Stop Words Research Articles

DeepGram: Combining Language Transformer and N-Gram based ML Models for YouTube Spam Comment Detection

Spam comments on YouTube videos are a persistent issue that can negatively impact the user experience and content creator's reputation. In this paper, we propose an algorithm called "DeepGram" for detecting YouTube spam comments using a combination of deep learning-based language transformer models and N-gram based machine learning models. The algorithm leverages the power of language transformers, which have shown significant success in various natural language processing tasks, along with N-gram based models that capture local context and patterns in the text data. The proposed algorithm goes through several stages, including data collection, text preprocessing, feature extraction, and model training. The collected YouTube comments are preprocessed by removing special characters, punctuation, and HTML tags, and converting them to lowercase. Common stop words are also removed, and stemming or lemmatization is applied to reduce dimensionality. The algorithm then extracts features from the preprocessed comments using a combination of language transformer models and N-gram based features. Finally, the features are fed into machine learning models for training and evaluation. Experimental results on a large dataset of YouTube comments show that the DeepGram algorithm achieves high accuracy and robust performance in detecting spam comments. The proposed algorithm can be potentially employed as an effective tool for YouTube content creators and platform moderators to combat spam comments and improve the quality of user interactions on YouTube videos.

Forecasting US Stock Prices Through Sentiment Analysis and Machine LearningA Case Study of Tesla Inc.

The financial sector is characterized by high volatility, and the accurate forecasting of stock prices is an actively pursued area of research and analysis. This study extends the scope of machine learning techniques, such as Artificial Neural Networks and fuzzy-based techniques, to enhance the precision of stock price predictions. The central focus of this research is algorithmic trading, which combines various qualitative factors in stock buying and selling decisions. More specifically, this study delves into the unique relationship between Elon Musks tweets and Teslas stock value. To identify patterns in the pre-processed dataset, which has had stop words removed, exploratory data analysis is used as the primary research methodology. The study conclusively demonstrates that a positive correlation exists between the number of tweets/engagements and Teslas closing price, and this correlation holds true in reverse: a decrease in tweets/engagements corresponds with a decrease in Teslas closing price. This research has broader implications for macroeconomic analysis of the US stock market by highlighting the role of technology and innovation in financial markets, as well as the importance of data-driven approaches in economic

Abstract 15133: Topic Models Using Notes From Electronic Medical Records Can Classify Non ST Elevation Myocardial Infarction Patients Based on the Presence of an Occluded Culprit Artery

Introduction: Approximately 33% of patients with a Non-ST-Elevation Myocardial Infarction (NSTEMI) have an occluded culprit artery which doubles their odds of adverse outcomes. Earlier identification of NSTEMI patients with occluded culprit artery may reduce their odds of adverse outcomes. Notes from the electronic health record (EHR) are rich sources of text data about signs and symptoms. Organizing text data into topics may help providers identify patients with NSTEMI and an occluded culprit artery sooner. Latent Dirichlet Allocation uses a deep-learning network to generate topics that can classify patients. Methods: Notes transcribed during the first 72 hours of hospitalization were extracted from the EHR among NSTEMI patients who presented to a medical center between 2015-2020. The data was preprocessed by removing punctuation, numbers, and stop words. We decided 5 topics based on computed coherence value. Two authors determined whether the first 100 terms were associated with an occluded artery. Analyses were completed in R using packages topicmodels and tidyverse . Results: We examined 45,025 notes from 2,223 patients (age=69 + 14 years old; 57% n=1267 male; 83% n=1,845 White) who experienced 2,685 hospitalizations (17%, n=462 NSTEMI re-hospitalizations). In Topic 1, 70% of the terms were associated with an occluded culprit artery including “chest pain” and “troponin”. In contrast, in Topic 5, 90% of terms were not associated with an occluded culprit artery such as “edema” and “pulmonary”. Figure 1 shows the frequency of terms associated with an occluded artery per topic. Conclusions: Topic 1 is most associated with an occluded culprit artery. Topic 5 is the least associated with an occluded culprit artery, instead associated with heart failure. Our results demonstrate it is feasible to develop topics to differentiate NSTEMI with and without an occluded culprit artery. Future research aims to identify patterns between signs and symptoms within topics.

