Abstract: Embodiments herein provide a method for managing Quality of Service (QoS) for multiple Evolved Multimedia Broadcast Multicast Service (eMBMS) services in a wireless communication system. The method includes detecting, by a network node, more than one Multimedia Broadcast-Multicast Service (MBMS) services in a group of MBMS services have a similar popularity index. Further, the method includes identifying, by the network node, which eMBMS service has more number of users facing Quality of Experience (QoE) issues, in response to detecting that the more than MBMS service in the group of MBMS services have the similar popularity index. Further, the method includes localizing, by the network node, whether the QoE issue is from at least one of the core network, the access network, the single eNB or a complete Multimedia Broadcast Single Frequency network (MBSFN) area. Furthermore, the method includes dynamically modifying, by the network node, QoS parameters based on the identification of the QoS issue.

Abstract: The present teaching generally relates to detecting providing pre-validated data buckets for online experiments. In a non-limiting embodiment, user activity data representing user activity for a first plurality of user identifiers may be obtained. A first set of values and a second values, representing first and second user engagement parameters, respectively, may be generated for each user identifier based on the user activity data. A first ranking and a second ranking may be determined for the first and second sets, respectively. A first exclusion range including a first number of values to be removed from the first and second sets may be determined. A homogenous value set may be generated by removing the first number of values from the first and second sets, where each value from the homogenous value set corresponds to a user identifier available to be placed in a data bucket for an online experiment.

Abstract: Disclosed are interest tag determining method and device for accurately promoting information for users in a group based on interest tags of the group. This includes obtaining a group identifier and a corresponding group name, obtaining user identifiers corresponding to the group identifier and first user contents of the user identifiers, concatenating the first user contents of the user identifiers corresponding to the group identifier to obtain a first group content of the group identifier, calculating probabilities that the first group content is under the topic identifiers, and determining interest tags of the group identifier according to the probabilities that the first group content is under the topic identifiers and an association relationship between the topic identifiers and the interest tags that is determined through pre-training.

Abstract: A website misclassification report is received, indicating that a website has been misclassified. A determination is made that a current classification model correctly classifies the reported website. The current classification model is different from a model that was previously used to classify the website. In response to a determination that the reported website should be reclassified using the current classification model, a reclassification operation is performed, using the current classification model, on a set of websites determined to be similar to the reported website.

Abstract: The present disclosure is directed toward systems, methods, and non-transitory computer readable media for providing digital content to users by applying a machine learning model based on composite utility scores reflecting multiple events categories. For example, the systems described herein can identify, from a digital content publisher, significance ratings of various event categories that a user can perform. The systems can analyze user activities to determine a composite utility score for user based on events that the users have performed. Furthermore, in one or more embodiments, the systems train a machine learning model based on training composite utility scores to identify additional users likely to have elevated composite utility scores. Moreover, the disclosed systems can utilize the trained machine learning model to provide targeted digital content to computing devices of these additional users.

Abstract: One embodiment provides a method, including: generating a plurality of ontologies wherein each ontology is generated by: monitoring interactions of a user with lineage information, wherein the monitoring comprises monitoring (i) filter interactions and (ii) access interactions; aggregating the monitored interactions of the user with monitored interactions of other users having a given business role; and generating an ontology for the given business role, wherein the subset comprises (i) event types, (ii) event constraints, (iii) event metadata, and (iv) event context; and upon a user having one of the plurality of business roles accessing lineage information on the data platform, providing a subset of the lineage information.

Abstract: A recipe pushing method includes receiving cooking device product information of a cooking device transmitted by a terminal device via a cooking service application, extracting a recipe preference characteristic of a user according to a history of the user browsing recipes in the cooking service application and a preset filtering condition, generating a recommended recipe matching the cooking device product information according to the recipe preference characteristic, and transmitting the recommended recipe to the cooking service application.

Abstract: One embodiment provides a method, including: receiving a query from a user requesting assistance regarding instructions for performing a task; identifying, within steps of the instructions, words that can be visualized, wherein the identifying comprises identifying relationships between terms within the query to generate a step query; retrieving, for each of the steps, a plurality of images representing the identified words; identifying at least one object occurring within the plurality of images corresponding to more than one of the steps; selecting an image for each of the steps of the instructions, wherein the selecting an image comprises selecting an image for each step such that the identified at least one object is represented similarly in each selected image including the identified at least one object; and presenting the instructions as visualized instructions by presenting the selected images for each of the steps in order.

