Abstract: A method for providing unsupervised entity resolution to a natural language processing system includes receiving a named entity for training from the natural language processing system, searching a corpus for a first undisambiguated named entity corresponding to the named entity, identifying a plurality of disambiguated named entities corresponding to the first undisambiguated named entity, identifying a plurality of aliases for each of the disambiguated named entities, training a classifier for the each of the disambiguated named entities utilizing the aliases identified for respective ones of the disambiguated named entities using the corpus, and resolving the named entity using the classifier, wherein resolving the named entity comprises selecting one of the disambiguated named entities from among the disambiguated named entities and returning, automatically, the selected disambiguated named entity to the natural language processing system.

Abstract: The present invention concerns a text analysis system, the text analysis system being adapted for utilizing a topic model to provide a document representation. The topic model is based on learning performed on a text corpus utilizing hidden layer representations associated to words of the text corpus, wherein each hidden layer representation pertains to a specific word of the text corpus and is based on a word environment including words occurring before and after the specific word in a text of the text corpus.

Abstract: Implementations of this disclosure provide methods and apparatuses for generating recommendation information. An example method includes matching text content from a text content library based on a plurality of predetermined scenario-related words; extracting keywords from the related text content, to generate a plurality of training samples; and for each training sample, providing a source sequence of a sequence pair corresponding to the training sample as an input to a recommendation information generation model, obtaining, from the recommendation information generation model, a predicted word, and adjusting a model parameter of the recommendation information generation model based on a comparison between the predicted word and a corresponding word in a target sequence of the sequence pair corresponding to the training sample, to train the recommendation information generation model.

Abstract: A non-transitory computer-readable storage medium storing a program that causes a computer to execute a process, the process includes learning distributed representations of words included in a word space of a first language using a learner for learning the distributed representations; classifying words included in a word space of a second language different from the first language into words common to words included in the word space of the first language and words not common to words included in the word space of the first language; and replacing distributed representations of the common words included in the word space of the second language with distributed representations of the words, corresponding to the common words, in the first language and adjusting a parameter of the learner.

Abstract: Regulating velocity of chat discourse can include determining, using computer hardware, a topic and emotive content from a plurality of chat messages of a group chat, determining, using the computer hardware, time deltas between different ones of the plurality of chat messages, determining, using the computer hardware, a tempo and a tempo adjustment for the group chat based on the topic, the emotive content, and the time deltas, and indicating, using the computer hardware, the tempo adjustment to a client device of a participant in the group chat.

Abstract: An example system includes a processor to receive a prefix of conversation and a text input. The processor is to also generate a completed response based on the prefix of conversation and the text input via a trained primal network. The primal network is trained to minimize a Lagrangian loss function representing a number of objectives and a dual network is trained to maximize the Lagrangian loss function.

Abstract: In an approach to generating natural language rules based on detected code snippets, one or more computer processors detect a code snippet. The one or more computer processors extract code information from the detected code snippet. The one or more computer processors feed the extracted code information into a cognitive model, wherein the cognitive model utilizes one or more historical code snippets based on the extracted code information and one or more natural language rules based on the extracted code information. The one or more computer processors generate, based on one or more calculations by the cognitive model, a natural language rule for the code snippet.

Abstract: A system configured to predict fine-grained affective states. The system comprising a processor configured to execute instructions to create training data comprising content conveying emotions, and to create a trained model by performing an emotion vector space model training process using the training data to train a model using a feed forward neural network that converts discrete emotions into emotion vector representations. The system uses the trained model to predict fine-grained affective states for text conveying an emotion.

Abstract: Provided is a system and method for retrieving a translation of a source word based on context. For example, the context may include other words in a same file as the source word. The context can be used to identify the correct semantic meaning of the source word when the word has multiple contextual meanings. In one example, the method may include identifying a source word from a data file, determining a plurality of translation candidates for the source word which translate the source word from a source language into a different language, retrieving a target translation candidate for the source word from among the plurality of translation candidates based on context of the source word included in the data file, and outputting the translated target translation candidate for display via a display device.

Abstract: Provided is a method for detecting deceptive e-commerce reviews based on a sentiment-topic joint probability, which belongs to the fields of natural language processing, data mining and machine learning. In the data of different fields, a STM model is superior to other reference models; compared with other models, the STM model belongs to a completely un-supervised (no label information) statistic learning method and shows great advantages in processing unbalanced large sample dataset. Thus, the STM model is more suitable for application in a real e-commerce environment.

Abstract: Provided is a processor-implemented method of generating a natural language, the method including generating a latent variable from an embedding vector that corresponds to an input utterance, determining attention information related to the input utterance by applying the generated latent variable to a neural network model, and outputting a natural language response that corresponds to the input utterance based on the calculated attention information.

