Retrieval-Augmented Generation (RAG) struggles in do- main-specific corpora with structured metadata and multimodal content, where text-only methods miss key information and similarity-based retrieval fails to ensure contextual compatibility. We propose a multimodal, localisation-aware RAG pipeline for industrial analytics that combines vision-to-text ingestion with metadata-aware retrieval. Evaluated on clinical market analysis presentations, retriever-filtered RAG improves retrieval quality (+0.145 NDCG) while remaining computationally efficient for production settings, whereas LLM-filtered RAG achieves higher precision but requires stronger reasoning capabilities

Localisation-Aware RAG for Industrial Data: A Case Study on Clinical Market Analysis / Noto, P., Monaco, S., Benedetto, I., Apiletti, D.. - 16950:(2027), pp. 463-467. (Applied Data Science Track, Demo Track and Industrial Track. ECML PKDD 2026 Napoli (IT) September 7–11, 2026) [10.1007/978-3-032-37685-5_56].

Localisation-Aware RAG for Industrial Data: A Case Study on Clinical Market Analysis

Noto, Pietro;Monaco, Simone;Benedetto, Irene;Apiletti, Daniele
2027

Abstract

Retrieval-Augmented Generation (RAG) struggles in do- main-specific corpora with structured metadata and multimodal content, where text-only methods miss key information and similarity-based retrieval fails to ensure contextual compatibility. We propose a multimodal, localisation-aware RAG pipeline for industrial analytics that combines vision-to-text ingestion with metadata-aware retrieval. Evaluated on clinical market analysis presentations, retriever-filtered RAG improves retrieval quality (+0.145 NDCG) while remaining computationally efficient for production settings, whereas LLM-filtered RAG achieves higher precision but requires stronger reasoning capabilities
2027
9783032376848
9783032376855
File in questo prodotto:
File Dimensione Formato  
RAGStrategies_ECML_IndustryTrack.pdf

embargo fino al 08/09/2027

Tipologia: 2. Post-print / Author's Accepted Manuscript
Licenza: Pubblico - Tutti i diritti riservati
Dimensione 260.67 kB
Formato Adobe PDF
260.67 kB Adobe PDF   Visualizza/Apri   Richiedi una copia
Pubblicazioni consigliate

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11583/3017907