CN109657832A - A kind of prediction technique and device of frequent customer - Google Patents
A kind of prediction technique and device of frequent customer Download PDFInfo
- Publication number
- CN109657832A CN109657832A CN201810421976.4A CN201810421976A CN109657832A CN 109657832 A CN109657832 A CN 109657832A CN 201810421976 A CN201810421976 A CN 201810421976A CN 109657832 A CN109657832 A CN 109657832A
- Authority
- CN
- China
- Prior art keywords
- user
- brand
- prediction
- collection
- prediction model
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
- G06Q30/0201—Market modelling; Market analysis; Collecting market data
- G06Q30/0202—Market predictions or forecasting for commercial activities
Landscapes
- Business, Economics & Management (AREA)
- Engineering & Computer Science (AREA)
- Strategic Management (AREA)
- Development Economics (AREA)
- Economics (AREA)
- Finance (AREA)
- Accounting & Taxation (AREA)
- Entrepreneurship & Innovation (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Marketing (AREA)
- Human Resources & Organizations (AREA)
- Game Theory and Decision Science (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Tourism & Hospitality (AREA)
- Data Mining & Analysis (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The application provides the prediction technique and device of a kind of frequent customer, which comprises load prediction model, wherein the prediction model carries the model parameter value that training obtains；The first user identifier and the user to be predicted for obtaining user to be predicted patronized the number one brand mark of primary brand；According to first user identifier and number one brand mark search the corresponding user characteristics collection of first user identifier from database, the number one brand identifies corresponding brand identity collection and first user identifier linked character collection corresponding with number one brand mark, and saves as fisrt feature set；The fisrt feature set is input to the prediction model, and the prediction result whether user to be predicted that the prediction model is calculated according to the model parameter value can become the frequent customer of the brand is obtained, so that businessman be helped to judge the potential frequent customer in user group.
Description
Technical field
This application involves technical field of data processing, in particular to the prediction technique and device, calculating of a kind of frequent customer is set
Standby and computer readable storage medium.
Background technique
As user's covering surface of (movement) internet is continuously increased, user can touch more food and drink product than before
Board and businessman, food and drink businessman can also cover more clients in wider array of geographic range, this is all with user --- and it is flat
Based on the connection of platform --- businessman.
As the businessman that the user volume of platform increases and move in platform increases, platform has the ability also do a one more
Good " intermediary " and " resource discovering person " helps user to be quickly found out the businessman for more meeting oneself hobby in magnanimity businessman.Together
When, in the case where multiple businessmans compete the meeting of platform display machine, the displaying limited opportunities that each household businessman obtains, platform can also be helped
Businessman makes full use of these chances, gives most probable approved user oneself brand promotion.
Businessman, which is contacted, with a kind of major issue in user is: coming in the primary user of certain brand, which user may
Is the chance come again big? does is which user less likely to come again? how to judge the potential frequent customer in user group, is interested in businessman
's.
Summary of the invention
In view of this, the embodiment of the present application provides the prediction technique and device, calculating equipment and calculating of a kind of frequent customer
Machine readable storage medium storing program for executing, to solve technological deficiency existing in the prior art.
The embodiment of the present application discloses the prediction technique of frequent customer a kind of, which comprises
Load prediction model, wherein the prediction model carries the model parameter value that training obtains；
The first user identifier and the user to be predicted for obtaining user to be predicted patronized the first of primary brand
Brand identity；
First user identifier is searched from database according to first user identifier and number one brand mark
Corresponding user characteristics collection, the number one brand identify corresponding brand identity collection and first user identifier and described the
The corresponding linked character collection of one brand identity, and save as fisrt feature set；
The fisrt feature set is input to the prediction model, and obtains the prediction model and is joined according to the model
Whether the user to be predicted that numerical value is calculated can become the prediction result of the frequent customer of the brand.
