Stop Words Research Articles

EMOTIONAL BEHAVIOR ANALYSIS OF NOVEL CHARACTERS BASED ON COMPLEX NETWORK AND WLDA ALGORITHM — TAKING HARUKI MURAKAMI'S NORWEGIAN FOREST AS AN EXAMPLE

Abstract Background In recent years, with the rapid development of public network, emotion analysis has always been a research hotspot in the field of natural language processing and data mining. The current research mainly focuses on various comments on the Internet, and there is relatively little analysis of the psychological activities of the characters and the emotional changes of the text in the novel. How to use computer technology to identify the emotional tendency and psychological activities of characters in literary works has important practical significance. Topics and Methods This paper uses complex network and wlda algorithm to analyze the mood of Norwegian forest. Using complex network is to preprocess the data, extract the information in the article by using word frequency statistics, then build a complex network according to word frequency, find out the key points of complex network according to the principle of structural hole, complex network, and analyze the emotional tendency to be expressed in the article. The wlda algorithm model is used to segment the data, remove the stop words, and then the algorithm is used to verify the emotional tendency of the novel. The corpus used in the experiment is Haruki Murakami's novel Norwegian forest. The emotion seed words used in the experiment are from the Chinese word set used for emotion analysis in the Internet HowNet. The algorithm parameters take the data commonly used in wlda model, where 50 is equal to 0.01, and the number of keywords to judge the subject's emotional tendency is C, which is equal to 100. Readers' emotion algorithm, this study uses the relevant scale to investigate. (1) Positive emotion scale. The Panas emotion scale developed by Wason et al. Is widely used to measure emotion. The scale includes two dimensions: positive emotion and negative emotion. There are 6 questions in this dimension. In addition, the boredom tendency questionnaire was used to investigate internet boredom. The boredom tendency questionnaire was prepared by Huang Shihua et al. In 2010. The research shows that the scale has high reliability and validity. The scale has 30 items and is scored by Likert 7 points (from 7 to 1 means “completely agree” to “completely disagree”, and 4 means neutral). The scale includes two sub questionnaires of external stimulation and internal stimulation. The external stimulus sub questionnaire includes four factors: monotonicity, loneliness, tension and restraint. The internal stimulation sub questionnaire contains two factors: self-control and creativity. The higher the questionnaire score, the higher the boredom tendency. Group learning burnout scale group learning burnout scale was compiled by Lian Rong et al in 2005. The research shows that the scale has high reliability and validity. The scale has 20 items and uses a 5-level scoring method (from 5 to 1 means “completely consistent” to “completely inconsistent”, and 3 means neutral). It includes three dimensions, including depression, improper behavior and low sense of achievement. Emotion regulation strategy scale emotion regulation style scale was compiled by gross et al in 2003. The Chinese version of the scale has been proved to have high reliability and validity. The scale has 10 items and adopts Likert 7-point scoring (from 7 to 1 means “fully agree” to “completely disagree”, and 4 means neutral). The scale includes two sub questionnaires: cognitive reappraisal and expression inhibition. Data Analysis Adopt spss16 0 and amos17 0 statistical software Line statistical processing. Results In the process of simplifying complex network, the frequency of low-frequency words was 1. Experiments show that the results of complex network and wlda algorithm model are basically consistent, and the effect of emotion analysis is obvious. Readers' emotional response and emotional effect are also basically the same. Conclusion Some high-frequency but meaningless stop words in the corpus have caused great interference to the reasoning of the model topic. Therefore, when analyzing the text, we need to preprocess the corpus and filter out low-frequency words, which affects the emotion extraction to a certain extent. In the process of simplifying complex networks, it is also necessary to adjust the threshold of filtered low-frequency words according to different work. The experimental results show that the negative tendency is greater than the positive tendency, and the whole text expresses the negative emotion, that is, the sadness and confusion of survival. Acknowledgements Supported by the doctoral startup Research (No.20rc15), the research on the influence of consumers' purchase intention of traceable agricultural products (No.20rc03), and the design of precision control system based on the Internet of things (No.202103001).

