https://github.com/ivanrj7j/ner-bilstm
This project showcases a custom Named Entity Recognition (NER) model using Bidirectional Long Short-Term Memory (BiLSTM) layers for precise entity predictions. The model is trained using a specific dataset and is integrated with a simple user interface built with HTML, CSS, and Flask.
https://github.com/ivanrj7j/ner-bilstm
ai lstm ner nlp tensorflow
Last synced: 2 months ago
JSON representation
This project showcases a custom Named Entity Recognition (NER) model using Bidirectional Long Short-Term Memory (BiLSTM) layers for precise entity predictions. The model is trained using a specific dataset and is integrated with a simple user interface built with HTML, CSS, and Flask.
- Host: GitHub
- URL: https://github.com/ivanrj7j/ner-bilstm
- Owner: ivanrj7j
- Created: 2024-06-27T13:25:47.000Z (about 2 years ago)
- Default Branch: main
- Last Pushed: 2024-06-27T14:06:57.000Z (about 2 years ago)
- Last Synced: 2025-04-08T20:23:55.541Z (about 1 year ago)
- Topics: ai, lstm, ner, nlp, tensorflow
- Language: Jupyter Notebook
- Homepage:
- Size: 70.6 MB
- Stars: 1
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# Custom Trained NER using Bidirectional LSTM
## Introduction
This project showcases a custom Named Entity Recognition (NER) model using Bidirectional Long Short-Term Memory (BiLSTM) layers for precise entity predictions. The model is trained using a specific dataset and is integrated with a simple user interface built with HTML, CSS, and Flask.
## Features
- **Custom NER Model**: Utilizes Bidirectional LSTM layers for accurate predictions.
- **User Interface**: Simple dark-themed UI for inputting text and viewing NER outputs.
- **Pre-trained Model**: The project includes the pre-trained NER model, ready for use.
## Technologies Used
- TensorFlow
- Flask
- HTML & CSS
## Dataset
The model is trained using the NER dataset provided by [Raj Nath Patel on Kaggle](https://www.kaggle.com/datasets/rajnathpatel/ner-data).
## Project Motivation
The primary motivation for this project was to gain hands-on experience with machine learning and NER models, and to showcase these skills in a practical application.
## Data Preprocessing
The provided dataset was already well-preprocessed, so the primary task was to create a custom data pipeline for feeding the data into the model.
## Model Architecture
The NER model is built using a series of Bidirectional LSTM layers with dropout for regularization. Below is a table summarizing the model architecture:
| Layer | Output Shape | Additional Parameters |
|----------------|--------------------|-----------------------------------------------------|
| Input | (None, maxLen) | `maxLen` = 35 |
| Embedding | (None, maxLen, 200)| `input_dim`=totalTokens, `output_dim`=200 |
| Bidirectional LSTM (LSTM1) | (None, maxLen, 256)| `units`=128, `return_sequences`=True, `recurrent_dropout`=0.1 |
| Bidirectional LSTM (LSTM2) | (None, maxLen, 256)| `units`=128, `return_sequences`=True, `recurrent_dropout`=0.1 |
| Bidirectional LSTM (LSTM3) | (None, maxLen, 256)| `units`=128, `return_sequences`=True, `recurrent_dropout`=0.1 |
| Dropout | (None, maxLen, 256)| `rate`=0.5 |
| TimeDistributed Dense (Classifier) | (None, maxLen, numClasses)| `units`=numClasses, `activation`="softmax" |
## Training Process
The model was trained for 10 epochs, which took approximately 15 minutes on the available hardware.
## Evaluation Metrics
The model achieved a validation accuracy of 97.47%.
## User Interface
The UI features a text input area for entering sentences and an output area for displaying the NER results. The labels are color-coded for easy identification of different entity types.
## Prerequisites
- TensorFlow
## Installation and Setup
1. Clone the repository:
```sh
git clone https://github.com/ivanrj7j/NER-BiLSTM
```
2. Install the required packages:
```sh
pip install -r requirements.txt
```
3. Navigate to the interface folder:
```sh
cd interface
```
4. Run the application:
```sh
python main.py
```
## Usage
- Open a browser and go to `http://127.0.0.1:5000/`.
- Enter text in the input area and click the "Submit" button.
- View the NER results in the output area.
## Known Issues
- The model can only process up to 35 tokens at once, which may result in loss of context during prediction.
## Acknowledgments
- [Raj Nath Patel](https://www.kaggle.com/rajnathpatel) for the dataset.
## License
This project is licensed under the MIT License.
## Contact
For any questions or suggestions, feel free to contact me at:
- **Email**: ivanrj7j@gmail.com
- **GitHub**: [ivanrj7j](https://github.com/ivanrj7j)
---
Feel free to make any adjustments or let me know if there's anything more you'd like to add.