Graphical Models of False Information and Fact Checking Ecosystems

arXiv:2208.11582 · cs.SI, cs.AI, cs.CR · Submitted 2026-08-11 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: AI Radio. Generated commentary on the latest Artificial Intelligence papers.

Tom: Next we'll be talking about the paper "Graphical Models of False Information and Fact Checking Ecosystems".

Jane: The paper was written by Haiyue Yuan, Enes Altuncu, Shujun Li, Can Başkent and Jason R.C. Nurse from University of Kent and Istanbul University-Cerrahpaşa and Middlesex University.

Tom: Stay tuned as we take you through the paper and discuss its implications.

Title: Tom: Welcome back to the show, everyone. Today we’re looking at a paper that’s been making the rounds on arXiv, and it’s called “A Graphical Model of the False Information and Fact-Checking Ecosystem.” Jane, I have to say, the title alone got me excited because it promises to map out something we all deal with every day.

Jane: Absolutely, Tom. And what I love about this title is that it’s not just about fake news or fact-checkers in isolation. It’s about the whole ecosystem — the connections between everyone involved. The authors, Haiyue Yuan, Enes Altuncu, Shujun Li, Can Başkent, and Jason Nurse, they’re from the University of Kent and Middlesex University, and they’ve basically built a visual map of how false information moves and how fact-checking fits into that picture.

Tom: And that’s the part that got me, Jane. We always hear about misinformation spreading, but we rarely see the full network laid out. This paper uses something called an enhanced entity-relationship model, which sounds technical, but really it’s just a way of drawing boxes and arrows to show who does what and how they connect.

Jane: Right, and the key insight is that the ecosystem includes everyone — not just the people spreading false information and the fact-checkers debunking it, but also media outlets, regulators, automated agents, even the platforms where all this happens. The paper’s authors argue that we’ve been studying pieces of this puzzle separately, but we haven’t had a single model that shows the whole thing at once.

Tom: So it’s like having a map of a city instead of just knowing a few streets. And the implications are huge, because if you can see the whole map, you can start to understand where the bottlenecks are, where the vulnerabilities are, and where fact-checking can actually make a difference.

Jane: Exactly. And the authors make a point that this model can be used by researchers and practitioners alike. It’s not just an academic exercise. It’s a tool that can help journalists, policymakers, and even the rest of us understand how false information flows and where interventions might work best.

Tom: I’m curious about the practical side, though. Lu, you’re our senior researcher — what do you think about this approach of modeling the ecosystem visually?

Lu: I think it’s a really smart move, Tom. In my work, I see how often we get lost in the details of individual algorithms or specific case studies. Having a graphical model forces you to step back and see the whole picture. And the fact that they’ve included automated agents and AI-generated content is particularly timely. That’s something older models didn’t really account for.

Jane: And that’s the hook for us, because next we’re going to dig into what the paper actually proposes in detail. Stay with us.

Summary: Tom: So we’re back, and we’re still talking about “A Graphical Model of the False Information and Fact-Checking Ecosystem.” Jane, we’ve covered the title and the authors, but now let’s get into what this paper actually does.

Jane: Right, Tom. So the core of the paper is a diagram — and I know that sounds simple, but it’s actually a really comprehensive diagram. The authors identified twenty-one different entity types, things like Person, Organisation, Media Outlet, Fact-checking Outlet, Account, Automated Agent, Service, and even things like Legal and Regulatory Documents. And then they mapped out all the relationships between these entities.

Tom: And those relationships are where the real insight lives. For example, they show that a Person can publish information, consume information, fact-check information, and also be a member of an organisation. And an Organisation can operate accounts, publish news, and also regulate or be regulated. It’s a web of connections, not just a simple chain.

Jane: Exactly. And one of the things I really appreciated is that they didn’t just draw the diagram and stop. They tested it against real-world scenarios. They modeled things like the BBC Breakfast incident from two thousand twenty-two where old footage of a Russian military parade was mistakenly used to illustrate the invasion of Ukraine. They showed how fact-checkers like Full Fact caught that error and published corrections.

Tom: And that’s where the model proves its worth, because when you see the incident laid out as a graph, you can spot things you might otherwise miss. Like the fact that the misinformation was already out there on Twitter before BBC used it, and that multiple fact-checkers were working on it simultaneously. The model makes those connections visible.

Jane: They also modeled the Trump Twitter suspension incident, the fact-checking of multimedia content, and even a case where PolitiFact fact-checked a pro-Russian “fact-checking” service called War on Fakes. That last one is fascinating because it shows that fact-checking itself can be a target for disinformation.

Lu: What I find compelling is that the model isn’t just descriptive. It’s generative. Once you have this graph, you can start asking questions like “What happens if this node disappears?” or “What if this relationship changes?” It becomes a tool for thought experiments.

