[Paper Review] A Visual Entity-Relationship Model for Constraint-Based University Timetabling

Islam Abdelraouf, Slim Abdennadher+1 more|arXiv (Cornell University)|Sep 28, 2011

Constraint Satisfaction and Optimization5 references3 citations

TL;DR

This paper proposes a visual entity-relationship modeling approach for constraint-based university timetabling, enabling non-experts to specify scheduling problems via intuitive node-and-link diagrams representing resources (lecturers, student groups) and events (lectures, labs). The system automatically compiles these diagrams into efficient constraint programs using global constraints, achieving a complete timetable for 2233 events and 400+ resources in under 3.5 minutes on a standard PC.

ABSTRACT

University timetabling (UTT) is a complex problem due to its combinatorial nature but also the type of constraints involved. The holy grail of (constraint) programming: "the user states the problem the program solves it" remains a challenge since solution quality is tightly coupled with deriving "effective models", best handled by technology experts. In this paper, focusing on the field of university timetabling, we introduce a visual graphic communication tool that lets the user specify her problem in an abstract manner, using a visual entity-relationship model. The entities are nodes of mainly two types: resource nodes (lecturers, assistants, student groups) and events nodes (lectures, lab sessions, tutorials). The links between the nodes signify a desired relationship between them. The visual modeling abstraction focuses on the nature of the entities and their relationships and abstracts from an actual constraint model.

Motivation & Objective

• Address the challenge of translating complex university timetabling problems into effective constraint models, a task traditionally requiring expert programming knowledge.
• Enable non-technical stakeholders—such as academic staff and timetable coordinators—to model scheduling problems using a high-level, visual abstraction.
• Decouple problem specification from low-level constraint implementation to improve accessibility and reduce modeling errors.
• Develop a scalable and modular system that supports dynamic updates and consistency checks during model construction.
• Demonstrate the feasibility and efficiency of the approach on a real-world university instance with over 2000 events and 400 resources.

Proposed method

• Model university timetabling problems using a visual entity-relationship diagram with two primary node types: resource nodes (lecturers, teaching assistants, student groups) and event nodes (lectures, tutorials, lab sessions).
• Represent relationships between nodes via links, where link semantics are determined by the types of connected nodes, enabling semantic-aware constraint generation.
• Perform online consistency checks during model construction by leveraging rich semantic information encoded in node types and link relationships.
• Compile the visual model into a constraint programming (CP) model using SICStus Prolog, with automatic integration of global constraints (e.g., count, cumulative) to improve efficiency.
• Support soft constraint handling by encoding lecturer preferences (e.g., day-off wishes) as count-based constraints and maximizing satisfaction via labeling heuristics.
• Use a domain-specific compilation pipeline that maps visual constructs to optimized CP code, including symmetry-breaking and constraint propagation enhancements.

Experimental results

Research questions

• RQ1 Can a visual, entity-relationship-based modeling interface effectively abstract away the complexity of constraint-based university timetabling for non-expert users?
• RQ2 How can semantic information from node types and link relationships be leveraged to generate correct and efficient constraint programming models automatically?
• RQ3 To what extent can this visual modeling approach support scalability and modularity in real-world, large-scale timetabling instances?
• RQ4 Can the system achieve both high solution quality and performance on complex, real-world university timetabling problems with hard and soft constraints?
• RQ5 How does the visual model compare to traditional constraint network representations in terms of clarity and maintainability for non-programming stakeholders?

Key findings

• The visual entity-relationship model enabled the specification of a full university timetabling instance at the German University in Cairo (GUC) in just 5 hours, significantly reducing the time compared to manual methods.
• The system generated a complete SICStus Prolog constraint program from the visual model in only 7 seconds, demonstrating fast and reliable compilation.
• The first five feasible solutions were found within 3.5 minutes for a large-scale instance involving 2233 events and over 400 resources, indicating strong performance on complex problems.
• The approach supports dynamic model updates: modifying node properties or adding links automatically triggers recompilation of the constraint model, preserving consistency.
• The use of semantic-rich nodes and link types enabled the automatic generation of global constraints (e.g., count, cumulative), improving solver efficiency and solution quality.
• The system achieved high-quality solutions by maximizing soft constraint satisfaction—such as lecturer preferences for day-offs—using a maximize(SCONS) labeling strategy with fcc heuristic.