ChaCha: Leveraging Large Language Models to Prompt Children to Share Their Emotions about Personal Events

Title of the Paper

ChaCha: Leveraging Large Language Models to Prompt Children to Share Their Emotions about Personal Events

Paper Information

• Domain: Human-Computer Interaction (HCI), Children's Emotional Expression, Conversational Agents
• Keywords: Chatbots, Children, Emotional Sharing, Large Language Models, Conversational Agents, Emotion Regulation, Human-centered Computing

Research Background and Problem

• Identified Issues or Challenges:

  • Communicating emotions with children is challenging, especially as they are still developing communication skills. Parents often lack support or tools to facilitate such communication, which impacts children's mental health.
  • Current technologies, such as emotion detection systems, primarily focus on collecting emotional data from children but lack mechanisms to help children practice and express their emotions.
  • COVID-19 has negatively impacted children's social activities and daily lives, further exacerbating challenges in emotional regulation.

• Significance:

  • Emotional expression is crucial for children's social and psychological development. The ability to accurately identify and share emotions not only helps children alleviate negative feelings but also enhances their communication and problem-solving skills.

• Research Motivation and Related Work:

  • Research on children's emotion-related technologies has primarily focused on emotion detection, such as using chatbots or materials to detect emotions, but lacks tools to help children express emotions in open conversations.
  • This study is inspired by John Gottman's emotion coaching guidelines and aims to design a technological tool to guide children in identifying and sharing emotions while communicating with their parents.
  • The application of Large Language Models (LLMs), such as GPT-4, has demonstrated potential for more natural human-machine conversations, making them suitable for assisting children in freely expressing emotions.

Solution

• Proposed Solution:

  • ChaCha, a chatbot powered by LLMs, aims to help children share personal events and related emotions through free-form conversations.
  • The chatbot includes multiple conversational stages designed for emotion identification, exploration of solutions for negative emotions, and encouraging children to share their emotions with parents.

• Innovations:

  1. Utilizes LLMs to enable smoother, more flexible free-form conversations rather than restrictive rule-based or template-based dialogues.
  2. Guides conversations through multiple stages to ensure gradual identification and processing of emotions, such as exploring events, defining emotion labels, solving problems, or recording happy moments.
  3. Integrates feedback from child psychology professionals to design a system that is guiding but does not replace the role of parents.

• Implementation Steps and Key Technologies:

  • Stage Design:
    • Explore: Establish rapport between the child and the chatbot and guide the sharing of key events.
    • Label: Assist in labeling the emotions associated with each event through open-ended questions and emotion options.
    • Find and Record: Provide solutions for negative emotions or suggestions for recording positive feelings.
    • Share: Encourage children to share their emotions with their parents and discuss the importance of sharing.
  • Core Technologies:
    • Use GPT-4 to generate conversational content and monitor the sequence and logic of the dialogue through a state machine.
    • Employ dynamic prompting techniques to design specific task scenarios, such as assisting in emotion label generation or problem-solving.
    • Create child-friendly conversational rules, including limiting information length, using simple language, and appropriately using emojis.

Research Outcomes

• Specific Outcomes:

  • ChaCha successfully guided 20 children to express emotions related to personal events, including family trips, achievements, daily activities, and experiences of anxiety.
  • During testing, children regarded ChaCha as a trustworthy friend and willingly shared feelings and stories they had not previously disclosed to their parents.

• Advantages:

  • Compared to traditional rule-based chatbots, ChaCha generates dynamic and human-like responses through LLMs, significantly improving children's engagement and trust.
  • Encourages children to reflect on events and emotions, and enhances emotional processing by sharing strategies for addressing negative emotions and recording positive experiences.

• Experiment or Evaluation Results:

  • Children engaged in conversations for an average of 30 minutes, with 83% expressing willingness to continue interacting with ChaCha in the future.
  • Most participants felt that ChaCha listened well and empathized with their emotions, with some gaining confidence in emotional expression through interactions with ChaCha.

• Limitations and Future Directions:

  • Limitations:
    • Testing was limited to a laboratory environment, and the effects of long-term natural use remain unverified.
    • The model occasionally generated contextually inappropriate content, requiring further refinement of LLM behavior control.
    • The study focused on Korean children, which may introduce cultural bias; broader validation across other countries or linguistic contexts is needed.
  • Future Directions:
    • Explore long-term application scenarios: ensuring consistency in emotional coaching and expanding to personalized services.
    • Enhance interaction design with parents to strengthen emotional communication between parents and children and provide dynamic care solutions.
    • Establish robust safeguards to prevent inappropriate content generation and ensure positive psychological impacts on children.

In summary, ChaCha offers an innovative and child-centric conversational framework, leveraging large language models to support children's emotional expression. It provides clear directions for designing future child-friendly chatbots.