Year: 2023
Type: dissertation
Abstract: The internet is a large-scale web of interconnected pages, which are ephemeral by nature. Most web pages have a short life span, which leads to a loss of data as they die out. To fight this knowledge gap, web archives were created, which occasionally crawl the internet to store web pages. Because of the sheer size of the Internet as well as their longitudinal nature, web archives are some of the largest agglomerations of human social knowledge. ArquivoPT is the largest Portuguese web archive, and easily the largest collection of unsorted Portuguese text. Its searching mechanisms focus on full-text search, domain and URL filtering, and time ranges. These methods scale well but fail to capture the implicit semantic connections across pages. Web pages, particularly news pages, are connected through mentioning the same entities, directly or indirectly, or through referring to the same or preceding/succeeding events. These relations between pages form an implicit graph structure that goes unexplored with conventional methods, but that theoretically allows for powerful exploration of knowledge in regards to entities and their relations. We hypothesize that by making these relations explicit through a knowledge graph, we can perform semantic-based, relation-oriented search and question answering over a news set. By using a natural language processing pipeline to extract entities and relations from Portuguese text, the sources of these connections are made apparent and, in combination with a well-defined model, structured into an RDF graph. This graph is later queried and used with specific algorithms which fulfill specific information-seeking use cases. We thus contribute with a framework that generalizes to other scenarios, archives, and even domains. It is accompanied by an evaluation methodology and dataset, to assess the quality of these knowledge bases. Because news are not self-contained and lack common facts, we also propose several ways to incorporate these into an entity-relation graph, when it is linked to external knowledge bases.
Author: Castanho, Daniel João Nunes
Language: English
Topic: Semantic Web and Ontologies
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7113138087