Cross-Domain Few-Shot Relation Extraction via Representation Learning and Domain Adaptation

Zhongju Yuan

Keywords: few-shot, domain adaptation, relation extraction

Abstract: Few-shot relation extraction aims to recognize novel relations with few labeled sentences in each relation. Previous metric-based few-shot relation extraction methods classify by comparing the embeddings of query sentence embedding with those prototypes generated by the few labeled sentences embedding using a learned metric function. However, the generalization ability of these methods on unseen relations in different domains is limited, since these domains always have significant discrepancies from those in the training dataset. Because the prototype is essential for extracting relations between entities in the latent space. To extract new relations in various domains more effectively, we propose to learn more interpretable and robust prototypes by learning from prior knowledge and intrinsic semantics of relations. We improve the prototype representation of relations more efficiently by using prior knowledge to explore the connections between relations. The geometric interpretability of the prototype is improved by making the classification margins between sentence embedding clearer through contrastive learning. Besides, for better-extracting relations in different domains, using a cross-domain approach makes the generation process of the prototype take into account the gap between other domains, which makes the prototype more robust. The experimental results on the benchmark FewRel dataset demonstrate the advantages of the proposed method over some state-of-the-art methods.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics

Submission Guidelines: Yes

Please Choose The Closest Area That Your Submission Falls Into: Applications (eg, speech processing, computer vision, NLP)