服裝代表了一個人的形象、品味、風格以及價值觀,消費者常透過流行雜誌學習服裝穿著搭配以尋找個人風格。隨著網際網路的發展,消費者可透過網路購買商品、蒐集資訊與建立人際關係,但面對網路上琳瑯滿目的商品,使用者不易選擇符合他們所需求的商品。而推薦系統的發展主要是協助消費者在龐大的選擇之中找尋適合他們的商品。推薦系統目前已經廣泛的應用在個人的活動領域,包含旅遊、遊戲、電影、休閒生活等,但是應用在服裝推薦是較少研究觸及的。此外,目前現有的推薦系統網站大多是將推薦結果以條列式敘述、圖片、表格等方式呈現,介面較不友善。 有鑑於此,本研究歸納適用於服裝推薦系統之視覺化設計,將蒐集的服裝風格語彙經由專家分析,界定出適用於服裝推薦系統中衡量服裝風格之語彙,以此作為多準則評分推薦系統的基礎,成果如下:(1) 使用者服裝推薦:系統即時產生服裝圖片推薦給使用者、(2) 提供使用者了解社群使用者的喜好服裝、(3) 提供使用者了解特定服裝的喜好社群、(4) 服裝風格視覺化呈現:提供使用者檢視相似的服裝風格、(5) 社群風格視覺化呈現:藉此檢視有共同嗜好的社群使用者。透過這些方式,系統化地表現服裝圖片與使用者之間分佈脈絡,透過社群網絡的概念,以彌補現有推薦網站的不足。最後,本研究利用平均絕對誤差衡量推薦方式的準確率以及使用者回饋問卷的方式,以了解使用者對推薦系統的整體滿意度。
Clothing is an icon for personal image, taste and style. To get the information about the current trend of clothing fashion, consumers usually search what they want to know thru Internet. However, they often found themselves lost in the large amount of information flooded over Internet. Recommendation system is a way to aid the consumers to search the products they really need. Recommendation systems are commonly used in topics such as travel, game, movie and leisure life. Clothing recommendation is a topic that is rarely researched. Aside from that, the interface of current recommendation systems are mainly composed of description lists, pictures and tables that are visually unfriendly. The purpose of this thesis is to design an information visualization system for recommending clothes for the user. Vocabularies about clothing styles are analyzed by experts to describe the clothing samples in the database. Multi-criteria rating model is designed and implemented. The major results of our system include: (1) recommending clothing to the current user, (2) displaying favorite clothing of a group of users with similar taste, (3) finding the group of users who like a specific clothing, (4) examining clothing with similar visual style, and (5) separating users into groups with similar taste. The recommended results are displayed as a visual network. Finally, Mean Absolute Error is calculated for recommended results. User feedback is collected and analyzed to reveal the satisfactory degree of the proposed recommendation system.