5

GRAPH DATA
SCIENCE BASICS
EVERYONE
SHOULD KNOW
5 Graph Data Science Basics Everyone Should Know

Why Graph Data Science and Why Now?

Commercial applications of graph data science are new and most data experts are still coming
up to speed on how to best use it in their organizations. While some data experts studied
graph theory, others have not had much exposure to it. Graph data science brings together
graph analytics, statistics, and AI and ML techniques to improve their predictive and prescriptive
models. This paper walks you through the basics of graph data science so you will feel confident
knowing when to use it in your daily work.

1. What is a graph?
Before you can understand graph data science, you need to understand graphs. At its most
fundamental, a graph is simply a different way of structuring data. Instead of rows and columns,
like in a traditional, relational database table or dataframe, graphs use nodes (nouns) and
relationships (verbs) as their primary structure.

In a graph, nouns – people, places, things, organizations – are nodes. The relationships between
them are verbs: friends, works for, likes, and so on.

Name Del Name Ash Graphs represent data

Born May 29, 1970 Born Dec 5, 1975 via relationships.
Twitter @del
Del LOVES Ash

LOVES Nodes represent

an entity in
LIVES WITH
the graph

Relationships
NS
DR

(edges/links) connect
OW

since Jan 10, 2011

IVE

nodes to each other

S

Car Properties describe

Brand Volvo a node or relationships
Model V70
name, age, height, etc.

2
5 Graph Data Science Basics Everyone Should Know

2. How is graph data science different

from traditional analytics?
When you analyze data in tabular form, like in a relational database, you try to make sense of
data points without a coherent way to analyze their connections. Ben Squire, a data scientist,
compared traditional methods to “trying to solve a Rubik’s Cube by only looking at one side.” A
graph gives data experts the ability to look at, understand, and analyze the connections between
each data point. This gives data context that is impossible to get from a tabular data model. You
can see how strong connections are, where groups of connections form, how important each
connection is, and how connections influence one another.

Graph data science brings together graph statistics, analytics, and ML to put data in context and
answer pressing questions.

Graph statistics, queries, and visualization drive exploration and insights. Graph statistics
provide basic measures about a graph, such as the number of nodes and the distribution of
relationships. Graph queries answer any question, no matter how deep, whether 6 or 600
degrees of separation. Graph visualization empowers data experts to see their data and
explore patterns that bear further investigation.

Graph analytics builds on graph statistics by answering specific questions and gaining
insights from connections in existing or historical data. Graph queries and algorithms are
typically applied together in “recipes” during graph analytics, and the results are used
directly for analysis.

Graph-enhanced ML is the application of graph data and analytics results to train ML models
or support probabilistic decisions within an AI system. Graph statistics and analytics are
often used in conjunction to answer certain types of questions about complex systems and
the subsequent insights, applied to improve ML.

3
5 Graph Data Science Basics Everyone Should Know

Quick Look: In-Graph ML Techniques That Do the Work for You

Interest in graph-based machine learning is exploding. By 2025, Gartner projects that graph
technology will fuel 80% of data and analytics innovations, up from 10% in 2021. Here are a few
advanced graph data science techniques your team may want to explore.

Link prediction fills in the blanks in Node embedding transforms the Node classification models predict the
your data and predicts changes in topology and features of your class of nodes in your graph. A class
your graph’s structure. Link prediction graph into a low-dimensional vector can be a binary indicator, like whether
is a common machine learning task representation of each node. These a user account is engaged in fraud, or
applied to graphs: training a model to vectors, also called embeddings, can a multivalued indicator, such as which
learn where relationships should exist be used for exploratory data analysis, market segment a customer belongs
between pairs of nodes in a graph. You similarity measurements, and ML. Node to. Node classification models can be
can think of link prediction as building a embeddings can aggregate information trained to predict which class nodes
model to predict missing relationships about a node’s position in the graph, (including any new nodes) belong to.
in your dataset or relationships that are its local neighbors, its centrality and Node classification can incorporate a
likely to form in the future. With graph influence, and in some cases, other broad range of input features, including
data science, you can train supervised numeric node properties. the network structure of your graph and
ML models based on the relationships properties from your source data.
and node properties in your graph to
predict the existence – and probability –
of relationships.

Graph Data Science

Unsupervised Graph Algorithms Supervised Machine Learning

Which parts of my
graph are connected to Which nodes are Where will connections What‘s the label
each other? most similar? form next? for this node?

Clustering Association Link Node

Prediction Classification
• Community • Similarity
Detection • Pathfinding

How important What‘s the missing

is each node? property value?

Dimension Reduction Property

(generalization) Regression

• Centrality
• Embeddings

4
5 Graph Data Science Basics Everyone Should Know

3. Why do graphs matter?

Across an organization, every department can benefit from graph data science to answer questions
like who or what is important, what should I do next, and what’s unusual about this?

While use cases for graph data science span industries and lines of business – from life sciences to
manufacturing – a few use cases are rapidly becoming the most popular among data scientists.

Finance Marketing Ops

• Customer 360 • Product Development
• Fraud Detection
• Influencer Strategy • Pipeline Acceleration
• Pricing Analysis
• Campaign Optimization • Supply Chain Optimization
• Budgeting
• Product Recommendations • Infrastructure Planning
• Forecasting

IT HR
• Network Monitoring • Training
• Cybersecurity • Upskilling & Retention
• DevOps • Promotions

Anomaly and fraud detection

Anomaly detection across corporate networks can help to identify cybersecurity attacks and
prevent data loss. The same strategy used to identify threat actors in a cybersecurity context
can be used to detect fraud in banking, insurance, and government programs by analyzing the
relationships and behaviors in your graph.

