DLSI-Volvam at RepLab 2013: Polarity Classification on Twitter Data.
Alejandro Mosquera, Javier Fernández Martínez, Jose Manuel Gomez, Patricio Martínez-Barco, Paloma Moreda · CLEF (Working Notes) · 2013
This paper describes our participation in the proling (po- larity classication) task of the RepLab 2013 workshop. This task is fo- cused on determining whether a given text from Twitter contains a pos- itive or a negative statement related to the reputation of a given entity. We cover three dierent approaches, one unsupervised and two unsuper- vised. They combine machine learning and lexicon-based techniques with an emotional concept model. These approaches were properly adapted to English and Spanish depending on the resources available for each lan- guage. We obtained promising results in the overall evaluations, reaching a F-score of 34% and a sensitivity of 40% in the best cases. The reason- able level of performance compared to other methods encourages us to continue working on the improvement of the proposed approaches.