This repository contains the code for our paper: Training with One2MultiSeq: CopyBART for Social Media Keyphrase Generation.

Dataset

The datasets can be downloaded from here

For more details about the Twitter dataset, please reference here or contact us at gaochunyang@mail.hfut.edu.cn

Prepocessing

To preprocess the source data, run: python One2MultiSeq_dataprocess.py

Training

To preprocess the source data, run: python train_One2MultiSeq.py After the training, you can change model_name in line 707 to the path of the trained model(for example, model_name = 'models/temp_model/CMKP/CopyBART_One2MultiSeq_base_epochs-10_learning_rate-5e-05_batch_size-32_seed-100') and set is_train = False in train_One2MultiSeq.py.

Note:

• Please download and unzip the datasets in the ./data directory first.