Abstract: Methods and apparatus are disclosed for extracting structured content, as graphs, from text documents. Graph vertices and edges correspond to document tokens and pairwise relationships between tokens. Undirected peer relationships and directed relationships (e.g. key-value or composition) are supported. Vertices can be identified with predefined fields, and thence mapped to database columns for automated storage of document content in a database. A trained neural network classifier determines relationship classifications for all pairwise combinations of input tokens. The relationship classification can differentiate multiple relationship types. A multi-level classifier extracts multi-level graph structure from a document. Disclosed embodiments support arbitrary graph structures with hierarchical and planar relationships. Relationships are not restricted by spatial proximity or document layout. Composite tokens can be identified interspersed with other content.

Abstract: A large language model can be used to implement a service assistant. Natural language commands can be sent to the large language model, which identifies intents and responds with actions and API payloads. The command can then be implemented by an appropriate API call. The assistant can support actions that span a plurality of applications. A wide variety of human languages can be supported, and the large language model can maintain context between commands. Useful functionality such as prompting for missing parameters and the like can be supported.

Abstract: Systems and methods described herein relate to the efficient handling of data purge requests in the context of a distributed storage system. A plurality of data purge requests is stored in a first data structure. The data purge requests may be grouped into batches that are processed at least partially in parallel. A first data purge request from the plurality of data purge requests is successfully processed, and is moved from the first data structure to a second data structure. Processing of a second data purge request from the plurality of data purge requests is unsuccessful. The second data purge request is retained in the first data structure. Purge status data is generated based on the first data purge request being in the second data structure and the second data purge request being in the first data structure. The purge status data may be presented at a user device.

Abstract: A partition-based method to implement idempotence in an event-driven architecture is provided, which acts to reduce the pressure of database reading and writing, thus improving the performance of event handling. In order to improve parallelism, a consumer cluster (a cluster of multiple consumers that can share the responsibility of processing events) is established such that the events are distributed among different partitions within the cluster. Each partition contains a different consumer worker (with its own cache). An event dispatch module is then used to dispatch events among the consumer workers in a way that keeps the load balanced. In an example embodiment, this is performed by generating a partition identification for an event based on a hash of the context identification of the session established between the user and the application when the event was generated, followed by a modulo operation based on the number of consumer workers in the cluster.

Abstract: Provided are systems and methods for creating histograms with distinct value sketches integrated therein and for query processing based on the histograms with distinct value sketches. In one example, the method may include storing a histogram that comprises a representation of a bucket of data from a database and that includes a distinct value sketch with a distinct value attribute that identifies an estimated number of distinct values within the bucket of data, receiving a database query, generating a query execution plan for the database query based on the distinct value attribute of the bucket within the distinct value sketch embedded within the histogram, and executing the database query on the bucket of data from the database based on the generated query execution plan.

Abstract: In an example embodiment, a common, composable abstraction is provided that allows components to work efficiently across programming languages and services without the need to write glue code. Application developers can concentrate on the application logic itself. The functionality of services can be developed by framework developers only once, using the programming language of their choice.

Abstract: A data annotation server accesses a request from a machine learning server for annotated images of a user interface containing a specified user interface element. The data annotation server programmatically determines whether user interfaces generated by an application server include the specified user interface element. If so, an image of the user interface is stored and a location or bounding box of the user interface element is determined. The stored image of the user interface is annotated with the determined location of the user interface element. The image and the annotation are provided to the machine learning server, which uses the images and annotations to train a machine learning model.

Abstract: A computer-implemented method may comprise creating a first view of a first data source comprising a first online analytical processing (OLAP) cube based on a first user input, creating a second view of a second data source based on a second user input, combining the first view and the second view, and creating metadata objects for elements of the first view and the second view. The method may further comprise generating a query execution plan comprising a first native query and a second native query based on a user-defined query specification and the metadata objects, executing the first native query on the first data source to retrieve a first dataset from the first data source and the second native query on the second data source to retrieve a second dataset from the second data source, and generating a federated dataset using the first dataset and the second dataset.

Abstract: A method, a system, and a computer program product for decompressing data. One or more compressed blocks in a set of stored compressed blocks responsive to a request to access data in the set of stored compressed blocks are identified. String prefixes inside the identified compressed blocks are decompressed using front coding. String suffixes inside the identified compressed blocks are decompressed using a re-pair decompression. Uncompressed data is generated.

Abstract: A method includes receiving a message query from an entity identifier participating in a social network. The message query specifies one or more entities, one or more requirements, and one or more constraints. A set of message query parameters is generated based on the message query. A set of queries for a semantic graph of the social network is generated based on the set of message query parameters. The set of queries is applied to the semantic graph to obtain a set of query results. A message context of the entity identifier is determined based on the set of query results and the set of message query parameters. A set of messages from a message repository is determined based on the message context. The set of messages can be presented on a client computer associated with the entity identifier.

