Results

Here we present the average and standard deviation results of the current benchmark version. We also point to the corresponding .yml configuration file used for each configuration so that users can consistently reproduce experiments or build new configurations based on one of them.

ml-100k

Experiment ran using the MovieLens-100k dataset with the following presented models and their configurations. The complete configuration can be found in config_files/run_ml-100k.yml:

  • Summarized results from experiment_results/ml-100k.csv:

Model

MAP@5

nDCG@5

Node2Vec based model + cosine similarity;q=1.0;p=1.0;embedding_size=64

.1198 ± .0041

.1642 ± .0047

Node2Vec based model + cosine similarity;q=0.6;p=0.8;embedding_size=64

.1176 ± .0031

.1622 ± .0031

TransE based model + cosine similarity;embedding_dim=150;scoring_fct_norm=1;epochs=25;seed=42;triples=ratings

.0045 ± .0007

.0068 ± .0012

TransH based model + cosine similarity;embedding_dim=150;scoring_fct_norm=2;epochs=25;seed=42;triples=ratings

.0011 ± .0004

.0016 ± .0004

TransR based model + cosine similarity;embedding_dim=150;relation_dim=90;scoring_fct_norm=2;epochs=25;seed=42;triples=all

.0041 ± .0006

.0063 ± .0006

TransD based model + cosine similarity;embedding_dim=150;epochs=25;seed=42;triples=ratings

.0058 ± .0010

.0097 ± .0016

TuckER based model + cosine similarity;embedding_dim=200;dropout_0=0.3;dropout_1=0.4;dropout_2=0.5;apply_batch_normalization=True;epochs=25;seed=42;triples=ratings

.0041 ± .0004

.0065 ± .0008

RESCAL based model + cosine similarity;embedding_dim=50;epochs=25;seed=42;triples=ratings

.0081 ± .0009

.0128 ± .0015

DistMult based model + cosine similarity;embedding_dim=50;epochs=25;seed=42;triples=all

.0066 ± .0004

.0106 ± .0008

ComplEx based model + cosine similarity;embedding_dim=100;epochs=25;seed=42

.0066 ± .0004

.0106 ± .0007

RotatE based model + cosine similarity;embedding_dim=200;epochs=25;seed=42;triples=all

.0061 ± .0007

.0093 ± .0013

EPHEN based model + cosine similarity;embedding_model=sentence-transformers/all-roberta-large-v1;embed_with=abstract;iterations=30;mi=0.5

.0108 ± .0011

.0155 ± .0013

EPHEN based model + cosine similarity;embedding_model=deepwalk_based;embedding_model_kwargs={‘walk_len’: 10, ‘p’: 1.0, ‘q’: 1.0, ‘n_walks’: 50, ‘embedding_size’: 64, ‘epochs’: 1};embed_with=graph;iterations=30;mi=0.5

.1203 ± .0065

.1648 ± .0072

Entity2Rec;embedding_model=deepwalk_based;embedding_model_kwargs={‘config’: {‘save_weights’: True}, ‘parameters’: {‘walk_len’: 10, ‘p’: 1.0, ‘q’: 1.0, ‘n_walks’: 50, ‘embedding_size’: 64, ‘epochs’: 1}};run_all=False;workers=6;iterations=1;collab_only=False;content_only=False

.0082 ± .0017

.0129 ± .0026

  • Summarized execution time results from experiment_results/ml-100k_times.csv (configuration: CPU: AMD EPYC 7502P 32-Core Processor; RAM: 94GB; GPUs: [‘NVIDIA A2’]):

Model

Execution Time (s)

Node2Vec based model + cosine similarity;q=1.0;p=1.0;embedding_size=64

7.819 ± .4210

Node2Vec based model + cosine similarity;q=0.6;p=0.8;embedding_size=64

7.527 ± .3735

TransE based model + cosine similarity;embedding_dim=150;scoring_fct_norm=1;epochs=25;seed=42;triples=ratings

40.57 ± .6235

TransH based model + cosine similarity;embedding_dim=150;scoring_fct_norm=2;epochs=25;seed=42;triples=ratings

67.11 ± 2.103

TransR based model + cosine similarity;embedding_dim=150;relation_dim=90;scoring_fct_norm=2;epochs=25;seed=42;triples=all

79.04 ± 1.977

TransD based model + cosine similarity;embedding_dim=150;epochs=25;seed=42;triples=ratings

74.56 ± 1.877

TuckER based model + cosine similarity;embedding_dim=200;dropout_0=0.3;dropout_1=0.4;dropout_2=0.5;apply_batch_normalization=True;epochs=25;seed=42;triples=ratings

378.6 ± 3.069

RESCAL based model + cosine similarity;embedding_dim=50;epochs=25;seed=42;triples=ratings

58.44 ± 2.144

DistMult based model + cosine similarity;embedding_dim=50;epochs=25;seed=42;triples=all

40.39 ± 1.722

ComplEx based model + cosine similarity;embedding_dim=100;epochs=25;seed=42

44.13 ± 1.502

RotatE based model + cosine similarity;embedding_dim=200;epochs=25;seed=42;triples=all

48.31 ± 1.420

EPHEN based model + cosine similarity;embedding_model=sentence-transformers/all-roberta-large-v1;embed_with=abstract;iterations=30;mi=0.5

85.05 ± .3143

EPHEN based model + cosine similarity;embedding_model=deepwalk_based;embedding_model_kwargs={‘walk_len’: 10, ‘p’: 1.0, ‘q’: 1.0, ‘n_walks’: 50, ‘embedding_size’: 64, ‘epochs’: 1};embed_with=graph;iterations=30;mi=0.5

50.65 ± .3341

Entity2Rec;embedding_model=deepwalk_based;embedding_model_kwargs={‘config’: {‘save_weights’: True}, ‘parameters’: {‘walk_len’: 10, ‘p’: 1.0, ‘q’: 1.0, ‘n_walks’: 50, ‘embedding_size’: 64, ‘epochs’: 1}};run_all=False;workers=6;iterations=1;collab_only=False;content_only=False

73506 ± 5979.

ml-100k_enriched

Experiment ran using the MovieLens-100k dataset with DBpedia enrichement and the following presented models and their configurations. The complete configuration can be found in config_files/run_ml-100k_enriched.yml:

  • Summarized results from experiment_results/ml-100k_enriched.csv:

Model

MAP@5

nDCG@5

Node2Vec based model + cosine similarity;q=1.0;p=1.0;embedding_size=64

.1819 ± .0106

.2339 ± .0097

Node2Vec based model + cosine similarity;q=0.6;p=0.8;embedding_size=64

.1775 ± .0096

.2301 ± .0088

TransE based model + cosine similarity;embedding_dim=150;scoring_fct_norm=1;epochs=25;seed=42;triples=ratings

