Recommender Systems

COMP3208: Social Computing Techniques

---

Contents

• Introduction
  • Task Description
  • Project Contents
• Running the Application

---

Introduction

Task Description

• Two datasets were given:
  • A set of predicted and ground truth user-ratings (10,000 entries each).
  • A set of user-ratings (100,000 entries, split into training and testing sets with a 90:10 split).
• Three tasks were set:
  • Task 1: Evaluate the performance of the predicted ratings against the ground truth ratings, according to standard evaluation metrics: MAE, MSE, RMSE.
  • Task 2: Develop a Collaborative Filtering recommender system to produce ratings for the test split of the dataset of user-ratings (either item- or user-based).
    • An item-based approach was taken, as it yeilded better resultss.
  • Task 3: Develop a Matrix Factorisation recommender system to produce ratings for the test split of the dataset of user-ratings.

Project Contents

Data

• Contained in the data directory is two directories:
  • predictions : The dataset of predicted and ground truth ratings.
  • dataset : The dataset of user ratings.

Source Code

All code is extensivley documented. In particular, the algorithms used for the two recommender system implementations are explained in detail.

• Directories:
  • General : Code used across the whole project.
  • IBCFRecommender : The implementation of the item-based collaborative filtering recommender.
  • MFRecommender : The implementation of the matrix factorisation recommender.
  • Tasks : Individual files/scripts for each of the tasks. These files make use of the project code (e.g., the relevant recommender system) to actually "carry out" the corrresponding task.
  • Tools : Non-specific tools used across the whole project.
• Files:
  • App.java : Program used to execute the code for each of the project's tasks (i.e., calls the relevant task file).
  • Test.java : Program used to test the performance of the recommender systems developed for the courswork.

---

Running the Application

• Only App.java and Test.java are runnable.
• App.java can be used to carry out the tasks.
• Test.java can be used to evaluate the performance of the recommender systems using the dataset of user-ratings.

---