Google Inc.

Lucent Technologies Inc.

Alcatel Lucent

Alcatel-lucent USA Inc.

Michael E Flaster

Paul Haahr

Yonghui Wu

Eran Gabber

Nitin Kumar Gupta

Steven D Baker

Srinivasan Venkatachary

Ruoming Pang

Shanmugavelayutham Muthukrishnan

Sundeep Tirumalareddy

Wenfei Fan

Philip L Bohannon

Eric Lehman

Randall G Keller

Peter Ciccolo

Rajeev R Rastogi

Clifford Eric Martin

Cuong Tran

April R Lehman

Suryanarayan Perinkulam

Neil C Fernandes

Wee Teck Ng

Jeffrey K Fellin

Fengrui Gu

Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for generating descriptive text for images. In one aspect, a method includes identifying a set of seed descriptors for an image in a document that is hosted on a website. For each seed descriptor, structure information is generated that specifies a structure of the document with respect to the image and the seed descriptor. One or more templates are generated for each seed descriptor using the structure information for the seed descriptor. Each template can include image location information, document structure information, image feature information, and a generative rule that generates descriptive text for other images in other documents. Descriptive text for other images is generated using the templates and the other documents. The descriptive text is associated with the images.

Generating descriptive text for images in documents using seed descriptors

Methods, systems, and apparatus, including computer program products, for generating synthetic queries using seed queries and structural similarity between documents are described. In one aspect, a method includes identifying embedded coding fragments (e.g., HTML tag) from a structured document and a seed query; generating one or more query templates, each query template corresponding to at least one coding fragment, the query template including a generative rule to be used in generating candidate synthetic queries; generating the candidate synthetic queries by applying the query templates to other documents that are hosted on the same web site as the document; identifying terms that match structure of the query templates as candidate synthetic queries; measuring a performance for each of the candidate synthetic queries; and designating as synthetic queries the candidate synthetic queries that have performance measurements exceeding a performance threshold.

Query generation using structural similarity between documents

A method for creating a hidden text data index for ranking computerized query search results that includes generating a render tree based on a document object model (DOM) tree for a web page. The render tree includes nodes that correspond to text that will be visually displayed by a client device when executed. The method includes comparing nodes corresponding to text of the DOM tree with the nodes corresponding to text of the render tree to identify the nodes in the DOM tree that will not be visually displayed when executed by the client device. The method also includes creating a hidden text data index for the nodes corresponding to text of the DOM tree not in the render tree. The hidden text data index identifies nodes corresponding to text of the DOM tree as hidden that will not be visually displayed when executed by the client device.

Hidden text detection for search result scoring

Methods, systems, and apparatus, including computer programs encoded on computer storage media, for generating synthetic descriptive text. One of the methods includes identifying a group of linking resources, wherein each of the linking resources includes a link to a respective target resource; determining, from a search engine index, that at least some of the target resources are associated with seed queries; generating term location information that identifies, for each seed query, locations of terms from the seed query in the linking resource that links to the target resource associated with the seed query; generating synthetic descriptive text for the target resources based on the term location information; and associating the synthetic descriptive text with the target resources in the search engine index.

Generating synthetic descriptive text

Methods, systems, and apparatus, including computer programs encoded on computer storage media, for using synthetic descriptive text to rank search results. One of the methods includes receiving a search query from a user device; receiving data identifying a plurality of search result resources and respective initial scores for each of the search result resources; determining, from a search engine index, that a particular search result resource of the plurality of search result resources is associated with one or more pieces of synthetic descriptive text, wherein each piece of synthetic descriptive text is generated by applying a respective template to a respective linking resource that links to the particular search result resource; computing a synthetic descriptive text score for the particular search result resource; and adjusting the initial score for the particular search result resource based at least in part on the synthetic descriptive text score.

Using synthetic descriptive text to rank search results

Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for evaluating website properties by partitioning user feedback data are disclosed. In one aspect, a method includes receiving a plurality of document query pairs (D-Qs) associated with a website, partitioning the plurality of D-Qs into one or more groups according to values for a partition parameter associated with the plurality of D-Qs, evaluating a property parameter of the website based on aggregated user feedback data of the D-Qs included within at least one of the one or more groups, and providing the evaluated property parameter as an input for ranking documents from the website as result documents for searches.

Evaluating website properties by partitioning user feedback

Methods and apparatus are provided for identifying constraint violation repairs in data that is comprised of a plurality of records, where each record has a plurality of cells. A database is processed, based on a plurality of constraints that data in the database must satisfy. At least one constraint violation to be resolved is identified based on a cost of repair and the corresponding records to be resolved and equivalent cells are identified in the data that violate the identified at least one constraint violation. A value for each of the equivalent cells can optionally be determined, and the determined value can be assigned to each of the equivalent cells. The at least one constraint violation selected for resolution may be, for example, the constraint violation with a lowest cost. The cost of repairing a constraint is based on a distance metric between the attributes values.

Equivalence class-based method and apparatus for cost-based repair of database constraint violations

Methods, systems, and apparatus, including computer program products, for clustering documents. A plurality of documents are identified from a set of documents, where the identified documents have the same top N terms by term frequency score for an integer N. A pattern string that is satisfied by at least a subset of the identified documents is identified. A document cluster is formed from at least the subset of the identified documents.

Efficient document clustering

Systems, methods and computer readable media for identifying transient paths within websites. Transient paths can be identified, for example, by identifying a path associated with known transient content and determining that the path exists on other pages associated with the website. If the path exists in other web pages associated with the website, the content associated with the path can be identified as transient content.

Identifying transient paths within websites

Systems, methods and computer readable media for identifying transient content in web pages. Transient content can be identified, for example, by parsing different versions of the same web page into tokens, and inserting fingerprints associated with the tokens into data structures. The data structures can be compared to each other to identify differences between the web pages, thereby identifying transient content associated with the web pages.

Identifying transient portions of web pages

Methods and apparatus are provided for mapping XML source documents to target documents using schema embeddings. According to one aspect of the invention, one or more edges in the one or more source schemas are mapped to one or more paths in at least one target schema. The disclosed mapping techniques ensure that (i) one or more source documents that conform to one or more of the source schemas can be recovered from one or more target documents that conform to the at least one target schema, if a mapping exists between the one or more of the source schemas and the at least one target schema; (ii) queries on one or more source documents that conform to one or more of the source schemas in a given query language can be answered on one or more target documents that conform to the at least one target schema; and (iii) the one or more target documents conform to a target schema.

Growing community of inventors

Menlo Park, CA, United States of America

Michael E Flaster