Abstract: The use of data from multiple data source provides inferred air quality indices with respect to a particular pollutant for multiple areas without the addition of air quality monitor stations to those areas. Labeled air quality index data for a pollutant in a region may be obtained from one or more air quality monitor stations. Spatial features for the region may be extracted from spatially-related data for the region. The spatially-related data may include information on fixed infrastructures in the region. Likewise, temporal features for the region may be extracted from temporally-related data for the region that changes over time. A co-training based learning framework may be further applied to co-train a spatial classifier and a temporal classifier based at least on the labeled air quality index data, the spatial features for the region, and the temporal features for the region.

Abstract: A method and system for identifying information about people is provided. The information system identifies groups of people that have relationships based on their relationships to documents or more generally to objects. The information system initially is provided with an indication of which people have which relationships to which documents. The information system then identifies clusters of people based on having a relationship to the same objects. The information system may also identify clusters of related objects associated with a cluster of people. When a user wants to identify information about a person, the user can provide the name of that person to the information system. The information system then can retrieve and display the names of the other people who are in the same cluster as the person.

Abstract: Disclosed herein are one or more embodiments that arrange a plurality of location entities into a hierarchy of location descriptors. One or more of the disclosed embodiments may determine whether one of the location entities is a spatial outlier based at least in part on presence of one or more other location entities within a predetermined distance of the one location entity. Also, the other location entities and the one location entity may share a location descriptor.

Abstract: A method and system for augmenting a training set used to train a classifier of documents is provided. The augmentation system augments a training set with training data derived from features of documents based on a document hierarchy. The training data of the initial training set may be derived from the root documents of the hierarchies of documents. The augmentation system generates additional training data that includes an aggregate feature that represents the overall characteristics of a hierarchy of documents, rather than just the root document. After the training data is generated, the augmentation system augments the initial training set with the newly generated training data.

Abstract: Implementations for identifying duplicate images in an image space are described. An image space is partitioned into a plurality of coarse clusters based on signatures of the images within the image space. The signatures are determined from compact descriptors of the images. Refined clusters that include one or more images of an individual coarse cluster are created based on pair-wise comparisons of the compact descriptors of images in the coarse cluster, and the refined clusters are identified as sets of duplicate images. The refined clusters are grown by searching in similar coarse clusters for images to add to the refined clusters.

Abstract: A method and system for detecting whether an outgoing communication contains confidential information or other target information is provided. The detection system is provided with a collection of documents that contain confidential information, referred to as “confidential documents.” When the detection system is provided with an outgoing communication, it compares the content of the outgoing communication to the content of the confidential documents. If the outgoing communication contains confidential information, then the detection system may prevent the outgoing communication from being sent outside the organization. The detection system detects confidential information based on the similarity between the content of an outgoing communication and the content of confidential documents that are known to contain confidential information.

Abstract: A method and system for identifying informative links of a web site for use in crawling the web site is provided. A forum crawler analyzes sample web pages of a web forum to identify informative links and then crawls the web forum by following links determined to be informative and not following other links. The forum crawler system determines whether links are informative based on whether they are part of the overall structure of the web site or are used to select sequential information that has been split onto multiple web pages.

Abstract: Methods and systems for selecting advertisements to present to a user of a computing device are provided. An advertisement system selects advertisements to display to a user based on the serving area of candidate advertisements. The advertisement system selects those candidate advertisements whose serving area encompasses the user's current location. The advertisement system may also select candidate advertisements to present to a user based on a map area currently being displayed to the user. The advertisement system may filter the candidate advertisements based on the provider location being within the map area that is currently being displayed to the user.

Abstract: An exemplary method includes defining a vocabulary for emotions; extracting descriptions for songs; generating distributions for the songs in an emotion space based at least in part on the vocabulary and the extracted descriptions; extracting salient words from a document; generating a distribution for the document in an emotion space based at least in part on the vocabulary and the extracted salient words; and matching the distribution for the document to one or more of the distributions for the songs. Various other exemplary methods, devices, systems, etc., are also disclosed.

Abstract: A method and system for determining relevance of a document having text and images to a text string is provided. A scoring system identifies image text associated with an image of the document. The scoring system calculates an image score indicating relevance of the image text to the text string. The image score may be used in many applications, such as searching, summary generation, and document classification, image search, and image classification.

Abstract: While browsing, a user may interact with a wide variety of images. The user may upload and share images taken with a digital camera and/or search for image using a search engine. Because images are rich in contextual information, it may be advantageous to provide additional information, such as adjacent market advertising based upon matching advertisements with contextual information of the images. Accordingly, a query image may be used to retrieve a video frame set. The video frame set may be expanded with related video frames corresponding to adjacent markets. The expanded video frame set may be grouped into clusters of similar frames. The clusters may be used to rank advertisements based upon how similar the advertisements are to the clusters and/or video frames within the clusters. In this way, one or more ranked advertisements may be presented with the query image.

