User information needs, interactions, and behavior are specific for each domain. The goal of this project is to model information needs of users interacting with collaborative sources, in particular focusing on Yahoo! Answers and Wikipedia, to understand the factors that influence user behavior and information access patterns. By using this information we can improve user interfaces and interaction models to support (and influence) future user behavior.
Explicit feedback (votes, stars) in Answers, and implicit feedback (# of accesses, edits, chatter) in Wikipedia pages indicate user interest and engagement with the content. There is a non-trivial relationship between user engagement and accuracy, and we attempt to identify the factors that cause users to engage with content, and to compare and contrast with factual accuracy or quality of the content. We hope to identify patterns in the topics and the language of the content that makes it more (or less) attractive to users.
Relevant Publications:
- The Influence of Caption Features on Clickthrough Patterns in Web Search,
Charlie Clarke, Eugene Agichtein, Susan Dumais and Ryen W. White,
in the ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR), 2007
- Learning User Interaction Models for Predicting Web Search Result Preferences [ slides ],
Eugene Agichtein, Eric Brill, Susan T. Dumais, and Robert Ragno,
in the ACM SIGIR Conference on Research and Development on Information Retrieval (SIGIR), 2006
People:
- Eugene Agichtein (Emory)
- Gilad Mishne (Yahoo! Applied Research)
- Rosie Jones (Yahoo! Research)
- Looking for a graduate student to work on this project!
Analyze the effectiveness of Answers vs. Wikipedia vs. Web for different types of queries, topics, and desired type of answer. Initially, focus on the TREC factoid questions. Classify questions/queries as appropriate to each kind of source, and develop better techniques to infer user information needs.
Application:
- Route queries/questions to best source
- Improve automatic question answering by appropriately integrating/weighting evidence.
People:
- Eugene Agichtein (Emory)
- Yandong Liu (Ph.D. student, Emory)
Relevant Publications:
- Question Answering over Implicitly Structured Web Content,
Eugene Agichtein, Chris Burges, and Eric Brill,
to appear in the IEEE/WIC/ACM International Conference on Web Intelligence (WI), 2007- Discovering Authorities in Question Answer Communities Using Link Analysis (short paper),
Pawel Jurczyk and Eugene Agichtein,
to appear in the ACM Conference on Information and Knowledge Management (CIKM), 2007- Improving Web Search Ranking by Incorporating User Behavior Information [ slides ],
Eugene Agichtein, Eric Brill, and Susan T. Dumais,
in the ACM SIGIR Conference on Research and Development on Information Retrieval (SIGIR), 2006