Emotional Regulation in Conversations: Developing a Dataset and Classifier

  • Subject:Emotions, Group chats, Dataset, Machine learning
  • Type:Bachelor / Master Thesis
  • Supervisor:

    Dr. Ivo Benke

This thesis contributes directly to contemporary research on emotion recognition based on collaboration scenarios. We invite students who are both interested in emotional research and motivated to work in a cutting-edge research environment.

Problem Description

The automatic recognition of emotional processes (e.g., empathy or emotional support) in collaborative scenarios such as video-meetings or group chats is complex. The usage of technology changes our expression and our interpretation of emotional signals. Limitations in display size and application forms make the emotional interpretation more difficult. Therefore, automatic recognition of emotional processes is beneficial to ease the collaboration in hybrid and remote work settings.

To allow for automated recognition labelled datasets for these scenarios are necessary due to the special interaction form of conversations (opposed to social media comments on which classifiers are often trained on). However, there exist only very few datasets with tagged labels which allow for the development of algorithms to recognize emotions in conversations. Such datasets require raw conversational data in a machine understandable form from a communication scenario together with labels by humans to classify emotional processes. Consequently, the task of this thesis is to develop a labelled dataset based on the theory of emotion regulation strategies and, subsequently, the development of a basic classifier. The basis of this thesis builds existing raw group chat data (~100.000 chat messages) which lacks preprocessing and tagged labels for emotions and emotional regulation.

Goal of the Thesis

The goal of this thesis is to develop a labelled dataset for emotions in group chats. This task includes preprocessing of the raw data and development of label structure. Optional for a master thesis, an initial classifier can be developed based on the developed dataset


  • (Very) good programmings experience in Python
  • Interest in research on emotions and emotion recognition
  • Proactive thinking
  • Very good time management and organizational skills
  • Fluent in English writing


If you are interested in this topic, please contact me directly via email: ivo.benke∂kit.edu.