Abstract: Systems, methods, and computer-readable media are disclosed for remote monitoring and dynamic document management. Example methods may include receiving a first document from a device driver associated with a hardware device, identifying a clinical trial identifier associated with the first document based at least in part on metadata associated with the first document, and determining a first document type of the first document by analyzing contents of the first document. Methods may include managing user permission to access the first document. Methods may further include identifying a folder associated with the clinical trial identifier at which to store the first document based at least in part on the first document type, and assigning a first document state to the first document based at least in part on the first document type of the first document.

Abstract: A source table organized into a set of batch units is accessed. The source table comprises a column of data corresponding to a semi-structured data type. One or more indexing transformations for an object in the column are generated. The generating of the one or more indexing transformation includes converting the object to one or more stored data types. A pruning index is generated for the source table based in part on the one or more indexing transformations. The pruning index comprises a set of filters that index distinct values in each column of the source table, and each filter corresponds to a batch unit in the set of batch units. The pruning index is stored in a database with an association with the source table.

Abstract: Embodiments provide a computer-executable method, computer system and non-transitory computer-readable medium for programmatically generating an association among two or more purchase categories based on purchase data of a plurality of consumers. The method includes programmatically accessing, from a dataset via a network device, prior purchase data associated with purchases of a plurality of commercial objects by a plurality of consumers. The method also includes programmatically identifying a plurality of categories associated with the plurality of commercial objects. The method also includes, for each consumer in the plurality of consumers, programmatically generating a total number of purchases by the consumer in each category in the plurality of categories.

Abstract: Disclosed are systems, apparatus, methods, and computer readable media for discontinuing the following of records in an on-demand database service environment. In one embodiment, one or more records are identified according to one or more designated criteria. The records are stored in a database. The records are being followed by one or more users in an on-demand database service environment such that updates to the records are capable of being stored as feed items in one or more feed tables in the database. The feed items are capable of being published in an information feed displayed on one or more devices associated with the one or more users. The following of the identified one or more records by the one or more users is discontinued.

Abstract: A system for extensible search includes a processor, non-transitory machine readable storage medium communicatively coupled to the processor, and a search application. The search application includes computer-executable instructions on the medium that are readable by the processor. The search application is configured to accept a search string, parse the search string to determine a category of data to be searched and determine a filter of results of a search, recommend a suggested search string based on the accepted search string and a previous search, accept selection of the suggested search string, modify the search string in response to acceptance of the suggested search string, and perform the search of the contents and settings of the plurality of clients for the category of data as limited by at least the filter.

Abstract: A process for securely processing a search query for homomorphically encrypted search results is provided. The process includes receiving a search query from a requesting device, the search query including a request for encrypted data stored in a data store operably coupled to the one or more processors. The process further includes executing a search function to access a set of search results from the data store, the search results including homomorphically encrypted information, and determining, based upon the search query, a transmission encryption technique for secure transmission of the search results to the requesting device. The search results can be additionally encrypted using the transmission encryption technique to generate an additionally encrypted search response that includes the homomorphically encrypted information. The process can further include causing transmission of the additionally encrypted search response to the requesting device.

Abstract: A data harvester enhances compliance audits by characterizing data sources, sampling data in one or more of the data sources to determine likelihood of success of the data harvest, estimating time for the data harvest, making recommendations from the samples based on machine learning relating to previous runs, then sampling additional data while estimated expected completion time. The harvested data may then be analyzed and compared to compliance requirements, and a compliance report may be generated.

Abstract: Disclosed herein are embodiments of systems, methods, and products comprises an analytic server for data rationalization and migration. The server receives data from a plurality of electronic data sources and stores the received data into a database within a central hub. The server demobilizes the data and performs enhancement on the data to prepare the data for complex calculation. The server copies the enhanced data to an operational data store, performs various calculation functions that filer, rationalize the data, and create data linkage. The server copies back the calculation result to the database for further processing and downstream consumption. The calculation results provide a consolidated view of data from different source systems and satisfy the reporting needs for different consumers. The server uses common data model to generate reports for different consumers. After the data migration and consolidation into a central hub, the server decommissions legacy components.

