Skip to content

Sachet12345/Music-Recommender-System

Repository files navigation

Music-Recommender-System

1 Description

Our study is based on Million Song Dataset Challenge in Kaggle. We would like to do a large, personalized music recommendation system with the goal of predicting the songs that a user is going to listen. We learn from users listening history and full information of all songs (metadata, audio content analysis and standardized identifiers). Our goal is make our system largescale and more personal to users

2 Project approach

We propose following algorithms for our task1. Popularity based The most trivial recommendation algorithm is to simply present each song in descending order of its popularity skipping those songs already consumed by the user, regardless of the user’s taste profile./n

1. Popularity based

The most trivial recommendation algorithm is to simply present each song in descending order of its popularity skipping those songs already consumed by the user, regardless of the user’s taste profile.

2. Same artist greatest hits

This simply produces the most popular songs by artists that the user has already listened to. This gives some level of personalization in the recommendation system.

3. Collaborative Filtering

It can be either user-based or item-based. In user-based recommendation, users who listen to the same songs in the past tend to have similar interests and will probably listen to the same songs in future. In the item-based recommendation strategy, songs that are often listened by the same user tend to be similar and are more likely to be listened together in future by some other user.

4. Latent factor model

The ratings are deeply influenced by a set of factors that are very specific to the domain (e.g. genre, artist). These factors are in general not obvious and we need to infer those so called latent factors from the rating data. Users and songs are characterized by latent factors and a latent factor model such as Singular Value Decomposition (SVD) can decompose rating matrix into the product of a user feature and an item(song) feature matrix.

3 Dataset

We would be using the database provided by Kaggle competition, refer https://www.kaggle.com/c/msdchallenge/data It includes metadata (e.g., artist identifiers, tags,etc) , audio content analysis and standardized identifiers.

4 Libraries

Please install the following libraries:
sklearn
pandas
numpy

5 Instructions to run the code:

Run any model with the command python model_name.py. The User to be used for predictions can be changed by changing the user_id variable in the code.

Group:

Sachet Sourav- @Sachet12345
Koushik Swaminathan- @imkoushik22
Vijay Krishna- @vijaykrishna

About

No description or website provided.

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages