Publication

Abstract : Livestreaming platforms are discernibly the most comprehensive sources of data in real time. Such websites enable users to broadcast content like the games which they are playing, while providing them the opportunity to interact with viewers watching the livestream. Twitch.tv is one of the most popular livestreaming platforms across the globe with millions of monthly active streamers and viewers. Owing to the COVID-19 pandemic, there has been a shift in the conventional lifestyle of the people, with them turning towards online alternatives like Twitch.tv for leisure. This change has led to an increase in the engagement of users in these livestreaming platforms by manifolds. Concurrently, a lot of data is generated from this sudden inflow, which can prove very useful in understanding the general consensus of the crowd. This data is very important, and there is a need to construe the true emotion of the people in real time, which is reflected in the comments made by them in the chat section of livestream. The streamers on Twitch.tv can consequently refine their content immediately based on the feedback that they can infer from the responses given by the users. But, due to the sheer volume of data and convoluted nature of the chat due to the use of emojis, emotes, and emoticons, there are bound to be inconsistencies, human errors, and other esoteric references which are exceedingly complex to dissect, making the task of language processing difficult and leading to incoherent results. Taking into account the hindrance posed by these issues, we have taken up the task to achieve fairly accurate emotion prediction by putting forward machine learning and deep learning techniques. This will involve the creation of a labelled dataset that can be used for training and evaluating the algorithms. Given how context-specific most comments are on the platform, this will be an extensive task. The project will also require the creation of an end-to-end system that performs emotion analysis and giving results in real time through feedback-loops.