데이터 기반으로 의사결정하기-푸쉬 노티피케이션 타임

학습목표

  1. 푸쉬 노티피케이션 타임 의사 결정 하기

푸쉬 노티피케이션 이란?

push notification은 단적으로 설명한다면 서버에서 발생한 Event를 특정 클라이언트에게 Event 발생 사실을 통지하는 기술입니다. 주변에서 볼 수 있는 가장 흔히 볼 수 있는 사례가 SNS application의 메시지 수신 알림입니다.

import numpy as np
import pandas as pd
# seaborn
import seaborn as sns
COLORS = sns.color_palette()

%matplotlib inline

def plot_bar(df, xlabel, ylabel, title, figsize=(20, 10), color=COLORS[-1], rotation=45):
    plot = df.plot(kind='bar', color=color, figsize=figsize)
    plot.set_xlabel(xlabel, fontsize=10)
    plot.set_ylabel(ylabel, fontsize=10)
    plot.set_title(title, fontsize=12)
    plot.set_xticklabels(labels=df.index, rotation=rotation)

dtypes = {
    'UnitPrice': np.float32,
    'CustomerID': np.int32,
    'Quantity': np.int32
}
retail = pd.read_csv('./OnlineRetailClean.csv', dtype=dtypes)
retail['InvoiceDate'] = pd.to_datetime(retail['InvoiceDate'], infer_datetime_format=True)
retail.head()
Unnamed: 0 InvoiceNo StockCode Description Quantity InvoiceDate UnitPrice CustomerID Country CheckoutPrice
0 0 536365 85123A WHITE HANGING HEART T-LIGHT HOLDER 6 2010-12-01 08:26:00 2.55 17850 United Kingdom 15.30
1 1 536365 71053 WHITE METAL LANTERN 6 2010-12-01 08:26:00 3.39 17850 United Kingdom 20.34
2 2 536365 84406B CREAM CUPID HEARTS COAT HANGER 8 2010-12-01 08:26:00 2.75 17850 United Kingdom 22.00
3 3 536365 84029G KNITTED UNION FLAG HOT WATER BOTTLE 6 2010-12-01 08:26:00 3.39 17850 United Kingdom 20.34
4 4 536365 84029E RED WOOLLY HOTTIE WHITE HEART. 6 2010-12-01 08:26:00 3.39 17850 United Kingdom 20.34

쿠폰 발송을 할때, push를 언제 보내는게 좋을까?

  • 고객에게 쿠폰 발송을 한다고 기획하고, 회의를 한다고 가정해보겠습니다.
  • A: 쿠폰을 언제보내는게 좋을까요?
  • B: 아침에 출퇴근 시간에 보내는게 좋을까요?
  • C: 점심 먹고 졸린데 그때 보내보죠?
  • D: 흠 자기전에 스마트폰 많이 하던데 그때는 어떨까요?
  • A: 그러면 평균 시간을 내볼까요?

  • K: 아 데이터를 확인해보는게 맞지 않을까요? 언제 고객이 주로 주문을 하는지?

  • 위에서 처럼 실제로 회의를 하다보면 의사결정이 본인/주변의 경험에 의해서 이뤄지는 것을 많이 볼 수 있습니다.
  • 주문이 이뤄지는 시간을 고려하지 않고 막무가내로 보낸다면 아무 의미가 없고, 추후 같은 이벤트 발생시에도 판단 근거가 없게 됨
  • 현상태에서는 가장 많이 주문이 일어나는 시점에서 하는 것이 가장 직관적인 판단
      1. 데이터로 파악
      1. 가설 제시
      1. 가설 검증
      1. 1-3 반복
  • 시간(hour, minute)과 주로 관련되기 때문에 역시 InvoiceDate가 중요한 feature
order_by_hour = retail.set_index('InvoiceDate').groupby(lambda date: date.hour).count()
order_by_hour
Unnamed: 0 InvoiceNo StockCode Description Quantity UnitPrice CustomerID Country CheckoutPrice
6 1 1 1 1 1 1 1 1 1
7 379 379 379 379 379 379 379 379 379
8 8690 8690 8690 8690 8690 8690 8690 8690 8690
9 21944 21944 21944 21944 21944 21944 21944 21944 21944
10 37997 37997 37997 37997 37997 37997 37997 37997 37997
11 49084 49084 49084 49084 49084 49084 49084 49084 49084
12 72065 72065 72065 72065 72065 72065 72065 72065 72065
13 64026 64026 64026 64026 64026 64026 64026 64026 64026
14 54118 54118 54118 54118 54118 54118 54118 54118 54118
15 45369 45369 45369 45369 45369 45369 45369 45369 45369
16 24089 24089 24089 24089 24089 24089 24089 24089 24089
17 13071 13071 13071 13071 13071 13071 13071 13071 13071
18 2928 2928 2928 2928 2928 2928 2928 2928 2928
19 3321 3321 3321 3321 3321 3321 3321 3321 3321
20 802 802 802 802 802 802 802 802 802 
order_by_hour = retail.set_index('InvoiceDate').groupby(lambda date: date.hour).count()['CustomerID']
order_by_hour

