Pre-processing Method for Conversation-Based Fashion Curating Automation Recommendation System

Abstract

Fashion curation is a system that recommends coordination outfits that are tailored to users’ situation and environment. Unlike traditional fashion recommendation systems that rely on user profile and survey information to recommend single clothing items, fashion curation assembles multiple items based on the user’s context and environment. This approach requires the availability of access to various types and wide range of user information. Personalized fashion curation can be automated by linking item characteristics and fabric attributes of user preferences based on Q&A interactions regardings between the AI coordinator and the user. Over time, the performance of the model can be continuously improved through the accumulation of conversation data, which enables ongoing retraining. Accordingly, a text-embedding pre-processing framework is essential, as it enables the AI coordinator to interpret and integrate both user conversations and fashion item information. This study proposes a conversational data pre-processing method for training an automated fashion curation model using a dataset that includes conversations between the AI coordinator and the user, as well as fashion item information. When using the proposed conversation data pre-processing method, both the Weighted Kendall Tau(WKT) Sum and WKT Avg exhibited significant improvements. The highest performance was achieved when both user conversations and fashion item information were pre-processed.