Method for detecting link spam in hyperlinked databases

• US 7,509,344 B1
• Filed: 08/18/2004
• Issued: 03/24/2009
• Est. Priority Date: 08/18/2003
• Status: Active Grant

• Current Assignee: Google LLC (Alphabet Inc.)
• Original Assignee: Google Inc. (Alphabet Inc.)
• Inventors: Kamvar, Sepandar D., Jeh, Glen M., Haveliwala, Taher H.
• Primary Examiner(s): Fleurantin; Jean B
• Application Number:
  US10/921,381
• Time in Patent Office: 1,679 Days
• Field of Search: 707 1- 10, 707100-1041, 707200-206, 709/218, 709/500.1, 709/215, 715/500.1
• US Class Current: 1/1
• CPC Class Codes: G06F 16/31 Indexing; Data structures therefor; Storage structures
• CPC Class Codes: G06F 16/951 Indexing; Web crawling techniques
• CPC Class Codes: G06F 16/9566 URL specific, e.g. using aliases, detecting broken or misspelled links
• CPC Class Codes: Y10S 707/99936 Pattern matching access
• CPC Class Codes: Y10S 707/99945 Object-oriented database structure processing
• CPC Class Codes: Y10S 707/99948 Application of database or data structure, e.g. distributed, multimedia, or image

Abstract

Methods for facilitating the identification of link spamming in a linked database include calculating a spam likelihood value for nodes in a directed graph of linked nodes are disclosed. The spam likelihood value is computed from an importance of the node and a derivative value of the importance function with respect to a coupling factor. The likelihood that the node'"'"'s importance is inflated by link spam is estimated by calculating the ratio of the magnitude of the derivative value for the node to the rank for the node. Alternatively, the spam likelihood may be computed directly from a component of the principal eigenvector of A evaluated at two values of the parameter c. The normalized derivative value can also be used to provide an order of importance in a list of nodes.

Claims

1. A computer-implemented method for identifying nodes that are beneficiaries of node importance inflating links in a directed graph of linked nodes, wherein the directed graph of linked nodes corresponds to a linked database, and wherein the nodes correspond to documents within the linked database, the method comprising:

computing, for each of at least a subset of the nodes in the directed graph, a respective quantity corresponding to a derivative of a node importance function;

for each node in the subset, comparing the respective computed quantity with a threshold; and

identifying at least a portion of the subset for which the comparison produces a predefined result.

2. The method of claim 1, further including normalizing the respective quantity in accordance with a respective rank.

3. The method of claim 1, wherein the computing computes the derivative of the node importance function with respect to a coupling factor.

4. The method of claim 3, wherein the computing includes:

computing at least two intermediate quantities of the derivative of the node importance function for each of the set of nodes, each of the intermediate quantities computed using a respective value of the coupling factor; and

averaging the intermediate quantities to produce the computed respective quantity.

5. The method of claim 3, wherein the coupling factor represents how tightly coupled the linked nodes are in the directed graph.

6. The method of claim 3, wherein the coupling factor is inversely related to a probability that a user will randomly traverse the directed graph.

7. The method of claim 3, further including:

generating an adjusted importance for at least one node of the portion.

8. The method of claim 7, wherein the generating adjusts, for the at least one node, the importance by an amount proportional to the respective computed quantity.

9. The method of claim 3, further including:

removing at least one node of the identified portion from the directed graph.

36. The method of claim 1, wherein the computing includes:

calculating a matrix A(c) representing the directed graph, the matrix A(c) being an N×

N matrix that is a function of a link coupling coefficient c, where N is a number of nodes in the directed graph;

calculating a principal eigenvector of A(c), denoted x(c);

calculating a derivative with respect to c of x(c), denoted x′

(c); and

for at least one node, calculating from x′

(c) and a rank for the node, an inflation likelihood value S.

37. The method of claim 36, wherein calculating A(c) comprises calculating A(c)=(cP+(1−

c)E)^T, where P is an N×

N matrix and E is an N×

N matrix.

38. The method of claim 37, wherein elements of P represent link coupling strengths between nodes of the linked database, and wherein elements of E represent link decoupling strengths between nodes of the linked database.