6         1
7       379
8      8690
9     21944
10    37997
11    49084
12    72065
13    64026
14    54118
15    45369
16    24089
17    13071
18     2928
19     3321
20      802
Name: CustomerID, dtype: int64

plot_bar(order_by_hour, 'hour', '# orders', 'Order by hour')

output_7_0

def half_an_hour(date):
    minute = ':00'
    if date.minute > 30:
        minute = ':30'
    hour = str(date.hour)
    if date.hour < 10:
        hour = '0' + hour
    
    return hour + minute

order_by_hour_half = retail.set_index('InvoiceDate').groupby(half_an_hour).count()['CustomerID']
order_by_hour_half

06:00        1
07:30      379
08:00     3145
08:30     5545
09:00     9364
09:30    12580
10:00    16950
10:30    21047
11:00    18925
11:30    30159
12:00    37174
12:30    34891
13:00    31131
13:30    32895
14:00    26958
14:30    27160
15:00    24227
15:30    21142
16:00    14316
16:30     9773
17:00     8889
17:30     4182
18:00     1715
18:30     1213
19:00     1534
19:30     1787
20:00      802
Name: CustomerID, dtype: int64

order_by_hour_half / order_by_hour_half.sum()

06:00    0.000003
07:30    0.000953
08:00    0.007904
08:30    0.013936
09:00    0.023534
09:30    0.031617
10:00    0.042600
10:30    0.052897
11:00    0.047564
11:30    0.075798
12:00    0.093429
12:30    0.087691
13:00    0.078241
13:30    0.082675
14:00    0.067753
14:30    0.068261
15:00    0.060890
15:30    0.053136
16:00    0.035980
16:30    0.024562
17:00    0.022341
17:30    0.010511
18:00    0.004310
18:30    0.003049
19:00    0.003855
19:30    0.004491
20:00    0.002016
Name: CustomerID, dtype: float64

plot_bar(order_by_hour_half, 'half an hour', '# orders', 'order by half an hour')

output_11_0

개인화된 push notification

  • 아마존을 필두로, 개인화(personalization)하여 맞춤으로 사용자마다 최적의 솔루션을 찾는것이 트렌드가 됨
  • 사용자별로 소비의 패턴이 다를 수 있기 때문에, 가장 많이 구매한 시간대를 찾아서 해당 시간대에 쿠폰을 발송!

사용자별 각 시간별 주문 량 계산하기

order_count_by_hour = retail.set_index('InvoiceDate').groupby(['CustomerID', lambda date: date.hour]).count()['StockCode']
order_count_by_hour #12346 ID 는 10 시에 1건 , 12347  ID는 8 시에 22건 

CustomerID    
12346       10     1
12347       8     22
            10    24
            12    47
            13    18
                  ..
18283       15     1
            16    56
            19    87
18287       9      3
            10    67
Name: StockCode, Length: 11205, dtype: int64

order_count_by_hour.loc[12347] #2시에 가장 많은 주문을 함

8     22
10    24
12    47
13    18
14    60
15    11
Name: StockCode, dtype: int64

사용자별 최대 주문 시간 계산하기

  • 가장 많은 주문량을 보인 시간을 계산
order_count_by_hour.groupby('CustomerID').idxmax() #idxmax 함수는 최댓값을 가진 index를 반환한다

CustomerID
12346    (12346, 10)
12347    (12347, 14)
12348    (12348, 19)
12349     (12349, 9)
12350    (12350, 16)
            ...     
18280     (18280, 9)
18281    (18281, 10)
18282    (18282, 13)
18283    (18283, 14)
18287    (18287, 10)
Name: StockCode, Length: 4338, dtype: object

idx = order_count_by_hour.groupby('CustomerID').idxmax() #idxmax 함수는 최댓값을 가진 index를 반환한다

해당 시간 indexing

result = order_count_by_hour.loc[idx]
result

CustomerID    
12346       10      1
12347       14     60
12348       19     17
12349       9      73
12350       16     17
                 ... 
18280       9      10
18281       10      7
18282       13      7
18283       14    201
18287       10     67
Name: StockCode, Length: 4338, dtype: int64

result.reset_index()
CustomerID level_1 StockCode
0 12346 10 1
1 12347 14 60
2 12348 19 17
3 12349 9 73
4 12350 16 17
... ... ... ...
4333 18280 9 10
4334 18281 10 7
4335 18282 13 7
4336 18283 14 201
4337 18287 10 67

