Inferring Stressors from Conversation: Towards an Emotional Support Robot Companion
Journal
International Journal of Social Robotics
Journal Volume
14
Journal Issue
7
Pages
1657
Date Issued
2022-09-01
Author(s)
Abstract
People build relationships with each other through emotional support. Appropriate care provided by friends or family may critically help people extricate themselves from negative emotional states. In this work, we implement a conversational robot that can provide emotional support to people facing stressful situations in their daily lives. The proposed robot system extracts meaningful keywords from the support recipients’ speech to identify the moments when people expose their concerns through self-disclosure. These keywords are then used to determine the underlying stressors in the recipients’ minds. By using a novel approach through a commonsense knowledge graph-based Bayesian network, the proposed system leverages Gibbs sampling to infer the posterior probability of various documented stressors. Moreover, we trained a neural network to determine an appropriate support strategy that should be offered to a recipient with a specific personality given a detected stressor. We evaluated our implementation through several human-in-loop experiments. Results show that this system can correctly identify the underlying stressors from speech and provide appropriate emotional support required to help people cope with difficult situations in one’s life.
Subjects
Conversational Agents | Emotional Support | Social Robots
Publisher
SPRINGER
Type
journal article