The paper proposes a novel Question-Aware Graph Convolutional Network (QAGCN) model for answering multi-relation questions over knowledge graphs.
The key highlights are:
QAGCN can perform single-step implicit reasoning to answer multi-relation questions, which is simpler, more efficient, and easier to adopt than existing explicit multi-step reasoning-based methods.
QAGCN includes a novel GCN architecture with controlled question-dependent message propagation to enable the implicit reasoning.
Extensive experiments show that QAGCN achieves competitive and even superior performance compared to state-of-the-art explicit-reasoning methods on widely used benchmark datasets.
QAGCN is easier to train than the state-of-the-art reasoning-based method NSM, requiring about half the number of training epochs.
The efficiency evaluation demonstrates that QAGCN can answer questions in real-time, with most steps taking less than 100ms on average.
To Another Language
from source content
arxiv.org
Thông tin chi tiết chính được chắt lọc từ
by Ruijie Wang,... lúc arxiv.org 04-01-2024
https://arxiv.org/pdf/2206.01818.pdfYêu cầu sâu hơn