Music Recommendation and Discovery in the Long TailOscar Celma
Music consumption is biased towards a few popular artists. For instance, in 2007 only 1% of all digital tracks accounted for 80% of all sales. Similarly, 1,000 albums accounted for 50% of all album sales, and 80% of all albums sold were purchased less than 100 times. There is a need to assist people to filter, discover, personalise and recommend from the huge amount of music content available along the Long Tail.
Current music recommendation algorithms try to accurately predict what people demand to listen to. However, quite often these algorithms tend to recommend popular -or well-known to the user- music, decreasing the effectiveness of the recommendations. These approaches focus on improving the accuracy of the recommendations. That is, try to make accurate predictions about what a user could listen to, or buy next, independently of how useful to the user could be the provided recommendations.
In this Thesis we stress the importance of the user's perceived quality of the recommendations. We model the Long Tail curve of artist popularity to predict -potentially- interesting and unknown music, hidden in the tail of the popularity curve. Effective recommendation systems should promote novel and relevant material (non-obvious recommendations), taken primarily from the tail of a popularity distribution.
The main contributions of this Thesis are: (i) a novel network-based approach for recommender systems, based on the analysis of the item (or user) similarity graph, and the popularity of the items, (ii) a user-centric evaluation that measures the user's relevance and novelty of the recommendations, and (iii) two prototype systems that implement the ideas derived from the theoretical work. Our findings have significant implications for recommender systems that assist users to explore the Long Tail, digging for content they might like.
The Quest for Musical Genres: Do the Experts and the Wisdom of Crowds Agree?Oscar Celma
This paper presents some findings around musical genres. The main goal is to analyse whether there is any agreement between a group of experts and a community, when defining a set of genres and their relationships. For this purpose, three different experiments are conducted using two datasets: the MP3.com expert taxonomy, and last.fm tags at artist level. The experimental results show a clear agreement for some components of the taxonomy (Blues, HipHop), whilst in other cases (e.g. Rock) there is no correlations. Interestingly enough, the same results are found in the MIREX2007 results for audio genre classification task. Thus, showing the fact that a musical genre could have a multi–faceted definition; using expert based classifications, dynamic associations derived from the community driven annotations, and content–based analysis would improve genre classification, as well as other relevant MIR tasks such as music similarity or music recommendation.
From hits to niches? ...or how popular artists can bias music recommendation ...Oscar Celma
This paper presents some experiments to analyse the popularity effect in music recommendation.
Popularity is measured in terms of total playcounts, and the Long Tail model is used in order to rank music artists.
Furthermore, metrics derived from complex network analysis are used to detect the influence of the most popular artists in the network of similar artists.
The results from the experiments reveal that, as expected by its inherent social component, the collaborative filtering approach is prone to popularity bias. This has some consequences on the discovery ratio as well as in the navigation through the Long Tail.
On the other hand, in both audio content-based and human expert-based approaches artists are linked independently of their popularity. This allows one to navigate from a mainstream artist to a Long Tail artist in just two or three clicks.
Music Recommendation and Discovery in...which Web?Oscar Celma
As the world of online music grows, music recommendation systems become an increasingly important way for music listeners to discover new music. Commercial recommenders such as Last.fm and Pandora have enjoyed commercial and critical success. But how well do these systems really work? How good are the recommendations? How far into the "long tail" do these recommenders reach? In this talk we look at the current state-of-the-art in music recommendation and music discovery in the Web.
But...which Web? the old-fashioned one that we all are used to? The "Web 2.0"? The "Semantic Web"?
And, how does these "environments" affect music recommendation strategies and social media interaction? This talk will present real examples that emphasize the uppers and downers of the different coexisting webs.
Furthermore, we will present the current tools that the Music Information Retrieval field offers to improve/refine the resulting recommendations, as well as easing the life to Content Providers when annotating huge music collections.
Annotating Music Collections: How Content-Based Similarity Helps to Propagate...Oscar Celma
In this paper we present a way to annotate music collections by exploiting audio similarity. In this sense, similarity is used to propose labels (tags) to yet unlabeled songs, based on the content–based distance between them. The main goal of our work is to ease the process of annotating huge music collections, by using content-based similarity distances as a way to propagate labels among songs.
We present two different experiments. The first one propagates labels that are related with the style of the piece, whereas the second experiment deals with mood labels. On the one hand, our approach shows that using a music collection annotated at 40% with styles, and using content– based, the collection can be automatically annotated up to 78% (that is, 40% already annotated and the rest, 38%, only using propagation), with a recall greater than 0.4. On the other hand, for a smaller music collection annotated at 30% with moods, the collection can be automatically annotated up to 65% (e.g. 30% plus 35% using propagation).
UPDATED VERSION (2011): http://www.slideshare.net/plamere/music-recommendation-and-discovery
As the world of online music grows, music 2.0 recommendation systems become an increasingly important way for music listeners to discover new music.
Commercial recommenders such as Last.fm and Pandora have enjoyed commercial and critical success. But how well do these systems really work? How good are the recommendations? How far into The Long Tail do these recommenders reach?
In this tutorial we look at the current stateof theart in music recommendation. We examine current commercial and research systems, focusing on the advantages and the disadvantages of the various recommendation strategies. We look at some of the challenges in building music recommenders and we explore some of the ways that MIR techniques can be used to improve future recommenders.