THE APPROACH AND THE SOFTWARE TOOL TO CALCULATE SEMANTIC QUALITY MEASURES OF BUSINESS PROCESS MODELS

Business process models are essential business process management artifacts that help describe visually ongoing business activities to facilitate communication between information technology and business stakeholders. Business process models are used to find inefficient spots within described workflows and resolve detected shortcomings by automation via configurable software solutions or unified workflow engines. However, this is impossible when using syntactically or semantically poor business process models. It is the same as building a house using the blueprint with windows on the floor and typos in text labels. Therefore, it is extremely important to keep created business process models clear and relevant to the actual workflows they describe. Hence, in this paper, we propose the approach and the software tool to calculate semantic quality measures of business process models. The proposed approach uses a special procedure to extract the modeling domain statements using natural language processing techniques. According to the proposed approach, the initial textual descriptions of business process models should be tokenized. Then obtained tokens should be turned to the lower case style and cleansed to remove non-alphabetic tokens and stop words. Finally, the remaining tokens should be stemmed and the existing duplicates should be removed. The same procedure is then repeated for text labels attached to the business process model activities. Then, tokens present in the result of textual description’s processing but missing in the result of labels’ processing are considered incomplete (i.e. incorrect in the modeling domain). Similarly, tokens present in the result of labels’ processing but missing in the result of textual description’s processing are considered invalid (i.e. irrelevant to the modeling domain). Therefore, respective semantic quality measures can be calculated. The software tool is created using the Python programming language because of its powerful natural language processing packages.

DILI C : An AI-Based Classifier to Search for Drug-Induced Liver Injury Literature.

Drug-induced liver injury (DILI) is a class of adverse drug reactions (ADR) that causes problems in both clinical and research settings. It is the most frequent cause of acute liver failure in the majority of Western countries and is a major cause of attrition of novel drug candidates. Manual trawling of the literature is the main route of deriving information on DILI from research studies. This makes it an inefficient process prone to human error. Therefore, an automatized AI model capable of retrieving DILI-related articles from the huge ocean of literature could be invaluable for the drug discovery community. In this study, we built an artificial intelligence (AI) model combining the power of natural language processing (NLP) and machine learning (ML) to address this problem. This model uses NLP to filter out meaningless text (e.g., stop words) and uses customized functions to extract relevant keywords such as singleton, pair, and triplet. These keywords are processed by an apriori pattern mining algorithm to extract relevant patterns which are used to estimate initial weightings for a ML classifier. Along with pattern importance and frequency, an FDA-approved drug list mentioning DILI adds extra confidence in classification. The combined power of these methods builds a DILI classifier (DILI C ), with 94.91% cross-validation and 94.14% external validation accuracy. To make DILI C as accessible as possible, including to researchers without coding experience, an R Shiny app capable of classifying single or multiple entries for DILI is developed to enhance ease of user experience and made available at https://researchmind.co.uk/diliclassifier/. Additionally, a GitHub link (https://github.com/sanjaysinghrathi/DILI-Classifier) for app source code and ISMB extended video talk (https://www.youtube.com/watch?v=j305yIVi_f8) are available as supplementary materials.

Fake news detection using machine learning

ID of the phony news is the pivotal advance. Calculation like SVM and NB are utilized in our task. Furthermore, we remove Fake news-explicit feeling information each Trend's instances, both labelled and unlabeled, and use it to enhance the understanding of Fake news-explicit opinion classifiers. News online has turned into the significant wellspring of data for individuals., much data showing up on the Internet is questionable and, surprisingly, planned to misdirect. Some phony news is so like the genuine ones that it is hard for human to distinguish them. robotized counterfeit news location devices like AI and profound learning models have turned into a fundamental necessity. additionally utilized stemming, lemmatization, stop word methods to get message portrayal for AI and profound learning models separately. The significant item perspectives are recognized in light of two perceptions. Fully intent on ordering words early on. This would permit to give a separated subset of phony news to end clients. We dissect and explore different avenues regarding a bunch of clear language-autonomous elements in view of the social spread of phony news to classify them into the presented typology

A Deep Learning Modified Neural Network(DLMNN) based proficient sentiment analysis technique on Twitter data

ABSTRACT The rapid enhancement in social media over the internet generates massive information in real-time scenarios, which has a striking impact on big data analysis. It resulted in the elevated usage of emotions and sentiments in social media. This paper proffers a proficient sentiment analysis technique in Twitter data. The Twitter database is preprocessed includes, stemming, tokenisation, number removal and stop word removal, etc. The preprocessed words are then passed into the HDFS (Hadoop Distributed File System) to reduce the repeated words and are eliminated using the MapReduce technique. The emoticons and the non-emoticons are extorted as features. The resulted features are ranked with their intended meaning. Then, the classification is performed utilising the DLMNN (Deep Learning Modified Neural Network). The experimental results were examined by using the output parameter such as Accuracy, Recall, Precision, F-Score and Execution Time with other conventional techniques such as ANN, SVM, K-Means and DCNN to show the greatest outcome of the proposed model. Evaluation result shows that DLMNN achieved the greatest performance in terms of precision (95.78%), Recall (95.84%), F-Score (95.87%) and Accuracy (91.65%) when compared with the existing models.

