Methods, apparatus, systems, and computer-readable media are provided for providing context specific schema files that allow an automated assistant to broker human-to-computer dialogs between a user and an application that is separate from the automated assistant. The context specific schema file can provide the automated assistant with sufficient data to be responsive to user queries without necessarily communicating with a remote device, such as a server. Multiple different context specific schema files can be made available to the automated assistant according to a context in which a user is interacting with the automated assistant. In this way, latency otherwise exhibited by the automated assistant can be mitigated by providing the automated assistant with the information needed to respond to a user without continually retrieving the information over a network.

Provided are a question answering method and language model training method, apparatus, device, and storage media, including: acquiring at least one candidate table matching a question to be queried, each candidate table includes a candidate answer corresponding to the question; processing the at least one candidate table to obtain at least one table text, the table text includes textual content of respective fields in the candidate table; inputting the question and each table text into a preset language model respectively to obtain a degree of matching between the question and each candidate table; and outputting a reply table according to the degree of matching of each candidate table, the reply table is a candidate table out of the at least one candidate table whose degree of matching with the question is greater than a preset value or a candidate table that corresponds to a maximum degree of matching.

In general, embodiments relate to a method for managing a technical support session, comprising: determining a technical support issue (TSI) for a technical support session; identifying a question path graph (QPG) associated with the TSI; and displaying at least a portion of the QPG to a technical support person (TSP) during the technical support session.

Techniques for mapping policy documents to regulatory documents to check for compliance between the policies and documents are provided. In one example, a computer-implemented method determining, by a system operatively coupled to a processor, an information input, a control framework, and a document from a first group consisting of a regulatory document and a policy document, wherein the information input is a corpora from a second group consisting of a domain corpora and a global corpora. The computer-implemented method can also comprise mapping, by the system, the received regulatory document or the received policy document to the control framework using a supervised machine learning technique.

Methods, systems and computer-program products are directed to a Privacy Engine for evaluating initial electronic documents to identify document content categories for portions of content within the electronic documents, with respect to extracted document structures and document positions, that may include privacy information for possible redaction via visual modification. The Privacy Engine builds a content profile based on detecting information at respective portions of electronic document content that indicate one or more pre-defined categories and/or sub-categories. For each respective portion of electronic document content, the Privacy Engine applies a machine learning model that corresponds with the indicated category (or categories and sub-categories) to determine a probability value of whether the respective portion of content includes data considered likely to be privacy information. The Privacy Engine recreates the one or more initial electronic documents according to one or more privacy information redactions at respective locations of the portions of content.

A method of generating a question-answer learning model through adversarial learning may include: sampling a latent variable based on constraints in an input passage; generating an answer based on the latent variable; generating a question based on the answer; and machine-learning the question-answer learning model using a dataset of the generated question and answer, wherein the constraints are controlled so that the latent variable is present in a data manifold while increasing a loss of the question-answer learning model.

Documents from a set of related documents in a domain are processed to identify keywords associated with each document. The documents are then further processed to identify the documents that are the most similar to each other. For each document, some or all of the keywords that are associated with the similar documents, but not the document itself, are selected as semantic tags for the document. These semantic tags determined for a document represent novel or hidden concepts and contexts that may relate to the document, but that do not actually appear in the document. The documents are used to train a model that generate semantic tags for a document or for keywords associated with the document. The generated model can then be used for a variety of purposes such the creation of an index for a set of documents or for query expansion.

There is a need for more effective and efficient cross-dataset field integration. In one example, a method comprises determining a primary integration feature vector for a primary dataset field; for each secondary dataset field of a plurality of secondary dataset fields, determining a secondary integration feature vector; determining, based at least in part on the primary integration feature vector and each secondary integration feature vector, an integration space; determining, for each secondary dataset field of the plurality of secondary dataset fields and based at least in part on the integration space, a distance measure between the primary dataset field and the secondary dataset field; determining, based at least in part on each distance measure between the primary dataset field and a secondary dataset field of the plurality of secondary dataset fields, a predefined number of the plurality of secondary dataset fields; and performing the cross-dataset field integration based at least in part on the a predefined number of the plurality of secondary dataset fields.

Techniques are disclosed relating to probabilistic data structures. A database node may maintaining a probabilistic data structure capable of encoding database keys. The probabilistic data structure may include a plurality of levels that are each capable of storing an indication of a transition between successive characters in a database key. The database node may insert a particular database key into the probabilistic data structure and the particular database key may comprise a series of characters. The inserting may include setting, for each transition between successive characters of the series of characters, an indication in a corresponding level of the plurality of levels that is indicative of that transition. The database node may further maintain lineage information specifying one or more lineages that correspond to the transition.

An information retrieval method includes obtaining Mi (i+1).sup.th-hop candidate documents based on a retrieval text query and Ki i.sup.th-hop candidate documents; obtaining a score of each candidate document in the Mi (i+1).sup.th-hop candidate documents; obtaining, based on a score of a candidate document Pjy(i+1) and a probability of a path L, a probability of a path corresponding to the candidate document Pjy(i+1); obtaining K(i+1) (i+1).sup.th-hop candidate documents based on probabilities of paths respectively corresponding to the Mi (i+1).sup.th-hop candidate documents; and obtaining, based on the K(i+1) (i+1).sup.th-hop candidate documents, a retrieval result corresponding to the query.