In the schematical embodiment of the application, the model parameter value of the prediction model passes through following
Method obtains:
The second user mark and target product of the frequent customer of target brand are screened from the historical data that database prestores
The third user identifier of the non-frequent customer of board；
It is identified according to the second user, the second user is searched from database and identifies corresponding user characteristics collection；
According to the third user identifier, the corresponding user characteristics collection of the third user identifier is searched from database；
According to the second brand identity of the target brand, the corresponding product of second brand identity are searched from database
Board feature set；
According to second user mark and second brand identity, the second user mark is searched from database
Linked character collection corresponding with second brand identity；
According to the third user identifier and second brand identity, the third user identifier is searched from database
Linked character collection corresponding with second brand identity；
By the second user identify corresponding user characteristics collection, the corresponding user characteristics collection of the third user identifier,
The corresponding brand identity collection of second brand identity, the second user identify association corresponding with second brand identity
Feature set and third user identifier linked character collection corresponding with second brand identity save as second feature collection
It closes；
The second feature set is input to prediction model to be trained, obtains and save the model of the prediction model
Parameter value.
In the schematical embodiment of the application, the model parameter value is calculated by the following formula:
Wherein,The biography of j-th of node for l layers in the prediction model (W, b) of i-th of node to l+1 layers
Defeated weight；
For the output biasing of l layers of i-th of node in the prediction model (W, b)；
Alpha is learning rate, controls the renewal speed of every wheel study model parameter value；
S is second feature set；
L (W, b；It S) is the loss function of the prediction model (W, b) on the second feature set S.
In the schematical embodiment of the application, it is being calculated whether the user to be predicted can become institute
After the prediction result for stating the frequent customer of brand, further includes:
By the prediction result and the user to be predicted whether can become the actual result of the frequent customer of the brand into
Row comparison, obtains predictablity rate.
In the schematical embodiment of the application, the user characteristics collection includes: customer attribute information, user
Have dinner record, user have dinner geography information, user have dinner brand message, user have dinner shops's information, user's time for eating meals information,
One or more in user's taste information；
The brand identity collection includes: in brand evaluation information, brand ranking information, brand taste and brand temporal information
One or more；
The linked character collection includes: first time dining information of the user in brand.
In the schematical embodiment of the application, the prediction model includes: input layer, output layer and position
At least one layer of hidden layer between the input layer and the output layer；
The input layer includes at least one node；
The hidden layer includes at least one node；
The output layer includes a node；
Every layer of each node is connected with each node of adjacent layer respectively.
This application discloses the prediction meanss of frequent customer a kind of, described device includes:
Prediction model loading module, for loading prediction model, wherein the prediction model carries the mould that training obtains
Shape parameter value；
Identifier acquisition module, the first user identifier and the user to be predicted for obtaining user to be predicted were patronized
The number one brand of primary brand identifies；
Fisrt feature set generation module, for being identified according to first user identifier and the number one brand from data
The corresponding user characteristics collection of first user identifier is searched in library, the number one brand identifies corresponding brand identity collection and
First user identifier linked character collection corresponding with number one brand mark, and save as fisrt feature set；
Prediction result generation module, for the fisrt feature set to be input to the prediction model, and described in acquisition
Whether the user to be predicted that prediction model is calculated according to the model parameter value can become the frequent customer of the brand
Prediction result.
In the schematical embodiment of the application, described device further include:
Predictablity rate generation module, for whether the prediction result and the user to be predicted can be become the product
The actual result of the frequent customer of board compares, and obtains predictablity rate.
The embodiment of the present application discloses a kind of calculating equipment, including memory, processor and storage are on a memory and can
The computer instruction run on a processor, the processor perform the steps of when executing described instruction
Load prediction model, wherein the prediction model carries the model parameter value that training obtains；
The first user identifier and the user to be predicted for obtaining user to be predicted patronized the first of primary brand
Brand identity；
First user identifier is searched from database according to first user identifier and number one brand mark
Corresponding user characteristics collection, the number one brand identify corresponding brand identity collection and first user identifier and described the
The corresponding linked character collection of one brand identity, and save as fisrt feature set；
The fisrt feature set is input to the prediction model, and obtains the prediction model and is joined according to the model
Whether the user to be predicted that numerical value is calculated can become the prediction result of the frequent customer of the brand.
