Sentiment Classification of Crisis Related Tweets using Segmentation
Candy Lalrempuii, Namita Mittal · 2016
Social media data has served as a quick and accessible means of communication which may convey many important event-related information. Social media such as Twitter gives users the ability to tweet any current situation to other people and in emergencies such as disaster-related events, it is important to know the sentiments of the people and their concerns from the tweets posted by users. It can be helpful in facilitating the needs of those affected by the disaster. The tweets in the original form include many grammatical errors and slang words because of the informal nature of tweets. Many sentiment classifications have been performed on the tweets using techniques like bag-of-words and using word sequences. A sentence-level sentiment classification can be conducted on the tweets using segmentation in addition to the features extracted using word sequences. Segmentation model produces segments that are generated using a tree structure from a phrase dictionary that further is classified using a classification model for predicting the sentiment polarity. We assess the polarity predicted by the classifier for each segmentation result of a particular tweet and using a majority rule approach we predict a new sentiment class. The primary goal is to help in improving the sentiment classification for crisis-related events.