Graph

The graph analytics features enable you to discover how items in an Elasticsearch index are related. You can explore the connections between indexed terms and see which connections are the most meaningful. This can be useful in a variety of applications, from fraud detection to recommendation engines.

For example, graph exploration could help you uncover website vulnerabilities that hackers are targeting so you can harden your website. Or, you might provide graph-based personalized recommendations to your e-commerce customers.

The graph analytics features provide a simple, yet powerful graph exploration API, and an interactive graph visualization app for Kibana. Both work out of the box with existing Elasticsearch indices—you don’t need to store any additional data to use these features.

The graph API provides an alternative way to extract and summarize information about the documents and terms in your Elasticsearch index. A graph is really just a network of related items. In our case, this means a network of related terms in the index.

The terms you want to include in the graph are called vertices. The relationship between any two vertices is a connection. The connection summarizes the documents that contain both vertices' terms.

The graph vertices are simply the terms that you’ve already indexed. The connections are derived on the fly using Elasticsearch aggregations. To identify the most meaningful connections, the graph API leverages Elasticsearch relevance scoring. The same data structures and relevance ranking tools built into Elasticsearch to support text searches enable the graph API to separate useful signals from the noise that is typical of most connected data.

This foundation lets you easily answer questions like:

But what about performance? The Elasticsearch aggregation framework enables the graph API to quickly summarize millions of documents as a single super-connection. Instead of retrieving every banking transaction between accounts A and B, it derives a single connection that represents that relationship. And, of course, this summarization process works across multi-node clusters and scales with your Elasticsearch deployment. Advanced options let you control how your data is sampled and summarized. You can also set timeouts to prevent graph queries from adversely affecting the cluster.

Use Graph to reveal the relationships in your data.

Apply custom colors and icons to vertices, configure the number of vertices that a search adds to the graph, block terms, and more.

Each vertex has a color, icon, and label. To change the color or icon of all vertices of a certain field, click it’s field, and then select Edit settings.

To change the color and label of selected vertices, click the style icon in the control bar.

By default, Graph is configured to tune out noise in your data. If this isn’t a good fit for your data, open Settings > Advanced settings, and then adjust the way Graph queries your data. You can tune the graph to show only the results relevant to you and to improve performance. For more information, see Graph troubleshooting.

You can configure the number of vertices that a search or expand operation adds to the graph. By default, only the five most relevant terms for any given field are added at a time. This keeps the graph from overflowing. To increase this number, click a field, select Edit Settings, and change Terms per hop.

Documents that match a blocked term are not allowed in the graph. To block a term, select its vertex and click the block icon in the graph toolbar. For a list of blocked terms, open Settings > Blocked terms.

With drilldowns, you can display additional information about a selected vertex in a new browser window. For example, you might configure a drilldown URL to perform a web search for the selected vertex term.

Use the drilldown icon in the graph toolbar to show the drilldown buttons for the selected vertices. To configure drilldowns, go to Settings > Drilldowns. See also Disabling drilldown configuration.

Graph uses a "force layout", where vertices behave like magnets, pushing off of one another. By default, when you add a new vertex to the graph, all vertices begin moving. In some cases, the movement might go on for some time. To freeze the current vertex position, click the pause icon in the graph toolbar.