An efficient graph modeling approach for storing and analyzing heterogeneous IoT data

management as a case study. 
3.1. A Graph-based View on IoT Data
A conceptual view of IoT data could be 
represented as in Figure 1. That is fused by a social 
graph, a spatial graph, and a things graph into one 
graph model, and incorporates the relationships 
among them. The graph components are explained 
in more detail as follows.
Figure 1. A conceptual view of IoT Graph Data
a) Things Graph
This graph represents entities including 
sensors and devices and their connectivity. Each 
node represents a sensor or a device with different 
attributes such as SensorID, Name, Type, Position, 
Status, Timestamp, and Value. An edge represents 
the relationship between two sensors/devices, and 
two types of edge-label are used in things graph 
including Connects and Links.
b) Spatial Graph
This graph represents locations and their 
proximity. Each node is a place with attributes such 
as LocationID, PlaceName, and Coordinates. Each 
edge indicates the proximity between two locations. 
Besides, a node in the Spatial Graph could be 
connected by nodes in the Things Graph, which 
indicates that some sensors/devices are employed at 
certain locations. This relation between a thing and 
a location is represented by using AsignedTo type 
edge. Also, a node in the Spatial Graph could be 
connected by another node from the Social Graph 
to show who is in a specific location. There are 
four edge types to represent these kinds of relations 
including WorksAt, WorksFor, StudiesAt, and 
LivesAt.
c) Social Graph
This graph represents people who are using 
IoT devices and their relationship. Each node is 
a person with some attributes such as ID, Name, 
Age, and Title. An edge represents the relationship 
between two people. Furthermore, a node of Social 
Graph could be connected to a node from Spatial 
Graph to show where a person is and connected to 
a node from Things Graph to indicate which things 
are used by a person.
3.2. IoT Graph Data Modeling
Graph data modeling is the translation of 
a dataset in a conceptual view to a graph model. 
During the graph modeling process, we determine 
which entities in the dataset should be nodes (or 
vertex), which should be edges, and which should be 
properties. The result is a blueprint of whole entities, 
relationships, and properties in the dataset. We can 
use that blueprint to create a visualization model.
In fact, an entity or a relationship could have 
several properties. For instance, a person is identified 
by his/her national ID, first name, last name, birth 
of date, and he/she might have a relationship as 
a colleague with another person since 2019. For 
representing data in detail and rich information, a 
comprehensive graph model is introduced which is 
named a property graph. The property graph is first 
introduced in [9], and a formal definition is given 
by Angles et al. in [10]. In the later one, a property 
graph is defined as a tuple (V, E, ρ, λ, δ), where V is 
a set of nodes and E is a set of edges in the graph, 
ISSN 2354-0575
Journal of Science and Technology24 Khoa học & Công nghệ - Số 27/Tháng 9 - 2020
ρ is a total function E → V × V, λ is a total function 
that defines labels on both V and E, δ is a partial 
function that maps a property of a node or an edge 
to a value. We present an extension of the property 
graph to support data modeling to be easy and more 
clear.
Property Graph. A property graph is a tuple G = 
(V, E, Σ, Θ, F, λ, P, ϑ, ϱ), where:
• V: is a finite set of nodes (vertices)
• E: is a finite set of edges
• Σ: is a finite set of labels for edges
• Θ: is a finite set of labels for nodes
• F: is the function mapping each node v ∈ V to a 
label from Θ.
• λ: is the function mapping each edge e ∈ E to a 
label from Σ.
• P: is a finite set of property names for vertices/edges
• ϑ: is the function mapping each node v ∈ V with a 
given property p ∈ P to a specific value.
• ϱ: is the function mapping each edge e ∈ E with a 
given property p ∈ P to a specific value.
Figure 2. An example of IoT graph data modeling
Figure 3. The format of nodes and edges in the 
property graph
Example: An illustration of a property graph is 
shown in Figure 2. In this example, the values of 
V, E, Σ, F, and λ are not difficult to recognize. Here, 
the property graph has three more parameters P, ϑ, 
and ϱ, where P = {name, age, no, time, since}, the 
example of mapping functions for node properties 
and edge properties (a few of them) are listed as the 
following:
ϑ(1, name) = Quyet ϑ(1, age) = 32
ϑ(6, name) = Computer Engineering 
ϑ(4, no) = 718
ϱ((1, 3), since) = 2019 
ϱ((5, 7), time) = 2019/05/01 2:00PM
Thus, we can understand that properties are 
name-value pairs which are used to add qualities 
(more information) to nodes and relationships 
(edges). A set of properties for each type of node/
edge is specified by using the format shown in 
Figure 3. The value part of the property can hold 
different data types such as string, number, and date 
time. Each node and edge can have zero or few 
properties. For example, node 1 has two properties 
including name and age, and edge (1,3) has only 
one property since, while edge (2,5) has no property 
(the value will be null when we map any property 
name on the edge (2,5)).
