Abstract: A method, system, and computer program product for implementing automated cognitive software application error detection is provided. The method includes receiving data associated with model based self-learning software code. The annotated data is automatically divided with respect to specified categorization and grouping attributes and categorized groups comprising portions of the annotated data are generated and analyzed. At least one incorrect annotation associated a group of the categorized groups is detected and filtered. Likewise, a correct annotation for the group is detected and retrieved from a database. The correct annotation is appended to the group.

Abstract: An approach is provided for improving a named entity recognition. Using a multi-label classification in a neural network, a sub-entity is identified in an original sentence. First and second labels are determined indicating first and second candidate types of the sub-entity. First and second replacement sentences are generated. The first replacement sentence replaces the sub-entity in the original sentence with a first sub-entity of the first candidate type. The second replacement sentence replaces the sub-entity in the original sentence with a second sub-entity of the second candidate type. First and second confidence scores for the first and second replacement sentences are determined. Based on the first confidence score exceeding the second confidence score by more than a threshold amount, the neural network is retrained by selecting the first instead of the second candidate type as the sub-entity type.

Abstract: A data-driven model compression technique is introduced that only targets to provide same accuracy as the original (not compressed) model in certain areas by reducing compression parameters. A compression engine relies on backpropagation to determine an extent of parameter value changes and designate certain parameters as key parameters. The model matrix is reshaped according to importance of each neuron. Only randomly generated parameter values of the reshaped parameter matrix are fine tuned to create a reliable compressed neural network model.

Abstract: Label-based document classification using artificial intelligence includes collecting, by one or more processors, a plurality of pre-trained classification models into a model pool and a plurality of documents into a document pool. The collected plurality of pre-trained classification models are applied in parallel to the plurality of documents in the document pool to generate a list of labels. Based on the list of labels, a final label result is generated according to which a baseline algorithm for document classification is generated by the one or more processors.

Abstract: A method, system, and computer program product for segmenting and processing documents for optical character recognition is provided. The method includes receiving a document and detecting different types of text data. The document is divided into a plurality of text regions associated with the different types of said text data. Optical noise is removed from each text region and differing optical character recognition software code is selected for application to each text region. The differing optical character recognition software code is executed with respect to each text region resulting in extractable computer readable text located within each said text region.

Abstract: Monitoring of an environmental location to detect waste disposal includes receiving, by a computer, a collection of images from aerial data acquisition sources, the collection of images corresponding to the environmental location and the waste disposed in the environmental location. The collection of images are processed by the computer to extract properties of the waste and first properties of the environmental location. Subsequently, the properties of the waste disposed in the environmental location are classified according to a class. Additional information from external data sources including second properties of the environmental location is received by the computer. The computer determines a behavior of waste disposal in the environmental location over a period of time based on the classified properties of the disposed waste and the first and second properties of the environmental location, and generated a remediation plan based on the determined behavior.

Abstract: Embodiments of the present disclosure present and approach for automatic generation of short names for a named entity. According to the approach, a standard text segment is obtained, which indicates a full name of a named entity. At least one feature representation of the standard text segment is extracted. A plurality of variant text segments are generated based on the at least one feature representation using a generative learning network. The plurality of variant text segments indicate a plurality of short names for the named entity, the generative learning network characterizing a generation of variants for an input text segment. The plurality of variant text segments are stored in association with the standard text segment into a data repository.

Abstract: An approach is provided for generating graph database parameter settings. Parameter settings for importing data into a graph database are determined. A speed of importing simulated data into the graph database and a system resource usage are determined by executing an importing of the simulated data using the parameter settings and a simulated hardware environment. A reward associated with the parameter settings is determined. Using a policy network that includes convolutional neural networks and based on the reward and the settings, candidates of adjusted parameter settings are determined. Using a Monte Carlo tree search in multiple iterations to estimate changes in speeds of importing the simulated data and changes in system resource usages for candidates of the adjusted parameter settings, rewards for the candidates are determined. Based on the rewards, a candidate is selected as including final parameter settings that optimize an importing speed and system resource usage.

Abstract: Provided is a method, computer program product, and system for fusing knowledge graphs to generate a larger knowledgebase for responding to cross document questions. A processor may extract contextual information from a plurality of documents. The processor may generate, based on the extracted contextual information, a knowledge graph for each document of the plurality of documents. The processor may analyze each knowledge graph to determine if one or more entities of each knowledge graph are linked. The processor may fuse, in response to an entity in a first knowledge graph being linked to an entity in a second knowledge graph, the first knowledge graph with the second knowledge graph to create a fused knowledge graph.

Abstract: Embodiments of the present disclosure relate to generation of natural language expression variants. In an embodiment, a computer-implemented method is disclosed. According to the method, a structured expression is determined for a source expression in a natural language by replacing a source key entity in the source expression with a predetermined symbol. At least one template structured expression is selected from a set of template structured expressions based on respective similarities between the structured expression and respective template structured expressions in the set. Each of the set of template structured expressions comprises the predetermined symbol to represent a key entity. At least one variant expression is generated for the source expression by replacing the predetermined symbol in the at least one selected template structured expression with the source key entity. In other embodiments, a system and a computer program product are disclosed.

