Learning Relation Prototype from Unlabeled Texts for Long-tail Relation Extraction

Relation Extraction (RE) is a vital step to complete Knowledge Graph (KG) by extracting entity relations from texts. However, it usually suffers the long-tail issue. This paper proposes novel approach learn relation prototypes unlabeled texts, facilitate RE transferring knowledge types with sufficient training data. We as an implicit factor between entities, which reflects meanings of and their proximities. construct co-occurrence graph capture both first-order second-order proximities for embedding learning. By optimize distance pairs corresponding prototypes, our method can be easily adapted almost arbitrary frameworks. Thus, learning infrequent or even unseen will benefit semantically proximate through entities large-scale textual information. Extensive experiments on two publicly available datasets present promising improvements (4.1% F1 average). Ablation studies relations, main components, different models demonstrate effectiveness learned prototypes. Finally, we analyze several example cases give intuitive impressions qualitative analysis. Our codes data found in https://github.com/CrisJk/PA-TRP.