Providing product recommendations through keyword extraction from negative reviews

• US 8,515,828 B1
• Filed: 05/29/2012
• Issued: 08/20/2013
• Est. Priority Date: 05/29/2012
• Status: Active Grant

• Current Assignee: Google LLC (Alphabet Inc.)
• Original Assignee: Google Inc. (Alphabet Inc.)
• Inventors: Semret, Nemo, Wolf, Noam, Zhu, Zutao, Baskin, Jacob
• Primary Examiner(s): Shah, Amee A
• Application Number:
  US13/483,037
• Time in Patent Office: 448 Days
• Field of Search: 705/26.1, 705/26.7
• US Class Current: 705/26.7
• CPC Class Codes: G06Q 30/02 Marketing; Price estimation or determination; Fundraising

Abstract

A product recommendation system and method identify product characteristics from customer reviews using a semantic analysis of the product review text. The semantic analysis identifies negative sentiment keywords associated with one or more product characteristics in the customer review and assigns each negative sentiment keyword a value. The value of each identified negative sentiment keywords is then used to calculate a score for the product characteristic to which the negative sentiment keyword is correlated. A product recommendation report comprising the identified products, their product characteristics, a list of scored product reviews, or a combination thereof is then presented to the end user.

Claims

1. A computer-implemented method for providing product recommendations based on identification of negative sentiment key phrases in customer reviews, the method comprising:

receiving, by a computer, a request for product information defining at least a product or product category;

identifying, by the computer, product reviews matching the product or product category to generate a review analysis set;

identifying, by the computer, negative sentiment key phrases in each review of the review analysis set;

calculating, by the computer, a weight for each identified negative sentiment key phrase by determining a relative frequency of the identified negative sentiment key phrase in the customer reviews of the review analysis set and calculating an inverse of the relative frequency;

correlating, by the computer, the negative sentiment key phrases with product characteristics using a classification rubric, wherein the classification rubric defines which negative sentiment key phrases are associated with a product characteristic;

calculating, by the computer, for each product characteristic, a score based on the weights of the negative sentiment key phrase or phrases correlated to the product characteristic;

generating, by the computer, an identity of each product reviewed in the review analysis set and, for each product, a summary of scored product characteristics; and

communicating, by the computer, at least a portion of the identities of each product reviewed in the review analysis set and the corresponding summary of scored product characteristics.

2. The method of claim 1, wherein the classification rubric is derived by conducting a semantic analysis of a product review sample set to identify sentiment key phrases and assigning each sentiment key phrase to a product characteristic.

3. The method of claim 2, wherein the semantic analysis is conducted using a supervised or unsupervised machine learning algorithm.

4. The method of claim 3, wherein the semantic analysis further identifies product characteristic categories for use in mapping the negative sentiment key words to a product characteristic in the classification rubric.

5. The method of claim 2, wherein the sentiment key phrases are classified as negative or positive sentiment key phrases.

6. The method of claim 5, wherein the positive sentiment key phrases are filtered out prior to assigning the sentiment key phrases to a product characteristic.

7. The method of claim 1, wherein the relative frequency of a negative sentiment key phrase is calculated by determining how frequently the negative sentiment key phrase is used in customer reviews of all products of the same product class.

8. The method of claim 1, wherein the relative frequency of a negative sentiment key phrase is calculated by determining how frequently the negative sentiment key phrase is used in all customer reviews regardless of product class.

9. The method of claim 1, wherein the review search criteria further specify a desired product characteristic.

10. The method of claim 9, wherein the desired product characteristic is good service, quality, or affordability.

11. A computer program product, comprising:

a non-transitory computer-readable storage device having computer-readable program code embodied therein that when executed by a computer cause the computer to perform a method for providing product information based on user-generated product reviews, the computer-readable program code comprising;

computer-readable program code for receiving a request for product information defining at least a product or product category;

computer-readable program code for identifying product reviews matching the review search criteria to generate a review analysis set;

computer-readable program code for identifying negative sentiment key phrases in each review of the review analysis set;

