Legal framework development through adversarial transfer learning for consumer grievance classification in algorithmic price discrimination contexts

The escalation of algorithmic price discrimination necessitates systematic analysis of consumer grievance data. Addressing the scarcity of annotated datasets in regulatory research, this study proposes a hybrid neural framework integrating BERT-based semantic encoding with optimised convolutional architectures. The model employs bidirectional recurrent layers to capture sequential dependencies while applying multi-head attention mechanisms for contextual feature fusion. A domain adaptation strategy combining adversarial training and transfer learning bridges feature distribution gaps between source and target domains. Through iterative parameter optimisation, the framework achieves cross-domain knowledge transfer while maintaining discriminative classification capabilities. Empirical validation demonstrates significant performance improvements, with macro-F1 scores increasing by 13.29% compared to baseline models. The classification outcomes inform three regulatory proposals addressing dynamic pricing oversight, algorithmic transparency requirements, and consumer compensation mechanisms. This dual technical-legal approach provides implementable solutions for governing emerging digital market practices while advancing domain adaptation methodologies in computational legal studies.