While implementing the cross-category recommendation of tops and bottoms, the natural sentence of the reasoning for recommendation was generated by GRU.
Input item images of different categories and explanations for them at the same time, and output compatibility scores and natural sentences. Two types of attention are used to describe the relationship between items and text.
Basic Information
Link
https://arxiv.org/pdf/1806.08977.pdf
Overview
Others
Reference (for understanding)