Sampath Jayarathna

Unified Relevance Feedback for Multi-Application User Interest Modeling

A user often interacts with multiple applications while working on a task. User models can be developed individually at each of the individual applications, but there is no easy way to come up with a more complete user model based on the distributed activity of the user. To address this issue, this research studies the importance of combining various implicit and explicit relevance feedback indicators in a multi-application environment. It allows different applications used for different purposes by the user to contribute user activity and its context to mutually support users with unified relevance feedback. Using the data collected by the web browser, Microsoft Word and Microsoft PowerPoint, combinations of implicit relevance feedback with semi-explicit relevance feedback were analyzed and compared with explicit user ratings. Our results are two-fold: first we demonstrate the aggregation of implicit and semi-explicit user interest data across multiple everyday applications using our Interest Profile Manager (IPM) framework. Second, our experimental results show that incorporating implicit feedback with semi-explicit feedback for page-level user interest estimation resulted in a significant improvement over the content-based models.

Rationale and Architecture for Incorporating Human Oculomotor Plant Features in User Interest Modeling

We present a conceptual framework expanding the use of eye movement as a source of implicit relevance feedback. While gaze time has been the primary feature to be incorporated in interest modeling, this work constructs a model of human oculomotor plant features during users interaction with multiple everyday applications with the goal of better interpreting user gaze data. The following presents the anatomical reasoning behind incorporating additional gaze features, the integration of the additional features into an existing interest modeling architecture, and a plan for assessing the impact of the addition of the features.

Grading Degradation in an Institutionally Managed Repository

It is not unusual for digital collections to degrade and suffer from problems associated with unexpected change. In an analysis of the ACM conference list, we found that categorizing the degree of change affecting a digital collection over time is a difficult task. More specifically, we found that categorizing this degree of change is not a binary problem where documents are either unchanged or they have changed so dramatically that they do not fit within the scope of the collection. It is, in part, a characterization of the intent of the change. In this work, we examine and categorize the various degrees of change that digital documents endure within the boundaries of an institutionally managed repository.

Analyzing the Perceptions of Change in a Distributed Collection of Web Documents

It is not unusual for documents on the Web to degrade and suffer from problems associated with unexpected change. In an analysis of the Association for Computing Machinery conference list, we found that categorizing the degree of change affecting digital documents over time is a difficult task. More specifically, we found that categorizing this degree of change is not a binary problem where documents are either unchanged or they have changed so dramatically that they do not fit within the scope of the collection. It is in part, a characterization of the intent of the change. In this paper, we present a case study that compares change detection methods based on machine learning algorithms against the assessment made by human subjects in a user study. Consequently, this paper will focus on two research questions. First, how can we categorize the various degrees of change that documents endure? And second, how did our automatic detection methods fare against the human assessment of change in the ACM conference list?

IPM-G: Enabling Collaborative Filtering Using Multi-Application Interest Models

He Interest Profile Manager (IPM) plays the central role in inferring user interest during document triage. The IPM collects information about interest-related activity from the potentially many triage applications. In this paper, we extend the IPM framework to enable community-based navigation using inferred user interests from information gathering tasks involving the use of multiple applications. We call IPM running on server, Global IPM (IPM-G), and IPM-G can generate similarity assessments, and thus recommendations, based on three different levels: tasks, documents, and annotations. As a result, CF methods can be applied to each level to get results at these three levels of granularity. By representing inferred interests based on the features of their tasks, documents, and annotations, we make possible six potential collaborative filtering (CF) modes in the IPM-G. This paper describes why collaborative filtering based on multi-application interest models is important, abstractly describes the representation of the interest models, and presents details of one of these filtering modes.