Learning non-verbal relations under open information extraction paradigm

• Ano de defesa: 2014
• Autor(a) principal: Xavier, Clarissa Castellã
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Pontifícia Universidade Católica do Rio Grande do Sul; Faculdade de Informáca; BR; PUCRS; Programa de Pós-Graduação em Ciência da Computação
• Palavras-chave em Português: INFORMÁTICA; PROCESSAMENTO DA LINGUAGEM NATURAL; CNPQ::CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO
• Link de acesso: http://tede2.pucrs.br/tede2/handle/tede/5275

The Open Information Extraction (Open IE) is a relation extraction paradigm in which the target relationships cannot be specified in advance, and it aims to overcome the limitations imposed by traditional IE methods, such as domain-dependence and scalability. In order to extend Open IE to extract relationships that are not expressed by verbs from texts in English, we introduce CompIE, a component that learns relations expressed in noun compounds (NCs), such as (oil, extracted from, olive) from olive oil, or in adjectivenoun pairs (ANs), such as (moon, that is, gorgeous) from gorgeous moon. CompIE input is a text file, and the output is a set of triples describing binary relationships. The architecture comprises two main tasks: NCs and ANs Extraction (1) and NCs and ANs Interpretation (2). The first task generates a list of NCs and ANs from the input corpus. The second task performs the interpretation of NCs and ANs and generates the tuples that describe the relations extracted from the corpus. In order to study CompIE s feasibility, we perform an evaluation based on hypotheses. In order to implement the strategies to validate each hypothesis we have built a prototype. The results show that our solution achieves 89% Precision and demonstrate that CompIE reaches its goal of extending Open IE paradigm extracting relationships within NCs and ANs.