.0041 ± .0005

.0068 ± .0009

TransH based model + cosine similarity;embedding_dim=150;scoring_fct_norm=2;epochs=25;seed=42;triples=ratings

.0010 ± .0003

.0015 ± .0007

TransR based model + cosine similarity;embedding_dim=150;relation_dim=90;scoring_fct_norm=2;epochs=25;seed=42;triples=all

.0034 ± .0004

.0056 ± .0007

TransD based model + cosine similarity;embedding_dim=150;epochs=25;seed=42;triples=ratings

.0063 ± .0008

.0101 ± .0010

TuckER based model + cosine similarity;embedding_dim=200;dropout_0=0.3;dropout_1=0.4;dropout_2=0.5;apply_batch_normalization=True;epochs=25;seed=42;triples=ratings

.0040 ± .0005

.0065 ± .0006

RESCAL based model + cosine similarity;embedding_dim=50;epochs=25;seed=42;triples=ratings

.0079 ± .0011

.0124 ± .0019

DistMult based model + cosine similarity;embedding_dim=50;epochs=25;seed=42;triples=all

.0037 ± .0010

.0060 ± .0014

ComplEx based model + cosine similarity;embedding_dim=100;epochs=25;seed=42

.0065 ± .0009

.0106 ± .0014

RotatE based model + cosine similarity;embedding_dim=200;epochs=25;seed=42;triples=all

.0065 ± .0006

.0104 ± .0010

EPHEN based model + cosine similarity;embedding_model=sentence-transformers/all-roberta-large-v1;embed_with=abstract;iterations=30;mi=0.5

.0141 ± .0006

.0252 ± .0005

EPHEN based model + cosine similarity;embedding_model=deepwalk_based;embedding_model_kwargs={‘walk_len’: 10, ‘p’: 1.0, ‘q’: 1.0, ‘n_walks’: 50, ‘embedding_size’: 64, ‘epochs’: 1};embed_with=graph;iterations=30;mi=0.5

.1812 ± .0078

.2338 ± .0076

Entity2Rec;embedding_model=deepwalk_based;embedding_model_kwargs={‘config’: {‘save_weights’: True}, ‘parameters’: {‘walk_len’: 10, ‘p’: 1.0, ‘q’: 1.0, ‘n_walks’: 50, ‘embedding_size’: 64, ‘epochs’: 1}};run_all=False;workers=6;iterations=1;collab_only=False;content_only=False

.0076 ± .0010

.0120 ± .0019

  • Summarized execution time results from experiment_results/ml-100k_enriched_times.csv (configuration: CPU: AMD EPYC 7502P 32-Core Processor; RAM: 94GB; GPUs: [‘NVIDIA A2’]):

Model

Execution Time (s)

Node2Vec based model + cosine similarity;q=1.0;p=1.0;embedding_size=64

26.31 ± 1.073

Node2Vec based model + cosine similarity;q=0.6;p=0.8;embedding_size=64

25.44 ± .7091

TransE based model + cosine similarity;embedding_dim=150;scoring_fct_norm=1;epochs=25;seed=42;triples=ratings

43.75 ± 1.923

TransH based model + cosine similarity;embedding_dim=150;scoring_fct_norm=2;epochs=25;seed=42;triples=ratings

72.83 ± 3.641

TransR based model + cosine similarity;embedding_dim=150;relation_dim=90;scoring_fct_norm=2;epochs=25;seed=42;triples=all

128.0 ± 2.616

TransD based model + cosine similarity;embedding_dim=150;epochs=25;seed=42;triples=ratings

79.85 ± 3.074

TuckER based model + cosine similarity;embedding_dim=200;dropout_0=0.3;dropout_1=0.4;dropout_2=0.5;apply_batch_normalization=True;epochs=25;seed=42;triples=ratings

381.6 ± 3.547

RESCAL based model + cosine similarity;embedding_dim=50;epochs=25;seed=42;triples=ratings

60.83 ± 2.265

DistMult based model + cosine similarity;embedding_dim=50;epochs=25;seed=42;triples=all

55.10 ± 2.484

ComplEx based model + cosine similarity;embedding_dim=100;epochs=25;seed=42

46.26 ± 2.255

RotatE based model + cosine similarity;embedding_dim=200;epochs=25;seed=42;triples=all

102.9 ± 1.780

EPHEN based model + cosine similarity;embedding_model=sentence-transformers/all-roberta-large-v1;embed_with=abstract;iterations=30;mi=0.5

177.4 ± 1.292

EPHEN based model + cosine similarity;embedding_model=deepwalk_based;embedding_model_kwargs={‘walk_len’: 10, ‘p’: 1.0, ‘q’: 1.0, ‘n_walks’: 50, ‘embedding_size’: 64, ‘epochs’: 1};embed_with=graph;iterations=30;mi=0.5

96.80 ± 1.006

Entity2Rec;embedding_model=deepwalk_based;embedding_model_kwargs={‘config’: {‘save_weights’: True}, ‘parameters’: {‘walk_len’: 10, ‘p’: 1.0, ‘q’: 1.0, ‘n_walks’: 50, ‘embedding_size’: 64, ‘epochs’: 1}};run_all=False;workers=6;iterations=1;collab_only=False;content_only=False

68685 ± 10848

ml-1m

Experiment ran using the MovieLens-1m dataset with the following presented models and their configurations. The complete configuration can be found in config_files/run_ml-1m.yml:

  • Summarized results from experiment_results/ml-1m.csv:

Model

MAP@5

nDCG@5

Node2Vec based model + cosine similarity;q=1.0;p=1.0;embedding_size=64

.1110 ± .0022

.1452 ± .0027

Node2Vec based model + cosine similarity;q=0.6;p=0.8;embedding_size=64

.1103 ± .0035

.1444 ± .0037

TransE based model + cosine similarity;embedding_dim=150;scoring_fct_norm=1;epochs=25;seed=42;triples=ratings

.0038 ± .0002

.0059 ± .0002

TransH based model + cosine similarity;embedding_dim=150;scoring_fct_norm=2;epochs=25;seed=42;triples=ratings

.0014 ± .0001

.0022 ± .0001

TransR based model + cosine similarity;embedding_dim=150;relation_dim=90;scoring_fct_norm=2;epochs=25;seed=42;triples=all

.0011 ± .0001

.0016 ± .0002

TransD based model + cosine similarity;embedding_dim=150;epochs=25;seed=42;triples=ratings

.0022 ± .0003

.0034 ± .0004

TuckER based model + cosine similarity;embedding_dim=200;dropout_0=0.3;dropout_1=0.4;dropout_2=0.5;apply_batch_normalization=True;epochs=25;seed=42;triples=ratings

