Introduction Bluesky is a decentralized social media platform similar to Twitter, where the data on the network is open for access to everyone. 📂 View project on GitHub What is the problem? Recommendation systems in social media platforms face unique challenges compared to traditional recommendation domains like movies or music. Unlike static user-item interactions, social media engagement is highly dynamic—users interact with new posts, trends emerge, and user preferences shift rapidly.

Why it matters? One of the biggest concerns for any platform is figuring out how to make customers stay on their platforms. One way to do this is by recommending items that a user would like to retain user’s attention. The same goes with Twitch. As a streaming platform, recommending streamers that users would enjoy could keep their interest and make them stay on the platform. However, live recommendation is different from traditional recommendation tasks.

Why it matters? Listening to music on YouTube has become an integral part of my life, yet a challenge persists. The music I prefer during workouts differs from my regular listening choices. However, the YouTube recommendation system fails to recognize this difference, resulting in the need to frequently switch songs between my workout sets. To address this, I decided to create my own music classifier. By accomplishing this, I expect the distraction of changing songs diminishes, allowing me to focus more on my workout routine and utilize the time saved on song selection more effectively.