Verifying relevance between keywords and web site contents

• US 7,260,568 B2
• Filed: 04/15/2004
• Issued: 08/21/2007
• Est. Priority Date: 04/15/2004
• Status: Expired due to Fees

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Corporation
• Inventors: Ma, Wei-Ying, Chen, Zheng, Najm, Tarek, Li, Li, Zhang, Benyu, Li, Ying, Zeng, Hua-Jun
• Primary Examiner(s): FLEURANTIN, JEAN B
• Application Number:
  US10/826,162
• Publication Number:
  US 20050234953A1
• Time in Patent Office: 1,223 Days
• Field of Search: 707 1- 10, 707100-1041, 707200-206, 382/218, 382/220, 382/225
• US Class Current: 707/711
• CPC Class Codes: G06F 16/3334 Selection or weighting of terms from queries, including natural language queries
• CPC Class Codes: G06F 16/3346 using probabilistic model
• CPC Class Codes: Y10S 707/99933 Query processing, i.e. searching
• CPC Class Codes: Y10S 707/99934 Query formulation, input preparation, or translation
• CPC Class Codes: Y10S 707/99935 Query augmenting and refining, e.g. inexact access

Abstract

Systems and methods for verifying relevance between terms and Web site contents are described. In one aspect, site contents from a bid URL are retrieved. Expanded term(s) semantically and/or contextually related to bid term(s) are calculated. Content similarity and expanded similarity measurements are calculated from respective combinations of the bid term(s), the site contents, and the expanded terms. Category similarity measurements between the expanded terms and the site contents are determined in view of a trained similarity classifier. The trained similarity classifier having been trained from mined web site content associated with directory data. A confidence value providing an objective measure of relevance between the bid term(s) and the site contents is determined from the content, expanded, and category similarity measurements evaluating the multiple similarity scores in view of a trained relevance classifier model.

Claims

1. A method for verifying relevance between terms and Web site contents, the method comprising:

retrieving site contents from a bid URL;

formulating expanded term(s) comprising at least one of semantically or contextually related to bid term(s), which are mined from a search engine in view of high-frequency of occurrence historical query terms;

generating content similarity and expanded similarity measurements from respective combinations of the bid term(s), the site contents, and the expanded terms, wherein the similarity measurements indicate relatedness between respective ones of the bid term(s), site contents, or expanded terms;

calculating category similarity measurements between the expanded terms and the site contents in view of a similarity classifier, wherein the similarity classifier has been trained from mined web site content associated with directory data;

calculating a confidence value from combined ones of multiple similarity measurements, wherein the combined ones comprise content, expanded, and category similarity measurements, wherein the confidence value provides objective measure of relevance between the bid term(s) and the site contents;

analyzing the confidence value to identify the bid term(s); and

using the bid term(s) identified to increase traffic to a site to obtain site exposure;

wherein generating the category similarity measurements further comprises;

extracting features from Web site content associated with the directory data, the features comprising a combination of at least one of title, metadata, body, hypertext link(s), visual feature(s), and summarization by page layout analysis information;

reducing dimensionality of the features via feature selection;

categorizing the features via a classifier model to generate the similarity classifier;

generating respective term vectors from the bid term(s), the site contents, and the expanded terms; and

calculating similarity between the respective term vectors as a function of the similarity classifier to determine the category similarity measurements.

2. A method as recited in claim 1, wherein the similarity classifier is based on a statistical n-gram based naï

ve Bayesian (N-Gram), a naï

ve Bayesian (NB), support vector machine (SVM), a nearest neighbor (KNN), a decision tree, a co-training, or a boosting classification model.

3. A method as recited in claim 1, wherein formulating the expanded terms further comprises generating term clusters from term vectors based on calculated term similarity, the term vectors being generated from historical queries, each historical query having a high frequency of occurrence, the term clusters comprising the expanded terms.

4. A method as recited in claim 1, wherein generating the content similarity measurements further comprise generating respective term vectors from the bid term(s) and the site contents, and calculating similarity between the respective term vectors to determine direct similarity between the bid term(s) and the site contents.

