Comparative Analysis of RAG-Based Methods for Multilingual SwarmLexAI

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

This research compares the effectiveness of different Retrieval-Augmented Generation (RAG) methods in processing multilingual AI-related legal documents, focusing on text summarization and content comprehension. Specifically, it contrasts standard RAG-based applications with the GraphRAG and Lazy GraphRAG approaches, conducting a cross-lingual analysis of legal texts in English, Spanish, German, and French. Using SwarmLexAI, an AI-native document analyzer, along with the Swarm API from OpenAI, spaCy, and fastText, legal materials are processed to extract key entities, keywords, and legal clauses. In addition, an AI-driven document converter improves the accuracy of summarization across languages, improving contextual understanding. A visualization system generates interactive knowledge graphs to analyze the relationships between extracted legal concepts that illustrate connections across linguistic and cultural boundaries. The study highlights trade-offs between GraphRAG, which offers in-depth, structured legal analysis, and Lazy GraphRAG, which is optimized for speed and high-level document insights. The results contribute to cross-lingual legal comparisons and enhance AI-native processing of legislation. The study aims to compare AI-related legislation across different countries by identifying these cross-lingual relationships.

Original languageEnglish
Title of host publicationIntelligent Systems and Applications - Proceedings of the 2025 Intelligent Systems Conference IntelliSys
EditorsKohei Arai
PublisherSpringer Science and Business Media Deutschland GmbH
Pages229-245
Number of pages17
ISBN (Print)9783032071088
DOIs
StatePublished - 2026
Event11th Intelligent Systems Conference, IntelliSys 2025 - Amsterdam, Netherlands
Duration: 28 Aug 202529 Aug 2025

Publication series

NameLecture Notes in Networks and Systems
Volume1660 LNNS
ISSN (Print)2367-3370
ISSN (Electronic)2367-3389

Conference

Conference11th Intelligent Systems Conference, IntelliSys 2025
Country/TerritoryNetherlands
CityAmsterdam
Period28/08/2529/08/25

Keywords

  • AI-native document analyzer
  • GraphRAG
  • Lazy GraphRAG
  • SwarmLexAI

Fingerprint

Dive into the research topics of 'Comparative Analysis of RAG-Based Methods for Multilingual SwarmLexAI'. Together they form a unique fingerprint.

Cite this