Abstract: Methods for identifying encrypted content in ones of a plurality of encrypted data files in a library of encrypted files without decrypting the data files utilize crowd sourcing for content identification. A method includes selecting, by a computer, content titles for presenting with ones of identifiers for the data files in a data structure. Each of the identifiers includes a hash of metadata for one of the data files and the content titles include a character string that identifies each file's content. The user selection data identifies the content titles that correspond to the data files. The computer determines which content titles satisfy a minimum confidence threshold for associating with one of the identifiers, based on a quality or quantity of the multiple independent clients supplying the user selection data. An apparatus for performing the method includes a memory holding instructions for performing steps of the method as summarized above.

Abstract: A method for automatically providing key moments in a multimedia content on an electronic device and an electronic device therefor are provided. The method includes determining a navigation behavior of each user of the multimedia content during playback, determining a plurality of key moments in the multimedia content based on the navigational behavior, the plurality of key moments including a positive key moment, a negative key moment, and a neutral key moment, storing the plurality of key moments, detecting a playback event of the multimedia content by a candidate user, retrieving at least one key moment from the plurality of key moments in the multimedia content based on the candidate user, and displaying an actionable user interface including the at least one key moment.

Abstract: An interactive voice response system for interpreting sentiment of a conversation in real time is provided. The system may include a receiver. The receiver may be operable to receive a plurality of sentiment signals and a first portion of a conversation. The first portion of the conversation may include a plurality of interaction pairs. Each interaction pair may include an utterance and a system response. The system may include an utterance sentiment category determination module that may be operable to determine a sentiment category for each utterance. The system may include a sequential module. The sequential module may determine a conversation sentiment category based on the utterance sentiment categories and the order in which the utterance sentiment categories are received. The system may include an output module. The output module may output the conversation sentiment category. The conversation sentiment category may be used to determine a future system response.

Abstract: A system that facilitates sharing of multimedia content includes a user interface, network circuitry; and one or more processors. The processors perform operations that include displaying, via the user interface, a list of one or more media playback devices available to be added to a social networking group. A selection of first and second media playback devices to be associated with a particular social networking group is received via the user interface. The first media playback device is associated with a first playback queue that specifies a first set of media items associated with first media content that can be streamed from a first streaming service to the first media playback device. An indication to the social networking system that specifies the selection of the first and second media playback devices to be associated with the group is communicated via the network circuitry.

Abstract: Embodiments of the present invention include a system, computer-implemented method, and a computer program product. A non-limiting example of the method includes a processor utilizing a model having a plurality of parameters. The processor compares a current value of a model parameter to a prior value of the model parameter. Based at least in part on comparing the current value of the model parameter to the prior value of the model parameter, a determination is made that the model being utilized by the processor has changed. The current value of the model parameter is transmitted by the processor.

Abstract: Example techniques relate to cloud queue tombstones. An example implementation may involve a computing system receiving instructions to remove a particular media item from a playback queue. Based on the instructions, the computing system removes the particular media item from the playback queue and adds, to the playback queue, a tombstone indicator indicating that the particular media item has been removed from the playback queue. The computing system receives, from the media playback system, a request for a window of one or more media items from the playback queue. Based on the request, the computing system sends, to the media playback system, data representing a particular window of media items from the playback queue, wherein the particular window includes the tombstone indicator. The computing system detects a given tombstone removal trigger from among multiple tombstone removal triggers and based on the detecting, removes the tombstone indicator from the playback queue.

Abstract: Certain aspects of the present disclosure provide techniques for determining content quality of a set of content item based on generating a score for each content item. An example technique includes using a trained content quality model to generate the score for each content item component. In such example, the model is trained using a calculated set of features associated with text and metadata of a content item in a training data set as well as a quality label for the content item. The model is trained by associating the set of features with the quality label for each content item in the training data.

Abstract: In one aspect, this disclosure details methods and apparatuses for influencing the service continuity threshold used for switching a multimedia session between Packet-Switched (PS) transport and Circuit-Switched (CS) transport, in dependence on the “robustness” of the media configuration in use for the multimedia session. Robustness, for example, may be defined in terms of packet loss tolerance, and it will be understood that the media configurations available for a given multimedia session may have different levels of absolute or relative robustness and that the media configuration selected for use in a multimedia session may change during the multimedia session. Correspondingly, the approaches proposed herein provide mechanisms for dynamically adjusting the service continuity threshold in dependence on which media configuration is currently in use for the multimedia session.