.0002 ± .0001

.0003 ± .0001

RESCAL based model + cosine similarity;embedding_dim=50;epochs=25;seed=42;triples=ratings

.0075 ± .0004

.0116 ± .0006

DistMult based model + cosine similarity;embedding_dim=50;epochs=25;seed=42;triples=all

.0012 ± .0005

.0019 ± .0005

ComplEx based model + cosine similarity;embedding_dim=100;epochs=25;seed=42

.0014 ± .0006

.0023 ± .0009

RotatE based model + cosine similarity;embedding_dim=200;epochs=25;seed=42;triples=all

.0036 ± .0003

.0057 ± .0004

EPHEN based model + cosine similarity;embedding_model=sentence-transformers/all-roberta-large-v1;embed_with=abstract;iterations=30;mi=0.5

.0050 ± .0006

.0073 ± .0017

EPHEN based model + cosine similarity;embedding_model=deepwalk_based;embedding_model_kwargs={‘walk_len’: 10, ‘p’: 1.0, ‘q’: 1.0, ‘n_walks’: 50, ‘embedding_size’: 64, ‘epochs’: 1};embed_with=graph;iterations=30;mi=0.5

.1098 ± .0016

.1438 ± .0026

  • Summarized execution time results from experiment_results/ml-1m_times.csv (configuration: CPU: AMD EPYC 7502P 32-Core Processor; RAM: 94GB; GPUs: [‘NVIDIA A2’]):

Model

Execution Time (s)

Node2Vec based model + cosine similarity;q=1.0;p=1.0;embedding_size=64

29.20 ± 1.310

Node2Vec based model + cosine similarity;q=0.6;p=0.8;embedding_size=64

32.17 ± 1.990

TransE based model + cosine similarity;embedding_dim=150;scoring_fct_norm=1;epochs=25;seed=42;triples=ratings

417.9 ± 19.26

TransH based model + cosine similarity;embedding_dim=150;scoring_fct_norm=2;epochs=25;seed=42;triples=ratings

704.0 ± 12.20

TransR based model + cosine similarity;embedding_dim=150;relation_dim=90;scoring_fct_norm=2;epochs=25;seed=42;triples=all

861.2 ± 16.92

TransD based model + cosine similarity;embedding_dim=150;epochs=25;seed=42;triples=ratings

848.3 ± 17.31

TuckER based model + cosine similarity;embedding_dim=200;dropout_0=0.3;dropout_1=0.4;dropout_2=0.5;apply_batch_normalization=True;epochs=25;seed=42;triples=ratings

3864. ± 20.52

RESCAL based model + cosine similarity;embedding_dim=50;epochs=25;seed=42;triples=ratings

588.4 ± 17.11

DistMult based model + cosine similarity;embedding_dim=50;epochs=25;seed=42;triples=all

397.9 ± 16.00

ComplEx based model + cosine similarity;embedding_dim=100;epochs=25;seed=42

465.11 ± 17.73

RotatE based model + cosine similarity;embedding_dim=200;epochs=25;seed=42;triples=all

689.4 ± 19.46

EPHEN based model + cosine similarity;embedding_model=sentence-transformers/all-roberta-large-v1;embed_with=abstract;iterations=30;mi=0.5

551.1 ± 6.458

EPHEN based model + cosine similarity;embedding_model=deepwalk_based;embedding_model_kwargs={‘walk_len’: 10, ‘p’: 1.0, ‘q’: 1.0, ‘n_walks’: 50, ‘embedding_size’: 64, ‘epochs’: 1};embed_with=graph;iterations=30;mi=0.5

450.4 ± 6.674

ml-1m_enriched

Experiment ran using the MovieLens-1m dataset with DBpedia enrichement and the following presented models and their configurations. The complete configuration can be found in config_files/run_ml-1m_enriched.yml:

  • Summarized results from experiment_results/ml-1m_enriched.csv:

Model

MAP@5

nDCG@5

Node2Vec based model + cosine similarity;q=1.0;p=1.0;embedding_size=64

.1625 ± .0021

.2014 ± .0026

Node2Vec based model + cosine similarity;q=0.6;p=0.8;embedding_size=64

.1624 ± .0038

.2016 ± .0040

TransE based model + cosine similarity;embedding_dim=150;scoring_fct_norm=1;epochs=25;seed=42;triples=ratings

.0041 ± .0002

.0063 ± .0002

TransH based model + cosine similarity;embedding_dim=150;scoring_fct_norm=2;epochs=25;seed=42;triples=ratings

.0014 ± .0002

.0021 ± .0004

TransR based model + cosine similarity;embedding_dim=150;relation_dim=90;scoring_fct_norm=2;epochs=25;seed=42;triples=all

.0008 ± .0001

.0012 ± .0001

TransD based model + cosine similarity;embedding_dim=150;epochs=25;seed=42;triples=ratings

.0021 ± .0002

.0033 ± .0002

TuckER based model + cosine similarity;embedding_dim=200;dropout_0=0.3;dropout_1=0.4;dropout_2=0.5;apply_batch_normalization=True;epochs=25;seed=42;triples=ratings

.0003 ± .0001

.0003 ± .0001

RESCAL based model + cosine similarity;embedding_dim=50;epochs=25;seed=42;triples=ratings

.0076 ± .0004

.0116 ± .0007

DistMult based model + cosine similarity;embedding_dim=50;epochs=25;seed=42;triples=all

.0020 ± .0006

.0029 ± .0008

ComplEx based model + cosine similarity;embedding_dim=100;epochs=25;seed=42

.0014 ± .0004

.0022 ± .0004

RotatE based model + cosine similarity;embedding_dim=200;epochs=25;seed=42;triples=all

.0034 ± .0002

.0055 ± .0002

EPHEN based model + cosine similarity;embedding_model=sentence-transformers/all-roberta-large-v1;embed_with=abstract;iterations=30;mi=0.5

.0020 ± .0002

.0030 ± .0001

EPHEN based model + cosine similarity;embedding_model=deepwalk_based;embedding_model_kwargs={‘walk_len’: 10, ‘p’: 1.0, ‘q’: 1.0, ‘n_walks’: 50, ‘embedding_size’: 64, ‘epochs’: 1};embed_with=graph;iterations=30;mi=0.5

.1622 ± .0040

.2013 ± .0036

  • Summarized execution time results from experiment_results/ml-1m_enriched_times.csv (configuration: CPU: AMD EPYC 7502P 32-Core Processor; RAM: 94GB; GPUs: [‘NVIDIA A2’]):

Model

Execution Time (s)

Node2Vec based model + cosine similarity;q=1.0;p=1.0;embedding_size=64

67.28 ± 2.997

Node2Vec based model + cosine similarity;q=0.6;p=0.8;embedding_size=64

74.52 ± 2.706

TransE based model + cosine similarity;embedding_dim=150;scoring_fct_norm=1;epochs=25;seed=42;triples=ratings

505.7 ± 136.6

TransH based model + cosine similarity;embedding_dim=150;scoring_fct_norm=2;epochs=25;seed=42;triples=ratings

834.3 ± 181.8

TransR based model + cosine similarity;embedding_dim=150;relation_dim=90;scoring_fct_norm=2;epochs=25;seed=42;triples=all

1404. ± 282.5

TransD based model + cosine similarity;embedding_dim=150;epochs=25;seed=42;triples=ratings

946.1 ± 112.5

TuckER based model + cosine similarity;embedding_dim=200;dropout_0=0.3;dropout_1=0.4;dropout_2=0.5;apply_batch_normalization=True;epochs=25;seed=42;triples=ratings

4475. ± 1067.

