Multi-View Empowered Structural Graph Wordification for Language Models

Authors: Zipeng Liu, Likang Wu, Ming He, Zhong Guan, Hongke Zhao, Nan Feng

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

Reproducibility Variable Result LLM Response
Research Type Experimental Our experimental evaluations on standard graph tasks demonstrate competitive performance against other state-ofthe-art (SOTA) approaches. Additionally, our framework ensures certain visual interpretability, efficiency, and robustness, marking the promising successful endeavor to achieve token-level alignment between LLMs and GNNs.
Researcher Affiliation Collaboration 1College of Management and Economics, Tianjin University 2Laboratory of Computation and Analytics of Complex Management Systems (CACMS), Tianjin University 3AI Lab, Lenovo Research
Pseudocode No The paper describes the methodology using prose and equations, but it does not contain a clearly labeled 'Pseudocode' or 'Algorithm' block.
Open Source Code Yes Code https://github.com/Timothy914/Dr.E
Open Datasets Yes To evaluate the efficacy of our framework, Dr.E is tested on three benchmark datasets: Cora (Mc Callum et al. 2000), Pub Med (Sen et al. 2008), and OGBN-Arxiv (Hu et al. 2020).
Dataset Splits Yes We adhere to the dataset splits commonly employed by other methods, such as those detailed in (He et al. 2023).
Hardware Specification Yes Our experiments are conducted using 2 NVIDIA A800-SXM4-80GB GPUs.
Software Dependencies No The paper mentions using "Llama2-7B" and "Lo RA PEFT adjustments" but does not provide specific version numbers for these or other software components.
Experiment Setup Yes We implement Lo RA PEFT adjustments for Llama2-7B and establish two distinct learning rates for the GNN encoder and LLM decoder, set at 1 10 3 and 1 10 4, respectively, with a weight decay of 5 10 4. The hidden dimension for the SAGE convolution is 4096, matching the token embedding of Llama.