Discovery Challenge “Chat Analytics for Twitch” (ChAT)

This Discovery Challenge was part of ECML-PKDD 2020 in Ghent, Belgium (online, due to Covid-19). The task was to estimate a user’s subscription status regarding Twitch channels they are interacting with.

We received 23 registrations, from which four submitted working approaches. Three of these also submitted a descriptive paper about their method. The proceedings of this Discovery Challenge can be found at CEUR-WS.

If you want to cite this Discovery Challenge, please use the following BibTeX entry:

@ARTICLE {kobs2020towards,
    author  = "Kobs, Konstantin and Potthast, Martin and Wiegmann, Matti and Zehe, Albin and Stein, Benno and Hotho, Andreas",
    title   = "Towards Predicting the Subscription Status of Twitch.tv Users --- ECML-PKDD ChAT Discovery Challenge 2020",
    journal = "Proceedings of ECML-PKDD 2020 ChAT Discovery Challenge",
    year    = "2020"


Streaming platforms for e-sports such as Twitch become more and more popular. Users can stream their gameplay while others watch and comment on the stream. It is also possible to donate to a streamer or subscribe to the channel, which cost monthly fees (4.99$, 9.99$, or 24.99$). Half of the money goes to the streamer, the other half to Twitch. Thus, there is a monetary motivation for both, streamers and Twitch, to get people to subscribe to a channel. In this challenge, you will build a binary classifier that predicts the subscription status of a user-channel combination based on the comments the user made in this and other channels. Given a model that predicts a user to be subscribed although he or she is not, we can suspect that the user could be interested in a subscription, which could be used for targeted advertising.

The organizers of this discovery challenge are by no means affiliated with Twitch.

Important Dates

This is the timeline for this challenge. Further information will be announced in the future. All dates are given in the AoE (anywhere on earth) timezone.



If you have any questions regarding the challenge, please write an email to ecml2020-dc@professor-x.de.

Task Committee


Legal Information/Impressum