Clothes Collocation Recommendations by Compatibility Learning

This paper introduces a simple, yet effective, framework for clothes collocation by considering compatibility between items. In particular, we treat title sentences as the features of clothing items, instead of using clothing images. For feature transformation, the long-short term memory (LSTM) network is utilized for mapping title sentences into a low-dimensional space. Features of query and candidate items learned by the Siamese LSTMs are synthesized into a style space by a compatibility matrix. We evaluate our framework on two large-scale datasets compiled from Amazon and Taobao, respectively. Extensive experimental results show the effectiveness of our method in comparison to several state-of-the-art methods.