This is the final project of CMPT 413: Computational Linguistics, instructed by Prof. Angel Chang of School of Computing Sciences, Simon Fraser Univerisity.
The topic of this project is "Prompt-based Text Matching Methods for Fake News Stance Detection", presented by group AWSL, consisting of Zeyong Jin, Yuqing Wu and Zhi Feng.
This project dedicated to using the BERT pretraining model to implement the classifier of stance detection related to fake news recognition. This paper uses the BERT model to process news headlines and news article body and based on fine-tuning the model to continue training the BERT model, and finally detects the stance relationship between news headlines and news article body. This paper implements a fake news classification model based on BERT and compares the results of the BERT model with the four baseline algorithms. BERT model is used to process FNC task data, and an accuracy of 90.37% is obtained. There are still some limitations in this project due to our shortages of related knowledge, for example, we did not have the flexibility to modify the optimal function of the BERT model. It is a promising field to explore other behaviours or methods which can be involved in fake news detection in future. Therefore, further research is strongly needed, and we will keep on doing research in this field when time permits.
Feel free to check further details from our paper.
You can view the video of the presentation here.
The slides of this presentation are as follows.
Bình luận