Note : Les descriptions sont présentées dans la langue officielle dans laquelle elles ont été soumises.
METHOD AND SYSTEM FOR INTELLIGENT MARKETING
BACKGROUND OF THE INVENTION
Technical Field
[0001] The present invention relates to the field of big data technology, and
more particularly to
a method and a system for intelligent marketing.
Description of Related Art
[0002] With the development of information technology and the incessant
expansion of
commodity marketing businesses, the traditional commodity marketing mode can
no
longer meet the requirements of informatization of the modern society, and the
intelligent
marketing mode is often used instead to push commodities to target customers,
so as to
enhance purchasing possibilities of users and hence enhance marketing
efficiency.
[0003] However, in such machine learning models of the logistical regression
type as the
intelligent marketing model, continuous variables are usually required to
possess better
monotonicity or least assume positive U-shapes, inverse U-shapes, so that the
model
achieves better prediction effect or interpretability; there are many types of
variables in
actual application, including such thousands of variables as user basic
information,
browsing behaviors, and purchasing behaviors, etc., and it is usually
impossible to require
the trend chart of each type of variables to exhibit strict monotonicity or
assume U-shape,
at this time it is needed to screen out variables with better degree of
monotonicity or U-
shape and then to input the same in the intelligent marketing model. The
currently
available method relies on manual inspection to check the variable trend
curve, and this
calls for a great deal of workload with low efficiency under the circumstance
in which
there are massive variables.
1
Date Recue/Date Received 2022-03-01
SUMMARY OF THE INVENTION
[0004] An objective of the present invention is to provide a method and a
system for intelligent
marketing, totally in place of manual work, to automatically screen out
variables with
better degree of monotonicity or U-shape and then input the same to an
intelligent
marketing model to acquire target customers, push commodities to the target
customers,
and enhance marketing efficiency.
[0005] In order to achieve the above objective, the present invention provides
the following
technical solutions.
[0006] There is provided a method for intelligent marketing, which method
comprises:
[0007] obtaining variables and drawing a variable trend curve, and obtaining
variables with
monotonicity features or U-shape features on the basis of the variable trend
curve and a
preset trend recognizing rule;
[0008] employing the variables with monotonicity features or U-shape features
to train an
intelligent marketing model, and simultaneously employing a stepwise
regression
operation to screen and obtain one or more target variable(s) adapted to the
intelligent
marketing model; and
[0009] employing the target variable(s) and the intelligent marketing model to
obtain a target
customer, and pushing a commodity to the target customer.
[0010] Preferably, the variables are binned to obtain a plurality of bins, and
the variable trend
curve is drawn on the basis of the bins.
[0011] Further, the step of obtaining variables with monotonicity features on
the basis of the
variable trend curve and a preset trend recognizing rule includes:
[0012] calculating a total variation TV of the variables on the basis of the
variable trend curve,
wherein the total variation TV of the variables is a sum total of amplitudes
between every
2
Date Recue/Date Received 2022-03-01
two adjacent bins of the variable trend curve;
[0013] calculating an absolute value of a difference between left and right
two endpoints of the
variable trend curve and marking the absolute value as a first difference
value ADi, and
obtaining a monotonicity index M index of the variables according to the total
variation
TV and the first difference value ADi, where M index = TV/ADi; and
[0014] screening out variables with monotonicity features on the basis of a
preset monotonicity
index threshold.
[0015] Preferably, the step of obtaining variables with U-shape features on
the basis of the
variable trend curve and a preset trend recognizing rule includes:
[0016] obtaining the maximum value and the minimum value of the variable trend
curve except
for left and right two endpoints;
[0017] calculating a sum of absolute values of difference values between the
left and right two
endpoints of the variable trend curve respectively with respect to the minimum
value of
the variable trend curve and marking the sum as a second difference value AD2,
and
obtaining a positive U-shape index U index 1 of the variables according to the
total
variation TV and the second difference value AD2, where U index 1 = TV/AD2;
and/or
[0018] calculating a sum of absolute values of difference values between the
left and right two
endpoints of the variable trend curve respectively with respect to the maximum
value of
the variable trend curve and marking the sum as a third difference value AD3,
and
obtaining an inverse U-shape index U index 2 of the variables according to the
total
variation TV and the third difference value AD3, where U index 2 = TV/AD3; and
[0019] screening out variables with U-shape features on the basis of preset
positive U-shape
index threshold and inverse U-shape index threshold.
