29. IEEE SİNYAL İŞLEME VE İLETİŞİM UYGULAMALARI KURULTAYI, İstanbul, Türkiye, 09 Haziran 2021
Keywords are salient words that best describe the content and the topic of a text document. As text documents are not usually annotated by their authors, automatic keyword extraction has become a challenging research topic. In this study, we investigate the performance of graph-centrality measures as initial node weights in graph-based keyword extraction on Tukish bank documents. To this aim, we focus on degree, eigenvector, betweenness, and closeness centralities and investigate their performance on keyword extraction on a 553 bank document dataset in Turkish. The experimental results show eigenvector centrality achieved the best results.