Abstract: Systems and methods for classification of log data at the source into sets of critical and non-critical data. Critical data may be indexed and processed normally, while non-critical data may be provided to and stored by the cloud-based log aggregation system without indexing, at significantly lower cost in terms of processing and storage. In the event that non-critical data is required for troubleshooting or analysis, the non-critical data may be indexed dynamically on request. Because the non-critical data is stored at the cloud-based log aggregation system, it may be quickly indexed and added to the critical data, without additional consumption of bandwidth or delays due to transmission. Dynamic selection and classification of critical and non-critical data may thus allow an enterprise to balance costs and availability of indexed data.

Abstract: A transaction information aggregation system includes raw transaction and enriched transaction databases. The raw transaction database comprises transaction information for a plurality of transactions, including for each transaction an account holder identifier and at least one transaction parameter. The enriched transaction database comprises an indexed transaction record for each of the plurality of transactions, each record including a plurality of search terms associated with each of the at least one transaction parameter.

Abstract: Disclosed herein are systems and methods for pruning external data. In an embodiment, a database platform receives a query directed at least in part to external data in an external table on an external data storage platform. The external table includes partitions that correspond to storage locations in a source directory of the external data storage platform. The storage locations contain files that contain the external data. The database platform identifies, from external-table metadata that is stored by the database platform and that maps the partitions of the external table to the storage locations in the source directory, a subset of the partitions as including data that potentially satisfies the query. The database platform identifies data that satisfies the query by scanning the identified subset of the partitions, and responds to the query at least in part with the identified data that satisfies the query.

Abstract: A browse node page is addressed by its path. As a result, a web crawler may be more likely to find the browse node than the corresponding parameter-based search page. The browse node may be further distinguished from a search result page by the use of a title or header metatag that indicates information about the browse node and distinguishes the browse node from a general search result page. The number of combinations of keywords, categories, and key-value pairs makes the creation of a browse node for every possible combination prohibitive in all but the simplest applications. Methods and systems to identify which search result pages should be converted to browse nodes are disclosed herein.

Abstract: Certain aspects of the present disclosure are directed to methods and apparatus for compressing video content using deep generative models. One example method generally includes receiving video content for compression. The received video content is generally encoded into a latent code space through an auto-encoder, which may be implemented by a first artificial neural network. A compressed version of the encoded video content is generally generated through a trained probabilistic model, which may be implemented by a second artificial neural network, and output for transmission.

Abstract: Described are improved systems, computer program products, and methods for an improved approach to access small to medium size objects (MOBs) stored in LOB data type columns of a RDBMS. The approach includes receiving a SQL statement comprising a retrieval of a large object (LOB). The approach also includes determining whether to return a value of the LOB or a reference to a storage location storing the value of the LOB based on: a data dictionary property of the LOB to return the value of the LOB, a function included in the SQL statement to return the value of the LOB, or a flag derived from a SQL operator tree propagation to return the value of the LOB.

Abstract: Search results are improved, in one embodiment, through the use of subscription information about the status of a subscription to content. The subscription information can be transmitted to a cloud storage service used by a user account of a device; later, a search query from the device causes the receipt of search results from a search using the subscription information and these search results can show the status of the subscription such as whether the content is playable.

Abstract: Technologies are provided for dynamically adaptive Bloom filters. After processing a lookup operation using a Bloom filter, a feedback message is received which can be used to determine whether the Bloom filter produced a false-positive lookup result. A Bloom filter accuracy score can be updated based on the feedback message. If the accuracy score for the Bloom filter falls below a specified threshold then the Bloom filter can be changed, or replaced with a different Bloom filter. In at least some embodiments, a plurality of Bloom filters can be created. One of the Bloom filters, designated as an active Bloom filter, can be used to process lookup operations. When a feedback message is received, accuracy scores can be updated for each of the Bloom filters. Based on the accuracy scores, the active Bloom filter can be deactivated and another Bloom filter, with a better accuracy score, can be activated.

Abstract: Conditionally appending an event in an ordered event stream (OES) is disclosed. An OES storage system can comprise an OES that can have a dynamic topology, e.g., comprising changing numbers of segments, segments that have changing routing key ranges, mixes of local and remote storage devices and processors, employing different types of storage, or combinations thereof. A conditional append operation can be predicated on determining an appropriate segment, for example, based on a routing key. Conditions of the conditional appends can be designated by a user. Failing conditions can trigger a corrective action, throwing of an error. Satisfied conditions can result in permitting the conditional append. Moreover, the conditional append can be dependent on populating a group of attributes for the correct segment based on attributes that can be determined for a different epoch of the OES, again for example, based on a routing key.

