A tensorflow implementation of the RecoGCN model in a CIKM'19 paper, titled with "Relation-Aware Graph Convolutional Networks for Agent-Initiated Social E-Commerce Recommendation".

python train.py 

Time elapsed = 6.89 mins, Training: loss = 389.51047, mrr = 0.63130, ndcg = 0.71369, hr1 = 0.50939, hr3 = 0.69945, hr5 = 0.78027, hr10 = 0.87522 | Val:loss = 2172.41870, mrr = 0.25467, ndcg = 0.40172, hr1 = 0.15110, hr3 = 0.25807, hr5 = 0.33136, hr10 = 0.45893

0	lr=0.0001,lamb=0.55,batch_size=400,numNegative=100,featEmbedDim=64,idenEmbedDim=64,outputDim=128,pathNum=7	Test loss:2033.5934; Test mrr:0.25339168; Test ndcg:0.3976466; Test hr1:0.14939758; Test hr3:0.2633283; Test hr5:0.34176204; Test hr10:0.46430722

adj = {0:{}, 1:{}, 2:{}, 3:{}}
with h5py.File(dataset, 'r') as f:
	adj[0][1] = f['adj01'][:]
	adj[1][0] = f['adj10'][:]
	adj[0][2] = f['adj02'][:]
	adj[2][0] = f['adj20'][:]
	adj[0][3] = f['adj03'][:]
	adj[3][0] = f['adj30'][:]

	train_sample = f['train_sample'][:]
	val_sample = f['val_sample'][:]
	test_sample = f['test_sample'][:]
		
	item_freq = f['item_freq'][:]
	user_feature = f['user_feature'][:]
	agent_feature = f['agent_feature'][:]
	item_feature = f['item_feature'][:]

	userCnt = f['userCnt'][()]
	agentCnt = f['agentCnt'][()]
	itemCnt = f['itemCnt'][()]

@inproceedings{xu2019relation,
  title={Relation-aware graph convolutional networks for agent-initiated social e-commerce recommendation},
  author={Xu, Fengli and Lian, Jianxun and Han, Zhenyu and Li, Yong and Xu, Yujian and Xie, Xing},
  booktitle={Proceedings of the 28th ACM International Conference on Information and Knowledge Management},
  pages={529--538},
  year={2019}
}