Merge pull request #1979 from moj-analytical-services/graph_definition

Add graph definition to docs
moj-analytical-services · Mar 28, 2024 · ae85c13 · ae85c13
2 parents d1acc1d + 7347fe6
commit ae85c13
Show file tree

Hide file tree

Showing 7 changed files with 24 additions and 0 deletions.
diff --git a/docs/img/clusters/basic_graph.drawio.png b/docs/img/clusters/basic_graph.drawio.png
diff --git a/docs/img/clusters/basic_graph_cluster.drawio.png b/docs/img/clusters/basic_graph_cluster.drawio.png
diff --git a/docs/img/clusters/basic_graph_cluster_person.drawio.png b/docs/img/clusters/basic_graph_cluster_person.drawio.png
diff --git a/docs/img/clusters/basic_graph_records.drawio.png b/docs/img/clusters/basic_graph_records.drawio.png
diff --git a/docs/img/clusters/graph.png b/docs/img/clusters/graph.png
diff --git a/docs/topic_guides/theory/linked_data_as_graphs.md b/docs/topic_guides/theory/linked_data_as_graphs.md
@@ -0,0 +1,23 @@
+# Linked data as graphs
+
+A graph is a collection of points (often referred to as nodes or vertices) connected by lines (often referred to as edges).
+
+![Basic Graph](../../img/clusters/basic_graph.drawio.png){:width="80%"}
+
+Then a group of interconnected nodes is referred to as a **cluster**.
+
+![Basic Cluster](../../img/clusters/basic_graph_cluster.drawio.png){:width="80%"}
+
+Graphs provide a natural way to represent linked data, where the nodes of a graph represent records being linked and the edges represent the links between them. So, if we have 5 records (A-E) in our dataset(s), with links between them, this can be represented as a graph like so:
+
+![Basic Graph - Records](../../img/clusters/basic_graph_records.drawio.png){:width="80%"}
+
+When linking people together, a cluster represents the all of the records in our dataset(s) that refer to the same person. We can give this cluster a new identifier (F) as a way of referring to this single person.
+
+![Basic Person Cluster](../../img/clusters/basic_graph_cluster_person.drawio.png){:width="80%"}
+
+!!! note
+
+    For clusters produced with Splink, every edge comes with an associated Splink score (the probability of two records being a match). The clustering threshold (`match_probability_threshold`) supplied by the user determines which records are included in a cluster, as any links (edges) between records with a match probability below this threshold are excluded.
+
+    Clusters, specifically cluster IDs, are the ultimate output of a Splink pipeline.
diff --git a/mkdocs.yml b/mkdocs.yml
@@ -138,6 +138,7 @@ nav:
         - Why do we need record linkage?: "topic_guides/theory/record_linkage.md"
         - Probabilistic vs Deterministic linkage: "topic_guides/theory/probabilistic_vs_deterministic.md"
         - The Fellegi-Sunter Model: "topic_guides/theory/fellegi_sunter.md"
+        - Linked data as graphs: "topic_guides/theory/linked_data_as_graphs.md"
       - Linkage Models in Splink:
         - Splink's SQL backends - Spark, DuckDB etc:
           - Backends overview: "topic_guides/splink_fundamentals/backends/backends.md"