Co-occurrence relation types can be created between two entity types that share a connection to at least one other entity type.

Relation Name

1. Relation Name
Enter the name of the co-occurrence dynamic relation type. 
  • The name should typically end in "Co-occurrence" or "Co-occ"
  • The name will be updated automatically based on the endpoints when you change the Entity 1 and Entity 2 dropdowns.
  • After you edit the relation type name, it will become bold, indicating that further changes to dropdowns will not change it.

Relation Endpoints

2. Relation Endpoints
Select the two entity types that are the endpoints of the co-occurrence.  A co-occurrence is an undirected relation type, so we call the endpoints Entity 1, Entity 2 rather than Src and Dst.
  • Co-occurrences can be created between the same two entity types (two senators) or two different entity types (a state and a bill).
  • When editing a co-occurrence that has already been created you cannot change the endpoints.  To do so the co-occurrence must be deleted and a new one can be created in its place.

Constituent Relations

3. Constituent Relations
The Based on Relations table lets you select which constituent relation types will be used to create the co-occurrence type.
At least one constituent relation check box must be selected.
A co-occurrence is formed by combining constituent relation types. 
In the image above, the "Senator Co-occurrence" has been created by combining "Bill Sponsor Relations" and "Bill Cosponsor Relations". Olympia Snowe and Mary Landrieu are two Senator type entities.  A senator-to-senator relation type was not directly present in any data table.  We create this relationship by connecting two senators with a co-occurrence any time they were both cosponsors of a bill, or when one of them is a sponsor, and the other a cosponsor.
Likewise there is no direct relationship between LA (Louisiana) and a bill sponsored by Mary Landrieu, but we can create that co-occurrence relation by combining the Senator-State and Senator-Bill relationships to form a State-Bill co-occurrence.
  • If Entity Type 1 and Entity Type 2 are the same (two senators), the constituent relations must have one endpoint (Src or Dst) be a Senator, and the other should be a different entity type (a bill).
  • If Entity Type 1 and Entity Type 2 are two different entity types (a bill and a state), the constituent relations must go from Entity Type 1 (a bill) to a third entity type (a senator), and also from the third entity type to Entity Type 2 (a state).
  • There can be more than one or two constituent type relations as long as they connect Entity Type 1 to Entity Type 2 through an intermediate types.  For instance a co-occurrence between senators can incorporate both Senator-Bill relation types and Senator-State relation types.

Dynamic Attributes

There are several dynamic attributes that can be created for dynamic relation types.  By default only Cooccurrence # and C-Rank are created.  The others need to be manually selected in the Dynamic Attributes dialog.
  • Co-occ #:  Co-occurrence # is the number of entities which make up the other endpoints of constituent relations emanating from the co-occurrences endpoints.  For instance, for a co-authorship co-occurrence between two authors we would count the number of documents that the two authors have written together, which would be their co-occurrence #.
  • C-Rank: C-Rank is a rank of co-occurrences by Co-occurrence #.  A rank of 1 indicates the strongest relationship and higher values of C-Rank indicate a weaker relationship.  For a co-occurrence between entities A and B, the C-Rank is computed by sorting co-occurrences between A and all entities of type B and looking at the order in which our co-occurrence appears.  We then compare this to the sorted list of co-occurrences for B and all entities of type A.  C-Rank is the minimum of these two numbers.
  • Signif: Significance is probability that A and B would have the co-occurrence # that they do divided by the expected value of the overlap between two entities having the same number of constituent relations as A and B but occurring at random.  I.e. it is the importance of the overlap rather than the size of the intersection.  For instance, if we have a list of 100 documents, and two authors which have written 50 each, then by random chance the authors should have written 25 together.  If they have co-authored more than 25 documents then the significance has a positive value.  If the two authors had each written 25 documents documents to begin with, and they again have a co-occurrence of 25 out of a pool of 100, then the significance of them having co-authored 25 documents is this case is much larger when they had written 50 documents each.
  • S-Rank: S-Rank is similar to C-Rank except it is computed by sorting co-occurrences by significance rather than by co-occurrence #.

Ok, Cancel

  • OK: Add the co-occurrence dynamic relation type or apply changes to the relation type and return to the Dynamic Relations page.
  • Cancel: Discard changes and return to the Dynamic Relations page.