With the spread use of social media, people use misspelling, typos or add specific word to post malicious content deliberately to evade the censor of the platform. Human can still understand the meaning of the text, because we not only read the text but also see it. Inspired by the language processing of human being, we propose to utilize visual text representation to strengthen the robustness of text classification tasks, e.g.: hate speech classification.
Further information can be found in the attached PDF.
Ausschreibung: Download (pdf)