[0020] Preferably, the monotonicity index threshold, the positive U-shape
index threshold and
the inverse U-shape index threshold are valuated in the range of [1, 1.51.
[0021] Further, when the maximum value of the variable trend curve except for
the left and right
3
Date Recue/Date Received 2022-03-01
two endpoints is smaller than values of the left and right two endpoints
simultaneously,
the inverse U-shape index U index 2 is not calculated; and
[0022] when the minimum value of the variable trend curve except for the left
and right two
endpoints is greater than values of the left and right two endpoints
simultaneously, the
positive U-shape index U index 1 is not calculated.
[0023] Preferably, the step of employing the variables with monotonicity
features or U-shape
features to train an intelligent marketing model, and simultaneously employing
a stepwise
regression operation to screen and obtain one or more target variable(s)
adapted to the
intelligent marketing model includes:
[0024] data-preprocessing the variables with monotonicity features or U-shape
features;
[0025] screening important variables out of the preprocessed variables; and
[0026] inputting the important variables in the intelligent marketing model,
and simultaneously
employing a stepwise regression operation to screen and obtain one or more
target
variable(s) adapted to the intelligent marketing model.
[0027] Preferably, the data-preprocessing includes filling missing values,
processing abnormal
values, and one-hot coding with respect to categorical variables.
[0028] Preferably, the step of screening important variables out of the
preprocessed variables
includes:
[0029] calculating IV values and PSI values of the variables respectively; and
[0030] screening out important variables whose IV values are greater than an
IV threshold and
whose PSI values are smaller than a PSI threshold.
[0031] There is provided a system for intelligent marketing, which system
comprises:
[0032] a first variable screening module, a second variable screening module,
and a marketing
pushing module, of which:
[0033] the first variable screening module is employed for obtaining variables
and drawing a
4
Date Recue/Date Received 2022-03-01
variable trend curve, and obtaining variables with monotonicity features or U-
shape
features on the basis of the variable trend curve and a preset trend
recognizing rule;
[0034] the second variable screening module is employed for employing the
variables with
monotonicity features or U-shape features to train an intelligent marketing
model, and
simultaneously employing a stepwise regression operation to screen and obtain
one or
more target variable(s) adapted to the intelligent marketing model; and
[0035] the marketing pushing module is employed for employing the target
variable(s) and the
intelligent marketing model to obtain a target customer, and pushing a
commodity to the
target customer.
[0036] In comparison with prior-art technology, the method and system for
intelligent marketing
provided by the present invention achieve the following advantageous effects.
[0037] The method for intelligent marketing provided by the present invention
utilizes machine
learning to replace the traditional step of manually checking the trend chart,
automatically
screens out variables with better monotonicity or U-shape degree from great
many
variables, inputs the screened variables to an intelligent marketing model to
acquire target
customers, pushes commodities to the target customers, recognizes the type of
the
variable trend curve and the monotonicity and U-shape degrees consistently as
the manual,
visually direct judgment, can replace manual work, and effectively enhances
working
efficiency on the basis of guaranteeing recognition quality.
[0038] The system for intelligent marketing provided by the present invention
employs the
aforementioned method for intelligent marketing, can automatically screen out
variables,
acquire target customers according to the screened variables, push commodities
to the
target customers, and effectively enhance working efficiency and customers'
purchasing
success rate after marketing.
Date Recue/Date Received 2022-03-01
BRIEF DESCRIPTION OF THE DRAWINGS
[0039] The drawings described here are meant to provide further comprehension
to the present
invention, constitute a portion of the present invention, and exemplary
embodiments of
the present invention and the descriptions thereof are meant to explain the
present
invention, rather than to restrict the present invention. In the drawings:
[0040] Fig. 1 is a flowchart schematically illustrating the method for
intelligent marketing in the
embodiments of the present invention;
[0041] Fig. 2 is a flowchart schematically illustrating the essential
screening method for
intelligent marketing in the embodiments of the present invention;
[0042] Figs. 3(a) ¨ 3(b) are views respectively illustrating examples of
Condition 1 in which the
variable trend curve is recognized as a positive U-shape and Condition 2 in
which the
variable trend curve is recognized as an inverse U-shape in the embodiments of
the
present invention;
[0043] Figs. 4(a)¨ 4(0 are views respectively illustrating circumstances
possibly occurring when
the type of the variable trend curve is judged; and
[0044] Figs. 5(a) ¨ 5(1) are views respectively illustrating examples of
judging the type of the
variable trend curve.
