Abstract: Apparatuses, arrangements and methods therein for generation of comfort noise are disclosed. In short, the solution relates to exploiting the spatial coherence of multiple input audio channels in order to generate high quality multi channel comfort noise.

Abstract: Methods and systems for generating representative data. A generator is configured to create, using a learning model, one or more generated records based on a plurality of training records obtained from a sensitive database. A discriminator is trained to identify the generated records as being generated based on the training records and a privacy adversary is trained to identify a training sample as being more similar to a distribution of the generated records than a distribution of the reference records.

Abstract: A computerized method for extracting domain specific insights from a corpus of files containing large documents comprising: breaking down large chunks of text into smaller sentences/short paragraphs in a domain specific way, identifying and removing domain noise; identifying the sentence intents of the non-noise sentences; tagging the sentences with other domain specific attributes; defining a semantic ontology using a graph database based on the sentence intents, a multitude of mini-dictionaries and domain attributes; applying a pre-defined ontology to tag documents with domain specific hashtags; and combining the hashtags using machine learning techniques into insights.

Abstract: A created sentence evaluating device 1 using a neural network unit 10 of an encoder/decoder model in which an encoder unit 100 inputs a sentence in a first language, and a decoder unit 101 sequentially outputs word candidates for a sentence in a second language corresponding to the sentence in the first language and likelihood of the word candidates includes: an encoder input unit 13 configured to input a created sentence created in the second language to the encoder unit 100 sequentially for each word; and an evaluation unit 17 configured to evaluate words of the created sentence on the basis of word candidates in the second language and the likelihood of the word candidates output by the decoder unit 101 on the basis of an input from the encoder input unit 13.

Abstract: A method for processing text data includes analyzing the text data to identify a plurality of keywords. The method also includes determining whether each of the plurality of keywords already exists in one or more databases. When a keyword in the plurality of keywords is not found in the one or more databases, the method includes tagging the keyword with a plurality of characters for storage. The plurality of characters includes at least a first character to indicate a start of the tagging, a second character to indicate a corresponding database for storing the keyword, and a third character to indicate an end of the tagging. The method also includes storing the tagged keyword in the corresponding database.

Abstract: Systems and processes for operating an intelligent automated assistant are provided. An example process includes receiving an utterance including a user request, determining, based on the user request, a domain associated with the user request, determining, based on the domain, a first subsequent user action and a second subsequent user action, determining, based on the domain, a first parameter for the first subsequent user action and a second parameter for the second subsequent user action, in accordance with a determination that a first score associated with the first subsequent user action is higher than a score associated with the second subsequent user action, selecting the first subsequent user action as a suggested subsequent user action, and providing the suggested subsequent user action.

Abstract: This disclosure provides systems, methods and apparatuses for converting natural language text in financial documents to mathematical operations for a financial management program. In some implementations, a computing device may obtain a first segment of computer readable text associated with a text capture in a financial document, parse only a portion of the first segment to generate a parsed segment, and determine translation information associated with the parsed segment to translate the parsed segment. The computing device may also store the parsed segment and the translation information in a data structure in memory, translate the parsed segment in the data structure based on the translation information, and output one or more computer readable mathematical operations corresponding to the translated segment. In this manner, the computing device may output the one or more computer readable mathematical operations without parsing all of the segment.

Abstract: An interactive control system according to an embodiment includes one or more hardware processors configured to function as a detection unit, a conversion unit, first and second estimation units, and an attribute determination unit. The detection unit detects an input keyword from input information. The conversion unit converts the input information into information including the input keyword and the input-keyword-excluded context information. The first estimation unit estimates, from the context information, at least a first estimate information piece in which the input keyword's first attribute is associated with a first likelihood of the first attribute. The second estimation unit estimates, from the input keyword, at least a second estimate information piece in which the input keyword's second attribute is associated with a second likelihood of the second attribute. The attribute determination unit determines the input keyword's attribute based on the first and second estimate information.

Abstract: A man-machine conversation method includes: extracting a keyword from input information of a user; determining a candidate article according to the keyword; obtaining a plurality of sentences in the candidate article, and determining a score of each of the plurality of sentences; extracting part of the plurality of sentences based on the score of each of the plurality of sentences; and generating return information that includes the extracted part of the plurality of sentences.

Abstract: Systems and methods for a boosted Latent Dirichlet Allocation model with predefined clusters and repelling force. Herein, clusters are pre-populated with seed words with amplified weights to predefine the characteristics of clusters. A repel force is also applied while assigning words to specific clusters in order to assign words more accurately to clusters in instances where clusters have similar characteristics. In addition to generating improved clustering of words, systems and methods may analyze words within queries to provide explainability output related to the performance of content or data.

