Graph Data Science 101: Using Graph Data Science in the Real World

5 min readNov 27, 2020

Today’s most pressing data challenges center around connections, not just tabulating discrete data. The ability for graph data science (GDS) to uncover and leverage network structure drives a range of use cases from fraud prevention and targeted recommendations to personalized experiences and drug repurposing.

We can’t overstate the impact of improved graph techniques such as new algorithms or the efforts of applied network scientists such as within computational biology. We don’t want you to overlook societal projects that use graphs, either. However, we believe that the recent explosion of graphs in the business world represents a shift in accessibility and opportunity to drive the democratization of graphs for everyone.

Graph technologies help organizations with many practical use cases across industries and domains. In the past, many businesses began exploring graph technology to create a 360-degree view of their customers or to unify master data, including customer, product, supplier, and logistics information. They may use this kind of tracking to improve customer experience or to meet compliance regulations of recent privacy acts such as the EU’s General Data Protection Regulation (GDPR) and the California Consumer Privacy Act (CCPA). This same kind of complete view and data lineage in graphs is also now used to understand and track data used in machine learning (ML) for more responsible artificial intelligence (AI) applications.

Today, businesses are just as likely to look at using graphs specifically for data science as they recognize the predictive power of relationships, the ability to use network structures to improve their ML, and their own need to innovate. The sections in this chapter highlight a few GDS use cases in areas of accelerating growth and significant commercial interest.

Looking at Graphs in Healthcare

It’s easy to see how any industry with biological roots would naturally comprehend the importance of interconnected systems. You can see this relationship in computational biology as well as healthcare and life sciences in how they view challenges as part of larger processes. Two examples stand out for serving health and commercial interests: more efficient drug discovery and better patient outcomes.

Discovering more efficient drugs

Safety, speed, and costs are paramount in making new drug solutions accessible. Graphs can help tackle the complexity of intertwined relationships between diseases, genes, drugs, side effects, and demographics — to name just a few considerations.

One impressive knowledge graph in the life sciences industry integrates over 50 years of biomedical data that includes genes, compounds, diseases, and other information such as symptoms and side effects. One of the projects from the graph predicts new uses for drugs by using the graph topology. The graph helps predict new uses for currently approved drugs by evaluating relationships, network structures, and similarities. Drug repurposing significantly reduces costs and time to market compared to developing and testing new drugs — not to mention the benefit of having more real-world information available about side effects and unexpected results when a drug is already in use.

Improving the patient journey

Another area of emerging interest is the use of graphs for mapping, evaluating, and improving patient journeys. When a patient doesn’t feel well, many factors are in play that may have evolved over a period of time. Likewise, treatments are rarely a single event, especially for chronic or serious illnesses. The tree of possible symptoms, visits, test, care givers, treatment plans, outcomes, and then secondary tests and treatments and so on can branch out into an immense number of possible paths. Imagine the patient treatment options that can be mapped with a graph to better see the sequence alternatives and path splits after each and every test result or visit. In fact, researchers and healthcare providers already employ graphs to better understand what influences patient journeys so they can improve individual outcomes as well as create and compare to optimal paths.

Recommendations and Personalized Marketing

Making relevant product and service recommendations requires correlating product, customer information, historic behavior, inventory, supplier, logistics, and even social sentiment data. Graph-powered recommendations and targeted marketing help companies provide more appropriate services and experiences to a wider range of users. For example, graph community detection algorithms are used to group customers with interactions or similar behavior for more relevant recommendations. Research shows that graph-enhanced ML can predict customer churn, for example, for uses such as targeted prevention or marketing.

Graph analytics are also used to help target offers to online users that are anonymous in name and demographics but not in site behavior. Insights from analysis performed offline are typically rolled into decision models used in production for real-time recommendations, which can include recommendations for products that ship faster based on shifting stock levels or instantly incorporating data from the customer’s current visit.

Fraud Detection

The amount of money lost to fraud each year is growing, despite increased use of AI and ML to detect and prevent it. To uncover more fraud while avoiding costly false positives, organizations look beyond individual data points to the connections and patterns that link them. Organizations use the network structure to augment existing ML pipelines as a practical approach to increase the amount of fraud detected and recovered.

Graph feature engineering allows businesses to extract predictive elements based on graph queries or algorithms and use that information to train ML models. Improving the predictive accuracy in fraud detection even small percentage points can result in tens of millions of dollars saved in just a few months. GDS enables companies to stay ahead of the ever-shifting patterns of fraud as well as recover more losses.

Head to “Read More” where we give you a detailed example of detecting fraud with GDS.