Graduation Semester and Year
Fall 2025
Language
English
Document Type
Dissertation
Degree Name
Doctor of Philosophy in Computer Science
Department
Computer Science and Engineering
First Advisor
Chengkai Li
Abstract
Constructing structured queries over large and complex knowledge graphs remains a significant challenge, especially for users who are unfamiliar with the underlying schema or the graph’s data model. While prior research has explored areas such as query recommendation, graph completion, and prediction of user query intent, few solutions provide interactive, real-time guidance that can assist users during the query formulation process. Existing visual query builders often require substantial prior knowledge of the graph structure or rely solely on static templates, limiting their accessibility and effectiveness for non-expert users.
To address these limitations, we present Orion, an interactive visual query builder designed to guide users in constructing complex query graphs through intelligent, system-generated suggestions. Orion provides real-time recommendations at each step of the query construction process, suggesting candidate edges that are ranked according to their estimated relevance to the user’s intended query. The ranking is powered by Random Decision Paths (RDP), a novel probabilistic algorithm that integrates co-occurrence patterns between node and edge types, extracted not only from the data graph itself but also from external textual knowledge sources such as Wikipedia. By leveraging this combination of internal and external evidence, RDP is able to predict likely query edges even in scenarios where direct co-occurrence data is sparse.
Building upon RDP, we further introduce Hybrid, a variant that combines the predictive accuracy of RDP with the computational efficiency of a simpler, data-driven baseline method. This hybrid approach balances recommendation quality with responsiveness, making it suitable for real-time interaction in a visual query interface. We evaluated Orion through extensive simulation experiments as well as a controlled user study using the Freebase knowledge graph. The results demonstrate that both systems significantly improve query formulation performance, with users completing queries more accurately and efficiently compared to baseline approaches.
In particular, Hybrid achieved the highest overall success rate of 73.5%, outperforming existing methods in both objective metrics and subjective usability ratings. Participants reported that the system’s edge suggestions made query construction more intuitive, reduced cognitive load, and helped them discover relevant connections they might have otherwise overlooked. These findings suggest that interactive visual query builders like Orion can substantially lower the barrier to accessing complex knowledge graphs, supporting both novice and experienced users in formulating accurate, expressive queries. Our work highlights the importance of integrating probabilistic modeling, external textual knowledge, and real-time interaction in the design of effective graph query systems.
Keywords
Data Management Systems, Knowledge Graphs, Visual Query Formulation, Statistical Modeling, Recommender System, User Interface, User Study
Disciplines
Databases and Information Systems | Graphics and Human Computer Interfaces | Systems Architecture
License
This work is licensed under a Creative Commons Attribution 4.0 International License.
Recommended Citation
Saeef, Mohammed Samiul, "On Interactive Query Building With System-Generated Suggestions for Knowledge Graphs" (2025). Computer Science and Engineering Dissertations. 428.
https://mavmatrix.uta.edu/cse_dissertations/428
Included in
Databases and Information Systems Commons, Graphics and Human Computer Interfaces Commons, Systems Architecture Commons