Connection sub-graphs in entity relationship graphs
Techniques are described herein for a connection sub-graph generator that identifies important relationships between entities in entity relationship graphs. The generator first identifies two entities in an entity relationship graphs. A plurality of candidate path relationships between the two entities may be determined. For each such candidate path relationship, the generator determines a path type based on types of edge relationships that make up the candidate path relationship and generates a score value for the candidate path. This score value is based in part on the path type of the candidate path relationship. One or more candidate path relationships that have score values that indicate a high likelihood of interest to a user is selected to be included in a connection sub-graph.
1. A method comprising:
establishing a first entity and a second entity in an entity relationship graph, wherein the entity relationship graph comprises a set of entities interconnected through a plurality of edge relationships and wherein the first entity and the second entity are included in the set of entities;
assigning an initial activation value to the first entity;
creating an activation value distribution in the entity relationship graph by spreading the initial activation value assigned to the first entity to other entities in the entity relationship graph;
determining an activation value at the second entity in the activation value distribution;
determining a plurality of candidate path relationships between the first entity and the second entity, wherein each of the plurality of candidate path relationships terminates at both the first entity and the second entity and comprises one or more edge relationships in the plurality of edge relationships;
for each candidate path relationship in the plurality of candidate path relationships:
determining a path type that is associated with a candidate path relationship; and
generating a score value for the candidate path relationship based in part on the path type that is associated with the candidate path relationship;
wherein the score value represents a likelihood that the candidate path relationship is of interest to a user; and
selecting, in the plurality of candidate path relationships, one candidate path relationship having a score value that satisfies one or more first criteria;
wherein the plurality of candidate path relationships is identified as one or more path relationships whose contributions to the activation value at the second entity satisfy one or more second criteria;
wherein the method is performed by one or more computing devices.
2. A method as recited in claim 1 , wherein the first entity and the second entity are specified in a query issued by the user.
3. A method as recited in claim 1 , wherein the entity relationship graph is a candidate entity relationship graph that is reduced from an original entity relationship graph.
4. A method as recited in claim 1 , wherein an entity in the entity relationship graph is of one of a plurality of entity types.
5. A method as recited in claim 1 , wherein an edge relationship in the entity relationship graph is of one of a plurality of edge relationship types.
6. A method as recited in claim 1 , wherein a candidate path relationship in the plurality of candidate path relationships is of a path type in the form of a tuplet that comprises one or more edge relationship types and wherein each edge relationship type in the one or more edge relationship types is a type of an associated edge relationship in the one or more edge relationships that forms the candidate path relationship.
7. A method as recited in claim 1 , wherein the score value for the candidate path relationship includes a first measure that indicates how strongly the first entity and the second entity are connected through the candidate path relationship.
8. A method as recited in claim 1 , wherein the score value for the candidate path relationship includes a second measure that indicates how rare the candidate path relationship is in the entity relationship graph.
9. A method as recited in claim 1 , wherein the score value for the candidate path relationship includes a third measure that indicates how prestigious one or more entities along the candidate path are.
10. A computer-readable storage medium carrying one or more sequences of instructions which, when executed by one or more processors, causes the one or more processors to perform the method recited in claim 1 .
11. A computer-readable storage medium carrying one or more sequences of instructions which, when executed by one or more processors, causes the one or more processors to perform the method recited in claim 2 .
12. A computer-readable storage medium carrying one or more sequences of instructions which, when executed by one or more processors, causes the one or more processors to perform the method recited in claim 3 .
13. A computer-readable storage medium carrying one or more sequences of instructions which, when executed by one or more processors, causes the one or more processors to perform the method recited in claim 4 .
14. A computer-readable storage medium carrying one or more sequences of instructions which, when executed by one or more processors, causes the one or more processors to perform the method recited in claim 5 .
15. A computer-readable storage medium carrying one or more sequences of instructions which, when executed by one or more processors, causes the one or more processors to perform the method recited in claim 6 .
16. A computer-readable storage medium carrying one or more sequences of instructions which, when executed by one or more processors, causes the one or more processors to perform the method recited in claim 7 .
17. A computer-readable storage medium carrying one or more sequences of instructions which, when executed by one or more processors, causes the one or more processors to perform the method recited in claim 8 .
18. A computer-readable storage medium carrying one or more sequences of instructions which, when executed by one or more processors, causes the one or more processors to perform the method recited in claim 9 .