Abstract: Systems and methods are disclosed herein that relate to secure monitoring or interception of traffic in a wireless communications system. In some embodiments, a method of operation of a network node comprises receiving a list of one or more obfuscated target identifiers from a monitoring node, where each obfuscated target identifier is a user identifier of a target user that is encrypted using a first encryption key that is unknown to the network node. The method further comprises receiving an encrypted packet from another network node and determining whether an encrypted user identifier of the encrypted packet matches one of the obfuscated target identifiers. The method further comprises, if the encrypted user identifier matches one of the obfuscated target identifiers, further encrypting the encrypted packet using a second encryption key negotiated between the network node and the monitoring node and transmitting the further encrypted packet to the monitoring node.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for preserving user privacy when collecting and analyzing user data. Methods include discarding user data after generating a privacy enhanced representation of user data and storing this privacy enhanced representation in a probabilistic data structure. Methods further preserve user privacy by inserting false negatives in the probabilistic data structure. Methods also include storing continuous value attributes in the data structure. Methods include storing date values associated with one or more user attributes. The date values may specify the date when one or more user attributes was last received.

Abstract: The present application discloses a data configuration query method and device. The method includes: determining at least two target data sets required to be queried by a query instruction and an ordered association between the target data sets, wherein the ordered association at least includes a unidirectional association and/or a bidirectional association; generating, on the basis of the at least two target data sets and the ordered association between the target data sets, a target association path; and outputting an OLAP model conforming to the target association path in a database. According to the present application, the technical problems of a large demand quantity of OLAP models included in an OLAP query system and a low utilization rate of the OLAP models in the related art can be solved.

Abstract: Embodiments of the present invention provide systems, methods, and computer storage media for sentence compression in which a provided sentence is compressed to fit within an allotted space. Portions of the input sentence are copied to generate the compressed sentence. Upon receipt of a sentence, top candidate compressed sentences may be determined based on probabilities of segments of the input sentence to be included in a potential compressed sentence. The top candidate compressed sentences are re-ranked based on grammatical accuracy scores for each of the candidate compressed sentences using a language model trained using linguistic features of words and/or phrases. The highest scoring candidate compressed sentence may be presented to the user.

Abstract: According to a data analysis method by which time required for data analysis is shortened and practicality is improved, a client terminal requests institution terminals to perform analysis of a matching attribute; each of the institution terminals encrypts an element belonging to the matching attribute within a database and sends the converted data to an outsource terminal; the outsource terminal integrates the plurality of converted data sent from the institution terminals and sends the integrated converted data to the institution terminals; and each of the institution terminals compares each matching attribute of a plurality of the elements within the database of the institution terminal against the integrated converted data, thereby identifying, as a common element, an element associated with the matching attribute and held in common by the institution terminals.

Abstract: A view can be presented with an augmented reality (AR) view of the space. The AR view can be augmented with imagery to indicate to the viewer environmental conditions that may not otherwise be known to the viewer. The viewer can also initiate alterations to the environment based on the information and recommendations presented in the AR view. Current conditions, past trends, and forecasted future trends can be included in the creation of the AR displays.

Abstract: A system and method including receiving a request to perform an operation relying on sets-related tables of a semantic layer universe; injecting in response to the received request, persisted Data Foundation (DF) objects stored in a dedicated data structure of a first set container into the in-memory representation of the semantic layer universe, each of the DF objects being automatically created based on sets-related tables of the semantic layer universe; injecting, by the processor and in response to the received request, persisted business layer (BL) objects stored in a dedicated data structure of the first set container into the in-memory representation of the semantic layer universe, each of the BL objects being automatically created based on the sets-related tables of the semantic layer universe; and executing the operation on the augmented semantic universe, including using the injected DF objects and the injected BL objects.

Abstract: Techniques are disclosed for a system that provides an intuitive way for merging or joining data from different datasets. The system may provide graphical interfaces to enable a user to combine or join datasets identified as having a relationship. In at least one embodiment, the system can determine options for joining datasets, such as by a left join, right join, or outer join. A graphical interface may display a visual representation (e.g., a “Glenn Diagram”) illustrate options for joining datasets based on identifying a relationship between the data sets. The representation may further illustrate one or more types of joins and information about the data, such as rows where data may be joined based on the type of join function for the relationship by columns. The visual representation may indicate where the datasets can be joined, such that they are not overlapping.