39. The method of claim 37, wherein calculating x′

(c) comprises computing x′

(c)=(I−

cP^T)<sup>−

1</sup>(P−

E)^Tx(c).

40. The method of claim 39, wherein computing x′

(c)=(I−

cP^T)<sup>−

1</sup>(P−

E)^Tx(c) is performed using a Jacobi relaxation technique.

41. The method of claim 39, wherein computing x′

(c)=(I−

cP^T)<sup>−

1</sup>(P−

E)^Tx(c) is performed using a Gauss-Seidel technique.

42. The method of claim 36, further comprising comparing the inflation likelihood value S with a predetermined spam threshold value.

43. The method of claim 36, wherein the rank for the node is computed using a link-based ranking technique.

44. The method of claim 43, wherein the link-based ranking technique is a technique selected from the group consisting of counting in-links to each node, calculating a principal eigenvector of a link database matrix A, and calculating a singular value decomposition of a link database matrix A.

45. The method of claim 36, wherein calculating the inflation likelihood value S comprises dividing the absolute value of a component of x′

(c) by the rank for the node.

46. The method of claim 36, wherein calculating the inflation likelihood value S comprises calculating several inflation likelihood values corresponding to various values of c.

47. The method of claim 36, wherein calculating the inflation likelihood value S further comprises averaging the several inflation likelihood values.

10. A computer-implemented method for selecting nodes in a directed graph of linked nodes, wherein the directed graph of linked nodes corresponds to a linked database, and wherein the nodes correspond to documents within the linked database, the method comprising:

computing, for each of at least a portion of the nodes in the directed graph, a respective quantity for the respective node corresponding to a derivative of a node importance function; and

selecting a predetermined percentage of nodes in accordance with the respective quantities.

11. The method of claim 10, wherein the computing computes the derivative of the node importance function with respect to a coupling factor.

12. The method of claim 10, wherein the subset is determined in response to a search request.

13. A computer program product for use with a computer system, the computer program product comprising a computer readable storage medium and a computer program mechanism embedded therein, the computer program mechanism comprising:

instructions for computing, for each of at least a subset of nodes in a directed graph of linked nodes, a respective quantity corresponding to a derivative of a node importance function;

instructions for comparing the respective computed quantity with a threshold for each node in the subset; and

instructions for identifying at least a portion of the subset for which the comparison produces a predefined result;

wherein the directed graph of linked nodes corresponds to a linked database, and wherein the nodes correspond to documents within the linked database.

14. The computer program product of claim 13, further including instructions for normalizing the respective quantity in accordance with a respective rank.

15. The computer program product of claim 13, wherein the instructions for computing include instructions for computing the derivative of the node importance function with respect to a coupling factor.

16. The computer program product of claim 15, wherein the instructions for computing include:

instructions for computing at least two intermediate quantities of the derivative of the node importance function for each of the set of nodes, each of the intermediate quantities computed using a respective value of the coupling factor; and

instructions for averaging the intermediate quantities to produce the computed respective quantity.

17. The computer program product of claim 15, wherein the coupling factor represents how tightly coupled the linked nodes are in the directed graph.

18. The computer program product of claim 15, further including:

instructions for generating an adjusted importance for at least one node of the portion.

19. The computer program product of claim 17, wherein the instructions for generating include instructions for adjusting, for the at least one node, the importance by an amount proportional to the respective computed quantity.

20. The computer program product of claim 13, further including:

instructions for removing at least one node of the portion from the directed graph.

21. A computer program product, for use with a computer system, the computer program product comprising a computer readable storage medium and a computer program mechanism embedded therein, the computer program mechanism comprising:

instructions for computing, for each of at least a portion of nodes in a directed graph of linked nodes, a respective quantity for the respective node corresponding to a derivative of a node importance function; and

instructions for selecting a predetermined percentage of nodes in accordance with the respective quantities;

wherein the directed graph of linked nodes corresponds to a linked database, and wherein the nodes correspond to documents within the linked database.