5. A method as recited in claim 1, wherein generating the expanded similarity measurements further comprises:

generating respective term vectors from the bid term(s), the site contents, and the expanded terms; and

calculating similarity between the respective term vectors to determine the expanded similarity measurements between the bid term(s) and the site contents.

6. A method as recited in claim 1, wherein calculating the confidence value further comprises:

training a combined relevance classifier with data of the form <

term(s), Web site content, accept/reject>

in view of an accept/reject threshold;

generating relevance verification similarity measurement (RSVM) feature vectors from the content, expanded, and category similarity measurements; and

mapping multiple scores from the RSVM feature vectors to the confidence value via the combined relevance classifier.

7. A method as recited in claim 1, wherein the method further comprises:

caching the bid term(s) and bid URL into a bidding database;

responsive to receipt of an search query, determining if terms of the search query are relevant to the bid term(s) in view of a possibility that the terms of the search query may not exactly match the bid term(s); and

if the term(s) of search query are determined to be relevant to the bid term(s), communicating the bid URL to the end-user.

8. A method as recited in claim 1, wherein the method further comprises:

determining proper name similarity measurements from the bid term(s) and site contents, the proper name similarity measurements indicating relatedness between any proper name(s) detected in the bid term(s) and the site contents in view a set of proper names; and

wherein the combined ones of multiple similarity measurements comprise the proper name similarity measurements.

9. A method as recited in claim 8, wherein determining the proper name similarity measurements further comprises:

responsive to detecting a proper name comprising at least one of the bid term(s) or the site contents, calculating a proper name similarity score as;

Prop_Sim(term, site contents),

wherein Prop_Sim(term, site contents) equals;

one (1) when a term contains a proper name P, and site contents contains a comfortable proper name Q;

zero (0) when a term contains a proper name P, and site contents contains only unconformable proper name(s);

or, zero-point-five (0.5).

10. A method as recited in claim 1, wherein the method further comprises:

determining that the confidence value is relatively low; and

responsive to the determining, identifying one or more other terms comprising at least one of,semantically or contextually related to the bid URL.

11. A method as recited in claim 10, wherein identifying further comprises:

generating a set of term clusters from term vectors based on calculated term similarity, the term vectors being generated from search engine results of submitted historical queries, each historical query having a relatively low frequency of occurrence as compared to other query terms in a query log; and

evaluating the site contents in view of term(s) specified by the term clusters to identify at least one or more semantically or contextually related terms, the terms being the one or more other terms.

12. A computer-readable storage medium comprising computer-executable instructions for verifying relevance between terms and Web site contents, the computer-executable instructions comprising instructions for:

retrieving site contents from a bid URL;

formulating expanded term(s) comprising at least one of semantically or contextually related to bid term(s), which are mined from a search engine in view of high-frequency of occurrence historical query terms;

generating content similarity and expanded similarity measurements from respective combinations of the bid term(s), the site contents, and the expanded terms, wherein the similarity measurements indicate relatedness between respective ones of the bid term(s), site contents, or expanded terms;

calculating category similarity measurements between the expanded terms and the site contents in view of a similarity classifier, wherein the similarity classifier has been trained from mined web site content associated with directory data;

calculating a confidence value from combined ones of multiple similarity measurements, wherein the combined ones comprise content, expanded, and category similarity measurements;

providing an objective measure of relevance between the bid term(s) and the site contents as indicated by the confidence value;

analyzing the confidence value to identify the bid term(s); and

using the bid term(s) identified to increase traffic to a site to obtain site exposure;

wherein the computer-executable instructions for generating the category similarity measurements further comprise instructions for;

extracting features from Web site content associated with the directory data, the features comprising a combination at least one of title, metadata, body, hypertext link(s), visual feature(s), and summarization by page layout analysis information;

reducing dimensionality of the features via feature selection;

categorizing the features via a classifier model to generate the similarity classifier;

generating respective term vectors from the bid term(s), the site contents, and the expanded terms; and

calculating similarity between the respective term vectors as a function of the similarity classifier to determine the category similarity measurements.

