Enhancing apparel data based on fashion theory for developing a novel apparel style recommendation system

Guan, C; Qin, S; Ling, W; Long, Y

doi:10.1007/978-3-319-77700-9_4

Enhancing apparel data based on fashion theory for developing a novel apparel style recommendation system

Lookup NU author(s): Dr Yang Long

Downloads

Accepted version [.pdf]

Licence

This is the authors' accepted manuscript of a conference proceedings (inc. abstract) that has been published in its final definitive form by Springer Verlag, 2018.

For re-use rights please refer to the publisher's terms and conditions.

Abstract

© Springer International Publishing AG, part of Springer Nature 2018. Smart apparel recommendation system is a kind of machine learning system applied to clothes online shopping. The performance quality of the system is greatly dependent on apparel data quality as well as the system learning ability. This paper proposes (1) to enhance knowledge-based apparel data based on fashion communication theories and (2) to use deep learning driven methods for apparel data training. The acquisition of new apparel data is supported by apparel visual communication and sign theories. A two-step data training model is proposed. The first step is to predict apparel ATTRIBUTEs from the image data through a multi-task CNN model. The second step is to learn apparel MEANINGs from predicted attributes through SVM and LKF classifiers. The testing results show that the prediction rate of eleven predefined MEANING classes can reach the range from 80.1% to 93.5%. The two-step apparel learning model is applicable for novel recommendation system developments.