Objective: The rapid growth of e-commerce in Iran has transformed the purchasing patterns of hedonic products, yet understanding of how emotional and environmental factors act simultaneously on this behavior remains limited. The contribution of this study is contextual and integrative: placing a corpus of real transactions alongside an independent emotional survey within a convergent (triangulation) design, documenting a previously unreported late-night purchasing pattern, and translating segmentation into a budget-constrained selective targeting model in Iran's emerging market.



Methodology: The study employed a hybrid intelligent approach and relied on 1,379 real transactions and 1,000 valid questionnaire responses from Iranian users (owing to item-level nonresponse, the number of valid answers for some items is slightly below 1,000 and is reported where applicable). The methods included Pearson correlation, goal programming, principal component analysis, K-means clustering (validated with multiple internal indices and the DBSCAN algorithm), and inferential tests (one-way analysis of variance, Tukey's post hoc test, and the chi-square test).



Findings: Hedonic purchases peak between 10 p.m. and 2 a.m. and are largely concentrated in metropolitan areas, particularly Tehran. They were influenced more by happiness, nostalgia, user interface design, and the authenticity of user reviews than by price or transactional factors (with a correlation close to zero). Balancing emotional and financial objectives under explicit budget and cardinality constraints, the selective goal programming model identified a subset of users with high emotional-response potential. K-means clustering revealed three consumer tendencies: review-driven emotional buyers, users sensitive to user experience, and mood-enhancing impulse buyers. Analysis of variance and chi-square tests confirmed significant differences among these tendencies.



Conclusion: This research provides new evidence of nighttime purchasing patterns in the Iranian market and illustrates the complementary value of combining intelligent methods with conventional statistical analysis. Given the weak cluster separation (silhouette = 0.120), the three groupings are treated as overlapping behavioral tendencies rather than sharply separated segments; the managerial recommendations are therefore indicative directions for field testing (e.g., A/B testing) rather than definitive prescriptions. These recommendations cover the design of personalized late-night campaigns, a focus on urban areas, investment in user interface and review authenticity, and clustering-based strategies.