A Comparative Assessment of Unsupervised Keyword Extraction Tools

The extraction of keywords is a critical task in natural language processing and information retrieval. It has become increasingly important in a wide range of applications, from search engines and e-commerce platforms to news and social media analysis. However, the evaluation of keyword extraction methods remains a challenging task due to the diverse range of data types and contexts used, as well as the complexity of the methodologies and techniques involved. In this regard, this study reviews the prior surveys on keyword extraction methods to comprehend the fundamental principles, difficulties in keyword extraction, and benchmark datasets. The reliability of evaluation techniques and an examination of their flaws remain two of the largest problems. Hence, this study includes the literature that performed a comparative analysis of popular keyword extraction methods. Furthermore, in this paper, we present a comparative evaluation of open-source unsupervised keyword extraction tools, analyzing their performance across a range of data types and under different testing conditions. The experimental analysis shows that, in terms of f-score, KPminer performs most consistently for different text lengths while KeyBERT(mmr) outperforms other tools. Considering the execution time, RAKE and YAKE are the fastest tools. Though graph-based tools tend not to perform well on long text, TopicRank and MultipartiteRank perform very well on long text as they use topics as nodes of the graph, which is another finding of this study. By highlighting the key factors that influence the performance of keyword extraction tools, our analysis contributes to directing the reader in selecting suitable keyword extraction tools for different applications.