Personalized information retrieval search with backoff

• US 7,783,636 B2
• Filed: 09/28/2006
• Issued: 08/24/2010
• Est. Priority Date: 09/28/2006
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Corporation
• Inventors: Mei, Qiaozhu, Church, Kenneth W.
• Primary Examiner(s): Pham; Khanh B
• Assistant Examiner(s): HARPER, ELIYAH STONE
• Application Number:
  US11/529,134
• Publication Number:
  US 20080082485A1
• Time in Patent Office: 1,426 Days
• Field of Search: 707/104.1, 707/3, 707/100, 707/10, 707/5, 707/999.003
• US Class Current: 707/732
• CPC Class Codes: G06F 16/3331 Query processing

Abstract

Query logs are accessed to obtain queries, user information that specifies a user from which the query was received, a long with a selected result that was selected by the specified user who authored the query. This query log information is used to identify classes of users that looked for a similar result given a similar query. Those classes can then be used by a search engine in order to rank or provide search results to a user in response to a query input by the user.

Claims

1. A computer implemented method of retrieving information, comprising:

receiving a query and user information indicative of a user that initiated the query;

receiving search results based on the query;

utilizing a computer processor that is a component of the computer to apply to the search results a user class model that models a first set of user classes by grouping a plurality of users into multiple classes based on a first portion of an IP address of the plurality of users, said first portion being less than the complete IP address, and wherein applying the user class model to the search results further comprises applying the user class model so as to compare the user information to the first set of user classes so as to determine which of the search results are most likely desired by the user;

applying to the search results a backoff model that models a second set of user classes by grouping the plurality of users into multiple classes based on a second, smaller portion of the IP address of the plurality of users, said second portion being less than the complete IP address, and wherein applying the backoff model to the search results further comprises applying the backoff model so as to compare the user information to the second set of user classes so as to determine which of the search results are most likely desired by the user;

ranking the search results based on said applications of the user class model and the backoff model, wherein ranking the search results comprises ranking the search results not based on a complete indication of the identity of the user as reflected in the complete IP address but instead ranking the search results based on an incomplete indication of the identity of the user as reflected in the user class model and the backoff model; and

returning the ranked search results to the user.

2. The method of claim 1, further comprising applying a plurality of additional backoff models that model successively larger user classes.

3. The method of claim 1 and further comprising:

generating and presenting the user with customized advertising selected based on a correlation to said ranking.

4. The method of claim 1 and further comprising:

presenting the user with a suggested user action selected based on said applications of the user class model and the backoff model.

5. A computer implemented method of managing a search engine, comprising:

accessing a query log including a record of queries previously executed, when each query in the record was processed, and a search result selected for each query in the record;

utilizing a computer processor that is a component of the computer to train a user class model that models a first set of classes by grouping the queries in the record into multiple groups based on when each query was processed;

weighting the user class model with weights;

utilizing the processor to train a backoff model that models a second set of classes by grouping the queries in the record into multiple groups based on when each query was processed;

weighting the backoff model using the weights, wherein weighting the backoff model comprises weighting the backoff model higher than the weighting of the user class model;

wherein there are more classes in the first set than the second set such that each class in the first set encompasses queries from a narrower time range than each class in the second set; and

configuring the search engine to receive an input query, to receive a time indication indicative of the query, and to apply the user class model and backoff model to the time indication so as to facilitate a scoring of search results that correspond to the input query.

6. The method of claim 5 wherein training a backoff model comprises:

training a plurality of backoff models; and

wherein weighting the backoff model comprises weighting each of the plurality of backoff models.

7. The method of claim 6 wherein combining comprises:

linearly combining the user class models with each of the plurality of backoff models.

8. A computer implemented method of retrieving information, comprising:

receiving a user query from a user along with user information indicative of a location from which the user submitted the query;

receiving search results based on the user query;

utilizing a computer processor that is a component of the computer to apply to the search results a geographic class model that models a first set of geographic classes by grouping a plurality of preexisting queries into multiple geographic classes based on which of a first category of geographic units is assigned to each of the preexisting queries, and wherein applying the geographic class model to the search results further comprises applying the geographic class model so as to compare the user information and the user query to the first set of geographic classes so as to determine which of the search results are most likely desired by the user;

applying to the search results a backoff model that models a second set of geographic classes by grouping the plurality of preexisting queries into multiple geographic classes based on which of a second category of geographic units is assigned to each of the plurality of preexisting queries, and wherein applying the backoff model to the search results further comprises applying the backoff model so as to compare the user information to the second set of geographic classes so as to determine which of the search results are most likely desired by the user, and wherein the first category of geographic units is smaller than the second category of geographic units in that a unit of the first geographic category encompasses less geographic space than a unit of the second geographic category;

ranking the search results based on said applications of the geographic class model and the backoff model; and

returning the search results to the user.

9. The method of claim 8, wherein applying the backoff model comprises applying a plurality of backoff models that model successively larger user classes.

10. The method of claim 8 and further comprising generating advertising based on application of the user class model.

11. The method of claim 8 and further comprising suggesting to the user a user action based on application of the user class model.

12. The method of claim 8 and further comprising weighting and combining the first and second sets of geographic classes.

13. The method of claim 12 wherein weighting comprises weighting the second set of geographic classes higher than the first set of geographic classes.

14. The method of claim 12 wherein combining comprises linearly combining the first and second sets of geographic classes.