Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

StePO-Rec: Towards Personalized Outfit Styling Assistant via Knowledge-Guided Multi-Step Reasoning

Yuxi Bi, Yunfan Gao, Haofen Wang

TL;DR

StePO-Rec addresses the limitations of traditional fashion recommendations by grounding personalized outfit suggestions in a principled knowledge base (PAFA) and guiding recommendations with knowledge-guided multi-step reasoning. It combines a semantic bipartite style-rule graph, professional design principles, and a tree-based reasoning process with a preference-trend re-ranking module to produce interpretable, trend-aware outfits. On IQON-3000, StePO-Rec achieves Recall@1 of 0.55 and MAP of 0.85, outperforming strong baselines by approximately 28% and 33%, demonstrating improved accuracy and explainability. The work advances practical personalized fashion assistants by providing transparent decision pathways and scalable integration of expert knowledge with user preferences, enabling more reliable and explainable styling.

Abstract

Advancements in Generative AI offers new opportunities for FashionAI, surpassing traditional recommendation systems that often lack transparency and struggle to integrate expert knowledge, leaving the potential for personalized fashion styling remain untapped. To address these challenges, we present PAFA (Principle-Aware Fashion), a multi-granular knowledge base that organizes professional styling expertise into three levels of metadata, domain principles, and semantic relationships. Using PAFA, we develop StePO-Rec, a knowledge-guided method for multi-step outfit recommendation. StePO-Rec provides structured suggestions using a scenario-dimension-attribute framework, employing recursive tree construction to align recommendations with both professional principles and individual preferences. A preference-trend re-ranking system further adapts to fashion trends while maintaining the consistency of the user's original style. Experiments on the widely used personalized outfit dataset IQON show a 28% increase in Recall@1 and 32.8% in MAP. Furthermore, case studies highlight improved explainability, traceability, result reliability, and the seamless integration of expertise and personalization.

StePO-Rec: Towards Personalized Outfit Styling Assistant via Knowledge-Guided Multi-Step Reasoning

TL;DR

StePO-Rec addresses the limitations of traditional fashion recommendations by grounding personalized outfit suggestions in a principled knowledge base (PAFA) and guiding recommendations with knowledge-guided multi-step reasoning. It combines a semantic bipartite style-rule graph, professional design principles, and a tree-based reasoning process with a preference-trend re-ranking module to produce interpretable, trend-aware outfits. On IQON-3000, StePO-Rec achieves Recall@1 of 0.55 and MAP of 0.85, outperforming strong baselines by approximately 28% and 33%, demonstrating improved accuracy and explainability. The work advances practical personalized fashion assistants by providing transparent decision pathways and scalable integration of expert knowledge with user preferences, enabling more reliable and explainable styling.

Abstract

Advancements in Generative AI offers new opportunities for FashionAI, surpassing traditional recommendation systems that often lack transparency and struggle to integrate expert knowledge, leaving the potential for personalized fashion styling remain untapped. To address these challenges, we present PAFA (Principle-Aware Fashion), a multi-granular knowledge base that organizes professional styling expertise into three levels of metadata, domain principles, and semantic relationships. Using PAFA, we develop StePO-Rec, a knowledge-guided method for multi-step outfit recommendation. StePO-Rec provides structured suggestions using a scenario-dimension-attribute framework, employing recursive tree construction to align recommendations with both professional principles and individual preferences. A preference-trend re-ranking system further adapts to fashion trends while maintaining the consistency of the user's original style. Experiments on the widely used personalized outfit dataset IQON show a 28% increase in Recall@1 and 32.8% in MAP. Furthermore, case studies highlight improved explainability, traceability, result reliability, and the seamless integration of expertise and personalization.

Paper Structure

This paper contains 36 sections, 15 equations, 4 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Realted Work
  3. Personalized Outfit Recommendation
  4. LLM-based Personalized Assistant
  5. PAFA Knowledge Base
  6. Metadata Layer
  7. Principle Layer
  8. Principle Library ($\mathcal{P}$)
  9. Metric Space ($\mathcal{M}$)
  10. Semantic Layer
  11. Method
  12. Problem Formulation
  13. StePO-Rec Framework
  14. Hybrid Knowledge Retrieval
  15. Tree Search Multi-Step Reasoning
  16. ...and 21 more sections

Figures (4)

  • Figure 1: Comparison between traditional clothing recommendation systems and LLM-based outfit styling assistants
  • Figure 2: The PAFA Knowledge Base centralizes domain-specific knowledge by integrating fashion magazines, trend reports, styling guides, and designer insights through multi-step processing. It includes six sub-libraries: Fashion Trends, User Preference, Scene Styles, Color Coordination, Silhouette Matching, and Outfit Case Studies.
  • Figure 3: The StePO-Rec framework combines hybrid knowledge retrieval, tree-search multi-step reasoning, and preference-trends re-ranking. Using knowledge from the PAFA Knowledge Base, it deduces complementary garment attributes through tree-structured reasoning, with decisions at each node guided by dynamic context shaped by professional outfit styling rules and user preferences, ensuring interpretability and traceability at every step.
  • Figure 4: Case study of StePO-Rec