Abstract: Described herein is a technology that facilitates efficient large-scale similarity-based retrieval. In several embodiments documents, images, and/or other multimedia files are compactly represented and efficiently indexed to enable robust search using a long-query in a large-scale corpus. As described herein, these techniques include performing decomposition of a file, e.g., an image, a document containing an image, or a document-like representation of an image. The techniques use dimension reduction to obtain three parts, low-dimensional representations (major semantics), file specific terms (minor semantics), and background words, representing the major semantics in a feature vector and the minor semantics as keywords. Using the techniques described, file vectors are matched in a topic model and the results ranked based on the keywords.

Abstract: A method and system for generating a search result for a query of hierarchically organized documents based on retrieval of subtrees that are key resources for topic distillation is provided. The retrieval system may identify documents relevant to a query using conventional searching techniques. The retrieval system then calculates a subtree feature for subtrees that have an identified document as their root. After the retrieval system calculates the subtree feature for the subtrees, the retrieval system may generate a subtree relevance score for each subtree based on its subtree feature. The retrieval system may then order the identified documents based on their corresponding subtree relevances.

Abstract: Implementations for identifying duplicate images in an image space are described. An image space is partitioned into a plurality of coarse clusters based on signatures of the images within the image space. The signatures are determined from compact descriptors of the images. Refined clusters that include one or more images of an individual coarse cluster are created based on pair-wise comparisons of the compact descriptors of images in the coarse cluster, and the refined clusters are identified as sets of duplicate images. The refined clusters are grown by searching in similar coarse clusters for images to add to the refined clusters.

Abstract: Automated rich presentation of a semantic topic is described. In one aspect, respective portions of multimodal information corresponding to a semantic topic are evaluated to locate events associated with the semantic topic. The probability that a document belongs to an event is determined based on document inclusion of one or more of persons, times, locations, and keywords, and document distribution along a timeline associated with the event. For each event, one or more documents objectively determined to be substantially representative of the event are identified. One or more other types of media (e.g., video, images, etc.) related to the event are then extracted from the multimodal information. The representative documents and the other media are for presentation to a user in a storyboard.

Abstract: A dual cross-media relevance model (DCMRM) is used for automatic image annotation. In contrast to the traditional relevance models which calculate the joint probability of words and images over a training image database, the DCMRM model estimates the joint probability by calculating the expectation over words in a predefined lexicon. The DCMRM model may be advantageous because a predefined lexicon potentially has better behavior than a training image database. The DCMRM model also takes advantage of content-based techniques and image search techniques to define the word-to-image and word-to-word relations involved in image annotation. Both relations can be estimated by using image search techniques on the web data as well as available training data.

Abstract: Names of entities, such as people, in an image may be identified automatically. Visually similar images of entities are retrieved, including text proximate to the visually similar images. The collected text is mined for names of entities, and the detected names are analyzed. A name may be associated with the entity in the image, based on the analysis.

Abstract: A peer-to-peer advertisement platform is provided to ubiquitously promote products or services supplied by advertisers across content-based applications executing on nodes in a peer-to-peer network. The peer-to-peer advertisement platform may include a registration component to register nodes in the peer-to-peer advertising platform, an advertisement submission component to receive advertisement data from the advertisers, and a distribution component to distribute the advertisement data to the nodes registered in the peer-to-peer advertisement platform. The peer-to-peer advertisement platform also includes a money sharing component to reward nodes based on a contribution level assigned to the node. Accordingly, the peer-to-peer advertisement platform stores the advertisement data locally at the plurality of nodes registered in the peer-to-peer advertising platform and shares a portion of the revenue generated from the advertisement data with the nodes registered in the peer-to-peer advertising platform.

Abstract: According to a cost-per-action advertising model, advertisers submit ads with cost-per-action bids. Ad auctions are conducted and winning ads are returned with contextually relevant search results. Each time a winning ad is selected by a user, resulting in the user being redirected to a website associated with the advertiser, a selected impression and a price is recorded for the winning ad. Periodically, an advertiser submits a report indicating a number of actions attributed to the ads that have occurred through the advertiser website. The advertiser is then charged a fee for each reported action based on the recorded prices for the winning ads and based on the number of selected impressions recorded for the winning ads.

Abstract: A “Name Disambiguator” provides various techniques for implementing an interactive framework for resolving or disambiguating entity names (associated with objects such as publications) for entity searches where two or more same or similar names may refer to different entities. More specifically, the Name Disambiguator uses a combination of user input and automatic models to address the disambiguation problem. In various embodiments, the Name Disambiguator uses a two part process, including: 1) a global SVM trained from large sets of documents or objects in a simulated interactive mode, and 2) further personalization of local SVM models (associated with individual names or groups of names such as, for example, a group of coauthors) derived from the global SVM model. The result of this process is that large sets of documents or objects are rapidly and accurately condensed or clustered into ordered sets by that are organized by entity names.