Abstract: Systems, methods, and non-transitory computer-readable media can obtain respective instances of verification information can be obtained from computing systems associated with a plurality of third-party vendors, wherein a computing system associated with a third-party vendor determines verification information based at least in part on an analysis of an identification document provided by a user to be authenticated. The respective instances of verification information can be interpreted, wherein interpreting an instance of verification information comprises assigning respective states to attribute fields included in the verification information. A set of interpreted instances of verification information can be evaluated in combination to determine whether to authenticate the user, wherein the set includes an interpreted instance of verification information from each of the plurality of third-party vendors.

Abstract: The disclosed computer-implemented method may include (1) for each tenant in a plurality of tenants within a multi-tenant service system, assigning a probability factor to the tenant that indicates a likelihood that the tenant will be selected when a resource of the multi-tenant service system is available, (2) detecting that the resource of the multi-tenant service system is available, (3) probabilistically selecting a tenant from the plurality of tenants by using the probability factors assigned to the tenants in the plurality of tenants, and (4) directing the multi-tenant service system to allocate the resource to the selected tenant for execution of a work item received from the selected tenant. Various other methods, systems, and computer-readable media are also disclosed.

Abstract: Systems, methods, and non-transitory computer-readable media can generate an initial alpha mask for an image based on machine learning techniques. A plurality of uncertain pixels is defined in the initial alpha mask. For each uncertain pixel in the plurality of uncertain pixels, a binary value is assigned based on a nearest certain neighbor determination.

Abstract: The disclosed computer-implemented method may include (1) for each tenant in a plurality of tenants within a multi-tenant service system, assigning a probability factor to the tenant that indicates a likelihood that the tenant will be selected when a resource of the multi-tenant service system is available, (2) detecting that the resource of the multi-tenant service system is available, (3) probabilistically selecting a tenant from the plurality of tenants by using the probability factors assigned to the tenants in the plurality of tenants, and (4) directing the multi-tenant service system to allocate the resource to the selected tenant for execution of a work item received from the selected tenant. Various other methods, systems, and computer-readable media are also disclosed.

Abstract: An online system receives data and processes the data in a data processing pipeline. To data loss in the data processing pipeline, the online system determines a time interval during which each item of data is received and associates a set of counters with each time interval. For each time interval, an input counter is incremented for each data item received during the time interval and an output counter is incremented for each data item received during the time interval that was processed by the data processing pipeline. The online system compares an input number from the input counter and an output number from the output counter for each time interval. Based on a difference between the input number and output number for a time interval, the online system determines if a loss of data received during the time interval occurred. Lost Data are identified and processed.

Abstract: The disclosed computer-implemented method may include (1) for each tenant in a plurality of tenants within a multi-tenant service system, assigning a probability factor to the tenant that indicates a likelihood that the tenant will be selected when a resource of the multi-tenant service system is available, (2) detecting that the resource of the multi-tenant service system is available, (3) probabilistically selecting a tenant from the plurality of tenants by using the probability factors assigned to the tenants in the plurality of tenants, and (4) directing the multi-tenant service system to allocate the resource to the selected tenant for execution of a work item received from the selected tenant. Various other methods, systems, and computer-readable media are also disclosed.

Abstract: An online system processes data in a distributed processing system. To detect loss and corruption of data, the online system periodically stores information describing states of data processed during various time intervals and updates the information to include changes occurring within a predetermined period. Based on states of data described by information stored at a time, the online system performs a modified process on data received or processed during a time interval. For each item of data on which the modified process was performed, the online system compares a modified state of the data item to a state described by information stored at an additional time to determine if data was lost or corrupted. Lost or corrupted data is identified and processed based on the state of data described by the information stored at the time.

Abstract: A report execution engine processes reports for evaluating advertising campaigns. The report execution engine maintains a plurality of generators, each generator including instructions to generate execution plans for one or more columns of a report. The report execution engine uses different generators for generating sub-plans for processing columns of the report. A generator generates a sub-plan for processing a column of the report and invokes other generators for generating sub-plans for remaining columns of the report. The generator receives a sub-plan for the remaining columns of the report from another generator and generates a combined execution plan based on the sub-plans. An execution plan comprises a hierarchical structure of nodes, each node representing a portion of computation of the report. The report execution engine generates the report based on the execution plan and sends the results for presentation.

Abstract: Systems, methods, and non-transitory computer-readable media can generate an initial alpha mask for an image based on machine learning techniques. A plurality of uncertain pixels is defined in the initial alpha mask. For each uncertain pixel in the plurality of uncertain pixels, a binary value is assigned based on a nearest certain neighbor determination.

Abstract: An online system processes data in a distributed processing system. To detect loss and corruption of data, the online system periodically stores information describing states of data processed during various time intervals and updates the information to include changes occurring within a predetermined period. Based on states of data described by information stored at a time, the online system performs a modified process on data received or processed during a time interval. For each item of data on which the modified process was performed, the online system compares a modified state of the data item to a state described by information stored at an additional time to determine if data was lost or corrupted. Lost or corrupted data is identified and processed based on the state of data described by the information stored at the time.

Abstract: An online system receives data and processes the data in a data processing pipeline. To data loss in the data processing pipeline, the online system determines a time interval during which each item of data is received and associates a set of counters with each time interval. For each time interval, an input counter is incremented for each data item received during the time interval and an output counter is incremented for each data item received during the time interval that was processed by the data processing pipeline. The online system compares an input number from the input counter and an output number from the output counter for each time interval. Based on a difference between the input number and output number for a time interval, the online system determines if a loss of data received during the time interval occurred. Lost Data are identified and processed.

