Supporting Information Seeking and Sensemaking in Issue-Based Knowledge Crystallization

Open Access
Author:
Sun, Feng
Graduate Program:
Information Sciences and Technology
Degree:
Doctor of Philosophy
Document Type:
Dissertation
Date of Defense:
November 16, 2018
Committee Members:
  • Guoray Cai, Dissertation Advisor
  • Guoray Cai, Committee Chair
  • Mary Beth Rosson, Committee Member
  • Xiaolong (Luke) Zhang, Committee Member
  • Alan M. MacEachren, Outside Member
Keywords:
  • Information Seeking
  • Topic Modeling
  • Civic Engagement
  • Design-based Research
  • Information Visualization
  • Text Analysis
  • Human Computer Interaction
Abstract:
People often face the challenge of having to sift through a large volume of data and make sense of them in order to achieve actionable insights on complex issues. We formalize this challenge as a class of problem, Issue-based Knowledge Crystallization (IBKC), and address this problem in the context of local civic engagement. This dissertation focuses on understanding the information seeking behavior of IBKC and identifying design issues for supporting effective knowledge crystallization. Towards these goals, we propose a conceptual framework that characterizes the information seeking processes and associated information and cognitive barriers in IBKC. It was informed by prior theories and models, and by observing practices of IBKC in Community Issue Review (CIR). We found that effectively extracting information nuggets in IBKC requires collaborative information foraging and sense-making where workspace awareness and activity awareness are important for judging the completeness of nugget extraction task. Based on such insight, we employed a design-based research approach to investigate the design issues of NuggetLens as a visual analytical method to support group information seeking in IBKC. The design of NuggetLens requires that we advance our understanding of two questions: (1) how do people judge completeness in nugget extraction tasks? (2) how should people be assisted in dealing with cold-start problems when social clues are lacking? The first question was answered by observing small-group nugget extraction experiments and semi-structured interviews with participants. The second question was partially answered by generating an initial information landscape through an interactive topic modeling method. Using a variety of datasets, we demonstrated that interactive topic modeling can enable non-expert users to quickly make sense of topics and generate good summaries. The resulting topics, serving as information scent, can help users better understand document space structurally, providing a solid foundation for nugget extraction. This work contributes to the science and solutions of information seeking by presenting new research frontiers in IBKC problems. Our findings have significant implications in enabling informed citizens engagement in public decision-making.