Evaluating Generative Ad Hoc Information Retrieval
This work addresses the need for reliable evaluation methods in generative ad hoc information retrieval, which is incremental as it builds on existing IR and NLP methodologies.
The paper tackles the problem of evaluating generative retrieval systems, which produce direct text responses instead of document rankings, by surveying literature, identifying tasks and architectures, and developing a new user model for operationalization.
Recent advances in large language models have enabled the development of viable generative retrieval systems. Instead of a traditional document ranking, generative retrieval systems often directly return a grounded generated text as a response to a query. Quantifying the utility of the textual responses is essential for appropriately evaluating such generative ad hoc retrieval. Yet, the established evaluation methodology for ranking-based ad hoc retrieval is not suited for the reliable and reproducible evaluation of generated responses. To lay a foundation for developing new evaluation methods for generative retrieval systems, we survey the relevant literature from the fields of information retrieval and natural language processing, identify search tasks and system architectures in generative retrieval, develop a new user model, and study its operationalization.