# Quick Lists: Enriched Playlist Embeddings for Future Playlist Recommendation

**Authors:** Brett Vintch  
**arXiv:** [2006.12382](https://arxiv.org/abs/2006.12382)  
**Date:** 2020-06-17  
**Summary page:** https://www.summarizepaper.com/en/paper/2006.12382v1/

## Key points

- Playlist recommendation is a challenge in digital music services
- Playlist embeddings are proposed to address this challenge
- Embeddings are invariant to playlist length and sensitive to track ordering
- Embeddings are enriched with user-specific data for personalized recommendations
- Enriched playlist embeddings help overcome the cold start problem
- The approach improves recommendation systems in digital music services

## Summary

The paper titled "Quick Lists: Enriched Playlist Embeddings for Future Playlist Recommendation" by Brett Vintch addresses the challenge of recommending playlists to users in the context of a digital music service. The authors highlight that a playlist is more than just the sum of its individual tracks, making it difficult to accurately recommend playlists to users. To overcome this challenge, the paper presents a novel method for generating playlist embeddings. The proposed method ensures that the generated embeddings are invariant to playlist length and sensitive to both local and global track ordering. This means that regardless of how long or short a playlist is, and regardless of the specific order in which tracks appear within the playlist, the embeddings capture meaningful information about the overall structure and sequencing of the playlist. Furthermore, these embeddings are enriched with side information about the user who created or interacts with the playlist. By incorporating user-specific data into the embeddings such as their preferences or listening history, a more personalized recommendation can be generated. The authors demonstrate that these enriched playlist embeddings are highly useful for generating next-best playlist recommendations. By leveraging captured information about playlist sequencing and incorporating user-specific data, accurate recommendations can be made even when faced with limited data or new users (known as the cold start problem). In summary, this paper introduces an innovative approach for generating playlist embeddings that address challenges related to playlist length, track ordering and personalization. The results show that these embeddings enable effective next-best playlist recommendations while also tackling issues associated with new users. Overall, this research contributes valuable insights into improving recommendation systems in digital music services.