Deep Interest-Shifting Network with Meta-Embeddings for Fresh Item Recommendation

Table 3

Testing AUC comparison of context-aware models on synthetic datasets.

Methods	Dataset: Book-Crossing			Dataset: MovieLens
	Auxiliary	Context	Full data	Auxiliary	Context	Full data
DeepFM	0.7836	0.7741	0.7840	0.7284	0.7586	0.7654
PNN	0.7854	0.7725	0.7857	0.7290	0.7584	0.7649
CFM	0.7730	0.7726	0.7707	0.7289	0.7571	0.7633
DisNet-Add	–	0.7745	0.7862	–	0.7576	0.7660
DisNet-COT	–	0.7733	0.7864	–	0.7559	0.7664
DisNet-NN	0.7858	0.7748	0.7878	0.7287	0.7577	0.7666

indicates whether the DisNet-NN variant is significantly superior to the coupling algorithm or not on each dataset (pairwise t-test at the 0.05 significance level).On the auxiliary-only data, the network architecture of DisNet is fixed, and we only report the performance once.