Multi-Channel Convolutional Neural Network for Twitter Emotion and Sentiment Recognition

Proceedings of NAACL-HLT 2019, pages 1355–1365 Minneapolis, Minnesota, June 2 - June 7, 2019. c⃝2019 Association for Computational Linguistics 1355 Multi-Channel Convolutional Neural Network for Twitter Emotion and Sentiment Recognition Jumayel Islam1, Robert E. Mercer1, Lu Xiao2 1Department of Computer Science, The University of Western Ontario 2School of Information Studies, Syracuse University jislam3@uwo.ca, mercer@csd.uwo.ca, lxiao04@syr.edu Abstract The advent of micro-blogging sites has paved the way for researchers to collect and analyze huge volumes of data in recent years. Twit- ter, being one of the leading social network- ing sites worldwide, provides a great oppor- tunity to its users for expressing their states of mind via short messages which are called tweets. The urgency of identifying emotions and sentiments conveyed through tweets has led to several research works. It provides a great way to understand human psychology and impose a challenge to researchers to ana- lyze their content easily. In this paper, we pro- pose a novel use of a multi-channel convolu- tional neural architecture which can effectively use different emotion and sentiment indicators such as hashtags, emoticons and emojis that are present in the tweets and improve the per- formance of emotion and sentiment identifi- cation. We also investigate the incorporation of different lexical features in the neural net- work model and its effect on the emotion and sentiment identification task. We analyze our model on some standard datasets and compare its effectiveness with existing techniques. 1 Introduction Social networking sites (e.g., Twitter) have be- come immensely popular in the last decade. User generated content (e.g., blog posts, statuses, tweets etc.) in social media provides a wide range of opinionated, emotional and sentimental con- tent which gives researchers a massive data source to explore. For example, Twitter, being one of the leading social networking giants, provides an online environment that allows people of various backgrounds and locations to share their opinions and views on different matters. As of July 2018, over 500 million tweets are sent per day having over 300 million monthly active users.1 1http://www.internetlivestats.com/twitter-statistics/ There is often a misconception about senti- ments and emotions as these subjectivity terms have been used interchangeably (Munezero et al., 2014). Munezero et al. (2014) differentiate these two terms along with other subjectivity terms and provide the computational linguistics community with clear concepts for effective analysis of text. While sentiment classification tasks deal with the polarity of a given text (whether a piece of text expresses positive, negative or neutral sentiment) and the intensity of the sentiment expressed, emo- tion mining tasks naturally deal with human emo- tions which in some end purposes are more de- sirable (Ren and Quan, 2012; Desmet and Hoste, 2013; Mohammad et al., 2015). Detecting emo- tion and sentiment from noisy twitter data is re- ally challenging due to its nature. Tweets tend to be short in length and have a diverse vocabulary making them harder to analyze due to the limited contextual information they contain. In this study, we are interested in tackling these two tasks with …

Content truncated to 500 words