Efficient calculation of node proximity on graphs with side information

1. A computer method for evaluating data, the method comprising performing operations on at least one data processing device, the operations comprising:

maintaining an embodiment of a weighted, directed graph representation of the data on a machine readable medium, the embodiment including representations of nodes, links, link weights, and random walk results indicating proximity of nodes to one another;

receiving indications of a source and target node within the graph;

receiving side information indicating positive and/or negative feedback relating to nodes of the graph; and

providing a machine readable embodiment of an indication of an updated proximity between the source and target node, responsive to the side information, without recalculating random walk results for the entire graph, the indication being embodied on a medium.

2. The method of claim 1 , wherein

the graph is maintained in matrix form embodied in the medium;

random walk results comprise

at least one pre-computed low rank approximation for a normalized adjacency matrix resulting from a singular value decomposition; and

a matrix inversion result.

3. The method of claim 2 , wherein providing comprises:

responsive to each negative node and a threshold neighborhood size, computing at least one respective intermediate matrix value;

responsive to positive nodes and at least one scaled column of the adjacency matrix, computing at least one respective further intermediate matrix value; and

responsive to the intermediate and further intermediate matrix values, updating the low rank approximation and the matrix inversion result, so that a full recomputation of the inversion result is not necessary.

4. The method of claim 1 , wherein

the graph is maintained in matrix form embodied in the medium;

random walk results comprise

at least one pre-computed low rank approximation;

at least one original matrix of ranking vectors; and

at least one original matrix inversion result; and

providing comprises

updating the low rank approximation and the matrix inversion result without a complete re-inversion operation; and

performing intermediate value matrix-vector multiplications.

5. The method of claim 4 , wherein updating comprises:

creating a revised low rank approximation responsive to negative nodes, partial influence of positive nodes and the pre-computed low rank approximation; and

creating a revised matrix inversion result, responsive to the negative nodes, partial influence of positive nodes, the original matrix inversion result; and the revised low rank approximation; and

the matrix-vector multiplications compute proximity and fix a remaining influence of positive nodes using the revised low rank approximation and the revised matrix inversion result.

6. The method of claim 1 , wherein the indication of proximity comprises a machine readable embodiment of a ranking vector.

7. The method of claim 1 , wherein the indication of proximity is presented over a network as an indication of relationship between content.

8. The method of claim 1 , wherein the indication of proximity is presented over a network as a ranking of content.

9. A system comprising:

at least one data processing device;

at least one network and/or user interface device for communicating with the data processing device;

at least one medium for embodying at least machine performable code and data in machine readable form; the code comprising instructions for causing the data processing device to perform operations on the data, the operations comprising:

maintaining an embodiment of a weighted, directed graph representation of the data on a machine readable medium, the embodiment including representations of nodes, links, link weights, and random walk results indicating proximity of nodes to one another;

receiving indications of a source and target node within the graph;

receiving side information indicating positive and/or negative feedback relating to nodes of the graph; and

providing a machine readable embodiment of an indication of an updated proximity between the source and target node, responsive to the side information, without recalculating random walk results for the entire graph, the indication being embodied on a medium.

10. The system of claim 9 , wherein

the graph is maintained in matrix form embodied in the medium;

random walk results comprise:

at least one pre-computed low rank approximation;

at least one original matrix of ranking vectors; and

at least one original matrix inversion result; and

providing comprises:

updating the low rank approximation and the matrix inversion result without a complete re-inversion operation; and

performing intermediate value matrix-vector multiplications.

11. The system of claim 10 , wherein updating comprises:

creating a revised low rank approximation responsive to negative nodes, partial influence of positive nodes and the pre-computed low rank approximation; and

creating a revised matrix inversion result, responsive to the negative nodes, partial influence of positive nodes, the original matrix inversion result; and the revised low rank approximation; and

the matrix-vector multiplications compute proximity and fix a remaining influence of positive nodes using the revised low rank approximation and the revised matrix inversion result.

12. The system of claim 11 , wherein the indication of proximity comprises a machine readable embodiment of a ranking vector.

13. The system of claim 9 , wherein the indication of proximity is presented over a network as an indication of relationship between content.

14. The system of claim 9 , wherein the indication of proximity is presented over a network as a ranking of content.

15. A computer program product for performing operations, the computer program product comprising at least one storage medium readable by a processing circuit and storing instructions to be run by the processing circuit for performing a method comprising:

maintaining a machine readable embodiment of a weighted, directed graph representation of the data, the embodiment including representations of nodes, links, link weights, and random walk results indicating proximity of nodes to one another;

receiving indications of a source and target node within the graph;

receiving side information indicating positive and/or negative feedback relating to nodes of the graph; and

providing a machine readable embodiment of an indication of an updated proximity between the source and target node, responsive to the side information, without recalculating random walk results for the entire graph.

16. The program product of claim 15 , wherein

the graph is maintained in matrix form embodied in the medium;

random walk results comprise:

at least one pre-computed low rank approximation;

at least one original matrix of ranking vectors; and

at least one original matrix inversion result; and

providing comprises:

updating the low rank approximation and the matrix inversion result without a complete re-inversion operation; and

performing intermediate value matrix-vector multiplications.

17. The program product of claim 16 , wherein updating comprises:

creating a revised low rank approximation responsive to negative nodes, partial influence of positive nodes and the pre-computed low rank approximation; and

creating a revised matrix inversion result, responsive to the negative nodes, partial influence of positive nodes, the original matrix inversion result; and the revised low rank approximation; and

the matrix-vector multiplications compute proximity and fix a remaining influence of positive nodes using the revised low rank approximation and the revised matrix inversion result.

18. The program product of claim 15 , wherein the indication of proximity comprises a machine readable embodiment of a ranking vector.

19. The program product of claim 15 , wherein the indication of proximity is presented over a network as an indication of relationship between content.

20. The program product of claim 15 , wherein the indication of proximity is presented over a network as a ranking of content.