Snippet extraction and ranking

• US 8,954,425 B2
• Filed: 06/08/2010
• Issued: 02/10/2015
• Est. Priority Date: 06/08/2010
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Corporation
• Inventors: Xiao, Rong, Hao, Qiang, Wang, Changhu, Cai, Rui, Zhang, Lei
• Primary Examiner(s): SYED, FARHAN M
• Application Number:
  US12/796,345
• Publication Number:
  US 20110302162A1
• Time in Patent Office: 1,708 Days
• US Class Current: 707/724
• CPC Class Codes: G06F 16/29 Geographical information databases
• CPC Class Codes: G06F 16/9535 Search customisation based on user profiles and personalisation
• CPC Class Codes: G06F 16/9536 Search customisation based on social or collaborative filtering

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 are described that facilitate automatically extracting snippets. These techniques include, for example, computer annotating travelogues with learned tags and images, performing topic learning to obtain an interest model, performing location matching based on the interest model, calculating geographic and semantic relevance scores, ranking snippets based on the geographic and semantic relevance scores, and searching snippets with a “location+context term” query.

Claims

1. A computer-readable storage medium having computer executable instructions encoded thereon, the computer executable instructions for execution by a processor to perform location-related mining operations on a travelogue, the operations comprising:

receiving a query comprising a particular location and a context term which describes the particular location;

identifying at least one snippet, the at least one snippet comprising a segment of user-generated content;

filtering out one or more biased context terms from the at least one snippet, the one or more biased context terms describing a plurality of locations;

extracting the particular location from the at least one snippet;

mining the context term which describes the particular location from the at least one snippet;

calculating a geographic relevance score for the at least one snippet based at least in part on the particular location;

calculating a semantic relevance score for the at least one snippet based at least in part on the context term which describes the particular location;

computing a snippet score based at least in part on the geographic relevance score and the semantic relevance score for the at least one snippet; and

ranking the at least one snippet and at least one other snippet based at least in part on the snippet score of the at least one snippet.

2. A computer-readable storage medium as recited in claim 1, wherein the context term which describes the particular location comprises one of a style word or a keyword corresponding to a local topic.

3. A computer-readable storage medium as recited in claim 1, wherein the query is received in a form of location-representative knowledge from an automated knowledge mining operation.

4. A computer-readable storage medium as recited in claim 1, the operations further comprising;

building a snippet collection from the at least one snippet and the at least one other snippet, wherein the at least one snippet and the at least one other snippet are assembled from one or more travelogues; and

generating, in response to the query, representative tags for the particular location.

5. A computer-readable storage medium as recited in claim 4, wherein building the snippet collection from one or more travelogues comprises:

identifying one or more particular locations mentioned in one or more travelogues;

segmenting the one or more travelogues into one or more snippets, each snippet containing at least one identified particular location;

determining whether a number of words of the one or more snippets meets a length constraint; and

compiling the snippet collection from the one or more snippets that meet the length constraint.

6. A computer-readable storage medium as recited in claim 5, wherein the length constraint comprises one or both of a minimum number of words or a maximum number of words.

7. A computer-readable storage medium as recited in claim 1, the operations further comprising enriching the travelogue with one or more images based at least in part on matching the particular location and the context term which describes the particular location.

8. A computer-readable storage medium as recited in claim 1, wherein the at least one snippet is represented by at least one particular location, words within the at least one snippet other than locations, and a snippet quality score.

17. A computer-readable storage medium as recited in claim 1, wherein calculating the geographic relevance score is based at least in part on how often the particular location appears in the at least one snippet.

18. A computer-readable storage medium as recited in claim 1, wherein mining the context term is performed with a latent Dirichlet allocation.

19. A computer-readable storage medium as recited in claim 1, wherein the filtering out is based at least in part on comparing the context terms to a list of biased context terms.

20. A computer-readable storage medium as recite in claim 19, wherein the list of biased context terms comprises context terms that correspond to at least one of transportation, accommodation, or opinion.

9. A computer-implemented method comprising:

identifying a travelogue for location-related mining;

decomposing the travelogue into a plurality of portions of the travelogue to identify a particular location and a context term, the context term characterizing the particular location without co-occurring with a plurality of locations;

filtering out one or more noisy context terms that do not characterize the particular location and co-occur with a plurality of locations from at least one portion of the travelogue;

extracting the particular location from the portion of the travelogue as decomposed;

mining the context term which characterizes the location from the portion of the travelogue as decomposed;

creating a geo-snippet comprising the particular location and the context term which characterizes the particular location;

obtaining a geographic relevance score corresponding to the geo-snippet;

obtaining a semantic relevance score corresponding to the geo-snippet; and

computing, by a processor, a snippet score corresponding to the geo-snippet based at least in part on the geographic relevance score and the semantic relevance score.

10. A computer-implemented method as recited in claim 9, further comprising ranking a plurality of geo-snippets based at least in part on the snippet score.

11. A computer-implemented method as recited in claim 9, further comprising building at least one tag from information comprising one or both of the location or the context term which characterizes the location.

12. A computer-implemented method performed on one or more travelogues comprising:

identifying at least one snippet from the one or more travelogues;

filtering out one or more noisy context terms from the at least one snippet, the noisy context terms co-occurring with a plurality of locations;

extracting a particular location from the at least one snippet;

mining a context term which characterizes the particular location from the at least one snippet;

calculating a geographic relevance score for the at least one snippet based at least in part on the particular location;

calculating a semantic relevance score for the at least one snippet based at least in part on the context term which characterizes the particular location;

computing a snippet score based at least in part on the geographic relevance score and the semantic relevance score for the at least one snippet;

ranking the at least one snippet based at least in part on the snippet score; and

assembling the at least one snippet into a snippet collection.

13. A method as recited in claim 12, wherein the semantic relevance score is calculated based at least in part on a sum of distances in a topic space from a query word to a plurality of unique words contained in the at least one snippet.

14. A method as recited in claim 12, wherein computing the snippet score based at least in part on the geographic relevance score and the semantic relevance score for the at least one snippet includes determining a quality score based at least in part on snippet length.

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

receiving the particular location and the context term which characterizes the particular location from a query; and

returning the at least one snippet ranked based at least in part on the snippet score responsive to the query.

16. A method as recited in claim 12, further comprising, where there is a plurality of snippets, normalizing the snippet score with a normalization factor that accounts for different lengths between the plurality of snippets.