Meng: But Lu, I have to ask — as someone who builds systems, how do you actually implement something like this? It’s one thing to draw a diagram, but another to turn it into something that runs.

Lu: That’s a fair question, Meng. The authors actually address that. They suggest that the model could be turned into a computational ontology, which is basically a machine-readable version of the diagram. Then you could use natural language processing and knowledge graph construction to automatically populate it with real-world data.

Meng: So you’re saying this could become a living map that updates itself as new information comes in?

Lu: Exactly. Imagine a fact-checking tool that not only checks a claim but also understands where that claim came from, who’s spreading it, and what regulatory frameworks apply. That’s the potential here.

Jane: And that’s a perfect setup for our next segment, where we talk about the improvements the paper suggests. Don’t go anywhere.

Improvements: Tom: We’re back with “A Graphical Model of the False Information and Fact-Checking Ecosystem,” and Jane, we’ve covered the model itself and how it works. Now let’s talk about what the authors think we should do next.

Jane: Right, Tom. The paper doesn’t just stop at proposing the model. The authors are pretty clear about what they see as the next steps, and the big one is building a computational ontology. That means taking this visual diagram and turning it into something a computer can actually use — a structured, machine-readable version.

Lu: And that’s the part that excites me the most. Because once you have a computational ontology, you can start doing automated reasoning. The system could infer new relationships, flag inconsistencies, and even predict where false information might spread next. It’s like giving the model a brain.

Meng: But I want to push back a little here. Turning a diagram into a working system is hard. You need to populate it with real data, and that means entity recognition, relationship extraction, and a lot of cleaning. The authors acknowledge this, but I’d love to see more specifics on how they’d actually do it.

Jane: That’s a fair point, Meng, and the authors do mention that advanced NLP techniques and large language models would probably be necessary. They’re not pretending it’s easy. But they also point out that the payoff is worth it — you could build fact-checking tools that are aware of the whole ecosystem, not just individual claims.

Tom: And there’s another improvement they suggest that I really like. They talk about using the model for agent-based simulation. So instead of just describing the ecosystem, you could simulate it — create virtual actors, let them interact, and see how false information spreads under different conditions.

Lu: That’s a powerful idea. You could test different intervention strategies in a simulated environment before rolling them out in the real world. For example, you could ask “What happens if we increase fact-checking speed?” or “What if we change the way platforms flag misinformation?” and get answers without any real-world risk.

Meng: So it’s like a flight simulator for misinformation. That could be genuinely useful for policymakers.

Jane: Exactly. And the authors also suggest using the model to guide literature reviews, which is something they demonstrate in the paper. They analyzed recent research and showed how the model can help identify gaps — areas of the ecosystem that haven’t been studied enough.

Tom: And what did they find? What’s the gap?

Jane: They found that most research focuses on the relationship between users and information — how people consume and share false information. But there’s much less work on the regulatory side, on the relationships between regulators and services, and on how automated agents fit into the picture.

Lu: That’s a really useful finding. It tells researchers where to focus their energy next.

Tom: And it tells us where the field is heading. But before we wrap up, let’s hear from Lalam about the bigger cultural picture.

Lalam: I’d like to add that this model has the potential to improve how we think about information literacy. When people can see the whole ecosystem — how false information is created, spread, and countered — they can become more discerning consumers of information. This isn’t just a technical tool; it’s a cultural one.

Jane: That’s a beautiful way to put it, Lalam. And it sets us up perfectly for our conclusion. Stay with us.

Conclusion: Tom: So we’ve reached the end of our discussion on “A Graphical Model of the False Information and Fact-Checking Ecosystem.” Jane, what’s the big takeaway for our listeners?

Jane: The big takeaway, Tom, is that this paper gives us a map — a comprehensive map of the entire false information and fact-checking ecosystem. It shows us who the players are, how they connect, and where interventions can make a difference. And it’s not just a static picture; it’s a foundation for building tools that can help us fight misinformation more effectively.

Tom: And we’ve seen it work in practice. The authors modeled real-world incidents, from the BBC Breakfast mistake to the Trump Twitter suspension, and the model held up. It revealed connections and insights that might have been missed otherwise.

Lu: I’d add that the model’s flexibility is its strength. It can be extended, adapted, and eventually turned into a computational system that could power the next generation of fact-checking tools.

Meng: And from a practical standpoint, the fact that they’ve already demonstrated how to use it for literature review means it’s not just theoretical. It has immediate applications for researchers.

Lalam: And for society, it offers a way to think about information ecosystems more holistically, which is essential if we want to build resilience against false information.

