Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

An Empathetic AI Coach for Self-Attachment Therapy

Lisa Alazraki, Ali Ghachem, Neophytos Polydorou, Foaad Khosmood, Abbas Edalat

TL;DR

This work tackles the challenge of delivering scalable, empathetic digital psychotherapy by augmenting a rule-based SAT chatbot with deep learning components for emotion recognition and fluent, empathetic utterance retrieval, plus selectable personas. It introduces the EMPATHETICP ERSONAS dataset (1,181 emotion expressions and 2,143 empathetic rewritings of 45 base utterances) and a multi-objective retrieval framework that optimizes empathy, fluency, and novelty, using a RoBERTa-based emotion classifier with $94.96\%$ accuracy and $95.10\%$ macro F1. A non-clinical trial with $N=16$ participants over five days shows higher perceived empathy, engagement, and usefulness compared with a baseline rule-based system, with clinicians noting improved emotion recognition but limited emotional scope. The work provides practical design guidance, a public dataset, and a reference implementation (SATbot) to advance empathetic digital coaching for SAT, highlighting avenues for safety, scalability, and broader clinical validation.

Abstract

In this work, we present a new dataset and a computational strategy for a digital coach that aims to guide users in practicing the protocols of self-attachment therapy. Our framework augments a rule-based conversational agent with a deep-learning classifier for identifying the underlying emotion in a user's text response, as well as a deep-learning assisted retrieval method for producing novel, fluent and empathetic utterances. We also craft a set of human-like personas that users can choose to interact with. Our goal is to achieve a high level of engagement during virtual therapy sessions. We evaluate the effectiveness of our framework in a non-clinical trial with N=16 participants, all of whom have had at least four interactions with the agent over the course of five days. We find that our platform is consistently rated higher for empathy, user engagement and usefulness than the simple rule-based framework. Finally, we provide guidelines to further improve the design and performance of the application, in accordance with the feedback received.

An Empathetic AI Coach for Self-Attachment Therapy

TL;DR

This work tackles the challenge of delivering scalable, empathetic digital psychotherapy by augmenting a rule-based SAT chatbot with deep learning components for emotion recognition and fluent, empathetic utterance retrieval, plus selectable personas. It introduces the EMPATHETICP ERSONAS dataset (1,181 emotion expressions and 2,143 empathetic rewritings of 45 base utterances) and a multi-objective retrieval framework that optimizes empathy, fluency, and novelty, using a RoBERTa-based emotion classifier with accuracy and macro F1. A non-clinical trial with participants over five days shows higher perceived empathy, engagement, and usefulness compared with a baseline rule-based system, with clinicians noting improved emotion recognition but limited emotional scope. The work provides practical design guidance, a public dataset, and a reference implementation (SATbot) to advance empathetic digital coaching for SAT, highlighting avenues for safety, scalability, and broader clinical validation.

Abstract

In this work, we present a new dataset and a computational strategy for a digital coach that aims to guide users in practicing the protocols of self-attachment therapy. Our framework augments a rule-based conversational agent with a deep-learning classifier for identifying the underlying emotion in a user's text response, as well as a deep-learning assisted retrieval method for producing novel, fluent and empathetic utterances. We also craft a set of human-like personas that users can choose to interact with. Our goal is to achieve a high level of engagement during virtual therapy sessions. We evaluate the effectiveness of our framework in a non-clinical trial with N=16 participants, all of whom have had at least four interactions with the agent over the course of five days. We find that our platform is consistently rated higher for empathy, user engagement and usefulness than the simple rule-based framework. Finally, we provide guidelines to further improve the design and performance of the application, in accordance with the feedback received.
Paper Structure (27 sections, 3 equations, 9 figures, 4 tables)

This paper contains 27 sections, 3 equations, 9 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Background
  3. Existing approaches to chatbot-assisted mental support
  4. Digital psychotherapy and self-attachment technique
  5. Empathy in digital psychotherapy
  6. Chatbot personification and user engagement
  7. Privacy and ethics
  8. Dataset and data collection
  9. Survey preparation
  10. Recruitment of survey respondents
  11. Criteria for the acceptance of responses
  12. Data analysis
  13. Creating personas from the data
  14. Empathy annotation
  15. Implementation
  16. ...and 12 more sections

Figures (9)

  • Figure 1: Age distribution for both sexes across samples in the E MPATHETICP ERSONAS dataset, showing that most samples belong to the middle age groups 30-39 and 40-49.
  • Figure 2: A three-sentence utterance rewriting in the E MPATHETICP ERSONAS dataset. Sentence 2 conveys the main question, while Sentences 1 and 3 reinforce the empathy of the message by expressing sympathy and compassion.
  • Figure 3: Bar chart showing the mean empathy score of rewritten utterances in the E MPATHETICP ERSONAS dataset by number of sentences that they contain (for details of the empathy annotation process see Section \ref{['empathyann']}).
  • Figure 4: Conversation flow of the SAT chatbot.
  • Figure 5: Appearance of the SAT chatbot web application.
  • ...and 4 more figures