Extractive Text Summarization for Turkish: Implementation of Tf-Idf and Pagerank Algorithms

Abstract

Due to the massive amount of information available on the web, reaching the desired content has become more and more difficult. Automatic text summarization helps to solve the problem by minimizing the document size while keeping its core information. In this study, two extractive single document automatic text summarization systems for Turkish are presented which implement the statistical-based TF-IDF algorithm as well as the combination of TF-IDF with the graph-based PageRank algorithm. The study aims to reveal the usability and effectiveness of these algorithms for Turkish documents. Moreover, the results of the TF-IDF implementation and the hybrid approach are compared using the co-selection measures, precision, recall, and F-score. In the evaluation phase, the system-generated summaries are categorized and tested based on their word sizes and the predetermined thresholds and compared against the human-generated summaries. The results indicate that the hybrid system performs better than the TF-IDF system even in lower thresholds, and also both systems are inclined to improve average F-scores in higher threshold generated summarization. © 2023, The Author(s), under exclusive license to Springer Nature Switzerland AG.