Supporting claim-making in democratic deliberation through an information recommendation system
Restricted (Penn State Only)
- Author:
- Tian, Ye
- Graduate Program:
- Informatics
- Degree:
- Doctor of Philosophy
- Document Type:
- Dissertation
- Date of Defense:
- July 31, 2020
- Committee Members:
- Andrea Tapia, Major Field Member
Jack Carroll, Major Field Member
Guoray Cai, Chair & Dissertation Advisor
John Gastil, Outside Unit & Field Member - Keywords:
- Information overload
recommendation systems
claim-making
democratic deliberation - Abstract:
- Deliberative decision analysis demonstrates great potential in fostering trustworthy and unbiased knowledge to inform the general electorate. Its key component involves building well-reasoned judgments and insights (called claim-making) on the public matter through rigorous deliberation. This task, however, is daunted by the severe and multi-faceted information overload problem. This dissertation aims to develop new knowledge on the design of recommendation systems (RS) to alleviate information overload in the claim-making task. To develop such knowledge, we aim to address two knowledge gaps: First, how do human deliberators manage information overload? Insights from this question provides the behavioral foundation and informs our investigation of the second issue: given the myriad of RS design patterns, strategies and techniques, which are the most effective ones in identifying the most relevant information for citizen deliberators? To address the first issue, we conduct a field study to observe claim-making sessions from citizen participants on local public issues. This study discovers that an effective RS should not only highlight topically relevant information items that serve as direct evidence for or against a claim, but also logically relevant ones that equip the user with the necessary background knowledge to effectively interpret information. For the second issue, we adopt the design science research methodology and conduct a series of design--evaluation--analysis iterations. This study shows that content-based recommendation is a viable approach, in which lexical profile-based item representation with standard TF-IDF weighting achieves an F-1 measure of 0.5; incorporating linguistic and domain knowledge systems enables the RS to further capture the logical relationship among information candidates and pushes the system's F-1 measure to 0.8. This research contributes to the field of information overload research by generating new insights on designing effective RS for decision analysis tasks. These insights help drive towards the grand goal of fostering informed public opinion via deliberative decision analysis practices.