At Google Cloud Next, we announced the preview of graph algorithms with Spanner Graph, bringing Google Research’s state-of-the-art graph mining capabilities natively to your database. These graph intelligence capabilities can help you derive valuable insights from graph data faster, cheaper, and at scale.
Enterprises are increasingly leveraging graph technologies to uncover complex relationships in data for use cases such as fraud detection, social network analysis, entity resolution, and healthcare research. Graph algorithms, such as node centrality and community detection, are the computational methods used to analyze these structures, and work by quantifying the patterns and strength of connections between entities. However, running graph algorithms at scale has historically been challenging and resource-intensive, often requiring complex ETL pipelines to dedicated analytic solutions or risking the transactional performance of the graph database.
We designed Spanner Graph algorithms to tackle demanding enterprise workloads without compromising on the performance of your operational database. This architecture provides several distinct advantages:
-
Tight integration with GQL: Directly invoke algorithms using ISO Graph Query Language (GQL) to run structural analytics across your data. By sequentially weaving algorithms and standard queries together, Spanner Graph minimizes complex data movement to external engines, simplifying your architecture and accelerating time-to-insight.
-
Near-zero transactional impact and lower TCO: Algorithm execution happens on dedicated compute resources, so as not to impact live production traffic. Spanner automatically provisions resources and securely routes data via Data Boost without having to create a custom ETL pipeline. Pay only for what you use, avoiding expensive licensing and operational overhead of legacy solutions.
-
Global insights on billion-edge graphs in minutes: Built for scale and speed, our engine can run algorithms on graphs with tens of billions of edges within minutes. Encoding topologies in a dense format that’s optimized for random access enables high-performance structural analytics on massive datasets.
While Google Research has published several research papers, held workshops, and released open-source projects based on its graph mining tools (e.g., for multi-core clustering), this is the first time that they are widely available to Google Cloud customers. Let’s take a deeper look at graph algorithms, and how you can use them with Spanner Graph.
Algorithms: Deeper insights for connected data
When we first launched Spanner Graph, our goal was to reimagine graph data management with a native graph database experience within Spanner, Google’s highly scalable, distributed database. Spanner Graph unifies relational and graph models, allowing developers to query connected data using the ISO GQL, while also interoperating with Spanner’s existing tabular, search, and vector capabilities. This allows you to build intelligent applications without creating complex data pipelines, duplicating data, or increasing security and governance risk.
Building on this foundation, Spanner Graph algorithms help you to extract even deeper insights from your connected data. Graph algorithms analyze the relationships and connections within data, revealing hidden patterns and insights that might be missed with traditional analytical methods. With this launch, you can analyze connectedness to, for example, detect fraud rings, conduct clustering for entity resolution, identify points of failure in complex networks, or recommend products based on the preferences of connected users.
We use graphs extensively at Google. In fact, many popular algorithms like PageRank, the foundational technology that powers Google Search, were invented here. With native algorithm support in Spanner Graph, we are bringing some of Google’s leading graph intelligence capabilities directly to Google Cloud customers, with a set of essential graph algorithms that help you easily uncover the hidden structures within your data:
-
Centrality: Pinpoint the most influential and central nodes within your network using betweenness centrality, closeness centrality, and PageRank.
-
Community detection: Automatically group highly connected entities to uncover hidden segments with label propagation, correlation clustering, modularity clustering, weakly connected components, and clique aggregator.
-
Similarity and path finding: Find optimal routes using set-to-set shortest paths, or measure node similarities using Jaccard, cosine, common neighbors, and total neighbors.
An integrated developer experience
You can invoke graph algorithms directly using GQL on the entire graph, subgraphs, or a select set of nodes and edges. Spanner offers an integrated workflow: results from graph algorithm runs can be written directly back to Spanner Graph. This lets you invoke algorithms and standard queries sequentially, using the output of one operation as input to the next. Additionally, you can also store results in Cloud Storage buckets.
Example: Uncovering the ringleader of a fraudulent network
Consider a scenario where you are analyzing financial transactions to combat money laundering. Fraudsters usually manipulate a set of “mule” accounts (intermediary accounts for money laundering) that interact with one another to collectively commit fraud. To capture the teamwork between detected and hidden mule accounts, anti-fraud experts usually resort to link analysis and community detection graph algorithms. Here’s how you can use algorithms and queries together in Spanner Graph to catch them.
Step 1: Identify communities of accounts (algorithm)
First, we apply a modularity clustering algorithm to cluster accounts into communities. We then write the resulting community_id directly back to the Account in Spanner Graph.
