Search

Digitalization of music is the new trend, and preferences of individuals are highly rated. Millions of songs are being streamed in the music applications. The companies providing these services need to sort and arrange a wide range of music tastes for all of its users. On top of that, fresh music from various artists in a wide spectrum of genres are popping up every day. To keep track of all this, a classification system can be handy. So, we propose an RNN based model based on Natural Language processing to classify the songs based on their lyrics into different genres [1]. Additionally, this tool can be handy to the music lovers for quickly identifying which genre a particular song belongs to. In this paper, we apply Long Short Term Memory (LSTM) model with both Universal Serial Embedder (USE) and Bert embedders. A comparative study is performed to understand which combination of models works based to classify the genres based on lyrics. From our results, on the basis of accuracy of the model, we found that USE embedder with LSTM [2] gives a slightly better performance than Bert embedder. The LSTM model with USE embedding gave the highest accuracy of 83.42% when trained over a range of five folds. [ABSTRACT FROM AUTHOR]

Recommendation systems play a vital role in our day to day lives whether it may provide recommendation on which movie to watch or which item to buy. Most of the popular e-commerce websites are now a day's using recommender system gathering customer preferences like Amazon and Flipkart for selling their products. Similar to that we have music streaming applications like wynk and gaana.com which also provide recommendations to its users. These recommendations systems will benefit both the business and the users, the business are getting benefits and users are getting what they like.In this paper we have investigated music recommendation systems like content based filtering and collaborative filtering which provides music recommendations[1].We also tried to draw correlation between users and songs to provide recommendations as to which user would prefer to listen to which music based on their listening history. For this investigation we used the million song dataset available from kaggle.We used both the content based filtering and collaborate filtering algorithms to provide best music recommendation. Finally we also compared the accuracy levels of both the recommendation systems[2]. [ABSTRACT FROM AUTHOR]