A Topic based Approach for Sentiment Analysis on Twitter Data

Twitter has grown in popularity during the past decades. It is now used by millions of users who share information about their daily life and their feelings. In order to automatically process and analyze these data, applications can rely on analysis methods such as sentiment analysis and topic modeling. This paper contributes to the sentiment analysis research field. First, the preprocessing steps required to extract features from Twitter data are described. Then, a topic based method is proposed so as to estimate the sentiment of a tweet. This method requires to extract topics from the training dataset, and train models for each of these topics. The method allows to increase the accuracy of the sentiment estimation compared to using a single model for every topic.