MartinXPN / abcde

ABCDE: Approximating Betweenness-Centrality ranking with progressive-DropEdge

ABCDE

ABCDE: Approximating Betweenness-Centrality ranking with progressive-DropEdge

This work was published in PeerJ Computer Science journal: https://peerj.com/articles/cs-699/

Link to the Overleaf project (PeerJ draft): https://www.overleaf.com/read/tphdqhycvwfk

ABCDE model architecture

Each Transition block is a set of {Linear → LayerNorm → PRelu → Dropout} layers, while each GCN is a set of {GCNConv → PReLU → LayerNorm → Dropout}. + symbol is the concatenation operation. Each MaxPooling operation extracts the maximum value from the given GCN block.

To reproduce the results

# This will run the ABCDE model on both real-world and synthetic datasets and report the results
docker run martin97/abcde:latest

To run the training script

# Modify the hyperparameters in `train.py` and run it
python train.py

To evaluate the model

Download the evaluation data

cd datasets
./download.sh

*Optionally download the model reported in the paper

mkdir models && cd models
wget https://github.com/MartinXPN/abcde/releases/download/v1.0.0/best.ckpt

Then run the prediction script on the target dataset

# For a specific dataset
python predict.py real --model_path experiments/latest/models/best.ckpt \
                       --data_test datasets/real/amazon.txt \
                       --label_file datasets/real/amazon_score.txt

# To run the whole evaluation
python predict.py all --model_path experiments/latest/models/best.h5py --datasets_dir datasets

Obtained results so far v1.0.0 and comparison with the original DrBC paper and sampling based benchmarks

Obtained results on Real datasets (Model was run on 512GB CPU machine with 80 cores):

Dataset	ABRA	RK	KADABRA	Node2Vec	DrBC	ABCDE
			Top-1%
com-youtube	95.7	76.0	57.5	12.3	73.6	77.1
amazon	69.2	86.0	47.6	16.7	86.2	92.0
Dblp	49.7	NA	35.2	11.5	78.9	79.8
cit-Patents	37.0	74.4	23.4	0.04	48.3	50.2
com-lj	60.0	54.2	31.9	3.9	67.2	70.9
			Top-5%
com-youtube	91.2	75.8	47.3	18.9	66.7	75.1
amazon	58.0	59.4	56.0	23.2	79.7	88.0
Dblp	45.5	NA	42.6	20.2	72.0	73.7
cit-Patents	42.4	68.2	25.1	0.29	57.5	58.3
com-lj	56.9	NA	39.5	10.35	72.6	75.7
			Top-10%
com-youtube	89.5	100.0	44.6	23.6	69.5	77.6
amazon	60.3	100.0	56.7	26.6	76.9	85.6
Dblp	100.0	NA	50.4	27.7	72.5	76.3
cit-Patents	50.9	53.5	21.6	0.99	64.1	64.9
com-lj	63.6	NA	47.6	15.4	74.8	78.0
			Kendall Tau
com-youtube	56.2	13.9	NA	46.2	57.3	59.8
amazon	16.3	9.7	NA	44.7	69.3	77.7
Dblp	14.3	NA	NA	49.5	71.9	73.7
cit-Patents	17.3	15.3	NA	4.0	72.6	73.5
com-lj	22.8	NA	NA	35.1	71.3	71.8
			Time/s
com-youtube	72898.7	125651.2	116.1	4729.8	402.9	26.7
amazon	5402.3	149680.6	244.7	10679.0	449.8	63.5
Dblp	11591.5	NA	398.1	17446.9	566.7	104.9
cit-Patents	10704.6	252028.5	568.0	11729.1	744.1	163.9
com-lj	34309.6	NA	612.9	18253.6	2274.2	271.0

Obtained results on Synthetic datasets (Model was run on 512GB CPU machine with 80 cores):

Scale	ABRA	RK	k-BC	KADABRA	Node2Vec	DrBC	ABCDE
			Top-1%
5000	97.8±1.5	96.8±1.7	94.1±0.8	76.2±12.5	19.1±4.8	96.5±1.8	97.5±1.3
10000	97.2±1.2	96.4±1.3	93.3±3.1	74.6±16.5	21.2±4.3	96.7±1.2	96.9±0.9
20000	96.5±1.0	95.5±1.1	91.6±4.0	74.6±16.7	16.1±3.9	95.6±0.9	96.0±1.2
50000	94.6±0.7	93.3±0.9	90.1±4.7	73.8±14.9	9.6±1.3	92.5±1.2	93.6±0.9
100000	92.2±0.8	91.5±0.8	88.6±4.7	67.0±12.4	9.6±1.3	90.3±0.9	91.8±0.6
			Top-5%
5000	96.9±0.7	95.6±0.9	89.3±3.9	68.7±13.4	23.3±3.6	95.9±0.9	97.8±0.7
10000	95.6±0.8	94.1±0.8	88.4±5.1	70.7±13.8	20.5±2.7	95.0±0.8	97.0±0.6
20000	93.9±0.8	92.2±0.9	86.9±6.2	69.1±13.5	16.9±2.0	93.0±1.1	95.2±0.8
50000	90.1±0.8	88.0±0.8	84.4±7.2	65.8±11.7	13.8±1.0	89.2±1.1	92.1±0.6
100000	85.6±1.1	87.6±0.5	82.4±7.5	57.0±9.4	12.9±1.2	86.2±0.9	89.7±0.5
			Top-10%
5000	96.1±0.7	94.3±0.9	86.7±4.5	67.2±12.5	25.4±3.4	94.8±0.7	97.6±0.4
10000	94.1±0.6	92.2±0.9	86.0±5.9	67.8±13.0	25.4±3.4	94.0±0.9	96.8±0.6
20000	92.1±0.8	90.6±0.9	84.5±6.8	66.1±12.4	19.9±1.9	91.9±0.9	94.9±0.5
50000	87.4±0.9	88.2±0.5	82.1±8.0	61.3±10.4	18.0±1.2	87.9±1.0	91.7±0.6
100000	81.8±1.5	87.4±0.4	80.1±8.2	52.4±8.2	17.3±1.3	85.0±0.9	89.4±0.5
			Kendall Tau
5000	86.6±1.0	78.6±0.6	66.2±11.4	NA	11.3±3.0	88.4±0.3	93.7±0.2
10000	81.6±1.2	72.3±0.6	67.2±13.5	NA	8.5±2.3	86.8±0.4	93.3±0.1
20000	76.9±1.5	65.5±1.2	67.1±14.3	NA	7.5±2.2	84.0±0.5	92.1±0.1
50000	68.2±1.3	53.3±1.4	66.2±14.1	NA	7.1±1.8	80.1±0.5	90.1±0.2
100000	60.3±1.9	44.2±0.2	64.9±13.5	NA	7.1±1.9	77.8±0.4	88.4±0.2
			Time/s
5000	18.5±3.6	17.1±3.0	12.2±6.3	0.6±0.1	32.4±3.8	0.3±0.0	0.5±0.0
10000	29.2±4.8	21.0±3.6	47.2±27.3	1.0±0.2	73.1±7.0	0.6±0.0	0.6±0.0
20000	52.7±8.1	43.0±3.2	176.4±105.1	1.6±0.3	129.3±17.6	1.4±0.0	0.9±0.0
50000	168.3±23.8	131.4±2.0	935.1±505.9	3.9±1.0	263.2±46.6	3.9±0.2	2.2±0.0
100000	380.3±63.7	363.4±36.3	3069.2±1378.5	7.2±1.8	416.2±37.0	8.2±0.3	3.2±0.0