International Journal of Data Science (4 papers in press)

Regular Issues

• Research on Carbon Emission Prediction Method of Cement Industry based on Electricity Data  
  by Xuejun Li, Yi Zhang, Xingwei Liao, Jinlin Xie, Shu Zhang, Yuanlin Cheng, Hu Yu 
  Abstract: In response to Chinas double carbon goal, the countrys national emissions trading system has proposed to include the cement industry. This paper proposes using machine learning techniques to predict cement energy consumption carbon emissions and production process carbon emissions respectively. The paper shows that these emissions can be predicted accurately based on the electricity purchase volume of the cement industry and thewaste heat power generated. The electricity-carbon emission prediction model of cement industry is established based on the least squares optimisation support vector machine (SVM), and Bayes linear regression, K-nearest neighbour (KNN), SVM, multiple linear regression and BP neural network are used for comparison. Through example simulation, the electrical input variables are reasonably selected, and the advantages of using machine learning to predict the carbon of cement industry through electricity data are analysed. The feasibility and reliability of the proposed algorithm are verified by taking the electricity data and carbon emission data of a cement factory in Hunan province as an example.
  Keywords: Cement Carbon Emission; Machine Learning Algorithms; Electricity Data.
  DOI: 10.1504/IJDS.2025.10075580
   
  
• Monitoring and Analysis of Public Opinion in Social Networks Based on Weibo Data Mining  
  by Wang Ning, Yulin Zhou, Yutao Li, Yingcai Ouyang 
  Abstract: This study addresses fragmented, multimodal Weibo comments and the frequent neglect of non-text elements in public-opinion analysis. Using DeepSeek-related hot-search events in early 2025 as a case, we propose a framework combining LDA topic modelling with Naive Bayes sentiment classification. Comments were collected via the Weibo API and web crawlers, then cleaned and segmented. Perplexity analysis set the number of LDA topics to five, revealing key themes such as ChinaUS science and technology interaction and cultural translation/communication. A baseline Naive Bayes classifier was then enhanced by adding emoji features, sentiment-word weighting, and neutral-decision rules. Experiments show clear gains: accuracy improves from 0.86 to 0.94 and F1 from 0.81 to 0.92. Negative-sentiment precision increases from 0.48 to 0.75, and its F1 rises from 0.65 to 0.85. The framework supports multimodal emotion recognition and social-media opinion monitoring.
  Keywords: Weibo comments; LDA topic model; Naive Bayes; sentiment analysis; emoji features.
  DOI: 10.1504/IJDS.2026.10077166
   
  
• Explainable Deep Learning for Misinformation Detection under Severe Class Imbalance  
  by Alexis Lazanas, Marios Sotirios Ntaflos 
  Abstract: The expansion of deceptive and satirical fake-news content through online media is a serious concern for information credibility and requires effective automated solutions. This paper presents an evaluation of explainable deep learning techniques for news credibility classification under severe class imbalance. Three neural architectures (CNN, LSTM, and transformer-based BERT) are tested with various resampling techniques and assessed using precision, recall, and F1 score with 95% Wilson confidence intervals. The results show that resampling strategies strongly influence recurrent models, whilst transformer-based models exhibit greater robustness; notably, BERT under aggressive under-sampling improves substantially with longer fine-tuning(F1: 0.487 0.826). . To gain insight into linguistic features that contribute to credibility, model-agnostic explainability methods (LIME and SHAP) are integrated across three representative use cases. This paper highlights the need to balance prediction performance and interpretability, and calls for the development of trustworthy, transparent methods for automatic news credibility evaluation.
  Keywords: Misinformation detection; Explainable artificial intelligence; Class imbalance; Deep learning; LIME; SHAP; BERT; Resampling techniques; Model interpretability; Fake-news detection.
  DOI: 10.1504/IJDS.2026.10080232
   
  
• Exploring the Affordances of Generative Conversational AI: Insights from Social Media Discourse on ChatGPT Use  
  by Amir Karami, Mahdi M. Najafabadi, Seyed Pouyan Eslami 
  Abstract: Despite the growing interest in Generative Conversational Artificial Intelligence (GCAI), particularly among users in creative and knowledge-intensive fields, and significant investments in the sector, a gap remains in understanding its significant affordances and their prevalence from the public's perspective. Bridging this gap is essential for advancing research and practical applications tailored to these user groups. Drawing from affordance theory, this research utilizes the popular ChatGPT as a case study to methodically analyse approximately 147,000 social media posts. Our analysis identified 15 distinct affordances, with programming assistance being the most discussed and game and sport engagement the least. Additionally, affordances related to geopolitics and law engagement, social and politics engagement, and song-writing and musical composition creativity were the most likely to be reshared during the early stages of ChatGPT adoption. Our findings offer insights that can guide researchers, policymakers, industry stakeholders, and the public in making informed decisions regarding GCAI technologies.
  Keywords: Generative AI; Social Media; ChatGPT; Affordance; Text Mining.
  DOI: 10.1504/IJDS.2025.10080458