Abstract: Systems and methods are provided for monitoring the performance of a network with respect to providing results for user requests. A user request can correspond to a search query, an entry of a uniform resource locator (URL) or other address for a document on a network, or another type of request. A plurality of user requests are aggregated, such as based on logs of search query or browsing activity. A representative group of user requests is selected and then submitted in order to evaluate the results provided. Based on a heuristic analysis of the results, an alert can be provided to indicate potential performance problems in the network environment.

Abstract: Web applications, systems and services, which are prone to cyber-attacks, can utilize an adaptive fuzzing system and methodology to intelligently employ fuzzer technology to test web site pages for vulnerabilities. A breadth first search and minimal fuzzing testing is performed on identified pages of a web site looking for either a vulnerability or the potential for a vulnerability. Heuristics are gathered and/or generated on each tested web page to determine a vulnerability score for the page that is an indication of the page's potential for hosting a vulnerability. When a page is discovered with a vulnerability score that indicates the page has the potential for a vulnerability a depth first search and expanded fuzzing testing is performed on one or more branches of the web site that begin with the page with the potential vulnerability.

Abstract: Query suggestions are generated based on a Breadth-First-Search having a configurable decaying radius. A computer system receives an initial set of semantically similar queries. The computer system expands the set to include related terms. The set of related terms is included in the initial set. The expansion process is repeated for each query or related term in the set. The radius may be reduced for each subsequent related term added to the query. The process may stop when the radius reaches a specified threshold, e.g., a predetermined number of queries or terms for the set is reached. The set includes the related terms and search queries. The set may be used for, among other things, suggesting related terms to a researcher, improving search engine performance, or selecting appropriate advertisements.

Abstract: Query suggestions are generated based on a Breadth-First-Search having a configurable decaying radius. A computer system receives an initial set of semantically similar queries. The computer system expands the set to include related terms. The set of related terms is included in the initial set. The expansion process is repeated for each query or related term in the set. The radius may be reduced for each subsequent related term added to the query. The process may stop when the radius reaches a specified threshold, e.g., a predetermined number of queries or terms for the set is reached. The set includes the related terms and search queries. The set may be used for, among other things, suggesting related terms to a researcher, improving search engine performance, or selecting appropriate advertisements.

Abstract: A web search system uses humans to rank the relevance of results returned for various sample search queries. The search results may be divided into groups allowing training and validation with the ranked results. Consistent guidelines for human evaluation allow consistent results across a number of people performing the ranking. After a machine learning categorization tool, such as MART, has been programmed and validated, it may be used to provide an absolute rank of relevance for documents returned, rather than a simple relative ranking, based, for example, on key word matches and click counts. Documents with lower relevance rankings may be excluded from consideration when developing related refinements, such as category and price sorting.

Abstract: Systems and methods are provided for monitoring the performance of a network with respect to providing results for user requests. A user request can correspond to a search query, an entry of a uniform resource locator (URL) or other address for a document on a network, or another type of request. A plurality of user requests are aggregated, such as based on logs of search query or browsing activity. A representative group of user is selected and then submitted in order to evaluate the results provided. Based on a heuristic analysis of the results, an alert can be provided to indicate potential performance problems in the network environment.

Abstract: Web applications, systems and services, which are prone to cyber-attacks, can utilize an adaptive fuzzing system and methodology to intelligently employ fuzzer technology to test web site pages for vulnerabilities. A breadth first search and minimal fuzzing testing is performed on identified pages of a web site looking for either a vulnerability or the potential for a vulnerability. Heuristics are gathered and/or generated on each tested web page to determine a vulnerability score for the page that is an indication of the page's potential for hosting a vulnerability. When a page is discovered with a vulnerability score that indicates the page has the potential for a vulnerability a depth first search and expanded fuzzing testing is performed on one or more branches of the web site that begin with the page with the potential vulnerability.