DETAILED DESCRIPTION OF THE INVENTION
[0045] In order to make apparent and comprehensible the aforementioned
objectives, features
and advantages of the present invention, the technical solutions in the
embodiments of
the present invention will be more clearly and comprehensively described below
with
6
Date Recue/Date Received 2022-03-01
reference to the accompanying drawings in the embodiments of the present
invention.
Apparently, the embodiments as described are merely partial, rather than the
entire,
embodiments of the present invention. All other embodiments obtainable by
persons
ordinarily skilled in the art on the basis of the embodiments in the present
invention
without spending creative effort in the process shall all be covered by the
protection scope
of the present invention.
[0046] Embodiment 1
[0047] Please refer to Fig. 1, a method for intelligent marketing, comprising:
[0048] obtaining variables and drawing a variable trend curve, and obtaining
variables with
monotonicity features or U-shape features on the basis of the variable trend
curve and a
preset trend recognizing rule;
[0049] employing the variables with monotonicity features or U-shape features
to train an
intelligent marketing model, and simultaneously employing a stepwise
regression
operation to screen and obtain one or more target variable(s) adapted to the
intelligent
marketing model; and
[0050] employing the target variable(s) and the intelligent marketing model to
obtain a target
customer, and pushing a commodity to the target customer.
[0051] The method for intelligent marketing provided by the present invention
utilizes machine
learning to replace the traditional step of manually checking the trend chart,
automatically
screens out variables with better monotonicity or U-shape degree from great
many
variables, inputs the screened variables to an intelligent marketing model to
acquire target
customers, pushes commodities to the target customers, enhances commodity
marketing
success rate, achieves automatic model training, and effectively enhances
working
efficiency.
[0052] Please refer to Fig. 2, the step of obtaining variables with
monotonicity features or U-
7
Date Recue/Date Received 2022-03-01
shape features on the basis of the variable trend curve and a preset trend
recognizing rule
in the embodiments of the present invention includes:
[0053] binning the variables to obtain a plurality of bins, and drawing a
variable trend curve of
positive sample rate (Target Rate) relevant variables on the basis of the
bins. The
variables are continuous variables, while discrete variables are not taken
into
consideration in the present invention.
[0054] Suppose there are altogether N bins on the x axis in a variable trend
curve of binned
continuous variables, the N bins are sequentially Pi. P2, ..., PN, positive
sample rates at
the N bins on the variable trend curve are respectively Target Ratei, Target
Rate2,
Target RateN, and the minimum positive sample rate (Target Rate.) and the
maximum
positive sample rate (Target _Rate) except for the positive sample rates
(Target Ratei
and Target RateN) at the left and right two endpoints are obtained.
[0055] The total variation (TV) of the variable trend curve is calculated, the
total variation (TV)
reflects the degree of fluctuation of the variable trend curve, and can be
obtained by
calculating a sum total of amplitudes between every two adjacent bins of the
variable
trend curve, namely TV = EiN_i I Target _ Rate i i ¨ Target_ Rate i I .
[0056] A first difference value (ADO of the variable trend curve is
calculated, the first difference
value means an absolute value of a difference between the positive sample
rates at the left
and right two endpoints on the variable trend curve, namely ADi = I Target_
RateN ¨
Target _Rated.
[0057] A second difference value (AD2) of the variable trend curve is
calculated, the second
difference value means a sum of absolute values of differences between
positive sample
rates at the left and right two endpoints on the variable trend curve
respectively with
respect to the minimum positive sample rate except for the positive sample
rates at the
left and right two endpoints, namely AD2 = I Target_ Ratemin ¨ Target_ Rate' I
+
8
Date Recue/Date Received 2022-03-01
I Target_ RateN ¨ Target_ Ratemin I.
[0058] A third difference value (AD3) of the variable trend curve is
calculated, the third
difference value means a sum of absolute values of differences between
positive sample
rates at the left and right two endpoints on the variable trend curve
respectively with
respect to the maximum positive sample rate except for the positive sample
rates at the
left and right two endpoints, namely AD3 = 'Target_ Ratemax ¨ Target_ Rate' I
+
I Target_ RateN ¨ Target_ Ratemax I .
