Course Hero Logo

poster.pptx - Building a Music Recommendation System Gautam...

Course Hero uses AI to attempt to automatically extract content from documents to surface to you and others so you can study better, e.g., in search results, to enrich docs, and more. This preview shows page 1 - 2 out of 3 pages.

Building a Music Recommendation SystemGautam Dudeja, Stephen Weldon, Chitta Mahapatra,Siddhartha Maharana and Tanmay SahooGeorgia TechGoodreadsis a social platform where users candiscuss and rate books on a scale from 1 to 5. We want to build a BookRecommender and find an efficient way to predict book ratings.With the rise of digital content distribution, we have access to a large musiccollection. With millions of songs to choose from, we sometimes feeloverwhelmed. Thus, an efficient music recommender system is necessary inthe interest of both music service providers and customers.Dataset:384,000 unique songs dataset48 million user ratings from 1 Millions usersU(average of 30 ratings peruser)5000 audio files crawledThe dataset is split between train (80% + 20% for cross validation) and test(20%).AbstractMatrix Factorizationturned out to be the best model to predict individualratings (RMSE) and obtain the most ideal rankingNeural Networksresults confirmed the intuition that the content of thebook is a better predictor of ratings and book quality than the cover imagealone. Results were still good for both models. Content basedrecommendation was able to address cold start problem and able torecommend music from different languages and artists from different partsof the world.Next steps:An hybrid model from all our different approaches could be aninteresting way to combine each model’s strength into a robust recommendersystem. We would also work on leveraging the Neural Network model toindividual users.We would like to include lyrics data from different music and use that totrain model for more contextual recommendations like patriotic songs, lovesongs or socio political songsl.ConclusionEvaluation MetricsAdd your information, graphs and images to this section.MaterialsMethodology : Collaborative FilteringAdd your information, graphs and images to this section.ResultsIdea:Listening histories are influenced by a set of factors specific to the domain(e.g. Genre, artist...)Matrix Factorization [2] consists in assuming there exist d latent features that can allowus to approach our U x V rating matrix R as the product of two matrices: Q of size U x d(users) and P of size d x V (Songs).

Upload your study docs or become a

Course Hero member to access this document

Upload your study docs or become a

Course Hero member to access this document

End of preview. Want to read all 3 pages?

Upload your study docs or become a

Course Hero member to access this document

Term
Fall
Professor
Pinedo
Tags
Cold start, Recommender system, mel spectrogram

Newly uploaded documents

Show More

  • Left Quote Icon

    Student Picture

  • Left Quote Icon

    Student Picture

  • Left Quote Icon

    Student Picture