Building Compassionate Chatbots Using Neural Network Models
Event details
| Date | 09.07.2019 |
| Hour | 15:30 › 17:30 |
| Speaker | Anuradha Welivita |
| Location | |
| Category | Conferences - Seminars |
EDIC candidacy exam
Exam president: Dr. Martin Rajman
Thesis advisor: Dr. Pearl Pu Faltings
Co-examiner: Prof. Robert West
Abstract
Open-domain conversational agents or chatbots have increasingly become popular in the natural language processing community. In recent years, extensive research has been conducted on building open-domain conversational models using neural networks. Even though these models can process natural language at a lexico-syntactic level, they often fail to identify subtle variations of emotion or affect in human conversations and respond in an emotionally appropriate manner. Our research goal is to identify novel approaches in building compassionate neural chatbots that can effectively identify and process emotional content present in user utterances and generate responses that mimic human social and emotional intelligence. We discuss three existing works related to neural conversational agents and emotion analysis and how they relate to our work. We first examine a basic sequence-to-sequence neural network model for generating human-like responses in a conversation. We then look at how this model can be augmented with affect information based on the traditional Valence-Arousal-Dissonance (VAD) affective notation, affect-based objective functions and affectively diverse decoding strategies to generate responses that are rich in emotion. Finally, we discuss a novel approach to learn richer representations of human emotions, using emojis used in social media.
Background papers
A neural conversational model, by O. Vinyals, and Q. Le. In Proceedings of the 31st International Conference on Machine Learning, volume 37, 2015.
Affective neural response generation, by N. Asghar, P. Poupart, J. Hoey, X. Jiang, and L. Mou. In European Conference on Information Retrieval, pages 154-166, Springer, Cham, 2018.
Using millions of emoji occurrences to learn any-domain representations for detecting sentiment, emotion and sarcasm, by B. Felbo, A. Mislove, A. Søgaard, I. Rahwan, and S. Lehmann. In Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing, pages 1615–1625, 2017.
Exam president: Dr. Martin Rajman
Thesis advisor: Dr. Pearl Pu Faltings
Co-examiner: Prof. Robert West
Abstract
Open-domain conversational agents or chatbots have increasingly become popular in the natural language processing community. In recent years, extensive research has been conducted on building open-domain conversational models using neural networks. Even though these models can process natural language at a lexico-syntactic level, they often fail to identify subtle variations of emotion or affect in human conversations and respond in an emotionally appropriate manner. Our research goal is to identify novel approaches in building compassionate neural chatbots that can effectively identify and process emotional content present in user utterances and generate responses that mimic human social and emotional intelligence. We discuss three existing works related to neural conversational agents and emotion analysis and how they relate to our work. We first examine a basic sequence-to-sequence neural network model for generating human-like responses in a conversation. We then look at how this model can be augmented with affect information based on the traditional Valence-Arousal-Dissonance (VAD) affective notation, affect-based objective functions and affectively diverse decoding strategies to generate responses that are rich in emotion. Finally, we discuss a novel approach to learn richer representations of human emotions, using emojis used in social media.
Background papers
A neural conversational model, by O. Vinyals, and Q. Le. In Proceedings of the 31st International Conference on Machine Learning, volume 37, 2015.
Affective neural response generation, by N. Asghar, P. Poupart, J. Hoey, X. Jiang, and L. Mou. In European Conference on Information Retrieval, pages 154-166, Springer, Cham, 2018.
Using millions of emoji occurrences to learn any-domain representations for detecting sentiment, emotion and sarcasm, by B. Felbo, A. Mislove, A. Søgaard, I. Rahwan, and S. Lehmann. In Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing, pages 1615–1625, 2017.
Practical information
- General public
- Free