This tutorial shows how traditional Entity/Relationship modeling and modern graph data modeling can be combined to bring forward well-designed graph data models that process workloads and maintain data integrity eficiently. Despite the popularity and growing maturity of graph database systems, they continue to rank well below trusted relational technology1. The reasons for this are not just historical. Indeed, an inhibitor to the further uptake of graph databases is the lack of principled methodologies for their design, rigorous schema and data integrity support. As a consequence, academics and practitioners have worked together to develop proposals for emerging standards of query languages [1, 2], schema [3], and data integrity [4]. For the emerging standards PG-Schema [3] for graph schemata and PG-Key [4] for graph integrity, it is not yet well understood which of their fragments supports which applications. PG-Schema, in particular, has been developed to support basic features of Entity/Relationship (E/R) models, but its expressiveness is well beyond those capabilities. Chen's E/R model [5] constitutes the best breed of conceptual data models. The model captures entities and their relationships in an easy-to-understand framework powerful enough to derive a formal data model. E/R models visually represent complex requirements for the target database [6]. In fact, the graphical depiction of an E/R diagram is invaluable for efective communication between experts with diferent expertise. E/R modeling is a methodology for generating well-designed E/R diagrams as they guarantee directed acyclicity, data integrity and do not exhibit redundancy of data or anomalies of updates in any instance [7, 8, 9]. Given these developments, the tutorial addresses the following ambitious research question.
What is a methodology for designing property graph schemata that can process workloads and maintain data integrity eficaciously on each of their graph instances?
The main goals are to show what Entity/Relationship modeling can do for graph data, and what graph data can do for Entity/Relationship modeling. An important point is to bring together diferent communities for their mutual benefit.
In what follows, Sec. 2 will describe the target audience and knowledge outcomes. The latter are fundamental for conceptual, graph and logical data modeling. A detailed outline and timetable of the tutorial will be given in Sec. 3. The tutorial method is explained in Sec. 4, before a short biography for the presenters is given in Sec. 5.
The specific target audience is people with an interest in conceptual data modeling or graph data modeling. The tutorial showcases fundamentals for these disciplines, and focuses on how they can bring out the best in one another when combined. The topic is timely as graph databases have grown larger in popularity among researchers and practitioners, but also because the design of graph databases has emerged as a new and important direction. The tutorial is accessible to students with a general computer science education at undergraduate level because fundamental concepts are introduced. This is even of interest to experts in the area, due to the diversity of models and languages available, which makes it necessary to fix concepts and notation.
The overarching outcome targeted for this tutorial is to survey the fundamental impact E/R modeling and graph databases have on one another.
Property graphs that comply with E/R diagrams, called E/R graphs, constitute the first E/R databases that are graphs. Hence, instead of translating E/R diagrams to other data models, we can directly manage E/R graphs as graph databases. That is, graph databases bring E/R modeling concepts to life. The tutorial illustrates how E/R diagrams form a core fragment of PG-Schema that captures well-designed (graph) databases. Hence, PG-Schema can be viewed as a general data modeling tool.
It is demonstrated that reasoning about PG-Key is infeasible, but keys of E/R models form a fragment of PG-Key for eficiently managing entity integrity in well-designed property graphs.
The audience learns about principled approaches to managing referential integrity eficiently in E/R graphs. This contrasts relational and graph-based approaches, and pinpoints what opportunities for improving integrity management in graph database systems exist.
Overall, the tutorial conveys how E/R diagrams and property graphs can be combined to unify conceptual, logical and graph data modeling, how integrity management can be taken to the next level by eliminating property redundancy, and why relational benchmarks perform very well when translated into property graphs. Indeed, the tutorial shows how a major inhibitor to the uptake of graph databases can be turned into a strong driver.
We will give a detailed outline of what the tutorial covers, and indicate how much time is spent on each topic. The following subsections form the logical units of the tutorial, in the sequence they are presented.
After some motivation and outline of the goals (5 minutes) [
The concept of property graphs is introduced next, both in terms of formal definitions and visualizations (10 minutes). We also illustrate the concepts on our running example.
We then discuss the recent proposal for the definition of key constraints in property graphs, called
PG-Key [
Next we give a concise overview of PG-Schema [
Our next aim is to show how traditional E/R modeling provides a methodology for designing property
graphs well (20 minutes). In the first of two steps, we illustrate that E/R diagrams are property graphs
themselves, but also form a fragment of PG-Schema. This situation is reminiscent of XML where
XML Schema definitions constitute XML documents themselves [
Next, we demonstrate that E/R graphs and diagrams ofer principled concepts for maintaining entity and referential integrity eficiently within graph database systems (20 minutes). In particular, translations of relational databases to E/R graphs ofer various benefits for data integrity management. We illustrate the notion of an E/R key and demonstrate that every E/R key is also a PG-Key, but not vice versa. We demonstrate that E/R keys form an eficient fragment of PG-Key, namely for well-designed property graphs. In terms of referential integrity, we illustrate two principled approaches: one that always duplicates key properties as foreign key properties (relational semantics), and one that always uses references by directed edges in graphs (graph semantics). The first approach can directly benefit from graph technology, while the second one can only make use of current technology in case all key properties are defined locally on nodes without use of references. Hence, we propose a third approach (mixed semantics) which maximizes the opportunity for references while complying with current technology limits.
We then showcase TPC-H as an industry-like use case that illustrates the main concepts and quantifies how well the E/R approach works at operational level (10 minutes). We present translations of the TPC-H schema, including keys and foreign keys, into an E/R diagram using the diferent semantics, translate its query and update operations, and compare the eficiency of their evaluation within Neo4j and MySQL. The results demonstrate to the audience how well our initial research question has been addressed. It also shows the audience how important data modeling is for core database operations.
Finally, we summarize main takeaways and outline future research directions (10 minutes).
The presentation is based on slides to convey the knowledge outcomes in a visually appealing way that draws the audience’s attention and promotes their understanding, but also facilitates interaction. Our timetable, proposed above, accommodates the presentation itself but also designates time for questions and discussions. In particular, we will prompt the audience to criticize our toy example and ofer alternative modeling approaches, but also ask them to convert between some classical relational database instances and graph instances to facilitate understanding further. Lots of examples and visualizations illustrate ideas, techniques and findings. A 90 minute version was presented at VLDB 2025. The ER audience is quite diferent, the topic well aligned with ER, and a presentation of 2 hours conveys more details, discussion, and interesting direction of future research for the modeling community.
We only use a standard projector and a board for exploring ideas interactively with the audience.
Philipp is a research fellow and teaching assistant with considerable experience teaching database topics to undergraduate and postgraduate students of diferent backgrounds. The topic of his PhD was on the design of graph databases.
Sebastian is a Professor of Computer Science with vast experience in presenting database topics to audiences of various backgrounds. He has presented extensively at top database conferences, particularly on topics of the tutorial.
The author(s) have not employed any Generative AI tools.