Search academic texts

Information × Registration Number 0224U032489, (0124U004319) , R & D reports Title Development of methods of neural network classification of textual and graphic documents popup.stage_title Розроблення теоретико-методичного забезпечення та формування практичних рекомендацій щодо нейромережевої класифікації текстово-графічних документів Head Udovenko Serhii H., Доктор технічних наук Registration Date 01-11-2024 Organization Simon Kuznets Kharkiv National University of Economics popup.description1 Improvement of theoretical and methodological support and formation of practical recommendations regarding methods of neural network classification of textual and graphic documents popup.description2  The report presents the results of research related to the classification of textual and graphic documents to be processed in publishing houses and printing companies. Important aspects of the analysis of these tasks, the development of promising methods for the classification of textual and graphic fragments of documents, as well as their complex use as part of a multimodal system for processing textual and graphic documents are considered. The analysis of the most promising methods and architectures of neural network classification of textual and graphic documents was carried out. A vector model for determining the similarity of text-graphic fragments based on visual characteristics of images automatically extracted from visual content is considered. Possibilities of improving the quality of neural network classification of text-graphic documents using SMOTE, AutoML technologies, as well as data augmentation and visual transformer are considered. The results of testing the built classifiers of textual and graphic documents according to the data of the test datasets are given. The results of testing the proposed methods confirm the expediency of their use in the analysis systems of electronic textual and graphic documents. The recommendations provided in the report can be useful for use in publishing houses and printing companies, since some types of text-graphic documents intended for further printing require specific processing (in particular, multiclass classification). Product Description popup.authors Andriienko Anna D. Teslenko Oleh V. Udovenko Serhii H. popup.nrat_date 2024-11-01 Close