Role-Playing System for Improving Friend Relationships Using LLM
A lack of self-expression becomes most apparent in situations where one needs to refuse a request.
PALINK is a system designed to enhance self-expression by allowing users to practice communication skills through simulated conversations with a virtual friend, focusing on the theme of "refusing requests."
A survey conducted with 57 Korean adolescents asked, "What do you think needs improvement in your interpersonal relationships? Please select the option that applies to you." The majority (47.4%) identified "awareness of others" as the key area for improvement. This tendency is closely linked to a lack of self-expression.
[Goal] The goal of this study is to develop a role-playing-based interpersonal practice system aimed at improving interpersonal relationship issues and to validate its effectiveness.
[Key Elements for Achieving the Goal]
- Conversation simulation with a virtual friend
- Practice of communication techniques
- Real-time conversation evaluation and guidance
1. Role-Playing Therapy Technique Role-playing therapy involves recreating interpersonal scenarios to experience therapeutic effects. Leveraging this technique, users engage in role-playing with AI characters in specific situations, such as practicing refusals.
2. Positive Mindset Formation Before loading screens or the start of a conversation, the system displays encouraging phrases to help users build confidence and foster a positive mindset.
3. Quest-Based Learning Users receive guidance through quests, allowing them to follow structured steps to learn effective refusal techniques. Quests are detected by prompts designed to identify refusal categories, which are then processed in the code to track quest progress.
4. Tip Assistance and Feedback If users struggle to find an appropriate response during a conversation, they can use a "Tip" button to receive textual guidance. Once the conversation ends, the dialogue history is analyzed using a prompt, and comprehensive final feedback is provided.
1. Response Generation Prompt
- Generates appropriate replies for the conversation based on the user's statements.
2. Refusal Detection Prompt
- Analyzes the user's statements to detect refusal patterns or categories.
3. Tip Generation Prompt
- Provides example responses or tips to guide the user when they struggle to reply.
4. Final Analysis Prompt
- Evaluates the entire conversation to deliver comprehensive feedback and insights for improvement.
- Demo
default.mp4
- 4개의 다른 캐릭터 페르소나 (Four Different Character Personas)
- 미연, 진혁, 세진, 현아 (Miyeon, Jinhyuk, Sejin, Hyuna)
- We have adjusted the prompts to create four different characters with distinct personas. This allows users to practice refusal scenarios tailored to each character type.
- 팁 기능 (Tip Feature via Button (Example Answer Generation))
- 퀘스트 제공 및 탐지 (Quest Provision and Detection Tailored to Each Persona) (퀘스트 달성 시 토스트 메시지 (퀘스트 3 ‘상대방이 처한 상황 파악하기’가 달성된 모습))
2.mp4
- Futter WebApp Deployment: Amazon S3 (Integrated with CloudFront)
- FastAPI Server Deployment : Amazon EC2
- DB : Amazon RDS
The Flutter web app was successfully deployed, but due to the cost issues with the AI API, the link has been set to private 🥲.
My flutter version : Flutter 3.23.0-8.0.pre.3