4338 rows × 3 columns

result.reset_index().groupby('level_1').groups #시간대별로 그룹화하기

{7: [73, 269, 319, 344, 375, 893, 1667, 2317], 8: [46, 58, 87, 126, 172, 179, 187, 260, 278, 279, 282, 292, 306, 347, 399, 429, 496, 503, 526, 533, 549, 552, 651, 671, 747, 755, 784, 792, 800, 803, 806, 821, 838, 877, 883, 920, 944, 947, 951, 954, 1008, 1093, 1106, 1120, 1138, 1172, 1173, 1217, 1251, 1397, 1422, 1424, 1436, 1472, 1512, 1616, 1621, 1666, 1668, 1678, 1687, 1734, 1759, 1761, 1774, 1791, 1815, 1827, 1846, 1859, 1895, 1900, 1903, 1996, 2018, 2023, 2054, 2085, 2108, 2117, 2167, 2172, 2253, 2380, 2383, 2403, 2404, 2417, 2427, 2462, 2464, 2643, 2749, 2776, 2781, 2896, 2936, 2949, 3021, 3130, ...], 9: [3, 9, 26, 30, 33, 35, 37, 48, 60, 66, 75, 84, 86, 90, 100, 106, 107, 121, 127, 135, 138, 142, 144, 146, 154, 159, 181, 199, 230, 240, 264, 265, 267, 277, 280, 286, 294, 298, 328, 333, 336, 342, 343, 352, 362, 366, 385, 402, 421, 459, 470, 475, 478, 482, 483, 509, 517, 519, 574, 603, 615, 630, 636, 642, 644, 691, 701, 706, 707, 746, 749, 752, 764, 770, 781, 783, 818, 825, 829, 844, 859, 874, 887, 925, 934, 950, 969, 981, 992, 998, 1003, 1004, 1016, 1032, 1038, 1045, 1050, 1053, 1063, 1082, ...], 10: [0, 11, 21, 27, 28, 41, 42, 45, 49, 51, 55, 61, 77, 93, 94, 103, 104, 105, 110, 113, 122, 132, 137, 140, 147, 150, 155, 156, 165, 168, 169, 174, 178, 182, 186, 195, 205, 206, 208, 216, 217, 222, 231, 233, 242, 251, 252, 255, 263, 275, 276, 287, 288, 290, 293, 301, 310, 314, 322, 331, 337, 339, 341, 348, 359, 360, 361, 363, 364, 365, 379, 381, 407, 437, 439, 441, 443, 450, 464, 465, 468, 471, 481, 499, 500, 511, 516, 529, 541, 553, 560, 563, 570, 578, 584, 586, 590, 591, 595, 596, ...], 11: [29, 32, 34, 57, 99, 102, 111, 124, 139, 148, 163, 171, 176, 188, 207, 220, 223, 228, 234, 246, 253, 254, 256, 266, 272, 311, 313, 315, 324, 326, 330, 346, 349, 355, 356, 380, 393, 400, 419, 423, 424, 427, 430, 431, 449, 458, 462, 485, 487, 515, 521, 528, 542, 545, 550, 559, 567, 569, 575, 605, 616, 635, 648, 650, 654, 658, 664, 677, 678, 680, 692, 693, 694, 702, 712, 729, 744, 748, 763, 765, 771, 778, 793, 798, 812, 819, 824, 828, 831, 837, 843, 846, 851, 856, 866, 868, 869, 873, 875, 903, ...], 12: [12, 20, 22, 36, 50, 62, 64, 67, 72, 74, 81, 116, 120, 123, 145, 151, 158, 160, 164, 189, 191, 193, 200, 203, 209, 226, 237, 238, 241, 243, 244, 245, 249, 259, 270, 271, 284, 297, 305, 308, 317, 327, 332, 335, 350, 357, 367, 371, 376, 377, 388, 390, 391, 397, 398, 403, 404, 414, 415, 418, 428, 432, 435, 436, 440, 451, 460, 473, 477, 488, 489, 490, 492, 495, 504, 510, 525, 540, 565, 568, 577, 582, 585, 594, 598, 599, 611, 612, 613, 622, 624, 625, 631, 634, 643, 649, 653, 655, 666, 675, ...], 13: [7, 8, 14, 16, 18, 23, 43, 44, 52, 59, 70, 71, 76, 82, 83, 97, 98, 108, 112, 114, 115, 119, 143, 149, 166, 167, 183, 190, 198, 201, 202, 204, 212, 213, 225, 