liuyuaa / RAM

Role-Aware Modeling for N-ary Relational Knowledge Bases

Role-Aware Modeling for N-ary Relational Knowledge Bases

This repository is the official implementation of our WWW'2021 paper "Role-Aware Modeling for N-ary Relational Knowledge Bases".

Requirements

To install requirements:

python 3.7.4
pytorch 1.1

Running a model

To train (and evaluate) the model in the paper, run this command:

python main.py --dataset FB-AUTO --num_iterations 200 --batch_size 64 --lr 0.005 --dr 0.995 --K 10 --rdim 50 --m 2 --drop_role 0.2 --drop_ent 0.4 --eval_step 1 --valid_patience 10 -ary 2 -ary 4 -ary 5

📋 The ary append for WikiPeople is

-ary 2 -ary 3 -ary 4 -ary 5 -ary 6 -ary 7 -ary 8 -ary 9

📋 The ary append for JF17K is

-ary 2 -ary 3 -ary 4 -ary 5 -ary 6

📋 The ary append for WN18/FB15k is

-ary 2

Evaluation interval is determined by the parameter "eval_step"

Hyperparameters Settings

Dataset	d	K	lr	dr	drop_role	drop_ent	batch_size
WikiPeople	25	10	0.003	0.995	0.0	0.2	64
JF17K	50	10	0.005	0.995	0.2	0.4	64
FB-AUTO	50	10	0.005	0.995	0.2	0.4	64
WN18	50	10	0.002	0.995	0.0	0.4	128
FB15k	100	50	0.001	0.99	0.2	0.0	128

Results

Our model achieves the following performance on WikiPeople, JF17K, FB-AUTO, WN18, and FB15k.

Dataset	MRR	Hits@10	Hits@1
WikiPeople	0.380	0.541	0.278
JF17K	0.539	0.690	0.463
FB-AUTO	0.830	0.876	0.803
WN18	0.947	0.952	0.943
FB15k	0.803	0.882	0.756

Reference

@inproceddings{liu2021ram,
	title 	  = {Role-Aware Modeling for N-ary Relational Knowledge Bases},
	author	  = {Liu, Yu and Yao, Quanming and Li, Yong},
	booktitle = {The World Wide Web Conference},
	year      = {2021},
}