[0059] A monotonicity degree index (M index) of the variable trend curve is
calculated, the
monotonicity degree index can be calculated by calculating a ratio of the
total variation
(TV) of the variable trend curve to the first difference value (ADO, namely M
index =
TV/ADi. If the variable trend curve is completely monotonous, then the ratio
TV/ADi =
1; if it is not monotonous, then the ratio TV/ADi > 1; the more the ratio
TV/ADi
approaches 1, the higher is the monotonicity degree of the variable trend
curve,
conversely, the lower is the monotonicity degree of the variable trend curve.
Accordingly,
the monotonicity degree index of the variable trend curve is valuated in the
range of
M_index E [1, +00), when M index = 1, this indicates that the variable trend
curve is
completely monotonous, and the higher the value of M index is, the lower is
the
monotonicity degree of the variable trend curve.
[0060] A U-shape degree index of the variable trend curve is calculated, the U-
shape degree
index can include a positive U-shape degree index (U index 1) and an inverse U-
shape
degree index (U index 2). The U-shape degree is essentially the same as the V-
shape
degree, each being an analysis model index that judges and predicts the
current status and
development trend of a certain event with a special development process.
[0061] When and only when Condition 1 (Target Rate.. < Target Ratei, and
Target Rate.. <
Target RateN) is established, the variable trend curve is possibly recognized
as a positive
9
Date Recue/Date Received 2022-03-01
U-shape, and hence its positive U-shape degree index (U index 1) can be
calculated by
calculating the ratio of the total variation (TV) of the variable trend curve
to the second
difference value (AD2), namely U index 1 = TV/AD2. If the variable trend curve
assumes a strict positive U-shape, then the ratio TV/AD2 = 1; if it is not a
strict positive
U-shape, then the ratio TV/AD2 > 1; the more the ratio TV/AD2 approaches 1,
the higher
is the positive U-shape degree of the variable trend curve, conversely, the
lower is the
positive U-shape degree of the variable trend curve. Accordingly, the positive
U-shape
degree index of the variable trend curve is valuated in the range of U_index_l
E [1, +00),
when U index 1 = 1, this indicates that the variable trend curve assumes a
strict positive
U-shape, and the higher the value of U index 1 is, the lower is the positive U-
shape
degree of the variable trend curve.
[0062] When and only when Condition 2 (Target Rate.. > Target Ratei, and
Target Rate.. >
Target RateN) is established, the variable trend curve is possibly recognized
as an inverse
U-shape, and hence its inverse U-shape degree index (U index 2) can be
calculated by
calculating the ratio of the total variation (TV) of the variable trend curve
to the third
difference value (AD3), namely U index 2 = TV/AD3. If the variable trend curve
assumes a strict inverse U-shape, then the ratio TV/AD3 = 1; if it is not a
strict inverse U-
shape, then the ratio TV/AD3 > 1; the more the ratio TV/AD3 approaches 1, the
higher is
the inverse U-shape degree of the variable trend curve, conversely, the lower
is the inverse
U-shape degree of the variable trend curve. Accordingly, the inverse U-shape
degree
index of the variable trend curve is valuated in the range of U_index_2 E [1,
+00), when
U index 2 = 1, this indicates that the variable trend curve assumes a strict
inverse U-
shape, and the higher the value of U index 2 is, the lower is the inverse U-
shape degree
of the variable trend curve.
[0063] When the U-shape degree index of the variable trend curve is
calculated, since the
variable trend curve is possibly recognized as both the positive U-shape and
the inverse
U-shape at the same time, the positive U-shape degree index and the inverse U-
shape
Date Recue/Date Received 2022-03-01
degree index of the variable trend curve must be both calculated in this case,
the lesser
one of the two indexes is selected to judge whether the variable trend curve
pertains to a
positive U-shape or an inverse U-shape, and the lesser one of the two indexes
is taken to
serve as the U-shape degree index U index of the variable trend curve.
[0064] Variables with monotonicity features and U-shape features are screened
out on the basis
of preset monotonicity index threshold, positive U-shape index threshold and
inverse U-
shape index threshold. In this embodiment, the monotonicity index threshold,
the positive
U-shape index threshold and the inverse U-shape index threshold are all
valuated in the
range of [1, 1.51. However, in the case the variable trend curve is excellent
in
monotonicity degree and has many variables, this threshold range can be
lessened,
conversely, the threshold range is enlarged.
