Emotion analysis on Indian politics

Abstract

Indian Politics and its coverage in media has changed drastically in the recent years. News published major media houses are read mostly over the social media. Where the news is reduced to smaller pieces and it reaches to a larger audience faster. With the advent of social media and smartphones, it is easier for the masses to express their opinion for government actions faster. One of the major channels for the same is Twitter. With the BJP government using Twitter to reach their followers the reactions and expressions of people are instantly captured in form of tweets and retweets. Twitter in India has now become a powerful tool to garner the voices of the public and understand the topics take is of focus to the people through trending topics. In our project, we try to understand the reaction of the section of people on social media to any government announcements. By using the Twitter API and searching for relevant strings, hashtag and query, we have collected the relevant tweets that are posted by people on the same topic. After having collected the data from Twitter, we have manually cleaned the data for noise and collected the applicable responses. Using the refined data, we have analysed the same to understand the overall emotion that is being expressed by the country. We have first separated the words in each of the tweets to find the marker words in each of the sentence which express the emotion or the sentiment. Each of the words tagged with the emotion they express, and then whole is collated to understand the emotion of the tweet. We have used the major trending topics in Twitter for national and regional news and curated the content. Additionally, we have also looked at the major political party handles and collated the tweets and replies to the same. We have also collected data from Kaggle related to Demonetisation, as that was one of the major political event in the last year, whose effect is still being felt in parts and a lot of people took to social media to voice their opinions regarding the same. We have developed a model for detection of emotion of the tweets which is explained later in the paper. Further course of work shall be to understand what kind of aggregate emotion was expressed by majority of people for each of the topics. From the literature survey, we came across multiple forms of emotion detection in linguistics but we had to develop a model of our own, owing to the fact that most often tweets are not fully formed proper sentences. They are shortened version of words as there is a 160-character limit on the tweets. And people use emoticons to express their emotions in tweets. The limitations of the model are also present, the model only works for English tweets, any vernacular language can’t be captured as the result we could not get a lot of data on usage of Hindi decision in the country. The paper has been divided into the following sections. In the section on literature survey, we describe the different existing works on emotion detection. These works have mostly focussed on unsupervised emotion detection as we have used unsupervised linguistic based emotion detection algorithms in our method. It is followed by the section of data collection where we describe the source of data and the methodology used to collect the data. It is accompanied by the constraints that have been faced in collecting the data. In the section of text pre-processing, the different approaches to clean the text data have been discussed. In the section on processing, the methodologies on entity extraction and emotion detection have been discussed in detail. The CoreNLP parser used in the emotion detection has been described in a subsection. The results section talks about the different insights that can be extracted from the tweets based on secondary analytics module on the extracted entities and emotions and coupling them with the tweet metadata. In the limitation section, we describe the different limitations in the current system and finally suggest the solutions to the limitations in the section on future work.