Abstract: Various embodiments of an end-to-end visual analytics system and related method thereof are disclosed.

Abstract: A system, method, and computer program product provides contextual contracts across various enterprise blockchain applications. In an aspect of the present disclosure, a method is disclosed including receiving a contextual contract for addition to a blockchain. The contextual contract includes at least one state dependent term. The method further includes appending a block to the blockchain based on the received contextual contract and in response to appending the block, activating a ledger sensor for the contextual contract. The ledger sensor is configured to monitor at least one of the blockchain and at least one other blockchain for transactions that correspond to the at least one term. The method further includes receiving from the ledger sensor data about transactions that corresponding to the at least one term and finalizing the contextual contract by appending a block to the blockchain based on the received data and the contextual contract.

Abstract: A method of resetting a stateful connection in a storage cluster with stateless load-balancing is provided. The method includes receiving a packet associated with a stateful connection between a client and a first storage node of the storage cluster. The method includes sending a reset packet from a second storage node to the client, through a stateless load balancer, responsive to the second storage node failing to recognize the packet associated with the stateful connection between the client and the first storage node, wherein the client resets the stateful connection responsive to receiving the reset packet. The method includes sending further reset packets from the load balancer to a plurality of storage nodes of the storage cluster, responsive to receiving, at the load balancer, the reset packet from the second storage node, wherein the plurality of storage nodes reset stateful connection information responsive to receiving the further reset packets.

Abstract: Methods, systems and computer readable media are provided for accessing data utilizing a multi-level table comprising generating a plurality of levels of the multi-level table, wherein a first level of the multi-level table includes a hyper-table with a plurality of hyper-cells each hyper-cell including information for a group of cells from an initial base table, wherein intermediate levels of the multi-level table each include a plurality of hyper-tables comprising hyper-cells with each hyper-table linked to and providing information for a corresponding hyper-cell of a hyper-table of a prior level, and wherein a plurality of tables of a terminal level includes information from cells of the initial base table with each table linked to and providing information for a corresponding hyper-cell. Data from the multi-level table is accessed by traversing links between the hyper-tables of the plurality of levels to access data within the tables of the terminal level.

Abstract: Methods, systems and computer readable media are provided for accessing data utilizing a multi-level table comprising generating a plurality of levels of the multi-level table, wherein a first level of the multi-level table includes a hyper-table with a plurality of hyper-cells each hyper-cell including information for a group of cells from an initial base table, wherein intermediate levels of the multi-level table each include a plurality of hyper-tables comprising hyper-cells with each hyper-table linked to and providing information for a corresponding hyper-cell of a hyper-table of a prior level, and wherein a plurality of tables of a terminal level includes information from cells of the initial base table with each table linked to and providing information for a corresponding hyper-cell. Data from the multi-level table is accessed by traversing links between the hyper-tables of the plurality of levels to access data within the tables of the terminal level.

Abstract: The method includes identifying a tree data structure. The method includes identifying one or more features in the identified tree data structure, wherein the one or more features comprise at least one of: a node of the tree data structure, an object of the tree data structure, an array of the tree data structure, an object property of the tree data structure, and a root of the tree data structure. The method includes determining whether one of the one or more identified features matches a feature that initiates execution of a rule, wherein the rule defines augmentations to the tree data structure based upon one or more features in the tree data structure. The method includes augmenting the identified tree data structure based upon the determined one or more matches of the one or more identified features and the feature that initiates execution of the rule.

Abstract: Systems and methods to extract a string from a communication. A method includes: receiving a communication comprising a plurality of strings; assigning a score to each of the strings, wherein the score assigned to each of the strings corresponds to a frequency of usage of the respective string for a first function relative to an overall frequency of usage of the respective string; determining a respective total sum for each of a plurality of sequences in the communication, the respective total sum determined as a sum of the scores for each string in the respective sequence; and extracting a first sequence of the sequences from the communication based on the total sum for the first sequence. In one embodiment, the total sum includes an additional score for each of a starting word and an ending word of the first word sequence, wherein each respective additional score is associated with a probability that the starting (or ending) word is used as the first (or last word) of an address.

