62 5. PROTOTYPE-WISE INTERPRETABLE COMPATIBILITY MODELING
RAND
ExIBR
BPR-DAE
Bi-LSTM
PAICM
RAND
ExIBR
BPR-DAE
Bi-LSTM
PAICM
0.50
0.45
0.40
0.35
0.30
0.25
0.20
0.15
0.10
0.05
0.00
0.50
0.45
0.40
0.35
0.30
0.25
0.20
0.15
0.10
0.05
0.00
0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100
MRR
MRR
Number of Bottom Candidates Number of Bottom Candidates
(a) Observed testing tops. (b) Unobserved testing tops.
Figure 5.8: Performance of different models.
numbers of bottom candidates in both scenarios, indicating the robustness and effectiveness of
PAICM in complementary fashion item retrieval.
5.5 SUMMARY
In this chapter, we present a prototype-guided interpretable compatibility modeling scheme,
PAICM, which is capable of not only determining the outfit compatibility, but also locating the
discordance of incompatible outfits as well as providing the alternative item suggestion. We em-
ploy the NMF to discover the latent compatible (incompatible) attribute interaction prototypes,
which are regarded as the templates to guide the discordant attribute interpretation and alter-
native item suggestion. Extensive experiments have been conducted on the real-world Dataset
I and the promising empirical results demonstrate the effectiveness of PAICM. In addition, we
found that the NMF has remarkable advantages of discovering latent factors in the context of
clothing matching.
..................Content has been hidden....................
You can't read the all page of ebook, please click here login for view all page.