Ranking documents based on user behavior and/or feature data

• US 8,117,209 B1
• Filed: 03/19/2010
• Issued: 02/14/2012
• Est. Priority Date: 06/17/2004
• Status: Active Grant

• Current Assignee: Google LLC (Alphabet Inc.)
• Original Assignee: Google Inc. (Alphabet Inc.)
• Inventors: Anderson, Corin, Battle, Alexis, Dean, Jeffrey A
• Primary Examiner(s): LIN, SHEW FEN
• Application Number:
  US12/728,015
• Time in Patent Office: 697 Days
• Field of Search: 707722-735, 707999002-999005
• US Class Current: 707/748
• CPC Class Codes: G06F 16/24578 using ranking
• CPC Class Codes: G06F 16/951 Indexing; Web crawling techniques
• CPC Class Codes: G06F 16/9535 Search customisation based on user profiles and personalisation
• CPC Class Codes: G06F 16/9538 Presentation of query results
• CPC Class Codes: G06F 40/134 Hyperlinking

Abstract

A system generates a model based on feature data relating to different features of a link from a linking document to a linked document and user behavior data relating to navigational actions associated with the link. The system also assigns a rank to a document based on the model.

Claims

1. A method performed by one or more server devices, comprising:

storing, in a memory associated with the one or more server devices, feature data associated with features of links within documents where the feature data specifies one or more values for at least some of the features, and where the features are independent of user behavior;

generating, by the one or more server devices, feature vectors for the links based on the stored feature data;

storing, in a memory associated with the one or more server devices, user data relating to the links, where the user data includes user class data and user behavior data;

generating, using one or more processors of the one or more server devices, a model that identifies a probability that a particular link, associated with one or more particular features, will be selected by a particular user, associated with one or more user classes or behaviors, where generating the model includes;

analyzing the generated feature vectors and the stored user data in relation to selections of links made by users while viewing a document that includes one of the links, and in relation to the one or more user classes or behaviors; and

analyzing the generated feature vectors and the stored user data in relation to users viewing a document that includes one of the links and not selecting the one of the links, and in relation to the one or more user classes or behaviors;

storing, in a memory associated with the one or more server devices, the model; and

calculating, by at least one of the one or more server devices and based on the generated model, a probability that the particular link will be selected by the particular user.

2. The method of claim 1, where generating the model further comprises:

generating rules that apply to all of the links.

3. The method of claim 2, where generating rules that apply to all of the links comprises:

generating a rule that indicates that first ones of the links associated with a first attribute are associated with a higher probability of being selected than second ones of the links associated with a second attribute.

4. The method of claim 1, where generating the model further comprises:

generating rules that apply to a first document that includes at least one of the links, and that do not apply to a second document that includes at least one of the links.

5. The method of claim 4, where generating the rules that apply to the first document comprises:

generating a rule that indicates that a first link associated with a first attribute within the first document is associated with a higher probability of being selected than a second link associated with a second attribute within the first document.

6. The method of claim 1, further comprising:

assigning a weight to the particular link, where the weight reflects the probability that the link will be selected.

7. The method of claim 6, further comprising:

ranking a document that includes the particular link based on the assigned weight.

8. The method of claim 6, further comprising:

updating the weight based on new user behavior data.

9. A system comprising:

a memory to store instructions executable by one or more processors; and

a processor to execute the instructions to;

store feature data associated with features of links within documents where the feature data specifies one or more values for at least some of the features, and where the features are independent of user behavior,

generate feature vectors for the links based on the stored feature data,

store user data relating to;

user navigational activity with regard to the documents accessed by one or more users, and

user class data associated with the one or more users;

generate, based on the feature data and the user data, a model that identifies a probability that a particular link, associated with one or more particular features, will be selected by a particular user, associated with a particular navigational activity or a particular user class, where, when generating the model, the processor is to further execute the instructions to;

analyze the generated feature vectors and the stored user data in relation to selections of links made by users while viewing a document that includes one of the links and in relation to the user class data or the user navigational activity, and

analyze the generated feature vectors and the stored user data in relation to users viewing a document that includes one of the links without selecting the one of the links and in relation to the user class data or the user navigational activity; and

store the model in the memory.

10. The system of claim 9, where the processor is to further execute the instructions to:

calculate a rank score for a document that includes the particular link based on the identified probability.

11. The system of claim 10, where the processor is to further execute the instructions to:

order the document, with regard to at least one other document, based on the calculated rank score.

12. The system of claim 9, where the feature data includes at least one of:

a font size of anchor text associated with a link,

a position of the link within a document,

a position of the link in a list of links,

a font color associated with the link,

particular word in the anchor text associated with the link,

a determination of commerciality of the anchor text

a type of the link,

a particular word before or after the link,

a topical cluster with which the anchor text is associated,

whether the link leads to a target document on a same host or domain as the document that includes the link,

or whether an address associated with the link includes an embedded address.

13. A non-transitory computer-readable medium that store instructions executable by at least one processor, the computer-readable medium comprising:

one or more instructions to store, in a memory, feature data associated with a features of links within documents, where the feature data specifies one or more values for at least some of the features, and where the features are independent of user behavior;

one or more instructions to generate feature vectors for the links based on the stored feature data

one or more instructions to store, in a memory, user data relating to the links, where the user data includes user class data and user behavior data

one or more instructions to store, in a memory, instance data that includes;

positive instances relating to when links were selected by a user while viewing a document that includes one of the links, and

negative instances relating to when links within the source documents that were not selected by the one or more users prior to ceasing to view a document that includes one of the links;

one or more instructions to generate a model that identifies a probability that a particular link, associated with one or more particular features, will be selected by a user, associated with one or more particular user classes or user behaviors, where the instructions to generate the rules further comprise;

one or more instructions to analyze the positive instances in relation to the generated feature vectors and the stored user data, and

one or more instructions to analyze the negative instances in relation to the generated feature vectors and the stored user data;

one or more instructions to store, in the memory, the model; and

one or more instructions to order, based on the model, a particular document that includes the particular link, with respect to at least one other document in a plurality of documents.

14. The non-transitory computer-readable medium of claim 13, where the rules for the model comprise:

a general rule applicable to a group of documents, and

a specific rule applicable only to a particular document.

15. The non-transitory computer-readable medium of claim 13, further comprising:

instructions to update, periodically, the rules for the model based on changes in the user data.