- Keywords: Belief propagation, probabilistic graphical models, probabilistic knowledge base, Multi-modal knowledge bases, multilingual NLP, event coreference, entity coreference, relation coreference
- TL;DR: We present LEAPFROG, a probabilistic graphical model based framework, that maintains alternative probabilistic interpretations for conflicting entities, events and relations across multiple modalities in the automatically constructed knowledge graph.
- Abstract: Populating a knowledge base (KB) from unstructured information has been a widely studied problem. Capturing complex events and relations is especially challenging. Even more challenging is providing coherent interpretations of uncertain and even contradictory information. In this work, we present a novel probabilistic framework — LEAPFROG, for automated knowledge base construction that maintains alternative probabilistic interpretations for entities, events, and relations. To the best of our knowledge, this work is the first attempt at capturing multiple uncertain alternatives. Furthermore, we allow for a domain expert to inject their beliefs and prior knowledge into the system. We show how the expert’s beliefs about the reliability of an information source affect information interpretation.
- Archival status: Archival
- Subject areas: Machine Learning, Natural Language Processing, Information Integration, Knowledge Representation