Abstract:
"Online sexual abuse is a significant and growing issue that plagues digital platforms. Its complexity of detection and negative impact on victims escalates this problem. To address this, prevention-based strategies particularly AI-driven solutions have emerged. Current machine learning algorithms have only been able to detect single types of online sexual abuse and provide very basic alerts to the user with no victim guidance.
The author introduces the novel approach SafeConvo, a comprehensive detection system capable of detecting multiple types of online sexual abuse and providing victim support. To achieve this, Natural Language Processing (NLP) techniques were used to fine-tune a pre-trained Bidirectional Encoder Representations from Transformers (BERT) model for this task.
The test results demonstrated that SafeConvo was able to distinctly detect various forms of online sexual abuse including sexual extortion, grooming and unwanted sexually explicit messages with an overall accuracy rate of 89%. These findings not only prove the effectiveness of the solution in addressing the challenge of online sexual abuse but also highlight the potential to be applied in real world chat platforms in future research."