RNSTweets: An AI-Driven Secure Social Media Platform for Cyberbullying Prevention in Academic Institution Nayana C P, Manish P, Manish P M, Nagarjun P

Abstract

Cyberbullying has become a significant concern across social media platforms and has affected students’ mental well-being,academicperformance,anddigitalsafety.Mainstream platformslackinstitution-specificcontrols,thusallowingharmful content to spread unbounded. RNSTweets is an AI-powered, closed-community microblogging platform designed for the RNS Institute of Technology. It integrates transformer-based NLP models such as BERT and HateBERT for real-time abusive language detection, supported by a demerit-based penalty mech- anism and an admin moderation dashboard. The platform usesa modular architecture with a React and Next.js frontend and a Next.js and MongoDB backend for secure student data storage. By combining automated moderation with secure authentication and controlled community access, RNSTweets offers a secure, collaborative environment for academic communication.

Introduction

Overview:
Social media facilitates communication but exposes students to cyberbullying, harassment, and hate speech. Mainstream platforms like Twitter and Instagram lack institution-level authentication, real-time moderation, and accountability, making students vulnerable. RNSTweets is a secure, RNSIT-exclusive microblogging platform addressing these challenges with AI-powered cyberbullying detection, domain-based authentication, demerit-based violation tracking, admin dashboards, and role-based access control.

Related Work:

1. Cyberbullying Detection:

   • Machine Learning: Early models (SVM, Naïve Bayes) rely on handcrafted features (n-grams, TF–IDF, sentiment) but struggle with context, sarcasm, and multilingual slang.

   • Deep Learning: CNNs, LSTMs, BiLSTMs with attention better capture semantic context.

   • Transformers: BERT, RoBERTa, HateBERT provide state-of-the-art detection of toxic content.

2. AI-Assisted Moderation:

   • Rule-based filtering is simple but prone to false positives.

   • Context-aware AI uses embeddings, sentiment, and conversation modeling for better moderation.

3. Digital Safety in Academia:

   • Students are particularly vulnerable; existing LMS platforms (Moodle, Canvas) are insufficient for peer interaction.

   • Need for institution-specific communication systems with monitoring.

4. Institution-Restricted Authentication:

   • Controlled access via domain-restricted emails and OTP-based verification enhances security.

5. Microblogging System Design:

   • RNSTweets adopts a familiar social media interface (posts, timelines, replies) while incorporating campus-specific safety, accountability, and admin oversight.

Objectives:

• Create a domain-restricted platform for verified students and faculty.

• Integrate AI for real-time cyberbullying and toxicity detection.

• Implement a demerit-based penalty system for behavioral tracking.

• Ensure secure authentication and role-based access.

• Provide a user-friendly microblogging interface.

• Enable admin oversight with dashboards for moderation.

• Use scalable technologies (Next.js, MongoDB) for future expansion.

Methodology & Architecture:

• Development Phases: Requirement analysis, system design, AI moderation, full-stack implementation.

• Authentication: Institutional email verification, OTP, bcrypt password hashing, JWT sessions.

• AI Moderation: Text preprocessing → embeddings via BERT/HateBERT → toxicity classification → demerit scoring → admin review.

• System Components: Frontend (Next.js + React + TailwindCSS), Backend/API (REST + GraphQL), Database (MongoDB + Mongoose), AI moderation (OpenRouter LLM), Admin Dashboard, Deployment (Vercel).

Implementation:

• Real-time moderation, posting, commenting, liking, and threaded conversations.

• Automated violation scoring with escalation based on severity.

• Admin dashboard for oversight and manual intervention.

Results:

• AI moderation showed accurate detection of toxic content.

• Platform demonstrated stability, low-latency responses, and effective demerit-based penalty enforcement.

• Student feedback indicated increased perceived digital safety and platform usability.

Limitations:

• Dependence on third-party AI APIs; potential latency and transparency issues.

• False positives/negatives in toxicity detection.

• Limited to institutional email ecosystem.

• Scalability challenges with serverless architecture.

• MongoDB real-time analytics constraints.

• No native mobile app yet.

• Admin moderation requires manual oversight.

