Transformer-based Purchase Intention Mining: A Comparative Study Using A Novel Dataset

The scarcity of high-quality, contextually relevant datasets is a significant impediment to progress in the field of purchase intention mining. The absence of comprehensive datasets that capture the subtleties of consumer intentions across various contexts often hinders the development of robust and accurate predictive models. This research addresses this critical gap by developing a novel dataset specifically designed to encapsulate the intricacies of consumer purchase intentions. The primary aim of this research is to develop and evaluate a new dataset tailored for purchase intention mining and to assess the effectiveness of transformer-based models in this domain. To achieve this, we collected, preprocessed and analyzed a new dataset and fine-tuned advanced transformer models, including RoBERTa, ALBERT, and DistilBERT, on the newly created dataset. These models were then rigorously compared against traditional machine learning algorithms and deep learning architectures, such as Logistic Regression, Support Vector Machines, LightGBM, and Convolutional Neural Networks (CNN) with and without LSTM layers. The results of our comprehensive evaluation demonstrate that transformer models significantly outperform traditional approaches, achieving near-perfect accuracy, precision, recall, and F1-scores across different purchase intention categories (Positive, Negative, and Neutral). This superior performance was consistent across both undersampled and oversampled versions of the dataset, underscoring the robustness of our proposed dataset in facilitating high-precision sentiment analysis tasks. In conclusion, this research not only highlights the effectiveness of transformer models in understanding and predicting consumer purchase intentions but also emphasizes the importance of developing specialized datasets to overcome the challenges posed by dataset scarcity. Our findings align with existing literature, reinforcing the dominance of transformer-based approaches in natural language processing applications and setting a new standard in the field of purchase intention mining.