非パーソナライズ型の提案
Pythonでつくるレコメンデーションエンジン
Robert O'Callaghan
Director of Data
ペアの特定
0 User_223 The Great Gatsby <---| 同一ユーザーが読了
1 User_223 The Catcher in the Rye <---|
2 User_131 The Lord of the Rings
3 User_965 Little Women <---| 同一ユーザーが読了
4 User_965 Fifty Shades of Grey <---|
... ...
順列と組合せの違い
| User | book_title |
|---|---|
| User_233 | The Great Gatsby |
| User_233 | The Catcher in the Rye |
To:
| Book A | Book B | |
|---|---|---|
| 0 | The Great Gatsby | The Catcher in the Rye |
| 1 | The Catcher in the Rye | The Great Gatsby |
The Great Gatsby と一緒に見られた本 -> The Catcher in the Rye
The Catcher in the Rye と一緒に見られた本 -> The Great Gatsby
ペア作成関数の作成
from itertools import permutations
def create_pairs(x):
return pairs
ペア作成関数の作成
from itertools import permutations
def create_pairs(x):
pairs = permutations(x.values, 2)
return pairs
permutations(list, length_of_permutations))すべての順列を含むイテラブルを生成
ペア作成関数の作成
from itertools import permutations
def create_pairs(x):
pairs = list(permutations(x.values, 2))
return pairs
permutations(list, length_of_permutations))すべての順列を含むイテラブルを生成list()このオブジェクトをリストに変換
ペア作成関数の作成
from itertools import permutations
def create_pairs(x):
pairs = pd.DataFrame(list(permutations(x.values, 2)),
columns=['book_a','book_b'])
return pairs
permutations(list, length_of_permutations))すべての順列を含むイテラブルを生成list()このオブジェクトをリストに変換pd.DataFrame()リストをbook_aとbook_b列のDataFrameに変換
関数をデータに適用
book_pairs = book_df.groupby('userId')['book_title'].apply(perm_function)
print(book_pairs.head())
book_a book_b
userId
User_223 0 The Great Gatsby The Catcher in the Rye
1 The Catcher in the Rye The Great Gatsby
User_965 0 Little Women 40 Shades of Grey
1 40 Shades of Grey Little Women
User_773 0 The Twilight Saga Harry Potter and the Sorcerer's Stone
...
結果の整形
book_pairs = book_pairs.reset_index(drop=True)
print(book_pairs.head())
book_a book_b
0 The Great Gatsby The Catcher in the Rye
1 The Catcher in the Rye The Great Gatsby
3 Little Women 40 Shades of Grey
4 40 Shades of Grey Little Women
5 The Twilight Saga Harry Potter and the Sorcerer's Stone
...
ペア数の集計
pair_counts = book_pairs.groupby(['book_a', 'book_b']).size()
book_a book_b
The Twilight Saga Fifty Shades of Grey 16
Pride and Prejudice 12
...
pair_counts_df = pair_counts.to_frame(name = 'size').reset_index()
print(pair_counts_df.head())
book_a book_b size
1 The Twilight Saga Fifty Shades of Grey 16
2 The Twilight Saga Pride and Prejudice 12
...
推薦の検索
pair_counts_sorted = pair_counts_df.sort_values('size', ascending=False)
pair_counts_sorted[pair_counts_sorted['book_a'] == 'Lord of the Rings']
book_a book_b size
137 Lord of the Rings The Hobbit 12
147 Lord of the Rings Harry Potter and the Sorcerer's Stone 10
143 Lord of the Rings The Colour of Magic 9
...
Passons à la pratique !
Pythonでつくるレコメンデーションエンジン
