Image search moves on the Web: An exploratory study
Abby Goodrum · Proceedings of the American Society for Information Science and Technology · 2002
Image seeking behavior is a complex interaction among many factors. One aspect of this behavior, are the search moves that users make when seeking images on the Web. Most of the research examining search moves has focused primarily on users seeking textual materials from databases (Bates, 1981; Chapman, 1981; Wildemuth, He, & Friedman 1992; Fidel, 1984, Xie, 2002). Fidel and Efthimiadis (2000) have extended this work by examining the effect of task on image searching behavior in an image database system. Jose, Furner & Harper (1998) have also investigated image retrieval using a simulated work task situation. Research examining image searching in the unstructured environment of the Web has focused primarily on term-level query analysis but has not examined patterns of search moves. (Goodrum & Spink, 2001). The current study examines the moves searchers make when seeking images on the Web with simulated tasks that are as close as possible to those they might have in a real situation, including a described role and context. This study is part of a multi-year research project examining image searching and image choice behavior. The specific focus is an examination of modifications made to search strategies in order to gain precision and on the search shifts users make between text and images. The purpose of the analysis is to: (1) drive the coding of image search moves, (2) describe a model of image searching on the Web and (3) develop transition matrices to examine image search patterns. This poster will present the theoretical framework for studying image search moves, the current state of the project, and results to date. 18 students in a graduate course on Visual Information Retrieval received 3 hours of training about how search engines on the web find images. Included in this training were: Google, Altavista, PictureQuest, WebSeek, and HotBot as well as several large scale image collections available for searching (i.e., American Memory, Corbis, and Associated Press). Because they had only 3 hours of training and no prior experience with image searching, these users were considered novices. Students were given one week to conduct searches for five images to illustrate a lecture on a selected topic. Prior to searching, students were asked to describe their image needs and to describe the kinds of images they hoped to find. For each image sought, the students logged their starting point, subsequent moves, queries, decision points for browsing, relevance criteria, and the amount of time spent. Students provided print-outs of the images selected, and provided a written post search reflection on their experience. 202 queries for 90 images are in the process of being examined along with accompanying search logs, pre-search questionnaire and post search essay. Preliminary data show that on average, students spent 20 minutes per image, and changed their initial queries frequently. Eighty-three percent of the initial queries were revised and resubmitted in some way before an image was selected. Students selected 16 search sites including major search engines and specialized sites such as PictureQuest, AP Archives, and Library of Congress Prints and Photos as beginning points for their searches, but most students used at least 2 different search sites in the course of looking for each image. Currently, the search logs are being examined and the moves are being coded. New categories will be defined as needed and state transition matrices developed.