13. A computer-readable storage medium as recited in claim 12, wherein the similarity classifier is based on a statistical n-gram based naï

ve Bayesian (N-Gram), a naï

ve Bayesian (NB), support vector machine (SVM), a nearest neighbor (KNN), a decision tree, a co-training, or a boosting classification model.

14. A computer-readable storage medium as recited in claim 12, wherein the computer-executable instructions for formulating the expanded terms further comprise instructions for generating term clusters from term vectors based on calculated term similarity, the term vectors being generated from historical queries, each historical query having a high frequency of occurrence, the term clusters comprising the expanded terms.

15. A computer-readable storage medium as recited in claim 12, wherein the computer-executable instructions for generating the content similarity measurements further comprise instructions for generating respective term vectors from the bid term(s) and the site contents, and calculating similarity between the respective term vectors to determine direct similarity between the bid term(s) and the site contents.

16. A computer-readable medium storage as recited in claim 12, wherein the computer-executable instructions for generating the expanded similarity measurements further comprise instructions for:

generating respective term vectors from the bid term(s), the site contents, and the expanded terms; and

calculating similarity between the respective term vectors to determine the expanded similarity measurements between the bid term(s) and the site contents.

17. A computer-readable storage medium as recited in claim 12, wherein the computer-executable instructions for calculating the confidence value further comprise instructions for:

training a combined relevance classifier with data of the form <

term(s), Web site content, accept/reject>

in view of an accept/reject threshold;

generating relevance verification similarity measurement (RSVM) feature vectors from the content, expanded, and category similarity measurements; and

mapping multiple scores from the RSVM feature vectors to the confidence value via the combined relevance classifier.

18. A computer-readable storage medium as recited in claim 12, wherein the computer-executable instructions further comprise instructions for:

caching the bid term(s) and bid URL into a bidding database;

responsive to receipt of an search query, determining if terms of the search query are relevant to the bid term(s) in view of a possibility that the terms of the search query may not exactly match the bid term(s); and

if the term(s) of search query are determined to be relevant to the bid term(s), communicating the bid URL to the end-user.

19. A computer-readable storage medium as recited in claim 12, wherein the computer-executable instructions further comprise instructions for:

determining proper name similarity measurements from the bid term(s) and site contents, the proper name similarity measurements indicating relatedness between any proper name(s) detected in the bid term(s) and the site contents in view a set of proper names; and

wherein the combined ones of multiple similarity measurements comprise the proper name similarity measurements.

20. A computer-readable storage medium as recited in claim 19, wherein the computer-executable instructions for determining the proper name similarity measurements further comprise instructions for:

responsive to detecting a proper name comprising at least one of the bid term(s) or the site contents, calculating a proper name similarity score as;

Prop_Sim(term, site contents) and

wherein Prop_Sim(term, site contents) equals;

one (1) when a term contains a proper name P, and site contents contains a conformable proper name Q;

zero (0) when a term contains a proper name P, and site contents contains only unconformable proper name(s);

or, zero-point-five (0.5).

21. A computer-readable storage medium as recited in claim 12, wherein the computer-executable instructions further comprise instructions for:

determining that the confidence value is relatively low; and

responsive to the determining, identifying one or more other terms that are semantically and/or contextually related to the bid URL.

22. A computer-readable storage medium as recited in claim 21, wherein the computer executable instructions for identifying further comprise instructions for:

generating a set of term clusters from term vectors based on calculated term similarity, the term vectors being generated from search engine results of submitted historical queries, each historical query having a relatively low frequency of occurrence as compared to other query terms in a query log; and

evaluating the site contents in view of term(s) specified by the term clusters to identify one or more semantically and/or contextually related terms, the terms being the one or more other terms.