Unified Fake News Detection System (UFNDS) Framework

The deliberate spread of misleading or inaccurate material pose as authentic news is known as "fake news." Its increasing prevalence calls for the creation of practical strategies to recognize and counteract its negative effects on people and society. Previous methods of identifying fake news depended on linguistic signals and stylistic components. However, these methods faced limitations in terms of their applicability and accuracy. To overcome these constraints, this study proposes the utilization of an extended stacking ensemble classification algorithm (ES-ECA), a machine learning technique designed specifically for detecting fake news. By employing this innovative approach, we aim to surpass the existing barriers and enhance our ability to combat misinformation. The ensemble classifier outperformed the individual classifiers, with an accuracy of 75.18% and an F1-score of 81.81%. These findings imply that the suggested algorithm can be utilized to lessen the negative effects of fake news on society and is efficient at identifying it. The EHT-DL model leverages a multi-step approach to effectively detect fake news. It begins with preprocessing steps such as text normalization, special character handling, stemming, stop word removal, tokenization, and lemmatization. This ensures the dataset is clean and ready for subsequent processing. Feature extraction is performed using TF-IDF, N-grams, and word embeddings scores to capture semantic information and word importance. After that, the dataset is divided into training and testing sets and the deep learning model Dl4jMlpClassifier is used to classify the data. To tackle the drawbacks of existing techniques, the EHT-DL model incorporates efficient hyperparameter tuning. It uses both Grid Search and Random Search methods to optimize the Dl4jMlpClassifier's hyperparameters. By using this method, the model's accuracy and capacity to distinguish between authentic and fraudulent news are both improved. The effectiveness of the EHT-DL model is shown by the experimental findings. Standard assessment criteria including accuracy, precision, recall, and F1-score are used to assess the model. In terms of accuracy and efficiency, comparisons with current methods demonstrate the superiority of the proposed model in identifying bogus news (83.27% accuracy, 80.62% precision, 71.57% recall, and 75.63% f1-score). To increase classification accuracy and resilience, OE-MDL combines the phases of optimized deep learning (ODL) and optimized machine learning (OML). An optimized Multilayer Perceptron serves as the Meta classifier in the OML phase, on top of base classifiers such as optimized RandomForest, optimized J48, optimized SMO, optimized NaiveBayes, and optimized IBk. The experimental findings show that the OE-MDL algorithm performs better than other methods with the maximum recall (85.18%), accuracy (84.27%), precision (74.17%), and F1-Score (79.29%), providing a practical means of halting the spread of false information. The framework for the Unified Fake News Detection System (UFNDS) is revealed. The final result of the research works and demonstrates how the three main stages—Deep Learning and Optimized Ensemble Machine (OE-MDL), Efficient Hyperparameter-Tuned Deep Learning Model (EHT-DL), and Enhanced Stacking Ensemble Classification Algorithm (ES-ECA)—are cohesively incorporated into the UFNDS framework. We examine the UFNDS framework's architectural design and show how flexible it is to the always changing problems associated with fake news identification.

Harnessing Machine Learning for Unmasking Deception: An In-Depth Analysis using ML Approaches for Fake News Identification in News Media

In an era, In which misleading data may spread quickly, it is critical to have an efficient fake news detection system. This study explores the field of text-based machine learning models with the goal of separating potentially unreliable news stories from legitimate news articles in daily newspapers. The dataset that was obtained from Kaggle provides the basis for this project. It includes a number of characteristics, such as article headings, authors, textual content, and labels identifying whether an article is "Fake News" or "Real News." A methodical strategy is used that includes data preparation, feature engineering, model selection, and hyperparameter tweaking to provide the best level of accuracy. Text data is tokenized, stemmed, and stop words are eliminated before being converted to numerical features using methods like TF-IDF and word embeddings. To assess model performance, the dataset is intelligently split into training and testing sets. Logistic regression, Naive Bayes, SVM, and sophisticated deep learning models like BERT and GPT are among the machine learning models that are taken into account. To improve accuracy, the project also uses ensemble learning strategies.

