Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in [1].
Heterogeneous graph adaptive flow network
Authors: Lu Yiqi, Feng Ji, Wee Peng Tay
TMLR 2024 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experiments on real-world datasets for vertex clustering and vertex classification demonstrate that Heta Flow outperforms other benchmark models and achieves state-of-the-art performance on commonly used benchmark datasets. The codes are available at https://github.com/Anonymized C/Heta Flow. ... In this section, we conduct experiments on the clustering and classification tasks using benchmark datasets. We compare the performance of Heta Flow to other state-of-the-art baseline models. |
| Researcher Affiliation | Academia | Lu Yiqi EMAIL Department of Electrical and Electronic Engineering Nanyang Technological University Ji Feng EMAIL Department of Electrical and Electronic Engineering Nanyang Technological University Tay Wee Peng EMAIL Department of Electrical and Electronic Engineering Nanyang Technological University |
| Pseudocode | Yes | Algorithm 1 Heta Flow forward propagation. Require: The node feature {xi, i V} , |
| Open Source Code | Yes | The codes are available at https://github.com/Anonymized C/Heta Flow. |
| Open Datasets | Yes | ACM. We choose the papers published in SIGMOD, KDD, Mobi COMM, SIGCOMM, and VLDB. ... DBLP. We construct a subset of DBLP by selecting ... IMDB. We construct a heterogeneous graph that consists of ... Freebase. Freebase is a huge knowledge graph of the world s information. ... Following the procedure of a previous survey Yang et al. (2020), we sample a subgraph of 8 categories of vertex types with about 1,000,000 edges. |
| Dataset Splits | No | During the robust test, only part of the training set is available, namely 20%, 40%, 60%, 80%, while the random split of training, validation, and testing sets is fixed. The paper mentions a 'random split of training, validation, and testing sets is fixed' but does not specify the explicit proportions (e.g., 80/10/10) of this fixed split for reproduction. The percentages mentioned are for varying training set sizes in robustness tests, not the primary data partitioning. |
| Hardware Specification | No | The paper does not provide specific details about the hardware (e.g., GPU models, CPU types, or memory) used for running the experiments. |
| Software Dependencies | No | We employ the Adam optimizer (Kingma & Ba, 2015) to optimize the model. The paper mentions the Adam optimizer but does not specify versions of programming languages or key software libraries used for implementation. |
| Experiment Setup | Yes | To find a suitable initial parameter setting, we test multiple initial hyperparameter settings, including: different dimensions of the semantic-level features h from 26 to 21; several numbers of the attention head K including {8, 4, 3, 2, 1}; different numbers of convolution layers from 1 to 4; and various dropout rates from 0.4 to 0.85 with a step size of 0.15; two learning rates including {0.01, 0.05}; two regularization parameters {0.001, 0.00005}. Furthermore, we apply early stopping with a patience of 100. |