LDM-DA

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

Official Implementation of "Large Language Models-guided Dynamic Adaptation for Temporal Knowledge Graph Reasoning".

LLM-DA first analyzes historical data to extract temporal rules and utilizes the powerful generative capabilities of LLMs to generate general rules. Subsequently, LLM-DA updates these rules using current data. Finally, the updated rules are applied to predict future events.

Requirements

pip install -r requirements.txt

Set your OpenAI API key in .env file

Mining Rules with LLM-DA

1. Temporal Logical Rules Sampling

python rule_sampler.py -d ${DATASET} -m 3 -n 200 -p 16 -s 12 --is_relax_time No

2. Rule Generation & Dynamic Adaptation

python Iteration_reasoning.py -d ${DATASET} --model_name gpt-3.5-turbo-0125 -f 50 -l 5 --is_rel_name Yes 

3. Rank Rules

python rank_rule.py -p copy_gpt-3.5-turbo-0125-top-0-f-10-l-10 -d ${DATASET} 

4. Candidate Reasoning

python reasoning.py -d ${DATASET} -r confidence.json -l 1 2 3 -p 8 --min_conf 0.01 --weight_0 0.5 --gpu 0 --top_k 20 --window 0

5. Evaluate

python evaluate.py -d ${DATASET} -c 'llm_test_apply_all_conf_cands_r[1,2,3]_w0_score_12[0.1,0.5,'\''TLogic'\'',0.0,0.01,0]_top_20_et_origin.json' --graph_reasoning_type TiRGN --rule_weight 0.9

Reproduce LLM-DA results with mined rules

ICEWS14

python rule_sampler.py -d icews14 -m 3 -n 200 -p 16 -s 12 --is_relax_time No

python Iteration_reasoning.py -d icews14 --model_name gpt-3.5-turbo-0125 -f 50 -l 5 --is_rel_name Yes 

python rank_rule.py -p copy_gpt-3.5-turbo-0125-top-0-f-10-l-10 -d icews14

python reasoning.py -p copy_gpt-3.5-turbo-0125-top-0-f-10-l-10 -d icews14 

python evaluate.py -d icews14  -c 'llm_test_apply_all_conf_cands_r[1,2,3]_w0_score_12[0.1,0.5,'\''TLogic'\'',0.0,0.01,0]_top_20_et_origin.json' --graph_reasoning_type TiRGN --rule_weight 0.9

ICEWS05-15

python rule_sampler.py -d icews0515 -m 3 -n 200 -p 16 -s 12 --is_relax_time No

python Iteration_reasoning.py -d icews0515 --model_name gpt-3.5-turbo-0125 -f 50 -l 5 --is_rel_name Yes 

python rank_rule.py -p copy_gpt-3.5-turbo-0125-top-0-f-10-l-10 -d icews0515

python reasoning.py -p copy_gpt-3.5-turbo-0125-top-0-f-10-l-10 -d icews0515 

python evaluate.py -d icews0515  -c 'llm_test_apply_all_conf_cands_r[1,2,3]_w0_score_12[0.1,0.5,'\''TLogic'\'',0.0,0.01,0]_top_20_et_origin.json' --graph_reasoning_type TiRGN --rule_weight 0.8

Results

Bibinfo

If you found this repo helpful, please help us by citing this paper:

@article{wang2024large,
  title={Large Language Models-guided Dynamic Adaptation for Temporal Knowledge Graph Reasoning},
  author={Wang, Jiapu and Sun, Kai and Luo, Linhao and Wei, Wei and Hu, Yongli and Liew, Alan Wee-Chung and Pan, Shirui and Yin, Baocai},
  journal={arXiv preprint arXiv:2405.14170},
  year={2024}
}