SYSTEM AND METHOD FOR CAPTURING SEASONALITY AND NEWNESS IN DATABASE SEARCHES

• US 20160092519A1
• Filed: 09/26/2014
• Published: 03/31/2016
• Est. Priority Date: 09/26/2014
• Status: Abandoned Application

• Current Assignee: Walmart Apollo, LLC (WalMart Inc.)
• Original Assignee: Wal Mart Stores Inc (WalMart Inc.)
• Inventors: Srivastava, Varun, Ruan, Yiye, Zheng, Yan
• Application Number:
  US14/498,417
• Publication Number:
  US 20160092519A1
• Time in Patent Office: Days
• US Class Current: 1/1
• CPC Class Codes: G06Q 30/0641 Shopping interfaces

Abstract

A system and method for capturing seasonality in a database system is disclosed herein. A database system may comprise a first database cluster H and a second database cluster L, with records being divided between the first database cluster H and the second database cluster L by likelihood of the record being accessed. Several criteria may be used to determine the likelihood of a record being accessed. For example, one may review access history to determine the activity for the record during various time periods, such as comparing access history for the record the same day one year ago, in an attempt to determine seasonality of the record. One may also review the newness of a record, such that the newness of a record is part of the criteria being reviewed when assigning record to a database cluster. Other embodiments are also disclosed herein.

Claims

1. A method comprising:

determining an access history for a record in a set of distinct records in a database system, the database system comprising a first database cluster H and a second database cluster L, where each record in the set of distinct records is placed in one of the first database cluster H or the second database cluster L based on a likelihood of each such record being accessed;

using the access history for the record in a predictive model to predict a future likelihood of the record being accessed;

storing the record in one of the first database cluster H or the second database cluster L based at least on the future likelihood of the record being accessed;

receiving a search request from a requester; and

presenting a search result containing one or more records of the set of distinct records to the requester;

wherein;

the access history for the record is analyzed at one or more discrete time periods, the one or more discrete time periods including an immediately prior year.

2. The method of claim 1 further comprising determining the one or more discrete time periods to be at least one of 1-day, 3-days, 7-days, 1-month, 3-months, 6-months, or one year.

3. The method of claim 2 wherein the one or more discrete time periods comprise at least two discrete time periods.

4. The method of claim 1 wherein:

the database system is used in an electronic commerce system;

each record in the set of distinct records represents a separate item for sale in the electronic commerce system;

the access history for the record is used to determine if the record has a seasonal pattern for popularity; and

if the record has the seasonal pattern for popularity, then using the access history further comprises;

using the seasonal pattern for popularity to predict the future likelihood of the record being accessed.

5. The method of claim 1 wherein:

the first database cluster H is stored on a first database server H;

the second database cluster L is stored on a second database server L; and

the first database server H has greater processing capabilities than the second database server L.

6. A method comprising:

for each record in a set of distinct records in a database system, the database system comprising a first database cluster H and a second database cluster L, where each such record in the set of distinct records is placed in one of the first database cluster H or the second database cluster L based on a likelihood of each such record being accessed;

reviewing a set of past feature vectors for the record;

counting the number of days that the set of past feature vectors for the record indicated that the record was available;

translating the number of days into a newness factor;

using the newness factor to determine a new feature vector for the record;

using the new feature vector to determine whether to store the record in the first database cluster H or the second database cluster L;

storing the record in the first database cluster H or the second database cluster L based at least in part on the new feature vector;

receiving a search request from a requester; and

presenting a search result containing one or more records of the set of distinct records to the requester.

7. The method of claim 6 wherein:

the database system is a part of an electronic commerce system;

each record within the set of distinct records contains information about a single item for sale in the electronic commerce system; and

translating the number of days comprises configuring the newness factor such that records with a lower newness factor are more likely to be placed in the first database cluster H.

8. The method of claim 6 wherein storing the record comprises storing the record in the first database cluster H if the record is more likely to be accessed than the records of the set of distinct records stored in the second database cluster L.