Deep learning and natural language processing in computation for offensive language detection in online social networks by feature selection and ensemble classification techniques

Offensive communications have made their way into social media posts. Using computational algorithms to distinguish objectionable content is one of the most effective ways to deal with this problem. One of the most effective approaches to deal with this issue is to use computational methods to distinguish undesirable content. This research aims to tackle MOLD_DL (Multilingual Offensive Language Detection using deep learning) techniques and natural language processing used in feature selection and classification. Here the dataset has been collected from YouTube, Twitter and Facebook, which has been pre-processed for noise removal, filtering and removing the stop words and segmented. The feature selection has been carried out for segmented data using Fuzzy based convolutional neural network (FCNN). Then the extraction of selected features and classification has been carried out using ensemble architecture of Bi-LSTM model with Naïve Bayes architecture hybrid with Support Vector Machines (SVM). Evaluation of offensive language detection is classified automatically based on the emotions of the text. Here the experimental analysis has been carried out for YouTube, Twitter and Facebook datasets in terms of accuracy of 98%, precision of 95%, recall of 90%, F-1 score of 92.5% and RMSE of 45% with the confusion matrix in detecting offensive text of various languages.

Creation of a Russian Stop Word List

Creation of a Russian Stop Word List

Depression prognosis using natural language processing and machine learning from social media status

<p><span>Depression is an acute problem throughout the world. Due to worst and prolong depression many people dies in every year. The problem is that most of the people are not concern of the fact that they are suffering from depression. In this research, our aim was to find out whether an individual is depressed or not by analyzing social media status. Therefore, we focused on real data. Our dataset consists of 2000 sentences, which was collected from different social media platforms Facebook, Twitter, and Instagram. Then, we have performed five data pre-processing approaches for natural language processing (NLP) such as tokenization, removal of stop words, removing empty string, removing punctuations, stemming and lemmatization. For our selected model, we considered that processed data as an input. Finally, we applied six machine learning (ML) classifiers multinomial Naive Bayes (NB), logistic regression, liner support vector classifier, random forest, K-nearest neighbour, and decision tree to achieve better accuracy over our dataset. Among six algorithms, multinomial NB and logistic regression performed well on our dataset and obtained 98% accuracy.</span></p>

Ensemble of Artificial Intelligence Techniques for Bacterial Antimicrobial Resistance (AMR) Estimation Using Topic Modeling and Similarity Measure

In recent times, bacterial Antimicrobial Resistance (AMR) analyses becomes a hot study topic. The AMR comprises information related to the antibiotic product name, class name, subclass name, type, subtype, gene type, etc., which can fight against the illness. However, the tagging language used to determine the data is of free context. These contexts often contain ambiguous data, which leads to a hugely challenging issue in retrieving, organizing, merging, and finding the relevant data. Manually reading this text and labelling is not time-consuming. Hence, topic modeling overcomes these challenges and provides efficient results in categorizing the topic and in determining the data. In this view, this research work designs an ensemble of artificial intelligence for categorizing the AMR gene data and determine the relationship between the antibiotics. The proposed model includes a weighted voting based ensemble model by the incorporation of Latent Dirichlet Allocation (LDA) and Hierarchical Recurrent Neural Networks (HRNN), shows the novelty of the work. It is used for determining the amount of “topics” that cluster utilizing a multidimensional scaling approach. In addition, the proposed model involves the data pre-processing stage to get rid of stop words, punctuations, lower casing, etc. Moreover, an explanatory data analysis uses word cloud which assures the proper functionality and to proceed with the model training process. Besides, three approaches namely perplexity, Harmonic mean, and Random initialization of K are employed to determine the number of topics. For experimental validation, an openly accessible Bacterial AMR reference gene database is employed. The experimental results reported that the perplexity provided the optimal number of topics from the AMR gene data of more than 6500 samples. Therefore, the proposed model helps to find the appropriate antibiotic for bacterial and viral spread and discover how to increase the proper antibiotic in human bodies