227, 232, 236, 239, 257, 258, 262, 300, 303, 312, 329, 340, 351, 353, 368, 369, 372, 374, 382, 383, 384, 394, 396, 406, 416, 417, 422, 438, 445, 448, 452, 455, 456, 466, 474, 493, 505, 506, 512, 534, 535, 537, 548, 551, 556, 561, 581, 601, 609, 610, 614, 617, 623, 632, 639, 647, 659, 660, 668, 669, 676, 681, 684, 685, 687, ...], 14: [1, 5, 25, 31, 38, 40, 54, 56, 69, 78, 79, 85, 88, 95, 96, 101, 109, 118, 125, 129, 130, 131, 141, 152, 162, 173, 175, 177, 196, 197, 215, 219, 221, 247, 273, 281, 291, 295, 296, 318, 325, 334, 354, 358, 389, 395, 401, 405, 408, 412, 413, 425, 433, 457, 461, 463, 480, 486, 491, 494, 501, 507, 520, 522, 524, 530, 538, 539, 555, 557, 562, 572, 573, 579, 583, 588, 589, 618, 626, 627, 640, 641, 645, 646, 661, 663, 665, 696, 697, 699, 720, 725, 726, 735, 745, 760, 761, 799, 801, 809, ...], 15: [13, 15, 17, 24, 65, 68, 91, 92, 117, 134, 136, 161, 170, 180, 184, 194, 211, 214, 218, 229, 235, 250, 268, 274, 285, 299, 304, 307, 309, 338, 345, 373, 378, 386, 392, 409, 410, 411, 434, 444, 446, 467, 476, 479, 497, 498, 502, 513, 514, 527, 531, 532, 536, 544, 564, 566, 576, 592, 600, 602, 607, 619, 620, 621, 629, 638, 674, 689, 705, 714, 734, 739, 740, 777, 787, 789, 791, 796, 804, 814, 823, 827, 832, 855, 857, 861, 882, 888, 900, 902, 935, 938, 941, 952, 953, 962, 972, 977, 979, 982, ...], 16: [4, 10, 19, 39, 53, 128, 133, 157, 192, 210, 224, 248, 302, 316, 323, 370, 387, 420, 442, 447, 454, 469, 472, 484, 518, 523, 543, 546, 554, 558, 580, 587, 604, 628, 657, 662, 672, 682, 704, 788, 794, 833, 834, 847, 850, 908, 930, 940, 964, 970, 999, 1029, 1036, 1048, 1067, 1096, 1107, 1115, 1116, 1144, 1174, 1177, 1189, 1219, 1224, 1239, 1270, 1273, 1279, 1288, 1314, 1322, 1331, 1343, 1355, 1359, 1366, 1377, 1380, 1468, 1473, 1477, 1484, 1488, 1490, 1500, 1506, 1526, 1564, 1566, 1574, 1585, 1638, 1676, 1692, 1772, 1799, 1820, 1834, 1848, ...], 17: [6, 63, 89, 153, 185, 261, 283, 289, 321, 426, 508, 547, 571, 593, 652, 670, 703, 719, 722, 754, 836, 845, 907, 936, 1019, 1088, 1140, 1188, 1240, 1296, 1379, 1489, 1540, 1578, 1588, 1590, 1603, 1628, 1640, 1642, 1679, 1739, 1742, 1889, 1906, 1940, 2058, 2156, 2169, 2274, 2279, 2340, 2374, 2408, 2443, 2515, 2566, 2568, 2583, 2594, 2602, 2621, 2644, 2661, 2853, 2856, 2886, 2928, 2948, 2978, 2995, 3004, 3069, 3102, 3141, 3173, 3185, 3226, 3230, 3277, 3323, 3401, 3404, 3425, 3443, 3461, 3462, 3466, 3512, 3543, 3559, 3598, 3639, 3659, 3677, 3690, 3716, 3727, 3736, 3739, ...], 18: [80, 320, 453, 637, 767, 862, 879, 1128, 1326, 1378, 1498, 1519, 1624, 1652, 1758, 1768, 1844, 2879, 3198, 3467, 3511, 3537, 3767, 3802, 3820, 3837, 4072, 4077, 4079, 4273], 19: [2, 47, 667, 1589, 1591, 1639, 1730, 1776, 1928, 2044, 2448, 2548, 2876, 3002, 3047, 3261, 3274, 3479, 3556, 3652, 3685, 3789, 3812, 4324], 20: [1646, 1943, 3804, 3838, 4050, 4110]}
  • 시간대별로 어떤 고객이 가장많은 주문을 했는지 확인가능