Qualitative Analysis of Text Summarization Techniques and Its Applications in Health Domain

Area of Science:

• Natural Language Processing
• Artificial Intelligence
• Information Retrieval

Background:

• The exponential growth of digital information necessitates efficient methods for data condensation.
• Manual summarization is impractical due to scale and time constraints.
• Automated text summarization offers a viable solution for accessing and utilizing large datasets.

Purpose of the Study:

• To conduct a qualitative analysis of prominent text summarization algorithms.
• To compare the performance of Term Frequency-Inverse Document Frequency (TF-IDF), LexRank, TextRank, BertSum, and PEGASUS.
• To identify the most effective algorithms for both abstractive and extractive summarization tasks.

Main Methods:

• Implementation of five distinct text summarization algorithms: TF-IDF, LexRank, TextRank, BertSum, and PEGASUS.
• Evaluation of algorithm performance on two diverse datasets: Reddit-TIFU and MultiNews.
• Quantitative assessment using the Recall-Oriented Understudy for Gisting Evaluation (ROUGE) metric.

Main Results:

• PEGASUS demonstrated superior performance for abstractive text summarization, achieving the highest average F-score across both datasets.
• TextRank proved most effective for extractive text summarization, yielding the best average F-score.
• Both algorithms showed strong results, indicating their suitability for specific summarization approaches.

Conclusions:

• PEGASUS is recommended for abstractive summarization tasks requiring nuanced understanding and generation.
• TextRank is the preferred choice for extractive summarization, focusing on key sentence extraction.
• The study provides valuable insights for selecting appropriate text summarization algorithms based on task requirements and desired output.