Ethnic minorities' mentality and homosexuality psychology in literature: A text emotion analysis with NRC lexicon

As a significant subfield of natural language processing (NLP), text emotion analysis has been extensively researched and applied in various domains, such as media, education, and medicine. It has shown significant results in annotating blog posts that rely on an extensive corpus of short phrases. However, in interdisciplinary fields like literary pragmatics, character emotion analysis in literature becomes crucial. Despite the importance of this topic, there are fewer studies, especially for niche subjects such as ethnic minorities' mentality and homosexuality psychology. This paper examines the effectiveness of the widely used lexicon National Research Council of Canada (NRC) in detecting metaphorical words in the famous homosexual novel Maurice. To increase the accuracy of the test, we classified and cleaned the stop words using the Natural Language Toolkit (NLTK) before the analysis step. Our results indicate that the lexicon is able to demonstrate reasonable emotional changes in the story.

Using Cluster Analysis for Author Classification of Albanian Texts: A Study on the Effectiveness of Stop Words

Cluster analysis is a statistical approach that identifies uniform clusters within data. The closeness of data is measured quantitatively using distance functions. Specifically for text data mining, clustering serves as a method of categorization of words based on the similarity of their occurrence within texts and classifying texts by topics or author. Hierarchical clustering is a powerful technique for identifying natural groupings within datasets, which can be especially useful for unsupervised text classification. This paper aims to utilize cluster analysis to establish Albanian texts clusters by authors. Using agglomerative hierarchical clustering we classify Albanian texts by authors according to the similarity of their word frequency. The similarity of texts is evaluated using cosine and Euclidean distances. Considering two study cases, respectively with and without Albanian stop words we conclude that the best clustering by authors of the Albanian documents is achieved with 87% accuracy using Ward’s method with cosine distance in the case of study by removing stop words.

Research on performance variations of classifiers with the influence of pre-processing methods for Chinese short text classification.

Text pre-processing is an important component of a Chinese text classification. At present, however, most of the studies on this topic focus on exploring the influence of preprocessing methods on a few text classification algorithms using English text. In this paper we experimentally compared fifteen commonly used classifiers on two Chinese datasets using three widely used Chinese preprocessing methods that include word segmentation, Chinese specific stop word removal, and Chinese specific symbol removal. We then explored the influence of the preprocessing methods on the final classifications according to various conditions such as classification evaluation, combination style, and classifier selection. Finally, we conducted a battery of various additional experiments, and found that most of the classifiers improved in performance after proper preprocessing was applied. Our general conclusion is that the systematic use of preprocessing methods can have a positive impact on the classification of Chinese short text, using classification evaluation such as macro-F1, combination of preprocessing methods such as word segmentation, Chinese specific stop word and symbol removal, and classifier selection such as machine and deep learning models. We find that the best macro-f1s for categorizing text for the two datasets are 92.13% and 91.99%, which represent improvements of 0.3% and 2%, respectively over the compared baselines.

