Information Search with LLMs

Human-centered exploration of how LLMs can augment human information-seeking behaviors.

In this project we propose an exploratory study in which we will investigate the integration of LLMs into the human information search and retrieval process. We will explore how LLMs can assist at various stages of information retrieval - from initial query formulation to the final evaluation of results. Utilizing an iterative design methodology, the study will develop and test two alternative information search interfaces that incorporate LLMs, evaluating their effectiveness in enhancing query refinement, generating contextually relevant queries, providing summarization and clarification of search results from diverse sources. The interfaces will allow for dynamic interaction between users and LLM(s), adapting to user feedback and search behavior. One interface will be based on sequential interaction approach, while the other multi-pane interaction approach. The first interface will be simpler and based on a turn-taking model where the user and the LLM interact in a sequential manner. The second interface will be more complex with multiple panes, each corresponding to different stages of the information retrieval process, allowing simultaneous access to various types of information and tools. Interface evaluation will use interaction logs, eye-tracking, retrospective think aloud and interviews.

We expect this study to contribute to the understanding of how AI can augment human information-seeking behaviors, with the potential to significantly optimize the efficiency and efficacy of information retrieval in online environments.