Abstract: A settings service running on a server remote from a computing device can identify at least one settings group to which the computing device belongs. The settings service can identify all settings that are included in the at least one settings group to which the device belongs. The settings service can compare all the settings and detect any conflicting settings. If there are conflicting settings, the service can choose which of the conflicting settings takes priority, in order to resolve the conflict. For example, a default setting can be overridden by a setting with a higher priority. After resolving the conflict, the settings service can merge the conflicting setting(s) which takes priority, as well as the non-conflicting settings, into a single set of settings to be applied to the device. The settings service can transmit the set of settings to the computing device to be applied.

Abstract: Persistent connections between multiple client devices and multiple back-end service components are managed using a consistent hashing-based approach to route distribution. A load balancer distributes the connections across multiple gateway servers. Each connection is associated with a device having an identifier, which can be hashed using a selected hashing algorithm. The gateway servers are assigned values over a hashing range. When a connection is established for a device, the hash value for that device can be mapped to a corresponding gateway server. The primary gateway server establishing the connection can store information (e.g., the port or interface) for the connection, and can send identifying information to the corresponding gateway server determined by the hash value. When a backend service wants to locate the connection, the service hashes the device identifier to determine the corresponding gateway server, which returns the identity of the primary gateway server hosting the connection.

Abstract: This disclosure describes systems, methods, and computer-readable media related to semantic processing of content for product identification. Content may be received from a user device. The content may be processed based at least in part on one or more content filters. At least a portion of the processed content may be analyzed with named-entity recognition to identify one or more product references. A confidence score associated with each of the one or more product references may be calculated. Data associated with the one or more product references may be obtained. The data associated with the one or more product references may be transmitted for presentation on the user device.

Abstract: Approaches are described for automatically generating new security credentials, such as security tokens, which can involve automatically re-authenticating a user (or client device) using a previous security token issued to that user (or device). The re-authentication can happen without any knowledge and/or action on the part of the user. The re-authentication mechanism can invalidate and/or keep track of the previous security token, such that when a subsequent request is received that includes the previous security token, the new security token can be invalidated, and the user caused to re-authenticate, as receiving more than one request with the previous security token can be indicative that the user's token might have been stolen.

Abstract: A demand side platform (DSP) performs lift analysis on one or more advertisements provided by an entity to determine how effective the one or more advertisements cause users to perform various actions. To reduce bias in the lift analysis, when the DSP identifies an opportunity to present an advertisement to a user, the DSP identifies candidate advertisements for presentation to the user and selects a candidate advertisement. If the candidate advertisement is undergoing lift analysis and the user is in a holdout group that is not presented with the candidate advertisement, the DSP selects an alternative advertisement and associates a bid amount for the selected advertisement with the alternative advertisement. If the alternative advertisement is presented to the user, the DSP may allocate an amount charged between an entity associated with the alternative advertisement and an entity associated with the selected advertisement.

Abstract: Different types of connections can be established between a device and an endpoint (e.g., an address associated with a service) based at least in part upon the recent connection history for that endpoint. A connection manager can monitor the connections, and properties of those connections, from a number of different applications of a device to one or more endpoints. If the information from the system as a whole indicates that properties for the connection should be modified, an appropriate can be established for that endpoint, which can be used by any or all of those applications. The connection can meet or exceed any policies for the request, as well as other requests that can utilize the connection. Various other factors can be used in determining which types of connections to provide, as may include existing connections, system load, user type, and other such information.

Abstract: A routing device can maintain a long-lived, bidirectional connection with a network server. The routing device can establish a TCP connection with each of multiple computing devices. One or more computing devices can make (e.g., HTTP) requests for network resources. The routing device can receive the requests and decide to direct the requests to the network server via the long-lived, bidirectional connection. Based on the requests, the network server can make requests to other servers for the resources and send modified (e.g., assembled, filtered, optimized, etc.) responses back to the routing device, which can transmit the responses to the computing devices. This process can be transparent to the computing devices. In some embodiments, the routing device can optionally establish a long-lived TCP connection with a computing device capable of the long-lived TCP connection option. The routing device can maintain the long-lived connection with the network server for the computing device.

Abstract: A report execution engine processes reports for evaluating advertising campaigns. The report execution engine maintains a plurality of generators, each generator including instructions to generate execution plans for one or more columns of a report. The report execution engine uses different generators for generating sub-plans for processing columns of the report. A generator generates a sub-plan for processing a column of the report and invokes other generators for generating sub-plans for remaining columns of the report. The generator receives a sub-plan for the remaining columns of the report from another generator and generates a combined execution plan based on the sub-plans. An execution plan comprises a hierarchical structure of nodes, each node representing a portion of computation of the report. The report execution engine generates the report based on the execution plan and sends the results for presentation.

Abstract: An online system maintains information describing user interactions with advertisements and other content. Additionally, the online system maintains various types of information identifying users and confidence values associated with each type of information identifying users that provide a measure of the likelihood of a type of information identifying a specific user. Example types of information identifying a user include an online system identifier, a client device identifier, and an application identifier. When the online system receives information indicating a user has performed a conversion, the online system retrieves interactions associated with at least one type of information identifying the user. An attribution model is applied to the retrieved interactions to determine contributions of various interactions to the conversion, accounting for the confidence values associated with types of information identifying the user associated with various interactions.