Automatic Recognition of Gender and Genre in a Corpus of Microtexts

In this paper, we focus on author’s gender and writing genre recognition solely on books titles. We analyse data extracted from the bibliography resources of the National Library of Poland. Within a paper, we compare different methods of text (title) representation and classification. It includes word embedding models such as word2vec, ELMo and classification algorithms such as linear models, multilayer perceptron and bidirectional LSTM. It is shown, that the writing genre (for defined 28 classes) could be automatically recognized based only on the book title with accuracy equal to 0.74. The best results were achieved by fastText methods with word n-grams.