Source-linked AI summary
MMFact: Simplifying Multimodal Misinformation Detection with Shared Resources and Tools
Yuchen Miao, Zijun Wang, Mingtai Zhang, Xiaoyi Chen, Sihan Deng, Xianglin Ding
TL;DR
Multimodal misinformation detection uses varied preprocessing and auxiliary resources, making comparisons difficult to interpret. MMFact brings shared materials together with detector implementations and common training and evaluation procedures; compatible methods can use the same inputs to support fairer comparisons while preserving specialized designs.
Problem
Different preprocessing pipelines and auxiliary resources make multimodal misinformation comparisons difficult to interpret.
Source: §1 · p. 1Method
Results
Takeaways & Limitations
The shared materials and tools provide a practical foundation for developing and evaluating multimodal misinformation detectors.
Source: §4 · p. 2Weibo-21 was reconstructed because the original official dataset was unavailable.
Source: §2 · p. 1
Abstract
from arXiv · showhide
Multimodal misinformation detection methods often rely on different preprocessing pipelines and auxiliary resources, making experimental comparisons difficult to interpret. We present MMFact, a collection of shared resources and tools that simplifies this research workflow. It brings together 47 detector implementations and baselines, three multimodal datasets, and prepared materials ranging from semantic features to OCR, image descriptions, and external evidence. Compatible methods can use the same materials, reducing variation introduced by preprocessing and supporting fairer comparisons. A modular training and evaluation framework preserves method-specific designs while providing common experimental procedures. By treating prepared materials as a shared research resource, MMFact helps researchers focus on developing and evaluating detection methods.
1 Introduction
Multimodal misinformation detectors may need diverse text, image, and external inputs, whose preparation requires separate models and services. MMFact combines reusable resources with configurable models, training, and evaluation to support fairer comparisons while retaining method-specific designs.
1 Introduction
- Separate preparation of semantic, OCR, visual-region, and external-evidence inputs can add substantial work before detectors are studied.Methods may use different image descriptions, feature extractors, or retrieved evidence, so performance differences can reflect input choices as well as detector designs.Source: §1 · p. 1
- MMFact combines reusable prepared materials with configurable models, training, and evaluation for multimodal misinformation detection.Its emphasis is on shared inputs for fairer comparisons while retaining specialized information and computation that distinguish methods.Source: §1 · p. 1
- MMRec [5] and FaKnow [6] illustrate common infrastructure for implementing and comparing models, which MMFact extends with prepared materials.Source: §1 · p. 1
2 Datasets and Reusable Materials
MMFact packages three datasets and reusable materials for multimodal misinformation detection, with fixed splits and auxiliary inputs for varied methods. Sharing compatible materials reduces variation from preparation choices while preserving method-specific requirements.
2 Datasets and Reusable Materials
- The collection covers Weibo-17, Weibo-21 [2], and GossipCop from FakeNewsNet, totaling 26,361 records with fixed training, validation, and test splits.Weibo-17 retains EANN’s released splits after cleaning; Weibo-21 uses a fixed 8:1:1 repartition.Source: §2 · p. 1
- Prepared materials include semantic and visual representations, regions, crops, frequency information, OCR, image descriptions, translations, summaries, entities, and linguistic annotations.Encyclopedia content, retrieval results, generated arguments, and rationales support methods using external information or multiple perspectives.Source: §2 · p. 1
- Weibo-21 was reconstructed because the original official dataset was unavailable.Source: §2 · p. 1
- Compatible methods can share prepared materials, reducing variation from feature extraction, generation, or retrieval choices and helping attribute experimental differences to methods.Fixed auxiliary content also prevents repeated runs from receiving changing responses from external services.Source: §2 · p. 2
- Shared resources preserve method-specific inputs and trainable encoders when required, supporting controlled comparisons without replacing detectors’ core designs.Source: §2 · p. 2
3 Architecture and Key Features
MMFact organizes data access, training, and evaluation through a common configuration system. It integrates 47 detector implementations and baselines while supporting varied training designs and consistent evaluation procedures.
3 Architecture and Key Features
- Shared data access aligns detector inputs and lets users obtain required materials without downloading every resource.Source: §3 · p. 2
- MMFact integrates 47 implementations: eight baselines and 39 models associated with research papers.A shared training framework supports different fusion strategies and objectives, and methods may use prepared features or learn from textual and visual inputs.Source: §3 · p. 2
- Common evaluation reports overall, class-wise, and domain-wise results where applicable, using standard classification metrics and validation-based model selection.Repeated runs with different random seeds support consistent reporting across methods.Source: §3 · p. 2
- A common configuration system controls dataset, model, encoder, and training choices within a consistent experimental framework.The toolbox organizes data access, model training, and evaluation as connected components.Source: §3 · p. 2
4 Conclusion
MMFact brings shared preprocessing materials and experimental tools together for multimodal misinformation detection. Its common procedures support fairer comparisons and reduce repeated preparation while preserving specialized methods.
4 Conclusion
- MMFact combines shared preprocessing materials with common training and evaluation procedures to support fairer comparisons and reduce repeated preparation work.Source: §4 · p. 2
- Its modular design preserves specialized methods and provides a practical foundation for developing and evaluating new detectors.Source: §4 · p. 2
- MMFact brings shared preprocessing materials and experimental tools together for multimodal misinformation detection.Source: §4 · p. 2