Neural Network Course, Assignment 3 (KDD Cup 2011 Final Report)
M9915811 Rinardi B. Sarean

Introduction

KDD Cup 2011 focuses on the issues related to the user rating in Yahoo Music, since yahoo music has collected billions of user ratings for musical pieces over a period of time. The KDD Cup contest releases over 300 million ratings performed by over 1 million anonymized users. The ratings are given to different types of items-songs, albums, artists, genres-all tied together. This competition is divided into two tracks. Track 1, learning to predict users's ratings of musical items. And, track 2, learning to separate tracks scored highly by specific users from tracks not scored by them.

Survey

After looking for good implementation of artificial intelligence. I found Encog Framework. This framework offers 2 implementation, Java and C# (.NET 3.5). Since I used C#, so I use the C# implementation, and recompile it to match with .NET 4 environment. These framework offers flexibility to create networks and training algorithm. In fact they have prepared common used training algorithms. They have write enough documentation and guide too.

Experiment

After a few weeks of experiments, there are issues that became obstruction to finding a good prediction result. All experiments use C# as programming language and MySQL as database to provide Encog.NET AI Framework with training data and test data. All experiments use resilient propagation as training algorithm and various network configuration. My experiments consist of 3 major progress, as follow :

Post-Experiments Issues

These are issues related to the problems in KDD Cup 2011:

Resources

Here I provide source codes and sql schema for my KDD Cup 2011 Track 1 Prediction application. I used Visual Studio 2010. And several additional libraries. All included in compressed file here. For MySQL schema, download here. And for sample query for retrieving data, download here.