CharacterChat: Learning towards Conversational AI with Personalized Social Support (2308.10278v1)

Abstract: In our modern, fast-paced, and interconnected world, the importance of mental well-being has grown into a matter of great urgency. However, traditional methods such as Emotional Support Conversations (ESC) face challenges in effectively addressing a diverse range of individual personalities. In response, we introduce the Social Support Conversation (S2Conv) framework. It comprises a series of support agents and the interpersonal matching mechanism, linking individuals with persona-compatible virtual supporters. Utilizing persona decomposition based on the MBTI (Myers-Briggs Type Indicator), we have created the MBTI-1024 Bank, a group that of virtual characters with distinct profiles. Through improved role-playing prompts with behavior preset and dynamic memory, we facilitate the development of the MBTI-S2Conv dataset, which contains conversations between the characters in the MBTI-1024 Bank. Building upon these foundations, we present CharacterChat, a comprehensive S2Conv system, which includes a conversational model driven by personas and memories, along with an interpersonal matching plugin model that dispatches the optimal supporters from the MBTI-1024 Bank for individuals with specific personas. Empirical results indicate the remarkable efficacy of CharacterChat in providing personalized social support and highlight the substantial advantages derived from interpersonal matching. The source code is available in \url{https://github.com/morecry/CharacterChat}.

CharacterChat: A Novel Framework for Personalized Social Support Conversations

The paper "CharacterChat: Learning towards Conversational AI with Personalized Social Support" provides detailed insights into the development of a novel Social Support Conversation (S2Conv) framework designed to address mental health concerns by enhancing emotional support systems. This research navigates beyond the traditional Emotional Support Conversations (ESC) by incorporating individual personality considerations through advanced AI methods.

Introduction to the Framework

The paper introduces CharacterChat, an innovative S2Conv system that leverages the Myers-Briggs Type Indicator (MBTI) for persona decomposition. The paper responds to the limitations of ESC in engaging diverse personalities and aims to enhance the efficacy of conversational AI in providing mental health support. The S2Conv framework involves virtual characters that emulate human-like attributes, using an interpersonal matching mechanism to connect users with persona-compatible supporters.

Technical Development

The methodology primarily hinges on three core components:

1. MBTI-1024 Bank: The researchers established this bank based on persona decomposition, featuring characters with specific MBTI profiles to facilitate personalized social support interactions. This process employs ChatGPT, directing it through bespoke persona decomposition prompts, resulting in a diverse spectrum of virtual personalities.
2. Role-playing Systems: An upgraded role-playing prompting technique with behavior presets and dynamic memory is proposed. This approach ensures that the characters maintain contextually relevant narratives without overwhelming the system with excessive memory use. By dynamically selecting context-related memories, conversations retain factual consistency and relevance.
3. CharacterChat System Development: Building upon the MBTI framework, they integrate memory selection and interpersonal matching models to optimize the connection between seekers and supporters. CharacterChat is the culmination of utilizing persona and memory-based dialogue models, based on the Llama2-7B backbone, ensuring highly personalized interactions.

Empirical Evaluation

The empirical results demonstrate significant effectiveness in personalized social support, with strong numerical validations, especially in emotional improvement (EI), problem-solving (PS), and active engagement (AE). With over 50% of characters exhibiting full alignment with their MBTI dimensions, the paper showcases the substantial advantages of using MBTI for character development and personalized support provision.

Evaluations using ChatGPT highlight CharacterChat's superiority in providing emotional enhancement compared to existing models such as BlenderBot-Joint and Vicuna-13b-v1.5. Further insights reveal substantial positive impacts of interpersonal matching on support quality.

Implications and Future Directions

Practically, CharacterChat holds promise for deployment in real-world scenarios where mental health support is pivotal. Theoretically, it paves new pathways in conversational AI for not only addressing information needs but also emotional well-being. The paper emphasizes the potential role of interpersonal matching in refining dialogue systems, suggesting further exploration into broader applications across AI domains.

The authors advocate for future research to explore expanded interpersonal matching mechanisms and further optimization of AI facilitation in mental health contexts. CharacterChat could serve as a prototype for evolving interactions within AI systems, emphasizing personalized and empathic dialogues, which are crucial for user-centric technology.

In conclusion, the CharacterChat project represents a seminal effort in intertwining personality-focused AI systems with mental health support, establishing a potential paradigm for future conversational technologies.