Совместные рекомендации (collaborative filtering)

import math as mt
import numpy as np
import collections as cl

user_items = [
    ["Hadoop", "Big Data", "HBase", "Java", "Spark", "Storm", "Cassandra"],
    ["NoSQL", "MongoDB", "Cassandra", "HBase", "Postgres"],
    ["Python", "scikit-learn", "scipy", "numpy", "statsmodels", "pandas"],
    ["R", "Python", "statistics", "regression", "probability"],
    ["machine learning", "regression", "decision trees", "libsvm"],
    ["Python", "R", "Java", "C++", "Haskell", "programming languages"],
    ["statistics", "probability", "mathematics", "theory"],
    ["machine learning", "scikit-learn", "Mahout", "neural networks"],
    ["neural networks", "deep learning", "Big Data", "artificial intelligence"],
    ["Hadoop", "Java", "MapReduce", "Big Data"],
    ["statistics", "R", "statsmodels"],
    ["C++", "deep learning", "artificial intelligence", "probability"],
    ["pandas", "R", "Python"],
    ["databases", "HBase", "Postgres", "MySQL", "MongoDB"],
    ["libsvm", "regression", "support vector machines"]
]

most_popular = cl.Counter([j for i in user_items for j in i]).most_common()

def popular_recommendation_user(user_id, mx=5):
    return [(k,v) for (k,v) in most_popular if k not in user_items[user_id]][:mx]


user_id = 0 # id клиента

print('Интересы клиента id{}: {}'.format(user_id, user_items[user_id]))
print('Рекомендации для клиента id{}: {}'.format(user_id, popular_recommendation_user(user_id)))

user_id = 1 # id клиента

print('Интересы клиента id{}: {}'.format(user_id, user_items[user_id]))
print('Рекомендации для клиента id{}: {}'.format(user_id, popular_recommendation_user(user_id)))

user_id = 2 # id клиента

print('Интересы клиента id{}: {}'.format(user_id, user_items[user_id]))
print('Рекомендации для клиента id{}: {}'.format(user_id, popular_recommendation_user(user_id)))

def cosim(v1,v2):
    return np.dot(v1,v2)/ mt.sqrt(np.dot(v1,v1)*np.dot(v2,v2))

unique_items = sorted({j for i in user_items for j in i})
unique_items[:5]

def make_vector_items(items):
    return [1 if i in items else 0 for i in unique_items]

user_items_matrix = [make_vector_items(items) for items in user_items]

def similar_users(user_ID):
    pairs = [(user_id, cosim(user_items_matrix[user_ID], user_interest)) 
             for user_id, user_interest in enumerate(user_items_matrix) 
             if user_id != user_ID
             and cosim(user_items_matrix[user_ID], user_interest) > 0]   
    return sorted(pairs, key = lambda x: x[1], reverse =1)

similar_users(0)

def user_user_recommendation(user_ID, mx = 5):
    recommendation = cl.defaultdict(float)
# calculate recommendations over similar users
    for user_id, cosim_similarity in similar_users(user_ID):
        for interest in user_items[user_id]:
            recommendation[interest] += cosim_similarity
# sort them out 
    recommendation = sorted(recommendation.items(), key = lambda x: x[1], reverse=1)
# exclude already existing items
    recommendation = [i for i in recommendation if i[0] not in user_items[user_ID]]  
    return recommendation[:mx]

user_id = 0

print('Интересы клиента id_{0}: \n{1}\n\n\
Рекомендации для клиента id_{0} по наиболее популярным запросам: \n{2}\n\n\
Рекомендации для клиента id_{0} по схожим клиентам: \n{3}'\
      .format(user_id,user_items[user_id],popular_recommendation_user(user_id),user_user_recommendation(user_id)))

user_items

item_users_matrix = [[1 if i in j else 0 for j in user_items] for i in unique_items]
item_users_matrix[:5]

def similar_items(interest_id):
    pairs = [(unique_items[i], cosim(item_users_matrix[interest_id], item))
             for i, item in enumerate(item_users_matrix)
             if i != interest_id and cosim(item_users_matrix[interest_id], item) > 0
            ]
    pairs = sorted(pairs, key = lambda x: x[1], reverse =1)
    return pairs

similar_items(0)[:5]

unique_items_position = {k:v for v,k in enumerate(unique_items)}
print(unique_items[:3])
print(unique_items_position['Big Data'])

def item_item_recommendation(user_id, mx=5):
    rec = cl.defaultdict(float)
    for item in user_items[user_id]:
        sim_items = similar_items(unique_items_position[item])
        for it, similarity in sim_items:
            rec[it] += similarity
    rec = sorted(rec.items(), key=lambda x: x[1], reverse = 1)
    rec = [i for i in rec if i[0] not in user_items[user_id]]
    return rec[:mx]

user_id = 0

print('Интересы клиента id_{0}: \n{1}\n\n\
Рекомендации для клиента id_{0} по наиболее популярным запросам: \n{2}\n\n\
Рекомендации для клиента id_{0} user-user: \n{3}\n\n\
Рекомендации для клиента id_{0} item-item: \n{4}'\
      .format(user_id,
              user_items[user_id],
              popular_recommendation_user(user_id),
              user_user_recommendation(user_id),
              item_item_recommendation(user_id)))