A Corpus of Realistic Known-Item Topics with Associated Web Pages in the ClueWeb09

摘要

Known-item finding is the task of finding a previously seen item. Such items may range from visited websites to received emails but also read books or seen movies. Most of the research done on known-item finding focuses on web or email retrieval and is done on proprietary corpora not publically available. Public corpora usually are rather artificial as they contain automatically generated known-item queries or queries formulated by humans actually seeing the known-item. In this paper, we study original known-item information needs mined from questions at the popular Yahoo! Answers Q&A service. By carefully sampling only questions with a related known-item web page in the ClueWeb09 corpus, we provide an environment for repeatable realistic studies of known-item information needs and how a retrieval system could react. In particular, our own study sheds some first light on false memories within the known-item questions articulated by the users. Our main finding shows that false memories often relate to mixed up names. This indicates that search engines not retrieving any result on a known-item query could try to avoid returning a zero-result list by ignoring or replacing names in respective query situations. Our publically available corpus of 2,755 known-item questions mapped to web pages in the ClueWeb09 includes 240 questions with annotated and corrected false memories.