See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/379825214

APPLICATIONS OF GRAPH THEORY IN MACHINE LEARNING

Article · April 2024

CITATIONS READS

0 1,841

2 authors:

Shaila shivaji Jadhav Shinde Varsha

Tuljaram Chaturchand College Tuljaram Chaturchand College
6 PUBLICATIONS 16 CITATIONS 1 PUBLICATION 0 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Shaila shivaji Jadhav on 15 April 2024.

The user has requested enhancement of the downloaded file.

Journal of the Maharaja Sayajirao University of Baroda
ISSN: 0025-0422

APPLICATIONS OF GRAPH THEORY IN MACHINE LEARNING

Jadhav Shaila S., Department of Mathematics, T. C. College, Baramati, Dist. Pune, Maharashtra
Email: [email protected], [email protected]
Shinde Varsha H., Department of Mathematics, T. C. College, Baramati, Dist. Pune, Maharashtra
Email: [email protected]
ABSTRACT:
In mathematics, Graph Theory is the study of relation between vertices where vertices are different
objects depending upon examples. In graph, inter-related vertices join by a line called as edge.
Graph theory becomes a key part of mathematics, due to its ability to model and analyze
relationships between objects. It is useful to solve distinct Social network, Molecular network,
Disease Network, and Network modeling problems. Now a day’s artificial intelligence (AI) become
a key part of human life. Machine learning benefits from the application of graph theory in several
ways. Mathematical prerequisites are inherent to machine learning. We lower the dataset's
dimension and simplify the analysis by applying several facets of Graph Theory to our dataset prior
to machine model fitting. Applications of graph theory to machine learning are the focus of this
paper.

Keywords: Graph, Machine learning, Graph-based Representation.

INTRODUCTION:
Machine learning (ML) is essential in many fields. Math, statistics, computer science, and
cognitive science are just a few of the fields that make up the background of machine learning.
Graph theory intersects with machine learning and databases in several ways, enabling efficient
representation, querying, and analysis of complex relational data. Graphical representation is an
effective approach to understand the concepts information retrieval, recommendation systems,
question-answering systems, community detection, Disease analysis, and document classification.
Graph theory and machine learning are strongly related with each other. The field of graph theory
offers robust tools and frameworks for the analysis and representation of intricate data structures.
Machine learning techniques utilize these representations to handle a wide variety of problems. The
applications of graph theory in different phases of machine learning processes are covered in these
papers.
Machine Learning Process:
Machine Learning Process

Data Collection

Data Preprocessing

Model selection

Model Training

Model Evaluation

Model Deployment and Monitoring

Volume-58, No.1(III) : 2024 1

Journal of the Maharaja Sayajirao University of Baroda
ISSN: 0025-0422

Using above steps machine learning algorithm performs. There is a lot of literature which
demonstrate the significance of graph theory in machine learning tasks. [3]: Proposes graph
Consideration Organizations (GATs), which use consideration instruments to learn node portrayals
by taking care of their neighbours. [2]: This paper stretches out GCNs to semi-regulated learning
settings, where just a little piece of nodes have marks. By integrating both marked and unlabelled
information, GCNs really proliferate data across the diagram, further developing execution on
errands like node characterization. [4]: This exhaustive review gives an outline of different chart
brain organization (GNN) models and their applications in AI. It covers techniques going from
diagram convolutional organizations to graph consideration organizations, and examines
applications like informal community examination, suggestion frameworks, and bioinformatics. So
literature altogether show the adaptability and adequacy of applying graph hypothesis in AI,
empowering progressions in errands going from node characterization and connection forecast to
suggestion frameworks and positioning. Now we demonstrate some graphical format used in
machine learning as follows:

D1

D2

D3
D4

Figure 1: 2D Convolution
Neural Network
Figure 2: Individual Disease network
diagram

Volume-58, No.1(III) : 2024 2

Journal of the Maharaja Sayajirao University of Baroda
ISSN: 0025-0422

Graph Neural Network Taxonomy:

explicit in nature
applications in:
Structural
Molecules,
Physical systems
Graph
structure

Non structural Implicit

Undirected Directed
Directed/
Undirected

Static/
Dynamic

Graph types Home

Graph Nural
Network
Knowledge graph x road

Garden

Inductive

Ex. Does x person mobile

adicted? Node classification
Node level
Node regression
Node clustering

Graph Netflix Vodeo

learning Edge level
tasks Link Prediction

Graph classification
Graph level
Graph Mathching

Now we discuss the different tasks of machine learning algorithms in which Graph Theory plays an
important role.
Role of Graph Theory in machine learning:
1. Learning via Representation: By combining data from nearby nodes, Graph Neural
Networks (GNNs) use the concepts of graph theory to learn representations of nodes in a
network. These representations are commonly employed in tasks like node classification,
link prediction, and graph classification because they capture both the structural and feature
information of the nodes.