Enhanced Spam Detection System for Twitter Social Networking Platform

Twitter social site is one of the most popular Online Social Networking Site (OSN) used by popular people such as Ministers, businessman, large companies, actors to share their information. In this site, around 500 million of tweets are posted monthly by the total 313 million Twitter active users. The widespread of Twitter has drawn the interest of spammers. These malicious actors exploit the platform for various nefarious purposes, including monitoring authentic users, disseminating harmful software, and promoting their agendas through URLs embedded in tweets. They engage in tactics like secret following and unfollowing legitimate users, all with the intent of gathering sensitive information.To resolve this problem, a secure spam detection based on machine learning approach is designed. The designed used stop word removal, word to vector model to refined and dimensionally reduced the data. To enhance the quality of the data Cosine similarity is also been applied to measure the similarity score among the tweets and based upon that Artificial Neural Network (ANN) is trained. Later on, it is used to test the efficiency by examining the performance parameters in terms of precision, recall and F-measure. Also, the comparative analysis has been performed to present the efficiency of the work. The average precision, recall and F measure of proposed spam detection model of 0.9252, 0.6107 and 0.734 are obtained.

LDA Topic Analysis for Product Reviews in Social Media Platform

Social media in recent years is used as platform for product reviews and customer feedback. Thus, to understand the topic which have been discussed, we utilized Latent Dirichlet Allocation for topic modeling. The topic modeling is important due to it can gain insights into the specific features that customers like or dislike about a particular product. Thus, by not using stop words due it have possibilities remove the time domain, the information can be valuable for businesses as it helps them understand customer preferences and inform product development and marketing strategies with the coherence score 0.621520, the topic modeling obtained 3 optimal topics, where the topic 0 discussed about price and time it will be available to purchase. In topic 1 it discussed about the product is hard to obtain due to it not available in market. In topic 2, it discussed about ownership (what they like after usage).

Emotion detection on social media status in Myanmar language

<span lang="EN-US">Many social media emerged and provided services during these years. Most people, especially in Myanmar, use them to express their emotions or moods, learn subjects, sell products, read up-to-date news, and communicate with each other. Emotion detection on social users makes critical tasks in the opinion mining and sentiment analysis. This paper presents the emotion detection system on social media (Facebook) user status or post written in Myanmar (Burmese) language. Before the emotion detection process, the user posts are pre-processed under segmentation, stemming, part-of-speech (POS) tagging, and stop word removal. The system then uses our preconstructed Myanmar word-emotion Lexicon, M-Lexicon, to extract the emotion words from the segmented POS post. The system provides six types of emotion such as surprise, disgust, fear, anger, sadness, and happiness. The system applies naïve Bayes (NB) emotion classifier to examine the emotion in the case of more than two words with different emotion values are extracted. The classifiers also classify the emotion of the users on their posts. The experiment shows that the system can detect 85% accuracy in NB based emotion detection while 86% in recurrent neural network (RNN).</span>

ALGORITHM AND SOFTWARE TO ASSESSS THE COMPLIANCE OF BUSINESS PROCESS MODELS WITH THEIR TEXTUAL DESCRIPTIONS

This paper is devoted to solve the problem of analyzing the compliance of business process models with their textual descriptions. The problem being business process models describe re-designed or completely new organizational activities, but “wrong” models that do not reflect correctly business process requirements may mislead involved business process participants and other stakeholders, and cause workflow errors followed by extra costs. Therefore, the research goal is to ensure the correctness of business process models by analyzing their compliance with textual descriptions formulated by business process owners or business analysts. In the work, a review of existing tools for modeling and analysis of business processes is outlined, as well as the main technologies of natural language processing are considered, including tokenization, search for stop words, and stemming. These technologies are proposed to be used to analyze the compliance of business process models with their textual descriptions. An approach to solving the problem of analyzing the compliance of business process models with their textual descriptions, using the selected natural language processing tools is proposed and the respective algorithm is developed. The process of analyzing the compliance of business process models with their textual descriptions is formalized using data flow modeling. The corresponding software that implements this process is developed. Sample calculations are demonstrated that confirm the performance of the proposed approach by analyzing the model of the goods dispatch business process and the corresponding textual description of this business process. Finally, conclusions are given and the directions for further work are determined.

