A Robustly Optimized BERT using Random Oversampling for Analyzing Imbalanced Stock News Sentiment Data

Abstract—This paper presents a hybrid sentiment analysis system that integrates lexicon-based, traditional machine learning, and deep learning techniques to classify textual data into positive, negative, or neutral sentiments. The system leverages Python-based libraries such as Scikit-learn, NLTK, and Transformers to pre- process text, extract features, and apply models in- cluding Support Vector Machine (SVM), Bidirectional Encoder Representations from Transformers (BERT), Linear Regression, and Valence Aware Dictionary and sEntiment Reasoner (VADER). The framework pro- cesses customer reviews from e-commerce platforms (Amazon, Flipkart) and social media (Instagram) using web scraping and provides actionable insights through sentiment summarization and visualizations (bar and pie charts). Experimental results demonstrate BERT’s superior performance with 92.3% accuracy, followed by SVM (85.6%), Linear Regression (81.2%), and VADER (76.8%). The system addresses challenges like sarcasm, class imbalance, and scalability, offering a scalable, user-friendly solution for real-world applications in e- commerce, social media analytics, and brand reputation management. Index Terms—Sentiment Analysis, Machine Learning, Deep Learning, Natural Language Processing, BERT, SVM, VADER, Text Classification, Data Visualization.

The rapid growth of financial technology (fintech) services in Indonesia has significantly transformed digital transaction behavior, with digital wallets such as DANA becoming widely adopted. Despite high usage rates, public discourse on social media frequently highlights both positive experiences and recurring concerns related to system reliability and data security. Twitter, as a real-time public communication platform, provides a rich source of user-generated content that reflects public perception toward fintech services. This study investigates public sentiment toward DANA fintech services in Indonesia through sentiment analysis of Twitter data using machine learning approaches. Purpose:This study aims to analyze public perception of DANA fintech services based on sentiment expressed on Twitter and to compare the performance of several machine learning classification algorithms in identifying positive, neutral, and negative sentiments in Indonesian-language social media texts. Methods/Study design/approach: The study employs a quantitative descriptive approach using text mining and sentiment analysis techniques. Twitter data were collected via the Twitter API over a specified period and processed through preprocessing stages including text cleaning, tokenization, stopword removal, and stemming. Feature extraction was performed using the Term Frequency–Inverse Document Frequency (TF-IDF) method. Sentiment classification was conducted using Support Vector Machine (SVM), K-Nearest Neighbor (KNN), and Random Forest algorithms. Model performance was evaluated using accuracy, precision, recall, and F1-score metrics. Result/Findings:The results indicate that public sentiment toward DANA is predominantly positive, driven by perceptions of transaction convenience and speed, followed by neutral sentiment, while negative sentiment is mainly associated with system disruptions and data security issues. Among the evaluated algorithms, SVM achieved the highest classification accuracy (91.4%), outperforming Random Forest and KNN. These findings demonstrate that SVM is more effective for sentiment classification of short Indonesian-language texts on social media platforms. Novelty/Originality/Value:This study contributes to the growing body of sentiment analysis research in Indonesia by providing empirical evidence on public perception of fintech services using machine learning techniques. By systematically comparing multiple classification algorithms in the context of Indonesian Twitter data, the study offers methodological insights and practical implications for fintech service providers in improving service quality, communication strategies, and user trust.

With the rapid growth of digital technology, social media has become a key platform for sharing information and opinions. Twitter, one of the most popular platforms in Indonesia, enables users to interact directly with public figures such as Najwa Shihab. This study aims to analyze public sentiment toward Najwa Shihab on Twitter using sentiment analysis, specifically employing the Naïve Bayes and Support Vector Machine (SVM) algorithms. Sentiment analysis is essential to understanding public opinion, as it classifies text into categories like positive, negative, or neutral, providing valuable insights into societal perspectives on public figures. In this study, 10,000 tweets related to Najwa Shihab were collected from January 1, 2023, to January 31, 2023. Data preprocessing steps such as data cleaning, tokenization, stopwords removal, and filtering were conducted to ensure high-quality data for analysis. The Naïve Bayes and SVM algorithms were applied using RapidMiner to classify the sentiment of the tweets. The performance of both algorithms was evaluated based on accuracy, precision, recall, and F1-score.The results revealed that SVM outperformed Naïve Bayes in all metrics, demonstrating its superior ability to classify sentiments correctly. The sentiment distribution indicated a majority of positive opinions toward Najwa Shihab, with fluctuations in negative sentiment during specific events. This study provides insights into public sentiment analysis and contributes to understanding social media opinions on public figures.