RESCAL based model + cosine similarity;embedding_dim=50;epochs=25;seed=42;triples=ratings

648.5 ± 55.35

DistMult based model + cosine similarity;embedding_dim=50;epochs=25;seed=42;triples=all

470.0 ± 19.09

ComplEx based model + cosine similarity;embedding_dim=100;epochs=25;seed=42

490.4 ± 23.20

RotatE based model + cosine similarity;embedding_dim=200;epochs=25;seed=42;triples=all

1289. ± 22.07

EPHEN based model + cosine similarity;embedding_model=sentence-transformers/all-roberta-large-v1;embed_with=abstract;iterations=30;mi=0.5

749.3 ± 6.818

EPHEN based model + cosine similarity;embedding_model=deepwalk_based;embedding_model_kwargs={‘walk_len’: 10, ‘p’: 1.0, ‘q’: 1.0, ‘n_walks’: 50, ‘embedding_size’: 64, ‘epochs’: 1};embed_with=graph;iterations=30;mi=0.5

549.0 ± 10.42

lastfm

Experiment ran using the Lastfm dataset with the following presented models and their configurations. The complete configuration can be found in config_files/run_lastfm.yml:

  • Summarized results from experiment_results/lastfm.csv:

Model

MAP@5

nDCG@5

Node2Vec based model + cosine similarity;q=1.0;p=1.0;embedding_size=64

.1171 ± .0034

.1628 ± .0054

Node2Vec based model + cosine similarity;q=0.6;p=0.8;embedding_size=64

.1178 ± .0018

.1621 ± .0027

TransE based model + cosine similarity;embedding_dim=150;scoring_fct_norm=1;epochs=25;seed=42;triples=ratings

.0003 ± .0001

.0003 ± .0001

TransH based model + cosine similarity;embedding_dim=150;scoring_fct_norm=2;epochs=25;seed=42;triples=ratings

.0002 ± .0001

.0002 ± .0002

TransR based model + cosine similarity;embedding_dim=150;relation_dim=90;scoring_fct_norm=2;epochs=25;seed=42;triples=all

.0002 ± .0001

.0003 ± .0002

TransD based model + cosine similarity;embedding_dim=150;epochs=25;seed=42;triples=ratings

.0002 ± .0001

.0003 ± .0001

TuckER based model + cosine similarity;embedding_dim=200;dropout_0=0.3;dropout_1=0.4;dropout_2=0.5;apply_batch_normalization=True;epochs=25;seed=42;triples=ratings

.0002 ± .0001

.0004 ± .0001

RESCAL based model + cosine similarity;embedding_dim=50;epochs=25;seed=42;triples=ratings

.0002 ± .0001

.0004 ± .0002

DistMult based model + cosine similarity;embedding_dim=50;epochs=25;seed=42;triples=all

.0003 ± .0002

.0003 ± .0002

ComplEx based model + cosine similarity;embedding_dim=100;epochs=25;seed=42

.0003 ± .0001

.0004 ± .0001

RotatE based model + cosine similarity;embedding_dim=200;epochs=25;seed=42;triples=all

.0003 ± .0001

.0003 ± .0002

EPHEN based model + cosine similarity;embedding_model=sentence-transformers/all-roberta-large-v1;embed_with=abstract;iterations=30;mi=0.5

.0001 ± .0000

.0001 ± .0000

EPHEN based model + cosine similarity;embedding_model=deepwalk_based;embedding_model_kwargs={‘walk_len’: 10, ‘p’: 1.0, ‘q’: 1.0, ‘n_walks’: 50, ‘embedding_size’: 64, ‘epochs’: 1};embed_with=graph;iterations=30;mi=0.5

.1174 ± .0021

.1624 ± .0043

  • Summarized execution time results from experiment_results/lastfm_times.csv (configuration: CPU: AMD EPYC 7502P 32-Core Processor; RAM: 94GB; GPUs: [‘NVIDIA A2’]):

Model

Execution Time (s)

Node2Vec based model + cosine similarity;q=1.0;p=1.0;embedding_size=64

65.50 ± 1.855

Node2Vec based model + cosine similarity;q=0.6;p=0.8;embedding_size=64

63.40 ± 3.157

TransE based model + cosine similarity;embedding_dim=150;scoring_fct_norm=1;epochs=25;seed=42;triples=ratings

60.20 ± 1.085

TransH based model + cosine similarity;embedding_dim=150;scoring_fct_norm=2;epochs=25;seed=42;triples=ratings

90.59 ± 2.066

TransR based model + cosine similarity;embedding_dim=150;relation_dim=90;scoring_fct_norm=2;epochs=25;seed=42;triples=all

453.1 ± 4.384

TransD based model + cosine similarity;embedding_dim=150;epochs=25;seed=42;triples=ratings

113.7 ± 2.467

TuckER based model + cosine similarity;embedding_dim=200;dropout_0=0.3;dropout_1=0.4;dropout_2=0.5;apply_batch_normalization=True;epochs=25;seed=42;triples=ratings

384.5 ± 2.255

RESCAL based model + cosine similarity;embedding_dim=50;epochs=25;seed=42;triples=ratings

108.8 ± 2.364

DistMult based model + cosine similarity;embedding_dim=50;epochs=25;seed=42;triples=all

55.70 ± 2.296

ComplEx based model + cosine similarity;embedding_dim=100;epochs=25;seed=42

66.67 ± 2.371

RotatE based model + cosine similarity;embedding_dim=200;epochs=25;seed=42;triples=all

155.3 ± 1.722

EPHEN based model + cosine similarity;embedding_model=sentence-transformers/all-roberta-large-v1;embed_with=abstract;iterations=30;mi=0.5

438.8 ± .7641

EPHEN based model + cosine similarity;embedding_model=deepwalk_based;embedding_model_kwargs={‘walk_len’: 10, ‘p’: 1.0, ‘q’: 1.0, ‘n_walks’: 50, ‘embedding_size’: 64, ‘epochs’: 1};embed_with=graph;iterations=30;mi=0.5

