Matrix factorization for recommender systems

In this section, we will go over traditional techniques for recommending systems. As we will see, these techniques are really easy to implement in TensorFlow, and the resulting code is very flexible and easily allows modifications and improvements.

For this section, we will use the Online Retail Dataset. We first define the problem we want to solve and establish a few baselines. Then we implement the classical Matrix factorization algorithm as well as its modification based on Bayesian Personalized Ranking.

Dataset preparation and baseline

Now we are ready to start building a recommender system.

First, declare the imports:

import tensorflow as tf
import pandas as pd
import numpy as np
import scipy.sparse as sp
from tqdm import tqdm

Let us read the dataset:

df = pd.read_excel('Online Retail.xlsx')

Reading xlsx files may take a while. To save time when we next want to read the file, we can save the loaded copy into a pickle file:

import pickle
with open...