studi di Torino

Civic social networks can be considered as systems able to support citizens in their daily life. Their design offers many challenges, including: integration of institutional data and user generated contents; improvement of social awareness, self-governance and self-organization of local communities; development of ontologies for representing social entities; analysis techniques for multi-facet, purpose, topic knowledge management systems. My research activity at the Social Computing group, Department of Computer Science, Università degli Studi di Torino, aims to continue research on social computing by studying novel indexing and aggregation techniques of multidimensional and georeferenced data, focusing on the identification of the patterns through which individuals organize themselves. Special focus will be dedicated to the improvements of accessibility, through the study of novel data visualization techniques.

Knowledge Representation And Integration of Heterogeneous Data Sources

RAI research centre and Università degli Studi di Torino

This research activity is focused on the study, design and development of new models for the integration of the heterogeneous and dynamic data coming from different knowledge sources. I used a time-dependent knowledge graph to model all heterogeneous aspects of this type of information in a homogeneous way, facing problems like incompleteness and inconsistency. The framework allows to recognize and represent entities (people, users, sentiment), and their relationships, within the knowledge graph, enabling the following analysis phase. We designed and implemented a query system based on the structure of the graph that can be used to efficiently mine large dataset, and extract time-dependent, context-aware information on entities, enabling the discovery of non-trivial cross-domain latent relationships. I demonstrated that this framework can be successfully used in the Social TV scenarios, capturing multiple aspects of the considered domain, from the semantic characterization of the TV content, to the temporal dimension of the problem, to the social characterization and the social perception of a TV event.

Multidimensional Data Analysis

Università degli Studi di Torino, Arizona State University

This research activity is focused on the study of mechanisms of information integration that express additional relationships between the entities involved in standard unsupervised multi-way data analysis techniques. In the set of co-clustering, different approaches about metadata integration were designed, developed and compared, in order to discover aggregations that reflect latent relationships between input data. Actually the research activity is focused on the study of novel mechanisms of metadata integration in high-order clustering techniques and, in particular on the tensor decomposition algorithms. We proposed a novel tensor decomposition algorithm based on a multi-resolution approach that allows to speed-up the decomposition process by leveraging available additional metadata.

Evaluation of Research and Collaborative Networks

Università degli Studi di Torino, Université Paris 8

Bibliometrics is a set of methods to quantitatively analyze academic literature and the collaborative environment in which they have been produced. While bibliometric methods are most often used in the field of library and information science, bibliometrics have wide applications in other areas. Among all the possible application, we study and explore novel methods to estimate the impact of a work, a researcher (or a set of researchers), a paper in the surrounding community. At high level, we focus our study on the structure, behaviour, and evolving dynamics of these networks of autonomous researchers that collaborate to better achieve common or compatible scientific goals.

Social Networks Analysis

Università degli Studi di Torino, Université Paris 8

In order to facilitate the analysis of user-generated contents (always enormous in number), I studied novel topic detection and tracking (TDT) techniques to analyze and follow the evolution of the information expressed by a social network. In order to do that, I studied novel metrics to identify the relationships that exist among users, pages and contents, and therefore I map these information in a social graph where it is possible to follow the most emergent topics (by considering temporal conditions) expressed by a social network community.

Classification, Recommendation And Navigation on Digital Documents

Università degli Studi di Torino, Université Paris 8, Arizona State University

The aim of this activity is to improve standard management techniques of huge data, in order to improve retrieval and fruition. The contribution is the definition of context-driven mechanisms of classification, recommendation and navigation among multimedia contents (documents, images and videos). We proposed two systems for content classification, CoSeNa and Immex: they allow to find semantic relationships among contents by highlighting latent links among concepts organized through taxonomies.