lpover/sourcep

SourceP

The project's code and dataset set will be open-sourced after the paper is published.

This paper has been accepted by ICASSP 2024.

Paper address: https://ieeexplore.ieee.org/document/10448439

Detection

python run_classifier.py parameters

Evaluation

python3 evaluator.py -a dataset/test.txt -p result/predictions.txt 2>&1| tee result/score.log

Experimental Results

RQ1: Performance comparison with state-of-the-art methods.

Method	Recall	Precision	F-score
XGBoost-TF-IDF	0.234	0.882	0.370
SadPonzi	0.52	0.59	0.55
SVM-NC	0.375	0.923	0.533
Ridge-NC	0.453	0.829	0.586
MulCas	0.674	0.951	0.789
SourceP	0.887	0.956	0.918

RQ2: Sustainability of the model compared to other state-of-the-art methods.

Method	Metric	P2	P3	P4	P5
SadPonzi	Precision	0.33	0.42	0.18	0.24
	Recall	1.0	0.71	0.25	0.18
	F-score	0.5	0.53	0.21	0.20
------------	------------	--------	--------	--------	--------
MulCas	Precision	0.88	0.96	0.81	0.95
	Recall	0.38	0.32	0.94	0.67
	F-score	0.53	0.48	0.87	0.79
------------	------------	--------	--------	--------	--------
SourceP	Precision	0.99	0.97	0.88	0.96
	Recall	0.55	0.89	0.90	0.89
	F-score	0.59	0.92	0.88	0.92

RQ3: Ablation experiments.

Method	Recall	Precision	F-score
SourceP	0.887	0.956	0.918
-w/o EdgePred	0.867	0.919	0.891
-w/o NodeAlign	0.821	0.914	0.860
-w/o Data Flow	0.806	0.909	0.847

RQ4: Generalization ability of SourceP.

Method	Recall	Precision	F-score
SourceP	0.90	0.92	0.91