23. A computing device for verifying relevance between terms and Web site contents, the computing device comprising:

a processor; and

a memory coupled to the processor, the memory comprising computer-program instructions executable by the processor for;

retrieving site contents from a bid URL;

formulating expanded term(s) comprising at least one of semantically or contextually related to bid term(s),

generating content similarity and expanded similarity measurements from respective combinations of the bid term(s), the site contents, and the expanded terms, wherein the similarity measurements indicate relatedness between respective ones of the bid term(s), site contents;

or expanded terms;

calculating a confidence value from combined ones of multiple similarity measurements, wherein the combined ones comprise content, expanded, and category similarity measurements;

providing an objective measure of relevance between the bid term(s) and the site contents as indicated by the confidence value;

analyzing the confidence value to identify the bid term(s); and

using the bid term(s) identified to increase traffic to a site to obtain site exposure;

wherein the computer-executable instructions for generating the category similarity measurements further comprise instructions for;

extracting features from web site content associated with the directory data, the features comprising a combination of at least one of title, metadata, body, hypertext link(s), visual feature(s), and summarization by page layout analysis information;

reducing dimensionality of the features via feature selection;

categorizing the features via a classifier model to generate the similarity classifier;

generating respective term vectors from the bid term(s), the site contents, and the expanded terms; and

calculating similarity between the respective term vectors as a function of the similarity classifier to determine the category similarity measurements.

24. A computing device as recited in claim 23, wherein the similarity classifier is based on a statistical n-gram based naï

ve Bayesian (N-Gram), a naï

ve Bayesian (NB), support vector machine (SVM), a nearest neighbor (KNN), a decision tree, a co-training, or a boosting classification model.

25. A computing device as recited in claim 23, wherein the computer-executable stored instructions for formulating the expanded terms further comprise instructions for generating term clusters from term vectors based on calculated term similarity, the term vectors being generated from historical queries, each historical query having a high frequency of occurrence, the term clusters comprising the expanded terms.

26. A computing device as recited in claim 23, wherein the computer-executable stored instructions for generating the content similarity measurements further comprise instructions for generating respective term vectors from the bid term(s) and the site contents, and calculating similarity between the respective term vectors to determine direct similarity between the bid term(s) and the site contents.

27. A computing device as recited in claim 23, wherein the computer-executable stored instructions for generating the expanded similarity measurements further comprise instructions for:

generating respective term vectors from the bid term(s), the site contents, and the expanded terms; and

calculating similarity between the respective term vectors to determine the expanded similarity measurements between the bid term(s) and the site contents.

28. A computing device as recited in claim 23, wherein the computer-executable stored instructions for calculating the confidence value further comprise instructions for:

training a combined relevance classifier with data of the form <

term(s), Web site content, accept/reject>

in view of an accept/reject threshold;

generating relevance verification similarity measurement (RSVM) feature vectors from the content, expanded, and category similarity measurements; and

mapping multiple scores from the RSVM feature vectors to the confidence value via the combined relevance classifier.

29. A computing device as recited in claim 23, wherein the computer-executable stored instructions further comprise instructions for:

determining proper name similarity measurements from the bid term(s) and site contents, the proper name similarity measurements indicating relatedness between any proper name(s) detected in the bid term(s) and the site contents in view a set of proper names; and

wherein the combined ones of multiple similarity measurements comprise the proper name similarity measurements.

30. A computing device as recited in claim 29, wherein the computer-executable stored instructions for determining the proper name similarity measurements further comprise instructions for:

responsive to detecting a proper name comprising at least one of the bid term(s) or the site contents, calculating a proper name similarity score as;

Prop_Sim(term, site contents) and

wherein Prop_Sim(term, site contents) equals;

one (1) when a term contains a proper name P, and site contents contains a conformable proper name Q;

zero (0) when a term contains a proper name P, and site contents contains only unconformable proper name(s);

or, zero-point-five (0.5).

31. A computing device as recited in claim 23, wherein the computer-executable stored instructions further comprise instructions for:

determining that the confidence value is relatively low; and

responsive to the determining, identifying one or more other terms comprising at least one of semantically or contextually related to the bid URL.

32. A computing device as recited in claim 31, wherein the computer-executable stored instructions for identifying further comprise instructions for:

generating a set of term clusters from term vectors based on calculated term similarity, the term vectors being generated from search engine results of submitted historical queries, each historical query having a relatively low frequency of occurrence as compared to other query terms in a query log; and

evaluating the site contents in view of term(s) specified by the term clusters to identify at least one or more semantically or contextually related terms, the terms being the one or more other terms.