[0065] Moreover, the monotonicity degree index M index and the U-shape degree
index
U index are utilized to judge the type of the variable trend curve, and such
judging
process involves positive sample rates (Target Ratei, Target_RateN) at the
left and right
endpoints and the extreme values (Target Rate.õ Target _Rate) except for the
positive
sample rates at the left and right endpoints.
[0066] When the positive sample rate at the left endpoint is greater than the
positive sample rate
at the right endpoint ( Target_Ratei
Target_RateN), there are the following six
circumstances concerning size relationships between the positive sample rates
at the left
and right endpoints and the extreme values except for the positive sample
rates at the left
and right endpoints:
[0067] Circumstance Al: Target_Ratemax E (-00, Target_RateN);
[0068] Circumstance A2: Target_Ratemax E [Target_RateN, Target_Ratei);
[0069] Circumstance A3: Target_Ratemax E [Target_Ratei, +00);
[0070] Circumstance B 1: Target_Ratemin E (-00, Target_RateN);
[0071] Circumstance B2: Target_Ratemin E [Target_RateN, Target_Ratei);
11
Date Recue/Date Received 2022-03-01
[0072] Circumstance B3: Target_Ratemin E [Target_Ratei, +00).
[0073] There are nine combinational circumstances for the above six
circumstances, namely:
[0074] Circumstance Al-Bl: Target_Ratemax E (-00, Target_RateN)
and
Target_Rate mmE (-00, Target_RateN);
[0075] Circumstance A1-B2: Target_Ratemax E (-00, Target_RateN)
and
Target_Ratemin E [Target_RateN, Target_Ratei);
[0076] Circumstance A1-B3: Target_Ratemax E (-00, Target_RateN)
and
Target_Ratemin E [Target_Ratei, +00);
[0077] Circumstance A2-B1: Target_Ratemax E [Target_RateN, Target_Ratei)
and
Target_Ratemin E (-00, Target_RateN);
[0078] Circumstance A2-B2: Target_Ratemax E [Target_RateN, Target_Ratei)
and
Target_Ratemin E [Target_RateN, Target_Ratei);
[0079] Circumstance A2-B3: Target_Ratemax E [Target_RateN, Target_Ratei)
and
Target_Ratemin E [Target_Ratei, +00);
[0080] Circumstance A3-B1: Target_Ratemax E [Target_Ratei, +00)
and
Target_Ratemin E (-00, Target_RateN);
[0081] Circumstance A3-B2: Target_Ratemax E [Target_Ratei, +00)
and
Target_Ratemin E [Target_RateN, Target_Ratei);
[0082] Circumstance A3-B3: Target_Ratemax E [Target_Ratei, +00)
and
Target_Ratemin E [Target_Ratei, +00).
[0083] In Circumstance Al-B1, since Target_Ratemax E (-00, Target_RateN),
Condition 2
of being recognized as an inverse U-shape is not satisfied, it is impossible
for the variable
trend curve to be an inverse U-shape, so it is not required to calculate the
inverse U-shape
index; the curve may be judged as a positive U-shape, so the positive U-shape
index
should be calculated; the variable trend curve may also be judged as monotony
decrease,
so the monotonicity index should also be calculated. After the positive U-
shape index
U index 1 and the monotonicity index M index have been calculated, the
magnitudes of
12
Date Recue/Date Received 2022-03-01
the two are compared, if U index 1 < M index, the variable trend curve is
judged as a
positive U-shape, otherwise it is judged as monotony decrease.
[0084] In Circumstance Al-B2, since Target_Ratemin > Target_Ratemax is
impossible to
occur, such circumstance is nonexistent.
[0085] In Circumstance Al-B3, since Target_Ratemin > Target_Ratemax is
impossible to
occur, such circumstance is nonexistent.
[0086] In Circumstance A2-B1, since Target_Ratemax E [Target_RateN,
Target_Ratei) ,
Condition 2 of being recognized as an inverse U-shape is not satisfied, it is
impossible
for the variable trend curve to be an inverse U-shape, so it is not required
to calculate the
inverse U-shape index; the variable trend curve may be judged as a positive U-
shape, so
the positive U-shape index should be calculated; the variable trend curve may
also be
judged as monotony decrease, so the monotonicity index should also be
calculated. After
the positive U-shape index U index 1 and the monotonicity index M index have
been
calculated, the magnitudes of the two are compared, if U index 1 <M index, the
variable
trend curve is judged as a positive U-shape, otherwise it is judged as
monotony decrease.
