Exploring Text Mining as a Vocabulary Evaluation Method: Focusing on Utterance Data from Children and Adolescents

Abstract

Objectives: Recently, awareness of the need for computerized language analysis using natural language processing technology is increasing, but interest has been focused on morpheme analysis and automatic measurement. The purpose of this study is to explore the possibility of evaluating language samples, especially vocabulary, using the text mining method, which is one of the big data analysis methods. Methods: A total of 14 elementary and middle school students with typical development participated in conversations on topics such as family, school, and hobbies; and the utterances collected in the conversation were analyzed by group, using web-based text-mining program for 1) frequency analysis and word cloud, 2) semantic network analysis based on connection centrality, 3) CONCOR analysis that clusters topics based on meaning. Then, the utterances of two children in each group were individually analyzed with the same procedure. Results: In the vocabulary cloud of elementary school and middle school students' group data, family names and school-related vocabulary appeared as top words, and topics such as daily routine, family time, relatives and holidays were clustered. By presenting the visualized result using the text mining method, it was possible to intuitively grasp the content. It was possible to understand the relationship between the vocabularies in order to understand the overall content structure. Conclusion: Text mining methods were confirmed to be viable tools for individualized and qualitative vocabulary evaluation and a supplement the traditional vocabulary evaluation method.