Knowledge Graph Reasoning with Reinforcement Learning for Explainable Fact-checking
Name
Gustav Nikopensius
Abstract
Manual fact checking can not keep up with the pace at which false claims are produced and spread across the web. Computers are much faster at checking facts than humans.
Automated fact checking usually involves comparing a fact claim to some set of knowledge. This comparison is oftentimes carried out by a machine learning algorithm. An effective way of representing knowledge that is also highly machine-readable is Knowledge Graphs. This study frames the problem of computational fact-checking as a reinforcement learning based knowledge graph reasoning problem.
The experimental results reveal that reasoning over a knowledge graph is an effective way of producing human readable explanations in the form of paths and classifications for fact claims. The paths may aid fact-checking professionals with highly readable clues, improving trust and transparency in AI systems.
The artificial intelligence aims to compute a path that either proves or disproves a factual claim, but does not provide a verdict itself. A verdict is reached by a voting mechanism which utilizes paths produced by the artificial intelligence. These paths can be presented to a human reader so that they themselves can decide whether or not the provided evidence is convincing or not. Understanding between AI and humans makes for trust and cooperation.
Automated fact checking usually involves comparing a fact claim to some set of knowledge. This comparison is oftentimes carried out by a machine learning algorithm. An effective way of representing knowledge that is also highly machine-readable is Knowledge Graphs. This study frames the problem of computational fact-checking as a reinforcement learning based knowledge graph reasoning problem.
The experimental results reveal that reasoning over a knowledge graph is an effective way of producing human readable explanations in the form of paths and classifications for fact claims. The paths may aid fact-checking professionals with highly readable clues, improving trust and transparency in AI systems.
The artificial intelligence aims to compute a path that either proves or disproves a factual claim, but does not provide a verdict itself. A verdict is reached by a voting mechanism which utilizes paths produced by the artificial intelligence. These paths can be presented to a human reader so that they themselves can decide whether or not the provided evidence is convincing or not. Understanding between AI and humans makes for trust and cooperation.
Graduation Thesis language
English
Graduation Thesis type
Bachelor - Computer Science
Supervisor(s)
Mohit Mayank, Rajesh Sharma
Defence year
2022