Publikation
Attention-based information retrieval - using eye tracking to capture attention evidence and personalize search
Georg Buscher
Dr. Hut Verlag, 8/2010.
Zusammenfassung
In the modern world, we face an immense information overload. Search engines overwhelm us with masses of more or less relevant results and in order to handle this overload, we focus our attention to these bits and pieces that are important or interesting to us. The key idea for this thesis is to record indications for the user's attention while searching and working with documents, and to utilize this information for personalizing search. Information about the user's visual attention is collected by an eye tracker, which allows tracking the user's eye movements and point of gaze on a computer monitor. For recording attention data, the focus lies particularly on identifying reading behavior. An attention-enhancedWiki system is introduced that automatically generates and stores attention evidence data about what parts of documents are read, skimmed, or skipped by the user. In two eye tracking user studies, eye movement patterns are more closely examined and relationships between several eye movement measures and individually perceived relevance are identified. The results show that some newly invented measures are strongly correlated with individually perceived relevance. These measures can form the basis for automatic relevance detection. With regard to evaluating the effectiveness of attention feedback on personalizing search, the results from two eye tracking user studies are reported. The results show that applying attention feedback leads to considerable improvements concerning the quality of the returned results from a search engine. Furthermore, prototypes and concepts are introduced demonstrating the potential of attention feedback for different kinds of applications. For example, desktop search scenarios, implicit knowledge transfer, and automatic document classifier learning are discussed. Finally, an outlook on attentive documents that keep track of what and how the user is reading concludes the thesis.