Hypergraphs as Weighted Directed Self-Looped Graphs: Spectral Properties, Clustering, Cheeger Inequality

Authors: Zihao Li, Dongqi Fu, Hengyu Liu, Jingrui He

TMLR 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Additionally, we provide extensive experiments to validate our theoretical findings from an empirical perspective. ... 5 Supportive Experiments In this section, we demonstrate the effectiveness of our Hyper Clus-G on real-world data. We first describe the experimental settings and then discuss the experiment results. Details are provided in Appendix C.
Researcher Affiliation Collaboration Zihao Li EMAIL University of Illinois Urbana-Champaign Dongqi Fu EMAIL Meta Hengyu Liu EMAIL University of Illinois Urbana-Champaign Jingrui He EMAIL University of Illinois Urbana-Champaign
Pseudocode Yes We name this algorithm as Hyper Clus-G, whose pseudo code is given in Algorithm 1 with complexity analyzed in Appendix B. ... Algorithm 1 Hyper Clus-G
Open Source Code No The paper does not explicitly provide a link to the source code for the Hyper Clus-G algorithm described in this paper, nor does it state that the code is being released. It only mentions using official code for some baselines.
Open Datasets Yes We use 9 datasets, all from the UC Irvine Machine Learning Repository. ... Mushroom (https://archive.ics.uci.edu/dataset/73/mushroom) ... Rice (https://archive.ics.uci.edu/dataset/545/rice+cammeo+and+osmancik) ... Car (https://archive.ics.uci.edu/dataset/19/car+evaluation) ... Digit-24 is a subset of Digit (https://archive.ics.uci.edu/dataset/80/optical+recognition+of+ handwritten+digits) ... Covertype (https://archive.ics.uci.edu/dataset/31/covertype) ... Zoo (https://archive.ics.uci.edu/dataset/111/zoo) ... Wine-567 (https://archive.ics.uci.edu/dataset/186/wine+quality) ... Letter (https://archive.ics.uci.edu/dataset/80/optical+recognition+of+handwritten+digits)
Dataset Splits Yes Among the 9 datasets, 5 of them have 2 classes, and 4 of them have at least 3 classes. For 2-class datasets, we apply Hyper Clus-G to partition the whole EDVW hypergraph into 2 clusters. For k-class datasets (k ≥ 3), we call Hyper Clus-G iteratively for k − 1 times to get a k-way clustering. ... The two clusters have 4208 and 3916 instances, respectively.
Hardware Specification Yes We run all our experiments on a Windows 11 machine with a 13th Gen Intel(R) Core(TM) i9-13900H CPU, 64GB RAM, and an NVIDIA RTX A4500 GPU.
Software Dependencies No The Python version in our environment is 3.11.4. In order to run our code, one has to install some other common libraries, including Py Torch, pandas, numpy, scipy, and ucimlrepo.
Experiment Setup No We did not modify other default hyperparameters in node2vec. ... We tune the training epochs near default so that (1) execution time is acceptable; (2) the NCut value and F1 value of the result are both near convergence. We did not modify other default hyperparameters in event2vec.