CONTEXT TRANSFER IN SEARCH ADVERTISING

• US 20100161605A1
• Filed: 12/23/2008
• Published: 06/24/2010
• Est. Priority Date: 12/23/2008
• Status: Active Grant

• Current Assignee: R2 Solutions LLC (Acacia Research Corporation)
• Original Assignee: Yahoo! Inc. (Apollo Global Management, Inc.)
• Inventors: Josifovski, Vanja, Pang, Bo, Gabrilovich, Evgeniy, Broder, Andrei, Becker, Hila
• Granted Patent: US 8,886,636 B2
• Time in Patent Office: Days
• US Class Current: 707/736
• CPC Class Codes: G06F 16/36 Creation of semantic tools, e.g. ontology or thesauri
• CPC Class Codes: G06Q 30/02 Marketing; Price estimation or determination; Fundraising

Abstract

A computer-implemented method is disclosed for determining a type of landing page to which to transfer web searchers that enter a particular query, the method comprising: classifying a landing page as one of a plurality of landing page classes with a trained classifier of a computer based on textual content of the landing page; determining, by the computer, characteristics of one or more query to be associated with the landing page; and choosing, with the computer, whether to retain or to change classification of the landing page to be associated with the one or more query based on relative average conversion rates of advertisements on a plurality of manually-classified landing pages when associated with the characteristics of the one or more query.

Claims

1. A computer-implemented method for determining a type of landing page to which to transfer web searchers that enter a particular query, the method comprising:

classifying a landing page as one of a plurality of landing page classes with a trained classifier of a computer based on content of the landing page;

determining, by the computer, characteristics of one or more query to be associated with the landing page; and

choosing, with the computer, whether to retain or to change classification of the landing page to be associated with the one or more query based on relative average conversion rates of advertisements on a plurality of previously-classified landing pages when associated with the characteristics of the one or more query.

2. The method of claim 1, wherein the classes are selected from the group consisting of homepage, search transfer, category browse, miscellaneous, or another category.

3. The method of claim 1, wherein the query characteristics comprise one or more of query frequency, query length, and query topic, wherein the query topic comprises a query class predicted by an automatic query classifier with respect to a commercial taxonomy.

4. The method of claim 1, wherein the classification by the classifier is based on feature extraction comprising:

observing particular characteristics on each landing page to distinguish among the plurality of landing page classes; and

detecting particular words within the text of each landing page to be correlated with the plurality of landing page classes.

5. The method of claim 1, wherein the classifier comprises a web crawler to extract hypertext markup language (HTML) content and a page rendering program to extract text from the HTML content of each landing page to be classified.

6. The method of claim 5, further comprising:

applying a supervised attribute selection technique to the classifier to reduce a feature space; and

training a machine model of the classifier using a learning algorithm.

7. A computer-implemented method for determining a type of landing page to which to transfer web searchers that enter a particular query, the method comprising:

classifying each of a plurality of landing pages into one of a plurality of classes with a classifier of a computer based on content of each of the plurality of landing pages;

gathering, with the computer, characteristics on one or more query associated with each landing page;

partitioning, by the computer, the landing pages according to a plurality of query characteristics;

determining, by the computer, conversion rates of advertisements on at least some of the plurality of landing pages;

correlating, with the computer, the plurality of classes of landing pages within each landing page partition with corresponding conversion rates; and

choosing a landing page of a particular class to associate with an identified query within a search engine based on the corresponding conversion rate of that landing page class as associated with the characteristics of the identified query.

8. The method of claim 7, wherein the classes are selected from the group consisting of homepage, search transfer, category browse, and miscellaneous, wherein miscellaneous includes landing pages not falling into a classification of homepage, search transfer, or category browse.

9. The method of claim 7, wherein the plurality of query characteristics comprise one or more of query frequency, query length, and query topic.

10. The method of claim 9, wherein the query topic comprises a query class predicted by an automatic query classifier with respect to a commercial taxonomy.

11. The method of claim 7, wherein the classifier comprises a web crawler to extract hypertext markup language (HTML) content and a page rendering program to extract text from the HTML content of each landing page to be classified.

12. The method of claim 11, wherein the classification by the classifier is based on feature extraction comprising:

observing particular characteristics on the landing pages to distinguish among the plurality of landing page classes; and

detecting particular words within the text of each landing page to be correlated with the plurality of landing page classes.