[0087] In Circumstance A2-B2, since Target_Ratemax E [Target_RateN,
Target_Ratei) ,
Condition 2 of being recognized as an inverse U-shape is not satisfied, it is
impossible
for the variable trend curve to be an inverse U-shape; since Target_Ratemin E
[Target_RateN, Target_Ratei, Condition 1 of being recognized as a positive U-
shape is
not satisfied, it is also impossible for the variable trend curve to be a
positive U-shape.
Therefore, this variable trend curve can only be judged as monotony decrease,
and it is
merely required to calculate the monotonicity index M index.
[0088] In Circumstance A2-B3, since Target_Ratemin > Target_Ratemax is
impossible to
occur, such circumstance is nonexistent.
13
Date Recue/Date Received 2022-03-01
[0089] In circumstance A3-B1, since the variable trend curve not only
satisfies Condition 1 of
being recognized as a positive U-shape but also satisfies Condition 2 of being
recognized
as an inverse U-shape, it may be recognized as a positive U-shape and may also
be
recognized as an inverse U-shape. Therefore, it is required not only to
calculate the
positive U-shape index U index 1 but also to calculate the inverse U-shape
index
U index 2. The variable trend curve may also be recognized as monotony
decrease, so
the monotonicity index M index should also be calculated. After the three
indexes have
been calculated, the magnitudes of the three are compared, the least one is
selected
therefrom, and the variable trend curve is judged to be of the corresponding
type.
[0090] In circumstance A3-B2, since the variable trend curve does not satisfy
Condition 1 of
being recognized as a positive U-shape but satisfies Condition 2 of being
recognized as
an inverse U-shape, so after the inverse U-shape index U index 2 has been
calculated,
the monotonicity index M index is further calculated, the magnitudes of the
two are
compared, the lesser one is selected therefrom, and the variable trend curve
is judged to
be of the corresponding type.
[0091] In circumstance A3-B3, since the variable trend curve does not satisfy
Condition 1 of
being recognized as a positive U-shape but satisfies Condition 2 of being
recognized as
an inverse U-shape, so after the inverse U-shape index U index 2 has been
calculated,
the monotonicity index M index is further calculated, the magnitudes of the
two are
compared, the lesser one is selected therefrom, and the variable trend curve
is judged to
be of the corresponding type.
[0092] When the positive sample rate at the left endpoint is smaller than the
positive sample rate
at the right endpoint (Target_Ratei < Target_RateN), the circumstance is
similar to the
aforementioned circumstance in which the positive sample rate at the left
endpoint is
greater than the positive sample rate at the right endpoint, and it suffices
to change
14
Date Recue/Date Received 2022-03-01
monotony decrease to monotony increase.
[0093] Moreover, the variable monotonicity degrees and the U-shape degrees are
sorted by
means of the monotonicity degree index M index and the U-shape degree index U
index.
Specifically, the monotonous variables are classified as one type (including
monotony
increase and monotony decrease), the U-shaped variables are classified as one
type
(including positive U-shape and inverse U-shape), these are sorted according
to an
ascending order of the corresponding indexes, so that the variables are sorted
in a
decreasing order of monotonicity degrees and in a decreasing order of U-shape
degrees.
[0094] The step of employing the variables with monotonicity features or U-
shape features to
train an intelligent marketing model, and simultaneously employing a stepwise
regression
operation to screen and obtain one or more target variable(s) adapted to the
intelligent
marketing model includes:
[0095] data-preprocessing the variables with monotonicity features or U-shape
features, wherein
the data-preprocessing includes filling missing values, processing abnormal
values, and
one-hot coding with respect to categorical variables;
[0096] the step of screening important variables out of the preprocessed
variables includes:
calculating IV values and PSI values of the variables respectively, and hence
screening
out important variables whose IV values are greater than an IV threshold and
whose PSI
values are smaller than a PSI threshold;
[0097] inputting the important variables in the intelligent marketing model,
and simultaneously
employing a stepwise regression operation to screen and obtain one or more
target
variable(s) adapted to the intelligent marketing model.
