Abstract: A dataset management system organizes datasets using a data relationship graph that serves as a representation of datasets that are related to one another. For example, the data relationship graph includes nodes that each represent a dataset as well as edges that each connect two nodes. Each edge represents a similarity in characteristics of the similar datasets such as a similarity in the datasets' origin, organizational schema, lineage, or data values. When a user is to be provided recommended datasets, the dataset management system identifies candidate datasets by traversing the nodes and edges of the data relationship graph. Amongst these candidate datasets, the dataset management system evaluates users that have accessed the candidate datasets as well as the context in which the candidate datasets were accessed to identify recommended datasets for presentation to the user.

Abstract: An online database system receives a tenant dataset from a tenant and automatically generates a mapping that can be used to convert the tenant dataset from a tenant schema to a database schema used by the online database system. To generate the mapping, the online database system analyzes individual entries of the tenant dataset received from the tenant and determines a likelihood of usage. The mapping maintains entries in the tenant dataset that are likely to be used as individual entries in the converted dataset and compresses entries that are less likely to be used into fewer entries to satisfy the database schema. The online database system converts the tenant dataset from the tenant schema to the database schema using the mapping.

Abstract: A dataset management system uses a façade API to automate access to data stores maintained by data tenants. The system receives a configuration request from a data tenant and determines whether the configuration request corresponds to an existing data tenant. If an existing data tenant is identified, the system updates the façade API associated with the existing data tenant based on the configuration request. When a client request for data associated with the data tenant is received, the façade API formats the direct query to include the updated parameters and executes the direct query to retrieve data from the data stores.

Abstract: A proxy API system intercepts and enriches API messages such as an API request or API response. The proxy API system enriches an API message by incorporating additional semantic meaning into the API message. More specifically, the proxy API system extracts features from a message and applies the features to a trained model. The trained model identifies annotations that are relevant for labeling the API message. Additionally, the proxy API system identifies reference data corresponding to the API message data and annotations to provide additional data related to content of the API message, such as additional detail about a data item in the message such as additional fields related to that item. The proxy API system enriches the API request or API response with the annotations and reference data.

Abstract: A dataset management system organizes datasets and tracks the changes to the dataset to adaptively respond to user queries. For a dataset, the dataset management system tracks the evolving schema of the dataset over time as new data values and/or updates to existing data values are incorporated into the dataset. When a query is received, the dataset management system accesses the schema of a dataset to understand how the dataset has changed over time. Given the changing schema of the dataset, the dataset management system can respond by providing recommendations as to suggested queries that can return improved results. As another option, the dataset management system can execute a query and return results that satisfy the query to the client device that provided the query.

Abstract: A code repository receives and analyzes changed code to determine whether and how to automatically instrument the code. The code repository identifies locations within the code, such as locations next to external function calls, to insert instrumentation markers. The code repository determines the information to be collected at the location. Redundant information or information that can be derived from other collected information need not be collected. When the instrumented code is executed, the execution of instrumentation markers causes the collection of information. The subsets of information can be used to troubleshoot the code.

Abstract: A code repository receives and analyzes changed code to determine whether and how to automatically instrument the code. The code repository identifies locations within the code, such as locations next to external function calls, to insert instrumentation markers. The code repository determines the information to be collected at the location. Redundant information or information that can be derived from other collected information need not be collected. When the instrumented code is executed, the execution of instrumentation markers causes the collection of information. The subsets of information can be used to troubleshoot the code.

Abstract: An online database system receives a tenant dataset from a tenant and automatically generates a mapping that can be used to convert the tenant dataset from a tenant schema to a database schema used by the online database system. To generate the mapping, the online database system analyzes individual entries of the tenant dataset received from the tenant and determines a likelihood of usage. The mapping maintains entries in the tenant dataset that are likely to be used as individual entries in the converted dataset and compresses entries that are less likely to be used into fewer entries to satisfy the database schema. The online database system converts the tenant dataset from the tenant schema to the database schema using the mapping.

Abstract: A dataset management system uses a façade API to automate access to data stores maintained by data tenants. The system receives a configuration request from a data tenant and determines whether the configuration request corresponds to an existing data tenant. If an existing data tenant is identified, the system updates the façade API associated with the existing data tenant based on the configuration request. When a client request for data associated with the data tenant is received, the façade API formats the direct query to include the updated parameters and executes the direct query to retrieve data from the data stores.

Abstract: An extract, transform, load (ETL) orchestration interface receives an API script defining an ETL orchestration to extract data using a plurality of data projections and integrate the extracted data. Each data projection is defined as an API call and is associated with a particular data store. The ETL orchestration interface also receives a schedule for executing the ETL orchestration. The API script and the schedule are stored in an orchestrations store. An execution service executes the stored API script at a particular point of tie determined based on the schedule. Executing the stored API involves translating the API call for each data projection into a data query for extracting data from the data store associated with the data projection, executing the translated queries to extract the data, and integrating the extracted data according to the ETL orchestration.

