Discriminative Predicate Path Mining for Fact Checking in Knowledge Graphs (1510.05911v2)

Abstract: Traditional fact checking by experts and analysts cannot keep pace with the volume of newly created information. It is important and necessary, therefore, to enhance our ability to computationally determine whether some statement of fact is true or false. We view this problem as a link-prediction task in a knowledge graph, and present a discriminative path-based method for fact checking in knowledge graphs that incorporates connectivity, type information, and predicate interactions. Given a statement S of the form (subject, predicate, object), for example, (Chicago, capitalOf, Illinois), our approach mines discriminative paths that alternatively define the generalized statement (U.S. city, predicate, U.S. state) and uses the mined rules to evaluate the veracity of statement S. We evaluate our approach by examining thousands of claims related to history, geography, biology, and politics using a public, million node knowledge graph extracted from Wikipedia and PubMedDB. Not only does our approach significantly outperform related models, we also find that the discriminative predicate path model is easily interpretable and provides sensible reasons for the final determination.

Discriminative Predicate Path Mining for Fact Checking in Knowledge Graphs

The paper "Discriminative Predicate Path Mining for Fact Checking in Knowledge Graphs," authored by Baoxu Shi and Tim Weninger, tackles the increasingly pivotal challenge of fact validation within massive datasets. In an era characterized by rapid information generation, conventional fact-checking techniques often fall short in pace and scale. The authors address this issue by conceptualizing fact checking as a link prediction task within a knowledge graph, thus leveraging the structural representation and relational integrity inherent to such graphs.

Overview

The proposed method pivots on discriminative path mining within knowledge graphs, where the authors have devised a procedure to ascertain the veracity of statements structured as (\textsf{subject}, \textsf{predicate}, \textsf{object}). This is accomplished by examining paths that define generalized statements, such as (\textsf{U.S. city}, \textsf{predicate}, \textsf{U.S. state}), and using mined rules that provide insight into the truthfulness of individual claims. Unlike some existing models hampered by constraints such as untyped graphs or reliance on predefined meta paths, this approach incorporates connectivity, type information, and predicate interactions, thus aiming to refine computational fact-checking paradigms.

Experimental Framework

The authors validate their model across thousands of factual claims from domains including history, geography, biology, and politics, using a knowledge graph composed of entities extracted from large-scale public resources like Wikipedia and PubMedDB. The efficiency of the model is benchmarked against existing link prediction methodologies such as Adamic/Adar and PageRank, as well as statistical relational learning models like RESCAL and TransE. Results indicate significant performance improvements, not only in predictive accuracy but also in interpretability — a crucial aspect for explaining reasoning in AI systems.

Technical Contributions

1. Discriminative Path Mining Algorithm: The paper introduces a novel algorithm capable of autonomously discovering "definitions" of RDF-style triples within large-scale knowledge graphs. This facilitates the use of mined paths for the prediction of truthfulness, avoiding reliance on exhaustive path enumeration or human annotated selections.
2. Interpretable Fact Checking Framework: The methodology developed allows using discriminative paths to model belief veracity in statements. This aligns computational processes closer to human reasoning models where context-dependent predicates can be adapted for fact qualification.
3. Empirical Validation: Comparative analysis on datasets from DBpedia and SemMedDB with the proposed framework resulted in outperforming alternative approaches in both prediction and execution time, ensuring scalability for practical applications.

Implications and Future Directions

The research establishes a precedent for integrating discriminative predicate paths into broader computational linguistics and AI frameworks. The interpretability of findings, crucial for accountability in AI-driven decisions, suggests significant potential in domains like automated knowledge graph construction and open-domain question answering. Furthermore, the capability to dynamically adjust to unseen predicates indicates versatility within rapid data contexts and evolving information landscapes.

Theoretically, this work invites exploration into adjacency adjustments in networks for higher-order logic representation, possibly enhancing systems that cumulatively learn and predict complex relational interactions. Moreover, enhancing entity representation and transitive fact qualification can expand the framework’s utility in dynamic knowledge environments beyond static predicate analysis.

By reframing fact verification within knowledge graphs with emphasis on discriminative path mining, this paper contributes substantially to computational fact-checking paradigms, offering a scalable and interpretable model that bridges structural learning with real-world applicability.