Abstract: An example system may include instructions to control processor(s) to receive text from content of a first web page, determine, based on the content, a first title topic indicator, a first sentiment indicator, and a first text subjectivity indicator, apply the first title topic indicator, the first sentiment indicator, and the first text subjectivity indicator to a credibility machine learning model to generate a first content credibility score and a first content bias score for the text of the first web page, the credibility machine learning model being trained on text from other web pages using known title topic indicators, known sentiment indicators, and known text subjectivity indicators, and known credibility scores and bias scores, generate a first graphical representation for the first content credibility score and the first bias credibility score, and provide the graphical representation to a first digital device.

Abstract: Methods and systems for providing a visual content gallery within a controlled environment are disclosed herein. A content gallery server receives a content submission from an inmate device within the controlled environment. Further, the content gallery server determines that the content submission does not include prohibited content based on comparing the content submission to a blacklist of prohibited information. When the content submission does not include prohibited content, the content gallery server adds the content submission to a network accessible content gallery corresponding to an inmate associated with the inmate device. Further, authorized friends and family of the inmate may view the content submission and provide comments on the content submission.

Abstract: A method for displaying landmark data from a search of a place name keyword, the method includes: inputting the place name keyword to a server to search for a plurality of landmark data, wherein each of the landmark data comprises fields of a landmark name, an objective level category, an address, and an address quoting frequency; sorting the landmark data by an electronic device to a display order, based on a characterized parameter for each of the landmark, wherein the characterized parameter is calculated based on at least a publicity, wherein the publicity is a calculation of the objective level category and the address quoting frequency with respectively weighting to the objective level category and the address quoting frequency; and displaying the landmark data by the electronic device according to the display order.

Abstract: One embodiment provides a method comprising receiving general private data identifying at least one type of privacy-sensitive data to protect, collecting at least one type of real-time data, and determining an inference privacy risk level associated with transmitting the at least one type of real-time data to a second device. The inference privacy risk level indicates a degree of risk of inferring the general private data from transmitting the at least one type of real-time data. The method further comprises distorting at least a portion of the at least one type of real-time data based on the inference privacy risk level before transmitting the at least one type of real-time data to the second device.

Abstract: In an example embodiment, factual question generation from freeform content is achieved through semantic role labeling and recurrent neural networks (RNNs). Specifically, semantic role labeling is used to identify an answer phrase so that it can be replaced with an appropriate question word. RNNs are then used to extract triples (Subject-Object-Predicate) from the sentence, and each of these triples can be used as an answer phrase/word. An RNN is then fed with training data to generate the questions more efficiently.

Abstract: An apparatus acquires a first input-table, based on target table information, and acquires a script for generating a first output-table from the first input-table, and a constraint condition between columns included in the first input-table and the first output-table. The apparatus converts the first input-table into a second input-table such that the second input-table obtained by converting a column structure of the first input-table and a second output-table to be obtained by converting a column structure of the target table comply with the constraint condition, and generates assignment information indicating a correspondence relationship between columns to be included in the second output-table and the target table. The apparatus generates the second output-table from the script and the second input-table, and converts the second output-table into the target table, based on the assignment information.

Abstract: Systems and associated methods are described for providing content recommendations. The system selects a first plurality of subsets of content categories, each subset of content categories comprising a first number of content categories. The subsets are assigned reward scores based on content popularity and duplication. The subset are then iteratively modified to increase the rewards scores. If the reward scores are still low, the process is repeated by selecting a second plurality of subsets of content categories, each subset of content categories comprising a second number of content categories, different from first number.

Abstract: Systems and methods for a network environment for client-side remote access of a server device from a client device may utilize a biometric sensor device of the client device and a pluggable authentication and authorization framework. The biometric sensor device may capture a gesture of a target user. The server device may authenticate the target user based on previously registered encrypted biometric information of the target user utilizing the pluggable authentication and authorization framework and a remote desktop protocol. When the target user has been authenticated, the client device may be authorized to access a service of the server device.

Abstract: A non-transitory computer readable medium storing a program causing a computer to execute a process for information processing, the process including: obtaining input data that represents attributes of each subject among subjects; generating a bipartite network which includes the subjects and the attributes included in the input data as nodes and in which a node that corresponds to each subject among the subjects are connected to nodes that correspond to the attributes of the subject via links; and performing clustering of a group of the nodes by performing iterative calculations of a stochastic process of transitions between the nodes via the links in the bipartite network.

