1. dataset statistics: (mainly about the number of words, images in the outfit, how many outfits an average user clicked).
2. word embedding matrix的大小
RQ1: performance comparison
1) VBPR, TBPR, VTBPR
2) KNNLS
3) GP-BPR
4) Lu zhi
最好有8个左右，和GP-BPR对比
RQ2: variants of the fashion-MKN
1) transE + fashionMKN....
2) non-linear, linear
3) relu, tanh, sigmoid
4) attention
5) hidden_1_2 v.s., hidden_1
5) modality (v+t+e+两两结合）
embedding_size, number of sampled neighbors,
window sizes
of filters and the number of filters m
6)不同loss效果
case study (RQ3)
.列一个有kg和没有kg的推荐结果区别对比图。case study。反正就是放一个图，图示说明
网友评论