An Extended AHP-Based Corpus Assessment Approach for Handling Keyword Ranking of NLP: An Example of COVID-19 Corpus Data

• Published in: Axioms 2023-07, Vol.12 (8), p.740
• Main Authors: Chen, Liang-Ching, Chang, Kuei-Hu
• Format: Article
• Language: English
• Subjects: Air pollution; Algorithms; Analytic hierarchy process; analytic hierarchy process (AHP); Computational linguistics; Coronaviruses; corpus assessment approach; COVID-19; Earth; Empirical analysis; environmental science; Hierarchies; Information retrieval; Keywords; Language processing; Linguistics; Methods; Multiple criteria decision making; Natural language; Natural language interfaces; Natural language processing; natural language processing (NLP); Nitrogen dioxide; Optimization; Outdoor air quality; Pandemics; Parameters; Random variables; Risk management; Software
• ISSN: 2075-1680; 2075-1680
• DOI: 10.3390/axioms12080740

The use of corpus assessment approaches to determine and rank keywords for corpus data is critical due to the issues of information retrieval (IR) in Natural Language Processing (NLP), such as when encountering COVID-19, as it can determine whether people can rapidly obtain knowledge of the disease. The algorithms used for corpus assessment have to consider multiple parameters and integrate individuals’ subjective evaluation information simultaneously to meet real-world needs. However, traditional keyword-list-generating approaches are based on only one parameter (i.e., the keyness value) to determine and rank keywords, which is insufficient. To improve the evaluation benefit of the traditional keyword-list-generating approach, this paper proposed an extended analytic hierarchy process (AHP)-based corpus assessment approach to, firstly, refine the corpus data and then use the AHP method to compute the relative weights of three parameters (keyness, frequency, and range). To verify the proposed approach, this paper adopted 53 COVID-19-related research environmental science research articles from the Web of Science (WOS) as an empirical example. After comparing with the traditional keyword-list-generating approach and the equal weights (EW) method, the significant contributions are: (1) using the machine-based technique to remove function and meaningless words for optimizing the corpus data; (2) being able to consider multiple parameters simultaneously; and (3) being able to integrate the experts’ evaluation results to determine the relative weights of the parameters.