Jane: Well said, everyone. So we’re saying goodbye to this paper, but we’re taking its ideas with us. It’s given us a new lens for understanding a problem that affects all of us.

Tom: And we’re already looking forward to the next paper on our list. Thanks for listening, and we’ll see you next time.

Jane: Take care, everyone.

Haiyue Yuan, Enes Altuncu, Shujun Li, Can Başkent, Jason R.C. Nurse

University of Kent · Istanbul University-Cerrahpaşa · Middlesex University

cs.SI, cs.AI, cs.CR

Submitted: 2026-08-11

Updated: 2026-08-12

Project page: https://twitter.github.io/birdwatch/18https://www.logically.ai

License: http://creativecommons.org/licenses/by/4.0/

Importance score: 53/100

The gist: "Despite a lot of research on online false information, we noticed a lack of conceptual models describing the complicated ecosystem of false information and fact-checking." The proposed model is an

Terminology

Summary

Summary

The paper introduces the first comprehensive graphical model of the ecosystem surrounding false information online and fact-checking, addressing a noted research gap: Despite a lot of research on online false information, we noticed a lack of conceptual models describing the complicated ecosystem of false information and fact-checking. The proposed model is an enhanced entity-relationship (EER) model that covers a wide range of entities and relationships, focusing on false information online in multiple contexts, including traditional media outlets, as well as user-generated and AI-generated content.

The authors justify the need for such a model by noting that false information is expected to become the most severe threat in the next two years according to the Global Risks Report published by the World Economic Forum in 2025. They also highlight that while many researchers have investigated fact-checking, there is a lack of conceptual models describing the ecosystem, and such models "can help develop a better understanding of different stakeholders, relationships and processes that are involved in the ecosystem and enable more comprehensive analyses of the problem of false information at the ecosystem level."

The paper reviews related work on false information types, distinguishing between misinformation (false information whose creator did not have the intention to cause harm) and disinformation (information that was intentionally created to cause harm via different approaches such as deceptive advertising, government propaganda, and doctored photographs). It also mentions malinformation, which describes the scenario when genuine information is shared to harm rather than serve the public interest. The authors also review fact-checking approaches, categorizing them into "experts/professionals fact-checker approach, crowd-sourced approach, machine learning approach, natural language processing technique, hybrid technique, expert-crowdsource approach, human-machine approach, graph-based method, deep learning approach, and recommendation system approach."

To construct the model, the authors adopted the Ontology Development 101 methodology, which is a simple guide with 7 key phases based on iterative design. The phases included determining the domain and scope, considering reusing existing ontologies, enumerating important terms, defining classes and hierarchies, defining properties and facets of slots, and creating instances. The authors note that they were inspired by an entity-type graph-based semantic model for illustrating the disclosure of users' personal data to different entities and the benefits of such data-sharing activities.

The resulting graphical model is formalized as a directed graph G = (V, E), where V represents a set of N nodes (entity types) and E represents a set of M edges (semantic relationships). The model contains N = 21 different entity types, which are defined in detail. The entity types are grouped and colour-coded, with a superclass entity called Actor covering all actors in the model. Entity types representing individual actors, organisational actors, and other entities having relationships with actors are coloured green, red, and blue, respectively.

The entity types include: Actor (an individual, organisation, account, or service involved in the ecosystem), with six subclasses: Person (an individual in the physical world, with subclasses including Journalist, Expert, and Fact-checking Individual), Organisation (an organised group of people with a particular purpose, with subclasses including Media Outlet, Fact-checking Outlet, Lawmaking Body, and Regulatory Body), Group/Association (a number of individuals and/or organisations that share common interests, with subclasses including Media Outlet Association, Journalist Association, and Fact-checking Association), Account (a virtual identity of an individual or organisation using online services), Automated Agent (software that performs actions autonomously, including AI agents and content moderation bots), and Service (a platform, tool, or dataset provided by an actor to the public). The model also includes four types of Information: News (a report of recent events or information that attracts public interest), Comment (an actor's opinion, feedback, or other information in response to published news or posted content), Fact-checking Report (the outcome of a veracity investigation for a piece of information), and Legal & Regulatory Document (a law, or a set of regulations and/or implementation rules).

The relationships in the model are examined concerning key roles: information generation and propagation, information consumption, fact-checking, and lawmaking and regulation. For information generation, the model supports scenarios where a Person generates information, an Organisation generates information, a Group/Association generates information, and an Automated Agent autonomously drives information generation and propagation. For information consumption, the model supports scenarios where a Person consumes information, an Organisation consumes information (e.g., for monitoring purposes), and an Automated Agent consumes information through mechanisms like data ingestion from APIs and web scraping. For fact-checking, the model uses a self-claimed fact-checker attribute to identify actors dedicated to fact-checking, and it covers fact-checking performed by independent fact-checkers, fact-checking outlets, media outlets with fact-checking units, and automated agents. For lawmaking and regulation, the model covers the relationships between Lawmaking Body, Regulatory Body, and Legal & Regulatory Document entity types.

