Large Language Models-guided Dynamic Adaptation for Temporal Knowledge Graph Reasoning

Authors: Jiapu Wang, Sun Kai, LINHAO LUO, Wei Wei, Yongli Hu, Alan Wee-Chung Liew, Shirui Pan, Baocai Yin

NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental results show that without the need of fine-tuning, LLM-DA significantly improves the accuracy of reasoning over several common datasets, providing a robust framework for TKGR tasks3.
Researcher Affiliation Academia Jiapu Wang1, Kai Sun1 , Linhao Luo2, Wei Wei3, Yongli Hu1 , Alan Wee-Chung Liew4, Shirui Pan4 , Baocai Yin1 1Beijing University of Technology, China, 2Monash University, Australia 3University of Hong Kong, China, 4Griffith University, Australia
Pseudocode No The paper describes methods in text and uses flow diagrams but does not contain a formal "Pseudocode" or "Algorithm" block.
Open Source Code Yes Code and data are available at: https://github.com/jiapuwang/LLM-DA.git
Open Datasets Yes Datasets. ICEWS14 [52] and ICEWS05-15 [52] are the subset of Integrated Crisis Early Warning System (ICEWS)...
Dataset Splits Yes Specifically, the historical data, current data and future data correspond to the training, validation, and test datasets of prior research [46, 22].
Hardware Specification Yes All experiments are implemented on a NVIDIA RTX 3090 GPU with i9-10900X CPU.
Software Dependencies No The paper mentions specific LLM models (Chat GPT4, Llama-2-7b-Co H, Vicuna-7b-Co H, Mixtral-8x7B-Co H, GPT-Neo X) and pre-trained models (Sentence-Bert), but does not provide specific version numbers for these or other software libraries/dependencies.
Experiment Setup Yes Additionally, LLM-DA sets the decay rate λ in Temporal Logical Rules Sampling and Candidate Generation, the threshold θ in Dynamic Adaptation, the min-confidence γ and the parameter α in Candidate Generation on both datasets as follows: λ = 0.1, θ = 0.01, α = 0.9 and γ = 0.01, except for α = 0.8 on ICEWS05-15. The number of iterations for the Dynamic Adaptation is set as 5.