GPT-2 Modelleri Kullanarak Haber Özetinden Başlık ve Başlıktan Haber Özeti Üretimi Generation of Headline from News Summary and News Summary from Headline Using GPT-2 Models

Generative Pre-trained Transformer 2 (GPT-2) has an important role among the generative language models used for next token prediction. Studies using GPT-2 models include tasks performed with text-based generative networks, such as text summarization, question answering, feature extraction, and prompt engineering. Fundamentally, prompt engineering and fine-tuning is the continued training of pre-trained models with appropriate textual input to generate the expected textual output from a large language model and evaluating it in terms of outputs. In our study to measure the understanding, interpretation, and meaningful text generation ability of GPT-2 models, large language models were used within the framework of prompt engineering and fine-tuning for the purpose of generating a news summary when its headline was given as input, and vice versa when news summary was given as input, to generate the headline. In our experiments, four GPT-2 language models ofdifferent sizes from the Cosmos Research Group of the Computer Engineering Department of Yildiz Technical University were used in various combinations with two different news datasets, with and without fine-tuning. Our experimental results show that generative models for generating headlines from news summaries and news summaries from headlines produce promising results.