# A Hybrid Approach to Music Playlist Continuation Based on Playlist-Song Membership

**Authors:** Andreu Vall, Matthias Dorfer, Markus Schedl, Gerhard Widmer  
**arXiv:** [1805.09557](https://arxiv.org/abs/1805.09557)  
**Date:** 2018-05-24  
**Summary page:** https://www.summarizepaper.com/en/paper/1805.09557v1/

## Key points

- The paper discusses automated music playlist continuation in music recommender systems
- Collaborative filtering models offer better playlist continuations compared to content-based approaches
- Pure collaborative filtering models have limitations: they can only extend trained playlists and misrepresent songs occurring in few playlists
- The authors propose a hybrid model based on "playlist-song membership" to address these limitations
- The model considers playlist-song pairs exclusively in terms of feature vectors
- Experimental results show that the proposed model performs comparably to state-of-the-art collaborative filtering models when extending trained playlists with frequently occurring songs
- The proposed model offers advantages such as extending non-profiled playlists and recommending infrequently or never occurred songs
- This research presents an innovative approach for music playlist continuation by incorporating "playlist-song membership" and leveraging feature vectors for improved recommendations.

## Summary

The paper titled "A Hybrid Approach to Music Playlist Continuation Based on Playlist-Song Membership" discusses the task of automated music playlist continuation in music recommender systems. The goal is to provide a suitable extension to a given playlist. Collaborative filtering models, which extract abstract patterns from curated playlists, have been found to offer better playlist continuations compared to content-based approaches. However, pure collaborative filtering models have certain limitations. Firstly, they can only extend playlists that were profiled during training. Secondly, they tend to misrepresent songs that occur in very few playlists. To address these limitations, the authors propose a novel hybrid playlist continuation model based on "playlist-song membership". This concept refers to whether a given playlist and song fit together. The proposed model considers each playlist-song pair exclusively in terms of feature vectors. After being trained on a collection of labeled playlist-song pairs, the model determines whether a pair fits together or not. Experimental results using two datasets of curated music playlists demonstrate that the proposed model performs comparably to state-of-the-art collaborative filtering models when extending playlists profiled at training time and featuring frequently occurring songs. In contrast to traditional collaborative filtering models, the proposed model offers additional advantages due to its general understanding of playlist-song pairs based on feature vectors. It can extend non-profiled playlists and recommend songs that occurred infrequently or never in training playlists. Overall, this research presents an innovative approach for music playlist continuation by incorporating the concept of "playlist-song membership" and leveraging feature vectors for improved recommendations. The findings highlight the potential benefits of this hybrid model in enhancing user experience and expanding the range of recommended songs beyond what traditional collaborative filtering methods can achieve.