palco3.0
TRANSCRIPT
Outline Music Recommendation Approaches Examples Conclusions and future work
Recent Development at INESC Porto on PalcoPrincipal 3.0
Beatriz Mora
May 13, 2010
Outline Music Recommendation Approaches Examples Conclusions and future work
Music Recommendation ApproachesContent-basedContext-basedHybrid
Examples
Conclusions and future work
Outline Music Recommendation Approaches Examples Conclusions and future work
Content-based: Music signal
• Frequency domain.
• Timbral descriptors extracted from the music audio signal.
Outline Music Recommendation Approaches Examples Conclusions and future work
Content-based: Distance audio matrix
• Euclidean distance.
• Matrix of distancesbetween all songs.
• Musical matching.
• Sorted distance songrank.
Outline Music Recommendation Approaches Examples Conclusions and future work
Context-based: Latent Semantic Analysis (LSA)
is a technique used in natural language processing, in particular invectorial semantics, for analyzing relationships between a set ofdocuments and the terms they contain by producing a set ofconcepts related to the documents and terms, Wikipedia.
• Word correlations.
• Reduce dimension.
Outline Music Recommendation Approaches Examples Conclusions and future work
Context-based: Latent Semantic Analysis (LSA)
• Word: Tag.
• Concept: Group oftags.
Outline Music Recommendation Approaches Examples Conclusions and future work
Hybrid: weighted and cascade
Hybrid is a combination of the two previous approaches. Thereare two methods:
• Hybrid-weighted.
Outline Music Recommendation Approaches Examples Conclusions and future work
Hybrid: weighted and cascade
Hybrid is a combination of the two previous approaches. Thereare two methods:
• Hybrid-weighted. • Hybrid-cascade.
Outline Music Recommendation Approaches Examples Conclusions and future work
Jazz seed song
Content-based LSA-based
Hybrid-weighted Hybrid-cascade
Outline Music Recommendation Approaches Examples Conclusions and future work
Jazz seed song
Content-based LSA-based
Same timbre content Same band
Different genre Same genre
Hybrid-weighted Hybrid-cascade
New bands Same band as seed song
Same genre Same genre
Outline Music Recommendation Approaches Examples Conclusions and future work
Rock seed song
Content-based LSA-based
Hybrid-weighted Hybrid-cascade
Outline Music Recommendation Approaches Examples Conclusions and future work
Rock seed song
Content-based LSA-based
Same genre Same band
Hybrid-weighted Hybrid-cascade
New bands New bands
Same genre
Outline Music Recommendation Approaches Examples Conclusions and future work
Conclusions
• Audio is relevant and gives us a very wide musicrecommendation.
• LSA limits the recommendations by “our” concepts.
• Which kind of music would we like to discover?.
Outline Music Recommendation Approaches Examples Conclusions and future work
Conclusions
• Audio is relevant and gives us a very wide musicrecommendation.
• LSA limits the recommendations by “our” concepts.
• Which kind of music would we like to discover?.
Outline Music Recommendation Approaches Examples Conclusions and future work
Conclusions
• Audio is relevant and gives us a very wide musicrecommendation.
• LSA limits the recommendations by “our” concepts.
• Which kind of music would we like to discover?.
Outline Music Recommendation Approaches Examples Conclusions and future work
Future work
• Optimize audio analysis.
• Look for new descriptors in the audio that can help todistinguish genres.
• Collaborative filtering.
Outline Music Recommendation Approaches Examples Conclusions and future work
Future work
• Optimize audio analysis.
• Look for new descriptors in the audio that can help todistinguish genres.
• Collaborative filtering.
Outline Music Recommendation Approaches Examples Conclusions and future work
Future work
• Optimize audio analysis.
• Look for new descriptors in the audio that can help todistinguish genres.
• Collaborative filtering.