Abstract: Graph databases directly relate data items in the data store with edges that represent relationships between the data items. The relationships link the data items together and often permit complex sets of related data items to be retrieved with a single operation. New query systems and techniques for graph databases provide prediction of non-explicit connections between data items that further enhances the efficiency and utility of graph databases, as well as extend their industrial applications.

Abstract: An item extraction unit extracts, when a user selects an item from a list of items to be selected by the user, at least one item from among items that are not selected by the user. A word acquisition unit acquires a word that is contained in text information associated with the item selected by the user and is not contained in text information associated with the item extracted by the item extraction unit. A search condition setting unit sets a search condition for searching for an item that is similar to the item selected by the user based on the word acquired by the word acquisition unit. A search execution unit executes a search based on the search condition.

Abstract: Systems and methods for creating personalized digital music and digital visual media content to users of media devices include uploading live-streamed digital visual media content to a host server, wherein the live-streamed digital visual media content is assigned a first URL stored in the host server. At least one digital music content file is selected to be streamed from an external audio content provider, wherein the at least one digital music content file is personalized to the user. The selected digital music content file has a second URL stored in the host server. The live-streamed digital visual media content is paired with the digital music content file to create paired digital music and live-streamed digital visual media content. The paired digital music and live-streamed digital visual media content is played by executing the first and second URLs simultaneously.

Abstract: Methods, systems, and computer program products for assessing a student in in an online discussion event using at least one processor of a computer includes obtaining discussion data comprising a plurality of discussion posts in a data file, each comprising a post author and post content, grouping the plurality of discussion posts of the data file into a plurality of discussion threads, classifying individual discussion posts of the plurality of discussion posts based on a post type, and assigning relative value units (RVUs) to at least one of the plurality of discussion posts, where the RVUs are time-normalized scores based on a complexity of the discussion post.

Abstract: Described herein are systems, methods, and non-transitory computer readable media for validating or rejecting automated detections of an entity being tracked within an environment in order to generate a track representative of a travel path of the entity within the environment. The automated detections of the entity may be generated by an artificial intelligence (AI) algorithm. The track may represent a travel path of the tracked entity across a set of image frames. The track may contain one or more tracklets, where each tracklet includes a set of validated detections of the entity across a subset of the set of image frames and excludes any rejected detections of the entity. Each tracklet may also contain one or more user-provided detections in scenarios in which the tracked entity is observed or otherwise known to be present in an image frame but automated detection of the entity did not occur.

Abstract: The present invention is a system and method that generates a Table of Contents (TOC) customized to the user knowledge about the concept(s) in the user query and the specific context and preferences of the user. The invention identifies search concepts within the search queries, receives search results, and splits the search results into one or more result segments. In a preferred embodiment, a correlation strength between concepts in one or more of the result segments and the user search query, along with reference to the user knowledge, context, and/or preferences determine which result segments are selected in sequenced to form the TOC.

Abstract: Systems and methods may include sending, to a network registrar, an extended duplicate address request (EDAR) message including a first nonce generated by a host computing device, and receiving, from the network registrar, an extended duplicate address confirmation (EDAC) message including a second nonce, the second nonce being signed by the network registrar via a private key of a first public key infrastructure (PM) key pair of the network registrar via a first signature. The method further includes sending a first neighbor advertisement (NA) message to the host computing device including the second nonce. The second nonce and the private key of the network registrar verifies the first signature from the network registrar, the verification of the first signature indicating that the router is not impersonating the network.

Abstract: The present disclosure relates to techniques for analysis of data from multiple different data sources to determine similarity amongst the datasets. Determining a similarity between datasets may be useful for downstream processing of those datasets for different uses. A graphical interface may be provided to display detailed results including: a similarity prediction, data similarity prediction, column order similarity prediction, document type similarity prediction, prediction of overlapping or related columns, orphaned column prediction (e.g., a left orphaned column or a right orphaned column).

Abstract: Contextual spelling methods and systems are provided that utilize natural language processing and n-gram frequencies to group documents into logical groups and to provide spelling correction suggestions. For example, a contextual spelling correction system may receive a set of documents, group the documents into separate logical groups, generate dictionaries associated with the logical groups, receive a user input, determine scores for potential spelling correction suggestions regarding the user input, and provide spelling correction suggestions based at least partly on the dictionaries associated with the logical groups.