143.9 ± 2.147

lastfm_enriched

Experiment ran using the Lastfm dataset with DBpedia enrichement and the following presented models and their configurations. The complete configuration can be found in config_files/run_lastfm-enriched.yml:

  • Summarized results from experiment_results/lastfm_enriched.csv:

Model

MAP@5

nDCG@5

Node2Vec based model + cosine similarity;q=1.0;p=1.0;embedding_size=64

.1408 ± .0043

.1923 ± .0054

Node2Vec based model + cosine similarity;q=0.6;p=0.8;embedding_size=64

.1390 ± .0035

.1946 ± .0035

TransE based model + cosine similarity;embedding_dim=150;scoring_fct_norm=1;epochs=25;seed=42;triples=ratings

.0003 ± .0001

.0003 ± .0001

TransH based model + cosine similarity;embedding_dim=150;scoring_fct_norm=2;epochs=25;seed=42;triples=ratings

.0002 ± .0002

.0002 ± .0003

TransR based model + cosine similarity;embedding_dim=150;relation_dim=90;scoring_fct_norm=2;epochs=25;seed=42;triples=all

.0001 ± .0001

.0001 ± .0001

TransD based model + cosine similarity;embedding_dim=150;epochs=25;seed=42;triples=ratings

.0002 ± .0001

.0003 ± .0001

TuckER based model + cosine similarity;embedding_dim=200;dropout_0=0.3;dropout_1=0.4;dropout_2=0.5;apply_batch_normalization=True;epochs=25;seed=42;triples=ratings

.0003 ± .0001

.0003 ± .0001

RESCAL based model + cosine similarity;embedding_dim=50;epochs=25;seed=42;triples=ratings

.0002 ± .0001

.0002 ± .0001

DistMult based model + cosine similarity;embedding_dim=50;epochs=25;seed=42;triples=all

.0003 ± .0001

.0003 ± .0002

ComplEx based model + cosine similarity;embedding_dim=100;epochs=25;seed=42

.0002 ± .0001

.0003 ± .0002

RotatE based model + cosine similarity;embedding_dim=200;epochs=25;seed=42;triples=all

.0003 ± .0001

.0003 ± .0002

EPHEN based model + cosine similarity;embedding_model=sentence-transformers/all-roberta-large-v1;embed_with=abstract;iterations=30;mi=0.5

.0021 ± .0011

.0023 ± .0013

EPHEN based model + cosine similarity;embedding_model=deepwalk_based;embedding_model_kwargs={‘walk_len’: 10, ‘p’: 1.0, ‘q’: 1.0, ‘n_walks’: 50, ‘embedding_size’: 64, ‘epochs’: 1};embed_with=graph;iterations=30;mi=0.5

.1383 ± .0017

.1903 ± .0026

  • Summarized execution time results from experiment_results/lastfm_enriched_times.csv (configuration: CPU: AMD EPYC 7502P 32-Core Processor; RAM: 94GB; GPUs: [‘NVIDIA A2’]):

Model

Execution Time (s)

Node2Vec based model + cosine similarity;q=1.0;p=1.0;embedding_size=64

153.3 ± 1.309

Node2Vec based model + cosine similarity;q=0.6;p=0.8;embedding_size=64

155.9 ± 1.798

TransE based model + cosine similarity;embedding_dim=150;scoring_fct_norm=1;epochs=25;seed=42;triples=ratings

74.97 ± 18.59

TransH based model + cosine similarity;embedding_dim=150;scoring_fct_norm=2;epochs=25;seed=42;triples=ratings

112.3 ± 24.47

TransR based model + cosine similarity;embedding_dim=150;relation_dim=90;scoring_fct_norm=2;epochs=25;seed=42;triples=all

1767. ± 530.4

TransD based model + cosine similarity;embedding_dim=150;epochs=25;seed=42;triples=ratings

143.8 ± 30.80

TuckER based model + cosine similarity;embedding_dim=200;dropout_0=0.3;dropout_1=0.4;dropout_2=0.5;apply_batch_normalization=True;epochs=25;seed=42;triples=ratings

513.9 ± 157.7

RESCAL based model + cosine similarity;embedding_dim=50;epochs=25;seed=42;triples=ratings

142.0 ± 34.93

DistMult based model + cosine similarity;embedding_dim=50;epochs=25;seed=42;triples=all

213.1 ± 68.10

ComplEx based model + cosine similarity;embedding_dim=100;epochs=25;seed=42

89.37 ± 21.57

RotatE based model + cosine similarity;embedding_dim=200;epochs=25;seed=42;triples=all

951.5 ± 283.4

EPHEN based model + cosine similarity;embedding_model=sentence-transformers/all-roberta-large-v1;embed_with=abstract;iterations=30;mi=0.5

629.6 ± 30.63

EPHEN based model + cosine similarity;embedding_model=deepwalk_based;embedding_model_kwargs={‘walk_len’: 10, ‘p’: 1.0, ‘q’: 1.0, ‘n_walks’: 50, ‘embedding_size’: 64, ‘epochs’: 1};embed_with=graph;iterations=30;mi=0.5

348.0 ± 6.012

douban-movie

Experiment ran using the Douban Movie dataset with the following presented models and their configurations. The complete configuration can be found in config_files/run_douban-movie.yml:

  • Summarized results from experiment_results/douban-movie.csv:

Model

MAP@5

nDCG@5

Node2Vec based model + cosine similarity;q=1.0;p=1.0;embedding_size=64

.5843 ± .0101

.6754 ± .0117

Node2Vec based model + cosine similarity;q=0.6;p=0.8;embedding_size=64

.5885 ± .0047

.6825 ± .0044

TransE based model + cosine similarity;embedding_dim=150;scoring_fct_norm=1;epochs=25;seed=42;triples=ratings

.3991 ± .0032

.4813 ± .0064

TransH based model + cosine similarity;embedding_dim=150;scoring_fct_norm=2;epochs=25;seed=42;triples=ratings

.3998 ± .0045

.4687 ± .0079

TransR based model + cosine similarity;embedding_dim=150;relation_dim=90;scoring_fct_norm=2;epochs=25;seed=42;triples=all

.4036 ± .0018

.4855 ± .0044

TransD based model + cosine similarity;embedding_dim=150;epochs=25;seed=42;triples=ratings

.4076 ± .0034

.4888 ± .0045

TuckER based model + cosine similarity;embedding_dim=200;dropout_0=0.3;dropout_1=0.4;dropout_2=0.5;apply_batch_normalization=True;epochs=25;seed=42;triples=ratings

.4095 ± .0024

.4903 ± .0052

RESCAL based model + cosine similarity;embedding_dim=50;epochs=25;seed=42;triples=ratings

