The rapid rise of Generative AI since the launch of ChatGPT in late 2022 has transformed the way people search for information, create content, solve problems, and complete professional tasks. Businesses, educational institutions, government agencies, and creative industries have increasingly embraced AI-powered tools to improve efficiency and innovation. However, the widespread adoption of these technologies has also sparked public debate regarding their social impact, ethical implications, and long-term consequences.
According to the research team from Bina Sarana Informatika University, understanding public opinion is essential for technology developers, educators, and policymakers. Sentiment analysis of social media conversations provides valuable insights into how society perceives AI technologies and helps organizations make better-informed decisions regarding AI implementation and governance.
To investigate public attitudes, the researchers collected 1,125 posts from Twitter/X discussing ChatGPT and Generative AI. Each post was categorized into one of three sentiment classes: positive, neutral, or negative. Before the analysis, the text data underwent several preprocessing steps, including converting all letters to lowercase, tokenizing sentences into words, removing irrelevant words, and reducing words to their root forms. These processes improved the quality of the dataset for machine learning classification.
The study then compared two of the most widely used machine learning algorithms for sentiment classification: Naive Bayes and Support Vector Machine (SVM). Both models were evaluated using repeated validation techniques to ensure that the performance results were reliable and not dependent on a single train-test data split.
The results clearly demonstrated that Support Vector Machine achieved an accuracy of 62.13%, while Naive Bayes reached only 52.44%. Overall, SVM outperformed Naive Bayes by 9.69 percentage points, making it the more effective algorithm for classifying public sentiment related to Generative AI adoption.
The researchers explained that SVM performed better because it is more capable of handling high-dimensional textual data containing thousands of unique words and complex linguistic patterns. In contrast, Naive Bayes assumes that each word contributes independently to a document's meaning, an assumption that often limits its ability to capture contextual relationships within social media discussions.
Beyond overall accuracy, SVM also delivered more balanced classification results. The algorithm successfully identified the vast majority of negative opinions, achieving a recall rate of 97.13%, while also improving the detection of positive sentiments compared with Naive Bayes. Both algorithms, however, struggled to classify neutral opinions accurately because neutral posts frequently contain both positive and negative expressions within the same message.
The findings reveal that public opinion surrounding Generative AI is far more nuanced than a simple division between supporters and critics. Many users acknowledged the benefits of AI tools such as ChatGPT for education and workplace productivity while simultaneously emphasizing the importance of ethical guidelines, responsible use, transparency, and appropriate government regulation.
According to Yosep Moleng and his research team, an improvement of nearly 10 percentage points in classification accuracy represents a meaningful advancement. Based on the 1,125 social media posts analyzed, SVM correctly classified approximately 109 more posts than Naive Bayes. In large-scale public opinion monitoring systems, this improvement could significantly enhance the reliability of data-driven decision-making.
The study offers practical implications across multiple sectors. Technology companies can use sentiment analysis to better understand user acceptance of AI-powered products and services. Educational institutions may rely on similar analytical approaches to develop responsible policies governing the use of Generative AI in teaching and learning. Government agencies can also utilize public sentiment monitoring to design balanced regulations that encourage technological innovation while addressing ethical and societal concerns.
The researchers noted that additional improvements remain possible. Future studies could incorporate more advanced language representation techniques such as Word2Vec, FastText, or transformer-based models like BERT to achieve higher classification accuracy. They also recommended expanding future research by collecting data from multiple online platforms—including discussion forums, news websites, and other social media services—to obtain a broader understanding of public attitudes toward Generative AI.
Overall, the study concludes that Support Vector Machine is a more reliable and effective algorithm than Naive Bayes for analyzing public sentiment regarding Generative AI adoption. Its stronger ability to process complex textual information makes it a valuable tool for organizations seeking accurate insights into public perceptions of rapidly evolving artificial intelligence technologies.
Author Profile
Yosep Moleng is a researcher at Bina Sarana Informatika University, Indonesia, specializing in artificial intelligence, machine learning, text mining, data mining, and sentiment analysis. This study was conducted in collaboration with Felix Wijaya, Manda Reksi Saputri, and Besus Maula Sulthon, who are also researchers at Bina Sarana Informatika University with expertise in artificial intelligence, data analytics, and intelligent information systems.
Research Source
Article Title: Sentiment Analysis of Generative AI Adoption: A Comparative Study of Naive Bayes and Support Vector Machine Algorithms
Authors: Yosep Moleng, Felix Wijaya, Manda Reksi Saputri, Besus Maula Sulthon
Journal: International Journal of Applied and Scientific Research (IJASR)
Volume 4, Issue 6 (2026), Pages 343–356
0 Komentar