SKEP-

Introduction

SKEP (SKEP: Sentiment Knowledge Enhanced Pre-training for Sentiment Analysis) is proposed by Baidu in 2020,

SKEP propose Sentiment Knowledge Enhanced Pre-training for sentiment analysis. Sentiment masking and three sentiment pre-training objectives are designed to incorporate various types of knowledge for pre-training model.

More detail: https://aclanthology.org/2020.acl-main.374.pdf

⚠️ attention

Compared with the full version of the ernie_1.0_skep_large_ch, we lost the task_embeddings part in order to adapt to the Bert framework.

Released Model Info

Model Name	Language	Model Structure
skep-ernie1-bert-large	Chinese	Layer:24, Hidden:1024, Heads:24

This released pytorch model is converted from the officially released PaddlePaddle SKEP model and a series of experiments have been conducted to check the accuracy of the conversion.

Official PaddlePaddle SKEP repo:
1. https://github.com/PaddlePaddle/PaddleNLP/blob/develop/paddlenlp/transformers/skep
2. https://github.com/baidu/Senta
Pytorch Conversion repo: Not released yet

How to use

from transformers import AutoTokenizer, AutoModel
tokenizer = AutoTokenizer.from_pretrained("Yaxin/ernie_1.0_skep_large_ch")
model = AutoModel.from_pretrained("Yaxin/ernie_1.0_skep_large_ch")

Citation

@article{tian2020skep,
  title={SKEP: Sentiment knowledge enhanced pre-training for sentiment analysis},
  author={Tian, Hao and Gao, Can and Xiao, Xinyan and Liu, Hao and He, Bolei and Wu, Hua and Wang, Haifeng and Wu, Feng},
  journal={arXiv preprint arXiv:2005.05635},
  year={2020}
}

reference:
https://github.com/nghuyong/ERNIE-Pytorch