A Low-code Approach to Identify Toxicity in MOBA Games

Abstract

In this work, we explore the use of KNIME to identify toxic behavior in the MOBA game DOTA 2. Using a dataset composed of 10530 messages taken from 1903 matches, we tested the use of KNIME to identify toxic messages obtaining an accuracy of 92% and 85% for toxic and non-toxic messages, respectively. The DOTA 2 game chat log was used to present a low-code approach to a supervised learning model for message classification. In addition to providing insight into the toxic behavior of MOBA players, our work supports the idea that low-code development can reach levels as good as traditional development. On the other hand, our study can also serve as a basis for more elaborate implementations that allow us to observe other aspects of toxic behavior from its detection, encouraging the construction of prevention and neutralization tools.