Volume-58, No.1(III) : 2024 3

Journal of the Maharaja Sayajirao University of Baroda
ISSN: 0025-0422

i. Data Graphs: It represents information in an organized frame where substances are

vertices, and connections between them are edges. They are broadly utilized for activities
like data recovery, suggestion frameworks, and question-answering frameworks.
ii. Cluster Graphs: Cluster representation is a significant task in machine learning, indicating
to group of similar data points based on certain highlights or qualities. Cluster charts are
graphical representations that portray the associations among data centres interior clusters.
These graphs are regularly utilized for visualization and interpretation of clustering.
a. Graph-Based Clustering: By grouping nodes in a graph according to their
connectedness or similarity, graph clustering algorithms divide the nodes into groups.
One method that uses the graph Laplacian matrix's eigenvalues and eigenvectors to
accomplish clustering is called spectral clustering.
b. K- Means Clustering: It is unsupervised machine learning. In this clustering there is no
labelled data. Also K numbers of clusters are present in this type. :

Healthcare
services

Primary
Emergen Service
cy s
Services

Diagnostic
Services

Figure 3: Cluster graph

Following diagram represents K- Clustering with the help of graph used in Healthcare
services which is further useful in some prediction algorithms.
2. Social Networks: Social network analysis involves studying the structure of social networks
using graph theory. It helps in tasks like community detection, influence analysis, and
recommendation systems. Social network analysis (SNA) includes considering the connections and
intelligent between people or substances inside a social arrange. Graph hypothesis gives a capable
system for speaking to and analyzing these systems.
3. Regression and Classification: In fields like social network analysis, recommendation systems,
and bioinformatics, graph-based features improve the performance of classification and regression
tasks by capturing relational information between entities.

4. Link Forecast: The purpose of link prediction challenges, which use graph theory, is to forecast
upcoming or missing links between nodes in a graph. Graph features like node degrees,
neighbourhood overlap, and structural similarities are used by algorithms like common neighbours,
Jaccard similarity, and preferential attachment to infer missing links. In recommendation systems,
social network analysis, and citation networks, link prediction is essential.
5. Finding anomalies: Graph-based anomaly detection methods are useful for identifying outliers
or anomalies in a graph. These algorithms find nodes or edges in the graph that substantially
diverge from the typical patterns of connectedness. Graph-based anomaly detection finds
application in fraud detection, infrastructure network monitoring, cyber security and network
intrusion detection.

Volume-58, No.1(III) : 2024 4

 Journal of the Maharaja Sayajirao University of Baroda
ISSN: 0025-0422

6. Recommendation Systems: Recommendation systems that rely on graphs model the

interactions between users and goods as a bipartite graph, in which nodes are users and items, and
edges are interactions (like ratings and purchases). PageRank, matrix factorization with graph
regularization, random walk with restart, and other graph algorithms take advantage of the network
structure to generate tailored suggestions based on item similarities and user preferences.
7. Graph Generation and Synthesis: Synthetic graphs that mimic the characteristics of actual
networks are created using graph theory. Graph theory principles are used by generative models,
including graph generative adversarial networks (GANs), graph self-encoders, and stochastic block
models, to understand the underlying distribution of graphs and produce new graphs with
comparable attributes. Data augmentation, network simulation, and benchmarking methods can all
benefit from graph generation. These exercises highlight graph theory's significance in relational
data modeling, capturing intricate dependencies, and deriving insightful conclusions from graph-
structured data, demonstrating its adaptability in a range of machine learning applications.

CONCLUSIONS:
The potential of graph theory for the development of machine learning techniques and its impact on
addressing complex real world problems is highlighted by this paper. The interdisciplinary nature
of graph theory and machine learning is also discussed in this paper. Using graph theory we can
develop an appropriate model that can effectively analyse interconnected data, leading to more
accurate predictions. Graph theory contributes to data science by providing a powerful framework
for representing and analysing complex structures. The need for further research to address existing
problems and explore untapped opportunities in this exciting intersection of disciplines would also
be highlighted.

REFERENCES:
1) Zhilin Yang, et al., “Revisiting semi-supervised learning with graph embeddings”, In
International Conference on Machine Learning, pp. 40–48, 2016.
2) Thomas N. Kipf et al., “Semi-Supervised Classification with Graph Convolutional
Networks", conference paper at ICLR, 2017.
3) Petar Velickovic et al., “Graph Attention Networks", conference paper at ICLR, 2018.
4) Jie Zhou et al., "Graph Neural Networks: A Review of Methods and Applications",
Elsevier journal of Keai, 2665-6510, 2021.
5) Geetha Narayanan Kannaiyan, et al., “A Review on Graph Theory in Network and
Artificial Intelligence”, Journal of Physics, 1742-6596, p.2-9, 2021.
6) Jonathan Bryan, et al., “Graph based machine learning improves just-in-time defect
prediction”, Research Article, p.1-19, 2023.
7) Bader Alwasel, et al., “Levaraging Graph-Based Representations to Enhance Machine
learning Performance in IIOT Network Security and attack Detection”, 2023, Applied
Science Journal, p.2-16, 2023.
8) Bharti Khemani, et al., “A review of graph neural networks: concepts, architectures,
techniques, challenges, datasets, applications, and further directions”, Journal of big data,
p.1-43, 2024.

Volume-58, No.1(III) : 2024 5

View publication stats