Scholarlink

<p>The rise of e-commerce has transformed consumer behaviour, making the prediction of buying intentions critical for digital marketing strategy. This study integrates Artificial Intelligence (AI) and advanced statistical methods to examine how machine learning (ML) models and Structural Equation Modelling (SEM) can jointly predict consumer purchase intentions. Using a dataset of consumer reviews and behavioural signals from leading e-commerce platforms, we employ sentiment analysis, TF-IDF text vectorization, and predictive modelling across Logistic Regression, Random Forest, XGBoost, and Neural Networks. We also propose and test a SEM framework linking consumer trust, product quality perception, and sentiment polarity to buying intentions. Results highlight that ensemble ML models outperform linear classifiers, while SEM provides theoretical grounding by validating latent constructs. The hybrid SEM-ML framework demonstrates both predictive accuracy and theoretical rigor. Findings offer practical insights for e-commerce managers on optimizing customer engagement strategies. To improve theoretical clarity and robustness we explicitly anchor the SEM within the Technology Acceptance Model (TAM) and Theory of Planned Behaviour (TPB) and specify testable hypotheses (H1–H4). Sentiment analysis was re-run using both VADER and a transformer-based model (DistilBERT) and validated on a manually annotated subsample (n = 1000; DistilBERT accuracy = 87%, VADER = 71%). Time-aware validation (forward-chaining and rolling windows) and a robustness check using verified purchase as an alternative dependent variable were also performed. SEM reporting has been expanded to include standardized loadings, standard errors, <i>p</i>-values, Composite Reliability (CR), Average Variance Extracted (AVE), and discriminant validity. SHAP-based interpretability (summary + dependence plots) and detailed error analysis (confusion matrices and misclassified examples) are included. Appendix contains full hyper-parameter grids, TF-IDF settings, confusion matrices, and SEM tables.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Leveraging artificial intelligence for predictive modelling of consumer buying intentions on E-Commerce platforms

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年1月11日

查看全文
  • C. K. Kotravel Bharathi,
  • K. Elakkiyan

摘要

The rise of e-commerce has transformed consumer behaviour, making the prediction of buying intentions critical for digital marketing strategy. This study integrates Artificial Intelligence (AI) and advanced statistical methods to examine how machine learning (ML) models and Structural Equation Modelling (SEM) can jointly predict consumer purchase intentions. Using a dataset of consumer reviews and behavioural signals from leading e-commerce platforms, we employ sentiment analysis, TF-IDF text vectorization, and predictive modelling across Logistic Regression, Random Forest, XGBoost, and Neural Networks. We also propose and test a SEM framework linking consumer trust, product quality perception, and sentiment polarity to buying intentions. Results highlight that ensemble ML models outperform linear classifiers, while SEM provides theoretical grounding by validating latent constructs. The hybrid SEM-ML framework demonstrates both predictive accuracy and theoretical rigor. Findings offer practical insights for e-commerce managers on optimizing customer engagement strategies. To improve theoretical clarity and robustness we explicitly anchor the SEM within the Technology Acceptance Model (TAM) and Theory of Planned Behaviour (TPB) and specify testable hypotheses (H1–H4). Sentiment analysis was re-run using both VADER and a transformer-based model (DistilBERT) and validated on a manually annotated subsample (n = 1000; DistilBERT accuracy = 87%, VADER = 71%). Time-aware validation (forward-chaining and rolling windows) and a robustness check using verified purchase as an alternative dependent variable were also performed. SEM reporting has been expanded to include standardized loadings, standard errors, p-values, Composite Reliability (CR), Average Variance Extracted (AVE), and discriminant validity. SHAP-based interpretability (summary + dependence plots) and detailed error analysis (confusion matrices and misclassified examples) are included. Appendix contains full hyper-parameter grids, TF-IDF settings, confusion matrices, and SEM tables.