which kind of features are good to extract from text for author identification calssifying

Question

I want to classify texts as their authors for the author identification task...
the features are may:
the author's text length
or the authors text lexical features... is there anybody to help that which kind of features can help to improve classification results? the sample data frame I gathered is like this...

text long is 4 sentences, and I have 18 authors at least, about classification, this task is my thesis and I can not "just" apply classification on text, the goal is to apply classification into features that are extracted from text... I want to know which kind of features can help me to improve classification accuracy...( with both mo approaches or neural networks

Answer 1

How long are your texts? You can try deriving tf-idfs for each document, and then perform a knn search over your dataset. A more sophisticated way it's to featurize your texts with a neural network, and then perform a knn by using those vectors. If your dataset is big enough, there are not so many authors and there are several texts for each author, you could try to fine-tune a neural network to classify your texts. But I would go for the knn over the neural net features.