Text Authorship Verification through Watermarking

While a plethora of digital contents are daily generated and shared online, authorship verification has become an imperative task. In comparison to other media watermarking techniques, text watermarking is a more challenging task. The changes in text would strongly affect the visual form and the meaning, text might be very short (eg. social media posts) and it cannot be always converted into image. In this paper we propose a novel text watermarking method for authorship verification based on Unicode confusable substitution. The proposed method substitutes latin symbols with homoglyph characters. It ensures length preservation and visual indistinguishability among the original text and the watermarked one. We successfully evaluate our approach using a real dataset of 1.8 million of New York Times articles. The results show the effectiveness of our method providing an average length of 101 characters needed to embed a 64bit password based watermark.