Datasets

For authors classification

https://archive.ics.uci.edu/ml/datasets/Victorian+Era+Authorship+Attribution
In total it has 53678 rows and 50 different authors
We simplified the task to 2990 rows and 4 different authors

For imdb sentiment classification

https://www.kaggle.com/lakshmi25npathi/imdb-dataset-of-50k-movie-reviews
In total it has 49582 rows and 2 label: positive and negative
We simplified the task to 3000 rows

Independent models

2 separately trained models

Multitask model

One model with 2 heads: author classification and imdb sentiment

Results

Authors

  48.   0.   0.   0.   

  1.   20.   0.   0.   

  1.   0.   4.   1.   

  2.   8.   0.   64.   

  Accuracy: 91%

Imdb

  70.   7.   

  4.   69.   

  Accuracy: 93%