Abstract: A system trains a classification model. Text windows are defined from tokens based on a window size. A network model including a transformer network is trained with the text windows to define classification information. A first accuracy value is computed. (A) The window size is reduced using a predefined reduction factor value. (B) Second text windows are defined based on the reduced window size. (C) Retrain the network model with the second text windows to define classification information. (D) A second accuracy value is computed. (E) An accuracy reduction value is computed from the second accuracy value relative to the first accuracy value. When the computed accuracy reduction value is ?an accuracy reduction tolerance value, repeat (A)-(E) until the accuracy reduction value is <the accuracy reduction tolerance value. Otherwise, increase the window size, define final text windows based on the increased window size, and retrain the network model.

Abstract: In a computer system, systems and methods for automatically answering natural language questions using deep semantics are provided. Methods include receiving a natural language question, mapping it into one or more deductive database queries that captures one or more intents behind the question, computing one or more result sets of the question using one or more deductive database queries and a deductive database and providing one or more result sets. Systems include natural language question compilers and deductive databases. The natural language question compiler is configured to receive a natural language question and map it into one or more deductive database queries that capture one or more intents behind the question. The deductive database is configured to receive the mapped one or more deductive database queries, compute one or more result sets of the question using the one or more deductive database queries, and provide one or more result sets.

Abstract: A method and apparatus for mining general text content, a server, and a storage medium, are disclosed. A specific embodiment of the method can include: acquiring a question including a target subject and a target characteristic; and inputting the target subject, the target characteristic and a target text into a pre-constructed answer prediction model, and determining a starting position and an ending position of an answer to the question in the target text by the answer prediction model. The answer prediction model is pre-trained based on a sample question including a sample subject and a sample characteristic, and a starting position and an ending position of a sample answer in a text. In the technical solution provided by the embodiments of the present disclosure, the starting position and the ending position of the answer in the target text may be correctly predicted, thereby increasing the accuracy of answer recognition.

Abstract: The present disclosure relates generally to providing an intent-driven contact center. The contact center according to some embodiments analyzes intents to determine to which device or agent to route a communication. The analyzed intent information can also be used to formulate reports and analyze the accuracy of the identified intents with respect to the received communication.

Abstract: An information display device is an information display device for allowing content to be output in a plurality of languages, and includes a management information acquisition unit, a priority determination unit, and an image generator. The management information acquisition unit acquires management information including language information including information indicating a language used by sojourners in an area where the information display device is used, and sojourn information including information indicating a sojourn situation of the sojourners in the area. Based on the acquired management information, the priority determination unit determines priority of the languages used in the information display device. The image generator generates a display image in accordance with the priority.

Abstract: An intelligent system and method for analyzing documents and suggesting corrections based on diversity criteria include a processing device to analyze a job document, using a machine learning model, to identify a first expression representing a first qualification requirement favorable to a first class of applicants than a second class of applicants according to a diversity metric, responsive to identifying the first expression, determine, using a semantic relation map, a second expression representing a second qualification requirement that is less favorable to the first class of applicants when compared to the first expression, and responsive to determining that the second expression, present the second expression on the interface device as a suggested replacement to the first expression in the job document.

Abstract: A computer-implemented method for fine-grained affective states prediction. The computer-implemented method creates training data comprising content conveying emotions. The method creates a trained model by performing an emotion vector space model training process using the training data to train a model using a feed forward neural network that converts discrete emotions into emotion vector representations. The trained model can be used to predict fine-grained affective states for text conveying an emotion.

Abstract: Systems and methods for domain classification in natural language processing based on domains are disclosed. The method includes generating a trigram corpus for the purpose of classification based on a trigram analysis of a domain model containing a hierarchical ontology and semantic construction that maps patterns of semantic tokens to syntactic patterns. An input string is parsed within each domain, tokenized in each domain. The resulting trigrams for the input text in each domain are looked up in the corresponding trigram corpus to determine the relevancy of each domain to the input text. The input string is thus classified based on the relevancy determination. The systems and methods avoids having to rely on existing annotated domain corpora for classification and allows for fast regeneration of the classifier when domain models are under frequent update and development.

Abstract: Systems for selective data capture and translation are provided. In some examples, a system, may receive data from one or more systems, networks, applications, devices, or the like. The data may include data associated with one or more issues occurring at the system, network, application, device, or the like. In some examples, a plurality of data containers may be generated. In some arrangements, each data container may be associated with a different issue, type of issue, system, application, or the like. The data containers may be generated in response to receiving data associated with an issue or may be pre-generated. In some arrangements, the received data may be evaluated (e.g., using machine learning) to determine whether it should be added to one or more data containers of the plurality of data containers. If so, the data may be added and, if not the data may be preserved and/or further evaluated to determine whether it should be added to a different data container.