Knowledge-Fusion-Model-Based-on-Attention

The Source Code for Knowledge Fusion Model Based on Attention in Postgraduate Thesis

Requirements

• PyTorch 1.0.1
• tensorboardX
• tqdm

Prepare data

mkdir datasets && cd datasets && wget https://sites.cs.ucsb.edu/~xwhan/datasets/webqsp.tar.gz && tar -xzvf webqsp.tar.gz && cd ..

Full KB setting

CUDA_VISIBLE_DEVICES=0 python train.py --model_id KAReader_full_kb --max_num_neighbors 50 --label_smooth 0.1 --data_folder datasets/webqsp/full/ 

Incomplete KB setting

Note: The Hits@1 should match or be slightly better than the number reported in the paper. More tuning on threshold should give you better F1 score.

30% KB

CUDA_VISIBLE_DEVICES=0 python train.py --model_id KAReader_kb_03 --max_num_neighbors 50 --use_doc --data_folder datasets/webqsp/kb_03/ --eps 0.05

10% KB

CUDA_VISIBLE_DEVICES=0 python train.py --model_id KAReader_kb_01 --max_num_neighbors 50 --use_doc --data_folder datasets/webqsp/kb_01/ --eps 0.05

50% KB

CUDA_VISIBLE_DEVICES=0 python train.py --model_id KAReader_kb_05 --num_layer 1 --max_num_neighbors 100 --use_doc --data_folder datasets/webqsp/kb_05/ --eps 0.12 --seed 3 --hidden_drop 0.05