Claim Missing Document
Check
Articles

Found 2 Documents
Search
Journal : Building of Informatics, Technology and Science

Deteksi Komentar dan Analisis Sentimen Promosi Judi Online pada Youtube Menggunakan IndoBERT dan XGBoost Putri, Naila Raihana; Kurniawan, Dedy; Tania, Ken Ditha
Building of Informatics, Technology and Science (BITS) Vol 7 No 3 (2025): December 2025
Publisher : Forum Kerjasama Pendidikan Tinggi

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.47065/bits.v7i3.8421

Abstract

YouTube, as a highly interactive platform, has become a medium for online gambling promotions, raising legal issues under the Electronic Information and Transactions (ITE) Law and social risks, particularly for adolescents. This study aims to analyse public responses to gambling-related comments and to develop an automatic detection system using Natural Language Processing (NLP). The research follows the Knowledge Discovery in Databases (KDD) stages, including web scraping, preprocessing, text transformation, model training, and evaluation. Sentiment analysis was performed on 999 comments labelled positive, negative, and neutral. Detection of promotional content was tested using IndoBERT and TF-IDF-based XGBoost, with 587 training samples and 885 external testing samples at an 80:20 ratio. The results show that the majority of comments (52.65%) are positive with a fairly high average confidence score (0.914), indicating public support for the eradication of online gambling. Meanwhile, negative comments (24.72%) with a confidence score of 0.888 generally contained criticism of the rampant practice of gambling promotion or YouTube's weak moderation system. For automatic detection, IndoBERT achieved superior performance with 0.94 accuracy and F1-score and only 10 misclassifications, significantly outperforming XGBoost, which reached 0.73 accuracy with 47 errors. This study highlights the effectiveness of transformer-based models in detecting gambling promotions while also indicating strong public support for eradication efforts. These findings provide an empirical foundation for advancing research on adaptive automated moderation systems capable of identifying concealed patterns of illicit content in digital platforms, particularly in the detection of online gambling promotional comments within the YouTube ecosystem.
Comparison of XGBoost and LSTM in Knowledge Discovery for GrokAI Mobile Application Sentiment Analysis Risyahputri, Aliyananda; Kurniawan, Dedy; Tania, Ken Ditha
Building of Informatics, Technology and Science (BITS) Vol 7 No 3 (2025): December 2025
Publisher : Forum Kerjasama Pendidikan Tinggi

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.47065/bits.v7i3.8651

Abstract

Generative AI has provided real benefits in key sectors of the public sector. However, the rapid expansion of AI assistant services also raises concerns about whether newly released products can consistently meet user expectations, especially as negative experiences are increasingly expressed through public reviews. Its positive impacts encourage competitive rivalry among AI assistant product developers, including xAI, which also participates by formulating the Grok AI application. As a relatively new product with over 50 million downloads, GrokAI needs to perform an evaluation to maintain its competitiveness. This condition leads to the research goal of analyzing user sentiment toward GrokAI application through reviews on Google Play Store and comparing the performance of Machine Learning and Deep Learning classification models within the framework of Knowledge Discovery in Databases (KDD). This study uses 11,108 review data classified using the VADER Lexicon method, resulting in 7,633 positive reviews and 3,475 negative reviews. The data is then tested on XGBoost (Extreme Gradient Boosting) and LSTM (Long-Short Term Memory) models. The results show that the XGBoost model performs slightly better with an accuracy of 87.22%, compared to LSTM, which reaches 86.58%. However, both models exhibit significant performance disparities in classifying negative classes due to the extreme difference in data quantity. The knowledge discovery process reveals that the majority of positive sentiment appreciates the free access and general functions of the application. Meanwhile, negative sentiment focuses on complaints related to response time, output quality, and specific features such as image and voice. The main recommendation is to maintain the advantage of free access also improve features and processing logic to sustain loyalty and service quality. Future research is suggested to test models with more balanced data and optimize dataset cleaning to improve accuracy in minority classes.