AI_Notebooks_30

Main repository for AI-related projects and notebooks. Find four Python notebooks graded 30 during our master degree AI4ST.

Table of Contents

1. Introduction
2. Recommender System from Scratch
3. Explainable AI
4. Graph Representation Learning
5. Attention-based Recommender System

Introduction

This repository contains a collection of notebooks and projects related to AI, including recommender systems, explainable AI, graph representation learning, and attention mechanisms.

Recommender System from Scratch

Description

Implementation of a recommender system from scratch, covering the basics of collaborative filtering, content-based filtering, and hybrid approaches.

Goals

• Build a recommender system from scratch
• Implement GMF and MLP
• Implement Neural Collaborative Filtering

Dataset

• Movielens Dataset: 100,000 ratings from 1000 users on 1700 movies

Implementation

• Embedded user and item vectors using GMF and MLP
• Concatenation of user and item vectors
• Fully connected layers for prediction

Explainable AI

Description

Exploration of explainable AI techniques, including model interpretability and feature importance, using libraries such as LIME and SHAP.

Goals

• Implement MLP model with more features
• Explain the model using LIME and/or SHAP

Dataset

• Movielens Dataset: 100,000 ratings from 1000 users on 1700 movies

Implementation

• Merging of data and preprocessing
• Reconstruction of dataset and functions
• MLP model with more features
• Explanation of model using LIME and SHAP

Graph Representation Learning

Description

Application of graph representation learning to movie recommendation, using Node2Vec.

Goals

• Graphs representation for movie
• Find the most similar movies to a target movie

Dataset

• Movielens Dataset: 100,000 ratings from 1000 users on 1700 movies

Implementation

• Construction of graph based on dataset
• Training of Word2Vec model
• Representation of 5 most similar movies

Attention-based Recommender System

Description

Implementation of an attention-based recommender system, using a multi-head attention layer to enhance model performance.

Goals

• Build a multi-head attention layer
• Compare model performance with and without attention

Dataset

• Movielens Dataset: 100,000 ratings from 1000 users on 1700 movies

Implementation

• Addition of attention layer after embedding and computation layers
• Concatenation of different heads for output representation
• Comparison of model performance with and without attention

Presentation

• AI_presentation.pdf: A presentation summarizing the key concepts and findings of the projects in this repository.