DEVELOPMENT OF AN INTELLIGENT QUESTION ANSWERING SYSTEM FOR PROCESSING LEGISLATIVE TEXTS USING TRANSFORMER MODELS

Abstract

This paper presents the development of an intelligent question answering system for processing legislative texts using transformer models. The proposed approach is based on natural language processing techniques and semantic similarity computation between user queries and normative legal documents. Pre-trained transformer-based language models, including KazBERT and XLM-RoBERTa, are employed to generate vector representations of textual data. Semantic similarity between questions and legislative articles is calculated using the cosine similarity metric, and the most relevant text fragments are selected using an extractive method. The system is implemented as a modular web-based software solution, enabling scalability and further model integration. Experimental evaluation conducted on a corpus of legislative acts of the Republic of Kazakhstan demonstrates satisfactory performance in terms of Accuracy, F1-score, and response time, confirming the applicability of the proposed system for automated intelligent legal information systems.