Using Turkish natural language processing tools in analysis of science textbooks for primary and secondary education: Comparative analysis of word and lemma frequencies

The aim of this thesis is based on the automatic analysis of science textbooks taught at primary and secondary school grade levels using natural language processing tools. Document analysis was used in this study in which quantitative research method was used. Two Science textbooks from each grade level in PDF format published in the Ministry of National Education Information Network for the 2020-2021 academic year were sampled. The units were converted from PDF format to Microsoft Word document and then to the plain text. The units, whose word frequency lists were obtained with the help of NKod software, were compared with the help of RStudio program by performing automatic text analysis with appropriate libraries and necessary coding in R programming language. 134915 words and 19363 noun stems in the texts were studied. In the analysis, firstly, it was concluded that the first 20 words of the word frequency lists of the textbooks were classified and helped to predict the content of the units. Secondly, since the word frequency lists obtained did not fit the normal distribution, the Spearman Rank Correlation Test was performed for the correlations between the units. The analyses revealed strong and moderate correlations between the same units of different books (0.3≤𝜌≤0.5), and a low-strength relationship between different units (𝜌<0.3) or no relationship (𝜌≅0) was shown. The nonparametric Chi-square test showed that all units at each grade level had statistically differentiated terms (𝜒2 >3.84 𝑎𝑛𝑑 𝑝<0.05) were obtained. The results were interpreted and discussed according to whether they were appropriate for unit-specific terms, objects specific to daily life or science courses, and other categories. As the grade level increases, there is a noticeable increase in the number of differentiated terms specific to the unit.