The embodiment of the present application discloses a kind of computer readable storage medium, is stored with computer instruction, and feature exists
In the step of instruction realizes method as described above when being executed by processor.
The prediction technique and device of frequent customer provided by the present application calculate equipment and computer readable storage medium, pass through
Prediction model is loaded, patronized the first of primary brand according to the first user identifier of user to be predicted and user to be predicted
Brand identity is searched from database and obtains fisrt feature set, and fisrt feature set is then input to prediction model and is returned
The prediction result of head visitor, so that businessman be helped to judge the potential frequent customer in user group.
Detailed description of the invention
Fig. 1 is the flow diagram of the prediction technique of the frequent customer of the embodiment of the present application；
Fig. 2 is the schematic diagram of prediction model in the prediction technique of the frequent customer of the embodiment of the present application；
Fig. 3 is the acquisition methods flow chart of the model parameter value of the prediction model of the embodiment of the present application；
Fig. 4 is the structural schematic diagram of the prediction meanss of the frequent customer of the embodiment of the present application；
Fig. 5 is the structural schematic diagram of the calculating equipment of the embodiment of the present application.
Specific embodiment
Many details are explained in the following description in order to fully understand the application.But the application can be with
Much it is different from other way described herein to implement, those skilled in the art can be without prejudice to the application intension the case where
Under do similar popularization, therefore the application is not limited by following public specific implementation.
The prediction technique and device, calculating equipment and computer-readable storage medium of a kind of frequent customer are provided in this application
Matter is described in detail one by one in the following embodiments.
The prediction technique of frequent customer disclosed in the present embodiment is applied in a platform referring to Fig. 1, this method, which makees
For contact user and businessman intermediary, can recorde and update businessman brand identity, record user characteristics and record user with
The linked character of brand, and the feature that will acquire is stored in database.
The prediction technique of frequent customer disclosed in the present embodiment includes the following steps 101~104:
101, prediction model is loaded, wherein the prediction model carries the model parameter value that training obtains.
It should be noted that prediction model has just been trained to and has been stored in database before loading prediction model
In.In the present embodiment, prediction model is a deep learning neural network.Prediction model is described in detail below.
Specifically, the structure of the prediction model in the present embodiment is as shown in Fig. 2, prediction model includes: input layer, output layer
And at least one layer of hidden layer between the input layer and the output layer；
The input layer includes at least one node；
The hidden layer includes at least one node；
The output layer includes a node；
Every layer of each node is connected with each node of adjacent layer respectively.
Referring to fig. 2, the prediction model in the present embodiment, input layer are 152 nodes, and each node corresponds to second feature collection
The feature closed；Hidden layer is three layers, and the node of every layer of hidden layer is 64, and each node input is upper one layer of each node
Weighted sum, activation primitive are nonlinear function (relu function)；Output layer is 1 node, and output activation primitive is sigmoid letter
Number.
Prediction model needs to carry out Primary Construction before training.The process of Primary Construction mainly selects following hyper parameter
(hyperparameters): the structure of model, the number of plies of model, every layer of number of nodes, every layer of activation primitive, every layer just
Then change parameter, the activation primitive of output layer, the optimization algorithm in training process, the target loss functional form of optimization algorithm, instruction
Experienced wheel number, using one when stochastic gradient optimization algorithm batch (batch) size.
After prediction model Primary Construction, it is necessary to which input data is trained, to determine model parameter.
Hyper parameter and model parameter difference, model parameter is that prediction model can learn automatically according to input data
Variable；Hyper parameter is empirically determined variable, and for determining some parameters of model, hyper parameter is different, and model is not
With.
