{"title":"基于计算智能的文本数据识别模型设计","authors":"O. Dmitrieva, E. Babenko","doi":"10.31474/1996-1588-2023-1-36-4-10","DOIUrl":null,"url":null,"abstract":"The article is devoted to the development, justification, software implementation and research of a text data recognition model based on the use of neural networks with a hybrid architecture. The proposed model allows you to gain new knowledge through a combination of different activities and methods of text analysis. The main tasks implemented in the work were to study the essence and relevance of text data recognition, determine the criteria for assessing the quality of recognition, design the architecture of the model and software application, develop and train a software model of text recognition. Software application testing and a comparative analysis of recognition quality in terms of accuracy, reliability, completeness and time were carried out with the involvement of other neural networks. As a criterion for assessing the quality of neural network training, the loss function was used, which characterized the normalized deviation of the results of the actual values obtained by the neural network from the expected ones. At the stages of training and testing, the model was tuned, the optimal number of training epochs was determined in terms of the reliability metric. Based on the results obtained, it can be argued that the proposed hybrid architecture of the ConvBiGRU neural network has the highest quality indicators, similar to the ConvBiLSTM model, but the time spent on the implementation of one epoch is less. There was also a tendency to equalize the times of realization of epochs for combined models compared to simple models when large arrays of text data were processed.","PeriodicalId":104072,"journal":{"name":"Scientific papers of Donetsk National Technical University. Series: Informatics, Cybernetics and Computer Science","volume":"17 1","pages":"0"},"PeriodicalIF":0.0000,"publicationDate":"1900-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":"{\"title\":\"DESIGNING A TEXT DATA RECOGNITION MODEL USING COMPUTATIONAL INTELLIGENCE\",\"authors\":\"O. Dmitrieva, E. Babenko\",\"doi\":\"10.31474/1996-1588-2023-1-36-4-10\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"The article is devoted to the development, justification, software implementation and research of a text data recognition model based on the use of neural networks with a hybrid architecture. The proposed model allows you to gain new knowledge through a combination of different activities and methods of text analysis. The main tasks implemented in the work were to study the essence and relevance of text data recognition, determine the criteria for assessing the quality of recognition, design the architecture of the model and software application, develop and train a software model of text recognition. Software application testing and a comparative analysis of recognition quality in terms of accuracy, reliability, completeness and time were carried out with the involvement of other neural networks. As a criterion for assessing the quality of neural network training, the loss function was used, which characterized the normalized deviation of the results of the actual values obtained by the neural network from the expected ones. At the stages of training and testing, the model was tuned, the optimal number of training epochs was determined in terms of the reliability metric. Based on the results obtained, it can be argued that the proposed hybrid architecture of the ConvBiGRU neural network has the highest quality indicators, similar to the ConvBiLSTM model, but the time spent on the implementation of one epoch is less. There was also a tendency to equalize the times of realization of epochs for combined models compared to simple models when large arrays of text data were processed.\",\"PeriodicalId\":104072,\"journal\":{\"name\":\"Scientific papers of Donetsk National Technical University. Series: Informatics, Cybernetics and Computer Science\",\"volume\":\"17 1\",\"pages\":\"0\"},\"PeriodicalIF\":0.0000,\"publicationDate\":\"1900-01-01\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"0\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"Scientific papers of Donetsk National Technical University. Series: Informatics, Cybernetics and Computer Science\",\"FirstCategoryId\":\"1085\",\"ListUrlMain\":\"https://doi.org/10.31474/1996-1588-2023-1-36-4-10\",\"RegionNum\":0,\"RegionCategory\":null,\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"\",\"JCRName\":\"\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"Scientific papers of Donetsk National Technical University. Series: Informatics, Cybernetics and Computer Science","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.31474/1996-1588-2023-1-36-4-10","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
DESIGNING A TEXT DATA RECOGNITION MODEL USING COMPUTATIONAL INTELLIGENCE
The article is devoted to the development, justification, software implementation and research of a text data recognition model based on the use of neural networks with a hybrid architecture. The proposed model allows you to gain new knowledge through a combination of different activities and methods of text analysis. The main tasks implemented in the work were to study the essence and relevance of text data recognition, determine the criteria for assessing the quality of recognition, design the architecture of the model and software application, develop and train a software model of text recognition. Software application testing and a comparative analysis of recognition quality in terms of accuracy, reliability, completeness and time were carried out with the involvement of other neural networks. As a criterion for assessing the quality of neural network training, the loss function was used, which characterized the normalized deviation of the results of the actual values obtained by the neural network from the expected ones. At the stages of training and testing, the model was tuned, the optimal number of training epochs was determined in terms of the reliability metric. Based on the results obtained, it can be argued that the proposed hybrid architecture of the ConvBiGRU neural network has the highest quality indicators, similar to the ConvBiLSTM model, but the time spent on the implementation of one epoch is less. There was also a tendency to equalize the times of realization of epochs for combined models compared to simple models when large arrays of text data were processed.