We built a end-to-end NLP System with RoBERTa, T5 and T5+Mistral. RoBERTa outperforms T5 and T5+Mistral in pinpointing precise answer spans. Although generative models like T5 and Mistral demonstrated their strength in providing more comprehensive information for high-level queries, the flexibility in generating more elaborate responses come with a trade-off: longer results and less accurate information. The remaining challenge that RoBERTa is faced with would be synthesizing information from multiple sources and reasoning over implicit knowledge.
Before running this project, you must install the required dependencies. This project's dependencies are listed in the requirements.txt file. To install them, run the following command:
pip install -r requirements.txt- Clone this repository to your local machine using git clone .
- Navigate to the cloned directory.
- Install the project dependencies using the command above.
- Proceed with the rest of the installation steps as described.
- Web Script Code: Source code for raw data extraction used in the project.
- data: Directory for training and test data of annotated Q&A pairs.
- system_outputs: Contains output data (answers) from three models.
- Mistral_Model.ipynb: Jupyter notebook for the Mistral model development.
- Model_Evaluation.ipynb: Jupyter notebook for evaluating the models.
- Roberta_Model.ipynb: Jupyter notebook for the Roberta model development.
- T5_Model.ipynb: Jupyter notebook for the T5 model development.
- reference.zip: A compressed file containing reference materials or datasets.
- Ella Liang (jinqiul)
- Ellen Song (yuhans2)
- Alicia Wang (chenqiw)
This project is licensed under the GNU General Public License v2.0 and Apache License v2.0- see the LICENSE folder for details.