Logo Università degli Studi di Milano



 
 
Notizie  

Sentence-level Sentiment Analysis Methods: Benchmarking and Improving Existing Approaches.

Martedì 23 maggio, ore 9.30,

Sede Milano sala Lauree

Relatore: Fabrício Benevenuto

Responsabile: Paolo Boldi

 

Abstract

In the last few years, thousands of scientific papers have investigated sentiment analysis, several startups that measure opinions on real data have emerged and a number of innovative products related to this theme have been developed. In this talk, we discuss a series of efforts for identifying the polarity (i.e., positive or negative) of a sentence or a short message. There are multiple existing methods for this particular task, usually including lexical-based and supervised machine learning methods. Therefore, first, I will describe a benchmark comparison of twenty-four popular sentiment analysis methods. Then, I will present a supervised and an unsupervised approach to improve current existing methods. Finally, we present IFeel, a system that implements and allows existing sentence-level methods to be easily tested and compared. In the last few years, thousands of scientific papers have investigated sentiment analysis, several startups that measure opinions on real data have emerged and a number of innovative products related to this theme have been developed. In this talk, we discuss a series of efforts for identifying the polarity (i.e., positive or negative) of a sentence or a short message. There are multiple existing methods for this particular task, usually including lexical-based and supervised machine learning methods. Therefore, first, I will describe SentiBench, a benchmark for sentence-level sentiment analysis and I will present an apple-to-apple comparison of twenty-four popular sentiment analysis methods. Then, I will present a supervised and an unsupervised approach to improve current existing methods. Finally, we present IFeel, a system that implements and allows existing sentence-level methods to be easily tested and compared.

Short Bio

Fabrício Benevenuto is associate professor in the Computer Science Department of Federal University at Minas Gerais (UFMG) and an affiliated member of the Brazilian Academy of Science. Fabrício is actively working on social computing and sentiment analysis related projects. His work on these topics has led to a number of important publications and widely cited papers. Recently, he has received a prestigious scholarship from Humboldt foundation through which he is currently visiting Max Planck Institute in Germany.  

17 maggio 2017
Torna ad inizio pagina