Search

Music recommender systems have become a key technology supporting the access to increasingly larger music catalogs in on-line music streaming services, on-line music shops, and private collections. The interaction of users with large music catalogs is a complex phenomenon researched from different disciplines. We survey our works investigating the machine learning and data mining aspects of hybrid music recommender systems (i.e., systems that integrate different recommendation techniques). We proposed hybrid music recommender systems robust to the so-called “cold-start problem” for new music items, favoring the discovery of relevant but non-popular music. We thoroughly studied the specific task of music playlist continuation, by analyzing fundamental playlist characteristics, song feature representations, and the relationship between playlists and the songs therein.

Cross-modality retrieval encompasses retrieval tasks where the fetched items are of a different type than the search query, e.g., retrieving pictures relevant to a given text query. The state-of-the-art approach to cross-modality retrieval relies on learning a joint embedding space of the two modalities, where items from either modality are retrieved using nearest-neighbor search. In this work, we introduce a neural network layer based on Canonical Correlation Analysis (CCA) that learns better embedding spaces by analytically computing projections that maximize correlation. In contrast to previous approaches, the CCA Layer (CCAL) allows us to combine existing objectives for embedding space learning, such as pairwise ranking losses, with the optimal projections of CCA. We show the effectiveness of our approach for cross-modality retrieval on three different scenarios (text-to-image, audio-sheet-music and zero-shot retrieval), surpassing both Deep CCA and a multi-view network using freely learned projections optimized by a pairwise ranking loss, especially when little training data is available (the code for all three methods is released at: https://github.com/CPJKU/cca_layer)., Comment: Preliminary version of a paper published in the International Journal of Multimedia Information Retrieval

Automated music playlist generation is a specific form of music recommendation. Generally stated, the user receives a set of song suggestions defining a coherent listening session. We hypothesize that the best way to convey such playlist coherence to new recommendations is by learning it from actual curated examples, in contrast to imposing ad hoc constraints. Collaborative filtering methods can be used to capture underlying patterns in hand-curated playlists. However, the scarcity of thoroughly curated playlists and the bias towards popular songs result in the vast majority of songs occurring in very few playlists and thus being poorly recommended. To overcome this issue, we propose an alternative model based on a song-to-playlist classifier, which learns the underlying structure from actual playlists while leveraging song features derived from audio, social tags and independent listening logs. Experiments on two datasets of hand-curated playlists show competitive performance compared to collaborative filtering when sufficient training data is available and more robust performance when recommending rare and out-of-set songs. For example, both approaches achieve a recall@100 of roughly 35% for songs occurring in 5 or more training playists, whereas the proposed model achieves a recall@100 of roughly 15% for songs occurring in 4 or less training playlists, compared to the 3% achieved by collaborative filtering.

Applying diversity to a recommendation list has been shown to positively influence the user experience. A higher perceived diversity is argued to have a positive effect on the attractiveness of the recommendation list and a negative effect on the difficulty to make a choice. In a user study we presented 100 participants with several personalized lists of recommended music artists varying in levels of diversity. Participants were asked to assess these lists on perceived diversity and attractiveness, the experienced choice difficulty and discovery (i.e., the extent the list enriches their taste). We found that recommendation list attractiveness is influenced by two effects: 1) by diversity mediated through discovery; diverse recommendation lists are perceived to be more attractive if they enrich the user's taste or 2) by the list familiarity; a higher list familiarity contributes to a higher list attractiveness. We additionally revealed how individual differences (i.e., familiarity) moderate the effects found. Our results have implications on the composition of diversified recommendation lists. Specifically recommended items should contribute in extending and/or deepening the user's taste for the diversification to be effective.

Providing diversity in recommendations has shown to positively influence the user's subjective evaluations such as satisfaction. However, it is often unknown how much diversity a recommendation set needs to consist of. In this work, we explored how music users of Last.fm apply diversity in their listening behavior. We analyzed a dataset with the music listening history of 53,309 Last.fm users capturing their total listening events until August 2014. We complemented this dataset with The Echo Nest features and Hofstede's cultural dimensions to explore how music diversity is applied across countries. Between 47 countries, we found distinct relationships between the cultural dimensions and music diversity variables. These results suggest that different country-based diversity measurements should be considered when applied to a recommendation set in order to maximize the user's subjective evaluations. The country-based relationships also provide opportunities for recommender systems to personalize experiences when user data is limited by being able to rely on the user's demographics.

The objective of this PhD research is to deepen the understanding of how people listen to music and construct playlists. We believe that further insights into such mechanisms can lead to enhanced music recommendations. We research on the exploitation of user-generated data in the context of on-line music services, since it constitutes a rich and increasing source of information of user behavior. The research carried out so far has centered on the scenario of producing a single artist recommendation. Concretely, in this paper we show how to mitigate the cold-start problem for new artists, elaborating on our findings on the combined effect of users' listening histories and users' tagging activity. As future research, we will investigate how improved techniques to exploit user-generated data can also be applied to the task of producing sequential recommendations, like playlists. We are particulary interested in creating music playlists similarly as users would do, and in finding mechanisms to make such music streams adapt to users' feedback on-line.

Music information retrieval and music recommendation are seeing a paradigm shift towards methods that incorporate user context aspects. However, structured experiments on a standardized music dataset to investigate the effects of doing so are scarce. In this paper, we compare performance of various combinations of collaborative filtering and geospatial as well as cultural user models for the task of music recommendation. To this end, we propose a geospatial model that uses GPS coordinates and a cultural model that uses semantic locations (continent, country, and state of the user). We conduct experiments on a novel standardized music collection, the ``Million Musical Tweets Dataset'' of listening events extracted from microblogs. Overall, we find that modeling listeners' location via Gaussian mixture models and computing similarities from these outperforms both cultural user models and collaborative filtering.