[0098] The method provided by the present invention is applicable to general
intelligent
marketing response degree models, a marketing model for attracting new
customers with
respect to cash loan products is taken for example, the model includes
variables of such
dimensions as user basic information, browsing behaviors and purchasing
behaviors, etc.
Date Recue/Date Received 2022-03-01
Since the ultimate objective of attracting new customers with respect to cash
loan
products is to grant credit, whether the credit is granted is taken here as
the judging
criterion of y label positive and negative samples. Empirically, once over 90%
of the users
apply for the line of credit and pass risk control card A within seven days
after marketing,
it can then be judged whether the marketing for attracting new customers is
succeeded,
so the performance period is selected as seven days here. If credit is granted
to a user
within seven days after marketing, the user is marked as a positive sample,
otherwise the
user is marked as a negative sample. The observation point of the training set
is a certain
marketing day, and the observation point of the testing set is a certain
marketing day after
the observation point of the training set.
[0099] The technical personnel conducted an AB test, i.e., compared the result
of the screening
method in the present invention with the result of the manually screening
method. In the
traditional, manually screening process, a total of 498 variables was removed
from 1061
variables screened out of the previous step, 563 variables were retained, it
took 55
minutes to complete the process, in which 14 variables were put in the model
after the
screened variables had been subjected to another round of screening through
stepwise
regression, the AUC (area under curve) on the training set was 0.91, and the
AUC on the
testing set was 0.9. In the process of automatic screening of monotonicity
degrees
according to the present invention, a total of 516 variables was removed from
1061
variables screened out of the previous step, 545 variables were retained, it
took 3 minutes
to complete the process, in which 15 variables were put in the model after the
screened
variables had been subjected to another round of screening through stepwise
regression,
the AUC on the training set was 0.92, and the AUC on the testing set was 0.91.
The
present invention enhances working efficiency, and the model effect is
slightly enhanced
as compared with that of the manual screening; moreover, manual screening
relies on the
judgment of human beings, so the recognition may not be precise, whereas the
current
automatic recognition algorithm makes use of monotonicity indexes to judge
monotonicity degrees and U-shape degrees, so the recognition result is made
more precise.
16
Date Recue/Date Received 2022-03-01
[0100] Embodiment 2
[0101] There is provided a system for intelligent marketing, which system
comprises: a first
variable screening module, a second variable screening module, and a marketing
pushing
module, of which the first variable screening module is employed for obtaining
variables
and drawing a variable trend curve, and obtaining variables with monotonicity
features
or U-shape features on the basis of the variable trend curve and a preset
trend recognizing
rule; the second variable screening module is employed for employing the
variables with
monotonicity features or U-shape features to train an intelligent marketing
model, and
simultaneously employing a stepwise regression operation to screen and obtain
one or
more target variable(s) adapted to the intelligent marketing model; and the
marketing
pushing module is employed for employing the target variable(s) and the
intelligent
marketing model to obtain a target customer, and pushing a commodity to the
target
customer.
[0102] The system for intelligent marketing provided by the present invention
employs the
method for intelligent marketing in the aforementioned Embodiment 1 to utilize
machine
learning to replace the traditional step of manually checking the trend chart,
automatically
screen out variables with better monotonicity or U-shape degree from great
many
variables, input the screened variables to an intelligent marketing model to
acquire target
customers, push commodities to the target customers, and effectively enhance
working
efficiency and customer purchasing success rate after marketing. As compared
with prior-
art technology, the system for intelligent marketing provided by this
embodiment of the
present invention achieves the same advantageous effects as achieved by the
method for
intelligent marketing provided by the aforementioned Embodiment 1, and the
other
technical features in the system for intelligent marketing are identical with
the features
disclosed by the method of the aforementioned Embodiment 1, so no repetition
is
redundantly made in this context.
17
Date Recue/Date Received 2022-03-01
[0103] The specific features, structures, materials or features described in
the above
embodiments are combinational in any suitable form with any one or more
embodiment(s)
or example(s).
[0104] What is described above is merely directed to specific embodiments of
the present
invention, but the protection scope of the present invention is not restricted
thereby. Any
variation or replacement easily conceivable to persons skilled in the art
within the
technical range disclosed by the present invention shall be covered within the
protection
scope of the present invention. Accordingly, the protection scope of the
present invention
shall be based on the Claims.
18
Date Recue/Date Received 2022-03-01