Abstract: Information from social networks may be used to identify a user's interests and predilections, and the information may be used to affect search results. In one example, social networks have pages that correspond to real entities, such as manufacturers and merchants. Entity pages in social networks are mapped to their corresponding real entities, and information that users leave on the pages (e.g., “likes”, or textual reviews) are extracted to determine users' sentiments about the entities. When users search for products with a search engine, user sentiment is then used to guide the results. Social networks' information about users (e.g., their affinities, such as schools, workplaces, interests) may be used to determine the relevance of specific users' sentiments—e.g., sentiments of users who went to a particular school may be used to influence search results, when the search is requested by someone who went to the same school.

Abstract: Outlier images—those images that differ substantially from other images in a set—can be automatically identified. One or more penalty values can be assigned to each image that quantifies how different that image is from others in the set. A threshold can be determined based on the set of penalty values. Each image whose penalty values are above the threshold is an outlier image. The penalty values can be the sum of per-pixel penalty values multiplied by the number of pixels with nonzero penalty values. A per-pixel penalty value can be the difference between a color value for that pixel and a predetermined range of color values, based on corresponding pixels in other images. The per-pixel penalty value can be determined for each component color and then optionally summed together. The threshold penalty values can be adjusted to provide for greater, or less, sensitivity to differences among the images.

Abstract: An image may be accepted from a vendor, and the image may be submitted to an image analysis system. The image analysis system may determine whether the image is a not found image or a true image. The determination may occur in a variety of ways by examining the color and intensity characteristics of an image. After the analysis, a determination is received from the image analysis system of whether the image is a not found image or a true image.

Abstract: Outlier images—those images that differ substantially from other images in a set—can be automatically identified. One or more penalty values can be assigned to each image that quantifies how different that image is from others in the set. A threshold can be determined based on the set of penalty values. Each image whose penalty values are above the threshold is an outlier image. The penalty values can be the sum of per-pixel penalty values multiplied by the number of pixels with nonzero penalty values. A per-pixel penalty value can be the difference between a color value for that pixel and a predetermined range of color values, based on corresponding pixels in other images. The per-pixel penalty value can be determined for each component color and then optionally summed together. The threshold penalty values can be adjusted to provide for greater, or less, sensitivity to differences among the images.

Abstract: An image may be accepted from a vendor, and the image may be submitted to an image analysis system. The image analysis system may determine whether the image is a not found image or a true image. The determination may occur in a variety of ways by examining the color and intensity characteristics of an image. After the analysis, a determination is received from the image analysis system of whether the image is a not found image or a true image.

Abstract: A web search system uses humans to rank the relevance of results returned for various sample search queries. The search results may be divided into groups allowing training and validation with the ranked results. Consistent guidelines for human evaluation allow consistent results across a number of people performing the ranking. After a machine learning categorization tool, such as MART, has been programmed and validated, it may be used to provide an absolute rank of relevance for documents returned, rather than a simple relative ranking, based, for example, on key word matches and click counts. Documents with lower relevance rankings may be excluded from consideration when developing related refinements, such as category and price sorting.

Abstract: Techniques are described herein for determining whether data sets of real information in databases indicate PII information. The data sets are stored in a first table and parsed for keywords related to the names of data items in the sets. The keywords are stored in the second table in a many-to-many relationship with related data items in the first table. The number of times the keywords are parsed from the data items is counted, as well as the number of times each keyword is associated with a PII-designated data item. The counted numbers are then used in analyzing new data sets to identify the likelihood that the new data sets contain any PII data items.

Abstract: Strategies are described for sanitizing a data set, having the effect of obscuring restricted data in the data set to maintain its secrecy. The strategies operate by providing a production data set to a sanitizer. The sanitizer applies a data directory table to identify the location of restricted data items in the data set and to identify the respective sanitization tools to be applied to the restricted data items. The sanitizer then applies the identified sanitization tools to the identified restricted data items to produce a sanitized data set. A test environment receives the sanitized data set and performs testing, data mining, or some other application on the basis of the sanitized data set. Performing sanitization on a sanitized version of the production data set is advantageous because it preserves the state of the production data set. The data directory table also provides a flexible mechanism for applying sanitization tools to the production data set.