Page aware retrieval improves French PDF question answering accuracy
Page-Aware Retrieval-Augmented Generation for EvalLLM 2026: A Five-Variant Study on French PDFs
Artificial Intelligence
Summary
Finding the right page in French PDF documents is key to answering questions accurately. The authors tested five ways to combine search methods, starting with a basic keyword search called BM25. They found that mixing methods and breaking down queries helped pick better pages and answers, though it made the system slower and caused some errors. This suggests improving page selection is more important than just understanding meanings better.
What this means in practice
- •For document management teams: Improve question answering systems that retrieve precise pages from French PDF files to support accurate responses.
- •For legal support teams: Enhance retrieval of specific pages in legal PDF documents to provide quicker, more accurate answers during case reviews.
Authors
Abdelhak kelious
Abstract
We study retrieval-augmented generation (RAG) for questions about French PDF documents when both the answer and its supporting document pages are evaluated. Five system variants add dense retrieval, rank fusion, reranking, and query decomposition to a BM25 baseline. On 595 challenge questions, the complete system scores 0.4450 MRR@10 and 0.4013 Recall@10, compared with 0.3430 and 0.2994 for BM25. Dense retrieval alone and a simple lexical--dense fusion both underperform BM25. Reranking improves the hybrid system, whereas adding query decomposition produces the largest further gain, with higher latency and more detected output artifacts. The complete system slightly exceeds the reported anonymous overall mean on two answer metrics but falls below it on most page-retrieval metrics. These results identify accurate page selection, rather than semantic retrieval in isolation, as the main opportunity for improvement in this setting.