Abstract: The present disclosure provides techniques and solutions for automatically and dynamically supplementing user prompts to large language models with information to be used by the large language model in formulating a response. In particular, entities are identified in the original prompt. A semantic framework is searched for information about such entities, and such information is added to the original user prompt to provide a modified user prompt. In a particular example, the identified entities comprise triples, and verbalized triples are added to provide the modified user prompt. The modified prompt may be hidden from the user, so that a response of the large language model appears to be in response to the original prompt.

Abstract: Certain aspects of the disclosure concern a computer-implemented method for improved data security in large language models. The method includes receiving a prompt query entered through a user interface, extracting a plurality of named entities from the prompt query and classifying the plurality of named entities into respective entity classes, tagging the plurality of named entities to be security compliant or security noncompliant based on the respective entity classes, and responsive to finding that one or more named entities are tagged to be security noncompliant, generating an alert on the user interface.

Abstract: Various examples are directed to systems and methods for installing a plugin to a cloud-implemented database management application. A shared container file system may be initiated at a cloud environment. A plugin container image may be accessed, where the plugin container image comprises plugin payload data describing a first plugin to the database management application and a copy executable. A plugin container may be started at the cloud environment, where the plugin container is based at least in part on the plugin container image. The plugin container may be mounted to the shared container file system. The copy executable may be executed to copy the plugin payload data to the shared container file system. The first plugin may be installed to a database management application instance executing at the cloud environment.

Abstract: A computer implemented method can receive an incoming query statement, identify a database object specified in the incoming query statement, and search a query hint registry for a first hint record which includes the database object and a first hint paired with the database object. Responsive to finding the first hint record in the query hint registry, the method can generate a modified query statement by appending the first hint to the incoming query statement and obtain a query execution plan based on the modified query statement.

Abstract: Automated digital assistant generation can be implemented via large language models. Domain-specific documents can be loaded into a large language model that is then prompted to generate intents, entities, exemplar utterances, and the like. Such configuration components can then be assembled into a digital assistant definition that can then be deployed as a digital assistant for the domain in question. Skills can be aggregated so that the digital assistant can address the domain along with other domains, whether closely related or not.

Abstract: In an example embodiment, a protocol for private set intersection is introduced that provides for two-party computation. Each party has a private data set and both parties want to securely compute the intersection of their sets, such that only the result is revealed and nothing else. Construction rules are provided that rely on the evaluation of a branching program (BP) using a fully homomorphic encryption (FHE) scheme. Using the properties of an FHE scheme, a non-interactive protocol is built with extendable functionalities. Thus, not only can the intersection be securely computed but the result can be used for further secure computations. Furthermore, the communication overhead for practical applications is independent of the server's set size, allowing for easy scalability.

Abstract: Methods, systems, and computer-readable storage media for generating an expected configuration checksum based on a configuration file associated with an ETL job, the ETL job being executable to provide a target entity for consumption by one or more consuming applications, the target entity including data of one or more source entities; retrieving an ETL timestamp indicating a last time that the ETL job was executed, and determining, based on one of the expected configuration checksum and the ETL timestamp, that a target data schema of the target entity is to be updated, and in response, providing target metadata for the target entity and updating the target data schema based on the target metadata to provide an updated target data schema.

Abstract: Methods, systems, and computer-readable storage media for receiving, through a set of user interfaces (UIs), user input including job configuration information and deployment information for an application, providing a CI/CD job for the application with a CI/CD service using the job configuration information and the deployment information, and triggering, in response to a commit of changes to the application in a repository, automated build of the application, and in response, automatically: generating a development descriptor file at least partially based on the user input, providing an archive file including the development descriptor file, and processing, by the CI/CD job executed by the CI/CD service, the archive file to deploy the application within the target environment.

Abstract: Systems, methods, and computer-readable media for performing feature engineering on a dataset for predictive modeling are disclosed. A dataset may comprise a plurality of features that are used for the predictive model. The dataset may be fed to a neural network to determine which features have the greatest impact on the predictive model and which features do not positively impact the predictive model. A deep reinforcement learning agent may select an action to perform on the dataset. The action may be applied to the dataset to generate new features and obtain a transformed dataset. Features that do not positively impact the predictive model may be removed from the dataset. A reward may be calculated for the transformed dataset. The transformed dataset and the reward may be passed to the neural network for further iteration and optimization of the features for the predictive model.

Abstract: A computing system configured to perform artificial intelligence-powered search and evaluation functionalities is disclosed. The computer system can receive a natural language search input entered through a user interface, request a search engine to perform a search based on the natural language search input, receive search results returned by the search engine, and submit, in runtime, structured data contained in the search results to a large language model. The structured data includes a table including one or more records and a plurality of fields. The computing system can prompt, in runtime, the large language model to generate scores for the one or more records and present the structured data and the scores on the user interface.