ABSTRACT
Research question
To better understand esports consumer demand, this study derived topics from real-time viewer comments during live streams of two professional esports leagues, King Pro League (KPL) and CrossFire Pro League (CFPL). We investigated structural factors that drew viewers’ attention to each topic.
Research methods
Using data from Douyu.com, we collected 1,111,177 real-time comments on 126 KPL matches and 233,869 comments on 96 CFPL matches. We employed a digital ethnography approach to discern the fan identification of viewers who contributed to comments. We applied guided Latent Dirichlet Allocation (LDA) to extract topics in these comments. We examined structural factors that affect viewers’ interest in each topic using a multinomial logit regression.
Results and findings
Only a small proportion of viewers who commented during live-streaming esports games were identified as fans of competing teams. Viewers paid most attention to the excitement of the game. Commentators and in-streaming activities also captured viewers’ attention. Esports fans tended to focus on players and teams whereas general observers were more attuned to game-related skills and knowledge.
Implications
We take the attention economy as a theoretical framework to understand esports consumers’ interests and demonstrate the structure of consumer demand for various aspects of esports. We provide insight to help esports organizations and live-streaming platforms effectively attract viewers.
Disclosure statement
No potential conflict of interest was reported by the author(s).
Notes
1 During regular seasons, the same group of commentators usually provided commentaries on two consecutive matches a night.
2 The Marimo stopwords list (https://github.com/koheiw/marimo) covers stop words in languages such as Chinese, English, and German. Baidu stopwords (http://www.baiduguide.com/baidu-stopwords/) were created by Baidu, a Chinese company specializing in internet-related services and artificial intelligence products.
3 We thank the anonymous reviewer for suggesting these robustness checks.
4 For example, the CFPL comments with the highest percentage of the topic ‘Team’ are repetitions of a team's name; the KPL comments with the highest percentage of the topic ‘Socialization’ are comments encouraging others to guess the match results.
5 We performed 11 tests for each league. In each test, we compared the estimates of the sample excluding one topic with the estimates of the full sample. The p-value in each test was above 0.95, which suggested that we failed to reject the null hypothesis that there is no systematic difference between the two sets of estimates.