PhD in Computer Science

Contacts

Supervisor

Phd thesis

Dependency Syntax in the Automatic Detection of Irony and Stance

The research goal of my PhD thesis is to explore the impact of morphosyntactic information in sentiment analysis related tasks. In particular, irony detection and stance detection are chosen as two principal case studies.
A focus on the importance of the formulation of a clear problem statement is presented, and therefore a computational modeling of it is proposed. Secondly, the experiences in the creation of annotated corpora for those phenomena and topics, and the contributions to the organization of shared tasks and the important lessons they can teach are highlighted.
Finally, an approach to solve both tasks by using shallow features based on dependency syntax is proposed. And additionally, a combination of different sets of dependency-based features -- encoded in UD format -- are paired with neural models and a handful of experiments is performed in a variety of settings.
In the whole thesis a multilingual scenario is kept in mind, exploring four different language settings: English, Spanish, French and Italian, for both irony detection and stance. Furthermore, due to the availability of benchmark datasets, in the chapter regarding stance detection experiments are carried also across a fifth language, i.e. Catalan.
The results obtained show how a dependency-based approach can offer meaningful cues to the detection of irony. And how this, in turn, could evolve in being a preprocessing step in the task of stance detection and mark a relevant change.

Research activities

I am currently a research fellow in the project "Be Positive!" supported by Google.org fundings at the Computer Science Department at the Università degli Studi di Torino.

I did my PhD studies (now waiting for the defense) in cotutelle between the Content-Centered Computing group inside the Computer Science Department at the Università degli Studi di Torino (supervision of Cristina Bosco) and inside the Pattern Recognition and Human Language Technology (PRHLT) research center at the Universitat Politècnica de València (co-supervision of Paolo Rosso).

I am particularly interested in the research areas of Natural Language Processing and Theoretical Linguistics. My current research focuses on the topics of irony and stance detection in social media. In particular, I'm assessing the impact of dependency-based syntactic features (exploiting the Universal Dependencies format) in such tasks.