• Behavioral indicators are preliminary and not clinically validated.

Conclusion

RNSTweets proves that an institution-restricted microblog- ging platform, supported by transformer-based NLP models, can meaningfully improve digital safety in an academic envi- ronment. By embedding BERT and HateBERT for real-time identification of toxic language, the system provides imme- diate intervention capabilities, reducing reliance on delayed manualreportingandenablinghealthieronlinecommunication among students. The closed-community access model reduces impersonation risks and prevents unauthorized participation, directly addressing limitations observed in public social sys- tems. Accountabilityisensuredthroughsecureauthenticationlay- ers, role-based access mechanisms, and a structured demerit- basedpenaltymodelthatencouragesresponsibleonlinebehav- ior. The violation tracking pipeline and moderation dashboard ensure appropriate escalation of repeated infractions and pro- vide historical data for faculty moderators to make informed human decisions, aligning with recent findings in hybrid AI moderation systems [10],[19]. Architecturally, RNSTweets achieved stable performancein concurrent usage through its modular full-stack design, responsivefrontend,andoptimizeddatabaseoperations.These results demonstrate that the platform is both technically feasi- ble and operationally effective for fostering a safer academic communication environment. Overall, RNSTweets offers a practical, scalable model for institutions seeking to implement secure, AI-assisted communication platforms that prioritize studentwell-being,accountability,andconstructiveinteraction [20].

References

[1] T. Gao et al., “Using Social Media to Automate the AuthenticationCeremony in Secure Messaging,” 2023. [2] A.Vaswanietal.,“AttentionIsAllYouNeed,”inProc.NeurIPS,2017. [3] J.Devlin,M.-W.Chang,K.Lee,andK.Toutanova,“BERT:Pre-trainingof Deep Bidirectional Transformers for Language Understanding,” inProc. NAACL-HLT, 2019. [4] V. Caselli et al., “HateBERT: Retraining BERT for Abusive LanguageDetection,” arXiv:2010.12472, 2020. [5] Y. Xu et al., “Cyberbullying Detection Using Machine Learning Tech-niques,” IEEE Access, 2021. [6] N. Chandra and R. Kaur, “AI-based Moderation Systems for SocialMedia Platforms,” International Journal of Computer Applications,2022. [7] Google,“FirebaseDocumentation,”2024. [8] MetaPlatformsInc.,“ReactDocumentation,”2024. [9] TwilioSendGrid,“SendGridDocumentation,”2024. [10] R.Kumaretal.,“Human-in-the-LoopModerationSystems,”ACMDigital Library, 2022. [11] P.FortunaandS.Nunes,“ASurveyonAutomaticDetectionofHateSpeech in Text,” ACM Computing Surveys, 2018. [12] Z. Zhang et al., “Detecting Cyberbullying on Social Media Using NLPTechniques: A Review,” IEEE Access, 2022. [13] A.SchmidtandM.Wiegand,“ASurveyonHateSpeechDetectionUsingNatural Language Processing,” in SocialNLP Workshop, 2017. [14] M.Dadvar,D.Trieschnigg,R.Ordelman,andF.deJong,“ImprovingCyberbullying Detection with User Context,” in Proc. ECIR, 2013. [15] L.VidgenandB.Derczynski,“DirectionsinAbusiveLanguageTrainingData: Garbage In, Garbage Out,” PLOS ONE, 2021. [16] Cloudflare,“UnderstandingServerlessatScale,”CloudflareDocs,2024. [17] MongoDBInc.,“MongoDBArchitectureGuide,”MongoDBDocumen-tation, 2024. [18] Vercel, “Next.js 16 Server Actions and App Router Documentation,”Vercel Docs, 2024. [19] A. Jhaver, D. Karpf, and A. Agrawal, “Human–AI Collaboration inContent Moderation,” in Proc. ACM Human-Computer Interaction,2023. [20] R. Salminen et al., “Developing Safe Online Communities for Students:DesignPrinciplesandModerationStrategies,”inProc.IEEEEDUCON,2021.

Copyright

Copyright © 2025 Nayana C P, Manish P, Manish P M, Nagarjun P V, Pavan Kumar H. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.