PTR: Prompt Tuning with Rules for Text Classification

Recently, prompt tuning has been widely applied to stimulate the rich knowledge in pre-trained language models (PLMs) serve NLP tasks. Although achieved promising results on some few-class classification tasks, such as sentiment and natural inference, manually designing prompts is cumbersome. Meanwhile, generating automatically also difficult time-consuming. Therefore, obtaining effective for complex many-class tasks still remains a challenge. In this paper, we propose encode prior of task into rules, then design sub-prompts according finally combine handle task. We name Prompt Tuning method with Rules “PTR”. Compared existing prompt-based methods, PTR achieves good trade-off between effectiveness efficiency building prompts. conduct experiments three including relation classification, entity typing, intent classification. The show that outperforms both vanilla baselines, indicating utilizing rules tuning. source code available at https://github.com/thunlp/PTR.