Scientific papers of Donetsk National Technical University. Series: Informatics, Cybernetics and Computer Science最新文献

{"title":"INTERFACES SIMILARITY ANALYSIS FOR PROGRESSIVE WEB APPS AND WEB-APPLICATIONS BASED ON DISTILBERT TRANSFORMER","authors":"H. Yehoshyna, S. M. Voronoy, O. Polikarovskykh, R. Gokhman","doi":"10.31474/1996-1588-2023-1-36-51-60","DOIUrl":"https://doi.org/10.31474/1996-1588-2023-1-36-51-60","url":null,"abstract":"An approach to automated testing of components of Progressive Web Applications interfaces by determining their relevance to elements of the corresponding web versions of applications is proposed. An analysis of modern trends and existing categories in the field of Web Mining was carried out. It is shown that the predominant trend in the analysis of the interface structures of modern web applications is the use of Deep Learning technologies. Features and functioning of the latest Transformers neural network architecture are considered. The choice of the Transformers type model to determine the correspondence between the site structure and the PWA application interface is justified. It is shown that in the comparison of fragments of the interfaces of the web service and the PWA application, some elements have more impact (weight) than others. It is proposed to use the mechanism of multidimensional \"self-attention\" to take into account this feature of the content. It is shown that the analysis of correspondence of interfaces is a task of binary classification. Features of transformers of the Bidirectional Encoder Representations (BERT) type are viewed. Pretrained BERT model can be configured with only one additional output layer to create modern and powerful models for a wide range of problems. It is proposed to use transfer learning, namely the DistilBERT model and its fine tuning using the DistilBertForSequenceClassification class. For the basic architecture of DistillBert (embedding and encoder layers), the weights of the English-language model \"distilbert-base-uncased-finetuned-sst-2-english\" were used. The model was optimized using a modification of the Adam stochastic gradient descent method. It is also suggested to use a low learning rate to avoid \"forgetting\". The features of data preprocessing using DistilBertTokenizer are shown. The architecture of the model was designed and its research was done based on data set of CSS properties, which provide styling and layout of interface elements.","PeriodicalId":104072,"journal":{"name":"Scientific papers of Donetsk National Technical University. Series: Informatics, Cybernetics and Computer Science","volume":"34 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1900-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129204700","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}