9. The method of claim 6 wherein:

using the new feature vector comprises inputting the new feature vector into a predictive model that is configured to attribute records of the set of distinct records in either the first database cluster H or the second database cluster L.

10. The method of claim 6 wherein:

the first database cluster H is stored on a first database server H;

the second database cluster L is stored on a second database server L; and

the first database server H has greater processing capabilities than the second database server L.

11. A method comprising:

receiving a request from a requesting party to produce a feature vector for a record, the feature vector being used to aid in placing the record in one of a first database cluster H or a second database cluster L;

retrieving a first access history for the record at a first time period, the first access history indicating a number of times the record was accessed within the first time period that is a predetermined time period away from a current day;

repeating retrieving one or more second access histories for the record for one or more additional time periods, each second access history indicating a number of times the record was accessed within the one or more additional time periods, each of which are different predetermined time periods away from the current day;

generating the feature vector based on the first access history and the one or more second access histories of the record; and

sending the feature vector to the requesting party.

12. The method of claim 11 further comprising determining each of the first time period and the additional one or more time periods to be 1-day, 3-days, 7-days, 1-month, 3-months, 6-months, or one year.

13. The method of claim 11 further comprising using the feature vector for the record in a set of distinct records in a database system, the database system comprising the first database cluster H and the second database cluster L, where each record in the set of records is placed in one of first database cluster H or second database cluster L based on a likelihood of the record being accessed.

14. The method of claim 11 wherein:

the first database cluster H and the second database cluster L comprise a database system that is a part of an electronic commerce system;

the record contains information for a single item for sale in the electronic commerce system; and

repeating retrieving the one or more second access histories for the record for the one or more additional time periods comprises configuring the one or more additional time periods such that the one or more additional time periods are indicative of a seasonality of the single item for sale.

15. The method of claim 11 wherein generating the feature vector based on the first access history and the one or more second access histories of the record comprises configuring the feature vector to be used by a machine learning algorithm to determine if the record should be placed in a first database cluster H or a second database cluster L.

16. A method comprising:

for each record in a set of distinct records in a database system, the database system comprising a first database cluster H and a second database cluster L, where each record in the set of distinct records is to be placed in one of the first database cluster H or the second database cluster L based on a likelihood of the record being accessed of distinct records, for each record;

determining if the record was placed in the second database cluster L for each smaller time period within a longer time period;

calculating a total number of the smaller time periods within the longer time period that the record was placed in the second database cluster L; and

using the total number of the smaller time periods to at least partially determine if the record should be stored in the first database cluster H or the second database cluster L;

storing the record in either the first database cluster H or the second database cluster L based on the determination;

receiving a search request from a requester; and

presenting a search result containing one or more records of the set of distinct records to the requester.

17. The method of claim 16 wherein:

determining if the record was placed in the second database cluster L for each smaller time period within the longer time period comprises setting the smaller time period to one day; and

setting the longer time period to one month.

18. The method of claim 16 wherein:

the database system is a part of an electronic commerce system; and

each record within the set of distinct records contains information about a single item for sale in the electronic commerce system.

19. The method of claim 18 wherein using the total number of the smaller time periods to at least partially determine if the record should be stored in the first database cluster H or the second database cluster L comprises configuring the database system to be more likely to store the record in the second database cluster L if the total number of the smaller time periods that the record was stored in the second database cluster L is higher than a threshold value.

20. The method of claim 16 wherein:

using the total number of the smaller time periods to at least partially determine if the record should be stored in the first database cluster H or the second database cluster L comprises using the total number of the smaller time periods as an input into a feature vector; and

the method further comprises using the feature vector as an input into a predictive model.

21. The method of claim 16 wherein:

the first database cluster H is stored on a first database server H;

the second database cluster L is stored on a second database server L; and

the first database server H has greater processing capabilities than the second database server L.