From the conceptual view of IoT data, we 
can categorize the entities in an IoT system into 
three main groups including People, Locations, and 
Things for the brevity of the explanation. Besides, 
there are a few other groups related Things such as 
Applications or Permissions could be considered for 
representing IoT data. It depends on the objectives 
of the IoT systems. In this paper, we consider the 
IoT data management for smart building evacuation 
systems as a case study, therefore, we will describe 
the main groups and entities related to such a kind 
of system. For a better data representation and data 
exploration, we specify all entities in each group, 
each of them is considered as a node type (or node 
label) in the IoT graph model, and the relationship 
between two nodes is represented as an edge. The 
descriptions of nodes, edges, and their relationships 
in our graph model are described in Table 1 and 
Table 2, respectively.
ISSN 2354-0575
Khoa học & Công nghệ - Số 27/Tháng 9 - 2020 Journal of Science and Technology 25
Table 1. Node Types Description
Table 2. Edge Types Description
4. Experimental Evaluation
Exp-1: Analysis of IoT Graph Data
In this experiment, we analyze the graph 
characteristics with the changes in heterogeneous 
IoT data. To do this, we first generate a graph 
database by using gMark [11]. This graph follows 
the model that we presented in the previous section. 
It has 36,000 nodes, 273,610 edges, and 19 edge-
labels. The occurrence of labels follows the given 
Zipfian or uniform distribution. We then extract 
from the graph to obtain other six smaller graphs 
which contain only one or two kinds of graph from 
things, social, and spatial graphs. Finally, we use 
Gephi [12] to analyze the changes of parameters 
of these graphs. Specifically, we consider the 
following graph parameters:
• Graph size: the number of nodes (|V|) and 
edges (|E|).
• Number of relationships (|L|): the number of 
different labels in the graphs.
• Average degree: in a directed graph, it is 
defined as the fraction of the number of edges to the 
number of nodes. 
• Average path length: the average number of 
steps along the shortest paths for all possible pairs 
of nodes.
• Diameter (D): the number of edges in the 
shortest path between the most distant nodes.
• Strongly connected components (|C|): the 
maximal strongly connected subgraph, in which, a 
subgraph is called a strongly connected component 
if there is a path between all pairs of nodes.
Table 3 illustrates the results of analyzing 
graph parameters. We observe that when different 
graphs are fused together, it could generate a more 
complex graph with the increase of the number of 
relationships, the average degree, the average path 
length, and the value of other parameters. This 
causes substantial searching cost and long response 
time due to the large size of the graph and/or 
complex queries.
Exp-2: Evaluation of Query Performance
We evaluate the efficiency of analyzing IoT 
data using graph query. To do this, we compare the 
query performance between T-SQL queries on a 
relational database and Cypher queries on a graph 
database. We use the IoT dataset generated in Exp-
1. We convert and import this dataset into 14 tables 
in MySQL with 256,318 records. The dataset is also 
imported to a graph database, Neo4j, with 36,000 
nodes and 273,610 edges.
In this experiment, we use four common types 
of query including Look Up, Range, Complex 
(Join/Nested), and Aggregation, which are often 
used to extract knowledge from IoT data We write 
twelve queries, each type of query has three queries. 
The queries are written in both SQL language for 
running on MySQL and Cypher language for 
running on Neo4J. The experimental results are 
illustrated in Figure 4.
ISSN 2354-0575
Journal of Science and Technology26 Khoa học & Công nghệ - Số 27/Tháng 9 - 2020
Table 3. Analysis of IoT Graph Characteristics
Figure 4. Query performance comparision between relational database and graph database
From the results, we found that using Cypher queries 
on Neo4J can obtain better performance comparing 
to using SQL queries on MySQL in all the cases in 
overall. Specifically, the Look Up queries (#1, #2, 
#4) and Range queries (#4, #5, #6) take a low cost 
on both relational databases and graph databases. 
In the case of testing complex queries like Nested 
queries (#Q7, #Q8, #9), the performance of using 
Cypher queries on graph databases is much faster 
than the one using SQL queries on relational 
databases. We observed that Cypher queries reduced 
the average execution time around 3, 6, 6 times than 
SQL queries corresponding to #Q7, #Q8, and #Q9, 
respectively. We also observed that Aggregation 
queries on graph databases often take high cost. 
Indeed, their performance is up to 3 times slower 
than the ones with SQL queries (#10, #11, #12).
5. Conclusion
This paper proposed a graph model for 
representing IoT data. The proposed graph model 
represented entities in IoT environment such as 
devices, locations, people with attributes and 
relationships between two entities. The efficiency 
of the proposed graph model was evaluated on 
a simulated smart building management dataset. 
Experimental results showed that the proposed 
model is more efficient than relational model in 
storing and analyzing IoT data. 
ISSN 2354-0575
Khoa học & Công nghệ - Số 27/Tháng 9 - 2020 Journal of Science and Technology 27
References
[1]. V. Arora, F. Nawab, D. Agrawal, and A. El Abbadi, “Multi-representation based data processing 
architecture for iot applications,” in 2017 IEEE 37th International Conference on Distributed 
Computing Systems (ICDCS). IEEE, 2017, pp. 2234–2239.
