why GAN is useful?

[paulpig]

Hello, I ignore the GAN model and only keep the reconstruction loss. I find that the performance is the same as the CF-GAN. P@5: 0.4928 on MovieLen 100K.