Abstract: Monitoring of an environmental location to detect waste disposal includes receiving, by a computer, a collection of images from aerial data acquisition sources, the collection of images corresponding to the environmental location and the waste disposed in the environmental location. The collection of images are processed by the computer to extract properties of the waste and first properties of the environmental location. Subsequently, the properties of the waste disposed in the environmental location are classified according to a class. Additional information from external data sources including second properties of the environmental location is received by the computer. The computer determines a behavior of waste disposal in the environmental location over a period of time based on the classified properties of the disposed waste and the first and second properties of the environmental location, and generated a remediation plan based on the determined behavior.

Abstract: In embodiments, a method includes receiving a first user input to a chat managed by a multiple-round dialogue system, receiving a second user input to the chat, the second user input following the first user input, and embedding the first and second user inputs into first and second nodes, respectively, each node including a multi-dimensional vector. The method further includes determining, based at least in part on the first node and the second node, if a context of the second user input is the same as a context of the first user input, and, based on the determination, generating a chat reply to the second user input.

Abstract: In embodiments, a method includes receiving a first user input to a chat managed by a multiple-round dialogue system, receiving a second user input to the chat, the second user input following the first user input, and embedding the first and second user inputs into first and second nodes, respectively, each node including a multi-dimensional vector. The method further includes determining, based at least in part on the first node and the second node, if a context of the second user input is the same as a context of the first user input, and, based on the determination, generating a chat reply to the second user input.

Abstract: In an approach for a text block recognition in a document, a processor detects characters in the document using an object detection technique. A processor identifies positions of the detected characters in the document. A processor analyzes semantic connectivity among the detected characters based on the positions and semantic connectivity of the characters. A processor recognizes text blocks of related characters based on the semantic connectivity analysis. A processor outputs the text blocks associated with the related characters.

Abstract: A method, system, and computer program product for reconstructing training data and building a new incremental learning model with the reconstructed training data that can be further trained. The method may include receiving new data to be inputted into a previously trained machine learning model, where the previously trained machine learning model has inaccessible training data. The method may also include generating simulated training data using a reverse form of the previously trained machine learning model. The method may also include verifying the simulated training data. The method may also include creating a new machine learning model using the simulated training data, where the new machine learning model includes a same structure as the previously trained machine learning model. The method may also include inputting the new data into the new machine learning model, where the new machine learning model is further trained with the new data.

Abstract: An aspect of the present invention discloses a method for extracting content from a document. The method includes one or more processors identifying a visual anchor corresponding to a text element depicted in a first document utilizing an edge detection analysis. The method further includes determining edge coordinates of the text element depicted in the first document. The method further includes determining text at a leading edge of the text element depicted in the first document and text at a trailing edge of the text element depicted in the first document, based on the determined edge coordinates. The method further includes extracting a complete version of the text element depicted in the first document, from a plain text version of the first document, utilizing the determined text at the leading edge of the text element and the determined text at the trailing edge of the text element.

Abstract: Described are techniques for determining statistical properties of time series data. The techniques include a method comprising graphing, from time series data, a time series data graph. The method further comprises iteratively segmenting the time series data graph into respective pluralities of subgraphs using respective segmentation schemes until a first plurality of subgraphs generated by a first segmentation scheme exhibits a similarity between respective subgraphs of the first plurality of subgraphs satisfying a similarity threshold. The first segmentation scheme can be selected from: an equidistant segmentation scheme, a local extrema segmentation scheme, and a windowed segmentation scheme. The method further comprises associating a classification to the time series data based on the first segmentation scheme. The classification can be indicative of one selected from: stationarity of the time series data, periodicity of the time series data, and trending of the time series data.

Abstract: A first set of features associated with a neural network are parameterized. A decision tree is generated from the first set of features. One or more adjustments for the neural network are received at the decision tree. A second set of features associated with the adjustments at the decision tree are parameterized. The parameterized first and second set of features are combined into a plurality of parameters. From the plurality, an adjusted neural network is generated.

Abstract: An approach for extracting non-textual data from an electronic document is disclosed. The approach includes receiving a request to extract a file and converting the file into pixels. The approach creates a pixel map of the converted file and determines one or more density clusters of the pixel map based on image clustering method. Furthermore, the approach determines one or more coordinates of the one or more density clusters and determines one or more candidate information regions based on the one or more coordinates, density of the one or more density clusters. Finally, the approach extracts one or more textual data based on the one or more candidate information regions and outputs the extracted one or more textual data.

Abstract: A set of candidate intent vectors is generated from an input intent vector. A validation of the set of candidate intent vectors is performed that selects as valid intent vectors any of the set of candidate intent vectors that are semantically similar to the input intent vector.