Sentiment analysis has become increasingly important in understanding user perceptions of digital platforms. This study focuses on analyzing TikTok application reviews from the Google Play Store in Indonesia using machine learning techniques. The research aims to investigate sentiment distribution and compare the performance of three popular machine learning models: Random Forest, Support Vector Machine (SVM), and Naive Bayes. The study employed a comprehensive methodology involving data collection, preprocessing, feature extraction, and model evaluation. A dataset of 10,000 TikTok reviews was collected and preprocessed using techniques such as case folding, tokenization, and stopword removal. The sentiment labeling process categorizes reviews into positive, negative, and neutral sentiments based on user ratings. The TF-IDF algorithm was used for feature extraction, and the SMOTE technique addressed class imbalance. Results revealed a predominance of negative sentiment (53.5%), followed by neutral (32.1%) and positive (14.4%) sentiment. Model performance comparisons at different data sharing ratios (80/20 and 70/30) demonstrated that Random Forest and SVM consistently outperformed Naive Bayes. At the 80/20 ratio, Random Forest achieved the highest accuracy of 83.73%, highlighting its effectiveness in sentiment classification. The research contributes to the field of sentiment analysis and natural language processing by providing insights into user experiences with the TikTok application in Indonesia. The findings can guide application developers in understanding user perceptions and improving user experience.

Sentiment analysis has become an important approach in understanding user opinions towards digital applications, especially on review platforms such as the Google Play Store. This study aims to compare the performance of three popular classification algorithms, namely K-Nearest Neighbor (KNN), Support Vector Machine (SVM), and Naive Bayes, in classifying sentiment in Traveloka application reviews. The methods used include collecting user review data from Google Play, preprocessing stages such as tokenization and stopword removal, and applying the three algorithms to the cleaned data. The evaluation was carried out using accuracy, precision, recall, and f1-score metrics. The dataset used consisted of 5000 review data that were evenly divided between positive and negative sentiments. The results showed that the SVM algorithm provided the best performance with an accuracy of 88%, followed by Naive Bayes at 86%, and KNN at 87%. The conclusion of this study states that SVM is more reliable in handling text-based sentiment analysis in the context of mobile application reviews. These findings are expected to be a reference in developing a more accurate sentiment analysis system for business needs and future research.

While analyzing health data is important for improving health outcomes, class imbalance in datasets poses major challenges to machine learning classification models. This work, therefore, considers the class imbalance problem in stroke prediction using models such as K‐nearest neighbors, support vector machine, logistic regression, random forest, and decision tree. This work balances the stroke dataset, thereby enhancing model performance, through various oversampling strategies: random oversampling (RO), ADASYN, SMOTE, and SMOTE–Tomek. Compared to the results of the imbalanced dataset, all applied oversampling techniques enhanced the correct classification of stroke events by the ML model. Among these, RO–SVM with RBF kernel was the best in terms of sensitivity, specificity, G‐mean, F1‐score, and accuracy values, offering the highest results with respective values of 89.87%, 94.91%, 92.36%, 89.64%, and 89.87%. After applying oversampling techniques, all the machine learning classifications were good enough to classify stroke status, especially for the minority class. This study has highlighted the importance of class imbalance issues in health datasets. Precise detection of instances of minority classes can be enhanced considerably by employing classification models with the implementation of hybrid strategies to effectively solve class imbalance issues, which, in turn, will help improve healthcare outcomes. Further research in integrating more advanced deep learning techniques into other health datasets with imbalances is encouraged to further validate or refine class imbalance approaches, as effective handling of imbalanced classes can substantially promote predictive model performance in the analysis of healthcare.

