BYTEME

Home page for video analysis
Results of evaluation
Final ranking and feedback from evaluation

Inspiration

The current TikTok rewards system primarily relies on engagement metrics such as likes, views, and comments. However, these metrics do not adequately assess video quality in terms of creativity, originality, or production value. Our project was inspired by the need to bridge this gap and provide a more comprehensive evaluation system that aligns with TikTok’s goals of fostering innovation and high-quality content creation.

What it does

BYTEME offers an advanced evaluation system for TikTok videos, analyzing content across visual, audio, and textual dimensions. It provides creators with objective ratings and actionable feedback to help improve their content and achieve higher ratings in TikTok’s rewards system. The system uses machine learning for automatic categorization, ensuring consistent assessments.

How we built it

Backend: We used Python with libraries like PyTorch, OpenCV, and Librosa for data processing and model training. Frontend: The interface was built with Lynx, providing a user-friendly dashboard for creators to view their scores and feedback. APIs: We integrated various APIs, including the TikTok API, yt-dlp for video download, and TF-IDF for text analysis.

Challenges we ran into

Feature extraction complexity: Combining data from visuals, audio, and text posed challenges in ensuring meaningful analysis. Data consistency: Ensuring that the analysis remained consistent across varying video formats and quality. Neural network training: Balancing accuracy with realistic feedback was a challenge, requiring fine-tuning of the model.

Accomplishments that we're proud of

Multi-modal analysis: Successfully integrated visual, audio, and textual features into a single system to evaluate video quality holistically. Real-time feedback: The system can process videos and provide actionable feedback almost immediately. Front-end integration: The Lynx-based interface offers a seamless user experience for creators to interact with their evaluation results.

What we learned

Combining data modalities: Integrating visual, audio, and text analysis provided deeper insights into video quality. Importance of real-time feedback: Creators benefit from actionable feedback that helps them improve and grow their content. Machine learning insights: Training models to predict human-like feedback was a nuanced task that involved continuous refinement.

What's next for BYTEME

Scaling: We plan to scale the system to handle more videos and integrate additional content types. New metrics: We aim to introduce new evaluation metrics like engagement potential and audience retention. TikTok rewards integration: We plan to work towards automatic reward tier adjustments based on the evaluation scores.