Data Science Chair


    Project Description

    KILiMod (KI-basierte kontextuelle Verlinkung und Moderation von deutschsprachigem Chat) is research project started in 2023 in collaboration with vAudience. It aims to improve the safety and quality of chat texts in German language platforms through AI-powered models for chat moderation and content enrichment. The objective is to create a scalable, real-time, and high-quality solution that can address the limitations of existing methods like profanity filters and human moderators.

    The project is motivated by the significant concern of offensive or misleading user-generated content in chat platforms, which poses a challenge for many businesses. Through KILiMod, the aim is to offer a more effective and cost-efficient way of moderating chat texts at scale to improve the overall quality of the online chat experience for users. The main goal is the research of neural networks for natural language processing (NLP) to detect problematic content in chat messages during live streams of sports and esports events. Further, we are employing deep metric learning to develop a recommender system, which can automatically process information related to the conversations and provide personalized, relevant content recommendations to users.

    There are unique characteristics of live stream chats that pose special challenges. For example, pictograms, such as emojis or specific emotes, provide an additional avenue for altering the meaning of a message and, in many cases, circumvent automatic moderation systems. A robust solution would need to be able to interpret these and similar features of the chats. To help our models understand the unique characteristics of live stream chat messages and overcome the resulting challenges, we are employing our extensive dataset of chat messages gathered as part of our previous work


    The following persons are involved in this project: