According to the FashionUnited, the global fashion and apparel industry is valued of three trillion dollars, which demonstrates people’s great demand of clothing and the huge benefits behind it. In daily life, people can improve his(her) appearance by wearing compatible outfits. However, not everyone is a natural-born fashion-stylist and hence the research towards automatic clothing matching has a broad application prospect and market. Moreover, with the rapid growth of mobile internet and e-commerce, there has accumulated numerous clothing data, including images and text descriptions, where such tremendous volume of clothing data provides a wonderful venue to investigate the code in clothing matching. Currently, we have made progress from the following four aspects.

1) Data-driven Compatibility Modeling: we effectively combined the multi-modal information of fashion items and further explored a latent non-linear space to model fashion compatibility by advanced deep neural networks.

2) Knowledge-guided Compatibility Modeling: We employed a teacher-student framework to seamlessly integrate fashion domain knowledge (matching rule) to guide the data-driven method and achieved better performance than several state-of-the-art methods.

3) Personalized Compatibility Modeling: We parsed and model the users’ shopping history to capture the users’ preference, and further combine the general compatibility of fashion items to conduct personalized recommendation.

4) Explainable Compatibility Modeling: We utilized matrix factorization to capture some compatible and incompatible prototypes, based on which we can provide comprehensive explanations from the attribute interaction perspective. In the future, we will continue to explore related researches in the fashion domain, such as online clothing try-on and interactive fashion retrieval.