Classification of Severe Maternal Morbidity from Electronic Health Records Written in Spanish Using Natural Language Processing

One stepping stone for reducing the maternal mortality is to identify severe maternal morbidity (SMM) using Electronic Health Records (EHRs). We aim to develop a pipeline to represent and classify the unstructured text of maternal progress notes in eight classes according to the silver labels defined by the ICD-10 codes associated with SMM. We preprocessed the text, removing protected health information (PHI) and reducing stop words. We built different pipelines to classify the SMM by the combination of six word-embeddings schemes, three different approaches for the representation of the documents (average, clustering, and principal component analysis), and five well-known machine learning classifiers. Additionally, we implemented an algorithm for typos and misspelling adjustment based on the Levenshtein distance to the Spanish Billion Word Corpus dictionary. We analyzed 43,529 documents constructed by an average of 4.15 progress notes from 22,937 patients. The pipeline with the best performance was the one that included Word2Vec, typos and spelling adjustment, document representation by PCA, and an SVM classifier. We found that it is possible to identify conditions such as miscarriage complication or hypertensive disorders from clinical notes written in Spanish, with a true positive rate higher than 0.85. This is the first approach to classify SMM from the unstructured text contained in the maternal EHRs, which can contribute to the solution of one of the most important public health problems in the world. Future works must test other representation and classification approaches to detect the risk of SMM.

A scoping review of topic modelling on online data

With <span>the increasing prevalence of unstructured online data generated (e.g., social media, online forums), mining them is important since they provide a genuine viewpoint of the public. Due to this significant advantage, topic modelling has become more important than ever. Topic modelling is a natural language processing (NLP) technique that mainly reveals relevant topics hidden in text corpora. This paper aims to review recent research trends in topic modelling and state-of-the-art techniques used when dealing with online data. Preferred reporting items for systematic reviews and meta-analysis (PRISMA) methodology was used in this scoping review. This study was conducted on recent research works published from 2020 to 2022. We constructed 5 research questions for the interest of many researchers. 36 relevant papers revealed that more work on non-English languages is needed, common pre-processing techniques were applied to all datasets regardless of language e.g., stop word removal; latent dirichlet allocation (LDA) is the most used modelling technique and also one of the best performing; and the produced result is most evaluated using topic coherence. In conclusion, topic modelling has largely benefited from LDA, thus, it is interesting to see if this trend continues in the future across languages.</span>

HINDI FAKE NEWS DETECTOR

Fake news has proliferated on the internet in recent decades. More people than ever before are creating and sharing knowledge because of social networks, many of which have no connection to reality. This has led to the rapid dissemination of false information used for various political and business objectives. Finding reliable news sources has become more difficult due to online newspapers. In this work, we gathered news articles of Hindi text from various news sources. Techniques for pre-processing, feature extraction, classification, and prediction are all extensively covered. A “Fake news detection system” has been developed in this project. Various Hindi news articles have been collected from multiple sources to help diversify the dataset and train the model better. The project first pre-processes the dataset and uses the pre-trained Bert model for feature extraction. Then, the data is classified from the dataset and prediction processes are employed on the dataset. Various machine learning algorithms and deep learning models like Naïve Bayes, Long Short-Term Memory, Logistic Regression have been employed in previous works for the purpose of detecting fake news. Pre-processing steps include data cleaning, stop words removal, tokenizing, stemming. The testing and training of the dataset include using the BERT for sequence classification model. The model is trained and tested against the validation dataset

Integrating Text Classification into Topic Discovery Using Semantic Embedding Models