Machine Learning for Multilabel Emotion Classification in Arabic Tweets

Multilabel emotion classification is a high priority because it mimics real-life scenarios in which people display a variety of emotions. The text could express a collection of emotions such as happiness, love, and optimism, or sadness, anger, and pessimism. In this framework, the Arabic tweets data provided by SemEval 2018-Task1, E-c subtask have been first preprocessed through different normalization steps, including stemming, stop word removal, special characters, and digits removal. An emotion lexicon has been built to replace the emotions with their meaning related to emotion classes. A word embedding pre-trained model Aravec has been implemented for the feature extraction process because word embedding performed better in this task than other features such as the N-gram model. In the classification process of our framework, different machine learning techniques have been implemented, including Multi-Layer Perceptron (MLP), Support Vector Machine SVM, K Nearest Neighbor (KNN), Ensemble Random Forest (RF), and Ensemble Extra Tree. The best performance was achieved using MLP, whereas SVM proved to perform best over other Traditional machine learning techniques such as KNN, RF, and Extra tree. Extra tree achieved a multilabel Jaccard accuracy of 26.2%, Nearest Neighbor (KNN) of 37.5%, Ensemble Random Forest (RF) of 29.1%, and SVM accuracy of 46.3%. A neural network model Multi-Layer Perceptron (MLP), achieved an accuracy of 48%. The proposed framework has been compared with different previous machine learning models built for this task; the results obtained by the proposed framework outperform other previous models in most cases.

Automated Fake News Detection by LSTM Enabled with Optimal Feature Selection

Fake news plays a major role by broadcasting misinformation, which influences people’s knowledge or perceptions and distorts their decision-making and awareness. Online forums and social media have stimulated the broadcast of fake news by embedding it with truthful information. Thus, fake news has evolved into the main challenge of better impact in the information-driven community for intense fakesters. The detection of fake news articles that is generally found by considering the quality of the information in their news feeds under uncertain authenticity calls for automated tools. However, designing such tools is a major problem because of the multiple faces of fakesters. This paper offers a new text-analytics-driven method for detecting fake news to reduce the risks impacted by the consumption of fake news. The methodology for improved fake news detection focusses on four phases: (a) pre-processing, (b) feature extraction, (c) optimal feature selection and (d) classification. The pre-processing of the text data will be initially done by stop word removal, blank space removal and stemming. Further, the feature extraction is performed by term frequency-inverse document frequency, and grammatical analysis is done using mean, Q25, Q50, Q75, Max, Min and standard deviation. Then, the optimal feature selection is developed, which minimises the number of input variables. It is intended to reduce the number of input variables to improve the model’s performance by minimising the computational cost of modelling. An improved meta-heuristic algorithm called successive position-based barnacles mating optimisation is used for optimal feature selection and classification. As the main contribution, the influence of deep learning is employed, which employs optimised long short-term memory. Finally, the result shows the superiority in terms of different significant measures by the proposed model over other methods for fake news detection experimentally done on a publicly available benchmark dataset.

Building a Sustainable Social Feedback Loop: A Machine Intelligence Approach for Twitter Opinion Mining

This paper presents a sustainable machine intelligence approach for Twitter opinion mining, focusing on building a socially responsible feedback loop. We propose a methodology that combines advanced machine learning algorithms with eco-conscious practices to extract sentiment-related insights from Twitter data while minimizing environmental impact. The preprocessing steps involve removing special characters, tokenization, stop word removal, handling user handles and URLs, and lemmatization or stemming. Sentiment classification is performed using the Extra Tree Classifier, an ensemble learning algorithm that incorporates random feature selection and bagging techniques. Experimental results demonstrate the effectiveness of our approach in accurately classifying tweets into positive, negative, and neutral sentiment categories. The visualizations of class distribution, number of tokens per tweet, and word clouds provide further insights into the sentiment landscape on Twitter. Our research contributes to the development of sustainable and inclusive approaches for Twitter opinion mining, ensuring minimal environmental impact while capturing valuable sentimental information.

