Skip to content

Latest commit

 

History

History
170 lines (141 loc) · 5.9 KB

Knowledge-based_recommendation.md

File metadata and controls

170 lines (141 loc) · 5.9 KB
Raw

Time and memory cost of knowledge-based recommendation models

Datasets information:

Dataset #User #Item #Interaction Sparsity #Entity #Relation #Triple
ml-1m 6,040 3,629 836,478 0.9618 79,388 51 385,923
ml-10m 69,864 10,599 8,242,124 0.9889 181,941 51 1,051,385
LFM-1b2013 28,150 64,583 1,907,900 0.9990 181,112 7 281,900

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) ml-1m dataset:

Time and memory cost on ml-1m dataset:

Method Training Time (sec/epoch) Evaluation Time (sec/epoch) GPU Memory (GB)
CKE 4.88 0.44 0.38
KTUP 3.76 1.70 0.47
RippleNet 35.85 0.84 7.26
KGAT 6.68 0.37 2.10
KGNN-LS 8.20 1.14 0.57
KGCN 3.56 1.14 0.56
MKR 4.36 5.57 3.68
CFKG 1.60 0.57 0.27

Config file of ml-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]
val_interval:
    rating: "[3,inf)"

# training and evaluation
epochs: 500
train_batch_size: 4096
eval_batch_size: 102400
valid_metric: MRR@10

# model
embedding_size: 64

Other parameters (including model parameters) are default value.

2）ml-10m dataset:

Time and memory cost on ml-10m dataset:

Method Training Time (sec/epoch) Evaluation Time (sec/epoch) GPU Memory (GB)
CKE 43.41 10.72 0.70
KTUP 33.82 38.87 0.66
RippleNet 360.16 23.35 7.38
KGAT - - CUDA out of memory
KGNN-LS 84.51 47.31 0.73
KGCN 20.13 53.33 0.74
MKR 31.74 207.12 3.85
CFKG 16.33 16.88 0.46

Config file of ml-10m 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]
val_interval:
    rating: "[3,inf)"

# training and evaluation
epochs: 500
train_batch_size: 4096
eval_batch_size: 102400
valid_metric: MRR@10

# model
embedding_size: 64

Other parameters (including model parameters) are default value.

3）LFM-1b dataset:

Time and memory cost on LFM-1b dataset:

Method Training Time (sec/epoch) Evaluation Time (sec/epoch) GPU Memory (GB)
CKE 8.73 41.84 0.69
KTUP 4.56 87.67 0.48
RippleNet 82.53 69.37 7.32
KGAT 15.69 40.75 4.19
KGNN-LS 15.65 436.84 0.61
KGCN 8.04 443.80 0.60
MKR 9.06 456.11 2.87
CFKG 3.81 50.48 0.45

Config file of LFM-1b 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]
val_interval:
  rating: "[10,inf)"
user_inter_num_interval: "[10,inf)"
item_inter_num_interval: "[10,inf)"

# training and evaluation
epochs: 500
train_batch_size: 4096
eval_batch_size: 102400
valid_metric: MRR@10

# model
embedding_size: 64

Other parameters (including model parameters) are default value.