Dense and hybrid models for information retrieval

• Autor(a) principal: Frias, José André Lopes
• Data de Publicação: 2022
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10773/38434
• topic: Information retrieval; Dense retrieval; Dimensionality reduction; Deep learning

As in the era of Big Data, there is the need of finding information in an easy and fast way, being imperative for a search system to understand more efficiently the user intent. Dense Retrieval focuses on this idea, by allowing the models to capture the underlying meaning of the queries and documents. Current models already surpass the classical BM-25 model in terms of accuracy. However, due to the use of a high number of dimensions to create representations of the queries and documents, the dense models are still not optimized in terms of their efficiency at a search level. This work focuses on evaluating the need for that high number of dimensions, by analyzing different dimensionality reduction methods, trained for different purposes, and comparing the trade-offs between efficiency and accuracy.