Scoring recommendations and explanations with a probabilistic user model

• US 7,676,400 B1
• Filed: 06/03/2005
• Issued: 03/09/2010
• Est. Priority Date: 06/03/2005
• Status: Active Grant

• Current Assignee: Versata Development Group, Inc. (ESW Capital, LLC)
• Original Assignee: Versata Development Group, Inc. (ESW Capital, LLC)
• Inventors: Dillon, Thomas H.
• Primary Examiner(s): Airapetian; Mila
• Application Number:
  US11/144,455
• Time in Patent Office: 1,740 Days
• Field of Search: 705/26, 705/27, 705/14
• US Class Current: 705/26.7
• CPC Class Codes: G06Q 30/02 Marketing; Price estimation or determination; Fundraising
• CPC Class Codes: G06Q 30/06 Buying, selling or leasing transactions
• CPC Class Codes: G06Q 30/0631 Item recommendations

Abstract

A data processing system generates recommendations for on-line shopping by scoring recommendations matching the customer'"'"'s cart contents using by assessing and ranking each candidate recommendation by the expected incremental margin associated with the recommendation being issued (as compared to the expected margin associated with the recommendation not being issued) by taking into consideration historical associations, knowledge of the layout of the site, the complexity of the product being sold, the user'"'"'s session behavior, the quality of the selling point messages, product life cycle, substitutability, demographics and/or other considerations relating to the customer purchase environment. In an illustrative implementation, scoring inputs for each candidate recommendation (such as relevance, exposure, clarity and/or pitch strength) are included in a probabilistic framework (such as a Bayesian network) to score the effectiveness of the candidate recommendation and/or associated selling point messages by comparing a recommendation outcome (e.g., purchase likelihood or expected margin resulting from a given recommendation) against a non-recommendation outcome (e.g., the purchase likelihood or expected margin if no recommendation is issued). In addition, a probabilistic framework may also be used to select a selling point message for inclusion with a selected candidate recommendation by assessing the relative strength of the selling point messages by factoring in a user profile match factor (e.g., the relative likelihood that the customer matches the various user case profiles).

Claims

1. A computer-based method of scoring recommendations for potential purchase by a customer, comprising:

receiving a recommendation context from a customer;

using the recommendation context to identify a plurality of candidate recommendations that match the recommendation context, where each candidate recommendation recommends at least one recommended item;

with a computer system, determining a score for each candidate recommendation by subtracting a first expected margin value factor for the recommended item that is based on the candidate recommendation not being displayed from a second expected margin value factor for the recommended item that is based on the candidate recommendation being displayed; and

ranking the plurality of candidate recommendations using the score for each candidate recommendation to identify at least a highest ranking candidate recommendation.

2. The method of claim 1, further comprising using a Bayesian network to determine the score for each candidate recommendation.

3. The method of claim 1, further comprising generating a selling point message for each candidate recommendation by:

identifying a plurality of selling point messages corresponding to the candidate recommendation, where each selling point messages is targeted to a predetermined user case profile;

identifying a first user case profile most likely matches the customer; and

selecting a first selling point message from the plurality of selling point messages that is targeted to the first user case profile for use with the candidate recommendation.

4. The method of claim 1, where the first expected margin value is calculated based at least in part on a relevance measure, an expected exposure measure and an expected clarity measure for the candidate recommendation.

5. The method of claim 1, where the second expected margin value is calculated based at least in part on a relevance measure, a post-recommendation exposure measure and a post-recommendation clarity measure for the candidate recommendation.

6. The method of claim 1, where the first expected margin value factor is calculated by multiplying a margin value factor for the recommended item by the probability that the recommended item will be purchased if the candidate recommendation is not displayed.

7. The method of claim 1, where the second expected margin value factor is calculated by multiplying a margin value factor for the recommended item by the probability that the recommended item will be purchased if the candidate recommendation is displayed.

8. The method of claim 1, where the step of determining a score for each candidate recommendation comprises:

using a Bayesian network to calculate a first purchase probability measure for a candidate recommendation item, given a relevance measure for the item, an exposure measure for the item if the candidate recommendation is displayed and a clarity measure for the item if the candidate recommendation is displayed;

using a Bayesian network to calculate a second purchase probability for the candidate recommendation item, given a relevance measure for the item, an exposure measure for the item if the candidate recommendation is not displayed and a clarity measure for the item if the candidate recommendation is not displayed;

subtracting the second purchase probability measure from the first purchase probability measure to form a first difference value; and

multiplying the first difference value by a margin value for the candidate recommendation item.

9. The method of claim 1, wherein the second expected margin value factor is based at least in part on a pitch strength value for a selling point message associated with the candidate recommendation.