EUVIMEDEUVIMED entdecken
Crossref

DualPath-SSA: Separate Prediction of Strength and Importance on Plural Eccentricity Standards for Explainable and Diverse Outfit Recommendation

Kousuke Yamashita, Hiromitsu Shimakawa, Fumiko Harada

Advances in Science, Technology and Engineering Systems Journal · 2026-11 · 11 (6) · 1-17

DOI: 10.25046/aj110601 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Advances in Science, Technology and Engineering Systems Journal, 2026-11. Verfasst von Kousuke Yamashita und 2 weiteren Autor:innen. Die Arbeit nennt 3 Literaturangaben.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

In this paper, we propose DualPath-SSA, a fashion-recommendation method that realizes diversity recommendation based on subjective impression while preserving preference fit. Because existing diversification methods rely on differences in objective features, their diversity appears as a blend of many factors and cannot control the continuous change of subjective impression a user perceives. In contrast, focusing on eccentricity as one such subjective impression, the proposed method quantifies the overall eccentricity of an outfit from the visual elements composing it, and thereby expresses a range of subjective impression even among outfits that share a common item. In experiments on 100 outfits sharing a blue shirt, rated for eccentricity by 82 raters as ground truth, MMR that maximizes distance in the embedding space kept the human-rated eccentricity range below random selection, whereas the proposed method, selecting by the predicted eccentricity, widened this range to about twice that of similarity-based recommendation and reached the level of an ideal selection using human ratings. The method also reproduced the population’s ordering of eccentricity on significantly different image pairs at 82.6% (human reproducibility ceiling 88.2%), showing that subjective impression can be quantified at a level approaching human ratings. These results show that DualPath-SSA is a new fashion-recommendation framework integrating the quantification and diversity control of subjective impression.

Zitieren

APA

Yamashita, K., Shimakawa, H., & Harada, F. (2026). DualPath-SSA: Separate Prediction of Strength and Importance on Plural Eccentricity Standards for Explainable and Diverse Outfit Recommendation. Advances in Science, Technology and Engineering Systems Journal, 11(6), 1-17. https://doi.org/10.25046/aj110601

Vancouver

Yamashita K, Shimakawa H, Harada F. DualPath-SSA: Separate Prediction of Strength and Importance on Plural Eccentricity Standards for Explainable and Diverse Outfit Recommendation. Advances in Science, Technology and Engineering Systems Journal. 2026;11(6):1-17. doi:10.25046/aj110601

BibTeX
@article{Yamashita2026,
  author = {Yamashita, Kousuke and Shimakawa, Hiromitsu and Harada, Fumiko},
  title = {DualPath-SSA: Separate Prediction of Strength and Importance on Plural Eccentricity Standards for Explainable and Diverse Outfit Recommendation},
  journal = {Advances in Science, Technology and Engineering Systems Journal},
  year = {2026},
  volume = {11},
  number = {6},
  pages = {1--17},
  doi = {10.25046/aj110601}
}
Kostenlos starten →