Predict Image Upload Tags

Business Context

PixShare, a consumer photo-sharing platform, receives about 1.8M image uploads per day. The search and discovery team wants an automated tag prediction system that suggests relevant tags (for example: beach, dog, sunset, indoor, food) immediately after upload to improve search recall and downstream recommendations.

Dataset

You are given a historical training set of user-uploaded images with moderator-approved tags. This is a multi-label image classification problem: each image can have zero, one, or multiple valid tags.

Success Criteria

A solution is considered good enough if it achieves strong ranking quality on common tags while maintaining usable recall on tail tags. Target performance is micro-F1 >= 0.68, macro-F1 >= 0.42, and Precision@3 >= 0.78 on a held-out test set.

Constraints

• Predictions must be returned in under 120 ms p95 per image in online inference.
• The system should output the top 3-5 tags with calibrated confidence scores.
• Retraining can run weekly; online serving budget does not allow a very large ensemble.
• The moderation team needs per-tag thresholding and error analysis for sensitive tags.

Deliverables

1. Build a multi-label tag prediction model for uploaded images.
2. Explain preprocessing, feature extraction, and model choice.
3. Handle class imbalance, label noise, and threshold selection.
4. Evaluate the model with appropriate multi-label metrics.
5. Propose a production-ready inference and retraining approach.