[2]. A. M. Ibrahim, I. Venkat, K. Subramanian, A. T. Khader, and P. D. Wilde, “Intelligent evacuation 
management systems: A review,” ACM Transactions on Intelligent Systems and Technology (TIST), 
2016, vol. 7, no. 3, p. 36.
[3]. Nguyen, Van-Quyet, et al. “A Scalable Approach for Dynamic Evacuation Routing in Large Smart 
Buildings.” 2019 IEEE International Conference on Smart Computing (SMARTCOMP). IEEE, 2019.
[4]. M. M. Rathore, A. Ahmad, A. Paul, and G. Jeon, “Effcient graph-oriented smart transportation 
using internet of things generated big data,” in 2015 11th International Conference on Signal-Image 
Technology & Internet-Based Systems (SITIS). IEEE, 2015, pp. 512–519.
[5]. J. Byun, S. H. Kim, and D. Kim, “Lilliput: Ontology-based platform for iot social networks,” in 
2014 IEEE International Conference on Services Computing. IEEE, 2014, pp. 139–146.
[6]. V.-Q. Nguyen and K. Kim, “Comparison of relational databases and graph databases for 
heterogeneous iot data management,” in Proceedings of KISM Spring Conference 2019, 2019, pp. 
194–204.
[7]. R. Jin, Y. Xiang, N. Ruan, and H. Wang, “Effciently answering reachability queries on very large 
directed graphs,” in Proceedings of the 2008 ACM SIGMOD international conference on Management 
of data, 2008, pp. 595–608.
[8]. Nguyen, Van-Quyet, and Kyungbaek Kim. “Estimating the evaluation cost of regular path 
queries on large graphs.” Proceedings of the Eighth International Symposium on Information and 
Communication Technology, 2017.
[9]. M. A. Rodriguez and P. Neubauer, “Constructions from dots and lines,” Bulletin of the American 
Society for Information Science and Technology, 2010, vol. 36, no. 6, pp. 35–41.
[10]. R. Angles, M. Arenas, P. Barceló, A. Hogan, J. Reutter, and D. Vrgoč, “Foundations of modern 
query languages for graph databases,” ACM Computing Surveys (CSUR), 2017, vol. 50, no. 5, p. 68.
[11]. G. Bagan, A. Bonifati, R. Ciucanu, G. H. Fletcher, A. Lemay, and N. Advokaat, “gmark: Schema-
driven generation of graphs and queries,” IEEE Transactions on Knowledge and Data Engineering, 
2017, vol. 29, no. 4, pp. 856–869.
[12]. M. Bastian, S. Heymann, M. Jacomy et al., “Gephi: an open source software for exploring and 
manipulating networks.” ICWSM, 2009, vol. 8, pp. 361–362.
MỘT CÁCH MÔ HÌNH HÓA BẰNG ĐỒ THỊ HIỆU QUẢ CHO VIỆC 
LƯU TRỮ VÀ PHÂN TÍCH DỮ LIỆU IOT HỖN HỢP
Tóm tắt:
 Trong môi trường Internet of Thing (IoT), các thực thể với các thuộc tính và số lượng khác nhau sẽ kết 
nối với nhau tạo thành một mạng lưới liên kết dày đặc. Cụ thể, không chỉ các thiết bị máy móc mà còn các 
thực thể khác như con người, địa điểm và ứng dụng được kết nối với nhau. Hầu hết các ứng dụng IoT phải 
đều phải đối diện với các thách thức khi một lượng lớn dữ liệu thay đổi nhanh chóng do các thực thể mới 
đang được thêm vào hệ thống hoặc trạng thái kết nối giữa các thực thể thay đổi thường xuyên. Điều này yêu 
cầu một mô hình dữ liệu cho phép dễ dàng trong việc biểu diễn các thực thể và hỗ trợ lưu trữ, thêm, xóa và 
cập nhật quan hệ giữa các thực thể mà không ảnh hưởng đến tính khả dụng của ứng dụng. Trong bài báo 
này, chúng tôi đề xuất một mô hình đồ thị chung có thể được sử dụng để thiết kế cơ sở dữ liệu đồ thị hỗ trợ 
hiệu quả cho việc lưu trữ và phân tích dữ liệu IoT. Chúng tôi biểu diễn dữ liệu IoT dựa trên mô hình đồ thị 
và lấy việc quản lý dữ liệu của tòa nhà thông minh là một trường hợp minh họa. Thông qua việc phân tích 
kết quả thực nghiệm và so sánh ở các khía cạnh khác nhau, chúng tôi thấy rằng phương pháp tiếp cận bằng 
mô hình đồ thì có thể áp dụng để lưu trữ và phân tích dữ liệu IoT hỗn hợp một cách hiệu quả.
Từ khóa: Mô hình hóa đồ thị, Cơ sở dữ liệu đồ thị, Truy vấn đồ thị, Dữ liệu kết nối, Quản lý dữ liệu IoT.