Representing queries and determining similarity based on an ARIMA model

• US 8,090,709 B2
• Filed: 06/28/2007
• Issued: 01/03/2012
• Est. Priority Date: 06/28/2007
• Status: Expired due to Fees

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Corporation
• Inventors: Liu, Ning, Yan, Jun, Zhang, Benyu, Chen, Zheng, Wang, Jian
• Primary Examiner(s): Ehichioya, Fred I
• Assistant Examiner(s): Mitiku, Berhanu
• Application Number:
  US11/770,307
• Publication Number:
  US 20090006326A1
• Time in Patent Office: 1,650 Days
• Field of Search: None
• US Class Current: 707/713
• CPC Class Codes: G06Q 30/02 Marketing; Price estimation or determination; Fundraising

Abstract

Representing queries and determining similarity of queries based on an autoregressive integrated moving average (“ARIMA”) model is provided. A query analysis system represents each query by its ARIMA coefficients. The query analysis system may estimate the frequency information for a desired past or future interval based on frequency information for some initial intervals. The query analysis system may also determine the similarity of a pair of queries based on the similarity of their ARIMA coefficients. The query analysis system may use various metrics, such as a correlation metric, to determine the similarity of the ARIMA coefficients.

Claims

1. A method in a computing device for determining similarity between queries, the method comprising:

storing frequencies of the queries during intervals, each frequency of a query for an interval representing a number of times the query was submitted by users to a search engine;

for each of the queries, generating autoregressive integrated moving average (“

ARIMA”

) coefficients for that query based on the stored frequencies for that query; and

for a pair of queries, calculating a similarity score for the queries based on a correlation between the ARIMA coefficients of the queries, the calculating including aggregating products, for each ARIMA coefficients, of a first factor of a first query of the pair and a second factor of a second query of the pair, a factor for an ARIMA coefficient of a query being a difference between the ARIMA coefficient and a mean of the ARIMA coefficients divided by a standard deviation of the ARIMA coefficients.

2. The method of claim 1 wherein the generating of ARIMA coefficients includes identifying a number of ARIMA coefficients for each query.

3. The method of claim 2 wherein the identifying of the number of ARIMA coefficients is based on an Akaike Information Criterion algorithm.

4. The method of claim 2 wherein the calculating of a similarity score normalizes the number of ARIMA coefficients for the pair of queries.

5. The method of claim 4 wherein the normalizing includes adding ARIMA coefficients with a zero value to the query with fewer ARIMA coefficients.

6. The method of claim 1 wherein the generating of the ARIMA coefficients generates the same number of coefficients for each query.

7. The method of claim 1 wherein the frequencies of the query are derived from analysis of a query log of a search engine service.

8. The method of claim 1 including predicting frequencies of a query based on frequencies of queries that are determined to be similar based on their similarity scores.

9. A computer-readable medium encoded with instructions for controlling a computing device to determine frequency of a query at an interval, by a method comprising:

storing frequencies of the query during intervals;

generating autoregressive integrated moving average (“

ARIMA”

) coefficients representing the query based on the stored frequencies;

estimating the frequency of the query at the interval based on the ARIMA coefficients for the query; and

determining similarity between a pair of queries based on the ARIMA coefficients for the queries, the determining including aggregating products, for each ARIMA coefficient, of a first factor of a first query of the pair and a second factor of a second query of the pair, a factor for an ARIMA coefficient of a query being a difference between the ARIMA coefficient and a mean of the ARIMA coefficients divided by a standard deviation of the ARIMA coefficients.

10. The computer-readable medium of claim 9 wherein the estimated frequency is a predicted frequency for a future interval.

11. The computer-readable medium of claim 9 wherein the estimated frequency is a regenerated frequency of a past interval.

12. A computing device for representing a query comprising:

a query frequency store having, for each of a plurality of intervals, a frequency of the query during the interval;

a component that generates autoregressive integrated moving average (“

ARIMA”

) coefficients for the query based on the frequencies of the query during the intervals;

an ARIMA coefficient store for storing the generated ARIMA coefficients representing the query; and

a component that determines similarity between a pair of queries based on the ARIMA coefficients of the queries, the determining including aggregating products, for each ARIMA coefficient, of a first of a first query of the pair and a second factor of a second query of the pair, a factor for an ARIMA coefficient of a query being a difference between the ARIMA coefficient and a mean of the ARIMA coefficients divided by a standard deviation of the ARIMA coefficients.

13. The computing device of claim 12 wherein the similarity is based on a similarity score derived from a correlation of the ARIMA coefficients of the queries.