Semantically Enriched Recommender Systems for Smart Closets

With increasingly low prices on clothes and a tremendous interest in fashion, results in that people's wardrobes keep piling up. This causes people to struggle with daily selection of an outfit that includes clothing items that are matching and that suits the day's weather. Moreover, the environmental sustainability could benefit from people recycling their textile waste. This thesis investigates how exploiting new technologies, such as Big Data, recommender systems, semantic web, and Internet of Things can guide people in organizing their wardrobes more efficiently. The thesis proposes the architecture of a system consisting of a smart closet where the usage history of the user's clothing items can be tracked using RFID technology. Through a mobile application, the user can view wardrobe inventory, and receive daily outfit recommendations and recommendations for clothing items to recycle. To generate the daily outfit recommendations, the thesis proposes an approach based on collaborative filtering enabled with a machine learning algorithm--carefully selected through an experiment comparing multiple models on a real-world dataset. In addition to daily outfit recommendations, the approach facilitates the possibility of targeted advertisement by clothing retailers. Moreover, the thesis proposes a traditional content-based approach for recycling recommendations. The content-based approach utilizes semantic web technology and results from an evaluation shows that it outperforms a baseline approach without semantic web technology. As a whole, the thesis includes the state of the art for recommender systems utilizing semantic web technology, with especially focus on the domain of fashion recommendation from people's physical wardrobes