Varun Ratnakar

Trusting Information Sources One Citizen at a Time

This paper describes an approach to derive assessments about information sources based on individual feedback about the sources. We describe TRELLIS, a system that helps users annotate their analysis of alternative information sources that can be contradictory and incomplete. As the user makes a decision on which sources to dismiss and which to believe in making a final decision, TRELLIS captures the derivation of the decision in a semantic markup. TRELLIS then uses these annotations to derive an assessment of the source based on the annotations of many individuals. Our work builds on the Semantic Web and presents a tool that helps users create annotations that are in a mix of formal and human language, and exploits the formal representations to derive measures of trust in the content of Web resources and their original source.

TRELLIS: An Interactive Tool for Capturing Information Analysis and Decision Making

TRELLIS provides an interactive environment that allows users to add their observations, opinions, and conclusions as they analyze information by making semantic annotations about on-line documents. TRELLIS includes a vocabulary and markup language for semantic annotations of decisions and tradeoffs, and allows users to extend this vocabulary with domain specific terms or constructs that are useful to their particular task. To date, we have used TRELLIS with a variety of scenarios to annotate tradeoffs and decisions (e.g., military planning), organize materials (e.g., search results), analyze disagreements and controversies on a topic (e.g., intelligence analysis), and handle incomplete and conflicting information (e.g., genealogy research).

Towards intelligent assistance for to-do lists

Assisting users with to-do lists presents new challenges for intelligent user interfaces. This paper presents a detailed analysis of to-do list entries jotted by users of a system that automates tasks for users that we would like to extend to assist users with their to-do entries. We also present four distinct stages of interpretation of to-do entries that can be accomplished and evaluated separately. A system that has good performance in any of these four stages can provide intelligent assistance that is useful to users.

Workflow matching using semantic metadata

Workflows are becoming an increasingly more common paradigm to manage scientific analyses. As workflow repositories start to emerge, workflow retrieval and discovery becomes a challenge. Studies have shown that scientists wish to discover workflows given properties of workflow data inputs, intermediate data products, and data results. However, workflows typically lack this information when contributed to a repository. Our work addresses this issue by augmenting workflow descriptions with constraints derived from properties about the workflow components used to process data as well as the data itself. An important feature of our approach is that it assumes that component and data properties are obtained from catalogs that are external to the workflow system, consistent with current architectures for computational science.

Metadata catalogs with semantic representations

Metadata catalogs store descriptive information about logical data items. These catalogs can then be queried to retrieve the particular logical data item that matches the criteria. However, the query has to be formulated in terms of the metadata attributes defined for the catalog. Our work explores the concept of virtual metadata, where catalogs can be queried using metadata attributes not originally defined in the catalog. We use semantic web standards, where new metadata attributes can be taken from shared ontologies and can include expressive axioms to define the new terms. We have implemented a virtual metadata catalog as an extension of the Metadata Catalog Service (MCS), using the Web Ontology Language (OWL) and a reasoning engine to map queries of temporal nature in several metadata catalogs.

IKRAFT: Interactive Knowledge Representation and Acquisition from Text

We propose a new approach to develop knowledge bases that captures at different levels of formality and specificity how each piece of knowledge in the system was derived from original sources, which are often Web sources. If a knowledge base contains a trace of information about how each piece of knowledge was defined, it will be easier to reuse, extend, and translate the contents of the knowledge base. We are investigating these issues with IKRAFT, an interactive tool to elicit from users the rationale for choices and decisions as they analyze information used in building a knowledge base. Starting from raw information sources, most of them originating on the Web, users are able to specify connections between selected portions of those sources. These connections are initially very high level and informal, and our ultimate goal is to develop a system that will help users to formalize them further.