Emotional states and personality profiles in Conversational AI
This project was for summer term 2020, you CAN NOT apply to this project anymore!
- Sponsored by: Steering Lab by Horváth & Partners GmbH
- Project Lead: Dr. Ricardo Acevedo Cabra
- Scientific Lead: Olena Schüssler, Dr. Inna Vasylchuk
- Co-Mentor: Michael Rauchensteiner
- Term: Summer semester 2020
Results of this project are explained in detail in the final documentation and presentation.
Affective Computing (else: Emotional AI) has the potential to humanize interactions with digital assistants and offers advantages in a wide range of business applications. The most prominent areas where Emotional AI could bring a lot of value include Customer Relationship Management (CRM), Human Resource Management (HRM), Marketing and Entertainment.
In last couple of years several emotional industry solutions that are mainly based on deep learning technology emerged. They allow AI to measure the emotional state of the user very precisely, based on speech patterns and voice timbre. However, to make real use of user emotions, we have to consider the main goal of each digital assistant: to make the user of digital assistant happy. To accomplish this goal the digital assistant must understand the personality of the user and moreover it has to have a human-like emotional intelligence itself. Analogically to humans, digital assistant can learn how to understand the emotional states of the user and his personality through series of interactions and provided user feedback. Reinforcement learning is one of the most promising AI technologies that allows to incorporate user feedback efficiently. This approach of machine learning is inspired by a behavioral psychologist and introduces learning similar to how a child learns to perform the task: it does not rely upon the data-recognition techniques but uses experience-driven decision making instead. The use of reinforcement learning has got a new attention in the industry and has to be considered for improvements in Conversational AI.
Goal of the project is to use current advances in linguistic analysis, deep learning, reinforcement learning and another state-of-the-art methodologies to achieve human-like behavior of digital assistants when dealing with the emotions of the user. The output of the project shall be the initial version of the prototype code which proves its feasibility in one of the application domains of high interest to our clients.