On Data Engineering and Knowledge Graphs: a Context-Aware Proposal for Web-Scale Knowledge Graph Completion
Year: 2023
Type: dissertation
Abstract: Nowadays, Knowledge Graphs are a widely used means to store structured \ninformation for a variety of different domains and applications. However, \ndue to the fact that they are usually constructed using automated information \nextraction techniques, they are often incomplete, either because these techniques failed \nto extract the relevant information, or because it was not present altogether in the \noriginal sources. \nThe problem that we address in this dissertation is how to find this missing \nknowledge and complete Knowledge Graphs in an automatic manner. In the literature, \nthere are already many proposals to perform this task. However, they have important \ndrawbacks, namely: they rely on embedded representations, which are \ncomputationally expensive to generate and demand frequent regenerations, they \nrequire human intervention or human-provided data, they rely on external sources of \ninformation, they cannot produce new knowledge on their own, or they do not scale \nproperly to very large Knowledge Graphs. \nIn this dissertation, we present a new automated proposal for completing \nKnowledge Graphs that does not suffer from any of the previous drawbacks. Our \ncontribution is threefold: CHAI, a technique for automatically generating tractable sets \nof candidate triples; CAFE, a high-accuracy triple classification proposal; and \nSciCheck, a technique specifically tailored for completing scientific Knowledge Graphs. \nOur theoretical and practical validation suggests that our proposal is very efficient and \neffective in practice, and that it is able to successfully complete Knowledge Graphs of \nvarying natures.
Author: Borrego Díaz, Agustín
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7036944320