Affect in Tweets using Experts Model

Affect in Tweets Using Experts Model Subba Reddy Oota IIIT-Hyderabad Hyderabad, India oota.subba@students.iiit.ac.in Adithya Avvaru IIIT-Hyderabad Hyderabad, India adithya.avvaru@students.iiit.ac.in Mounika Marreddy IIIT-Hyderabad Hyderabad, India mounika.marreddy@research.iiit.ac.in Radhika Mamidi IIIT-Hyderabad Hyderabad, India radhika.mamidi@iiit.ac.in Abstract Estimating the intensity of emotion has gained significance as modern textual inputs in po- tential applications like social media, retail markets, psychology, advertisementse- etc., carry a lot of emotions, feelings, expres- sions along with its meaning. However, the approaches of traditional sentiment analysis primarily focuses on classifying the sentiment in general (positive or negative) or at an as- pect level(very positive, low negative, etc.) and cannot exploit the intensity information. Moreover, automatically identifying emotions like anger, fear, joy, sadness, disgust etc., from text introduces challenging scenarios where single tweet may contain multiple emotions with different intensities and some emotions may even co-occur in some of the tweets. In this paper, we propose an architecture, Experts Model, inspired from the standard Mixture of Experts (MoE) model. The key idea here is each expert learns different sets of features from the feature vector which helps in better emotion detection from the tweet. We com- pared the results of our Experts Model with both baseline results and top five perform- ers of SemEval-2018 Task-1, Affect in Tweets (AIT). The experimental results show that our proposed approach deals with the emotion de- tection problem and stands at top-5 results. 1 Introduction Sentiment analysis is one of the most famous Nat- ural Language Processing (NLP) tasks. This task was used in social network services (Pang et al., 2008; Asur and Huberman, 2010), e-retailing, ad- vertising (Qiu et al., 2010; Jin et al., 2007), ques- tion answering systems (Somasundaran et al., 2007; Stoyanov et al., 2005) and many other domains. It focuses on the automatic prediction of polarity or sentiment on tweets or reviews. While most com- puter science research in this field has focused on strict positive/negative sentiment analysis, the three dominant theories (Marsella et al., 2010; Stelmack and Stalikas, 1991) of emotion agree that humans express or operate with much more nuanced emotion representations. In other words, tweets or reviews, in recent times, include non-standard representa- tions of emotion like emoticons, emojis etc. This task of sentiment analysis became increasingly com- plex due to an addition of creatively spelt words (for eg, “gm” for “good morning”, “hpy” for “happy” etc,.) and hashtags, particularly in case of tweets. The current research of Sentiment Analysis is gearing towards evaluating emotion intensity in a text to identify and quantify discrete emotions which can help in above mentioned applications mentioned and many new ones. Here, intensity refers to the de- gree or quantity of an emotion such as anger, fear, joy, or sadness. For example, consider the three statements “The product is awesome and delivery is before time”, “It was waste of money and time” and “This TV is ok ok product at this budget range”. The above 3 statements respectively express the level of satisfaction as very happy, very sad and moderately happy. This illustrates …

Content truncated to 500 words