Graph-Hist: Graph Classification from Latent Feature Histograms with Application to Bot Detection

Authors: Thomas Magelinski, David Beskow, Kathleen M. Carley5134-5141

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

Reproducibility Variable Result LLM Response
Research Type Experimental We show that Graph-Hist improves state of the art performance on true social media benchmark datasets, while still performing well on other benchmarks. Finally, we demonstrate Graph-Hist s performance by conducting bot detection in social media. We apply Graph-Hist to classify these conversational graphs. In the process, we confirm that social media graphs are different than most baselines and that Graph-Hist outperforms existing botdetection models. The mean accuracy and its standard deviation is reported for each dataset in Table 3. Graph-Hist advances state-of-the-art classification in all 3 of the social media benchmarks. It also beats stateof-the-art results for IMDB-B, and obtains second place results for the remaining two datasets.
Researcher Affiliation Academia Thomas Magelinski, David Beskow, Kathleen M. Carley CASOS, School of Computer Science, Carnegie Mellon University 5000 Forbes Ave. Pittsburgh, PA, 15213 {tmagelin, dbeskow, kathleen.carley}@cs.cmu.edu
Pseudocode No The paper describes its architecture and procedures through text and mathematical equations, but it does not include a clearly labeled pseudocode block or algorithm.
Open Source Code No The paper does not provide any statement about making its source code publicly available or provide a link to a code repository.
Open Datasets Yes The datasets have been obtained from Kersting et al s collection, but were created by Yanardag and Vishwanathan (Kersting et al. 2016; Yanardag and Vishwanathan 2015).
Dataset Splits Yes We terminated training after 9 consecutive epochs without progress in the testing loss. We then tuned parameters to each dataset in the search space h [2, 4, 6, 8], c [32, 64, 128, 256], ρ [0.2, 0.8]. Parameters were selected based on their performance on the test set. We performed 10-fold cross-validation on each of the datasets using the parameters in Table 2. The new stopping threshold was given by F1 score in the validation set.
Hardware Specification No The paper states that Graph-Hist was implemented in PyTorch, but it does not provide any specific details about the hardware (e.g., CPU, GPU models, memory) used to run the experiments.
Software Dependencies No The paper mentions that "Graph-Hist was implemented in Py Torch." However, it does not specify the version of PyTorch or any other software dependencies with version numbers.
Experiment Setup Yes We used the Reduce LROn Plateau scheduler with an initial learning rate of α = 1e 4, a factor of 0.5, a patience of 2, a cooldown of 0, and a minimum learning rate of 1e 7. We used stochastic gradient descent with a mini-batch size of 32. We terminated training after 9 consecutive epochs without progress in the testing loss. We then tuned parameters to each dataset in the search space h [2, 4, 6, 8], c [32, 64, 128, 256], ρ [0.2, 0.8]. Parameters were selected based on their performance on the test set. The final parameters are shown in Table 2.