Andrew J Litteken

Twitch Tox Bot


The internet has afforded many different opportunities for interaction betweem communities; however, there tends to be problems in the toxicity of chats or comments sections. Along with a Jonathan Baker, John Westhoff, Connor Higgins and Joshua Huseman and the use of IBM Watson Tone Analyzation, Personality Analysis and Translation Modules we developed an "Intelligent Chat Bot" to monitor a chat stream.

We intended to create a chat bot that could determine the sentiment of a user and score the comment based on the overall scope of the message. With the data collected from the modules, we trained a nueral net to identify more toxic comments based off of Twitch stream data. The data was used to create profiles for each user to score the general toxicity of a user for more intelligent tracking of the toxicity of a channel.

In addition to the text analysis, an interface was created to show a dynamically changing graph of the most active, and most toxic users. In addition, a streamer could directly block, message, timeout or grant privileges to any one user.

We anticipate that a similar design could be applied in similar situations, such as suicide prevention or general safety for parents hoping to monitor what their younger children are exposed to.

This was an entry in the AT&T Hackathon at Notre Dame, where it won first place. You can view the project here.