Rich linguistic processing for multilingual opinion mining

Carlos Gómez Rodríguez, University of A Coruña
Room: 5A09
Time: Fri 19 May 2017 at 11:00-12:00
Host: Željko Agić (part of ML Seminar Series)

Identifying the sentiment present in texts published on the Web and social networks like Twitter can provide an invaluable source of information about the public perception of products, issues or events.
In this talk, I will present an overview of our recent developments at the LyS research group in the field of opinion mining, using natural language processing and machine learning techniques to identify positive or negative sentiment on texts from the Web and social networks like Twitter. These include both rule-based and supervised approaches. The rule-based approach estimates the polarity of a text (i.e., whether its content expresses a positive or a negative opinion, or neither) based on its syntactic structure, using compositional syntax-based rules to account for relevant linguistic phenomena like intensification, adversative clauses and negation. The approach can be applied to multiple languages with minimal effort thanks to universal dependencies. The supervised approaches solve the same problem applying machine learning techniques, which can be enriched with syntactic and psychometric features or with feedback from the rule-based system.