Learning embeddings for fashion recommendation

In this work, we present a novel methodology to recommend items that are compatible with a given item of clothing. Compatibility is a hard notion to capture because of its diversity and subjectivity. We propose an embedding based approach to solve this problem, and perform recommendation based on product-closeness to the given clothing item. We perform this by first decomposing the notion of product-closeness into two inter-related notions of product similarity and product compatibility. Then, we incorporate product type into our embedding mechanism, and learn different embedding networks for different product types. We evaluate our proposed strategy extensively, and demonstrate that it performs better than the baseline, and is an effective method for performing few-shot transfer to compatibility prediction tasks.

Name thesis:degree_name

M.S.

Level thesis:degree_level

Thesis

Discipline thesis:degree_discipline

Computer Science

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2018

Author dc:creator

• Rajpal, Shreya

Contributors dc:contributor

• Forsyth, David A.

• embeddings
• fashion
• fashion recommendation
• computer vision

Statement dc:rights

• Copyright 2018 Shreya Rajpal

Language dc:language

en

Handle dc:identifier

http://hdl.handle.net/2142/100947

OAI identifier oai:identifier

oai:www.ideals.illinois.edu:2142/100947