Conceptions of Features and Semantic Clusters as Search Mechanisms: A Pilot Study 1
Barbara M. Wildemuth, Meng Yang, Gary Geisler, Tom Tolleson, Jonathan L. Elsas, Jei Luo, Gary Marchionini · 2004
Search mechanisms are the interface tools that people can use to search the items in a collection; a common example is a text box for entering search terms. In this pilot study, the Open Video team investigated two mechanisms for providing access to a collection of news videos. The first provided direct access to the subsets of the collection containing particular video features, as identified by other TREC VID participants; the second provided direct access to the subsets of the collection that were identified as semantic clusters, using latent semantic indexing approaches to analyzing the video transcripts. Each access mechanism was depicted in the interface as a series of labeled checkboxes. In the pilot study, four team members completed the TREC VID topic searches and responded to measures of their perceptions of the experience of using each search mechanism. Reasonably high precision was achieved on the user searches across all three systems (0.67-0.74), but none of the systems achieved high recall (0.10-0.11). Mean average precision across three runs (as calculated by NIST) ranged from 0.06-0.09. Searches were completed in approximately 8 minutes across all three systems. User satisfaction with the two experimental systems was mixed. Lessons learned from conducting this pilot study will contribute to the design of a follow-up study investigating the ways in which users of digital video retrieval systems conceptualize search mechanisms that incorporate access to subsets of the collection based on video features or semantic clusters of transcript content. 1