TEXTSHIELD: Robust text classification based on multimodal embedding and neural machine translation

Text-based toxic content detection is an important tool for reducing harmful interactions in online social media environments. Yet, its underlying mechanism, deep learning-based text classification (DLTC), is inherently vulnerable to maliciously crafted adversarial texts. To mitigate such vulnerabilities, intensive research has been conducted on strengthening English-based DLTC models. However, the existing defenses are not effective for Chinese-based DLTC models, due to the unique sparseness, diversity, and variation of the Chinese language. In this paper, we bridge this striking gap by presenting TEXTSHIELD, a new adversarial defense framework specifically designed for Chinese-based DLTC models. TEXTSHIELD differs from previous work in several key aspects: (i) generic - it applies to any Chinese-based DLTC models without requiring re-training; (ii) robust - it significantly reduces the attack success rate even under the setting of adaptive attacks; and (iii) accurate - it has little impact on the performance of DLTC models over legitimate inputs. Extensive evaluations show that it outperforms both existing methods and the industry-leading platforms. Future work will explore its applicability in broader practical tasks.