Finding Similar Music Artists for Recommendation

18
Finding Similar Music Artists for Recommendation Presented by :Abhay Goel, Prerak Trivedi

description

Finding Similar Music Artists for Recommendation. Presented by : Abhay Goel , Prerak Trivedi. Introduction. In this project, we present similar music artists based on their genre, social ratings and the artist’s era - PowerPoint PPT Presentation

Transcript of Finding Similar Music Artists for Recommendation

Page 1: Finding Similar Music Artists for Recommendation

Finding Similar Music Artists for Recommendation

Presented by :Abhay Goel, Prerak Trivedi

Page 2: Finding Similar Music Artists for Recommendation

Introduction• In this project, we present similar music artists

based on their genre, social ratings and the artist’s era

• We find the related contents using information retrieval from the Y! webscope dataset (for ratings) and Y!Music database.

• It is much like Similar video recommendations. However, that works on finding similar videos based on visual and textual similarity where as this works on more parameters such as user ratings, artist era and the genre.

Page 3: Finding Similar Music Artists for Recommendation

Technologies Deployed

• MySQL – To store the Yahoo! Webscope dataset R1- For music artists and ratings.(Approx 97,000 artists and 3.25 million User ratings )

• Yahoo Music API/JSON (http://developer.yahoo.com/music/) – For getting artist Genre and Era(5000 queries per day per IP address)

• JSP/JSON – Connectivity between MySQL and HTML, to get User Ratings

• Apache Tomcat – Web Server used for JSP pages

• HTML/JavaScript /AJAX– Front end UI for entering artist query and displaying results

• SQL– Executing of queries.

Page 4: Finding Similar Music Artists for Recommendation

Model Approach• Gather Information about artist from Y!Music • Use the Y! Music API and an application ID,

identify to access the data.• The API is REST-based and returns the data in

JSON,XML format.• Collected data includes: artist information,

category, releases, top similar artists, top tracks, era and events.

• Use the Ratings data from the Y! Webscope dataset and apply the algorithm.

Page 5: Finding Similar Music Artists for Recommendation

Method / Algorithm• We implement three steps of computation in

order to obtain more detailed results, as defined below:

Compute similarity measure based on artist’s genre and era (Using Jaccard’s Similarity Coefficient and Nearest Neighbor Search)

Construct external db from the webscope dataset to evaluate the artist’s ratings

Combine the similarity measure and the artist reputation for the improved searching of the similar artists.

Page 6: Finding Similar Music Artists for Recommendation

Computing Artist Information

• Consider genre and album releases as basis for computing similar artists using these steps:

a. Applying comparison between artists’ genre using distance measure

b. Applying user rating to artistsc. Applying Nearest-Neighbor search on

artists’ releases album

Table: Parameter Properties

Page 7: Finding Similar Music Artists for Recommendation

Computing Similarity Measure• The Similarity Level (SL) between the Artist

and Genre is the Jaccard’s Similarity coefficient between |ax| and ac as defined in formula below.

o α represents the total number of genres that is not presented in |ax| but appears in ac

o β represents the total number of genres that is not presented in ac but appears in |ax|

o γ represents the total number of genres presented in both |ax| and ac

o |ax| represents the genre which the artist belongs to, where ax(gy) = 1

Page 8: Finding Similar Music Artists for Recommendation

Computing User Ratings for Artist

• The User Rating (Rax) is calculated from the Yahoo! Webscope Dataset R1 and is the average score for that particular artist for every user rating received.

• The above formula represents a set of ru_ax of users, who give rating to artist ax, and nax is the number of users in ru_ax

Page 9: Finding Similar Music Artists for Recommendation

Computing New Similarity Level

• The Similarity Level (SL2) is recalculated by combining the result from SL and Rax using the formula as follows:

o WSL1 is weight for the corresponding similarity level in Equation for SL

o WRa is the weight for the corresponding artist rating in Equation for Rax

Page 10: Finding Similar Music Artists for Recommendation

Computing Nearest-Neighbor Search

• The Nearest Neighbor Search (kNN(q)) is performed to find the closest object to the given query object q. The general equation to find the k nearest neighbors is as follows:

o kNN(q) query retrieves the k nearest-neighbors of the object qo R is the distance range (r) used for searching, where p Є S with

d(q, p) ≤ r.

Page 11: Finding Similar Music Artists for Recommendation

PROJECT RESULTS WITH SCREENSHOTS

Page 12: Finding Similar Music Artists for Recommendation

Screenshot of Y! Webscope dataset from MySql Database for ratings

Page 13: Finding Similar Music Artists for Recommendation

Screenshot of Data Collected from Yahoo Music API

Page 14: Finding Similar Music Artists for Recommendation

User Interface Screenshot(1)

Page 15: Finding Similar Music Artists for Recommendation

User Interface Screenshot(2)

Page 16: Finding Similar Music Artists for Recommendation

User Interface Screenshot(3)

Page 17: Finding Similar Music Artists for Recommendation

DEMO

Page 18: Finding Similar Music Artists for Recommendation

References

-Reference from Yahoo Webscope Publication “Finding Similar Music Artists for Recommendation”.

-Yahoo Developer Network