An information processing apparatus includes a processor. The processor receives an input of a graph structure. The graph structure has nodes including text and edge. The processor assigns the nodes to one or more clusters. The processor partitions the text into words. The processor classifies the words into 1) a word representing a subject or target of an operation, 2) a word representing a content or state of the operation, and 3) other words. The processor extracts a frequent word by counting a frequency of occurrence of one or more words classified as the words representing the subject or target of the operation and extracts a frequent word by counting a frequency of occurrence of one or more words classified as the words representing the content or state of the operation, for the respective clusters.

Technology is described herein for generating questions using a neural network. The technology generates the questions in a three-step process. In the first step, the technology selects, using a first neural network, a subset of textual passages from an identified electronic document. In the second step, the technology generates, using a second neural network, one or more candidate answers for each textual passage selected by the first neural network, to produce a plurality of candidate passage-answer pairs. In the third step, the technology selects, using a third neural network, a subset of the plurality of candidate passage-answer pairs. The technology then generates an output result that includes one or more output questions chosen from the candidate passage-answer pairs produced by the third neural network. The use of the first neural network reduces the processing burden placed on the second and third neural networks. It also reduces latency.

A computing device may represent dialog for output by a chatbot as a response graph. The graph may consist of nodes and edges, both of which may have attributes. The graph may be linked with another graph. There may be one or more traversal paths through the graph. The computing device may represent each traversal path as a row in a database table. Each column in the table row may correspond to: one or more nodes in the traversal path, a condition value, weight, and/or other filtering condition. A computing device may display a graphical user interface that allows a user to add, edit, and/or delete nodes and/or edges of the graph. The user interface may also allow the user to compare two or more graphs. The user interface may generate visualizations of traversal paths of a response graph.

Methods, systems, and apparatus, including computer programs encoded on computer storage media, for responding to a query. In some implementations, a computer obtains a query. The computer determines a meaning for each term in the query. The computer determines user data for the user that submitted the query. The computer identifies one or more ontologies based on the meanings for at least some of the terms. The computer identifies a knowledge graph based on the identified ontologies and the user data. The computer generates a response to the query by traversing a path of the identified knowledge graph to identify items in the knowledge graph based on the determined meaning for each of the terms. The computer generates path data that represents the path taken by the computer through the identified knowledge graph. The computer provides the generated response and the path data to the client device.

Methods, systems, and computer program products for accelerating inference of transformer-based models are provided herein. A computer-implemented method includes obtaining a machine learning model comprising a plurality of transformer blocks, a task, and a natural language dataset; generating a compressed version of the machine learning model based on the task and the natural language dataset, wherein the generating comprises: obtaining at least one set of tokens, wherein each token in the set corresponds to one of the items in the natural language dataset, identifying and removing one or more redundant output activations of different ones of the plurality of transformer blocks for the at least one set of tokens, and adding one or more input activations corresponding to the one or more removed output activations into the machine learning model at subsequent ones of the plurality of the transformer blocks; and outputting the compressed version of the machine learning model to at least one user.

Provided is a device including a memory storing information about sequences of a plurality of registered words; an input unit comprising input circuitry configured to receive an input of a text comprising a first eojeol not belonging to the plurality of registered words, wherein, in the first eojeol, a first word is attached to a first registered word that belongs to the plurality of registered words; and a controller configured to detect the first registered word from the first eojeol, to determine a predicted eojeol to be input after the text, based on the information about the sequences of the plurality of registered words and the detected first registered word and to control a display to display the predicted eojeol.

Provided is a device including a memory storing information about sequences of a plurality of registered words; an input unit comprising input circuitry configured to receive an input of a text comprising a first eojeol not belonging to the plurality of registered words, wherein, in the first eojeol, a first word is attached to a first registered word that belongs to the plurality of registered words; and a controller configured to detect the first registered word from the first eojeol, to determine a predicted eojeol to be input after the text, based on the information about the sequences of the plurality of registered words and the detected first registered word and to control a display to display the predicted eojeol.

A computing system identifies an incoming voice call from a user device to an agent device associated with the computing system. The computing system generates a transcription of the incoming voice call using one or more natural language processing techniques. The computing system extracts a problem description from the transcription. The problem description indicates a topic for the incoming voice call. A first machine learning model estimates a situation vector from the problem description. A second machine learning model identifies a pre-existing situation vector that closely matches the estimated situation vector. The computing system retrieves a situation description that corresponds to the identified pre-existing situation vector.

An electronic document is provided for presentation via a graphical user interface (GUI). A first region of the electronic document includes content objects. A second region of the electronic document includes a drawing object. A determination is made, based on first characteristics associated with the content objects and second characteristics associated with the drawing object, that the drawing object corresponds to the content objects. A mapping is generated between the content objects and the drawing object. A modification to the content objects in the first region of the electronic document is identified. In response to the modification to the content objects, a presentation of the drawing object in the second region of the electronic document is modified in view of the generated mapping.

Conversational artificial intelligence techniques with live agent engagement based on automated frustration level monitoring are disclosed. For example, a method comprises obtaining, via a conversational artificial intelligence system, a frustration level metric associated with a user participating in a conversation with the conversational artificial intelligence system, The method further comprises managing, via the conversational artificial intelligence system, human agent engagement in the conversation based on the frustration level metric.