Pre-training and multi-task training for federated keyword extraction

The generalization ability of supervised model is relatively weak in keyword extraction technology. In order to effectively improve the robustness and accuracy model, key overcome this problem collect more data for training process. However, text as an privacy information, it harder collect. To solve problem, we apply federal learning use user locally performance. integrate unlabeled utterances semantic parsing proposed a Pre-training multi-task based federated model. models learn local information via unsupervised setting core learns setting.