Adding Graphs to Machine Learning for Improved Accuracy, Performance, and Explainability

Representing data as a graph captures information on how data is connected. For example, bank accounts and transactions between them can be modeled as a graph, with the bank accounts as nodes in the graph and the cash transfers between them as edges in the graph. Modeling data as a graph enables the use of graph algorithms to generate embeddings for data. Instead of working with data as isolated entities, these embeddings enable a machine learning model to use information from the topology of the graph, which is based on how a data item is connected to other data items. This can lead to better accuracy and better performance when making predictions. Also, graphs make ML more explainable by providing a score for the importance of each node used in the prediction. Graph ML algorithms such as DeepWalk, Pg2Vec, and GraphWise can be used to generate graph embeddings that can be consumed by downstream models. When a property graph can have data updates in real-time, such as in Oracle Database, the most up-to-date data can be used to generate embeddings. In this session we will talk about how graphs can enhance machine learning, and the importance of using a graph data model that can capture the latest connections.