Covid-19 pandemic urges countries to limit interaction of their people to reduce transmission. Indonesia requires people to do activities at home, one of which is online school. Many people share their thoughts through social media Twitter. Therefore, authors conducted sentiment analysis using supervised machine learning algorithm to determine distribution of words used in commenting on online schools, relationship between sentence, length and sentiment, and best algorithms that can be used to get most accurate results. In this study, authors used the method of crawling with RapidMiner to get data from Twitter. Then authors do data cleansing, data processing with classification methods using Random Forest Classifier , Logistic Regression , BernoulliNB and SVC algorithm. After that authors evaluate using confusion matrix, accuracy rate and classification report. In this research, authors found there are positive, negative, and neutral sentiments expressed on the online school implementation through comments. Authors ranked top three most used words used to express positive sentiments which includes bahagia, rajin and senang. On negative sentiments, top three words are capek, muak and bosen. On neutral sentiments, top three words are tidur, capek, and buka. Lengthy Tweets are usually imbued with negative remarks. On the other hand, the tweet tends to be positive and neutral tweet is usually stable. Authors conclude that the weakness of online school is the amount of workload that makes students tired alongside ineffective teaching method which makes it hard for students to understand the material given by school. However, on the positive side, some people agree with policies that are implemented and they feel like they gained some benefits from the implementation. From the four supervised machine learning algorithms that have been tested, Logistic Regression shows the highest accuracy, 0,87. The analysis shows that society tends to be neutral to the implementation of online school.

The Google Maps platform provides a platform for visitors to express their opinions through reviews. This study aims to compare the performance of the Support Vector Machine and XGBoost algorithms in sentiment analysis of Baturraden Tourism Park visitor reviews. Data were collected using scraping techniques and obtained 4,096 reviews. After going through preprocessing stages including cleaning, tokenization, normalization, stopword removal, and stemming, the data used in the analysis process amounted to 2,912 reviews. Word weighting was carried out using the TF-IDF method, and the SMOTE technique was applied to address class imbalance. The results showed that the Support Vector Machine algorithm performed better than XGBoost with an accuracy rate of 94.52% before SMOTE and 94.86% after SMOTE, while XGBoost obtained an accuracy of 92.80% before SMOTE and 93.15% after SMOTE. These findings indicate that the Support Vector Machine is more effective in classifying positive and negative sentiments. This study is expected to contribute to the application of machine learning methods to understand visitor opinions on the Google Maps platform. Especially in the context of tourist attractions.

The Pintu app is a cryptocurrency investment platform in Indonesia designed to buy, sell, store and send cryptocurrencies easily. The app targets the younger generation and retail investors through its ease of use and user-friendly features. This study compares the performance of two machine learning methods, Naïve Bayes and Support Vector Machine (SVM), in sentiment analysis of Pintu app user reviews. 10,000 user review data were collected through web scraping and labelled as positive or negative sentiment. The preprocessing stage includes Case Folding, Stopword Removal, Tokenizing, and Stemming, followed by TF-IDF transformation for data vectorisation. Data imbalance is addressed using the SMOTE technique, and model validation is performed using K-Fold Cross Validation. The results show that SVM has superior performance compared to Naïve Bayes. SVM achieved 92.55% Accuracy, 93.20% Recall, and 95.06% F1-score, while Naïve Bayes recorded 91.50% Accuracy, 91.07% Recall, and 94.28% F1-score. However, Naïve Bayes excelled in Precision, with a value of 97.71% compared to SVM at 97%. Overall, SVM showed an optimal balance between Accuracy, Recall, and F1-score, making it the more effective method for sentiment analysis of Pintu app user reviews.

Enhanced Corporate Bond Yield Modelling Incorporating Macroeconomic News Sentiment

Capital city in every country has an important role. One of them is for the central of government. A capital city is selected by considering many aspects. These include location which is in the middle of the country, and history which was happened before. On the middle of 2019, Indonesian's President announced that Indonesian's capital city will be shifted from Jakarta to East Kalimantan. The statement makes a lot of public opinions especially in social media Twitter. The opinions can be used for sentiment analysis. This research aims to classify the information of public sentiments about Indonesian's capital city changes. The classification divided by two class which are positive sentiments and negative sentiments. The classification method that used in this sentiment analysis research is Support Vector Machine (SVM) with three function method, which are linear function, Radial Basis Function (RBF), and sigmoid function. The process include text mining, text preprocessing, analyze using SVM, data classification, and visualization. From the result of classification, linear function is the best function on this research with the accuracy value is 96,93% with 97,81% specificity value and 95% sensitivity value.

