Time-Aware Ranking Adapted to a Search Engine Application

• US 20130091128A1
• Filed: 10/11/2011
• Published: 04/11/2013
• Est. Priority Date: 10/11/2011
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Corporation
• Inventors: Radinsky, Kira, Dumais, Susan T., Svore, Krysta M., Horvitz, Eric J., Teevan, Jaime Brooks
• Granted Patent: US 9,244,931 B2
• Time in Patent Office: Days
• US Class Current: 707/723
• CPC Class Codes: G06F 16/17 Details of further file system functions
• CPC Class Codes: G06F 16/24578 using ranking
• 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: G06N 20/00 Machine learning
• CPC Class Codes: G06N 5/045 Explanation of inference; Explainable artificial intelligence [XAI]; Interpretable artificial intelligence

Abstract

Techniques provide time-aware ranking, such as ranking of information, files or URL (uniform resource locator) links. For example, time-aware modeling assists in determining user intent of a query to a search engine. In response to the query, results are ranked in a time-aware manner to better match the user intent. The ranking may model query, URL and query-URL pair behavior over time to create time-aware query, URL and query-URL pair models, respectively. Such models may predict behavior of a query-URL pair, such as frequency and timing of clicks to the URL of the pair when the query of the pair is posed to the search engine. Results of a query may be ranked by predicted query-URL behavior. Once ranked, the results may be sent to the user in response to the query.

Claims

1. One or more computer-readable media storing computer-executable instructions that, when executed, cause one or more processors to perform acts comprising:

modeling query behavior over time to create a time-aware query model;

modeling uniform resource locator (URL) behavior over time to create a time-aware URL model;

determining a query-URL relationship based at least in part on the time-aware query model and the time-aware URL model;

performing a search in response to a query; and

ranking results of the search according to the query-URL relationship.

2. One or more computer-readable media as recited in claim 1, wherein:

modeling the query behavior is based on a first time series; and

modeling the URL behavior is based on a second time series that is different than the first time series.

3. One or more computer-readable media as recited in claim 1, wherein determining the query-URL relationship comprises:

learning rules governing model selection; and

replacing the time-aware query model and the time-aware URL model according to the rules.

4. One or more computer-readable media as recited in claim 1, further comprising:

learning rules governing model selection, the learning based on the query behavior over time and the URL behavior over time; and

replacing, based on the learned rules governing model selection;

the time-aware query model with a second time-aware query model; and

the time-aware URL model with a second time-aware URL model.

5. One or more computer-readable media as recited in claim 1, further comprising:

configuring a plurality of time-aware models;

configuring a decision tree with rules; and

operating the decision tree to select from among the plurality of time-aware models, as indicated by the rules of the decision tree.

6. One or more computer-readable media as recited in claim 1, further comprising:

switching between two time-aware query models based on a change in the query behavior over time; and

switching between two time-aware URL models based on a change in the URL behavior over time.

7. One or more computer-readable media as recited in claim 1, further comprising:

learning rules governing model selection, wherein the rules change over time based on query behavior over time and URL behavior over time;

switching between two time-aware query models based on operation of the rules; and

switching between two the time-aware URL models based on operation of the rules.

8. One or more computer-readable media as recited in claim 1, wherein:

the determined query-URL relationship indicates, for the query, a future change that elevates a first URL to a popularity level greater than a second URL; and

the ranked results include the first URL ranked higher than the second URL.

9. A method of modeling query-uniform resource locator (URL) pairs, comprising:

under control of one or more processors configured with executable instructions;

receiving a query;

modeling a plurality of query-URL pairs to create a plurality of query-URL models for each of the plurality of query-URL pairs;

selecting between the plurality of query-URL models that model a query-URL pair to select a query-URL model for use;

predicting user intent based on the query-URL model selected for use; and

ranking URLs found in response to the query based on the predicted user intent.

10. The method of claim 9, wherein modeling the plurality of query-URL pairs comprises:

using a time series to create the plurality of query-URL models for each of the plurality of query-URL pairs.

11. The method of claim 9, wherein modeling the plurality of query-URL pairs comprises:

modeling query behavior over time to create a time-aware query model based on a first time series;

modeling URL behavior over time to create a time-aware URL model based on a second time series; and

using the time-aware query model and the time-aware URL model to model the plurality of query-URL pairs.

12. The method of claim 9, wherein selecting between the plurality of query-URL models comprises:

selecting based on static and temporal characteristics of the query-URL pair.

13. The method of claim 9, wherein selecting between the plurality of query-URL models comprises:

learning rules governing model selection, wherein the rules evolve over time based on query behavior and URL behavior; and

selecting between the plurality of query-URL models based the learned rules.

14. The method of claim 9, wherein selecting between the plurality of query-URL models comprises:

learning rules governing model selection;

navigating a decision tree based on the learned rules; and

selecting between the plurality of query-URL models as indicated by the decision tree.

15. The method of claim 9, wherein ranking URLs found in response to the query comprises:

learning rules governing model selection; and

switching from the query-URL model selected for use to a second query-URL model selected for use based on the rules learned.

16. A search service, comprising:

a processor;

a memory, in communication with the processor;

a plurality of query-uniform resource locator (URL) models, defined in the memory, for each of a plurality of query-URL pairs;

a selection module, defined in the memory, to select between the plurality of query-URL models for each of the plurality of query-URL pairs;

a prediction module, defined in the memory, to predict user intent based on the selected query-URL models;

a ranking module, defined in the memory, to rank URLs found in response to a query based on the predicted user intent; and

an interface to send ranked URLs to a client.

17. The search service of claim 16, wherein the plurality of query-URL models comprise:

state space models, the state space models including;

information derived from a plurality of query models formulated as a plurality of time series, respectively; and

information derived from a plurality of URL models formulated as a plurality of time series, respectively.

18. The search service of claim 16, wherein the selection module comprises:

rules developed in response to a query model and a URL model; and

a decision tree, with branches governed by the rules, for switching between query-URL models.

19. The search service of claim 16, wherein the selection model comprises:

a time series to model each of the plurality of query-URL models for each of the plurality of query-URL pairs.

20. The search service of claim 16, additionally comprising:

a models library, defined in the memory, containing the plurality of query-URL models for each of the plurality of query-URL pairs, wherein the models library updates each of the plurality of query-URL models for each of the plurality of query-URL pairs based on queries submitted to the search engine and URLs sent by the search engine and clicked by a client.