Abstract: Systems, methods, and computer media for securing data accessible through software applications are provided herein. By capturing path data such as returned results for a query and displayed results provided by an application (e.g., to or by a web browser) for an operation, it can be determined if the query returned more data than was needed for what was displayed. The query can be refined to limit the data returned and reduce the security risk of such over-provisioning of data.

Abstract: The present disclosure is directed to training, and providing recommendations via, a machine learning model architected to balance relevance and diversity of sets of recommendations. For example, a neural network can be provided with user profile features and can output probabilities for each of a number of recommendations. This can be converted into a ranked list of recommendations. The ranked list of recommendations is provided to a diversity model that maximizes an optimization objective having a first objective that quantifies relevance of a recommendation and a second objective that measures diversity of a set of recommendations. The output of the diversity model is a set of recommendations that have both high relevance and high diversity.

Abstract: Systems, methods, and computer-readable media are disclosed for dynamically determining highlights of media content based on user interaction metrics and/or social media metrics. In one embodiment, an example method may include determining media content streamed to user devices over a time period and corresponding to a live event, determining user interaction data associated with the media content and indicative of user interactions with the user devices, determining, using the user interaction data, a quality score for a time interval during the time period, the quality score indicative of user engagement in a segment of the media content, determining, using the time interval and segment-by-segment metadata, a stream start time and a stream end time for the segment, generating, using the stream start time and the stream end time, a clip of the segment, and determining, using the quality score, that the clip is to be included in a highlight.

Abstract: Disclosed is an electronic system for amending and studying codes of regulation, including an Internet-connected electronic device receiving amendments of texts from a server. The electronic device includes a display with a timeline highlighting the text modification dates, labeling the successful reading or the failure thereof with symbols and/or colors. The electronic device includes an electronic pen allowing supplementing the text with highlights, remarks, notes, external links, images, etc. The electronic pen includes the following buttons: “mode”, causing the reversible switching from a “highlighter” mode, to a “pen” mode, to an “eraser” mode; “timeline” to compare the different versions of the laws; “users” to view and share comments on the legislative texts with other users; “writing” to modify the writing settings and adapt the space available for the user for the remarks; and a tip to write with precision on the law texts and loaded documents.

Abstract: A data transformation system includes a schema inference module and an export module. The schema inference module is configured to dynamically create a cumulative schema for objects retrieved from a first data source. Each of the retrieved objects includes (i) data and (ii) metadata describing the data. Dynamically creating the cumulative schema includes, for each object of the retrieved objects, (i) inferring a schema from the object and (ii) selectively updating the cumulative schema to describe the object according to the inferred schema. The export module is configured to output the data of the retrieved objects to a data destination system according to the cumulative schema.

Abstract: A search apparatus includes a character string acquisition unit that acquires a character string recognized by a character recognition device, a search unit that sends to an information source the character sting as a search key and retrieves search result information responsive to the search key out of information stored on the information source, a determining unit that determines whether an error factor including at least one of a cause for an erroneous search and an erroneous search result is present, and a job switching unit that requests a manual input job to be performed to input a search key that is to replace the character string if the determining unit has determined that the error factor is present, and requests a verification job to be performed on the search result information if the determining unit has determined that the error factor is not present.

Abstract: Systems and method for annotating variants within a genome can call variants from reads or receive called variants directly and associate the called variants with functional annotations and interpretive annotations. A summary report of the called variants, the associated functional annotations, and the associated interpretive annotations can be generated.

Abstract: Methods for querying a database and database systems include optimizing a database query for parallel execution using spatial and temporal information relating to elements in the database, the optimized database query being split into sub-queries with sub-queries being divided spatially according to host and temporally according to time window. The sub-queries are executed in parallel. The results of the database query are outputted progressively.

Abstract: A mechanism is provided to implement a summarization mechanism for summarizing an identified natural language document using hyperbolic embeddings. Responsive to receiving a query from a user for a summarization of the identified natural language document, the summarization mechanism produces a hyperbolic embedding model of embeddings of the query. The summarization mechanism compares the embeddings of the query to each of a set of embeddings associated with a set of sentences of the identified natural language document. Responsive to identifying a subset of embeddings associated with the set of sentences of the identified natural language document having a semantic specificity to a subset of embeddings associated with the query, the summarization mechanism adds the sentence to a summary of the identified natural language document. The summarization mechanism then outputs the summary to the user.