Exploiting entities for query expansion

• Ano de defesa: 2013
• Autor(a) principal: Wladmir Cardoso Brandao
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Universidade Federal de Minas Gerais
• Palavras-chave em Português: Aprendizado por computador; Computação; Sistemas de recuperação de informação; Expansão de consultas; Aprendizagem para ranqueamento; Wikipédia; Feedback de relevância; Reconhecimento de entidades
• Link de acesso: https://hdl.handle.net/1843/ESBF-9GMJW2

A substantial fraction of web search queries contain references to entities, such as persons, organizations, and locations. In this work, we propose entity-oriented query expansion approaches that exploit semantic sources of evidence devising discriminative term features and machine learning techniques that effectively combines these features to rank candidate expansion terms. Particularly, our unsupervised approach (UQEE) uses taxonomic features devised by the semantic structure implicitly provided by infobox templates, while our learning to rank approach (L2EE) considers semantic evidence encoded in the content of Wikipedia article fields to automatically labels training examples proportionally to their observed retrieval effectiveness. Lastly, we propose a self-supervised approach to autonomously generate infoboxes for Wikipedia articles (WAVE). Experiments attest the effectiveness of our approaches, with significantly gains compared to state-of-the-art PRF and ePRF approaches.