Referring to Fig. 3, the model parameter value of prediction model is obtained by following steps 301~308:
301, the second user mark and mesh of the frequent customer of target brand are screened from the historical data that database prestores
Mark the third user identifier of the non-frequent customer of brand.
It is to be understood that the frequent customer of target brand described in the present embodiment refer to the target brand to shop number
User more than or equal to 2, the non-frequent customer of target brand refer to the target brand to shop number be 1 and it is last to shop when
Spacing the present is more than the user of a measurement period.Wherein, measurement period can be 365 days, and the present embodiment does not limit statistics
The specific time in period.
By this step 301, the positive sample and negative sample of the available target brand, wherein positive sample includes: second
User identifier, target brand identity and positive sample label, negative sample include: third user identifier, target brand identity and negative sample
This label.
302, it is identified according to the second user, the second user is searched from database and identifies corresponding user characteristics
Collection.
303, according to the third user identifier, the corresponding user characteristics of the third user identifier are searched from database
Collection.
304, according to the second brand identity of the target brand, it is corresponding that second brand identity is searched from database
Brand identity collection.
305, according to second user mark and second brand identity, the second user is searched from database
Identify linked character collection corresponding with second brand identity.
306, according to the third user identifier and second brand identity, the third user is searched from database
Identify linked character collection corresponding with second brand identity.
It should be noted that the execution sequence of above-mentioned steps 302~306 can be any, it can be according to from step 302 to step
Rapid 306 sequence successively executes, and sequence, such as the sequence executed side by side can also be executed according to other.
307, the second user is identified into corresponding user characteristics collection, the corresponding user characteristics of the third user identifier
Collection, the corresponding brand identity collection of second brand identity, the second user mark are corresponding with second brand identity
Linked character collection and third user identifier linked character collection corresponding with second brand identity save as the second spy
Collection is closed.
Specifically, in this step 307 the following steps are included:
1) positive negative sample is merged, the table of positive negative sample mixing is formed, containing " the second brand identity of user identifier-- is just
Three dimensions of negative label ", are denoted as DATA table；
2) according to user identifier as connecting key, will with connect in user identifier in DATA table matched " user characteristics collection "
(inner join) is into DATA table；
It 3), will be with the matched " brand identity of the second brand identity in DATA table according to the second brand identity as connecting key
Connection (inner join) is into DATA table in collection "；
4) joint connecting key is used as according to (user identifier, the second brand identity), it will be with (user identifier, the in DATA table
Two brand identities) in matched " user+brand linked character collection " connection (inner join) into DATA table；
5) user identifier and the second brand identity are deleted, forms " user characteristics collection-brand identity collection-user+brand
Linked character collection-positive and negative sample label " data, be stored in DATA table.
Wherein, user characteristics collection includes: that have dinner record, user of customer attribute information, user has dinner geography information, user just
Meal brand message, user has dinner shops's information, user's time for eating meals information, one or more in user's taste information；
Specifically:
Customer attribute information includes: user's gender, age of user etc.；
User's record of having dinner includes: for the first time using platform time for eating meals, last time platform time for eating meals etc.；
User have dinner geography information include: how many a cities have dinner record, which city have dinner number at most etc.；
User's brand message of having dinner includes: maximum/minimum shops number for the brand once having dinner etc.；
User's shops's information of having dinner includes: highest/minimum average price for the shops having dinner etc.；
User's time for eating meals information includes: to have dinner in the four seasons to record shared individual and have dinner the ratio of record, on weekdays/weekend
Proportion etc.；
User's taste information includes: the style of cooking number distribution etc. having dinner.
Brand identity collection includes: one in brand evaluation information, brand ranking information, brand taste and brand temporal information
It is a or multiple；
Specifically:
Brand evaluation information includes: each shops's average price mean value, taste marking mean value etc.；
Brand ranking information includes: brand scale in national ranking etc.；
Brand taste includes: brand style of cooking etc.；
Brand temporal information include: brand queuing amount respectively four seasons proportion, brand queuing amount on weekdays/weekend
Proportion etc..
