Text Style Transfer in Persian Language using Deep Learning

EasyChair Preprint no. 4690

Abstract

In recent years, Deep Learning has taken control of a wide range of Natural Language Processing tasks among which Style Transfer is one of the most recent. The purpose of this task is to change the style of input text from its source style to a target style while maintaining the content and fluency of the text. Recent studies have mostly addressed this problem in English, whereas the Persian Language has dropped behind such advancements, mainly due to its complex structure, lack of data, and computational power. In this research, we propose the first instance of Text Style Transfer in the gender domain and examine the effects of words and their Parts of Speech in determining the author's gender. In order to demonstrate our methods' success., we begin with training a neural network for gender classification purposes. Then by transferring the style, we aim to deceive the classifier with transferred text and compare the results. The most powerful proposed method of this research achieved a 32.9% reduction in accuracy.

Keyphrases: Bag of Words, deep learning, gender identification, heuristic search, Natural Language Processing, text style transfer, Word classification

@Booklet{EasyChair:4690,
  author = {Reza Khan Mohammadi and Seyed Abolghasem Mirroshandel},
  title = {Text Style Transfer in Persian Language using Deep Learning},
  howpublished = {EasyChair Preprint no. 4690},

  year = {EasyChair, 2020}}