Lexical Classification of Digital Humanities Publications Based on a Taxonomic Approach
Digital Humanities, Topic Modeling, Machine Learning
The use of automated text analysis is one of the main computational tools in classifying academic publications. Among other attributes, a relevant motivation for the use of these tools lies in their ability to help the academic community in the search and recovery of bibliographic references in the current midst of the informational flood. Starting with a relatively small amount of publications classified by experts, the use of an automated classification tool greatly extends that amount. In this context, we propose the use of the superseded version of the lexical classification method called Latent Dirichlet Allocation (LDA). to carry out the in two steps. The training stage of the method is carried out with a corpus of publications classified by experts in the field acquired from the group Doing Digital Humanities - A DARIAH Bibliography available on the Zotero platform. The classes used in this classification by hand correspond to categories of the hierarchical semantic structure of the TaDiRAH taxonomy, created by an international initiative for the classification of publications in Digital Humanities. In order to produce a more accurate model, we propose procedures to deal with unbalanced amounts of samples in each class and with overlapping publications in multiple categories. The model thus produced is then used to classify a corpus of publications taken from the Web Of Science. We aim to contribute to the development of the field in order to encourage the use of TaDiRAH towards the universalization of the organization, storage and retrieval of research resources.