Zhang, Xiaozhong
(2023)
Leveraging Interactive User Feedback for Personalized Data Visualization Recommendation.
Doctoral Dissertation, University of Pittsburgh.
(Unpublished)
Abstract
Data visualization is a powerful tool to communicate information clearly and efficiently, and is widely used in data analysis to support data-driven decision making. However, selecting a good visualization (i.e., view) that is both usable and useful is not a trivial task, because the search space of visualization configurations (e.g., variables to visualize, data transformations, visual encodings) is prohibitively large. In order to aid the user, a variety of utility functions have been proposed to estimate the utility of the views, so as to provide view recommendations, such as usability and usefulness. Different utility functions assess different aspects of a view’s utility and can also be combined to form multi-objective utility functions.
Traditional view recommendation systems typically do not consider user preferences of utility functions or user expectation of view values, leading to further suboptimal recommendation. A predefined utility function is not likely to fit all users’ needs at all times. User expectation also plays a role in the utility of a view. For example, if a peculiar pattern is expected by the user, then they will not find it very interesting in a recommended view. In order to support effective view recommendation for a given analysis, in this thesis, we propose a new paradigm, called Interactive View Recommendation (IVR), in which the system iteratively interacts with the user to elicit user preference information and expectations to formulate the most suitable utility function. IVR has two main objectives: high recommendation quality and low user interaction effort. However, these two objectives are usually in conflict, making the IVR problem even more challenging than traditional view recommendations.
As a proof-of-concept, we propose ViewSeeker, an active learning-based IVR framework and prototype system, which intelligently selects informative questions for user feedback to achieve a satisfactory recommendation quality while minimizing user interaction effort. ViewSeeker implements three main functionalities: (1) utility function preference learning, (2) user expectation learning, and (3) external knowledge base utilization. Simulated and real user studies were conducted to evaluate ViewSeeker’s effectiveness in leveraging interactive user feedback for personalized view recommendation, showing that with acceptable user effort, ViewSeeker can outperform current non-IVR recommenders.
Share
| Citation/Export: |
|
| Social Networking: |
|
Details
| Item Type: |
University of Pittsburgh ETD
|
| Status: |
Unpublished |
| Creators/Authors: |
|
| ETD Committee: |
|
| Date: |
6 September 2023 |
| Date Type: |
Publication |
| Defense Date: |
15 December 2022 |
| Approval Date: |
6 September 2023 |
| Submission Date: |
8 July 2023 |
| Access Restriction: |
No restriction; Release the ETD for access worldwide immediately. |
| Number of Pages: |
104 |
| Institution: |
University of Pittsburgh |
| Schools and Programs: |
Dietrich School of Arts and Sciences > Computer Science |
| Degree: |
PhD - Doctor of Philosophy |
| Thesis Type: |
Doctoral Dissertation |
| Refereed: |
Yes |
| Uncontrolled Keywords: |
Data Exploration, Data Visualization, Personalized Recommendation, SQL View, View Utility Function, Active Learning, User Expectation Learning, Knowledge Base Utilization |
| Date Deposited: |
07 Sep 2023 01:36 |
| Last Modified: |
07 Sep 2023 01:36 |
| URI: |
http://d-scholarship.pitt.edu/id/eprint/45061 |
Metrics
Monthly Views for the past 3 years
Plum Analytics
Actions (login required)
 |
View Item |
![[feed]](/images/feed-icon-32x32.png){kind=icon}