Footprints: A Visual Search Tool that Supports Discovery and Coverage Tracking

Abstract

Searching a large document collection to learn about a broad subject involves the iterative process of figuring out what to ask, filtering the results, identifying useful documents, and deciding when one has covered enough material to stop searching. We are calling this activity "discoverage," discovery of relevant material and tracking coverage of that material. We built a visual analytic tool called Footprints that uses multiple coordinated visualizations to help users navigate through the discoverage process. To support discovery, Footprints displays topics extracted from documents that provide an overview of the search space and are used to construct searches visuospatially. Footprints allows users to triage their search results by assigning a status to each document (To Read, Read, Useful), and those status markings are shown on interactive histograms depicting the user's coverage through the documents across dates, sources, and topics. Coverage histograms help users notice biases in their search and fill any gaps in their analytic process. To create Footprints, we used a highly iterative, user-centered approach in which we conducted many evaluations during both the design and implementation stages and continually modified the design in response to feedback.

Bridging Text Visualization and Mining: A Task-Driven Survey

“…[25], [27], [35], [39], [42], [67], [75], [80], [86], [96], [116], [125], [127], [142], [144], [154], [163], [177], [183], [187], [208], [212], [220], [233], [242], [253], [258], [264], [365] word cloud (56) [5], [22], [26], [37], [53], [54], [56], [64], [81], [87], [96], [128], [147], [157], [161], [196], [173], [186], [189], [192], [203], [217...…”

Ontology-assisted provenance visualization for supporting enterprise search of engineering and business files

“…The similarity measurement used in this work is based on the vector space model. Footprints [33] was developed for analysts to help them search documents and articles.…”

Text Visualization and Close Reading for Journalism with Storifier

“…Providing top-down awareness on a corpus level (i.e., distant reading techniques) enables not only to inform corpus trends and high-level patterns but also to provide entry points for further exploration [3,22]. Common approaches include topics [3,12,14,22,26], extracted entities [14,22,33,38], relevant keywords [13,15,27], and aggregate statistics over terms and metadata [15], sometimes organized over time [13,14,27]. These elements are often interactive and serve as content filters, linking back to particular text mentions.…”

“…Reading is time-intensive, and having the means to efficiently and effectively search for relevant content is key. Past assessments found that journalists [5] and intelligence analysts [22] prefer "high recall" queries, i.e., starting with a broad search that retrieves a larger set of documents for quick browsing and later re-filtering. This implies that filters should be compoundable, allowing for multiple filtering criteria at once.…”

“…This implies that filters should be compoundable, allowing for multiple filtering criteria at once. This is often achieved via metadata filters [3,5,13,15,16] and entity/topic filters [12,14,22,27]. A less common but powerful approach to complement keyword search is to use ontology concepts [11,34], where a higher level hypernym may replace multiple keyword searches of alternative spellings of the same concept.…”