Creating a Content-Based Product Similarity Matrix for a Recommender System

product_similarity.py

import pandas as pd

# calculates the content-based product similarity matrix
# items is a Pandas dataframe containing all product details available for comparison
# returns the feature matrix and the correlation matrix
def product_similarity(items):
  
    # drop multicollinear columns and columns not considered for similarity
    items = items.drop(['Item', 'Style', 'Product', 'On Sale'], axis=1)
    print(items.nunique())
    
    # select continuous and categorical features for correlation
    similarity_features = items[
      ['URL', 'Original', 'Discount', 'Gender', 'Made In', 'Category']]
    similarity_features = similarity_features.set_index('URL')
    
    # dummy encoding of categorical variables
    similarity_features = pd.get_dummies(
      similarity_features, columns=['Gender', 'Made In', 'Category'])
    
    # calculate correlation matrix
    similarity_matrix = similarity_features.T
    similarity_matrix = similarity_matrix.corr(method='pearson')
    
    return similarity_features, similarity_matrix

# requests a content-based recommendation
# similarity_matrix is the pre-computed correlation matrix
# top_favorite is the user-rated favorite product
# returns the i-th product most-correlated with top_favorite
def content_based_similarity(similarity_matrix, top_favorite, i):
    
    # sort all product correlations related to top_favorite
    recommendations = similarity_matrix[top_favorite].sort_values(ascending = False)
    
    # drop auto-correlation of top_favorite
    recommendations = recommendations.drop([top_favorite], axis=0).index
    
    # return the product with the highest correlation
    new_recommendation = recommendations[i]
    
    return new_recommendation