Recognizing user interest and document value from reading and organizing activities in document triage

Abstract

People frequently must sort through and identify relevant materials from a large set of documents, such as looking through the results of a web search. During this process of document triage there is reading and organizing activity. Moreover, these tasks can occur in different applications. A user's interests can be identified from reading and organizing activity and used as a basis for providing cues to other potential documents of interest in the set. To most effectively identify related documents of interest, activity data must be collected from all applications used in document triage. In this paper we present a common framework (the Interest Profile Manager) for collecting and analyzing user interest. We also present models for identifying user interest based only on reading activity, only on organizing activity, and models incorporating both reading and organizing activity. A study comparing document values calculated using the different models shows that incorporating interest information from both reading and organizing activity more accurately estimates users' valuation of documents than using either type of activity alone.

An Empirical Study of User Navigation during Document Triage

“…The current research into triage is fragmented, and this no doubt explains the lack of such a model. Available reports focus on very specific aspects or external aspects rather than document triage generally [1,2,3,5,18].…”

Supporting document triage via annotation‐based visualizations

“…Also, real world activity is not limited to the confines of a single application. Recognizing this, we found that models using a combination of the implicit information from multiple applications work better than relying on the information obtained from a single application in document triage [1].…”

“…Our prior study of document triage practice [1] showed that data that describes users' activities while they read, for example the time they spend on a page and how much they scroll, in conjunction with document attributes, such as document length or the number of embedded hypertext links, can be a meaningful source of evidence for inferring user interest. Thus the second major role of WebAnnotate is to collect this type of data on users' interactions with Web pages (such as scrolling, mouse clicks, and changes in focus) and the corresponding data about the characteristics of the web pages themselves (page length, number embedded links, and number of images).…”

“…Thus, users often interact with multiple applications during triage: for example, a Web browser or information workspace that displays an overview of search results; a reader for viewing the content of individual documents; and possibly a third tool to keep track of what has been read and for recording brief notes or assessments. To better support document selection and skimming, we have begun inferring user interest based on user interaction with multiple triage-related applications [1]. Our past research shows that combining evidence from an information workspace and a document reader improves the overall quality of the user interest model.…”

Statistical user model supported by R-Tree structure

“…In first place, the content-based approach [6] predicts user behavior through the past behavior based on Information Retrieval methods such as survey techniques or implicit methods to obtain and store profiles about user preferences and likes [5]. Some problems have been detected: the filtering and partial content analysis [46], the over-specialization [47,54], and the new user problem [52].…”