Aspect-based sentiment summarization

• US 8,799,773 B2
• Filed: 03/19/2008
• Issued: 08/05/2014
• Est. Priority Date: 01/25/2008
• Status: Active Grant

• Current Assignee: Google LLC (Alphabet Inc.)
• Original Assignee: Google Inc. (Alphabet Inc.)
• Inventors: Reis, George, Blair-Goldensohn, Sasha, McDonald, Ryan T.
• Primary Examiner(s): Bashore, William
• Assistant Examiner(s): SHREWSBURY, NATHAN K
• Application Number:
  US12/051,798
• Publication Number:
  US 20090193328A1
• Time in Patent Office: 2,330 Days
• Field of Search: 715255-256, 715/200, 715/772, 715/821, 715/254
• US Class Current: 715/254
• CPC Class Codes: G06F 16/345 Summarisation for human users
• CPC Class Codes: G06N 5/025 Extracting rules from data

Abstract

Phrases in the reviews that express sentiment about a particular aspect are identified. Reviewable aspects of the entity are also identified. The reviewable aspects include static aspects that are specific to particular types of entities and dynamic aspects that are extracted from the reviews of a specific entity instance. The sentiment phrases are associated with the reviewable aspects to which the phrases pertain. The sentiment expressed by the phrases associated with each aspect is summarized, thereby producing a summary of sentiment associated with each reviewable aspect of the entity. The summarized sentiment and associated phrases can be stored and displayed to a user as a summary description of the entity.

Claims

1. A method of summarizing sentiment expressed by reviews of an entity, comprising:

identifying sentiment phrases in the reviews expressing sentiment about the entity;

identifying reviewable aspects of the entity, wherein identifying reviewable aspects of the entity includes identifying one or more static aspects of the entity based on a type of the entity and identifying one or more dynamic aspects of the entity based on text of one or more of the reviews;

associating the sentiment phrases with the reviewable aspects of the entity to which the sentiment phrases pertain;

summarizing the sentiment expressed by the sentiment phrases associated with the reviewable aspects of the entity;

receiving a request for the summarized sentiment for the entity;

selecting one or more reviewable aspects of the entity, the selected reviewable aspects a subset of the identified reviewable aspects and being selected to identify the more relevant of the identified reviewable aspects;

wherein the selecting the one or more reviewable aspects of the entity includes selecting from the static aspects based on first criteria and selecting from the dynamic aspects based on second criteria, the second criteria unique from the first criteria;

wherein a given reviewable aspect of the selected reviewable aspects is selected based at least in part on how many unique user opinion sources provided the sentiment phrases associated with the given reviewable aspect; and

storing the selected reviewable aspects in response to the received request.

2. The method of claim 1, wherein identifying sentiment phrases in the reviews expressing sentiment about the entity comprises:

identifying syntactically coherent phrases in the reviews of the entity; and

generating sentiment scores for the syntactically coherent phrases, a sentiment score for a phrase representing a polarity and magnitude of sentiment expressed by the phrase.

3. The method of claim 1, wherein identifying reviewable aspects of the entity comprises:

identifying an entity type associated with the entity; and

identifying a set of reviewable aspects of the entity responsive at least in part to the entity type.

4. The method of claim 1, wherein identifying reviewable aspects of the entity comprises:

identifying n-grams appearing with a high-frequency in the reviews of the entity; and

filtering the identified n-grams to produce a set of reviewable aspects for the entity.

5. The method of claim 1, wherein associating the sentiment phrases with the reviewable aspects of the entity to which the sentiment phrases pertain comprises:

using a classifier to classify the sentiment phrases with the reviewable aspects of the entity to which the sentiment phrases pertain.

6. The method of claim 5, wherein using the classifier comprises:

selecting a set of random phrases expressing sentiment;

labeling the phrases in the random set to identify the reviewable aspects of the entity to which the phrases in the random set pertain;

using the labeled phrases to train the classifier to predict the reviewable aspects of the entity to which the sentiment phrases pertain.

7. The method of claim 1, wherein associating the sentiment phrases with the reviewable aspects of the entity to which the sentiment phrases pertain comprises:

determining whether a sentiment phrase mentions a reviewable aspect of the entity; and

associating the sentiment phrase with the reviewable aspect responsive to a positive determination that the sentiment phrase mentions the reviewable aspect of the entity.

8. The method of claim 1, wherein summarizing the sentiment expressed by the sentiment phrases associated with the reviewable aspects of the entity comprises:

identifying the sentiment phrases associated with a reviewable aspect of the entity;

scoring the sentiment expressed by each of the sentiment phrases associated with the reviewable aspect of the entity; and

determining a summary score for the reviewable aspect of the entity responsive at least in part to the scored sentiment expressed by each of the sentiment phrases associated with the reviewable aspect of the entity.

9. The method of claim 8, further comprising:

mapping the summary score to a rating for the reviewable aspect of the entity.

10. The method of claim 1, further comprising:

selecting one or more of the sentiment phrases for display in association with the selected reviewable aspects; and

generating a display of the selected reviewable aspects and the selected sentiment phrases in response to the received request.

11. The method of claim 10, wherein selecting one or more of the sentiment phrases for display in association with the selected reviewable aspects comprises:

selecting a subset of sentiment phrases associated with the reviewable aspect that express a representative sample of the sentiment expressed by a full set of sentiment phrases associated with the reviewable aspect.

