FIELD: information technology.
SUBSTANCE: invention relates to a method for text generation based on machine learning. Method includes creating expounders designed to recognize the semantic content of individual text fragments of the analyzed document included in the digital content, each of which can initiate the involvement of other expounders to ensure the best accuracy, use of several separate models of neural networks, each of which is trained on its individual data set, to recognize a separate text element, combining the required set of neural models into a conveyor contained in the expander, to solve the recognition task, the expander forming several possible versions of the recognized individual text elements, wherein if the expounder does not contain a neural network for recognizing any text element, then creates a request to create a new neural network with other training data obtained from another expounder, to which, in case of its addition to the pipeline, the execution of the recognition request is delegated, the selection of the most suitable version from the recognized individual text elements, which satisfies the given criteria, formation of a new text document based on the selected recognized text elements, taking into account a given template, which sets the design style, wherein the stage of combining the required set of neural models into the pipeline is carried out by parameterizing existing expounders with new neural networks and training data sets for formation by expander of several solutions for each task of recognition of separate text elements, where solutions are required neural networks and training data sets, wherein expander in real time performing dynamic combination of new neural networks and training data sets and subsequent solutions of problems of recognition of text documents or their fragments in parallel streams in a multithreaded environment.
EFFECT: high speed of generating a target text document owing to implementation of a multithreaded parallel computing environment.
1 cl, 6 dwg, 2 tbl
Title | Year | Author | Number |
---|---|---|---|
METHOD FOR RECOGNITION OF TEXT IN IMAGES OF DOCUMENTS | 2021 |
|
RU2768544C1 |
DETECTING TEXT FIELDS USING NEURAL NETWORKS | 2018 |
|
RU2699687C1 |
IDENTIFICATION OF FIELDS AND TABLES IN DOCUMENTS USING NEURAL NETWORKS USING GLOBAL DOCUMENT CONTEXT | 2019 |
|
RU2723293C1 |
METHOD AND SYSTEM FOR SEARCHING GRAPHIC IMAGES | 2022 |
|
RU2807639C1 |
DETECTING SECTIONS OF TABLES IN DOCUMENTS BY NEURAL NETWORKS USING GLOBAL DOCUMENT CONTEXT | 2019 |
|
RU2721189C1 |
METHOD FOR GENERATING MATHEMATICAL MODELS OF A PATIENT USING ARTIFICIAL INTELLIGENCE TECHNIQUES | 2017 |
|
RU2720363C2 |
IMAGE RECOGNITION SYSTEM: BEORG SMART VISION | 2020 |
|
RU2777354C2 |
METHOD FOR CONTROLLING A DIALOGUE AND NATURAL LANGUAGE RECOGNITION SYSTEM IN A PLATFORM OF VIRTUAL ASSISTANTS | 2020 |
|
RU2759090C1 |
MACHINE TRAINING | 2005 |
|
RU2391791C2 |
RECOGNITION OF EVENTS ON PHOTOGRAPHS WITH AUTOMATIC SELECTION OF ALBUMS | 2020 |
|
RU2742602C1 |
Authors
Dates
2024-06-26—Published
2023-07-18—Filed