CGRS: Collaborative Knowledge Propagation Graph Attention Network for Recipes Recommendation

Figures & data

Figure 1. The overall framework of CGRS.

Figure 2. Graph Attention Feature Learning Network.

Table 1. Data details.

Graph	Attributes	Recipe
User-recipe interaction	#User	6285
	#Recipe	7827
	#Interactions	108248
Knowledge graph	# Entities	26809
	# Relations	9
	# Triples	149127

Figure 3. Comparison of different models in Top-K recommendation (Recall@K and Precision@K).

Figure 4. Comparison of different models in NDCG@K.

Table 2. CTR predicts AUC and F1 outcomes.

	AUC	F1
BPRMF	0.7518	0.6674
KPRN	0.7151	0.6238
MKR	0.8699	0.7731
Ripple Net	0.8776	0.7790
KGCN	0.8815	0.7925
KGNN-LS	0.8812	0.7841
CKAN	0.8821	0.7944
CGRS	0.8968	0.8157

Table 3. AUC and F1 experimental results at different depths.

K-layer	AUC	F1
CGRS-1	0.8824	0.7905
CGRS-2	0.8872	0.7993
CGRS-3	0.8955	0.8123
CGRS-4	0.8900	0.8060
CGRS-5	0.8843	0.7975

Table 4. AUC and F1 experimental results at different embedding dimensions.

Dimension	AUC	F1
8	0.8818	0.7952
16	0.8837	0.7972
32	0.8851	0.7984
64	0.8903	0.8052
128	0.8922	0.8075
256	0.8984	0.8169
512	0.8906	0.8036

Table 5. AUC and F1 experimental results at the different aggregator.

Aggregators	AUC	F1
CGRS-GCN	0.8653	0.7798
CGRS-BI	0.8927	0.8088

Table 6. AUC and F1 experimental results at different component.

Aggregators	AUC	F1
CGRS-Concat	0.8713	0.7842
CGRS-Attention	0.8941	0.8106

Data availability statement

Some data from the Ta-da dataset are used in this study and will be gradually opened after the completion of subsequent laboratory projects. The sample data is part of the Ta-da data set and can be accessed at https://github.com/Eimo-Bai/Ta-da-recipe-dataset, visited on December 17, 2022.