Summarization of turkish news with machine learning

Technological advancements in media led to digital platforms offering quicker access to real-time information. As a result, people prefer to read news from credible internet resources compared with print media. Innovative media organizations develop search engine-friendly content to stay competitive and prominent in digital journalism, thus having greater accessibility to consumers. This led to an increase in the search engine optimized contents of online news lengthier than traditional news. On the contrary, people in the digitalized world desire to interact with systems that provide more information within the shortest possible time. To cope with such challenges, automatic text summarization is a technique mainly utilized to shorten the text length without compromising the integrity of the information. Through this method, this thesis summarized news produced in the Turkish language with BERT and Seq2Seq models. These models' validity and overall performances compared with traditional summarization methods, tf-idf, LSA, TextRank, and LexRank. Ultimately, this research project led to a web-based automatic text summarizer named Özetleyici. The BERT model used in this summarizer scored an average ROUGE-L score of 0.25 in a data set consisting of Turkish news.