A Hybrid Approach to Automatic Text Summarization of Turkish Texts: Integrating Extractive Methods with LLMs

Text summarization, which involves reducing the length of long texts without losing their semantic meaning, is an increasingly important topic in this era where time is of the essence. Automatic text summarization is a relatively challenging topic in the field of Natural Language Processing. In summarization studies conducted in the Turkish language, it is observed that extractive methods are predominantly used. Particularly with the recent proliferation of Large Language Models, the number of studies in this area has increased. In this study, a method combining abstractive and extractive approaches for the automatic text summarization of Turkish texts is proposed. The Turkish news articles section of the multi-lingual summarization dataset was chosen as the dataset. The method employed the TextRank and Latent Semantic Analysis algorithms as the extractive method and the GPT algorithm, one of the most popular large language models, as the abstractive method. The generated summaries obtained using the hybrid method were evaluated with ROUGE and BLEU metrics, and promising results were achieved.