Knowledge-aware Prompt-tuning with Synergistic Optimization for Relation Extraction

Recently, prompt-tuning has achieved promising results on some few-shot classification tasks. The core idea of prompt-tuning is to insert text pieces, i.e., templates, into the input and transform a classification task into a masked language modeling problem. However, as for relation extraction, determining the appropriate prompt template requires domain expertise. Single label word handcrafted or auto-searched is cumbersome and time-consuming to verify their effectiveness in non-few-shot scenarios. Further, there exist abundant semantic knowledge among the entities and relation labels which cannot be ignored. To this end, we focus on incorporating knowledge into prompt-tuning for relation extraction and propose a Knowledge-aware prompt-tuning with synergistic optimization (KNIGHT) approach. Specifically, we inject entity and relation knowledge into prompt construction with learnable virtual template words and answer words and jointly optimize their representation with knowledge constraints. Extensive experimental results on five datasets with standard and low-resource settings demonstrate the effectiveness of our approach.