2001 - Item based Collaborative Filtering Recommendation Algorithms

If you are interested in the paper, you can find it here.

Introduction

Back to 2001, in the rapidly evolving landscape of recommender systems, item-based collaborative filtering emerges as a game-changer. This paper highlights how item-based techniques outperform traditional user-based methods in both performance and quality, making them indispensable for large-scale applications. By focusing on item relationships, these algorithms ensure high-quality, real-time recommendations, even with sparse data and millions of users. As the demand for personalized experiences grows, item-based collaborative filtering stands out as a scalable and efficient solution, promising to enhance user satisfaction and engagement across various online platforms.

Code Reproduction

Data Source

https://grouplens.org/datasets/movielens/100k/

"""
2001-Item based Collaborative Filtering Recommendation Algorithms
Code Reproduction
"""

import os
import sys

sys.path.append(os.path.join(os.path.dirname(__file__), ".."))

import copy
import pandas as pd
import numpy as np
from util import sprint

# Load data
data_path = "../DATASETS/MovieLens100K/u.data"
raw = pd.read_csv(
    data_path,
    sep="\t",
    header=None,
    names=["user_id", "item_id", "rating", "timestamp"],
)
sprint(raw.head(), raw.shape)

# Create user-item matrix
n_users = raw.user_id.unique().shape[0]
n_items = raw.item_id.unique().shape[0]
sprint("Number of users =", n_users)
sprint("Number of items =", n_items)

data = copy.deepcopy(raw)
data = data.sort_values(by=["user_id", "item_id"])

ratings = np.zeros((n_users, n_items))
for row in raw.itertuples():
    ratings[row[1] - 1, row[2] - 1] = row[3]
sprint(ratings, ratings.shape)


# Cosine Similarity
def cosine_similarity(ratings):
    sim = ratings.dot(ratings.T)
    norms = np.array([np.sqrt(np.diagonal(sim))])
    # norms / norms.T == np.outer(norms, norms)
    return sim / norms / norms.T


# ratings is a user-item matrix, so we need to transpose it
cosine_similarity_res = cosine_similarity(ratings.T)
sprint(cosine_similarity_res, cosine_similarity_res.shape)


# Correlation-based Similarity
def pearson_similarity_item(ratings):
    mean_ratings = np.mean(ratings, axis=0)  # 每个物品的平均评分
    ratings_diff = ratings - mean_ratings
    sim = ratings_diff.T.dot(ratings_diff)
    norms = np.array([np.sqrt(np.sum(ratings_diff**2, axis=0))])
    return sim / norms / norms.T


pearson_similarity_item_res = pearson_similarity_item(ratings)
sprint(pearson_similarity_item_res, pearson_similarity_item_res.shape)


# Adjusted Cosine Similarity
def pearson_similarity_item_improve(ratings):
    mean_ratings = np.mean(ratings, axis=1)  # 每个用户的平均评分
    ratings_diff = ratings - mean_ratings[:, np.newaxis]
    sim = ratings_diff.T.dot(ratings_diff)
    norms = np.array([np.sqrt(np.sum(ratings_diff**2, axis=0))])
    return sim / norms / norms.T


pearson_similarity_item_improve_res = pearson_similarity_item_improve(ratings)
sprint(pearson_similarity_item_improve_res, pearson_similarity_item_improve_res.shape)


# Prediction
def predict(ratings, similarity):
    dot = ratings.dot(similarity)
    abs_sim = np.sum(np.abs(similarity), axis=0)
    sprint(dot.shape, abs_sim.shape)
    return dot / abs_sim


cosine_prediction_item = predict(ratings, cosine_similarity_res)
sprint(cosine_prediction_item, cosine_prediction_item.shape)

pearson_prediction_item = predict(ratings, pearson_similarity_item_res)
sprint(pearson_prediction_item, pearson_prediction_item.shape)

pearson_prediction_item_improve = predict(ratings, pearson_similarity_item_improve_res)
sprint(pearson_prediction_item_improve, pearson_prediction_item_improve.shape)