基於色彩學改善個人化服裝推薦;Application of Chromatics on Personalized Clothing Recommendation

NCUIR > School of Management at National Central University > Graduate Institute of Information Management > Electronic Thesis & Dissertation > Item 987654321/93194

Please use this identifier to cite or link to this item: http://ir.lib.ncu.edu.tw/handle/987654321/93194

Title:	基於色彩學改善個人化服裝推薦;Application of Chromatics on Personalized Clothing Recommendation
Authors:	李紹齊;Lee, Shao-Chi
Contributors:	資訊管理學系
Keywords:	服裝推薦;色彩美學特徵;服裝物件擷取
Date:	2023-07-18
Issue Date:	2024-09-19 16:47:14 (UTC+8)
Publisher:	國立中央大學
Abstract:	服裝推薦問題是近年諸多學者研究的議題，尤其是個人化服裝推薦任務有大量方法進行探討，其主要結合商品圖像、文字特徵，以及用戶評分、評論等。本研究針對過往在服裝推薦中鮮少討論的色彩學，以及使用者美學偏好作為基礎，建立個人化的線上服裝推薦系統。過去研究將色彩視為單純圖片特徵，缺乏消費者對其色彩學構成的心理因素，而本研究討論服裝圖片的色彩學帶給消費者美學概念，以達成個人化美學偏好推薦，並增加推薦合理性。本研究以 OB 嚴選具設計風格的穿搭圖片進行整體色彩、服裝色彩、單品色彩風格提取，其依據為使用者會受圖片中不同元素吸引，而對其有不同的美學印象及偏好。研究架構分為內容導向、模型式協同過濾、混合式推薦方法。內容導向推薦討論三大風格色彩如何影響商品間相似度，協同過濾推薦將使用者進行美學偏好分群，再推薦與其同偏好使用者的商品選擇，混合式推薦則先將使用者進行偏好分群，再依據其推薦候選找出與目標商品相似度高的服裝商品，最後評估推薦方法間的 Hit Ratio。結果發現混合式及基於分群模型的推薦方法效果較好，驗證了色彩學及美學概念應用於使用者偏好建模及服裝推薦的可行性及合理性。;Clothing recommendation has always been a popular topic, especially for clothing personalization. Generally, clothing recommendation requires images or texts information, and comments or ratings as well. Our study focuses on the chromatics features on images, which is seldom discussed by previous studies. We aim to build an online personalized clothing recommendation system based on users’ aesthetic preference. Colors were regarded as just simple features of images by previous studies, which lacked a user’s psychological factors affected by images’ chromatics. Therefore, our study discusses the aesthetic concepts that images’ chromatics brings to users and build a personalized aesthetic recommendation with higher interpretability. Users tend to be attracted by elements in images, so we thus extract the overall, clothing, and single-color style from OB Design’s clothing dataset to build users’ aesthetic preference. There are three recommendation systems, which include content- based, model-based collaborative filtering, and hybrid methods. To elaborate, content- based method is for computation on products similarity based on chromatics, model- based collaborative filtering is to cluster user’s aesthetic preference, and hybrid method uses both methods. We conduct three experiments and assess the results by Hit Ratio, and we find the third system using both methods perform the best among all systems. In conclusion, our methods prove that applying chromatics and aesthetic features to model users’ preference and recommend clothing items is feasible and reasonable.
Appears in Collections:	[Graduate Institute of Information Management] Electronic Thesis & Dissertation

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	0	View/Open

社群 sharing

Loading...