Abstract: This disclosure describes, in part, techniques for performing multi-path calculations for energy levels on an electronic device. For instance, the electronic device may include a first circuit and a second circuit, where the first circuit uses less power than the second circuit. As such, when operating in a standby mode, the electronic device may use the first circuit to calculate energy levels at the electronic device, such as speech-energy values and ambient-energy values. Additionally, while operating in an active mode, the electronic device may active the second circuit and then use the second circuit to calculate the energy levels at the electronic device. The first circuit and the second circuit can send/receive current energy levels between one another so that the electronic device can continually calculate the energy levels even when the electronic device switches between modes of operation.

Abstract: Methods and systems for efficiently generating tagged training data for machine learning models. In conventional systems, all of the raw data (e.g., each sentence) has to be manually tagged. Instead, the methods and systems generate a representative sample for multiple portions of raw data, e.g., a representative sentence for multiple, similar sentences. Only the representative sample is tagged and used for training, thereby realizing a significant efficiency in both tagging the data and training the machine learning models.

Abstract: Disclosed is a method for performing fine-grained text sentiment analysis based on the users' harshness. According to the method, product attributes are automatically extracted according to comments in a data set, and the comments are split according to the product attributes, so as to achieve the effect of fine-grained analysis. Then, an inference model, which takes users' harshness into consideration, is used, and the inference model takes the users' harshness as hidden variables and solves same by means of an expectation-maximization (EM) algorithm, so as to obtain evaluation results of several product attributes of each product, thereby further improving the accuracy of sentiment analysis. According to the method, the problems in sentiment analysis of analysis granularity being too large and the users' harshness differing are solved, such that high-quality sentiment analysis can be performed on product comments.

Abstract: A document processing system is configured to identify, for each accessed electronic document in a first set of multiple electronic documents, a set of identified multi-word phrases determined to be in ordered text information in the accessed electronic document, each multi-word phrase of the set of identified multi-word phrases including adjacent words in the ordered text information; and determine, for each accessed electronic document in the first set of multiple electronic documents, a selected document type from the first set of document types based at least on an analysis of the set of identified multi-word phrases with respect to multi-word-phrase characteristics identified by a first definition and associated with each document type in a first set of document types associated with a first document-set type.

Abstract: Devices and techniques are generally described for determining an information content score for a natural language input. In various examples, data representing a natural language input may be received. A first probability of the natural language input may be determined by comparing words of the natural language input to a first data structure representing a plurality of natural language phrases. A score for the natural language input may be determined based at least in part on the first probability. In some examples, the score may represent an amount of information content included in the natural language input. In various examples, the natural language input may be sent to a remote computing device based at least in part on the score.

Abstract: The invention relates to a method and a system for improving performance of text summarization and has an object of improving performance of a technique for generating a summary from a given paragraph. According to the invention to achieve the object, a method for improving performance of text summarization includes: an a step of generating an embedding vector by vectorizing a natural language-based context; a b step of generating a graph by using the embedding vector; a c step of assigning a weight depending on whether or not a keyword corresponding to at least one node included in the graph is present in the context; and a d step of selecting a path having a highest likelihood in the graph and generating a summary based on the path.

Abstract: Disclosed is a method for performing QA with multi-modal information. Specifically, according to the present disclosure, a computing device determines core text information from a video based on question data, determines core object information or core frame information from the video based on the core text information, and performs QA for the video based on the determined core text information, and the determined core object information or core frame information by utilizing a QA model.

Abstract: A server accesses a natural language query corresponding to one of a plurality of natural languages. The server maps, using a query-to-vector engine configured to leverage word embeddings in each of the plurality of natural languages to map natural language queries in the plurality of natural languages to vectors corresponding to meanings of the natural language queries, the natural language query to a vector. The server matches the vector to an intent representing a prediction associated with the natural language query. The server provides a response to the natural language query based on the intent.

Abstract: A method for securing electronic transactions includes associating a mobile electronic device with a first user. A first computer system retrievably stores registration data relating to the first user, including a device identifier that is unique to the mobile electronic device. A security application that supports in-application push notifications is installed on the mobile electronic device. The first computer system sends a push notification to the mobile electronic device, the push notification prompting the first user to provide a confirmation reply via a user interface of the security application for activating the mobile electronic device as a security token. The mobile electronic device is activated as a security token for the first user in response to receiving at the first computer system, from the mobile electronic device, the confirmation reply from the first user.

Abstract: In one example, a system can receive a set of text samples and generate a set of summaries based on the set of text samples. The system can then generate a training dataset by iteratively executing a training-sample generation process. Each iteration can involve selecting multiple text samples from the set of text samples, combining the multiple text samples together into a training sample, determining a text category and a summary corresponding to a selected one of the multiple text samples, and including the text category and the summary in the training sample. After generating the training dataset, the system can use it to train a model. The trained model can then receive a target textual dataset and a target category as input, identify a portion of the target textual dataset corresponding to the target category, and generate a summarization of the portion of that target textual dataset.