Linked character collection includes: first time dining information of the user in brand；
Specifically, user the first time dining information of brand include: type of service (queuing/reservation/is ordered/paid),
Season, the working day/weekend having dinner have dinner, time for eating meals point, average price of having dinner etc..
Three above feature set after the completion of calculating, respectively be stored in database in, and with the development of business, platform every
A period of time updates primary.
308, the second feature set is input to prediction model to be trained, obtain and saves the prediction model
Model parameter value.
Specifically, in this step 308, by second feature set-partition at training dataset (train Data), verifying
Data set (validation Data), test data set (test Data) guarantee three data using the modes such as randomly selecting
The positive and negative sample distribution concentrated is uniform.
It, can be according to the accuracy rate and loss function after each training on verifying collection during training prediction model
Trained model structure and model parameter value are finally stored in database by performance optimization hyper parameter.
In the specific embodiment of the application, for existing training dataset S={ (x1, y1), (x2, y2)…
(xm, ym), include m sample, constructs loss function L (W, b on training dataset；S), learn for measuring current depth
Predictablity rate of the network on training dataset.
By the training to network model, predictablity rate of the prediction model on training dataset is improved, is protected simultaneously
Certain generalization ability is held also to keep higher accuracy rate on following data set.
Model parameter value is calculated by the following formula:
Wherein,The biography of j-th of node for l layers in the prediction model (W, b) of i-th of node to l+1 layers
Defeated weight；
For the output biasing of l layers of i-th of node in the prediction model (W, b)；
Alpha is learning rate, controls the renewal speed of every wheel study model parameter value；
S is second feature set；
L (W, b；It S) is the loss function of the prediction model (W, b) on the second feature set S.
102, it obtains the first user identifier of user to be predicted and the user to be predicted patronized primary brand
Number one brand mark.
It should be noted that user to be predicted is for a certain target brand, neither frequent customer is also not non-frequent customer
User, that is to say, that user to be predicted be a nearest measurement period in only patronized the primary user of the brand.Wherein,
The measurement period can be 365 days, the application to specific time of the measurement period without limitation.
Since user to be predicted had consumer record on platform, so its first user identifier and user's light to be predicted
The number one brand mark for caring for primary brand is stored in database.
103, first user is searched from database according to first user identifier and number one brand mark
Identify corresponding user characteristics collection, the number one brand identifies corresponding brand identity collection and first user identifier and institute
It states number one brand and identifies corresponding linked character collection, and save as fisrt feature set.
Wherein, the particular content of user characteristics collection, brand identity collection and linked character collection has solved in detail in foregoing teachings
It released, just repeated no more herein.Three above feature set is stored in table different in database respectively after the completion of calculating, and
With the development of business, platform updates once at regular intervals.
104, the fisrt feature set is input to the prediction model, and obtains the prediction model according to the mould
Whether the user to be predicted that shape parameter value is calculated can become the prediction result of the frequent customer of the brand.
Optionally, after the completion of step 104,10-Fold cross validation is done, calculates the Average Accuracy of 10 predictions, is made
For the accuracy rate of prediction model.
The prediction technique of frequent customer provided by the present application is used by loading prediction model according to the first of user to be predicted
The number one brand that family mark and user to be predicted patronized primary brand, which identifies to search from database, obtains fisrt feature
Set, is then input to prediction model for fisrt feature set and obtains the prediction result of frequent customer, so that businessman be helped to judge to use
Potential frequent customer in the group of family.
Optionally, in another embodiment of the application, after step 104, the prediction technique of the frequent customer of the application
Further include:
105, whether the prediction result and the user to be predicted can be become to the practical knot of the frequent customer of the brand
Fruit compares, and obtains predictablity rate.
