Name
Abstract | ||
|---|---|---|
Recognizing relations between entities is a pivotal task of relational learning. Previously, the learning of relation representations from unstructured entity-linked corpora is less studied because of the rich, complicated expressions of relations in human language. This paper aims to learn predictive, interpretable, and robust relation representations from textual data that are effective in different settings, including supervised, distantly supervised, and few-shot learning. Instead of solely relying on the supervision from labels (which could be noisy), we propose to infer a latent prototype for each relation from contextual information to best explore the intrinsic semantics of relations. Prototypes are representations in a latent space abstracting canonical relevance between entities in the textual data. We learn the prototypes with a collaborative metric learning approach that uses hybrid metric functions to measure prototype-statement and statement-statement similarities. With the collaborative strategy, this approach not only helps us to train effective encoders but also outputs meaningful, interpretable latent prototypes for the final classification. Experimental results on several relation learning tasks show that our model significantly outperforms the previous state-of-the-art models. |
Year | Venue | DocType |
|---|---|---|
2021 | ICLR | Conference |
Citations | PageRank | References |
0 | 0.34 | 0 |
Authors | ||
10 | ||
Name | Order | Citations | PageRank |
|---|---|---|---|
| Ning Ding | 1 | 1 | 1.71 |
| Xiaobin Wang | 2 | 94 | 12.59 |
| Yao Fu | 3 | 0 | 0.34 |
| Guangwei Xu | 4 | 9 | 9.18 |
| Rui Wang | 5 | 54 | 7.75 |
| Pengjun Xie | 6 | 0 | 0.68 |
| Shen Ying | 7 | 73 | 23.48 |
| Fei Huang | 8 | 506 | 56.44 |
| Zheng Hai-Tao | 9 | 142 | 24.39 |
| Rui Zhang | 10 | 1208 | 67.26 |