Big Graph Analytics in Caixabank

Machine Learning is a powerful tool to study and perform predictive analytics. Graph technology superbly complements ML and IA by offering an insight into interconnections. Any relational information of a database based on tables and relationships can be seen in a model based on nodes and edges of a database in graphs. The Graphs also offer exceptional functionality for the Recommendation of Products, the Identification of important people or Influencers, detection of communities and recognition of Graph patterns.

In addition, these graphs allow us to execute algorithms that run the graph in parallel in a very efficient way, such as algorithms for the calculation of Page Rank, Path Finding, Link Prediction,… as well as operations for the creation of subgraphs, filtering, aggregation,… and that also has a query language called PGQL very similar to SQL.

This is the success story of the Caixabank Graphs project and the production of a graph of some 2,000 M of nodes and edges with the aim of improving: the 360º vision of customers, marketing and fraud detection.

Caixabank started this project by carrying out a Benchmark with the objective of choosing the best Graph technology for the project. Later we designed a solution architecture that will integrate with the Caixabank Data Lake and that would be efficient loading data massively as incrementally. We provide Notebook based on Data Science and graphic tools to business analysts. Next, we continue with the Data Science training.

During the course of this talk we will also present some of the use cases started with Fraud Detection, in which we execute queries of the order of x6000 times faster than with other previous technologies, visualizing all the relationships between clients and the contracts of the Bank in addition to much greater analysis detail.

The second use case that was implemented in the Analytical Graph was that of Security. The objective of this department is to investigate and avoid any criminal or dangerous act for our clients or the company, detecting any risk using the appropriate technologies efficiently.
They use an application and cannot scale on two levels of analysis.
The benefits it brings:
Using the Corporate Graph we can:
• Detect and visualize all relationships between affected parties and discover different patterns
• Expand a complete network community to identify all parties involved
• Provide a more intuitive report to people who do not have deep IT knowledge
And it also greatly facilitates administration and operation.

The architecture of the On-line Graph and the use case for real-time fraud detection will also be discussed.
We will continue with the following use case plans to be implemented such as the Money Laundering Prevention and others that will come in the medium term.

Finally, it will end with an exhibition of the benefits of having both Machine Learning and Graph technologies working in a complementary way.