Abstract: A proxy API system intercepts and enriches API messages such as an API request or API response. The proxy API system enriches an API message by incorporating additional semantic meaning into the API message. More specifically, the proxy API system extracts features from a message and applies the features to a trained model. The trained model identifies annotations that are relevant for labeling the API message. Additionally, the proxy API system identifies reference data corresponding to the API message data and annotations to provide additional data related to content of the API message, such as additional detail about a data item in the message such as additional fields related to that item. The proxy API system enriches the API request or API response with the annotations and reference data.

Abstract: A dataset management system organizes datasets using a data relationship graph that serves as a representation of datasets that are related to one another. For example, the data relationship graph includes nodes that each represent a dataset as well as edges that each connect two nodes. Each edge represents a similarity in characteristics of the similar datasets such as a similarity in the datasets' origin, organizational schema, lineage, or data values. When a user is to be provided recommended datasets, the dataset management system identifies candidate datasets by traversing the nodes and edges of the data relationship graph. Amongst these candidate datasets, the dataset management system evaluates users that have accessed the candidate datasets as well as the context in which the candidate datasets were accessed to identify recommended datasets for presentation to the user.

Abstract: A dataset management system organizes datasets and tracks the changes to the dataset to adaptively respond to user queries. For a dataset, the dataset management system tracks the evolving schema of the dataset over time as new data values and/or updates to existing data values are incorporated into the dataset. When a query is received, the dataset management system accesses the schema of a dataset to understand how the dataset has changed over time. Given the changing schema of the dataset, the dataset management system can respond by providing recommendations as to suggested queries that can return improved results. As another option, the dataset management system can execute a query and return results that satisfy the query to the client device that provided the query.

Abstract: One or more techniques and/or systems are disclosed for detecting and/or mitigating a potentially compromised online user account. One or more baselines can be established for a user's online account to determine a normal usage pattern for the account by the user (e.g., frequency of incoming/outgoing emails, text messages, etc.). The online user account can be periodically or continually monitored for use of the same resources used to determine the baseline(s). If a deviation from the baseline is detected, the deviation may be compared against a threshold to determine whether the deviation indicates that the account may be compromised. When an indication of a potentially compromised account is detected, the user can be notified of the indication, so that one or more actions can be taken to mitigate the potentially compromised account.

Abstract: One or more techniques and/or systems are disclosed for mitigating machine solvable human interactive proofs (HIPs). A classifier is trained over a set of one or more training HIPs that have known characteristics for OCR solvability and HIP solving pattern from actual use. A HIP classification is determined for a HIP (such as from a HIP library used by a HIP generator) using the trained classifier. If the HIP is classified by the trained classifier as a merely human solvable classification, such that it may not be solved by a machine, the HIP can be identified for use in the HIP generation system. Otherwise, the HIP can be altered to (attempt to) be merely human solvable.

Abstract: Embodiments of message organization and spam filtering based on user interaction are presented herein. In an implementation, user interaction with a plurality of messages in a user interface is monitored, which includes analyzing an amount of time spent by a user in interacting with each message. Subsequent messages may then be filtered based on the monitored user interaction. In another implementation, messages are processed that are received via a network using a spam filter that was generated based on monitored interaction of a user with previous messages. The processing results in a value describing a relative likelihood of importance of each of the processed message to the user. The processed messages are then arranged for display in an order, one to another, in a user interface based on respective values.

Abstract: One or more techniques and/or systems are disclosed for detecting a potentially abusive action for an online social network of a user. A network expansion profile and/or communications profile may be determined for the online social network of the user, by tracking user actions while building and/or maintaining the online social network. A reputation profile, such as a rating, for the online social network can be determined by combining information from the network expansion profile for the online social network and/or the communication profile for the online social network. Based upon the determined reputation profile, an action of the online social network, such as sending a communication, may be identified as potentially abusive and thus be treated as such (e.g., be accompanied by a warning to intended recipients that the communication may be from a spammer).

Abstract: One or more techniques and/or systems are disclosed for detecting and/or mitigating a potentially compromised online user account. One or more baselines can be established for a user's online account to determine a normal usage pattern for the account by the user (e.g., frequency of incoming/outgoing emails, text messages, etc.). The online user account can be periodically or continually monitored for use of the same resources used to determine the baseline(s). If a deviation from the baseline is detected, the deviation may be compared against a threshold to determine whether the deviation indicates that the account may be compromised. When an indication of a potentially compromised account is detected, the user can be notified of the indication, so that one or more actions can be taken to mitigate the potentially compromised account.

Abstract: One or more techniques and/or systems are disclosed for mitigating machine solvable human interactive proofs (HIPs). A classifier is trained over a set of one or more training HIPs that have known characteristics for OCR solvability and HIP solving pattern from actual use. A HIP classification is determined for a HIP (such as from a HIP library used by a HIP generator) using the trained classifier. If the HIP is classified by the trained classifier as a merely human solvable classification, such that it may not be solved by a machine, the HIP can be identified for use in the HIP generation system. Otherwise, the HIP can be altered to (attempt to) be merely human solvable.

Abstract: Functionality is described for addressing a threat to the security of a user device that utilizes a network-accessible service. The functionality operates by assessing the likelihood that the user device is infected by the undesirable item. When the user device makes a request to access the network-accessible service, the functionality can interact with the user device in a manner that is governed by the assessed likelihood that the user device is infected by the undesirable item.