This study aims to analyze the sentiment of the Hindu community toward religious issues on social media, with a specific focus on Kapuas Regency, Central Kalimantan. The development of social media as a digital public sphere has positioned platforms such as Instagram and Twitter as primary spaces for the spontaneous and open expression of opinions, perceptions, and religious attitudes. This study is important because religious issues in digital spaces often influence social harmony and shape religious discourse within society. The research employs a quantitative approach using text mining and sentiment analysis based on Natural Language Processing (NLP). Primary data were collected through web scraping techniques, utilizing Apify for Instagram and asynchronous Python scripts for Twitter, with relevant keywords, hashtags, and geographic indicators. The analysis process includes text preprocessing (cleaning, tokenization, stopword removal, and stemming), followed by sentiment classification using a lexicon-based approach with the InSet dictionary into three categories: positive, negative, and neutral. The analysis results were evaluated using a confusion matrix, along with precision, recall, and F1-score metrics to assess model reliability. The findings indicate that positive sentiment predominates on both Instagram and Twitter, followed by neutral sentiment, while negative sentiment appears in only a small proportion. Positive sentiment is generally associated with expressions of prayer, gratitude, tolerance, and calls for togetherness, whereas negative sentiment tends to emerge in discussions related to ritual differences or responses to socio-religious controversies. The sentiment analysis model achieved an accuracy of 100% on Instagram data (self-evaluation) and 74.4% on Twitter data (manual evaluation), with relatively high precision and recall values, indicating that the results are statistically reliable.

The sentiment analysis of online product review has been a pressing concern in perception and support of sound decision-making using information provided by e-commerce portals. Through a random forest (RF) classifier and a sophisticated preprocessor pipeline, this paper proposes a useful sentiment classification model of Amazon product reviews. The data collection is made in the beginning of the research process with Amazon Reviews (2018). It is pre-processed in several stages, such as special character removal, stop-word removal, tokenization and part-of-speech (POS) tagging to normalize and standardize the textual data. TF-IDF is an extraction tool which converts text into meaningful numerical processes. RF model is then trained to predict customer sentiment and its ensemble structure is employed to overfitting and maximize predictive stability. The measures of performance in the model include accuracy, precision, recall, F1-score and loss. RF classifier is an excellent accuracy (ACC) rate of 98.83, a high precision (PRE) rate of 97.70, a high recall (REC) rate of 98.64, and a balanced F1-score (F1) rate of 98.17 compared to other traditional and deep learning classifiers such as the Logistic Regression, SVM, NB-SVM, and CNN-RNN. Overall, the RF-based sentiment analysis model is a highly trustworthy, scalable and efficient way of obtaining insights using a huge volume of e-commerce reviews data.

Stock news plays an important role on market considering the timely dissemination of news will likely to affect a stock's price changes. Classifying stock news is very helpful for investors to make investment decisions. In this paper, we design and implement a structured data based classification model, which could be effectively applied to the field of stock news classification problem. The model firstly calculates the news correlation by using structured data, and then makes use of the parameter fitting method along with the output value of softmax regression model, after that, the final probability sequence is presented. The multi-label classification is realized by the method of selecting multi-labels. Finally, we compare the novel algorithm with a general algorithm, verifying the availability and accuracy of our classification method.

This study presents a comparative analysis of Support Vector Machine (SVM) and Naïve Bayes algorithms for sentiment analysis of TikTokShop application user reviews. As TikTokShop emerges as an innovative platform integrating social media with e-commerce, understanding user sentiments becomes crucial for both consumers and businesses. A balanced dataset of 3,000 user reviews (1,000 positive, 1,000 neutral, and 1,000 negative) was collected through web scraping from Google Play Store. Following comprehensive preprocessing including cleansing, case folding, normalization, tokenization, stopword removal, and stemming, the data was vectorized using TF-IDF. Performance evaluation utilized accuracy, precision, recall, F1-score, confusion matrix, and 10-fold cross-validation. Results demonstrate that SVM consistently outperformed Naïve Bayes with higher accuracy (68.86% vs. 64.48%), precision (68.43% vs. 64.19%), and F1-score (68.58% vs. 62.46%). SVM exhibited balanced classification across all sentiment categories, while Naïve Bayes excelled at identifying negative sentiments (94.1% accuracy) but struggled significantly with neutral reviews (38.5%). Despite SVM's superior performance, Naïve Bayes demonstrated remarkable computational efficiency, with training time 224 times faster than SVM. The study reveals complementary strengths between the algorithms, suggesting potential value in ensemble approaches. These findings contribute to the understanding of sentiment analysis in video-based e-commerce platforms and provide valuable insights for businesses seeking to leverage user feedback for improved decision-making.