Graduate Thesis Or Dissertation

 

Knowledge Base Completion Leveraging Natural Language Data Public Deposited

Downloadable Content

Download PDF
https://ir.library.oregonstate.edu/concern/graduate_thesis_or_dissertations/b2774341q

Descriptions

Attribute NameValues
Creator
Abstract
  • Information about named entities (real-world objects) is usually harvested from different sources and organized as a multiple relational directed graph in Knowledge Bases (KBs). KBs play essential roles in many NLP modules including question answering, fact-checking, search engines, etc. KBs are big but still incomplete: relational information among entities is missing. This phenomenon leads to the degradation of the performance for the consumers of a KB. In this work, we aim to mitigate the incompleteness issue by proposing models in two different paradigms of work: i) Extracting information from unstructured/unlabeled textual modality and integrating them into a KB ii) Improving expressiveness of current link prediction models developed on an observed relational graph. For the first paradigm, we propose 1) A novel global Named Entity Disambiguation (NED) model that formulates NED as a structure prediction problem in a limited discrepancy search framework. This model starts searching from a complete solution constructed by a local model and conducts a search in the space of few possible corrections leading to improved assignments from a global point of view. 2) A novel local NED model based on a state-of-the-art language model to efficiently leverage contextual cues for disambiguation. 3) A novel approach for relation extraction model with distant supervision to predicts relation types based on right reasons leading to improved transparency and robustness. 4) A novel mechanism for unsupervised relation learning. 5) Improved fusion model for integrating relations extracted from text into observed KB. For the second paradigm, we propose a novel approach to enforce logical rules into state-of-the-art link prediction models that is important for the transparency of such models.
License
Resource Type
Date Issued
Degree Level
Degree Name
Degree Field
Degree Grantor
Commencement Year
Advisor
Committee Member
Academic Affiliation
Rights Statement
Publisher
Peer Reviewed
Language

Relationships

Parents:

This work has no parents.

In Collection:

Items