Sign in to view source links and access this dataset
Description
313 long documents averaging 65.1 pages serve as an evaluation set for multimodal document information retrieval. The documents are categorized into ten domains including research reports, academic papers, financial reports, and government documents. The dataset, created by MMDocIR, features a modality distribution of 60.4% text, 18.8% image, and includes tables.
Use Cases
Benchmarking multimodal document retrieval systems based on the mix of text, images, and tables.
Evaluating domain-specific retrieval performance based on the ten distinct document categories.
Training or testing models for long-document understanding based on documents averaging 65.1 pages.
Strengths
Contains 313 documents, providing a substantial evaluation corpus.
Documents average 65.1 pages, focusing on long-form content.
Covers ten distinct domains, enabling cross-domain evaluation.
Provides a quantified modality distribution: 60.4% text, 18.8% image, and tables.
Limitations
Column-level documentation is absent; field semantics must be inferred after download.
Row count is unknown, which may limit suitability assessment for certain tasks.
Provenance
Source
MMDocIR
Freshness
Last updated 2025-05-28 03:08:56; freshness should be verified.
License is unknown; terms of use must be verified before application.