Event detection (ED) aims at detecting event trigger words in sentences and classifying them into specific event types . In real-world applications, ED typically does not have sufficient labelled data, thus can be formulated as afew-shot learning problem . We propose a novel knowledge-based few-shot event detection method which uses a definition-based encoder to introduce external event knowledge asthe knowledge prior of event types. We introduce adaptive knowledge-enhanced Bayesian meta-learning method to dynamicallyadjust the knowledge prior . Experiments show our method consistently outperforms a number of baselines by at least 15absolute F1 points under the same few shot settings .
Author(s) : Shirong Shen, Tongtong Wu, Guilin Qi, Yuan-Fang Li, Gholamreza Haffari, Sheng BiLinks : PDF - Abstract
Code :
Keywords : event - knowledge - shot - method - learning -