Mining topic-related aspects from user generated content

• US 8,458,115 B2
• Filed: 06/08/2010
• Issued: 06/04/2013
• Est. Priority Date: 06/08/2010
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Corporation
• Inventors: Cai, Rui, Hao, Qiang, Wang, Changhu, Xiao, Rong, Zhang, Lei
• Primary Examiner(s): Holmes, Michael B
• Application Number:
  US12/796,303
• Publication Number:
  US 20110302124A1
• Time in Patent Office: 1,092 Days
• Field of Search: 706/52
• US Class Current: 706/52
• CPC Class Codes: G06F 16/353 into predefined classes

Abstract

Described herein is a technology that facilitates efficient automated mining of topic-related aspects of user generated content based on automated analysis of the user generated content. Locations are automatically learned based on dividing documents into document segments, and decomposing the segments into local topics and global topics. Techniques described herein include, for example, computer annotating travelogues with learned tags, performing topic learning to obtain an interest model, and performing location matching based on the interest model.

Claims

1. A memory having computer executable instructions encoded thereon, the computer executable instructions executed by a processor to perform location-related mining operations, the operations comprising:

identifying a particular travelogue;

decomposing the particular travelogue by identifying at least two non-overlapping segments of the particular travelogue, each segment including a representation of at least one location;

representing a collection of travelogues with a term-document matrix, the collection of travelogues comprising the particular travelogue, and each word of the particular travelogue representing;

a location, a local topic, and a term in a sequence;

or

a global topic and a term in a sequence;

using a probabilistic topic model, decomposing the term-document matrix into one or more matrices comprising;

a term-local topic matrix;

a local topic-location matrix;

or

a location-document matrix; and

representing a particular location by a multinomial distribution over local topics while associating a document with a multinomial distribution over global topics.

2. A memory as recited in claim 1, wherein the one or more matrices further comprise at least one of a term-global topic matrix or a global topic-document matrix.

3. A memory as recited in claim 1, wherein words in a segment share a multinomial distribution over locations, the multinomial distribution being affected by a Dirichlet prior derived from locations extracted from the segment.

4. A memory as recited in claim 1, wherein the decomposing the term-document matrix further comprises estimating parameters for the probabilistic topic model.

5. A memory as recited in claim 1, further comprising based at least on the multinomial distribution of the particular location, enriching a travelogue browsing experience by associating one or more images with the particular location.

6. A method comprising:

identifying a travelogue for location-related mining;

decomposing the travelogue;

representing a decomposed travelogue with a term-document matrix, wherein each word from the travelogue represents one of;

a local topic;

or

a global topic;

selecting a candidate set of travelogues based at least on the local topic;

ranking the travelogues in the candidate set of travelogues based at least on the local topic; and

returning travelogues in the candidate set of travelogues based at least on the ranking.

7. A method as recited in claim 6, wherein a file for the location-related mining comprises local topic words and global topic words.

8. A method as recited in claim 6, wherein a location is identified based at least on a part of an index of a configurable number of local topic words.

9. A method as recited in claim 6, wherein the candidate set of travelogues comprises a compound representation of travelogues having undergone dimension reduction.

10. A method as recited in claim 6, further comprising determining a location associated with the local topic.

11. A method as recited in claim 6, further comprising determining a travelogue topic corresponding to each travelogue of the candidate set of travelogues.

12. A method as recited in claim 6, further comprising:

determining a location-related topic corresponding to at least one of the local topic;

determining a travelogue topic corresponding to each travelogue of the candidate set of travelogues; and

searching among the travelogues topics corresponding to each of the candidate set of travelogues to find one or more travelogue topics similar to the location-related topic; and

selecting those travelogues from the candidate set of travelogues having travelogue topics corresponding to the location-related topic.

13. A method as recited in claim 6, further comprising providing recommendations based at least on the ranking.

14. A memory having computer-executable instructions encoded thereon to program a computer to perform the method of claim 6 upon execution.

15. A computer-implemented method comprising:

in response to Internet browsing activities, identifying a collection of user generated content;

searching an image library for images having associated descriptive data that is similar to text in the collection of user generated content;

processing the descriptive data of the images to derive a topic for the collection of user generated content;

selecting a recommendation based at least in part on the topic derived; and

in further response to the Internet browsing activities, presenting the recommendation.

16. A method as recited in claim 15, wherein the Internet browsing activities comprise blogging.

17. A method as recited in claim 15, wherein user generated content comprises one or more travelogues.

18. A method as recited in claim 15, wherein processing the descriptive data comprises searching a vertical topic space based on the descriptive data.

19. A method as recited in claim 15, wherein searching the image library comprises:

deriving image content information from a plurality of entries of user generated content that have associated images in the collection of user generated content; and

comparing the image content information derived to image content information of images in the image library.

20. A method as recited in claim 15, wherein processing the descriptive data of the images comprises:

mining the descriptive data of the images for one or more local topics; and

applying the one or more local topics to derive at least one location for the collection of user generated content.