An Improved Cat Swarm Search-Based Deep Ensemble Learning Model for Group Recommender Systems

Recommender systems are often employed in different fields such as music, travel, and movies. The recommender systems are broadly utilised nowadays due to the emergence of social activities, in which particular recommendations are offered by group recommender systems. It is a system for recommending the items to a set of users together based on their preferences. The user preferences are used from the behavioural and social aspects of group members to enhance the quality of products recommended in various groups for generating the group recommendations. These group recommender systems solve the cold start problem, which is raised in an individual recommendation system. The ultimate aim of this paper is to design and develop a new Improved Deep Ensemble Learning Model (ID-ELM) for the group recommender systems concerning different application-oriented datasets. Initially, the datasets from different applications like healthcare, e-commerce, and e-learning are gathered from benchmark sources and split the data into various groups. These data are given to the pre-processing for making it fit for further processing. The pre-processing steps like stop word removal, stemming, and punctuation removal are performed here. Then the features are extracted using the Continuous Bag of Words Model (CBOW), and Principal Component Analysis (PCA) is used for dimension reduction. These features are fed to the ID-ELM, in which the optimised Convolutional Neural Network (CNN) extracts the significant features from the pooling layer, and the fully connected layer is replaced by a set of classifiers termed Neural Networks (NN), AdaBoost, and Logistic Regression (LR). Finally, the ranking of the ensemble learning model based on the group reviews extends the recommendation outcome. The optimised CNN is proposed by the Adaptive Seeking Range-based Cat Swarm Optimisation (ASR-CSO) for attaining better results. This model is validated on the benchmark datasets to show the efficiency of the designed model with different meta-heuristic-based algorithms and classification algorithms.

The Design of Automatic Summarization of Indonesian Texts Using a Hybrid Approach

This study aims to design a model for automatic text summarization in Indonesian. Automatic text summarization is a system that reduces the number of sentences without losing important information in the document. There are three approaches in making automatic text summarization, namely: extractive, abstractive, and hybrid. The extractive approach is to select a core sentence without changing it to a new sentence. The abstractive approach is to construct new sentences that describe the contents of the document. Meanwhile, the hybrid approach designed in this study is a combination of extractive and abstractive approaches. By designing automatic text summarization in Indonesian with a hybrid approach, it is hoped that the results issued by the system will be more like man-made summaries and have higher readability. The design made is specifically for the input of a document. The stages of automatic text summarization are divided into two, namely preprocessing and process. In the Preprocessing stage, sentence separation, tokenization, coreference resolution, stop words removal, feature extraction are carried out. There are two stages of making the summary, namely the extraction stage, to select important sentences and the abstraction stage, to select words and arrange them into summary sentences. Further research can be carried out for input of longer documents and the input is in the form of multi documents

An Efficient Data Classification Decision Based on Multimodel Deep Learning

A single model is often used to classify text data, but the generalization effect of a single model on text data sets is poor. To improve the model classification accuracy, a method is proposed that is based on a deep neural network (DNN), recurrent neural network (RNN), and convolutional neural network (CNN) and integrates multiple models trained by a deep learning network architecture to obtain a strong text classifier. Additionally, to increase the flexibility and accuracy of the model, various optimizer algorithms are used to train data sets. Moreover, to reduce the interference in the classification results caused by stop words in the text data, data preprocessing and text feature vector representation are used before training the model to improve its classification accuracy. The final experimental results show that the proposed model fusion method can achieve not only improved classification accuracy but also good classification effects on a variety of data sets.

Radial Restricted Boltzmann Machines with Functional Neural Network for Classification of the Fake and Real News Analysis

News methods are the main source of news for people around the world. However, there was a surprising rise in social media spread in later periods due to the significant increase in technology and updates. Fake news is as old as the news industry - misinformation, hoax, propaganda and satire have been around for a long time. So fake news is untrue and false information without proper evidence. Auto-detection technology is being built and studied on artificial intelligence and deep learning to address fake news’s rise and spread. This method’s efforts have been made to automate the process of fake news detection. This existing machine learning method has less classification accuracy, and multiple combination Machine learning methods have not consumed memory usage; it provides a higher false rate. Overcome this problem by introducing a deep learning method, namely Radial Restricted Boltzmann Machines with Functional Neural Network (R2BM-FNN), to reduce memory usage from a large set of training data. Initially, to apply the Count Victories distribution for data preprocessing to remove the stop words and unrelated text from data sets. In this proposed method neural network to train the neurons with the help of stochastic gradient descent. This proposed GRU-RLU method results in classification accuracy, precision, sensitivity, specificity, and execution time to evaluate system performance.

Emotion Analysis Model of Microblog Comment Text Based on CNN-BiLSTM.

