FakeKG: A Knowledge Graph of Fake Claims for Improving Automated Fact-Checking (Student Abstract)

Gautam Kishore ShahiAAAI 2023 · DOI

TL;DR

This paper introduces FakeKG, a novel knowledge graph designed to improve automated fact-checking by centralizing and organizing previously debunked claims. By leveraging semantic technologies and a SPARQL endpoint, FakeKG aims to help fact-checkers quickly identify and categorize claims, thereby reducing workload and improving efficiency, especially during high-demand periods like elections or crises.

What Problem Does This Paper Address?

The paper tackles the growing challenge of misinformation, particularly the "3V problem" (Volume, Velocity, Variety) faced by fact-checking organizations. During critical events like pandemics or elections, the sheer number of claims, the speed at which they spread, and their multimodal nature (text, images, videos) overwhelm human fact-checkers. Existing solutions, like manual verification by experts or crowd-sourced platforms, struggle with scalability and consistency. While initiatives like Google's ClaimReview and Fact Check Markup Tool exist, they lack a real-time, centralized corpus of fact-checked articles and often contain errors or limited information. The authors identify a critical gap: the absence of a structured, queryable knowledge base that can aid in the automated and efficient verification of fake claims.

Methods

The core of the proposed solution is FakeKG, a knowledge graph built using a semantic-technology-based approach. The process involves:

  • Data Collection: Gathering a large dataset of previously debunked fake claims and their ratings. The authors collected approximately 113,006 fact-checked data points across 40 languages from 105 countries, stored in a MySQL database.
  • Knowledge Graph Creation: This involves building a "Fake Ontology" (a custom ontology) to conceptualize the data. Relational tables are mapped to RDF statements using R2RML mapping rules. A tool called SDM-RDFizer is used to transform these table rows into RDF data.
  • Entity Linking: Entities within the knowledge graph are linked to external data sources using owl:sameAS to enrich the information.
  • SPARQL Endpoint: A SPARQL endpoint is provided, allowing for complex queries to retrieve information about claims, similar claims, or topics. This enables automated fact-checking by machines.

The pipeline for creating FakeKG is illustrated in Figure 1, showing the flow from data sources to the final knowledge graph with a SPARQL endpoint.

Datasets

The paper mentions the collection of around 113,006 fact-checked data points from various sources, collected using the AMUSED framework. This data spans 40 languages and 105 countries. The authors also reference the FakeCovid dataset (Shahi and Nandini 2020) as a related resource. No new benchmark datasets are introduced; rather, the paper focuses on creating a structured knowledge graph from existing fact-checked information.

Key Insights and Ideas

  • Centralized Knowledge for Efficiency: The most significant insight is the potential of a centralized, queryable knowledge graph to drastically improve the efficiency of fact-checking. Instead of re-verifying known claims, systems can quickly retrieve existing judgments.
  • Semantic Technologies for Scalability: Leveraging ontologies and SPARQL allows for a structured and scalable approach to managing and querying vast amounts of fact-checking data, overcoming the limitations of manual methods.
  • Reducing Fact-Checker Workload: By prescreening claims and filtering out those already debunked, FakeKG can significantly reduce the workload on human fact-checkers, allowing them to focus on novel or complex cases.
  • Cross-Lingual and Multimodal Potential: The proposed approach aims to facilitate cross-lingual and multimodal fake news analysis, a crucial step towards tackling the "variety" aspect of misinformation.

Key Results

The paper is a student abstract proposing an approach, and as such, it does not present quantitative results from a deployed system. The key "result" is the proposal of the FakeKG approach and its functional pipeline. The authors state that the proposed FakeKG will be useful in analyzing the propagation of fake news across different news and will be available as open source. The primary benefit highlighted is the ability to answer complex queries about similar claims or topics, which is difficult with current unstructured data.

Limitations and Future Work

Limitations: * The paper is a student abstract, and the proposed approach has not yet been fully implemented and evaluated. * The effectiveness of the FakeKG in real-world scenarios and its performance against existing fact-checking methods are not yet demonstrated.

Future Work: * The authors plan to implement the proposed approach of FakeKG. * They intend to provide a list of SPARQL queries as use cases to demonstrate its functionality. * The goal is to make the tool available as open source. * Future work will also focus on implementing the approach to analyze the propagation of fake news across different news and languages.

Summary updated

BibTeX
@inproceedings{shahi2023fakekg,
  author = {Gautam Kishore Shahi},
  title = {{FakeKG: A Knowledge Graph of Fake Claims for Improving Automated Fact-Checking (Student Abstract)}},
  booktitle = {AAAI Conference on Artificial Intelligence},
  year = {2023},
  doi = {10.1609/aaai.v37i13.27020},
}
Figure