Mining Behavioural Purchase Patterns for Personalized Product Recommendation in E-Commerce Platforms Using Association Rules

The rise of e-commerce has generated vast amounts of transactional and behavioural data, paving the way for personalized product suggestions. However, traditional association-rule mining mainly captures static co-purchase relationships and therefore fails to reflect changes in consumer preferences and individual purchasing behaviour. In contrast, recent neural and sequential recommendation methods can capture intricate behavioural relationships but typically require high computational costs and lack interpretability. To address these limitations, this paper introduces a new Temporal-Contrastive Neuro- Symbolic Association Rule Network (TC-NSAR), which integrates lightweight neural preference learning with interpretable association-rule knowledge. Its symbolic basis is the framework of association rules, which is used to represent the explicit product relationships in terms of support, confidence and lift, and a contrastive neural encoder is used to learn user representations discriminatively that capture the core nature of the user's interactions while separating out the patterns that are behaviourally different from the user's preferences. Temporal weighting dynamically adjusts rule strength based on interaction recency, allowing the learned recommendation knowledge to evolve as purchasing behaviour changes. A symbolic rule representation and a neural preference embedding are then combined to calculate personalized product relevance scores and to output a Top-N list of recommendations with interpretable rule-level evidence. Experimental testing on a large-scale ecommerce behavioural dataset yields Recall@10 of 0.184, NDCG of 0.112, MRR of 0.079, and Hit Rate of 0.271, and covers 86.4% of recommendations. The proposed temporal weighting and contrastive preference learning mechanism is shown to be effective through comparisons with, and ablation studies relative to, the traditional static approach of association-rule recommendation. The framework offers an adaptive, interpretable and computation-efficient approach for personalized product discovery and crossselling in dynamic e-commerce settings.