Methods and systems for implementing a compositional recommender framework

• US 8,566,274 B2
• Filed: 01/10/2011
• Issued: 10/22/2013
• Est. Priority Date: 05/12/2010
• Status: Active Grant

• Current Assignee: Salesforce.com, Inc.
• Original Assignee: Salesforce.com, Inc.
• Inventors: Koister, Jari
• Primary Examiner(s): Chaki, Kakali
• Assistant Examiner(s): RIOJAS RAMIREZ, MARIO
• Application Number:
  US12/987,957
• Publication Number:
  US 20110282814A1
• Time in Patent Office: 1,016 Days
• Field of Search: None
• US Class Current: 706/62
• CPC Class Codes: G06F 16/9535 Search customisation based on user profiles and personalisation
• CPC Class Codes: G06N 5/02 Knowledge representation; Symbolic representation
• CPC Class Codes: G06Q 30/02 Marketing; Price estimation or determination; Fundraising

Abstract

A compositional recommender framework using modular recommendation functions is described. Each modular recommendation function can use a discrete technology, such as using clustering, a database lookup, or other means. A first recommendation function can recommend to a user items, such as books to check out, automobiles to purchase, people to date, etc. Another modular recommendation function can be daisy chained with the first to recommend items that are similar or related to the first recommended items, such as users who have also checked out the same recommended book, trailers that can be towed by the recommended automobiles, or vacations booked by people that were recommended as people to date. The modular recommendation functions can be used to build customized recommendation engines for different industries.

Claims

1. A method of building a recommendation engine using a compositional recommender framework, the method comprising:

selecting a first modular recommendation function, the first recommendation function configured to accept a first input object and output at least one first recommended object based on the first input object;

selecting a second modular recommendation function, the second recommendation function configured to accept a second input object and output at least one second recommended object based on the second input object, wherein an output object from either modular function is compatible as an input object to another modular recommendation function;

configuring, using a processor operatively coupled with a memory, the modular functions so that one of the at least one first recommended objects from the first modular recommendation function is an input object to the second modular recommendation function, the configuring to build a recommendation engine such that a recommendation from the recommendation engine is based on an output from the second modular function, which is based on an output from the first modular function; and

reconfiguring the first and second modular recommendation functions so that one of the at least one second recommended objects from the second modular recommendation function is an input object to the first modular recommendation function, the reconfiguring to build a reconfigured recommendation engine such that a recommendation from the reconfigured recommendation engine is based on an output from the first modular function, which is based on an output from the second modular function.

2. The method of claim 1 wherein the first input object represents a document.

3. The method of claim 2 wherein the at least one first recommended object represents at least one document.

4. The method of claim 2 wherein the at least one first recommended object represents at least one user.

5. The method of claim 1 wherein the first input object represents a user.

6. The method of claim 5 wherein the at least one first recommended object represents at least one document.

7. The method of claim 5 wherein the at least one first recommended object represents at least one user.

8. The method of claim 1 further comprising:

mapping multiple first recommended objects from the first modular recommendation function to input one at a time as second input objects into the second modular recommendation function.

9. The method of claim 1 wherein the first recommendation function includes a clustering function, the method further comprising:

clustering the first input object to determine similar objects and outputting at least one of the similar objects as the at least one first recommended object.

10. The method of claim 1 wherein the first recommendation function includes a database lookup function, the method further comprising:

looking up objects in a database based on the first input object to determine related objects and outputting at least one of the related objects as the at least one first recommended object.

11. The method of claim 1 further comprising:

inputting an object into the recommendation engine; and

executing the recommendation engine to generate recommended objects.

12. The method of claim 1 wherein a recommended object from either modular recommendation function is compatible as an input to the other modular recommendation function.

13. The method of claim 1 further comprising:

selecting a third modular recommendation function, the third modular recommendation function configured to accept a third input object and output at least one third recommended object based on the third input object, wherein an output object from the third modular recommendation function is compatible as an input object to another modular recommendation function; and

reconfiguring the modular functions so that one of the at least one second recommended objects from the second modular recommendation function is an input object to the third modular recommendation function.

14. The method of claim 1 wherein the operations are performed in the following order:

selecting a first modular recommendation function;

selecting a second modular recommendation function; and

configuring the modular functions.

15. The method of claim 1 wherein each operation is performed by the computer processor operatively coupled to the memory.

16. A computer system executing instructions in a computer program, the computer program instructions comprising program code for performing the operations of claim 1.

17. A machine-readable tangible storage medium embodying a computer program having instructions for causing one or more machines to perform the operations of claim 1.

18. A method of using a recommendation engine built using a compositional recommender framework, the method comprising:

selecting an object of interest;

inputting the object of interest into a first modular recommendation function, the first recommendation function configured to accept a first input object and output first recommended objects based on the first input object;

mapping the first recommended objects from the first modular recommendation function to input one at a time as second input objects into a second modular recommendation function;

inputting, automatically without user interaction, using a processor operatively coupled with a memory, the second input objects into the second modular recommendation function, the second recommendation function configured to accept the second input objects and output at least one second recommended object based on the second input objects;

receiving data representing at least one output object from the second recommendation function as a recommendation, the recommendation based upon the object of interest; and

reconfiguring the first and second modular recommendation functions so that one of the at least one second recommended objects from the second modular recommendation function is an input object to the first modular recommendation function, the reconfiguring to build a reconfigured recommendation engine such that a recommendation from the reconfigured recommendation engine is based on an output from the first modular function, which is based on an output from the second modular function.