Abstract: Provided is a method and an apparatus for displaying a live broadcast room, a device, and a storage medium. The method includes: acquiring a live broadcast screenshot of at least one live broadcast room under a target classification label; inputting the live broadcast screenshot of the at least one live broadcast room into an image detection model to obtain a live broadcast screenshot that satisfies a set type condition; adding a display identifier to a live broadcast room corresponding to the live broadcast screenshot of the set type condition; and arranging and displaying the at least one live broadcast room according to the display identifier in a display interface corresponding to the target classification label.

Abstract: A method of searching a database of medical research for documentation in support of a specific treatment for a patient, the method including: identifying a medical condition; identifying attributes of the patient; searching the database for treatment regimens for the medical condition; scoring identified treatment regimens based on similarity of the attributes of the patient to cohorts treated with the treatment regimens and outcomes; and providing a recommendation for a treatment regimen for the patient based on the attributes of the patient.

Abstract: The disclosed computer-implemented method for crowd sourced application advisory may include determining at least one characteristic associated with an application, collecting, from a plurality of computing devices, a sentiment associated with the application, wherein the sentiment reflects an assessment of the application in light of the at least one characteristic by at least one user of the application, presenting, via a user interface, the at least one characteristic associated with the application, the presentation of the at least one characteristic being modified by the sentiment collected from the plurality of computing devices, and receiving, via the user interface, at least one of a decision to install the application or a decision to uninstall the application. Various other methods, systems, and computer-readable media are also disclosed.

Abstract: Disclosed systems and methods for the automatic creation of multiple layouts that maintain a design aesthetic of an input design document. In an example, a document processing application determines a set of document layout parameters such as balance or equilibrium from an input document. The application calculates, for each document layout parameter of the input document, a weight representing a prominence of the respective layout parameter. The application selects templates having an output size and a number of object containers equal to the number of objects of the document. The application further calculates a score for each template by applying the weights of the document layout parameters to the template layout parameters. The application further selects a template with a highest score and places the object on the template, thereby creating the new design document that maintains the design aesthetic.

Abstract: A system accesses content structure that includes a first attribute table including a first list of attributes of a first object, and a first mapping including first attribute values. The first list of attributes of the first object also includes a quality attribute indicating a first quality. After a request to modify quality is received, the system searches a plurality of content structures for a suitable second content structure that comprises a second attribute table including a second list of attributes of a second object. The suitable content structure has another attribute that matches a corresponding attribute of the first list of attributes of the first object and a quality attribute indicating a second quality. The system modifies the first attribute table to include the second list of attributes of the second object. In this way content is generated that is of higher or lower quality than the original content.

Abstract: Provided is a natural language processing system for determining a merchant based on a natural language query. The system may include a processor to receive a natural language query, convert at least one word of the natural language query to a vector using at least one neural network to form a set of vectors, determine a vector distance from the set of vectors to each profile in a set of profiles, rank the set of profiles based on the vector distance of each profile to the set of vectors, communicate merchant data associated with at least one merchant included in the set of profiles to the user device, receive a selection of a first merchant associated with the merchant data from the user device, and schedule an appointment with the first merchant for a user of the user device. A computer program product and method are also disclosed.

Abstract: Techniques to configure the network distribution of media compositions for transmission are described. An apparatus may comprise a user interface component of a messaging client on a client device operative to receive a media composition via a composition interface, the messaging client associated with a user account with a messaging system; and receive a transmission command via the composition interface; and a messaging component operative to transmit a media composition package from the client device to the messaging system in response to the transmission command, the media composition package comprising the media composition, the media composition package addressed to a media composition distribution group for the user account, the media composition distribution group determined based on, at least in part, a messaging history for the user account with the messaging system. Other embodiments are described and claimed.

Abstract: A method for routing interactions arriving at a contact center. The method may include: providing a cost schedule for the interactions that includes a classifier in which categories of interaction types are defined and a cost value corresponding to each of the interaction types; providing a capacity value for agents; and routing the interactions in accordance with a routing process. The routing process may include: receiving the first interaction; pursuant to the classifier, classifying the first interaction as being a first interaction type; determining the cost value of the first interaction as being the cost value corresponding to the first interaction type in the cost schedule; calculating for each of the agents an available capacity; and routing the first interaction to a first agent based on the available capacity of the first agent being sufficient given the cost value.

