This article compares traditional and large language models (LLM)-based search engines in the context of image geolocation. It studies user interactions and query formulation strategies. The study involved 60 participants, who used either traditional or LLM-based search engines to determine the location where an image was captured. The findings revealed that participants using traditional search engines were more accurate in predicting the image’s location compared to those using LLM-based search engines. Moreover, the query strategies differed based on the type of search engine used. Participants using LLM-based search engines issued longer, more natural language queries, but had shorter search sessions. On the other hand, those using traditional search engines added more terms to their initial queries when reformulating them.
Publication date: 19 Jan 2024
Project Page: https://doi.org/10.1145/3627508.3638305
Paper: https://arxiv.org/pdf/2401.10184
