Google Inc.

Zhen Lin

Kenneth Eric Vasilik

Ariel H Bardin

Dorothy Ann Nelson

Chirag Khopkar

Keith A Stevens

Andrei Pascovici

Theodore Vassilakis

Wei Shao

Xiaodong Fu

Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for semantic document analysis. In one aspect, methods include the actions of segmenting a document into segments; generating semantic representations, each corresponding to one of the segments; determining a corresponding segment score for one or more of the segments based on the corresponding semantic representation, such that each segment score represents a change in the corresponding segment; comparing each segment score to a threshold score, such that the threshold score represents an expectation of change for the document; and identifying segments having segment scores that indicate the change in the corresponding segment deviates from the expectation of change for the document based on the comparison.

Semantic document analysis

Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for scheduling resource crawls. In one aspect, a framework is provided for scheduling resource crawls such that a crawl scheduler determines the health of a document, i.e., whether it can be crawled, the popularity of the document, and the frequency of 'interesting,' i.e., substantive, content changes, and based on this information, estimates an appropriate crawl interval for each web resource to improve crawl resource utilization.

Scheduling resource crawls

Methods, systems, and apparatus, including computer program products for monitoring an experiment relating to the conversion rates of landing page variants to test landing page designs. One method includes computing respective conversion rates for each of a plurality of corresponding landing page variants, based on the number of times a conversion page is reached from the corresponding landing page variant, and based on the number of times the corresponding landing page variant is reached from one or more advertisements; and computing a confidence interval for each conversion rate. A user monitoring the experiment can be notified if winning or losing variants emerge. A time to completion of the experiment can be estimated. Presentation to browsing users of winning variants can be favored, and losing variants can be disfavored, during the experiment as they emerge.

Monitoring landing page experiments

Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for classifying changes to resources. In one aspect, a method includes obtaining a first version of a resource. The method includes obtaining a second version of the resource, the second version representing a change to the first version of the resource. The method includes determining feature data that describes changes from the first version to the second version. The method includes creating a feature vector from the feature data. The method includes providing the feature vector to a support vector machine, the support vector machine trained to classify the second version of the resource as having one of changes requiring a change notification or changes not requiring a change notification. The method includes classifying the second version as requiring a change notification or not requiring a change notification based on output from the support vector machine.

Classifying changes to resources

Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for managing system resources. In one aspect, a system monitors the number of items, and deallocates system resources for the least-important items when the number of items exceeds a predefined capacity. The importance of each item is determined, for example, based on the item's subscription count such that system resources are utilized for the benefit of the greatest number of users or to improve the quality of service.

Managing system resources

Methods, systems, and apparatus, including computer program products for defining landing pages to test different landing page designs. User input specifies a landing page having sections, each section having an original version of content. User input also specifies one or more additional versions of content, the original and the additional versions together constituting alternative versions, and the landing page with the alternative versions together constituting an experiment. Page fragments are determined. The fragments are added to the landing page and are operable, when interpreted by a client-side browser, to choose one of the alternative versions of each identified section and make each chosen alternative version the content of the corresponding section before the landing page is displayed by the browser. A designer can insert the page fragments into the landing page to produce an instrumented landing page.

Starting landing page experiments

Methods, systems, and apparatus, including computer program products for showing different landing page designs to a user. One method includes displaying an advertisement on a starting page to a user interacting with a client, the advertisement having a hyperlink to an instrumented landing page; in response to the user selecting the advertisement, downloading the instrumented landing page to the client, the instrumented landing page having one or more sections, each section having a corresponding plurality of alternative versions; choosing in the client a combination of one alternative version for each of the one or more sections; generating in the client a landing page variant, each section of the landing page variant being the corresponding alternative version of the section defined by the combination; notifying a server that the instrumented landing page was reached; notifying the server which combination was chosen; and displaying the landing page variant to the user.

Growing community of inventors

Bellevue, WA, United States of America

Zhen Lin