This project focuses on creating a machine learning model to automatically generate concise and engaging news titles from article abstracts using the New York Times (NYT) dataset. The goal is to produce titles that effectively capture the essence of the articles by leveraging advanced natural language processing (NLP) techniques.
- Data Preprocessing: Implements robust data cleaning and preparation techniques for the NYT dataset.
- Model Training: Uses a combination of LSTM/GRU networks with attention mechanisms to grasp the context and semantics of the abstracts.
- Title Generation: Utilizes a pointer-generator network to produce both extractive and abstractive titles, enhancing title summarization accuracy.
- Evaluation Metrics: Applies BLEU and ROUGE scores to evaluate the performance of generated titles against human-written ones.
- Python 🐍
- TensorFlow/Keras
- LSTM/GRU Networks
- Pointer-Generator Networks
- Natural Language Processing (NLP)
- BLEU, ROUGE Metrics
- Clone the repository and install the required dependencies.
- Load and preprocess the NYT dataset for model training.
- Train the LSTM/GRU model with attention mechanisms and evaluate the performance using BLEU and ROUGE scores.
- Generate automatic news titles for new article abstracts using the trained model.
This project showcases the power of deep learning in summarization and title generation, offering a solution that can be extended to various domains beyond news articles.