A Hybrid Deep Learning and Sentiment-driven Framework for Copyright Infringement Detection and Enforcement on YouTube

Read full text Summarise with VIRAT
Authors: M. Thanga Subha Devi, R. Suji Pramila, J. E. Judith
Cited by: 0
References: 22

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

The tremendous growth of user-generated content in digital space has intensified global concerns surrounding copyright violations.As millions of videos are uploaded every day, websites such as YouTube increasingly face challenges in identifying inappropriate usage of copyrighted music, video clips and images.Traditional monitoring mechanisms are insufficient, slow and error prone, leading to revenue losses and increased legal complexities.This has led to an urgent need for automated, scalable and intelligent systems capable of accurately detecting copyright infringement.This study proposes a hybrid deep learning (DL) and sentiment-driven framework for automated multi-class copyright infringement detection and enforcement on YouTube.The research incorporates a curated collection of 16,215 videos with metadata, interaction statistics and user remarks, labeled using verifiable platform-level copyright enforcement signals.After rigorous preprocessing, metadata tokens are embedded and processed through a one-dimensional Convolutional Neural Network (1D CNN) with multiple kernel sizes to extract phrase-level copyright cues from titles and tags.A Transformer encoder with multi-head self-attention then models contextual relationships among these cues, capturing ownership-and reuse-related dependencies.In parallel, a sentiment analysis branch aggregates polarity scores from user comments, weighted by engagement credibility, to derive behavioural indicators of suspected infringement.The contextual metadata embeddings and sentiment vector are concatenated and fed to a fully connected Softmax classifier that assigns each video to No Infringement, Music Infringement, Video Infringement or Image Infringement, with predicted classes mapped to enforcement measures including monetisation restriction, upload blocking or thumbnail replacement.Extensive experimental evaluation, including class-aware metrics, alternative imbalance-handling strategies and leakage-resistant data splits, demonstrates that the proposed model achieves an overall accuracy of 98.90% with a macro F1-score of 0.97.By jointly leveraging lexical patterns, contextual semantics and crowd sentiment while ensuring robust evaluation and generalization, the proposed framework reduces false positives and missed infringements, making it suitable for scalable and proactive copyright governance on large video platforms.

Fields of Study & Subjects

Hate Speech and Cyberbullying Detection Copyright and Intellectual Property Law AI and Intellectual Property

Publication details

Journal / source
International journal of intelligent engineering and systems
Volume
19
Issue
4
Pages
442–464
Publication date
23 March 2026
Language
EN