Skip to content

Latest commit

 

History

History
222 lines (184 loc) · 8.91 KB

Knowledge-based_recommendation.md

File metadata and controls

222 lines (184 loc) · 8.91 KB
Raw

Time and memory cost of knowledge-based recommendation models

Datasets information:

Dataset #User #Item #Interaction Sparsity #Entity #Relation #Triple
MovieLens-1m 6,034 3,096 832,104 95.5458% 10,234 54 206,844
Amazon-Books (2018) 160,383 4,000 344,601 99.9463% 10,302 22 152,882
Lastfm-track 45,987 38,439 3,118,764 99.8236% 94,104 12 774,508

Device information

OS:                   Linux
Python Version:       3.8.10
PyTorch Version:      1.8.1
cudatoolkit Version:  10.1
GPU:                  TITAN V（12GB）
Machine Specs:        14 CPU machine, 256GB RAM

1) MovieLens-1m dataset:

Time and memory cost on MovieLens-1m dataset:

Method Training Time (sec/epoch) Evaluation Time (sec/epoch) Training GPU Memory (GB) Evaluation GPU Memory (GB)
CFKG 3.03 0.30 0.05 18.15
CKE 9.32 0.22 0.29 0.58
KGAT 9.39 0.18 0.46 1.74
KGCN 5.61 2.07 0.07 0.19
KGIN 53.89 0.48 0.35 0.42
KGNNLS 10.63 2.2 0.22 0.22
KTUP 6.57 2.31 0.16 0.16
MCCLK 667.11 2.87 3.55 3.55
MKR 18.07 19.30 0.58 0.58
RippleNet 81.82 0.69 6.67 34.12

Config file of MovieLens-1m dataset:

# dataset config
field_separator: "\t"
seq_separator: " "
USER_ID_FIELD: user_id
ITEM_ID_FIELD: item_id
RATING_FIELD: rating
HEAD_ENTITY_ID_FIELD: head_id
TAIL_ENTITY_ID_FIELD: tail_id
RELATION_ID_FIELD: relation_id
ENTITY_ID_FIELD: entity_id
NEG_PREFIX: neg_
LABEL_FIELD: label
load_col:
    inter: [user_id, item_id, rating]
    kg: [head_id, relation_id, tail_id]
    link: [item_id, entity_id]

# data filtering for interactions
val_interval:
    rating: "[3,inf)"    
unused_col: 
    inter: [rating]

user_inter_num_interval: "[10,inf)"
item_inter_num_interval: "[10,inf)"

# data preprocessing for knowledge graph triples
kg_reverse_r: True
entity_kg_num_interval: "[5,inf)"
relation_kg_num_interval: "[5,inf)"

# training and evaluation
epochs: 500
train_batch_size: 4096
eval_batch_size: 40960000
valid_metric: NDCG@10
train_neg_sample_args: 
    distribution: uniform
    sample_num: 1
    dynamic: False

# model
embedding_size: 64

Other parameters (including model parameters) are default value.

2）Amazon-Books (2018) dataset:

Time and memory cost on Amazon-Books (2018) dataset:

Method Training Time (sec/epoch) Evaluation Time (sec/epoch) Training GPU Memory (GB) Evaluation GPU Memory (GB)
CFKG 66.53 1.92 16.79 16.79
CKE 64.80 2.09 1.1 1.1
KGAT 2.42 0.56 1.95 1.95
KGCN 1.43 6.74 0.36 0.36
KGIN 18.85 2.24 1.05 1.67
KGNNLS 1.89 6.92 0.36 0.36
KTUP 1.79 1.33 36.6 36.6
MCCLK 243.87 7.08 4.06 4.37
MKR 3.27 15.02 0.82 0.82
RippleNet 6.36 0.81 25.54 25.54

Config file of Amazon-Books (2018) dataset:

# dataset config
field_separator: "\t"
seq_separator: " "
USER_ID_FIELD: user_id
ITEM_ID_FIELD: item_id
RATING_FIELD: rating
HEAD_ENTITY_ID_FIELD: head_id
TAIL_ENTITY_ID_FIELD: tail_id
RELATION_ID_FIELD: relation_id
ENTITY_ID_FIELD: entity_id
NEG_PREFIX: neg_
LABEL_FIELD: label
load_col:
    inter: [user_id, item_id, rating]
    kg: [head_id, relation_id, tail_id]
    link: [item_id, entity_id]

# data filtering for interactions
val_interval:
    rating: "[3,inf)"    
unused_col: 
    inter: [rating]

user_inter_num_interval: "[10,inf)"
item_inter_num_interval: "[10,inf)"

# data preprocessing for knowledge graph triples
kg_reverse_r: True
entity_kg_num_interval: "[5,inf)"
relation_kg_num_interval: "[5,inf)"

# training and evaluation
epochs: 500
train_batch_size: 4096
eval_batch_size: 40960000
valid_metric: NDCG@10
train_neg_sample_args: 
    distribution: uniform
    sample_num: 1
    dynamic: False

# model
embedding_size: 64

Other parameters (including model parameters) are default value.

3）Lastfm-track dataset:

Time and memory cost on Lastfm-track dataset:

Method Training Time (sec/epoch) Evaluation Time (sec/epoch) Training GPU Memory (GB) Evaluation GPU Memory (GB)
CFKG 48.34 583.24 0.25 0.25
CKE 68.14 1.78 0.48 1.1
KGAT 161.36 2.01 5.89 5.89
KGCN 378.91 2186.54 0.38 0.38
KGIN 660.62 2.54 1.53 2.14
KGNNLS 66.58 264.76 0.92 0.92
KTUP 120.53 457.34 0.32 0.32
MCCLK - - - -
MKR 30.93 550.61 0.52 1.7
RippleNet 275.47 131.51 6.82 6.82

Config file of Lastfm-track dataset:

# dataset config
field_separator: "\t"
seq_separator: " "
USER_ID_FIELD: user_id
ITEM_ID_FIELD: item_id
RATING_FIELD: rating
HEAD_ENTITY_ID_FIELD: head_id
TAIL_ENTITY_ID_FIELD: tail_id
RELATION_ID_FIELD: relation_id
ENTITY_ID_FIELD: entity_id
NEG_PREFIX: neg_
LABEL_FIELD: label
load_col:
    inter: [user_id, item_id, rating]
    kg: [head_id, relation_id, tail_id]
    link: [item_id, entity_id]

# data filtering for interactions
val_interval:
    rating: "[3,inf)"    
unused_col: 
    inter: [rating]

user_inter_num_interval: "[10,inf)"
item_inter_num_interval: "[10,inf)"

# data preprocessing for knowledge graph triples
kg_reverse_r: True
entity_kg_num_interval: "[5,inf)"
relation_kg_num_interval: "[5,inf)"

# training and evaluation
epochs: 500
train_batch_size: 4096
eval_batch_size: 40960000
valid_metric: NDCG@10
train_neg_sample_args: 
    distribution: uniform
    sample_num: 1
    dynamic: False

# model
embedding_size: 64

Other parameters (including model parameters) are default value.