Abstract: Systems and methods for managing resources in a multi-tenancy grid computing environment are provided. The methods include an auxiliary service scheduler managing the lifecycle of auxiliary services by policies to optimize the resource allocation; coordinating with an application scheduler to dispatch tasks and an auxiliary service to the same set of servers; and using a multiple layer auxiliary service map to manage and record auxiliary service status. The methods optimize dynamical resource sharing between applications with a dependency on an auxiliary service running aside on the same server.

Abstract: A system for determining qualified contributors of a document is provided. A system can analyze user activities associated with a document to provide meaningful insights indicating the key contributors of the document based on a contributor policy. The contributor policy defines contribution factors and criteria that is used to determine when a user is deemed as a qualified contributor. The contribution factors can be used in an analysis of one or more user activities to determine a contribution score for each user associated with a document. The contribution scores of the users can be compared with a threshold to identify qualified contributors. The identified users are listed as qualified contributors of the document and associated with the document in the metadata. The contributor information can be utilized to inform readers of the document, identity expertise of these contributors, or recommend relevant documents to the contributors.

Abstract: A bioinformatics system for autonomous hypothesis generation for relatedness of seemingly independent concepts having tenuous or no prior known relationship in a knowledgebase. Independent concept pairs are identified in a question, and they are evaluated in light of lists of their similar concepts, and in light input documents, word embeddings, and co-occurrence matrices that are organized as a time series. Relationships between the independent concepts are hypothesized based on temporal analysis of relatedness measures of their similar concepts, measured via candidate connecting concepts.

Abstract: Implementations set forth herein relate to providing a subsidiary application GUI via a client interface. The GUI can be rendered when a user is accessing a first party system via an application that is provided by the first party or a separate entity. The subsidiary application GUI can be rendered in response to the user providing a search query to the first party server—such as a search query that is in furtherance of initializing receiving certain search results. The server can identify, based on the search query, one or more entities that offer primary and/or subsidiary applications, and request subsidiary data for visibly rendering corresponding subsidiary applications for each entity. The subsidiary applications can optionally provide access to application functions that would not otherwise be available at the client without a corresponding application being installed.

Abstract: A method for processing media data includes receiving metadata associated with media data; identifying another metadata associated with another media data, wherein the size of the media data is smaller than the size if the another media data; determining a score based on the metadata and another metadata, wherein the score is within a threshold level; and presenting information associated with a product or service associated with the another media data.

Abstract: A method for guiding installation of smart-home devices may include capturing, by a camera of a mobile computing device, a view of an installation location for a smart-home device; identifying a wire in the view of the installation location for the smart-home device; determining an instruction for connecting the wire to the smart-home device; and displaying the view of the installation location for a smart-home device with the instruction for connecting the wire to the smart-home device.

Abstract: The present teaching generally relates to detecting providing pre-validated data buckets for online experiments. In a non-limiting embodiment, user activity data representing user activity for a first plurality of user identifiers may be obtained. A first set of values and a second values, representing first and second user engagement parameters, respectively, may be generated for each user identifier based on the user activity data. A first ranking and a second ranking may be determined for the first and second sets, respectively. A first exclusion range including a first number of values to be removed from the first and second sets may be determined. A homogenous value set may be generated by removing the first number of values from the first and second sets, where each value from the homogenous value set corresponds to a user identifier available to be placed in a data bucket for an online experiment.

Abstract: A computer-implemented method comprises: receiving a first address and a second address, the first address including n morphemes and the second address including m morphemes, wherein a morpheme is a smallest semantic unit in the address, and wherein n and m are both natural numbers; determining first correlation values between the n morphemes and the m morphemes; obtaining, based on the first correlation values and a preset algorithm, a second correlation value between the first address and the second address; and analyzing a correlation between the first address and the second address based on the second correlation value.

Abstract: A journal of operations persistently stores a collection of user and machine interactions with documents, concepts within documents, and relationships/mentions among such documents and concepts. A knowledge graph or other visual expression can be used to provide an interactive format for user interactions with this information, and to visually display the accumulated information in the journal of operations on user's device of choice. Such accumulations are often called aggregations or aggregates. Both human users and computer users such as automated recommendation engines, machine learning algorithms, and the like can interact with the journal of operations to facilitate computer-assisted discovery and organization of data around a topic of interest to the human user. The resulting journal of operations and aggregations of those operations such as a knowledge graph or other visual expression(s) also provides a useful medium for sharing knowledge with other users and collaborating on investigation of a topic.