33. A computing device for verifying relevance between terms and Web site contents, the computing device comprising:

retrieving means to obtain site contents from a bid URL;

formulating means to identify expanded term(s) comprising at least one of semantically or contextually related to bid term(s),

generating means to create content similarity and expanded similarity measurements from respective combinations of the bid term(s), the site contents, and the expanded terms, wherein the similarity measurements indicating indicate relatedness between respective ones of the bid term(s), site contents, or expanded terms;

calculating means to determine category similarity measurements between the expanded terms and the site contents in view of a similarity classifier, wherein the similarity classifier has been trained from mined web site content associated with directory data;

calculating means to generate a confidence value from combined ones of multiple similarity measurements, wherein the combined ones comprise content, expanded, and category similarity measurements, wherein the confidence value provides an objective measure of relevance between the bid term(s) and the site contents;

analyzing means to analyze the confidence value to identify the bid term(s); and

increasing means to increase traffic to a site by using the bid term(s) identified;

wherein the generating means further comprise;

extracting means to obtain features from Web site content associated with the directory data, the features comprising a combination of at least one of title, metadata, body, hypertext link(s), visual feature(s), and summarization by page layout analysis information;

reducing means to lessen dimensionality of the features via feature selection;

categorizing means to organize the features via a classifier model to generate the similarity classifier;

generating means to create respective term vectors from the bid term(s), the site contents, and the expanded terms; and

calculating means to identify similarity between the respective term vectors as a function of the similarity classifier to determine the category similarity measurements.

34. A computing device as recited in claim 33, wherein the computer formulating means further comprise generating means to create term clusters from term vectors based on calculated term similarity, the term vectors being generated from historical queries, each historical query having a high frequency of occurrence, the terms clusters comprising the expanded terms.

35. A computing device as recited in claim 33, wherein the generating means further comprise creating means to generate respective term vectors from the bid term(s) and the site contents, and calculating similarity between the respective term vectors to determine direct similarity between the bid term(s) and the site contents.

36. A computing device as recited in claim 33, wherein the generating means further comprise:

creating means to generate respective term vectors from the bid term(s), the site contents, and the expanded terms; and

calculating means to determine similarity between the respective term vectors to determine the expanded similarity measurements between the bid term(s) and the site contents.

37. A computing device as recited in claim 33, wherein the calculating means further comprise:

training means to train a combined relevance classifier with data of the form <

term(s), Web site content, accept/reject>

in view of an accept/reject threshold;

generating means to generate relevance verification similarity measurement (RSVM) feature vectors from the content, expanded, and category similarity measurements; and

mapping means to correlate multiple scores from the RSVM feature vectors to the confidence value via the combined relevance classifier.

38. A computing device as recited in claim 33, wherein the computing device further comprises:

determining means to determine proper name similarity measurements from the bid term(s) and site contents, the proper name similarity measurements indicating relatedness between any proper name(s) detected in the bid term(s) and the site contents in view a set of proper names; and

wherein the combined ones of multiple similarity measurements comprise the proper name similarity measurements.

39. A computing device as recited in claim 38, wherein the determining means to determine the proper name similarity measurements further comprise responsive to detecting a proper name comprising at least one of the bid term(s) or the site contents, calculating means to calculate a proper name similarity score.

40. A computing device as recited in claim 33, wherein the computing device further comprises:

determining means to determine that the confidence value is relatively low; and

responsive to the determining, identifying means to identify one or more other terms comprising at least one of semantically or contextually related to the bid URL.

41. A computing device as recited in 40, wherein the identifying means further comprise:

generating means to generate a set of term clusters from term vectors based on calculated term similarity, the term vectors being generated from search engine results of submitted historical queries, each historical query having a relatively low frequency of occurrence as compared to other query terms in a query log; and

evaluating means to evaluate the site contents in view of lean(s) specified by the term clusters to identify at least one or more semantically or contextually related terms, the terms being the one or more other terms.