13. The method of claim 11, further comprising:

applying a supervised attribute selection technique to the classifier to reduce a feature space; and

training a machine model of the classifier using a learning algorithm.

14. The method of claim 13, wherein the machine model comprises a support vector machine model and wherein the learning algorithm comprises a sequential minimal optimization (SMO) algorithm of the support vector machine model.

15. A server for determining a type of landing page to which to transfer web searcher that enter a particular query, the server comprising:

a landing page classifier to classify a landing page as one of a plurality of landing page classes with a trained classifier of a computer based on content of the landing page;

a query classifier coupled with the landing page classifier to determine characteristics of one or more query to be associated with the landing page; and

a processor coupled with the landing page and query classifiers to choose whether to retain or to change classification of the landing page to be associated with the one or more query based on relative average conversion rates of advertisements on a plurality of previously-classified landing pages when associated with the characteristics of the one or more query.

16. The server of claim 15, wherein the classes are selected from the group consisting of homepage, search transfer, category browse, miscellaneous, or another category, and wherein the query characteristics comprise one or more of query frequency, query length, and query topic, wherein the query topic comprises a query class predicted by an automatic query classifier with respect to a commercial taxonomy.

17. The server of claim 15, wherein the processor observes particular characteristics on each landing page to distinguish among the plurality of landing page classes, the server further comprising:

an extractor to extract particular words within the text of each landing page to be correlated as features with the plurality of landing page classes.

18. The server of claim 15, wherein the classifier comprises a web crawler to extract hypertext markup language (HTML) content and a page rendering program to extract text from the HTML content of each landing page to be classified, wherein the processor:

applies a supervised attribute selection technique to the classifier to reduce a feature space; and

trains a machine model of the classifier using a learning algorithm, wherein the machine model comprises a support vector machine model and wherein the learning algorithm comprises a sequential minimal optimization (SMO) algorithm of the support vector machine model.

19. A computer-implemented system for determining a type of landing page to which to transfer web searchers that enter a particular query, the system comprising:

a landing page classifier to classify each of a plurality of landing pages into one of a plurality of classes based on content of each of the plurality of landing pages;

a processor coupled with a memory and with a communication interface;

a query logs database coupled with the processor, wherein the processor;

gathers from the query logs database characteristics on one or more query associated with each landing page;

partitions the landing pages according to a plurality of query characteristics;

retrieves conversion rates of advertisements on at least some of the plurality of landing pages;

correlates the plurality of classes of landing pages within each landing page partition with corresponding conversion rates; and

chooses a landing page of a particular class to associate with an identified query within a search engine based on the corresponding conversion rate of that landing page class as associated with the characteristics of the identified query.

20. The system of claim 19, wherein the classes are selected from the group consisting of homepage, search transfer, category browse, miscellaneous, or another category, and wherein the query characteristics comprise one or more of query frequency, query length, and query topic.

21. The system of claim 20, wherein the query topic comprises a query class predicted by an automatic query classifier with respect to a commercial taxonomy.

22. The server of claim 19, wherein the classifier comprises a web crawler to extract hypertext markup language (HTML) content and a page rendering program to extract text from the HTML content of each landing page to be classified, wherein the processor observes particular characteristics on each landing page to distinguish among the plurality of landing page classes, the server further comprising:

an extractor to extract particular words within the text of each landing page to be correlated as features with the plurality of landing page classes.

23. The system of claim 19, wherein the processor:

applies a supervised attribute selection technique to the landing page classifier to reduce a feature space; and

trains a machine model of the landing page classifier using a learning algorithm.

24. A computer-implemented method for training a landing page classifier, the method comprising:

extracting and counting a plurality of words from each of a plurality of manually-classified landing pages in a dataset of a computer;

selecting, with the computer, one or more words as features from each of the plurality of landing pages;

assigning values, with the computer, to each feature according to a scoring schema to create feature vectors thereof; and

feeding the feature vectors into a machine learning algorithm based on support vector machines (SVM) to induce a landing page classifier with an ability to automatically classify, with the computer, unclassified landing pages consistent with the manually-classified landing pages.

25. The method of claim 24, wherein the feature vectors comprise a word vector together with a class label.

26. The method of claim 24, wherein assigning the values to each feature is according to a term frequency-inverse document frequency (“

tf-idf”

) scheme.

27. The method of claim 24, wherein the machine learning algorithm comprises a sequential minimal optimization (SMO) algorithm of the support vector machine model.