Publication
Employing retrieval augmented generation to optimize LIMS for the legal domain: evaluating methods to improve chatbot performance
| Summary: | This paper explores the application of Large Language Models (LLMs) in the legal domain, uti lizing Retrieval Augmented Generation (RAG) to optimize the performance of Llama 2. First, we demonstrate that integrating RAG with various prompting methods and a classification step significantly enhances Llama 2-Chat’s effectiveness. Furthermore, we show RAG’s capability in document selection and ranking, proving its utility in legal document analysis. Our findings affirm the potential and merit of employing LLMs in legal settings. The study also opens avenues for future research, including Query Expansion, further integration of ranking models with chatbots, combining LLMs with tools, and exploring hybrid retrieval mechanisms. |
|---|---|
| Main Authors: | Schumann, Lorenzo Oliver |
| Subject: | Large language models Retrieval augmented generation Prompt engineering Embedding reranking & classification |
| Year: | 2024 |
| Country: | Portugal |
| Document type: | master thesis |
| Access type: | open access |
| Associated institution: | Universidade Nova de Lisboa |
| Language: | English |
| Origin: | Repositório Institucional da UNL |
| Summary: | This paper explores the application of Large Language Models (LLMs) in the legal domain, uti lizing Retrieval Augmented Generation (RAG) to optimize the performance of Llama 2. First, we demonstrate that integrating RAG with various prompting methods and a classification step significantly enhances Llama 2-Chat’s effectiveness. Furthermore, we show RAG’s capability in document selection and ranking, proving its utility in legal document analysis. Our findings affirm the potential and merit of employing LLMs in legal settings. The study also opens avenues for future research, including Query Expansion, further integration of ranking models with chatbots, combining LLMs with tools, and exploring hybrid retrieval mechanisms. |
|---|
Funded activities
No funded projects found.