Comparative Experiments for Multilingual Sentiment Analysis Using Machine Translation

1st International Workshop on Sentiment Discovery from Affective Data (SDAD 2012)

Alexandra Balahur ; Marco Turchi

Others  :  http://ceur-ws.org/Vol-917/SDAD2012_8_Balahur.pdf
PID  :  43175

来源: CEUR

【 摘 要 】

Sentiment analysis is the Natural Language Processing (NLP) task dealing with sentiment detection and classification from text. Given the importance of user-generated contents on the recent Social Web, this task has received much attention from the NLP research community in the past years. Sentiment analysis has been studied in different types of texts and in the context of distinct domains. However, only a small part of the research concentrated on dealing with sentiment analysis for languages other than English, which most of the times lack or have few lexical resources In this context, the present article proposes and evaluates the use of machine translation and supervised methods to deal with sentiment analysis in a multilingual context. Our extensive evaluation scenarios, for German, Spanish and French, using three different machine translation systems and various supervised algorithms show that SMT systems can start to be employed to obtain good quality data for other languages. Subsequently, this data can be employed to train classifiers for sentiment analysis in these languages, reaching performances close to the one obtained for English.