.4091 ± .0007

.4910 ± .0036

DistMult based model + cosine similarity;embedding_dim=50;epochs=25;seed=42;triples=all

.4067 ± .0019

.4877 ± .0044

ComplEx based model + cosine similarity;embedding_dim=100;epochs=25;seed=42

.4043 ± .0021

.4857 ± .0049

RotatE based model + cosine similarity;embedding_dim=200;epochs=25;seed=42;triples=all

.4091 ± .0024

.4882 ± .0066

EPHEN based model + cosine similarity;embedding_model=sentence-transformers/all-roberta-large-v1;embed_with=abstract;iterations=30;mi=0.5

.4517 ± .0078

.5284 ± .0045

EPHEN based model + cosine similarity;embedding_model=deepwalk_based;embedding_model_kwargs={‘walk_len’: 10, ‘p’: 1.0, ‘q’: 1.0, ‘n_walks’: 50, ‘embedding_size’: 64, ‘epochs’: 1};embed_with=graph;iterations=30;mi=0.5

.5833 ± .0043

.6755 ± .0059

Entity2Rec;embedding_model=deepwalk_based;embedding_model_kwargs={‘config’: {‘save_weights’: True}, ‘parameters’: {‘walk_len’: 10, ‘p’: 1.0, ‘q’: 1.0, ‘n_walks’: 50, ‘embedding_size’: 64, ‘epochs’: 1}};run_all=False;workers=6;iterations=1;collab_only=False;content_only=False

.4142 ± .0084

.4971 ± .0120

  • Summarized execution time results from experiment_results/douban-movie_times.csv (configuration: CPU: AMD EPYC 7502P 32-Core Processor; RAM: 94GB; GPUs: [‘NVIDIA A2’]):

Model

Execution Time (s)

Node2Vec based model + cosine similarity;q=1.0;p=1.0;embedding_size=64

5.199 ± .6045

Node2Vec based model + cosine similarity;q=0.6;p=0.8;embedding_size=64

5.240 ± .1954

TransE based model + cosine similarity;embedding_dim=150;scoring_fct_norm=1;epochs=25;seed=42;triples=ratings

47.84 ± .6285

TransH based model + cosine similarity;embedding_dim=150;scoring_fct_norm=2;epochs=25;seed=42;triples=ratings

70.69 ± 1.761

TransR based model + cosine similarity;embedding_dim=150;relation_dim=90;scoring_fct_norm=2;epochs=25;seed=42;triples=all

78.39 ± 2.203

TransD based model + cosine similarity;embedding_dim=150;epochs=25;seed=42;triples=ratings

79.52 ± 3.933

TuckER based model + cosine similarity;embedding_dim=200;dropout_0=0.3;dropout_1=0.4;dropout_2=0.5;apply_batch_normalization=True;epochs=25;seed=42;triples=ratings

348.6 ± 7.303

RESCAL based model + cosine similarity;embedding_dim=50;epochs=25;seed=42;triples=ratings

62.87 ± 2.368

DistMult based model + cosine similarity;embedding_dim=50;epochs=25;seed=42;triples=all

44.91 ± 1.734

ComplEx based model + cosine similarity;embedding_dim=100;epochs=25;seed=42

49.91 ± 2.109

RotatE based model + cosine similarity;embedding_dim=200;epochs=25;seed=42;triples=all

51.02 ± 1.804

EPHEN based model + cosine similarity;embedding_model=sentence-transformers/all-roberta-large-v1;embed_with=abstract;iterations=30;mi=0.5

47.73 ± .6584

EPHEN based model + cosine similarity;embedding_model=deepwalk_based;embedding_model_kwargs={‘walk_len’: 10, ‘p’: 1.0, ‘q’: 1.0, ‘n_walks’: 50, ‘embedding_size’: 64, ‘epochs’: 1};embed_with=graph;iterations=30;mi=0.5

41.90 ± .5456

Entity2Rec;embedding_model=deepwalk_based;embedding_model_kwargs={‘config’: {‘save_weights’: True}, ‘parameters’: {‘walk_len’: 10, ‘p’: 1.0, ‘q’: 1.0, ‘n_walks’: 50, ‘embedding_size’: 64, ‘epochs’: 1}};run_all=False;workers=6;iterations=1;collab_only=False;content_only=False

9479. ± 301.5

douban-movie_enriched

Experiment ran using the Douban Movie dataset with DBpedia enrichement and the following presented models and their configurations. The complete configuration can be found in config_files/run_douban-movie_enriched.yml:

  • Summarized results from experiment_results/douban-movie_enriched.csv:

Model

MAP@5

nDCG@5

Node2Vec based model + cosine similarity;q=1.0;p=1.0;embedding_size=64

.5739 ± .0093

.6627 ± .0134

Node2Vec based model + cosine similarity;q=0.6;p=0.8;embedding_size=64

.5722 ± .0055

.6609 ± .0084

TransE based model + cosine similarity;embedding_dim=150;scoring_fct_norm=1;epochs=25;seed=42;triples=ratings

.4035 ± .0029

.4830 ± .0044

TransH based model + cosine similarity;embedding_dim=150;scoring_fct_norm=2;epochs=25;seed=42;triples=ratings

.3949 ± .0029

.4695 ± .0056

TransR based model + cosine similarity;embedding_dim=150;relation_dim=90;scoring_fct_norm=2;epochs=25;seed=42;triples=all

.4062 ± .0022

.4872 ± .0066

TransD based model + cosine similarity;embedding_dim=150;epochs=25;seed=42;triples=ratings

.4076 ± .0033

.4883 ± .0040

TuckER based model + cosine similarity;embedding_dim=200;dropout_0=0.3;dropout_1=0.4;dropout_2=0.5;apply_batch_normalization=True;epochs=25;seed=42;triples=ratings

.4045 ± .0032

.4866 ± .0055

RESCAL based model + cosine similarity;embedding_dim=50;epochs=25;seed=42;triples=ratings

.4102 ± .0056

.4916 ± .0036

DistMult based model + cosine similarity;embedding_dim=50;epochs=25;seed=42;triples=all

.4011 ± .0011

.4810 ± .0040

ComplEx based model + cosine similarity;embedding_dim=100;epochs=25;seed=42

.4056 ± .0037

.4871 ± .0062

RotatE based model + cosine similarity;embedding_dim=200;epochs=25;seed=42;triples=all

.4086 ± .0016

.4895 ± .0028

EPHEN based model + cosine similarity;embedding_model=sentence-transformers/all-roberta-large-v1;embed_with=abstract;iterations=30;mi=0.5

.3876 ± .0038

.4601 ± .0036

EPHEN based model + cosine similarity;embedding_model=deepwalk_based;embedding_model_kwargs={‘walk_len’: 10, ‘p’: 1.0, ‘q’: 1.0, ‘n_walks’: 50, ‘embedding_size’: 64, ‘epochs’: 1};embed_with=graph;iterations=30;mi=0.5

