Scholarlink

<p>This study proposes a novel uncertainty-aware sustainability recommendation framework that integrates hesitant fuzzy user clustering with adaptive association rule mining, while treating sustainability as a core utility factor. The model is designed to operate in two phases. Initially, it employs hesitant fuzzy agglomerative hierarchical clustering to group users based on their preferences and behaviors. Subsequently, it utilizes hesitant fuzzy adapted association rule mining to group items and identify significant relationships among them. Incorporating sustainability metrics, such as environmental impact and eco-friendly attributes, into the rule-utility calculation in the second phase ensures that items aligned with sustainability goals are prioritized during rule generation. Because sustainability attributes and user intent are often imprecise or partially conflicting, we adopt hesitant fuzzy sets (HFS) to retain multiple plausible membership degrees throughout clustering and rule selection, while preserving standard top-<i>k</i> relevance. On a real-world fashion e-commerce dataset, we instantiate two configurations of our framework: (i) a baseline that ranks association rules without using sustainability information (utility-free), and (ii) a sustainability-aware variant that reweights rule utilities with SDG-aligned eco-scores. We evaluate both configurations with standard top-<i>k</i> recommendation metrics, precision@5 and NDCG@5, computed per user cluster and on average. The eco-aware variant improves overall recommendation quality (average P@5 increasing from 0.096 to 0.148 and NDCG@5 from 0.358 to 0.600), while keeping runtime and memory usage in the same order of magnitude as the baseline. To characterize computational behavior, we additionally report runtime and peak memory across increasing sample sizes. Empirically, our results show that top-confidence rules skew toward higher eco-score items, suggesting potential benefits for satisfaction and longer-term engagement.</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

Enhancing recommendation systems for sustainability: a two-phase approach with Hesitant Fuzzy clustering and adaptive association rule mining

  • 期刊论文

  • 发表日期：

    2026年4月23日

查看全文
  • Elmira Farrokhizadeh,
  • Başar Öztayşi

摘要

This study proposes a novel uncertainty-aware sustainability recommendation framework that integrates hesitant fuzzy user clustering with adaptive association rule mining, while treating sustainability as a core utility factor. The model is designed to operate in two phases. Initially, it employs hesitant fuzzy agglomerative hierarchical clustering to group users based on their preferences and behaviors. Subsequently, it utilizes hesitant fuzzy adapted association rule mining to group items and identify significant relationships among them. Incorporating sustainability metrics, such as environmental impact and eco-friendly attributes, into the rule-utility calculation in the second phase ensures that items aligned with sustainability goals are prioritized during rule generation. Because sustainability attributes and user intent are often imprecise or partially conflicting, we adopt hesitant fuzzy sets (HFS) to retain multiple plausible membership degrees throughout clustering and rule selection, while preserving standard top-k relevance. On a real-world fashion e-commerce dataset, we instantiate two configurations of our framework: (i) a baseline that ranks association rules without using sustainability information (utility-free), and (ii) a sustainability-aware variant that reweights rule utilities with SDG-aligned eco-scores. We evaluate both configurations with standard top-k recommendation metrics, precision@5 and NDCG@5, computed per user cluster and on average. The eco-aware variant improves overall recommendation quality (average P@5 increasing from 0.096 to 0.148 and NDCG@5 from 0.358 to 0.600), while keeping runtime and memory usage in the same order of magnitude as the baseline. To characterize computational behavior, we additionally report runtime and peak memory across increasing sample sizes. Empirically, our results show that top-confidence rules skew toward higher eco-score items, suggesting potential benefits for satisfaction and longer-term engagement.