Data representation in sentiment analysis task
Hani Hamdan, Pierre Vigier, Frédéric Wantiez · 2017
Sentiment analysis, or opinion mining, refers to the task of exctracting subjective information contained within text materials. The basic form of sentiment analysis is polarity classification. Intuitively, it aims to determine whether a piece of text expresses a positive sentiment or not with respect to a given subject. This process is already used in the industry. It has applications in a wide range of sectors such as marketing, online sales, or opinion forecasting. The technics of sentiment analysis have gained popularity thanks to the essort of modern social networks websites and the ever growing amount of labeled data. Most industrial applications use traditionnal classification algorithm combined with refined data representations such as Logistic Regression or Random Forest. Even with not so complex classifiers, industrial applications achieve great score on classification tasks. This is mainly due to the effort put in preprocessing and representing input data. We thus discuss how refined representations of text materials can lead to great performances.