Predicting Popularity of Video Streaming Services with Representation Learning: A Survey and a Real-World Case Study.

Area of Science:

• Computer Science
• Data Science
• Machine Learning

Background:

• The internet's growth has led to a surge in online video content.
• A small fraction of videos garners most user views, making popularity prediction challenging.
• Accurate prediction aids infrastructure planning and content recommendation.

Purpose of the Study:

• To examine machine learning approaches for predicting web content popularity.
• To propose and evaluate a novel method combining attribute engineering and word embeddings for video popularity prediction.

Main Methods:

• Literature survey and taxonomy of existing machine learning models for content popularity prediction.
• Development of a case study using engineered features and word embeddings (from titles and descriptions).
• Experimentation with the Random Forest algorithm on a dataset from GloboPlay.

Main Results:

• A taxonomy of predictive features and state-of-the-art techniques was established.
• The combined approach of engineered features and word embeddings achieved the highest accuracy.
• The Random Forest model utilizing this combined approach reached an 87% accuracy rate.

Conclusions:

• Machine learning, particularly with combined feature engineering and word embeddings, is effective for predicting video popularity.
• This predictive capability offers significant value for content providers and platform management.
• The proposed method demonstrates a practical and accurate solution for a key industry challenge.