User's Interest Detection through Eye Tracking for Related Documents Retrieval

In multiple studies, various implicit and explicit feedback types were used to quantify document quality and to identify user's interest in document content on both document and sub-document level. In this study we use eye movement data retrieved while reading a document to identify sections, user is most interested in. We describe multiple patterns in eye movement of users reading documents that can be used to identify important document fragments. The identified patterns are used to determine the user's interest in text fragments and to compare effectiveness of extracted document fragments and user created in-text highlights in the task of related document retrieval. Identified eye movement patterns allowed us to extract more important document sections compared to manual annotations with greater precision and consistency. Both types of compared interest indicators achieved similar performance in related document retrieval with the single most important difference: gaze based important section identification do not require active participation of the reader.