Lulwah Alkulaib
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Lulwah Alkulaib has a total of 34 co-authors. Here are the top co-authors by number of shared publications:
- Chang-Tien Lu
Virginia Tech, Alexandria, USA
10 shared publications - Shailik Sarkar
Florida Polytechnic Institute, Lakeland, FL, USA
7 shared publications - Abdulaziz Alhamadani
Florida Polytechnic University
4 shared publications - Abdulaziz Alhamadani
Virginia Tech
3 shared publications Chang-Tien Lu
Virginia Tech, Alexandria, VA, USA
3 shared publications- Mariam Almutairi
Virginia Tech, Alexandria, VA, USA
3 shared publications
Latest Publications
- Open Access
- 6 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 4076-4086
GeoRisk-RAG: A Hierarchy-Aware Risk Framework for Improving RAG Reliability through Selective Answering
- 6 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 4076-4086
Open AccessGeoRisk-RAG: A Hierarchy-Aware Risk Framework for Improving RAG Reliability through Selective Answering
- Open Access
- 6 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 116-126
Query-Graph Compatibility: A Formal Pre-Condition for Faithful Knowledge-Grounded Language Model Generation
Knowledge-graph-based retrieval-augmented generation (KG-RAG) grounds large language model (LLM) outputs in structured facts by retrieving a query-relevant subgraph before generation. However, existing KG-RAG systems often assume that the retrieved graph contains sufficient evidence to answer the query. When this assumption fails, the LLM may produce an answer that is factually plausible but not faithfully grounded in the KG. To address this problem, we introduce Query-Graph Compatibility (QGC), a formal, model-independent pre-condition for faithful KG-grounded generation. Given a natural-language query parsed into a typed conjunctive query and a KG with schema, QGC checks whether the query has a valid typed match in the graph, formalized as a typed homomorphism. If QGC=0, the KG lacks the structural evidence required by the query, so no generation model, regardless of architecture or scale, can produce a faithful answer from the KG alone-a consequence that follows directly from our definitions but yields a checkable pre-generation condition. QGC inherits the classical complexity of conjunctive-query evaluation: NP-complete in general, yet tractable for the bounded-width query structures that cover all KGQA queries in our evaluation. We validate QGC across six datasets and four domains. Under controlled edge deletion on MetaQA, QGC predicts LLM failure with AUC 0.78-0.93 for 1-2 hop queries across five LLMs, with accuracy gaps of +55 to +86 percentage points. Under natural incompleteness in clinical KGs (MIMIC-III, CORAL), all five evaluated models show positive QGC accuracy gaps (+5 to +31 percentage points), while PrimeKGQA exposes a boundary condition: structurally simpler QGC=0 contexts can yield higher correctness despite lacking faithful grounding. QGC checking runs in under 0.1 ms per query, making it practical as a real-time pre-filter or retrieval stopping criterion.
- 6 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 116-126
Open AccessQuery-Graph Compatibility: A Formal Pre-Condition for Faithful Knowledge-Grounded Language Model Generation
Knowledge-graph-based retrieval-augmented generation (KG-RAG) grounds large language model (LLM) outputs in structured facts by retrieving a query-relevant subgraph before generation. However, existing KG-RAG systems often assume that the retrieved graph contains sufficient evidence to answer the query. When this assumption fails, the LLM may produce an answer that is factually plausible but not faithfully grounded in the KG. To address this problem, we introduce Query-Graph Compatibility (QGC), a formal, model-independent pre-condition for faithful KG-grounded generation. Given a natural-language query parsed into a typed conjunctive query and a KG with schema, QGC checks whether the query has a valid typed match in the graph, formalized as a typed homomorphism. If QGC=0, the KG lacks the structural evidence required by the query, so no generation model, regardless of architecture or scale, can produce a faithful answer from the KG alone-a consequence that follows directly from our definitions but yields a checkable pre-generation condition. QGC inherits the classical complexity of conjunctive-query evaluation: NP-complete in general, yet tractable for the bounded-width query structures that cover all KGQA queries in our evaluation. We validate QGC across six datasets and four domains. Under controlled edge deletion on MetaQA, QGC predicts LLM failure with AUC 0.78-0.93 for 1-2 hop queries across five LLMs, with accuracy gaps of +55 to +86 percentage points. Under natural incompleteness in clinical KGs (MIMIC-III, CORAL), all five evaluated models show positive QGC accuracy gaps (+5 to +31 percentage points), while PrimeKGQA exposes a boundary condition: structurally simpler QGC=0 contexts can yield higher correctness despite lacking faithful grounding. QGC checking runs in under 0.1 ms per query, making it practical as a real-time pre-filter or retrieval stopping criterion.
- 14 July 2026
- Published by Springer Nature
- p. 420-432
AdaptiveEdge: Efficient Representation Learning for Structured Spatio-Temporal Graphs
- 14 July 2026
- Published by Springer Nature
- p. 420-432
AdaptiveEdge: Efficient Representation Learning for Structured Spatio-Temporal Graphs
- 4 February 2026
- Published by Springer Nature
- p. 155-167
TrendScope: A Temporal Hypergraph Framework for Food Trend Discovery
- 4 February 2026
- Published by Springer Nature
- p. 155-167
TrendScope: A Temporal Hypergraph Framework for Food Trend Discovery
- 22 May 2025
- Published by SAGE Publications in Journal of Planning Education and Research
- Vol. 45 (4) , 760-772
Unveiling the Ethical Dimension of Urban Digital Twins: A Framework for Responsible and Sustainable Urban Planning
This paper establishes a comprehensive ethical framework for urban digital twins (UDTs) to address the limited integration of ethical considerations in their design and implementation. It addresses a critical gap in the literature by providing a structured approach to embed ethics throughout the UDT development lifecycle. The research includes a systematic literature review, an analysis of UDT initiatives, and the development of an ethical framework that integrates co-creation and ethical checks. By embedding ethics at the core of UDT development, this framework provides urban planners and decision-makers with a practical tool to create responsible and sustainable urban ecosystems.
