前4天,通过购物车购买
precision:0.0153313777615
recall:0.177594903106
f1:0.0282260616417
前4天,通过收藏夹购买
precision:0.00326368668608
recall:0.0270772498009
f1:0.00582524271845
前4天 , 通过购物车&收藏夹购买
precision:0.0166666666667
recall:0.00955667640032
f1:0.0121477982116
前两天,通过购物车购买
precision:0.0256597238047
recall:0.149455800372
f1:0.0437995954567
前两天,通过收藏夹购买
precision:0.00478835472132
recall:0.0199097425007
f1:0.00772002058672
前2天 , 通过购物车&收藏夹购买
precision:0.0256166982922
recall:0.00716750730024
f1:0.0112009956441
前一天,通过购物车购买
data1
precision:0.0310559006211
recall:0.0876028670029
f1:0.0458556242618
data2
precision:0.0271345235991
recall:0.0793897077838
f1:0.0404452934589
前一天,通过收藏夹购买
precision:0.00696774193548
recall:0.0143350146005
f1:0.0093774420422
前一天,通过购物车&收藏夹购买
precision:0.0308370044053
recall:0.00371648526679
f1:0.00663349917081
可见,仅一天数据对于购物车行为来说是最好的,收藏夹在短期内是个坑跌货,甚至取交集也是个坑跌货。
所以,保留一天数据,购物车数据
去掉没有购买行为的id,基于前一天购买
data1
precision:0.0317643661565
recall:0.0876028670029
f1:0.0466233399265
data2
precision:0.0276277897768
recall:0.0793897077838
f1:0.0409907203418
有略微提升
去掉买了又买
data1
precision:0.0371799368643
recall:0.0844173082028
f1:0.0516233766234
data2
precision:0.0319275159098
recall:0.0765451254202
f1:0.0450601309179
准确率及f1值提升很多
基于一天购物车数据,8点后效果
313
precision:0.0340513489991
recall:0.0830899920361
f1:0.0483061964658
10点后效果
275
precision:0.0328672164456
recall:0.0730023891691
f1:0.0453271798253
12点后效果
234
precision:0.032298136646
recall:0.0621183966021
f1:0.0424990918997
16点后效果
184
precision:0.0360501567398
recall:0.048845234935
f1:0.0414834855146
20点后效果
109
precision:0.0400440852314
recall:0.0289354924343
f1:0.0335953151487
22点效果
56
precision:0.0683760683761
recall:0.0148659410672
f1:0.0244221543829
9点效果
296
precision:0.0335182878496
recall:0.0785771170693
f1:0.046991585966
7点后效果
320
precision:0.0334868145668
recall:0.0849482346695
f1:0.0480372288524
8点后效果最可靠,网传仅交最后几小时数据效果也就3左右的样子
data2
所有时间
296
precision:0.0271435121504
recall:0.0765451254202
f1:0.0400758191172
6点后
292
precision:0.0283165244375
recall:0.0755107318335
f1:0.0411876719092
7点后
291
precision:0.0288461538462
recall:0.0752521334368
f1:0.0417054819061
8点后
289
precision:0.0293938161107
recall:0.0747349366434
f1:0.0421928607928
9点后
286
precision:0.0301942567568
recall:0.0739591414533
f1:0.0428817752455
10点后
282
precision:0.0314556609035
recall:0.0729247478666
f1:0.0439526184539
11点后
280
precision:0.0336215177714
recall:0.0724075510732
f1:0.0459204592046
12点后
275
precision:0.035378875595
recall:0.0711145590897
f1:0.0472508591065
13点后
247
precision:0.0352756355327
recall:0.0638738039824
f1:0.0454503634189
15点后
231
precision:0.0387129210659
recall:0.0597362296354
f1:0.0469798657718
16点后
precision:0.0401098901099
recall:0.0566330488751
f1:0.0469604374397
17点后
206
precision:0.0409136047666
recall:0.0532712697181
f1:0.0462817344417
一个在8点,一个在12点,有时间多测几组数据
删除一个月内没有被购买的物品的记录
原数据:
320
precision:0.0334868145668
recall:0.0849482346695
f1:0.0480372288524
删除后数据:
43
precision:0.0452155625657
recall:0.0114149190337
f1:0.0182280627384
不跑不知道,一跑吓一跳啊。。
准确率是上去了,召回率呢。。。召回率。。。回率。。。率。。。
原数据
320
precision:0.0334868145668
recall:0.0849482346695
f1:0.0480372288524
删除购买天数《1的数据
320
precision:0.0342099636519
recall:0.0849482346695
f1:0.0487767700633
删除购买天数《2的数据
317precision:0.0354189944134
recall:0.0841518449695
f1:0.0498545254384
删除购买天数《3的数据
305
precision:0.0357142857143
recall:0.0809662861694
f1:0.0495652880475
《4
282
precision:0.0358687356907
recall:0.0748606318025
f1:0.0484994410525
删除购买天数《5的数据
198
precision:0.0309423347398
recall:0.0525617202018
f1:0.0389533739917
删除购买天数《10的数据
105
precision:0.028665028665
recall:0.0278736395009
f1:0.028263795424
data2
同丄
原数据:
219
precision:0.0401098901099
recall:0.0566330488751
f1:0.0469604374397
《1
219
precision:0.040835353347
recall:0.0566330488751
f1:0.0474539544962
《2
214
precision:0.0410433448408
recall:0.0553400568916
f1:0.0471313731968
《3
211
precision:0.0431492842536
recall:0.0545642617016
f1:0.048190019413
《4
207
precision:0.0462259937472
recall:0.0535298681148
f1:0.0496105452367
《5
201
