In many natural language processing applications, keyword extraction plays a crucial role in information retrieval, document classification, and sum- marization. This study investigates the efficacy of three cutting-edge keyword extraction methods: KeyBERT, YAKE (Yet Another Keyword Extractor), and RAKE (Rapid Automatic Keyword Extraction), along with a newly designed model, KeyBERT(+), which removes duplicates and offers improved perfor- mance. A comparative analysis was conducted to assess the performance of these techniques in identifying keywords from student and reference answers—a sce- nario particularly relevant to educational feedback and assessment systems. The comparison is based on two key metrics: the number of key points extracted and the extraction time. The findings demonstrate that KeyBERT(+) outperforms the other methods, providing valuable guidance for selecting appropriate keyword extraction techniques in educational contexts.