Human computation, games with a purpose, image retrieval, image tagging, query analysis, user search behavior, ESP game
Library and Information Science
Human computation game-based image labeling has proven effective in collecting image descriptions for use in improving image searching, browsing, and accessibility. ESP and Phetch are two successful image description games: the ESP game is designed for collecting keyword descriptions while the Phetch game for sentence descriptions. In this paper we propose and implement an image retrieval game for collecting query-like descriptions. The idea is that a player is presented with a target image for a short time and then needs to construct a query based on his recall of the image, with the goal of making the target image ranked as high as possible by an image search engine. We have conducted preliminary analysis of the queries collected from players who were recruited through the Amazon Turk. Our result shows that the image retrieval game enabled us to collect queries that are comparable to the real queries reported in existing studies, and it can also be used to collect informative tags while avoiding undesired ones such as trivial color words.
Jun Wang and Bei Yu (2011) Labeling Images with Queries: A Recall-based Image Retrieval Game Approach. Proceedings of the SIGIR 2011 Workshop on Crowdsourcing for Information Retrieval.