Topic discovery involves identifying the main ideas within large volumes of textual data. It indicates recurring topics in documents, providing an overview of the text. Current topic discovery models receive the text, with or without pre-processing, including stop word removal, text cleaning, and normalization (lowercase conversion). A topic discovery process that receives general domain text with or without processing generates general topics. General topics do not offer detailed overviews of the input text, and manual text categorization is tedious and time-consuming. Extracting topics from text with an automatic classification task is necessary to generate specific topics enriched with top words that maintain semantic relationships among them. Therefore, this paper presents an approach that integrates text classification for topic discovery from large amounts of English textual data, such as 20-Newsgroups and Reuters Corpora. We rely on integrating automatic text classification before the topic discovery process to obtain specific topics for each class with relevant semantic relationships between top words. Text classification performs a word analysis that makes up a document to decide what class or category to identify; then, the proposed integration provides latent and specific topics depicted by top words with high coherence from each obtained class. Text classification accomplishes this with a convolutional neural network (CNN), incorporating an embedding model based on semantic relationships. Topic discovery over categorized text is realized with latent Dirichlet analysis (LDA), probabilistic latent semantic analysis (PLSA), and latent semantic analysis (LSA) algorithms. An evaluation process for topic discovery over categorized text was performed based on the normalized topic coherence metric. The 20-Newsgroups corpus was classified, and twenty topics with the ten top words were identified for each class. The normalized topic coherence obtained was 0.1723 with LDA, 0.1622 with LSA, and 0.1716 with PLSA. The Reuters Corpus was also classified, and twenty and fifty topics were identified. A normalized topic coherence of 0.1441 was achieved when applying the LDA algorithm, obtaining 20 topics for each class; with LSA, the coherence was 0.1360, and with PLSA, it was 0.1436.

Arabic ChatGPT Tweets Classification Using RoBERTa and BERT Ensemble Model

ChatGPT OpenAI, a large-language chatbot model, has gained a lot of attention due to its popularity and impressive performance in many natural language processing tasks. ChatGPT produces superior answers to a wide range of real-world human questions and generates human-like text. The new OpenAI ChatGPT technology may have some strengths and weaknesses at this early stage. Users have reported early opinions about the ChatGPT features, and their feedback is essential to recognize and fix its shortcomings and issues. This study uses the ChatGPT tweets Arabic dataset to automatically find user opinions and sentiments about ChatGPT technology. The dataset is preprocessed and labeled using the TextBlob Arabic Python library into positive, negative, and neutral tweets. Despite extensive works for the English language, languages like Arabic are less studied regarding tweet analysis. Existing literature about Arabic tweet sentiment analysis has mainly focused on machine learning and deep learning models. We collected a total of 27,780 unstructured tweets from Twitter using the Tweepy SNscrape Python library using various hash-tags such as # Chat-GPT, #OpenAI, #Chatbot, Chat-GPT3, and so on. To enhance the model’s performance and reduce computational complexity, unstructured tweets are converted into structured and normalized forms. Tweets contain missing values, URL and HTML tags, stop words, punctuation, diacritics, elongations, and numeric values that have no impact on the model performance; hence, these increase the computational cost. So, these steps are removed with the help of Python preprocessing libraries to enhance text quality and consistency. This study adopts Transformer-based models such as RoBERTa, XLNet, and DistilBERT that automatically classify the tweets. Additionally, a hybrid transformer-based model is proposed to obtain better results. The proposed hybrid model is developed by combining the hidden outputs of the RoBERTA and BERT models using a concatenation layer, then adding dense layers with “Relu” activation employed as a hidden layer to create non-linearity and a “softmax” activation function for multiclass classification. They differ from existing state-of-the-art models due to the enhanced capabilities of both models in text classification. Hybrid models combine the different models to make accurate predictions and reduce bias and enhanced the overall results, while state-of-the-art models are incapable of making accurate predictions. Experiments show that the proposed hybrid model achieves 96.02% accuracy, 100% precision on negative tweets, and 99% recall for neutral tweets. The performance of the proposed model is far better than existing state-of-the-art models.

