- Mayank Bhandari
- Ouc-Houang Fogoum Philippe Jacques
- Ramesh Singh
November 23, 2023
Text summarization is a crucial aspect of natural language processing, aiming to condense lengthy texts while retaining essential information. In this project, the focus was on utilizing the google/pegasus-cnn dailymail model for abstractive summarization.
The primary goal of the project was to automatically generate concise and coherent summaries for given texts using the abstractive summarization capabilities of the selected model.
Samsum Dataset https://huggingface.co/datasets/samsum
The google/pegasus-cnn dailymail model was chosen for its specialization in abstractive summarization. This model has been pre-trained on a diverse dataset, including the CNN/Daily Mail dataset, making it suitable for a wide range of summarization tasks.
We employed a fine-tuned Pegasus model tokenizer to break down input text into meaningful units. This tailored tokenizer, designed for optimal compatibility with the Pegasus model, enhances the text summarization pipeline’s effectiveness, ensuring coherent and meaningful abstractive summaries.
The model’s performance was evaluated using standard ROUGE scores, including "rouge1," "rouge2," "rougeL," and "rougeLsum." These metrics provided a comprehensive assessment of the quality of the generated summaries compared to human-authored reference summaries.
For the model evaluation, we are using ROUGE score.
| Rouge Names | Result |
|---|---|
| rouge1 | 0.015465 |
| rouge2 | 0.000297 |
| rougeL | 0.015503 |
| rougeLsum | 0.015514 |
Figure 1: Text Summary
The google/pegasus-cnn dailymail model consistently produced coherent and relevant summaries across a diverse set of input texts, as evidenced by the robust ROUGE scores.
In conclusion, the integration of the google/pegasus-cnn dailymail model, fine-tuned and evaluated using ROUGE scores, proved successful in addressing the text summarization task. The abstractive summarization approach enhances the model’s ability to distill essential information from input texts, making it a valuable tool for a variety of applications.