Music Recommendation and Discovery in the Long TailOscar Celma
Music consumption is biased towards a few popular artists. For instance, in 2007 only 1% of all digital tracks accounted for 80% of all sales. Similarly, 1,000 albums accounted for 50% of all album sales, and 80% of all albums sold were purchased less than 100 times. There is a need to assist people to filter, discover, personalise and recommend from the huge amount of music content available along the Long Tail.
Current music recommendation algorithms try to accurately predict what people demand to listen to. However, quite often these algorithms tend to recommend popular -or well-known to the user- music, decreasing the effectiveness of the recommendations. These approaches focus on improving the accuracy of the recommendations. That is, try to make accurate predictions about what a user could listen to, or buy next, independently of how useful to the user could be the provided recommendations.
In this Thesis we stress the importance of the user's perceived quality of the recommendations. We model the Long Tail curve of artist popularity to predict -potentially- interesting and unknown music, hidden in the tail of the popularity curve. Effective recommendation systems should promote novel and relevant material (non-obvious recommendations), taken primarily from the tail of a popularity distribution.
The main contributions of this Thesis are: (i) a novel network-based approach for recommender systems, based on the analysis of the item (or user) similarity graph, and the popularity of the items, (ii) a user-centric evaluation that measures the user's relevance and novelty of the recommendations, and (iii) two prototype systems that implement the ideas derived from the theoretical work. Our findings have significant implications for recommender systems that assist users to explore the Long Tail, digging for content they might like.
The Quest for Musical Genres: Do the Experts and the Wisdom of Crowds Agree?Oscar Celma
This paper presents some findings around musical genres. The main goal is to analyse whether there is any agreement between a group of experts and a community, when defining a set of genres and their relationships. For this purpose, three different experiments are conducted using two datasets: the MP3.com expert taxonomy, and last.fm tags at artist level. The experimental results show a clear agreement for some components of the taxonomy (Blues, HipHop), whilst in other cases (e.g. Rock) there is no correlations. Interestingly enough, the same results are found in the MIREX2007 results for audio genre classification task. Thus, showing the fact that a musical genre could have a multi–faceted definition; using expert based classifications, dynamic associations derived from the community driven annotations, and content–based analysis would improve genre classification, as well as other relevant MIR tasks such as music similarity or music recommendation.
From hits to niches? ...or how popular artists can bias music recommendation ...Oscar Celma
This paper presents some experiments to analyse the popularity effect in music recommendation.
Popularity is measured in terms of total playcounts, and the Long Tail model is used in order to rank music artists.
Furthermore, metrics derived from complex network analysis are used to detect the influence of the most popular artists in the network of similar artists.
The results from the experiments reveal that, as expected by its inherent social component, the collaborative filtering approach is prone to popularity bias. This has some consequences on the discovery ratio as well as in the navigation through the Long Tail.
On the other hand, in both audio content-based and human expert-based approaches artists are linked independently of their popularity. This allows one to navigate from a mainstream artist to a Long Tail artist in just two or three clicks.
Music Recommendation and Discovery in...which Web?Oscar Celma
As the world of online music grows, music recommendation systems become an increasingly important way for music listeners to discover new music. Commercial recommenders such as Last.fm and Pandora have enjoyed commercial and critical success. But how well do these systems really work? How good are the recommendations? How far into the "long tail" do these recommenders reach? In this talk we look at the current state-of-the-art in music recommendation and music discovery in the Web.
But...which Web? the old-fashioned one that we all are used to? The "Web 2.0"? The "Semantic Web"?
And, how does these "environments" affect music recommendation strategies and social media interaction? This talk will present real examples that emphasize the uppers and downers of the different coexisting webs.
Furthermore, we will present the current tools that the Music Information Retrieval field offers to improve/refine the resulting recommendations, as well as easing the life to Content Providers when annotating huge music collections.
Annotating Music Collections: How Content-Based Similarity Helps to Propagate...Oscar Celma
In this paper we present a way to annotate music collections by exploiting audio similarity. In this sense, similarity is used to propose labels (tags) to yet unlabeled songs, based on the content–based distance between them. The main goal of our work is to ease the process of annotating huge music collections, by using content-based similarity distances as a way to propagate labels among songs.
We present two different experiments. The first one propagates labels that are related with the style of the piece, whereas the second experiment deals with mood labels. On the one hand, our approach shows that using a music collection annotated at 40% with styles, and using content– based, the collection can be automatically annotated up to 78% (that is, 40% already annotated and the rest, 38%, only using propagation), with a recall greater than 0.4. On the other hand, for a smaller music collection annotated at 30% with moods, the collection can be automatically annotated up to 65% (e.g. 30% plus 35% using propagation).
UPDATED VERSION (2011): http://www.slideshare.net/plamere/music-recommendation-and-discovery
As the world of online music grows, music 2.0 recommendation systems become an increasingly important way for music listeners to discover new music.
Commercial recommenders such as Last.fm and Pandora have enjoyed commercial and critical success. But how well do these systems really work? How good are the recommendations? How far into The Long Tail do these recommenders reach?
In this tutorial we look at the current stateof theart in music recommendation. We examine current commercial and research systems, focusing on the advantages and the disadvantages of the various recommendation strategies. We look at some of the challenges in building music recommenders and we explore some of the ways that MIR techniques can be used to improve future recommenders.