Abstract
The task of Information Retrieval (IR) is the foundation of question-answering and information search systems. The complexity of these retrieval systems evolved to enable an effective search of a variety of documents based on users’ information needs. However, the quality of the retrieval results highly depends on the question phrasing and structure heavily influenced by the user’s age, expertise, background, and aim. This paper aims to close the gap between user-specific query structures and the content of various documents by employing the approaches Hypothetical Document Embeddings (HyDE) and Hypothetical Query Embeddings (HyQE). Both approaches achieve promising results in terms of effectiveness and robustness. An evaluation indicates that HyDE and HyQE outperform traditional retrieval methods such as BM25 and plain embedding methods.
Recommended Citation
Jostmann, Marten and Winkelmann, Hendrik, "Evaluation of Hypothetical Document and Query Embeddings for Information Retrieval Enhancements in the Context of Diverse User Queries" (2024). Wirtschaftsinformatik 2024 Proceedings. 115.
https://aisel.aisnet.org/wi2024/115