12. A non-transitory computer-readable storage medium having computer-executable code encoded therein for summarizing sentiment expressed by reviews of an entity, comprising:

a sentiment classification module configured to identify sentiment phrases in the reviews expressing sentiment about the entity;

an aspect module configured to identify reviewable aspects of the entity, wherein the aspect module identifies one or more static aspects of the entity based on a type of the entity and identifies one or more dynamic aspects of the entity based on text of one or more of the reviews;

an association module configured to associate the sentiment phrases with the reviewable aspects of the entity to which the sentiment phrases pertain;

a summary module configured to summarize the sentiment expressed by the sentiment phrases associated with the reviewable aspects of the entity; and

a sentiment display engine configured to;

receive a request for the summarized sentiment for the entity;

select one or more reviewable aspects of the entity, the selected reviewable aspects a subset of the identified reviewable aspects and being selected to identify the more relevant of the identified reviewable aspects;

wherein the selecting the one or more reviewable aspects of the entity includes selecting from the static aspects based on first criteria and selecting from the dynamic aspects based on second criteria, the second criteria unique from the first criteria;

wherein a given reviewable aspect of the selected reviewable aspects is selected based at least in part on how many unique user opinion sources provided the sentiment phrases associated with the given reviewable aspect; and

store the selected reviewable aspects in response to the received request.

13. The computer-readable storage medium of claim 12, wherein identifying reviewable aspects of the entity comprises:

identifying syntactically coherent phrases in the reviews of the entity; and

generating sentiment scores for the syntactically coherent phrases, a sentiment score for a phrase representing a polarity and magnitude of sentiment expressed by the phrase.

14. The computer-readable storage medium of claim 12, wherein identifying reviewable aspects of the entity comprises:

identifying an entity type associated with the entity; and

identifying a set of reviewable aspects of the entity responsive at least in part to the entity type.

15. The computer-readable storage medium of claim 12, wherein identifying reviewable aspects of the entity comprises:

identifying n-grams appearing with a high-frequency in the reviews of the entity; and

filtering the identified n-grams to produce a set of reviewable aspects for the entity.

16. The computer-readable storage medium of claim 12, wherein associating the sentiment phrases with the reviewable aspects of the entity to which the sentiment phrases pertain comprises:

using a classifier to classify the sentiment phrases with the reviewable aspects of the entity to which the sentiment phrases pertain.

17. The computer-readable storage medium of claim 16, wherein using the classifier comprises:

selecting a set of random phrases expressing sentiment;

labeling the phrases in the random set to identify the reviewable aspects of the entity to which the phrases in the random set pertain;

using the labeled phrases to train the classifier to predict the reviewable aspects of the entity to which the sentiment phrases pertain.

18. The computer-readable storage medium of claim 12, wherein associating the sentiment phrases with the reviewable aspects of the entity to which the sentiment phrases pertain comprises:

determining whether a sentiment phrase mentions a reviewable aspect of the entity; and

associating the sentiment phrase with the reviewable aspect responsive to a positive determination that the sentiment phrase mentions the reviewable aspect of the entity.

19. The computer-readable storage medium of claim 12, wherein summarizing the sentiment expressed by the sentiment phrases associated with the reviewable aspects of the entity comprises:

identifying the sentiment phrases associated with a reviewable aspect of the entity;

scoring the sentiment expressed by each of the sentiment phrases associated with the reviewable aspect of the entity; and

determining a summary score for the reviewable aspect of the entity responsive at least in part to the scored sentiment expressed by each of the sentiment phrases associated with the reviewable aspect of the entity.

20. The computer-readable storage medium of claim 19, wherein the summary module is further configured to map the summary score to a rating for the reviewable aspect of the entity.

21. A computer-implemented system for summarizing sentiment expressed by reviews of an entity, comprising:

a non-transitory readable storage medium having computer-executable code encoded therein, the code comprising;

a sentiment classification module configured to identify sentiment phrases in the reviews expressing sentiment about the entity, wherein the aspect module identifies one or more static aspects of the entity based on a type of the entity and identifies one or more dynamic aspects of the entity based on text of one or more of the reviews;

an aspect module configured to identify reviewable aspects of the entity;

an association module configured to associate the sentiment phrases with the reviewable aspects of the entity to which the sentiment phrases pertain; and

a summary module configured to summarize the sentiment expressed by the sentiment phrases associated with the reviewable aspects of the entity; and

a sentiment display engine configured to;

receive a request for the summarized sentiment for the entity;

select one or more reviewable aspects of the entity, the selected reviewable aspects a subset of the identified reviewable aspects and being selected to identify the more relevant of the identified reviewable aspects;

wherein the selecting the one or more reviewable aspects of the entity includes selecting from the static aspects based on first criteria and selecting from the dynamic aspects based on second criteria, the second criteria unique from the first criteria; and

wherein a given reviewable aspect of the selected reviewable aspects is selected based at least in part on how many unique user opinion sources provided the sentiment phrases associated with the given reviewable aspect; and

store the selected reviewable aspects in response to the received request; and

a processor configured to execute the code stored in the storage medium.

22. The system of claim 21, wherein identifying reviewable aspects of the entity comprises:

identifying n-grams appearing with a high-frequency in the reviews of the entity; and

filtering the identified n-grams to produce a set of reviewable aspects for the entity.

23. The system of claim 21, wherein associating the sentiment phrases with the reviewable aspects of the entity to which the sentiment phrases pertain comprises:

determining whether a sentiment phrase mentions a reviewable aspect of the entity; and

associating the sentiment phrase with the reviewable aspect responsive to a positive determination that the sentiment phrase mentions the reviewable aspect of the entity.

24. The system of claim 21, wherein summarizing the sentiment expressed by the sentiment phrases associated with the reviewable aspects of the entity comprises:

identifying the sentiment phrases associated with a reviewable aspect of the entity;

scoring the sentiment expressed by each of the sentiment phrases associated with the reviewable aspect of the entity; and

determining a summary score for the reviewable aspect of the entity responsive at least in part to the scored sentiment expressed by each of the sentiment phrases associated with the reviewable aspect of the entity.