Investigating Examination Behavior of Image Search Users

摘要

Image search engines show results di‚erently from general Webrnsearch engines in three key ways: (1) mostWeb-based image searchrnengines adopt the two-dimensional result placement instead of thernlinear result list; (2) image searches show snapshots instead of snippetsrn(query-dependent abstracts of landing pages) on search enginernresult pages (SERPs); and (3) pagination is usually not (explicitly)rnsupported on image search SERPs, and users can view results withoutrnhaving to click on the “next page” buŠon. Compared with thernextensive study of user behavior in general Web search scenarios,rnthere exists no thorough investigation how the di‚erent interactionrnmechanism of image search engines a‚ects users’ examinationrnbehavior. To shed light on this research question, we conducted anrneye-tracking study to investigate users’ examination behavior inrnimage searches. We focus on the impacts of factors in examinationrnincluding position, visual saliency, edge density, the existence ofrntextual information, and human faces in result images. Œree interestingrn€ndings indicate users’ behavior biases: (1) instead of therntraditional “Golden Triangle” phenomena in the user examinationrnpaŠerns of general Web search, we observe a middle-position bias,rn(2) besides the position factor, the content of image results (e.g.,rnvisual saliency) a‚ects examination behavior, and (3) some popularrnbehavior assumptions in general Web search (e.g., examination hypothesis)rndo not hold in image search scenarios. We predict users’rnexamination behavior with di‚erent impact factors. Results showrnthat combining position and visual content features can improvernprediction in image searches.