Choose an application

• Reference Manager
• EndNote
• RefWorks (Direct export to RefWorks)

Choose an application

• Reference Manager
• EndNote
• RefWorks (Direct export to RefWorks)

Choose an application

• Reference Manager
• EndNote
• RefWorks (Direct export to RefWorks)

Choose an application

• Reference Manager
• EndNote
• RefWorks (Direct export to RefWorks)

Choose an application

• Reference Manager
• EndNote
• RefWorks (Direct export to RefWorks)

Corpus cleaning is essential when conducting research, but it can be difficult to find the correct method as much depends on the goal of the corpus. This research aims to provide a solution for identifying non-language containing files in a corpus. In addition, it attempts to further break down the concept of language into textual and non-textual language. To this end, definitions of those three concepts are provided in order to convert them into (linguistic) features. These features are then used to train and test classification models, in this case logistic regressions and decision trees. The operationalisation of those definitions and features can, however, have a vast influence on the results of any classifiers. The results indicate that linguistic parsing aids in the scores, whereas balancing the class weight of the underrepresented labels mainly lowers the results. Only in multi-class classification does balancing the data ensure a better identification of those underrepresented classes. Both logistic regressions and decision trees yielded great results and are therefore useful in corpus cleaning purposes, particularly the models based on unbalanced and parsed data.