- 22 May 2025
- Published by SAGE Publications in Journal of Planning Education and Research
- Vol. 45 (4) , 760-772
Unveiling the Ethical Dimension of Urban Digital Twins: A Framework for Responsible and Sustainable Urban Planning
This paper establishes a comprehensive ethical framework for urban digital twins (UDTs) to address the limited integration of ethical considerations in their design and implementation. It addresses a critical gap in the literature by providing a structured approach to embed ethics throughout the UDT development lifecycle. The research includes a systematic literature review, an analysis of UDT initiatives, and the development of an ethical framework that integrates co-creation and ethical checks. By embedding ethics at the core of UDT development, this framework provides urban planners and decision-makers with a practical tool to create responsible and sustainable urban ecosystems.
- 25 January 2025
- Published by Springer Nature
- p. 119-136
Empowering Airline Route Decisions with LLM-Generated Pseudo-labels and Zero-Shot Review Prediction
- 25 January 2025
- Published by Springer Nature
- p. 119-136
Empowering Airline Route Decisions with LLM-Generated Pseudo-labels and Zero-Shot Review Prediction
- 25 January 2025
- Published by Springer Nature
- p. 171-186
HyperSMOTE-MC: Enhancing Multiclass Bot Detection on X Through Hypergraph-Based Resampling
- 25 January 2025
- Published by Springer Nature
- p. 171-186
HyperSMOTE-MC: Enhancing Multiclass Bot Detection on X Through Hypergraph-Based Resampling
- 24 January 2025
- Published by Springer Nature
- p. 262-277
Enhancing School Success Prediction with FRC and Merged GNN
- 24 January 2025
- Published by Springer Nature
- p. 262-277
Enhancing School Success Prediction with FRC and Merged GNN
- Open Access
- 22 November 2024
- Published by Association for Computing Machinery (ACM)
- p. 1-7
FHIRViz: Multi-Agent Platform for FHIR Visualization to Advance Healthcare Analytics
The shift to electronic health records (EHRs) has enhanced patient care and research, but data sharing and complex clinical terminology remain challenges. The Fast Healthcare Interoperability Resource (FHIR) addresses interoperability issues, though extracting insights from FHIR data is still difficult. Traditional analytics often miss critical clinical context, and managing FHIR data requires advanced skills that are in short supply. This study presents FHIRViz, a novel analytics tool that integrates FHIR data with a semantic layer via a knowledge graph. It employs a large language model (LLM) system to extract insights and visualize them effectively. A retrieval vector store improves performance by saving successful generations for fine-tuning. FHIRViz translates clinical queries into actionable insights with high accuracy. Results show FHIRViz with GPT-4 achieving 92.62% accuracy, while Gemini 1.5 Pro reaches 89.34%, demonstrating the tool's potential in overcoming healthcare data analytics challenges.
- 22 November 2024
- Published by Association for Computing Machinery (ACM)
- p. 1-7
Open AccessFHIRViz: Multi-Agent Platform for FHIR Visualization to Advance Healthcare Analytics
The shift to electronic health records (EHRs) has enhanced patient care and research, but data sharing and complex clinical terminology remain challenges. The Fast Healthcare Interoperability Resource (FHIR) addresses interoperability issues, though extracting insights from FHIR data is still difficult. Traditional analytics often miss critical clinical context, and managing FHIR data requires advanced skills that are in short supply. This study presents FHIRViz, a novel analytics tool that integrates FHIR data with a semantic layer via a knowledge graph. It employs a large language model (LLM) system to extract insights and visualize them effectively. A retrieval vector store improves performance by saving successful generations for fine-tuning. FHIRViz translates clinical queries into actionable insights with high accuracy. Results show FHIRViz with GPT-4 achieving 92.62% accuracy, while Gemini 1.5 Pro reaches 89.34%, demonstrating the tool's potential in overcoming healthcare data analytics challenges.
- Open Access
- 1 January 2024
- Published by Association for Computational Linguistics (ACL)
- p. 11-26
Synthetic Arabic Medical Dialogues Using Advanced Multi-Agent LLM Techniques
Mariam ALMutairi, Lulwah AlKulaib, Melike Aktas, Sara Alsalamah, Chang-Tien Lu. Proceedings of The Second Arabic Natural Language Processing Conference. 2024.
- 1 January 2024
- Published by Association for Computational Linguistics (ACL)
- p. 11-26
Open AccessSynthetic Arabic Medical Dialogues Using Advanced Multi-Agent LLM Techniques
Mariam ALMutairi, Lulwah AlKulaib, Melike Aktas, Sara Alsalamah, Chang-Tien Lu. Proceedings of The Second Arabic Natural Language Processing Conference. 2024.