Douban.md

Experimental setting

Source domain dataset: Douban-Books

Target domain dataset: Douban-Music

Evaluation: all users in target dataset, ratio-based 8:1:1, full sort

Metrics: Recall, Precision, NDCG, MRR, Hit

Topk: 10, 20, 50

Properties:

field_separator: "\t"
source_domain:
  dataset: DoubanBook
  USER_ID_FIELD: user_id
  ITEM_ID_FIELD: item_id
  RATING_FIELD: rating
  TIME_FIELD: timestamp
  NEG_PREFIX: neg_
  LABEL_FIELD: label
  load_col:
    inter: [user_id, item_id, rating]
  user_inter_num_interval: "[5,inf)"
  item_inter_num_interval: "[5,inf)"
  val_interval:
    rating: "[3,inf)"
  drop_filter_field: True

target_domain:
  dataset: DoubanMovie
  USER_ID_FIELD: user_id
  ITEM_ID_FIELD: item_id
  RATING_FIELD: rating
  TIME_FIELD: timestamp
  NEG_PREFIX: neg_
  LABEL_FIELD: label
  load_col:
    inter: [user_id, item_id, rating]
  user_inter_num_interval: "[5,inf)"
  item_inter_num_interval: "[5,inf)"
  val_interval:
    rating: "[3,inf)"
  drop_filter_field: True

epochs: 500
train_batch_size: 4096
eval_batch_size: 409600
valid_metric: NDCG@10

For fairness, we restrict users' and items' embedding dimension as following. Please adjust the name of the corresponding args of different models.

embedding_size: 64

Dataset Statistics

Dataset	#Users	#items	#Interactions	Sparsity
Douban-Book	18085	33067	809248	99.86%
Douban-Movie	22041	25802	2552305	99.55%

Number of Overlapped User: 15434

Number of Overlapped Item: 0

Evaluation Results

Method	Recall@10	Precesion@10	NDCG@10	MRR@10	Hit@10
CoNet	0.1034	0.058	0.1011	0.1538	0.3224
CLFM	0.0885	0.0515	0.0861	0.1328	0.2948
DTCDR	0.0937	0.0582	0.0956	0.1487	0.3126
DeepAPF	0.067	0.0471	0.0737	0.1218	0.2626
BiTGCF	0.1124	0.063	0.109	0.1651	0.3485
CMF	0.0976	0.0588	0.0985	0.1531	0.3246
EMCDR	0.1169	0.067	0.1169	0.177	0.3568
NATR	0.081	0.0481	0.0757	0.1141	0.2774
SSCDR	0.1068	0.0614	0.1021	0.1483	0.3299
DCDCSR	0.0948	0.0531	0.0928	0.1464	0.3101

Method	Recall@20	Precesion@20	NDCG@20	MRR@20	Hit@20
CoNet	0.1581	0.0477	0.1108	0.1606	0.42
CLFM	0.1393	0.0434	0.096	0.1396	0.3937
DTCDR	0.149	0.0481	0.105	0.1555	0.4117
DeepAPF	0.1063	0.0393	0.0799	0.1277	0.3481
BiTGCF	0.1734	0.0522	0.1207	0.1721	0.4503
CMF	0.1521	0.0489	0.1086	0.1598	0.4216
EMCDR	0.1793	0.0545	0.1276	0.1838	0.4564
NATR	0.1341	0.042	0.0874	0.1213	0.3813
SSCDR	0.1695	0.0529	0.1153	0.1559	0.4408
DCDCSR	0.1458	0.0434	0.102	0.1529	0.4044

Method	Recall@50	Precesion@50	NDCG@50	MRR@50	Hit@50
CoNet	0.2687	0.0351	0.1332	0.1653	0.5653
CLFM	0.2433	0.033	0.1182	0.1442	0.5372
DTCDR	0.2591	0.0359	0.1272	0.1601	0.5551
DeepAPF	0.1943	0.0301	0.0977	0.1318	0.4771
BiTGCF	0.2891	0.0387	0.1452	0.1766	0.5903
CMF	0.262	0.0368	0.132	0.1643	0.561
EMCDR	0.2936	0.0393	0.1504	0.1883	0.5943
NATR	0.2359	0.0324	0.1098	0.126	0.5262
SSCDR	0.291	0.0404	0.1412	0.1609	0.5927
DCDCSR	0.2471	0.0319	0.1235	0.1574	0.5424

Hyper-parameters

Method	Best hyper-parameters
CoNet	learning_rate=0.005 mlp_hidden_size=[64,32,16,8] reg_weight=0.01
CLFM	learning_rate=0.0005 share_embedding_size=48 alpha=0.1 reg_weight=0.0001
DTCDR	learning_rate=0.0005 mlp_hidden_size=[64,64] dropout_prob=0.2 alpha=0.1 base_model=NeuMF
DeepAPF	learning_rate=0.0005
BiTGCF	learning_rate=0.0005 n_layers=2 concat_way=mean lambda_source=0.8 lambda_target=0.8 drop_rate=0.1 reg_weight=0.01
CMF	learning_rate=0.0005 lambda=0.9 gamma=0.1 alpha=0.1
EMCDR	learning_rate=0.001 mapping_function=non_linear mlp_hidden_size=[64] overlap_batch_size=100 reg_weight=0.01 latent_factor_model=BPR loss_type=BPR
NATR	learning_rate=0.001 max_inter_length=100 reg_weight=1e-5
SSCDR	learning_rate=0.0005 lambda=0 margin=0.2 overlap_batch_size=1024
DCDCSR	learning_rate=0.0005 mlp_hidden_size=[128] k=10