precision:0.0499378881988
recall:0.0519782777347
f1:0.0509376583882
《6
185
precision:0.0513888888889
recall:0.0478407033876
f1:0.0495513593143
《7
163
precision:0.0510811657788
recall:0.0421515386605
f1:0.0461887220176
删除购买天数小于5/2分别达到峰值
源数据
320
precision:0.0334868145668
recall:0.0849482346695
f1:0.0480372288524
总购买/总购物车比值处于0.2~5
235
precision:0.0465992464803
recall:0.0623838598354
f1:0.0533484676504
0.2~4
235
precision:0.0467196819085
recall:0.0623838598354
f1:0.053427304763
0.2~3
234
precision:0.0466321243523
recall:0.0621183966021
f1:0.0532726237906
0.3~3
200
precision:0.0552791597568
recall:0.0530926466684
f1:0.054163845633
0.3~4
201
precision:0.0553719008264
recall:0.0533581099018
f1:0.0543463566311
0.1~2
296
precision:0.039414114514
recall:0.0785771170693
f1:0.0524962312672
0.1~3
297
precision:0.0392182754523
recall:0.0788425803026
f1:0.052380952381
0.1~4
298
precision:0.0392880685564
recall:0.079108043536
f1:0.0525017618041
两数据对比,0.1~0.4可提升0.5的f1值,
data2
元数据
219
precision:0.0401098901099
recall:0.0566330488751
f1:0.0469604374397
0.3~4
115
precision:0.0610403397028
recall:0.0297388156193
f1:0.0399930446879
0.2~4
143
precision:0.0528260066494
recall:0.0369795707267
f1:0.0435047155461
0.1~4
201
precision:0.0492164544564
recall:0.0519782777347
f1:0.0505596780279
0.1~3
201
precision:0.0493009565857
recall:0.0519782777347
f1:0.0506042296073
0.1~2
200
precision:0.0494682166708
recall:0.051719679338
f1:0.0505689001264
0.1~4为最佳效果
原数据
320
precision:0.0334868145668
recall:0.0849482346695
f1:0.0480372288524
3天没有购买行为的
149
precision:0.0216286834083
recall:0.039554021768
f1:0.0279654654655
3天 购买/点击
》0.00001
149
precision:0.0216286834083
recall:0.039554021768
f1:0.0279654654655
》0.0001
149
precision:0.0216286834083
recall:0.039554021768
f1:0.0279654654655
》0.001
143
precision:0.0210417892878
recall:0.0379612423679
f1:0.0270756413898
》0.01
62
precision:0.0154305624689
recall:0.0164587204672
f1:0.0159280667951
》0.1
0
0
0
0
此数据暂不用
购买/点击
元数据
320
precision:0.0334868145668
recall:0.0849482346695
f1:0.0480372288524
》0.001
319
precision:0.0345986984816
recall:0.0846827714362
f1:0.049126049126
》0.005
286
precision:0.0386643233743
recall:0.0759224847359
f1:0.0512361160874
》0.008
235
precision:0.0399116847826
recall:0.0623838598354
f1:0.0486794407043
》0.003
311
precision:0.0365495357856
recall:0.0825590655694
f1:0.0506679700228
》0.004
300
precision:0.0377216144851
recall:0.0796389700027
f1:0.0511945392491
》0.002
319
precision:0.0359598692368
recall:0.0846827714362
f1:0.0504826713088
最佳为0.005
data2
元数据
219
precision:0.0401098901099
recall:0.0566330488751
f1:0.0469604374397
》0.001
217
precision:0.0409665848594
recall:0.0561158520817
f1:0.0473592317765
》0.002
216
precision:0.0426877470356
recall:0.055857253685
f1:0.048392517083
》0.003
215
precision:0.0442477876106
recall:0.0555986552883
f1:0.0492780197112
》0.004
213
precision:0.0458458889367
recall:0.055081458495
f1:0.050041113591
》0.005
207
precision:0.0479499652536
recall:0.0535298681148
f1:0.0505865102639
》0.006
193
precision:0.0490096495683
recall:0.0499094905612
f1:0.0494554772582
两个数据最佳都是0.005,平均提升f1值0.4%
对于依赖购物车购买的用户:交集,集合来自邹
121
precision:0.0432451751251
recall:0.0321210512344
f1:0.0368621477532
对于不使用购物车购买的用户:去掉,集合来自邹
320
precision:0.0338123415046
recall:0.0849482346695
f1:0.0483712493387
有微提升,实际集合中没有这些用户,估计已经被去掉了
元数据
320
precision:0.0334868145668
recall:0.0849482346695
f1:0.0480372288524
点击量大于200
319
precision:0.0339145226451
recall:0.0846827714362
f1:0.0484323996053
点击量大于300
316
precision:0.034095813552
recall:0.0838863817361
f1:0.0484848484848
点击量大于400
310
precision:0.0341146693078
recall:0.0822936023361
f1:0.0482340127587
元数据
219
precision:0.0401098901099
recall:0.0566330488751
f1:0.0469604374397
点击大于400
213
precision:0.0417483339867
recall:0.055081458495
f1:0.0474969338834
大于300
213
precision:0.0406101048618
recall:0.055081458495
f1:0.0467515364355
大于200
216
precision:0.0401935243766
recall:0.055857253685
f1:0.0467481874256
大于100
218
precision:0.0401251610528
recall:0.0563744504784
f1:0.0468817204301
数据不一致,暂不用