Those using graph data science

to curb fraud have seen detection
improvements of over 300%, saving
millions every year.

5
5 Graph Data Science Basics Everyone Should Know

Customer 360
Across the globe, businesses try to better understand their customers and improve lifetime value
(LTV). With graph data science, customer knowledge can become more accurate and complete
through entity resolution. This process looks at all the database entries and identifies duplicates.
Creating a complete, master database entry for each customer instead of having multiple,
incomplete entries improves LTV and deepens customer knowledge, allowing for optimized
marketing programs and offers.

Those using graph data science

for customer 360 can increase
customer knowledge by 30%.

Recommendation engines
Recommendation engines became well known through Netflix and online shopping experiences.
However, recommendation engines have uses across the business. From product development
to human resources for retaining employees through upskilling training, recommendation engines
power some of the most important parts of a business.

Supply chain management

Improving a supply chain leads to savings, not just in dollars, but also in carbon emissions. Every
optimized route, perfect timing, and perfect delivery mean happier customers and less waste
across time, infrastructure, and emissions. Graph data science helps optimize supply chain
routes by finding the best path, balancing cost and efficiency with customer satisfaction and
sustainability.

An ROI of tens of millions for one

organization using graph data
science for route optimization is
nothing compared to the 60,000
tons of carbon emissions eliminated
by using those optimized routes.

6
5 Graph Data Science Basics Everyone Should Know

4. What are the big questions

graph data science helps answer?
Graph data science helps you answer big questions to make critical business decisions. If you
hear questions like these, you may benefit from using graph data science to answer them:

What’s unusual? What should I recommend?

To understand anomalies and hidden patterns in your To build a recommendation engine, consider using
graph, consider using community detection. Communities similarity. Similarity identifies repeating patterns in your
are clusters within your graph, and community detection graph. Similarity algorithms employ set and distance
algorithms can be used to discover and identify these comparisons to score how alike individual nodes are based
clusters. Detecting communities helps you uncover on their neighbors or properties. This approach is used in
unusual patterns, predict similar behavior, find duplicate applications such as personalized recommendations and
entities, or simply prepare data for other analyses. developing categorical hierarchies.

What content is the most important? What is the optimal route?

Who is the most influential? To understand route optimization, use pathfinding.
Where is our biggest risk? Pathfinding algorithms find the best routes across your
To understand what’s important, use centrality algorithms. connected data. Pathfinding algorithms are foundational to
Centrality metrics like PageRank help you identify what’s graph analytics and find the most efficient or shortest paths
important. Centrality algorithms reveal which nodes to traverse between nodes. They can be used to understand
are important based on graph topology. They identify complex dependencies and evaluate routes for uses such as
influential nodes based on their position in the larger physical logistics and least-cost call or IP routing.
network, including their connections. These algorithms
are used to infer group dynamics such as credibility,
cascading vulnerability, and bridges between groups.

7
5 Graph Data Science Basics Everyone Should Know

5. Who uses graph data science tools?

Anyone who works with data can take advantage of graph data science to help find answers to
difficult questions. Here are some of the roles currently using graph data science.

Data scientists are typically the primary user of graph data science
tools because they are practitioners of data science with deep
knowledge of algorithms and models.

Machine learning engineers work to scale, improve,

embed, integrate or operate machine learning models that are
developed by data scientists.

Data engineers identify trends in data sets, create data pipelines,

and leverage graph algorithms to transform and enrich graph data
to make it more useful to the organization and other data experts.

Citizen data scientists apply graph data science

techniques without the need for deep analytics and ML expertise,
using a low-code/no-code experience.

Business data analysts visualize, build upon, and report on the

analysis done by a data scientist for business users.

Application developers and software

architects who are learning about graph databases often use
graph queries, typically using Cypher query language. They learn
graph algorithms as they seek to analyze patterns across their
graph and use those techniques in their applications. (Graph data
science is often the secret sauce in differentiating applications.)

8
5 Graph Data Science Basics Everyone Should Know

Stay Ahead of the Curve

Commercial applications of graph data science are new and data experts are still coming up to
speed on how to best use graph data science in their organizations. When considering use cases,
data experts and data scientists should keep in mind that graph data science helps answer big
questions like what’s important, what’s unusual, and what’s next. Using this framing, it is easier
to identify opportunities to use graph data science to improve models and make predictions.

Organizations of all sizes, all industries, and within all departments are using graph data science
to make recommendations, identify anomalies and find fraudsters, improve customer knowledge,
and optimize supply chains.

So what will you do with graph data science?

Ready to try it out? Activate a free graph data science sandbox from Neo4j with prepopulated
data for common scenarios.

Read more about popular

Build Recommendation
graph data science use cases:
Engines

Detect Fraud and

Anomalies

Improve Customer
Knowledge

Optimize Supply Chains

Neo4j is the world’s leading graph data platform. We help organizations – including Comcast, ICIJ, NASA, UBS, and Questions about
Volvo Cars – capture the rich context of the real world that exists in their data to solve challenges of any size
and scale. Our customers transform their industries by curbing financial fraud and cybercrime, optimizing global Neo4j? Contact us
networks, accelerating breakthrough research, and providing better recommendations. Neo4j delivers real-time around the globe:
transaction processing, advanced AI/ML, intuitive data visualization, and more. Find us at neo4j.com and follow us
at @Neo4j.
[email protected]
© 2022 Neo4j, Inc. neo4j.com/contact-us