computer-readable program code for calculating a weight for each identified negative sentiment key phrase by determining a frequency of the negative sentiment key phrase in the review analysis set and calculating an inverse of the frequency;

computer-readable program code for correlating the negative sentiment key phrases with product characteristics using a classification rubric, wherein the classification rubric defines which negative sentiment key phrases are associated with a product characteristic of a product class;

computer-readable program code for calculating for each product characteristic a score based on the weights of the negative sentiment key phrase or phrases correlated to the product characteristic;

computer-readable program code for generating an identity of each product reviewed in the review analysis set and, for each product, a summary of scored product characteristics; and

computer-readable program code for communicating at least a portion of the identities of each product reviewed in the review analysis set and the corresponding summary of scored product characteristics.

12. The computer program product of claim 11, wherein the classification rubric is derived by conducting a semantic analysis of a product review sample set to identify sentiment key phrases and assigning each sentiment key phrase to a product characteristic.

13. The computer program product of claim 12, wherein the semantic analysis is conducted using a supervised or unsupervised machine learning algorithm.

15. The computer program product of claim 12, wherein the sentiment key phrases are classified as negative or positive sentiment key phrases.

16. The computer program product of claim 13, wherein the positive sentiment key phrases are filtered out prior to assigning the sentiment key phrases to a product characteristic.

17. The computer program product of claim 11, wherein the relative frequency of the negative sentiment key phrase is calculated by determining how frequently the negative sentiment key phrase is used in customer reviews of all products of the same product class.

18. The computer program product of claim 11, wherein the relative frequency of a negative sentiment key phrase is calculated by determining how frequently the negative sentiment key phrase is used in all customer reviews regardless of product class.

14. The computer program product of 13, wherein the semantic analysis further identifies product characteristic categories for use in mapping the negative sentiment key words to a product characteristic in the classification rubric.

19. A system for providing product information based on identification of negative sentiment key phrases in customer reviews, the system comprising:

a storage device;

a network communication device; and

a processor communicatively coupled to the storage device and the network communication device, wherein the processor executes application code instructions that are stored in the storage device and that cause the system to;

receive a request for product information defining at least a product or product category;

identify product reviews matching the product or product category to generate a review analysis set;

identify negative sentiment key phrases in each review of the review analysis set;

calculate a weight for each identified negative sentiment key phrase by determining a frequency for the identified negative sentiment key phrase in the review analysis set and calculating an inverse of the frequency;

correlate the negative sentiment key phrases with product characteristics using a classification rubric, wherein the classification rubric defines which product characteristics are associated with a negative sentiment key phrase; and

calculate, for each product characteristic, a score based on the weights of the negative sentiment key phrases correlated to the product characteristic.

20. The system of claim 19, wherein the processor further executes application code instructions that are stored in the storage device and that cause the system to:

generate an identity for each product review in the review analysis set and, for each product, a summary of scored product characteristics; and

communicate at least a portion of the identities of each product reviewed in the review analysis set and the corresponding summary of scored product characteristics to a network device for display on the network device.

21. The system of claim 19, wherein the classification rubric is derived by conducting a semantic analysis of a product review sample set to identify sentiment key phrases and assigning each sentiment key phrase to a product characteristic.

23. The system of claim 19, wherein the classification rubric is derived by conducting a semantic analysis of a product review sample set to identify sentiment key phrases and assigning each sentiment key phrase to a product characteristic, and wherein the sentiment key phrases are classified as negative or positive sentiment key phrases.

24. The system of claim 19, wherein the positive sentiment key phrases are filtered out prior to assigning the sentiment key phrases to a product characteristic.

25. The system of claim 19, wherein the relative frequency of the negative sentiment key phrase is calculated by determining how frequently the negative sentiment key phrase is used in customer reviews of all products of the same product class.

26. The system of claim 19, wherein the relative frequency of a negative sentiment key phrase is calculated by determining how frequently the negative sentiment key phrase is used in all customer reviews regardless of product class.

22. The system of 19, wherein the semantic analysis further identifies product characteristic categories for use in mapping the negative sentiment key words to a product characteristic in the classification rubric.