The paper demonstrates the usefulness of the proposed model through two example applications. The first application models seven real-world scenarios, including: 1) the BBC Breakfast incident where a video clip of military planes was misidentified as Russian military flying into Ukraine; 2) a person using fact-checking services and resources from organisations like Full Fact, Snopes, Logically, and NewsGuard; 3) voluntary regulators of the UK media, including IPSO and the Editors' Code Committee; 4) journalists and their associations in the US, including SPJ, RTDNA, and IRE; 5) Trump's Twitter account suspension incident; 6) fact-checking multimedia content using tools like Google Image Search and TinEye; and 7) fact-checking fact-checkers, based on the incident where PolitiFact reviewed fact-checks published by the pro-Russian War on Fakes service and judged them to be Russian disinformation.

The second example application demonstrates how the model can be used to review research literature. The authors performed a search in Google Scholar with the query "misinformation" OR "disinformation" OR "false information" OR "fake news" OR "fact-checking" and selected the top 10 peer-reviewed English articles published in 2023. They generated tags in the format Entity1 Relationship Entity2 for each paper, which allowed them to identify key themes in the literature, including false information sharing, debunking and prebunking, and false information regulation. The authors note that the tags shown in Table 1 indicate that user-information relationships were more studied than other portions of the graphical model and that most studies focused on information generation or consumption, rather than fact-checking.

The authors discuss potential future work, including defining a computational ontology and automatically building it from data sources, which "can formalise the proposed model, facilitating large-scale Internet measurement studies to better understand how false information spreads and how different actors behave, and allowing automated reasoning to automatically discover new phenomena related to false information and fact-checking. They also suggest that the model can be used for modelling and computer-based simulation of the false information and fact-checking ecosystem, especially agent-based models, and that the many different types of conflicting and cooperative relationships between different entities" can help researchers use theoretical tools such as game theory and epistemic logic.

The paper concludes that "the usefulness of such a conceptual model is demonstrated using some example real-world scenarios and an example literature review, and its wider applications for studying false information online and fact-checking are discussed. It is our hope that the work can stimulate more work on conceptual modelling of false information online and fact-checking."

Improvements for AI systems

Based on the paper, here are the specific improvements I can make to AI systems:

Improvement: Enhance fact-checking AI with a structured knowledge graph based on the 21 entity types and 15+ relationship types defined in the EER model.

What the improved AI can do:

  • Automatically classify actors in a claim into Person, Organisation, Account, Automated Agent, etc.

  • Trace information propagation paths (e.g., Person → Account → Service → Media Outlet)

  • Distinguish between misinformation (unintentional) and disinformation (intentional) by analyzing creator intent attributes

  • Detect when an AI agent or bot (Automated Agent entity) is the source of false information

  • Model the complete fact-checking workflow: from claim detection → fact-checker assignment → report publication → audience consumption

Improvement: Implement the Fact-checking Fact-checkers capability (Section 4.7) into AI systems.

Improvement: Build AI systems that leverage the Service entity type and its relationships (Provides, Consumes) to integrate external verification tools.

Improvement: Incorporate the Lawmaking Body, Regulatory Body, and Legal & Regulatory Document entities into AI systems.

Improvement: Use the time attributes (shown in examples like 2022-02-25T07:46:00) to build AI systems that analyze information spread chronologically.

Improvement: Convert the EER model into a computational agent-based simulation framework.

Improvement: Use the model as a systematic tagging framework for AI-powered literature analysis.

Abstract

The wide spread of false information online, including misinformation and disinformation, has become a major problem for our highly digitised and globalised society. A lot of research has been done to better understand different aspects of false information online such as behaviours of different actors and patterns of spreading, and also on better detection and prevention of such information using technical and socio-technical means. One major approach to detecting and debunking false information online is to use fact-checkers. Despite a lot of research on online false information, we noticed a lack of conceptual models describing the complicated ecosystem of false information and fact-checking. In this paper, we introduce the first comprehensive graphical model of the ecosystem, focusing on false information online in multiple contexts, including traditional media outlets, as well as user-generated and AI-generated content. The proposed enhanced entity-relationship (EER) model covers a wide range of entities and relationships, and it can be a new useful tool for researchers and practitioners to study false information online and the effects of fact-checking. To demonstrate its usefulness, we provide two example applications of our proposed model, modelling real-world scenarios and reviewing the research literature.

Sources

Related papers