RA
Richmond Ampah-MensahResearch & Software
AboutResearchSoftwareAchievements
Let's talk
RA
Richmond Ampah-MensahResearch & Software
AboutResearchSoftwareAchievements
Let's talk
All research

Research Square preprint · 2026

Visual, OCR-Text, and Hybrid Evidence Retrieval for Document Visual Question Answering: A Controlled Failure Analysis

A controlled failure analysis that separates evidence retrieval from answer generation on DocVQA and MP-DocVQA.

All research
EvidenceRoleSource

Evidence explorer

Read the study by layer.

Move between the research question, method, recorded result, and interpretation boundary without losing context.

This study compares OCR-text, dense-text, visual, and reciprocal-rank-fusion retrieval under document-disjoint evaluation. It then measures how retrieved evidence affects a fixed vision-language answer generator, so retrieval failures are not confused with generation failures.

Richmond's roleFirst-listed author and repository maintainer
Authors
  • Richmond Ampah-Mensah
  • Muhammad Munsarif
  • Muhammad Sam’an
Methods
BM25BGE-M3MiniLMColQwenLocked reciprocal-rank fusionQwen2.5-VL
View article or demoView source repository
Richmond Ampah-MensahApplied AI research and software engineering
ResearchSoftwareAchievementsEmail