https://github.com/HeathSun/MediQA-Model

2022 May-Aug, NLP Summer Research Assistant at CS & AI Lab, UNNC. Explored the feasibility of Bert-like models for machine reading comprehension of small text in medical areas.
https://github.com/HeathSun/MediQA-Model

deep-learning natural-language-processing pytorch tensorflow

Last synced: 8 months ago
JSON representation

2022 May-Aug, NLP Summer Research Assistant at CS & AI Lab, UNNC. Explored the feasibility of Bert-like models for machine reading comprehension of small text in medical areas.

• Host: GitHub
• URL: https://github.com/HeathSun/MediQA-Model
• Owner: HShawnSun
• Created: 2022-06-23T06:44:17.000Z (about 4 years ago)
• Default Branch: gh-pages
• Last Pushed: 2025-01-16T03:41:14.000Z (over 1 year ago)
• Last Synced: 2025-01-16T04:36:10.214Z (over 1 year ago)
• Topics: deep-learning, natural-language-processing, pytorch, tensorflow
• Language: Python
• Homepage:
• Size: 102 KB
• Stars: 3
• Watchers: 2
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

          


# **Bert-like Models Feasibility Exploration on Medical Short-Text Q&A Tasks**

In this project, I explored the feasibility of leveraging Bert-like models for machine reading comprehension (MRC) in specialized domains, specifically medical short-text Q&A tasks. The work focused on blending different classes of Bert-like models to improve performance when handling small medical datasets extracted from medical illustration videos.

Key accomplishments and findings include:

- **Model Exploration:** Selected and blended various Bert-like models to enhance performance on small text data from medical contexts.

- **Field-Specific MRC Model Training:** Attempted to train a specialized MRC model tailored to medical Q&A tasks, adjusting hyperparameters and evaluating performance.

- **Performance Testing:** Rigorous testing and evaluation of the model’s ability to comprehend and respond to medical questions.

- **Skill Development:** Improved my Natural Language Processing (NLP) capabilities by working closely with my supervisor and Ph.D. students. Gained a deeper understanding of deep learning (DL) and reinforcement learning (RL) formulas.

- **Research Contribution:** This research contributed to the Ningbo 2025 Science and Technology Innovation Program, advancing knowledge in the application of Bert-like models to medical tasks.

---

### **Prediction Steps**

To execute predictions with the trained model, follow these steps:

1. **Run the Prediction Scripts:**

   - For **关黄母颗粒** question-answering prediction:

     ```bash

     sh test_bert_ghm.sh

     ```

   - For **丁苯酚** question-answering prediction:

     ```bash

     sh test_bert_dbf.sh

     ```

2. **Adjust Storage Path**: When executing the script, make sure to change the storage address as prompted by `metrics.py` at line 730.

---

### **Parameters Overview**

- **`--lm`**: Specify the folder containing the pre-trained model to be loaded.

- **`--do_train`**: Activate training mode.

- **`--evaluate_during_training`**: Enable validation during training.

- **`--do_test`**: Activate prediction mode.

- **`--version_2_with_negative`**: Adapt the model for datasets that may contain both questions with answers and questions without answers.

- **`--threads`**: Specify the number of threads for data processing.