Towards emotional reasoning by dialogue agents
Loading...
Authors
Bhaumik, Ankita
Issue Date
2025-05
Type
Electronic thesis
Thesis
Thesis
Language
en_US
Keywords
Computer science
Alternative Title
Abstract
Emotions play an important role in human interactions. They determine how people feel, express or respond in different situations. Therefore, emotions are an essential part of building human-like dialogue agents in various applications like healthcare, customer service, or psychotherapy. Despite major advancements in building emotion detection models, this remains an active research area as human emotions are very specific to people and domains. A significant limitation of state-of-the-art deep learning models for emotion detection is their inability to generalize across different application areas. These models rely on domain-specific fine-tuning data and predefined emotion labels that often fail to capture the nuanced emotional state when applied to new domains and situations. Our research addresses these limitations by developing the context-aware and generalizable emotion detection framework that enables dialogue agents to understand and respond to the user emotions without domain-specific fine-tuning. The major contributions of our research are:- Development of a generalizable emotion detection framework that is adaptable to unknown domains without re-training on domain-specific training data
- A generative approach to emotion detection and reasoning, inspired by the intuitive emotional reasoning process of humans in specific situations
- We focus on a task-oriented domain to demonstrate how emotion state tracking across dialogues can improve interactions in goal-driven dialogues. We show that adapting dialogue strategies based on the emotional state of the user significantly enhances the probability of task success, particularly in persuasion dialogues, where emotions influence decision making. By incorporating emotional reasoning into dialogue agents, this thesis contributes to the development of more emotion-aware adaptive dialogue agents for use across many real-life applications.
Description
May2025
School of Science
School of Science
Full Citation
Publisher
Rensselaer Polytechnic Institute, Troy, NY