22. The computer program product of claim 21, wherein the instructions for computing includes instructions for computing the derivative of the node importance function with respect to a coupling factor.

23. The computer program product of claim 21, further including instructions for providing the subset in response to a search request.

24. A system for identifying nodes that are beneficiaries of node importance inflating links in a directed graph of linked nodes, the system comprising:

memory;

one or more processors;

one or more programs stored in the memory and configured for execution by the one or more processors, the one or more programs including;

instructions to compute, for each of at least a subset of the nodes in the directed graph, a respective quantity corresponding to a derivative of a node importance function;

instructions to compare, for each node in the subset, the respective computed quantity with a threshold; and

instructions to identify at least a portion of the subset for which the comparison produces a predefined result;

wherein the directed graph of linked nodes corresponds to a linked database, and wherein the nodes correspond to documents within the linked database.

25. The system of claim 24, further including instructions to normalize the respective quantity in accordance with a respective rank.

26. The system of claim 24, wherein the instructions to compute include instructions to compute the derivative of the node importance function with respect to a coupling factor.

27. The system of claim 26, wherein the instructions to compute include:

instructions to compute at least two intermediate quantities of the derivative of the node importance function for each of the set of nodes, each of the intermediate quantities computed using a respective value of the coupling factor; and

instructions to average the intermediate quantities to produce the computed respective quantity.

28. The system of claim 26, wherein the coupling factor represents how tightly coupled the linked nodes are in the directed graph.

29. The system of claim 26, further including:

instructions to adjust an importance for at least one node of the portion.

30. The system of claim 29, wherein the instructions to adjust include instructions to adjust, for the at least one node, the importance by an amount proportional to the respective computed quantity.

31. The system of claim 26, further including:

instructions to remove at least one node of the identified portion from the directed graph.

32. A system for ordering nodes in a directed graph of linked nodes, the system comprising:

memory;

one or more processors;

one or more programs stored in the memory and configured for execution by the one or more processors, the one or more programs including;

instructions to compute, for each of at least a portion of the nodes in the direct graph, a respective quantity for the respective node corresponding to a derivative of a node importance function; and

instructions to select a predetermined percentage of nodes in accordance with the respective quantities;

wherein the directed graph of linked nodes corresponds to a linked database, and wherein the nodes correspond to documents within the linked database.

33. The system of claim 32, wherein the instructions to compute include instructions to compute the derivative of the node importance function with respect to a coupling factor.

34. The system of claim 32, including instructions to return the subset in response to a search request.

35. A system for identifying nodes that are beneficiaries of node importance inflating links in a directed graph of linked nodes, wherein the directed graph of linked nodes corresponds to a linked database, and wherein the nodes correspond to documents within the linked database, the system comprising:

means for computing, for each of at least a subset of the nodes in the directed graph, a respective quantity corresponding to a derivative of a node importance function;

means for, for each node in the subset, comparing the respective computed quantity with a threshold; and

means for identifying at least a portion of the subset for which the comparison produces a predefined result.

48. A computer-implemented method for identifying link spam in a linked database, the method comprising:

calculating a matrix A(c) representing the linked database, where the matrix A(c) is an N×

N matrix that is a function of a link coupling coefficient c, where N is a number of nodes in the linked database, wherein the nodes correspond to documents within the linked database;

calculating a principal eigenvector of A(a), denoted x(a) and a principal eigenvector of A(b), denoted x(b) where a and b are two distinct values of the coefficient c; and

for a node, calculating a spam likelihood value S from components of x(a) and x(b) corresponding to the node.

49. The method of claim 48, wherein calculating the spam likelihood value S comprises calculating a logarithm of a ratio of the two components.

50. The method of claim 48, wherein calculating the spam likelihood value S comprises calculating a difference between a logarithm of one of the components and a logarithm of another of the components.

51. The method of claim 48, wherein A(c)=(cP+(1−

c)E)^T, where P is an N×

N matrix and E is an N×

N matrix.

52. The method of claim 51, wherein elements of P represent link coupling strengths between nodes of the linked database, and wherein elements of E represent link decoupling strengths between nodes of the linked database.