[Paper Review] Between an Arena and a Sports Bar: Online Chats of eSports Spectators

Denis Bulygin, Ilya Musabirov+3 more|arXiv (Cornell University)|Jan 9, 2018

Digital Games and Media19 references16 citations

TL;DR

This paper analyzes massive live chat communication during The International 2017 Dota 2 esports tournament on Twitch.tv using structural topic modeling and cross-correlation to show that in-game events drive topical and emotional patterns in spectator chats. It finds that chat dynamics resemble both a sports arena and a sports bar, proposing design enhancements like sentiment visualization and event-highlighted clip generation to improve viewer engagement.

ABSTRACT

Hundreds of thousands of spectators use Twitch.tv to watch The International, a Dota 2 eSports tournament and communicate in massive chats. In this paper, we analyse crowd behavior in these chats, disentangle features of social communication, such as contextual meanings of emojis and short messages. We apply structural topic modelling and cross-correlation analysis to investigate topical and temporal patterns of chat messages and their relation to in-game events. We show that in-game events drive the communication in the massive chat and define its emergent topical structure to a various extent. Following the discussion in communication and social computing literature, we discuss these findings in the framework of analysis of communication of physical sports crowds and outline some limitations of the 'stadium' metaphor, suggesting a complementary metaphor of 'sports bar' as a useful analytical and design device.

Motivation & Objective

• To understand the emergent topical and emotional structure of massive live chat during esports events.
• To investigate how in-game events shape real-time spectator communication on Twitch.tv.
• To examine the role of emojis, emotes, and copypastas in maintaining coherence in high-volume chat environments.
• To evaluate the metaphor of 'sports bar' as a complementary framework to the traditional 'sports arena' model for understanding online spectator communities.
• To derive design implications that enhance viewer engagement and social coordination in live esports chat.

Proposed method

• Applied structural topic modeling (STM) to extract latent topics from 1.2 million chat messages during The International 2017.
• Used cross-correlation analysis to identify temporal relationships between in-game events and shifts in chat topics.
• Trained a topic model on chat data to disentangle contextual meanings of emojis, abbreviations, and short messages.
• Analyzed changes in topic diversity and dominance across different chat sizes and tournament stages.
• Identified event-driven bursts of communication (e.g., hero deaths) and their emotional or discursive triggers.
• Proposed design solutions based on observed patterns, including sentiment visualization and automated highlight generation.

Experimental results

Research questions

• RQ1 How do in-game events influence the topical and emotional structure of live esports chat?
• RQ2 What role do emojis, emotes, and short messages play in maintaining coherence in massive, fast-moving chat streams?
• RQ3 To what extent does the structure of chat communication resemble that of a physical sports arena versus a sports bar?
• RQ4 How does chat topic diversity and dominance shift across different stages of a tournament and chat size?
• RQ5 What design interventions can enhance viewer experience by leveraging real-time chat dynamics?

Key findings

• In-game events such as hero deaths or team kills trigger significant bursts of chat activity, with topics shifting rapidly to reflect emotional reactions or strategic discussion.
• The most dominant topic in the chat during the TI7 finals was cheering for Team Liquid, occupying a large portion of screen space and demonstrating high topical inequality.
• Chat communication is highly event-driven: periods of inactivity lead to boredom-related memes and emotes, while intense in-game moments trigger synchronized emotional responses.
• Despite high volume, chat maintains coherence through repeated use of emotes, copypastas, and shared abbreviations, forming collective 'voices' that mimic social synchronization.
• Topic diversity is higher in smaller chats, while larger chats show a concentration on fewer dominant topics, especially as the tournament progresses.
• The study confirms that the 'sports bar' metaphor complements the 'sports arena' metaphor, as spectators express emotions without direct feedback, yet still find value in collective expression.