CIAN

This repo shows the source code of EMNLP 2022 paper: Learning Inter-Entity Interaction for Few-Shot Knowledge Graph Completion. In this work, we propose a Cross Interaction Attention Network (CIAN) for few-shot knowledge graph completion.

Running the Experiments

Requirements

• Python 3.6.7
• PyTorch 1.10.0
• cuda 11.1
• GPU 3090

Dataset

We use NELL-One and Wiki-One to test our MetaR, and these datasets were firstly proposed by xiong. The orginal datasets and pretrain embeddings can be downloaded from xiong's repo.

How to run

NELL-One

# NELL-One, 5-shot,
python main.py --fine_tune --lr 8e-5 --few 5 --prefix nelllr8e-5.5shot```

# NELL-One, 3-shot,
python main.py --fine_tune --lr 8e-5 --few 3 --prefix nelllr8e-5.3shot```

Wiki-One

# Wiki-One, 5-shot,
python main.py --fine_tune --lr 2e-4 --few 5 --prefix wikilr2e-4.5shot```

# Wiki-One, 3-shot,
python main.py --fine_tune --lr 2e-4 --few 3 --prefix wikilr2e-4.3shot```

Here are explanations of some important args,

--data_path: "directory of dataset"
--few:       "the number of few in {few}-shot, as well as instance number in support set"
--prefix:    "given name of current experiment"
--fine_tune  "whether to fine tune the pre_trained embeddings"
--device:    "the GPU number"

Normally, other args can be set to default values. See params.py for more details about argus if needed.