Aiming at the problems of over reliance on labor and low generalization of traditional emotion analysis methods based on dictionary and machine learning, an emotion analysis model of microblog comment text based on deep learning is proposed. Firstly, text is obtained through microblog crawler program. After data preprocessing, including data cleaning, Chinese word segmentation, removal of stop words, and so on, the Skip-gram model is used for word vector training on a large-scale unmarked corpus, and then the trained word vector is used as the text input of CNN-BiLSTM model, which combines Bidirectional Long-Short Term Memory (BiLSTM) neural network and Convolution Neural Network (CNN). Considering the historical context information and the subsequent context information, BiLSTM can better use the temporal relationship of text to learn sentence semantics. CNN can extract hidden features from the text and combine them. Finally, after Adamax optimization training, the emotion type of microblog comment text is output. The proposed model combines the learning advantages of BiLSTM and CNN. The overall accuracy of text emotion analysis has been greatly improved, with an accuracy of 0.94 and an improvement of 8.51% compared with the single CNN model.

Fake news detection on Hindi news dataset

With the increase in social networks, more number of people are creating and sharing information than ever before, many of them have no relevance to reality. Due to this, fake news for various political and commercial purposes are spreading quickly. Online newspaper has made it challenging to identify trustworthy news sources. In this work, Hindi news articles from various news sources are collected. Preprocessing, feature extraction, classification and prediction processes are discussed in detail. Different machine learning algorithms such as Naïve Bayes, logistic regression and Long Short-Term Memory (LSTM) are used to detect the fake news. The preprocessing step includes data cleaning, stop words removal, tokenizing and stemming. Term frequency inverse document frequency(TF-IDF) is used for feature extraction. Naïve Bayes, logistic regression and LSTM classifiers are used and compared for fake news detection with probability of truth. It is observed that among these three classifiers, LSTM achieved best accuracy of 92.36%.

Illusion of Truth: Analysing and Classifying COVID-19 Fake News in Brazilian Portuguese Language

Public health interventions to counter the COVID-19 pandemic have accelerated and increased digital adoption and use of the Internet for sourcing health information. Unfortunately, there is evidence to suggest that it has also accelerated and increased the spread of false information relating to COVID-19. The consequences of misinformation, disinformation and misinterpretation of health information can interfere with attempts to curb the virus, delay or result in failure to seek or continue legitimate medical treatment and adherence to vaccination, as well as interfere with sound public health policy and attempts to disseminate public health messages. While there is a significant body of literature, datasets and tools to support countermeasures against the spread of false information online in resource-rich languages such as English and Chinese, there are few such resources to support Portuguese, and Brazilian Portuguese specifically. In this study, we explore the use of machine learning and deep learning techniques to identify fake news in online communications in the Brazilian Portuguese language relating to the COVID-19 pandemic. We build a dataset of 11,382 items comprising data from January 2020 to February 2021. Exploratory data analysis suggests that fake news about the COVID-19 vaccine was prevalent in Brazil, much of it related to government communications. To mitigate the adverse impact of fake news, we analyse the impact of machine learning to detect fake news based on stop words in communications. The results suggest that stop words improve the performance of the models when keeping them within the message. Random Forest was the machine learning model with the best results, achieving 97.91% of precision, while Bi-GRU was the best deep learning model with an F1 score of 94.03%.

Methodology for solving problems of classification of appeals/requests of citizens to the “hotline” of the President of the Russian Federation

The use of neural networks for the classification of text data is an important area of digital transformation of socio-economic systems. The article is devoted to the description of the methodology for classifying citizens' appeals. The proposed technique involves the use of a convolutional neural network. The stages of processing citizens' appeals in the amount of 7000 appeals are described. In order to reduce the dimension of the problem, methods of filtering and removing stop words were applied. The resulting data set allows you to choose the best classifier in terms of accuracy, specificity, sensitivity. Training and test samples were used, as well as cross-validation. The article shows the effectiveness of using this method to distribute requests on 15 topics of citizens' appeals to the “hotline” of the President of the Russian Federation. Automating the classification of received appeals by topic allows them to be processed quickly for further study by the relevant departments. The purpose of the study is automation of the distribution of citizens' appeals to the President's hotline by category based on the use of modern machine learning methods. Materials and methods. The development of software that automates the process of distributing citizens into categories is carried out using a convolutional neural network written in the Python programming language. Results. With the help of the prepared data set, the pre-trained model of NL BERT and sciBERT was trained by the deep learning method. The model shows an accuracy of 86% in the estimates of quality metrics. Conclusion. A pre-trained model was trained using a convolutional neural model using a prepared data set. Even if the forecast does not match the real category, the model gives a minor error, correctly determines the category of the appeal. The results obtained can be recommended for practical application by authors of scientific publications, scientific institutions, editors and reviewers of publishing houses.