Is Graph Data Science the Secret to Accelerating Machine Learning?

In the past decade, the field of machine learning has developed by leaps and bounds and has become one of the most intelligent frontier fields of artificial intelligence. With the continuous increase in the demand for data analysis in various industries in the era of big data, the volume of data has grown unprecedentedly, and new types of data are constantly emerging. Machine learning is increasingly moving towards the direction of intelligent data analysis.

Graph data science has a surprisingly long history as an effective means of processing complex information: it was first proposed in the 18th century by mathematical genius Leonhard Euler, and has become particularly important more recently as Google revolutionized search using a graph-based approach to page ranking.

Today, graph technology is no longer the exclusive secret of network pioneers with internal expertise and resources. In the past, only leading companies with well-trained R&D teams had the ability to handle large amounts of connected data. Today, any organization that values ​​the value of data can use this powerful innovation to discover connections between data through unique algorithms and embeddings.

Graph-based data work is rapidly becoming mainstream in business. As a core part of the toolbox of enterprise data scientists, graph data science will become an important feature of the next decade. Gartner pointed out in the "Top 10 Data and Analysis Technology Trends in 2021" report: "By 2025, graph technology will be used for 80% of data and analysis innovations, which is higher than 10% in 2021. Graph technology will facilitate rapid decision-making throughout the organization."

Gartner has also previously surveyed some companies on the use of artificial intelligence and machine learning technologies. As many as 92% of respondents said they plan to adopt graph technology within five years. Academic research focusing on this field is also increasing, with more than 28,000 peer-reviewed scientific papers on graph-driven data science published in recent years.

Leveraging associations to make more accurate and predictable predictions  

Enterprises are accelerating their adoption of graph data science, a powerful innovation that uses graph algorithms to reason about the “shape” of the context of each piece of data.

Why would a developer want to know this? It’s because graph data science enables better and richer machine learning predictions. Graph data science is revolutionizing the way businesses make predictions in different scenarios, from fraud detection to tracking customers or patients, by leveraging the associations between data nodes to achieve more accurate and predictable predictions. In the drug discovery use case, it means finding new possible associations between genes, diseases, drugs, and proteins, while providing adjacent context to evaluate the relevance or validity of any such findings. For customer recommendations, it means learning from the customer journey, making accurate recommendations for future purchases, and building confidence in relevant recommendations by showing historical purchase records.

This ability to quickly “learn” generalized, predictive features from data is enabling enterprises to take machine learning to a whole new level. While some teams are still learning how to leverage connected data in existing machine learning workflows, the number of real-world use cases is growing rapidly. Graph technology adopters are finding that graph technology empowers them in everything from querying to support domain experts in discovering patterns to identifying high-value features to train machine learning models.

Emerging Graph Technology Success Stories  

Let's look at some examples of the above trends. In Europe, graph data science is already being used by government departments, and data scientists have deployed the first machine learning model built with graph technology. The resulting system automatically recommends content from government online resources to users based on the pages they visit. The application displays continuous features of nodes and uses them for various machine learning tasks, such as content recommendations.

A government data scientist noted, “Through this process, we learned that creating the underlying data to support model training and deployment was the most time-consuming part.” In another area of ​​the graph database ecosystem, a senior data scientist from Meredith, a leading media and marketing services company, noted that the use of graph algorithms allows billions of page views to be converted into millions of pseudonymous identifiers with rich browsing profiles: “Delivering relevant content to online users who are not authenticated is critical to our business… Instead of ‘advertising in the dark,’ we are now better able to understand our customers, which will not only significantly increase revenue but also provide better services to consumers.”

Graph data science can also support the medical supply chain. Boston Scientific, a global medical device manufacturer, uses graph data science to find the cause of product failures. In this case, multiple teams in different countries and regions often have to work together to solve the same problem in parallel, and engineers must analyze data in different spreadsheets. This creates inconsistencies and makes it difficult to find the root cause of the problem. Boston Scientific said that turning to graph technology provides a more effective way to analyze, coordinate and improve manufacturing processes across all regions of the company.

Now, users can conduct meaningful, science-enhanced data searches. The analysis query time has been reduced from two minutes to 10 to 55 seconds, an improvement that helps improve overall efficiency and simplify the analysis process. Specific links that are more likely to fail can be identified. Another benefit is that the graph data model is very simple and easier to communicate. "Everyone involved in the project, from business stakeholders to technical implementers, is able to understand each other because they all speak the same language," said Eric Wespi, a data scientist at the company. The company generates higher business value by using natural language processing to analyze the raw text of the failure in detail, extracting and correlating topics to investigate the root cause of the failure.

At Caterpillar, an international manufacturing leader, graph data science enables more efficient natural language processing of a large database of repair technical documentation. Faced with valuable data captured in more than 27 million documents but not accessible, the company set out to create a processing tool that could reveal underlying connections and trends. The resulting graph-based machine learning classification tool learns from portions of the data that have been annotated with terms such as “reason” or “complaint” and can be applied to other data. It parses the text on its own and quickly finds patterns and connections, builds hierarchies, and adds ontologies.

Enhanced Insight

Another example of graph data science being applied is in healthcare. The analytics team at New York-Presbyterian Hospital used graph technology to track infections and take strategic action to control them. Its developers found that graph data science gave them a flexible way to connect all the dimensions of an event—the “what,” “when,” and “where” it occurred. With this insight, the team created a “time” and “space” tree to model all the patients in the wards being treated on site. This initial model revealed a large number of interrelationships, but that alone would not meet the project’s goals. By connecting the time and location trees with an event entity, the resulting data model meant the analytics team was able to analyze everything that happened in the model and proactively identify and control illness before it spread.

It is indisputable that graph data science will become a key part of business analytics and provide useful business insights after 2021. Gartner's data industry team predicts that a quarter of the Fortune 1000 companies will include graph technology in their plans for advanced data processing and analysis within three years.

There is no doubt that the application of graph data science has far exceeded the 18th century and entered the business field. It is time to explore the huge potential of graph data science to provide solutions to business problems. As a pioneer and leader in graph data science, Neo4j has helped many institutions and organizations accelerate the development of machine learning to intelligent analysis through graph data science to make predictions that drive business growth.