Abstract: Systems, methods, and other embodiments associated with updating data aggregation configurations and controlling data aggregation are described. Illustratively, a hierarchy of nodes is generated and displayed in a graphical representation of an energy supply network. The nodes store energy consumption data, and are selectable to identify where aggregations of the energy consumption data are to be performed. A settlement point is created to define an aggregation of the energy consumption data based on at least data from a plurality of the nodes identified from the hierarchy. The created settlement includes one or more characteristics selected from the database. The aggregation of the energy consumption data according to the settlement point is executed and controlled, and an electronic report including the aggregated energy consumption data is generated.

Abstract: Systems, methods, and non-transitory computer readable media configured to generate session information based on information regarding items of a plurality of item types associated with interactions performed by active users of a social networking system. A graph is generated based on the session information. At least a first item of the items is assigned to a cluster based on similarity between the item and the cluster. The cluster is provided to a recommender system to facilitate selection of relevant information for potential presentation to a user.

Abstract: Systems and methods for the display and navigation of a social network or set forth. According to one embodiment a method comprising identifying a first profile in a social network, identifying associated profiles associated with the first profile, ranking the associated profiles, wherein ranking is not based exclusively on a degree of separation, and outputting the associated profiles based at least in part on the ranking is set forth. According to another embodiment a method comprising identifying a user profile, identifying a member profile, determining an association path for the user profile and the member profile, and outputting the association path is set forth.

Abstract: One or more computing devices, systems, and/or methods for displaying videos based upon selectable inputs associated with tags are presented. For example, a video may be identified. A transcript, associated with the video, may be determined. The transcript may comprise a plurality of text segments. The transcript may be analyzed to generate a plurality of sets of tags associated with the transcript. A plurality of selectable inputs may be generated based upon the plurality of sets of tags. A video interface, comprising the plurality of selectable inputs, may be displayed on a first device. A selection of a first selectable input may be received via the video interface. The first selectable input may be associated with a first tag of the plurality of sets of tags and a first time of the video. A second device may display the video based upon the first time of the video.

Abstract: Disclosed are systems, apparatus, methods and computer-readable media for updating information stored in a database system over a network. In some implementations, first contact data is retrieved from a first virtual portion of a database system, where the first contact data provides first contact information associated with at least one entity. In some instances, the first contact data is compared with second contact data, where the second contact data provides second contact information associated with the at least one entity. In some instances, at least some of the second contact information is retrieved from a social networking system. In various implementations, at least one difference between the first contact data and the second contact data is identified, where the at least one difference is capable of being presented in a user interface displayed at a computer system. In some instances, a selection identifying contact data to store is received.

Abstract: A computer-implemented method of auditing drug supply chain data gathered from a distributed ledger is disclosed. The method includes receiving a population of drug product records from the distributed ledger. The method includes receiving a first set of drug product criteria. The method includes determining a weighted probability for one or more drug product records of the population of drug product records. The method includes generating a randomized first subset of drug product records from the population of drug product records based on the weighted probability of the one or more drug product records. Other methods, systems, and the like for unbiased drug selection for audit are also disclosed.

Abstract: According to embodiments of the present invention, an entity may be represented by an unstructured feature vector comprising a plurality of features extracted from unstructured data using semantic analysis and a structural feature vector comprising a plurality of features from chemical structure data. A similarity matrix may be used to compare entities and generate a similarity score, based on both the unstructured feature vector and the structural feature vector for each entity. In some aspects, a user may enter a query (from which a chemical structural feature vector is dynamically generated) to compare against entities having unstructured and/or structural feature vectors, stored in a database.

Abstract: In some implementations, a method includes, for each of multiple hosts: identifying visual leaf pages hosted by the host that are each a web page including image data defining an image or a video that is prominently displayed relative to all other content of the web page, identifying a set of hub pages hosted by the host that each link to at least one of the visual leaf pages through an image-based link, and for each hub page, generating cluster data representing the visual leaf pages to which the hub page links by determining, for each visual leaf page, a set of feature values that each indicate pre-defined features of the visual leaf page, and generating, from the sets of feature values, a set of central feature values as the cluster data for the hub page that indicate a central tendency of each respective pre-defined feature.