The problem CyberXpert solves

Cyberbullying casts a long shadow on the digital landscape, posing a significant threat to the well-being of individuals, particularly children and teenagers.Here's how a cyberbullying detection backend service steps in to address these problems:

Identifying Harmful Behavior: Through advanced language analysis techniques, the service can detect potential cyberbullying incidents.Alerting Relevant Parties: The detection system can trigger various actions depending on the severity of the situation:Notifications: Alerting users or moderators about potential cyberbullying, allowing them to intervene and address the situation.Content Flagging: Flagging abusive content for further review and potential removal from platforms.

A browser extension can prompt users to reconsider potentially harmful language before posting. Image captioning expands detection beyond text, while audio analysis tackles cyberbullying in voice chats. This combined effort offers real-time intervention and broader detection. However, accuracy and user experience require careful calibration. To address these challenges and safeguard user privacy, collaboration with experts and transparent communication regarding the system's limitations are crucial. Ultimately, this fusion of technological advancements and social initiatives like promoting responsible online behavior paves the way for a safer digital environment, empowering users and fostering a culture of empathy.

Challenges we ran into

Merging multiple cyberbullying detection models into a single backend service proved challenging. Firstly, each model likely operates independently with different data formats and outputs, requiring a unified framework. Secondly, building data pipelines to seamlessly transfer information between the models and ensuring efficient data flow is crucial. Furthermore, standardizing the diverse model outputs into a format the backend understands necessitates data transformation. Finally, integrating this combined output with the backend's core functionalities requires establishing clear communication channels.

Obtaining meaningful output in the backend presents its own hurdles. Inaccurate or misleading results limited training data, language complexities, and potential biases were a isssue. Additionally, grasping the true intent behind the analyzed content, especially when solely relying on text snippets or captions without broader context, is difficult. Finally, transforming the raw output into actionable insights requires further processing and interpretation for intervention or reporting purposes.