AI Assistant for Sustainable and Effective Solutions of Fast Fashion based on Recommendation System and Deep Learning
A Skyline Query Framework with NLP-Derived Sustainability Scoring and Text-to-Metric Transformation Architecture
DOI:
https://doi.org/10.26821/IJSHRE.14.03.2026.140303Keywords:
Fast Fashion, Skyline Query, Deep Learning, NLP, Sustainability Scoring, Recommendation System, BERT, Sentiment Analysis, Multi-Objective OptimizationAbstract
The fast fashion industry generates massive volumes of unstructured text—product descriptions, consumer reviews, supply chain reports, social media discourse, and corporate sustainability pledges—yet current recommendation systems process only structured numerical features, discarding the rich semantic information embedded in natural language. This paper proposes SAFR-NLP (Sustainability-Aware Fashion Recommender via Natural Language Processing), a novel framework that extracts quantitative sustainability and effectiveness metrics directly from textual data sources through a multi-stage NLP pipeline, then applies Skyline Query optimization to identify Pareto-optimal product recommendations. I introduce the Text-to-Metric Transformation (T2M) architecture, which converts heterogeneous text inputs into six numerical dimensions constituting our Sustainability-Effectiveness Score (SES). The T2M pipeline comprises: (1) a fine-tuned SustainBERT module for extracting carbon, water, and recyclability indicators from product and supply chain text; (2) a Review Sentiment Decomposer for mapping consumer review language into preference alignment scores; (3) a Trend Language Detector that identifies emerging fashion concepts from social media corpora. The Skyline Query layer then computes the Pareto frontier over these NLP-derived metrics to produce recommendations that are simultaneously sustainable and commercially effective. Experiments on a dataset of 1.2 million text-product pairs demonstrate that SAFR-NLP achieves a 27.1% improvement in Hit@10 over text-unaware baselines and reduces sustainability deficit by 44.3%, while an end-to-end business deployment framework projects $15M ARR within 36 months across four market segments.
References
Börzsönyi, S., Kossmann, D., & Stocker, K. (2001). The Skyline Operator. In Proceedings of ICDE, pp. 421–430.
Chen, T., Xu, R., He, Y., & Wang, X. (2019). Aspect-Level Sentiment Classification with Deep Memory Network. In Proceedings of EMNLP.
Gao, D., et al. (2020). FashionBERT: Text and Image Matching with Adaptive Loss. In Proceedings of SIGIR, pp. 2251–2260.
Green, A., Shah, R., & Thompson, K. (2021). Predicting Product Carbon Footprints from Text. Journal of Cleaner Production, 312.
He, X., et al. (2020). LightGCN: Simplifying Graph Convolution for Recommendation. In Proceedings of SIGIR.
Kang, W. & McAuley, J. (2018). Self-Attentive Sequential Recommendation. In Proceedings of ICDM.
Li, H., Zhang, Y., & Chen, X. (2021). Multi-Aspect Sentiment Analysis for Fashion Reviews via BERT. In Proceedings of NAACL.
Lin, X., Yuan, Y., Wang, W., & Lu, H. (2005). Stabbing the Sky: Efficient Skyline over Sliding Windows. In Proceedings of ICDE.
Pei, J., Jiang, B., Lin, X., & Yuan, Y. (2007). Probabilistic Skylines on Uncertain Data. In Proceedings of VLDB.
Rendle, S., et al. (2009). BPR: Bayesian Personalized Ranking from Implicit Feedback. In Proceedings of UAI.
Scaria, K., et al. (2023). InstructABSA: Instruction Learning for ABSA. In Proceedings of NAACL.
Ibersinke, N., et al. (2022). ClimateBERT: A Pretrained Language Model for Climate-Related Text. In AAAI Workshop.
Yuan, Y., et al. (2005). Efficient Computation of the Skyline Cube. In Proceedings of VLDB.
Zhang, Y., Chen, W., & Li, X. (2022). Adversarial Skyline Queries. Information Systems, 108.
Zheng, S., et al. (2018). Joint Entity and Relation Extraction Based on a Hybrid Neural Network. Neurocomputing, 257.
