Content based method for product-peer filtering

• US 6,356,879 B2
• Filed: 10/09/1998
• Issued: 03/12/2002
• Est. Priority Date: 10/09/1998
• Status: Expired due to Term

• Current Assignee: eBay Inc.
• Original Assignee: International Business Machines Corporation
• Inventors: Yu, Philip S., Aggarwal, Charu C.
• Primary Examiner(s): Hafiz, Tariq R.
• Assistant Examiner(s): Jeanty, Romain
• Application Number:
  US09/169,029
• Publication Number:
  US 20010049623A1
• Time in Patent Office: 1,250 Days
• Field of Search: 705/14, 705/10, 705/26, 705/74, 705/27, 709/219, 455/4.2, 707/3, 707/102
• US Class Current: 705/26.62
• CPC Class Codes: G06Q 30/02 Marketing; Price estimation or determination; Fundraising
• CPC Class Codes: G06Q 30/0253 During e-commerce, i.e. online transactions
• CPC Class Codes: G06Q 30/0625 Directed, with specific intent or strategy
• CPC Class Codes: G06Q 30/0631 Item recommendations
• CPC Class Codes: Y10S 707/99933 Query processing, i.e. searching
• CPC Class Codes: Y10S 707/99943 Generating database or data structure, e.g. via user interface

Abstract

The present invention derives product characterizations for products offered at an e-commerce site based on the text descriptions of the products provided at the site. A customer characterization is generated for any customer browsing the e-commerce site. The characterizations include an aggregation of derived product characterizations associated with products bought and/or browsed by that customer. A peer group is formed by clustering customers having similar customer characterizations. Recommendations are then made to a customer based on the processed characterization and peer group data.

Claims

1. A method for providing product recommendations to customers of an e-commerce web-site, the e-commerce web site having a plurality of products offered for sale there at, said method comprising the steps of:

deriving product characterizations only for products that have actually been browsed or purchased by the customers of the e-commerce web-site, the products comprised in the plurality of products, the product characterizations being based on text descriptions of the products;

creating individual customer characterizations for each of the customers based on the concise product characterizations, each of the individual customer characterizations corresponding to the products that were either browsed or purchased by a corresponding customer;

clustering the individual customer characterizations based on similarities there between to form peer groups;

categorizing each of the customers into one of the peer groups;

providing product recommendations to a given customer based on an individual customer characterization of the given customer and information from a peer group to which the given customer is categorized; and

re-evaluating a previous categorization of the current session customer into one of the peer groups, based on similarities between the updated individual customer characterization of the current session customer and the individual customer characterizations of other customers.

2. The method according to claim 1, wherein said step of deriving product characterizations further includes the steps of:

(a) finding a frequency of occurrence for each word in said text descriptions of said products browsed or purchased by each of said customers;

(b) dividing the total frequency for each word by the frequency of occurrence for said word for all of said customers;

(c) finding the standard deviation for each word;

(d) selecting words having larger standard deviations; and

(e) expressing a product characterization by using said selected words in said text descriptions.

3. The method of claim 1, further including the steps of:

(a) creating an individual customer text characterization by weighted concatenation of the product characterizations corresponding to said products browsed or purchased by the customer in a current session of the customer on the electronic commerce site;

(b) computing a cluster centroid for each of said peer groups;

(c) selecting the peer groups whose cluster centroid is closest to the individual customer text characterization created in step (a); and

(d) generating one of product, peer and profile recommendations based on said selected peer groups.

4. The method of claim 3, wherein said cluster centroid is computed by concatenating text characterizations of all of said individual customer text characterizations in each peer group.

5. The method of claim 3, wherein the recommendations comprise a weighted concatenation of text-characterizations of products bought and browsed in said current on-line session.

6. The method according to claim 1, wherein the predefined criteria comprises removing words corresponding to a predefined list of common words.

7. The method according to claim 1, wherein the predefined criteria comprises maintaining words indicative of customer behavior and discarding all other words.

8. The method according to claim 1, further comprising the step of automatically performing said deriving, creating, and clustering steps at pre-specified time intervals to update the peer groups.

9. The method according to claim 1, further comprising the step of automatically performing said deriving, creating, and clustering steps when the e-commerce web-site is experiencing a degree of traffic less than a predefined threshold.

