A carregar...
Publicação
LLM’s Dialogue with Persona and Long-term Memory
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 1.08 MB | Adobe PDF |
Autores
Orientador(es)
Resumo(s)
This master thesis explores the integration of long-term memory and persona dialogue capabilities into LLMs', addressing key limitations in current technologies such as lack of contextual retention, personalization, and consistency over extended interactions. The research identifies significant gaps in LLMs' ability to store and retrieve contextually relevant information, which hinders their effectiveness in dynamic, user-driven applications.
The objectives of this study include designing and implementing a framework that enables LLMs to dynamically manage user-specific information and sustain coherent, personalized interactions. To achieve this, the methodologies employed the Design Science Research (DSR) and PRISMA approach, which facilitated the iterative development and evaluation of the proposed solution. The framework leverages state-of-the-art techniques such as Retrieval-Augmented Generation (RAG) and Reflexion mechanisms process to ensure adaptive, efficient, and ethically managed dialogue systems. Notably, the design avoids fine-tuning by employing advanced prompting and metadata strategies, ensuring scalability and feasibility across diverse applications. The results demonstrate that the proposed system significantly improves LLMs' ability to maintain
long-term context, adapt to user preferences, and deliver coherent persona-driven responses. Through the implementation of mechanisms such as a topic manager and a selective forgetting system, the artifact provides an enhanced conversational experience by balancing memory efficiency and relevance. This research contributes to the advancement of conversational AI by offering a scalable and adaptable framework for enhancing LLMs. The findings underscore the potential for these systems to redefine applications in areas such as customer service, personalized education, and virtual assistance, fostering more natural and engaging human-AI interactions.
Descrição
Dissertation presented as the partial requirement for obtaining a Master's degree in Data Science and Advanced Analytics, specialization in Data Science
Palavras-chave
Large Language Models Long-term Memory Persona Dialogue Chatbot