Summarization of short comments

1. A system comprising:

a memory to store a plurality of comments, the plurality of comments respectively comprising an overall rating of an entity and at least one phrase, the at least one phrase comprising a head term and a modifier associated with the head term; and

one or more processors to implement:

an aspect module to map respective head terms of a portion of the plurality of comments to an aspect cluster corresponding to an attribute of the entity, and

a rating module to determine an aspect rating corresponding to the attribute of the entity based on the respective overall rating of the portion of the plurality of comments, and

a module to cause presentation, on a client machine, of a graphical representation of the determined aspect rating corresponding to the attribute of the entity.

2. The system of claim 1 , wherein the one or more processors are further to implement an extraction module to extract one or more phrases from the portion of the plurality of comments, the extracted phrases corresponding to the aspect cluster.

3. The system of claim 1 , wherein the aspect module maps the portion of the plurality of comments to the aspect cluster by decomposing the plurality of comments into a plurality of aspect clusters, wherein the aspect cluster is included in the plurality of aspect clusters.

4. The system of claim 1 , wherein the aspect module further comprises a k-means calculator to identify the aspect cluster using a k-means clustering algorithm.

5. The system of claim 1 , wherein the aspect module further comprises an unstructured probabilistic latent semantic analysis (PLSA) calculator to identify the aspect cluster using an unstructured PLSA algorithm.

6. The system of claim 1 , wherein the aspect module further comprises a structured probabilistic latent semantic analysis (PLSA) module to identify the aspect cluster using a structured PLSA algorithm.

7. The system of claim 1 , wherein the aspect module further comprises an aspect estimator to incorporate a topic model corresponding to the aspect cluster.

8. The system of claim 1 , further comprising an extraction module to extract at least one representative phrase from the portion of the plurality of comments.

9. The system of claim 1 , wherein the rating module further comprises a global predictor to determine the aspect rating based on respective modifiers of the head terms comprising the at least one phrase of respective comments of the comments collection.

10. The system of claim 1 , wherein the one or more processors are further to implement an evaluation module to evaluate the at least one phrase using a precision metric and a recall metric.

11. A method comprising:

using one or more computer processors, identifying a plurality of aspect clusters based on a plurality of comments, the plurality of comments respectively comprising an overall rating of an entity and at least one phrase, the identifying of the plurality of aspect clusters including:

identifying a head term in the at least one phrase of each comment in the plurality of comments, and

mapping each head term to one of a plurality of attributes of the entity, each aspect cluster of the plurality of aspect clusters corresponding to an attribute of the entity; and

determining an aspect rating corresponding to a first aspect cluster of the plurality of aspect clusters based on the overall rating of respective comments from a portion of the plurality of comments; and

causing presentation of a graphical representation of the aspect rating on a client machine.

12. The method of claim 11 , further comprising evaluating an aspect coverage of the plurality of aspect clusters.

13. The method of claim 11 , further comprising evaluating the plurality of aspect clusters using a user agreement on clustering accuracy.

14. The method of claim 11 , further comprising evaluating the aspect rating using an aspect rating correlation.

15. The method of claim 11 , further comprising evaluating the aspect rating using a ranking loss.

16. The method of claim ll, further comprising evaluating the at least one phrase using a precision metric and a recall metric.

17. The method of claim 11 , wherein the identifying of the plurality of the aspect clusters is performed using at least one of k-means clustering, unstructured probabilistic latent semantic analysis (PLSA), or structured PLSA.

18. The method of claim 11 , wherein the identifying of the plurality of the aspect clusters is based on a topic model corresponding to a respective aspect cluster.

19. The method of claim 11 , wherein the determining of the aspect rating is performed using local prediction or global prediction.

20. A computer-readable storage medium having no transitory signals and embodying instructions executable by a processor for performing operations comprising:

identifying a plurality of aspect clusters based on a plurality of comments, the plurality of comments respectively comprising an overall rating of an entity and at least one phrase, the identifying of the plurality of aspect clusters including:

identifying a head term in the at least one phrase of each comment in the plurality of comments, and

mapping each head term to one of a plurality of attributes of the entity, each aspect cluster of the plurality of aspect clusters corresponding to an attribute of the entity; and

determining an aspect rating corresponding to a first aspect cluster of the plurality of aspect clusters based on the overall rating of respective comments from a portion of the plurality of comments; and

causing presentation of a graphical representation of the aspect rating on a client machine.