.5736 ± .0018

.6617 ± .0061

Entity2Rec;embedding_model=deepwalk_based;embedding_model_kwargs={‘config’: {‘save_weights’: True}, ‘parameters’: {‘walk_len’: 10, ‘p’: 1.0, ‘q’: 1.0, ‘n_walks’: 50, ‘embedding_size’: 64, ‘epochs’: 1}};run_all=False;workers=6;iterations=1;collab_only=False;content_only=False

.4184 ± .0137

.5016 ± .0119

  • Summarized execution time results from experiment_results/douban-movie_times_enriched.csv (configuration: CPU: AMD Ryzen 5 7600 6-Core Processor; RAM: 31GB; GPUs: [‘NVIDIA GeForce RTX 4060’]):

Model

Execution Time (s)

Node2Vec based model + cosine similarity;q=1.0;p=1.0;embedding_size=64

2.824 ± .5424

Node2Vec based model + cosine similarity;q=0.6;p=0.8;embedding_size=64

2.839 ± .2007

TransE based model + cosine similarity;embedding_dim=150;scoring_fct_norm=1;epochs=25;seed=42;triples=ratings

58.86 ± 1.623

TransH based model + cosine similarity;embedding_dim=150;scoring_fct_norm=2;epochs=25;seed=42;triples=ratings

81.12 ± 1.741

TransR based model + cosine similarity;embedding_dim=150;relation_dim=90;scoring_fct_norm=2;epochs=25;seed=42;triples=all

95.10 ± 2.397

TransD based model + cosine similarity;embedding_dim=150;epochs=25;seed=42;triples=ratings

92.93 ± 2.962

TuckER based model + cosine similarity;embedding_dim=200;dropout_0=0.3;dropout_1=0.4;dropout_2=0.5;apply_batch_normalization=True;epochs=25;seed=42;triples=ratings

160.5 ± 3.752

RESCAL based model + cosine similarity;embedding_dim=50;epochs=25;seed=42;triples=ratings

58.57 ± 2.213

DistMult based model + cosine similarity;embedding_dim=50;epochs=25;seed=42;triples=all

46.99 ± 2.122

ComplEx based model + cosine similarity;embedding_dim=100;epochs=25;seed=42

51.79 ± 2.264

RotatE based model + cosine similarity;embedding_dim=200;epochs=25;seed=42;triples=all

56.15 ± 2.132

EPHEN based model + cosine similarity;embedding_model=sentence-transformers/all-roberta-large-v1;embed_with=abstract;iterations=30;mi=0.5

51.79 ± 43.51

EPHEN based model + cosine similarity;embedding_model=deepwalk_based;embedding_model_kwargs={‘walk_len’: 10, ‘p’: 1.0, ‘q’: 1.0, ‘n_walks’: 50, ‘embedding_size’: 64, ‘epochs’: 1};embed_with=graph;iterations=30;mi=0.5

22.54 ± 1.364

Entity2Rec;embedding_model=deepwalk_based;embedding_model_kwargs={‘config’: {‘save_weights’: True}, ‘parameters’: {‘walk_len’: 10, ‘p’: 1.0, ‘q’: 1.0, ‘n_walks’: 50, ‘embedding_size’: 64, ‘epochs’: 1}};run_all=False;workers=6;iterations=1;collab_only=False;content_only=False

9266. ± 104.2

yelp

Experiment ran using the Yelp Challenge dataset with the following presented models and their configurations. The complete configuration can be found in config_files/run_yelp.yml:

  • Summarized results from experiment_results/yelp.csv:

Model

MAP@10

nDCG@10

Node2Vec based model + cosine similarity;q=1.0;p=1.0;embedding_size=64

.01689 ± .00017

.03860 ± .00036

Node2Vec based model + cosine similarity;q=0.6;p=0.8;embedding_size=64

.01688 ± .00029

.03852 ± .00078

TransE based model + cosine similarity;embedding_dim=150;scoring_fct_norm=1;epochs=25;seed=42;triples=ratings

.00001 ± .00000

.00004 ± .00000

TransH based model + cosine similarity;embedding_dim=150;scoring_fct_norm=2;epochs=25;seed=42;triples=ratings

.00010 ± .00001

.00026 ± .00003

TransR based model + cosine similarity;embedding_dim=150;relation_dim=90;scoring_fct_norm=2;epochs=25;seed=42;triples=all

.00000 ± .00000

.00001 ± .00000

TransD based model + cosine similarity;embedding_dim=150;epochs=25;seed=42;triples=ratings

.00001 ± .00000

.00003 ± .00000

TuckER based model + cosine similarity;embedding_dim=200;dropout_0=0.3;dropout_1=0.4;dropout_2=0.5;apply_batch_normalization=True;epochs=25;seed=42;triples=ratings

.00001 ± .00000

.00002 ± .00001

RESCAL based model + cosine similarity;embedding_dim=50;epochs=25;seed=42;triples=ratings

.00006 ± .00001

.00017 ± .00002

DistMult based model + cosine similarity;embedding_dim=50;epochs=25;seed=42;triples=all

.00001 ± .00000

.00002 ± .00000

ComplEx based model + cosine similarity;embedding_dim=100;epochs=25;seed=42

.00002 ± .00000

.00007 ± .00001

RotatE based model + cosine similarity;embedding_dim=200;epochs=25;seed=42;triples=all

.00001 ± .00000

.00004 ± .00001

EPHEN based model + cosine similarity;embedding_model=sentence-transformers/all-roberta-large-v1;embed_with=abstract;iterations=30;mi=0.5

.00001 ± .00000

.00004 ± .00001

EPHEN based model + cosine similarity;embedding_model=deepwalk_based;embedding_model_kwargs={‘walk_len’: 10, ‘p’: 1.0, ‘q’: 1.0, ‘n_walks’: 50, ‘embedding_size’: 64, ‘epochs’: 1};embed_with=graph;iterations=30;mi=0.5

.01683 ± .00015

.03858 ± .00059

  • Summarized execution time results from experiment_results/yelp_times.csv (configuration: CPU: AMD EPYC 7502P 32-Core Processor; RAM: 94GB; GPUs: [‘NVIDIA A2’]):

Model

Execution Time (s)

Node2Vec based model + cosine similarity;q=1.0;p=1.0;embedding_size=64

1307.4 ± 333.51

Node2Vec based model + cosine similarity;q=0.6;p=0.8;embedding_size=64

1417.7 ± 555.66

TransE based model + cosine similarity;embedding_dim=150;scoring_fct_norm=1;epochs=25;seed=42;triples=ratings

