Multi-Scale Anomaly Detection on Attributed Networks

Authors: Leonardo Gutiérrez-Gómez, Alexandre Bovet, Jean-Charles Delvenne678-685

AAAI 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental The performance of our method is assessed on synthetic and realworld attributed networks and shows superior results concerning state of the art algorithms. Finally, we show the scalability of our approach in large networks employing Chebychev polynomial approximations. We conduct experiments on synthetic and real world benchmarks showing that our method allows to not only recover and rank the so called ground truth anomalies, but also to discover new anomalies jointly with their contexts.
Researcher Affiliation Academia 1Institute for Information and Communication Technologies, Electronics and Applied Mathematics (ICTEAM) 2Center for Operations Research and Econometrics (CORE) Universit e catholique de Louvain, Louvain-la-Neuve, Belgium {alexandre.bovet, jean-charles.delvenne}@uclouvain.be 3Luxembourg Institute of Science and Technology (LIST), Esch-sur-Alzette, Luxembourg leonardo.gutierrez@list.lu
Pseudocode No The paper describes the MADAN approach conceptually and mathematically but does not include structured pseudocode or an algorithm block.
Open Source Code Yes MADAN Python code can be found at https://github.com/leoguti85/MADAN
Open Datasets Yes The Disney and Books datasets (M uller et al. 2013) are co-purchase networks extracted from Amazon. Enron (Metsis and et al. 2006) is a communication network with edges indicating email transmission between people. ... This dataset has been extensively used as benchmark for spam detection (Metsis and et al. 2006).
Dataset Splits No The paper describes dataset generation and properties, and uses metrics for evaluation, but it does not specify any explicit training, validation, or test dataset splits or percentages for any of the datasets used.
Hardware Specification Yes All computations were done on a standard computer Intel(R) Core(TM) i7-4790CPU, 3.60GHz I with 16G of RAM.
Software Dependencies No The paper mentions that the MADAN code is in Python and available on GitHub, but it does not specify any particular software dependencies with version numbers (e.g., specific libraries like PyTorch, scikit-learn, or their versions).
Experiment Setup Yes For each time step, we run the Louvain algorithm 100 times with different random initializations and use Eq. 5 as a scoring function to rank partitions H. We chose m = 30 as in (Hammond, Vandergheynst, and Gribonval 2011). In all cases, we set the parameter σ in Eq. 1 as the standard deviation of the distribution of pairwise distances between node attributes.