Compressed Heterogeneous Graph for Abstractive Multi-Document Summarization
Authors: Miao Li, Jianzhong Qi, Jey Han Lau
AAAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experimental results over MULTI-NEWS, WCEP-100, and ARXIV show that HGSUM outperforms state-of-the-art MDS models. We test our proposed model HGSUM and compare it against state-of-the-art abstractive MDS models over several datasets. We also report the results of an ablation study to show the effectiveness of the components of HGSUM. |
| Researcher Affiliation | Academia | Miao Li, Jianzhong Qi, Jey Han Lau School of Computing and Information Systems, The University of Melbourne miao4@student.unimelb.edu.au, {jianzhong.qi, laujh}@unimelb.edu.au |
| Pseudocode | No | The paper does not contain any pseudocode or clearly labeled algorithm blocks. |
| Open Source Code | Yes | The code for our model and experiments is available at: https://github.com/oaimli/HGSum. |
| Open Datasets | Yes | We use MULTI-NEWS (Fabbri et al. 2019), WCEP-100 (Ghalandari et al. 2020), and ARXIV (Cohan et al. 2018) as benchmark English datasets. |
| Dataset Splits | No | The paper mentions tuning hyperparameters based on a "development set" but does not specify the train/validation/test dataset splits (e.g., percentages or sample counts) needed for reproduction. For example: "All other hyper-parameters are tuned based on the development set." |
| Hardware Specification | Yes | All experiments are run on Intel(R) Xeon(R) Gold 6326 CPU @ 2.90GHz with NVIDIA Tesla A100 GPU (40G). |
| Software Dependencies | No | The paper mentions using the "Hugging Face library" but does not provide specific version numbers for any software dependencies, libraries, or frameworks used (e.g., Python, PyTorch, TensorFlow, etc.). |
| Experiment Setup | Yes | The hyper-parameter β is set to 0.5 to balance two loss functions. All other hyper-parameters are tuned based on the development set. We use beam search decoding with beam width 5 to generate the summary. To alleviate overfitting, we apply label smoothing during training with a smoothing factor of 0.1. |