6423.5 ± 1493.2

TransH based model + cosine similarity;embedding_dim=150;scoring_fct_norm=2;epochs=25;seed=42;triples=ratings

9996.2 ± 2948.8

TransR based model + cosine similarity;embedding_dim=150;relation_dim=90;scoring_fct_norm=2;epochs=25;seed=42;triples=all

28568. ± 7317.5

TransD based model + cosine similarity;embedding_dim=150;epochs=25;seed=42;triples=ratings

13663. ± 3201.2

TuckER based model + cosine similarity;embedding_dim=200;dropout_0=0.3;dropout_1=0.4;dropout_2=0.5;apply_batch_normalization=True;epochs=25;seed=42;triples=ratings

15845. ± 140.81

RESCAL based model + cosine similarity;embedding_dim=50;epochs=25;seed=42;triples=ratings

3734.3 ± 163.13

DistMult based model + cosine similarity;embedding_dim=50;epochs=25;seed=42;triples=all

7292.4 ± 215.27

ComplEx based model + cosine similarity;embedding_dim=100;epochs=25;seed=42

6905.9 ± 406.81

RotatE based model + cosine similarity;embedding_dim=200;epochs=25;seed=42;triples=all

41117. ± 231.11

EPHEN based model + cosine similarity;embedding_model=sentence-transformers/all-roberta-large-v1;embed_with=abstract;iterations=30;mi=0.5

8067.5 ± 75.832

EPHEN based model + cosine similarity;embedding_model=deepwalk_based;embedding_model_kwargs={‘walk_len’: 10, ‘p’: 1.0, ‘q’: 1.0, ‘n_walks’: 50, ‘embedding_size’: 64, ‘epochs’: 1};embed_with=graph;iterations=30;mi=0.5

4243.7 ± 139.98

amazon-video_games-5

Experiment ran using the Amazon Video-Games core-5 dataset with the following presented models and their configurations. The complete configuration can be found in config_files/run_amazon-video_games-5.yml:

  • Summarized results from experiment_results/amazon-video_games-5.csv:

Model

MAP@5

nDCG@5

Node2Vec based model + cosine similarity;q=1.0;p=1.0;embedding_size=64

.0280 ± .0004

.0367 ± .0006

Node2Vec based model + cosine similarity;q=0.6;p=0.8;embedding_size=64

.0278 ± .0005

.0366 ± .0006

TransE based model + cosine similarity;embedding_dim=150;scoring_fct_norm=1;epochs=25;seed=42;triples=ratings

.0001 ± .0000

.0002 ± .0000

TransH based model + cosine similarity;embedding_dim=150;scoring_fct_norm=2;epochs=25;seed=42;triples=ratings

.0000 ± .0000

.0001 ± .0000

TransR based model + cosine similarity;embedding_dim=150;relation_dim=90;scoring_fct_norm=2;epochs=25;seed=42;triples=all

.0001 ± .0000

.0001 ± .0000

TransD based model + cosine similarity;embedding_dim=150;epochs=25;seed=42;triples=ratings

.0001 ± .0000

.0001 ± .0000

TuckER based model + cosine similarity;embedding_dim=200;dropout_0=0.3;dropout_1=0.4;dropout_2=0.5;apply_batch_normalization=True;epochs=25;seed=42;triples=ratings

.0001 ± .0000

.0001 ± .0000

RESCAL based model + cosine similarity;embedding_dim=50;epochs=25;seed=42;triples=ratings

.0001 ± .0000

.0001 ± .0000

DistMult based model + cosine similarity;embedding_dim=50;epochs=25;seed=42;triples=all

.0001 ± .0000

.0002 ± .0000

ComplEx based model + cosine similarity;embedding_dim=100;epochs=25;seed=42

.0001 ± .0000

.0002 ± .0000

RotatE based model + cosine similarity;embedding_dim=200;epochs=25;seed=42;triples=all

.0001 ± .0000

.0002 ± .0000

EPHEN based model + cosine similarity;embedding_model=sentence-transformers/all-roberta-large-v1;embed_with=abstract;iterations=30;mi=0.5

.0002 ± .0001

.0003 ± .0001

EPHEN based model + cosine similarity;embedding_model=deepwalk_based;embedding_model_kwargs={‘walk_len’: 10, ‘p’: 1.0, ‘q’: 1.0, ‘n_walks’: 50, ‘embedding_size’: 64, ‘epochs’: 1};embed_with=graph;iterations=30;mi=0.5

.0280 ± .0008

.0367 ± .0007

  • Summarized execution time results from experiment_results/amazon-video_games-5.csv (configuration: CPU: AMD EPYC 7502P 32-Core Processor; RAM: 94GB; GPUs: [‘NVIDIA A2’]):

Model

Execution Time (s)

Node2Vec based model + cosine similarity;q=1.0;p=1.0;embedding_size=64

740.5 ± 18.09

Node2Vec based model + cosine similarity;q=0.6;p=0.8;embedding_size=64

736.7 ± 19.41

TransE based model + cosine similarity;embedding_dim=150;scoring_fct_norm=1;epochs=25;seed=42;triples=ratings

570.4 ± 46.34

TransH based model + cosine similarity;embedding_dim=150;scoring_fct_norm=2;epochs=25;seed=42;triples=ratings

862.5 ± 56.28

TransR based model + cosine similarity;embedding_dim=150;relation_dim=90;scoring_fct_norm=2;epochs=25;seed=42;triples=all

11119 ± 119.3

TransD based model + cosine similarity;embedding_dim=150;epochs=25;seed=42;triples=ratings

1192. ± 69.81

TuckER based model + cosine similarity;embedding_dim=200;dropout_0=0.3;dropout_1=0.4;dropout_2=0.5;apply_batch_normalization=True;epochs=25;seed=42;triples=ratings

2375. ± 95.82

RESCAL based model + cosine similarity;embedding_dim=50;epochs=25;seed=42;triples=ratings

473.3 ± 42.67

DistMult based model + cosine similarity;embedding_dim=50;epochs=25;seed=42;triples=all

2966. ± 58.25

ComplEx based model + cosine similarity;embedding_dim=100;epochs=25;seed=42

726.1 ± 57.64

RotatE based model + cosine similarity;embedding_dim=200;epochs=25;seed=42;triples=all

18211 ± 164.9

EPHEN based model + cosine similarity;embedding_model=sentence-transformers/all-roberta-large-v1;embed_with=abstract;iterations=30;mi=0.5

2763. ± 65.41

EPHEN based model + cosine similarity;embedding_model=deepwalk_based;embedding_model_kwargs={‘walk_len’: 10, ‘p’: 1.0, ‘q’: 1.0, ‘n_walks’: 50, ‘embedding_size’: 64, ‘epochs’: 1};embed_with=graph;iterations=30;mi=0.5

1964. ± 24.11