10. The method according to claim 1, wherein said deriving step derives the product characterizations to the exclusion of any of the plurality of products that have been offered for sale but have not browsed or purchased by the customers of the e-commerce web-site.

11. The method according to claim 1, wherein said deriving step comprises the step of excluding, from the product characterizations, any of the plurality of products that have been offered for sale but have not been browsed or purchased by the customers of the e-commerce web-site.

12. The method according to claim 1, wherein said creating step comprises, for each of the individual customer characterizations, the step of generating word/number pairs, each of the word/number pairs comprising a word that occurs in the concise product characterizations and a number of occurrences of the word.

13. The method according to claim 1, wherein said deriving step is performed in real-time as the customers browse or purchase the products offered for sale at the e-commerce web-site.

14. The method according to claim 1, wherein the product recommendations are provided independent of customer ratings.

15. The method according to claim 1, further comprising the step of filtering a product recommendation list corresponding to the product recommendations to exclude all products that are not identified on a pre-specified promotion list.

16. The method according to claim 1, wherein said method is implemented by a program storage device readable by machine, tangibly embodying a program of instructions executable by the machine to perform said method steps.

17. The method according to claim 1, further comprising the steps of:

computing a cluster centroid for each of the peer groups by concatenating text characterizations of all of the individual customer characterizations in each of the peer groups;

selecting a particular peer group having the cluster centroid closest to the new or existing individual customer characterization; and

providing peer recommendations to the current session customer based on data from the particular peer group, the peer recommendations comprising names of customers who are members of the particular group.

18. The method according to claim 1, further comprising the steps of:

computing a cluster centroid for each of the peer groups by concatenating text characterizations of all of the individual customer characterizations in each of the peer groups;

selecting a particular peer group having the cluster centroid closest to the new or existing individual customer characterization; and

providing at least one concise product characterization to the current session customer based on data from the particular peer group.

19. A method for providing product recommendations to customers of an e-commerce web-site, the e-commerce web site having a plurality of products offered for sale there at, said method comprising the steps of:

deriving product characterizations only for products that have actually been browsed or purchased by the customers of the e-commerce web-site, the products comprised in the plurality of products, the product characterizations being based on text descriptions of the products;

creating individual customer characterizations for each of the customers based on the concise product characterizations, each of the individual customer characterizations corresponding to the products that were either browsed or purchased by a corresponding customer;

clustering the individual customer characterizations based on similarities there between to form peer groups;

receiving a query from a current session customer;

creating and updating, in real-time, text characterizations for the products that were either browsed or purchased by the current session customer;

creating a new individual customer characterization for the current session customer based on the text characterizations, when the current session customer is new to the e-commerce web-site;

updating an existing individual customer characterization for the current session customer based on the text characterizations, when the current session customer has previously browsed or purchased at least one of the plurality of products offered for sale at the e-commerce web-site;

re-evaluating a previous categorization of the current session customer into one of the peer groups, based on similarities between the updated individual customer characterization of the current session customer and the individual customer characterizations of other customers categorizing the current session customer into one of the peer groups, based on similarities between the new or existing individual customer characterization of the current session customer and the individual customer characterizations of other customers; and

responding to the query from the current session customer with at least one product recommendation, based on information from a peer group to which the current session customer is categorized.

20. A method for providing product recommendations to customers of an e-commerce web-site, the e-commerce web site having a plurality of products offered for sale there at, said method comprising the steps of:

deriving product characterizations only for products that have actually been browsed or purchased by the customers of the e-commerce web-site, the products comprised in the plurality of products, the product characterizations being based on text descriptions of the products;

creating individual customer characterizations for each of the customers based on the concise product characterizations, each of the individual customer characterizations corresponding to the products that were either browsed or purchased by a corresponding customer;

clustering the individual customer characterizations based on similarities there between to form peer groups;

receiving a query from a current session customer during a current session;

updating an individual customer characterization for the current session customer, based on the products that were either browsed or purchased by the current session customer during the current session, when the individual customer characterization for the current session customer exists prior to the current session;

re-evaluating a previous categorization of the current session customer into one of the peer groups, based on similarities between the updated individual customer characterization of the current session customer and the individual customer characterizations of other customers;

re-categorizing the current session customer into another one of the peer groups, if necessary, based on a result of said re-evaluating step; and

responding to the query from the current session customer with at least one product recommendation, based on information from a peer group to which the current session customer is categorized.