The Contribution of Selected Linguistic Markers for Unsupervised Arabic Verb Sense Disambiguation

Word sense disambiguation (WSD) is the task of automatically determining the meaning of a polysemous word in a specific context. Word sense induction is the unsupervised clustering of word usages in a different context to distinguish senses and perform unsupervised WSD. Most studies consider function words as stop words and delete them in the pre-processing step. However, function words can encode meaningful information that can help to improve the performance of WSD approaches. We propose in this work a novel approach to solve Arabic verb sense disambiguation that is based on a preposition-based classification that is used in an automatic word sense induction step to build sense inventories to disambiguate Arabic verbs. However, in the wake of the success of neural language models, recent works obtained encouraging results using BERT pre-trained models for English-language WSD approaches. Hence, we use contextualized word embeddings for an unsupervised Arabic WSD that is based on linguistic markers and uses sentence-BERT Transformer pre-trained models, which yields encouraging results that outperform other existing unsupervised neural AWSD approaches.

DZ-SMS: An Authentic Corpus of Algerian SMS

In this article, a complete methodology of a corpus realization of authentic Short Message Service (SMS) from Algerian dialect and which are transcribed in Latin characters or symbols is presented. A linguistic material constituted by 6,000 SMS coming from the different geographical regions of Algeria (Middle, East, and West) corresponding to 42 administrative and geographical departments, have been collected. The coexistence of several dialects through these three regions simultaneously has obliged us to consider and operate a classification of the data for each dialect. This data classification has yielded three extracted regional dialectic corpora, each of them covering a specific number of administrative departments. These treatments are based on the so-called Data-n-gram tokenization targeting the suppression of the stop words, the stemming and the imbalance of the classes linked to the nature of the SMS. Consequently, three text classifiers based on three linear classifiers, namely, Stochastic Gradient Descent (SGD), The Ridge Regression (RDG), and Linear Support Vector Machines, to find out the number of significant corpora to extract from the collected data. A deep analysis of the results has shown that the 5-grams data representation is more representative whereas the stop-words removal and stemming process has generated an information loss that has subsequently inferred an alteration of the recognition rate of about 2%. The emerging problem of classes imbalance has been treated by using three techniques: Random Oversampling, Synthetic Minorities Oversampling Technique (SMOTE), and Adaptive Synthetic (ADASYN). This treatment produced interesting results and enhancements; particularly, the classification by region with the oversampling process SMOTE by using the RDG technique has reached a better percentage of 55.93% whereas the classification by department with the oversampling process ADASYN associated with the SGD has only yielded a maximum score of about 17.11%. The results, which undoubtedly are in favor of the classification by region, have compelled us to create three Subdialectal regional corpora, each, covering a certain number of Algerian departments.

Will the Relaxation of COVID-19 Control Measures Have an Impact on the Chinese Internet-Using Public? Social Media-Based Topic and Sentiment Analysis.

Objective: In December 2022, the Chinese government announced the further optimization of the implementation of the prevention and control measures of COVID-19. We aimed to assess internet-using public expression and sentiment toward COVID-19 in the relaxation of control measures in China. Methods: We used a user-simulation-like web crawler to collect raw data from Sina-Weibo and then processed the raw data, including the removal of punctuation, stop words, and text segmentation. After performing the above processes, we analyzed the data in two aspects. Firstly, we used the Latent Dirichlet Allocation (LDA) model to analyze the text data and extract the theme. After that, we used sentiment analysis to reveal the sentiment trend and the geographical spatial sentiment distribution. Results: A total of five topics were extracted according to the LDA model, namely, Complete liberalization, Resource supply, Symptom, Knowledge, and Emotional Outlet. Furthermore, sentiment analysis indicates that while the percentages of positive and negative microblogs fluctuate over time, the overall quantity of positive microblogs exceeds that of negative ones. Meanwhile, the geographical dispersion of public sentiment on internet usage exhibits significant regional variations and is subject to multifarious factors such as economic conditions and demographic characteristics. Conclusion: In the face of the relaxation of COVID-19 control measures, although concerns arise among people, they continue to encourage and support each other.