Graph analytics is a powerful tool for financial crime detection, but choosing the right combination of techniques is key to success. There are two main techniques in graph analytics:
- Entity resolution and network filtering
- Graph representation learning (GRL)
Entity resolution, network filtering, and social connections
Entity resolution and network filtering are important for financial institutions with multiple customer and transaction systems. Entity resolution removes duplicate data records, while network filtering eliminates unnecessary connections.
Entity resolution can sometimes be challenging because real datasets can be imperfect. For example, twins could have the same initial, surname, and date of birth, but they are not the same person. In cases like this, network filtering can help by creating a “social network” of connections among different entities like people, addresses, companies, and bank accounts.
Filtering networks down to those representing social connections requires tuning to identify strong social connections. It can sometimes be challenging to identify all possible social relationships. Enter graph representation learning.
Graph representation learning, ‘nodes,’ and ‘edges’
Graph representation learning (GRL) is a deep-learning technique that analyzes transaction graphs. It uses “nodes” to represent customers, merchants, and transactions, and “edges” to connect customers to their transactions. GRL algorithms like GraphSAGE aggregate features to create a learned representation. This can be combined with raw features to improve accuracy in financial crime detection, such as card or payments fraud.
Understanding and overcoming the challenges
Implementing graph analytics can involve challenges. For example:
- Over-smoothing can occur when every transaction in a learned graph representation looks similar. But entity resolution and network filtering can help by identifying risky connections in complex business scenarios.
- Graph analytics can be resource intensive. But recent breakthroughs like GraphSAGE have reduced hardware requirements and costs.
- GRL can be difficult to interpret due to its deep-learning nature. But entity resolution and network filtering offer advanced visualization tools for investigators.
To maximize the benefits of graph analytics, consider the profile of your business, the complexity of relationships, and the investigative user journey. Low-volume organizations benefit from entity resolution and network filtering, while high-volume ones should consider GRL. Understanding relationship complexity helps choose the right technique. Investigators need tools that provide all relevant information in one place for efficient decision-making.
Graph analytics is a powerful tool for financial crime detection. By understanding the techniques and selecting the appropriate combination for your organization, you can enhance your anti-money laundering program and reduce financial crime risk. Embracing GRL can also improve the customer experience.
Contact SymphonyAI Sensa-NetReveal to learn more about our case management solutions and the benefits of graph analytics in your institution.
About the Author
Thomas Saminaden, a senior business manager at SymphonyAI Sensa-NetReveal, has extensive experience in graph analytics for financial crime detection. He is currently researching temporally aware graph representation learning techniques for card fraud detection as part of his artificial intelligence master’s program.