When calculating, it can be calculated by predictablity rate=correctly predicted number/test data sum, it is accurate by predicting
Rate, it can be determined that the superiority and inferiority degree that prediction model is established.
The embodiment of the present application also discloses the prediction meanss of frequent customer a kind of, and referring to fig. 4, described device includes:
Prediction model loading module 401, for loading prediction model, wherein the prediction model carries training and obtains
Model parameter value；
Identifier acquisition module 402, for obtain user to be predicted the first user identifier and user's light to be predicted
Cared for the number one brand mark of primary brand；
Fisrt feature set generation module 403, for according to first user identifier and the number one brand mark from
The corresponding user characteristics collection of first user identifier is searched in database, the number one brand identifies corresponding brand identity collection
And first user identifier linked character collection corresponding with number one brand mark, and save as fisrt feature set；
Prediction result generation module 404 for the fisrt feature set to be input to the prediction model, and obtains institute
State whether the user to be predicted that prediction model is calculated according to the model parameter value can become turning one's head for the brand
The prediction result of visitor.
Wherein, about the framework of prediction model, the acquisition methods of the model parameter value of prediction model, user characteristics collection, product
The preceding method part of the meaning of board feature set and linked character collection, the present embodiment has explained in detail, and just repeats no more herein.
Optionally, model parameter value is calculated by the following formula:
Wherein,The biography of j-th of node for l layers in the prediction model (W, b) of i-th of node to l+1 layers
Defeated weight；
For the output biasing of l layers of i-th of node in the prediction model (W, b)；
Alpha is learning rate, controls the renewal speed of every wheel study model parameter value；
S is second feature set；
L (W, b；It S) is the loss function of the prediction model (W, b) on the second feature set S.
Optionally, the prediction meanss of the frequent customer of the present embodiment further include:
Predictablity rate generation module, for whether the prediction result and the user to be predicted can be become the product
The actual result of the frequent customer of board compares, and obtains predictablity rate.
The prediction meanss of the frequent customer of the application are by load prediction model, according to the first user identifier of user to be predicted
And user to be predicted patronized primary brand number one brand identify from database search obtain fisrt feature set, so
Fisrt feature set is input to prediction model afterwards and obtains the prediction result of frequent customer, so that businessman be helped to judge in user group
Potential frequent customer.
A kind of exemplary scheme of the prediction meanss of above-mentioned frequent customer for the present embodiment.It should be noted that should be later
The technical solution of the technical solution of the prediction meanss of visitor and the prediction technique of above-mentioned frequent customer belongs to same design, frequent customer's
The detail content that the technical solution of prediction meanss is not described in detail may refer to the technical side of the prediction technique of above-mentioned frequent customer
The description of case.
Fig. 5 is to show the structural block diagram of the calculating equipment 500 according to one embodiment of the application.The calculating equipment 500
Component includes but is not limited to memory 510 and processor 520.Processor 520 is connected with memory 510.
Although being not shown in Fig. 5, it will be appreciated that calculating equipment 500 can also include network interface, network interface
Enable and calculates equipment 500 via one or more network communications.The example of these networks includes local area network (LAN), wide area
The combination of the communication network of net (WAN), personal area network (PAN) or such as internet.Network interface may include wired or wireless
One or more of any kind of network interface (for example, network interface card (NIC)), such as IEEE802.11 wireless local area
Net (WLAN) wireless interface, worldwide interoperability for microwave accesses (Wi-MAX) interface, Ethernet interface, universal serial bus (USB) connect
Mouth, cellular network interface, blue tooth interface, near-field communication (NFC) interface, etc..
In one embodiment of the application, unshowned other component can also in above-mentioned and Fig. 5 of calculating equipment 500
To be connected to each other, such as pass through bus.It should be appreciated that calculating device structure block diagram shown in fig. 5 is merely for the sake of exemplary
Purpose, rather than the limitation to the application range.Those skilled in the art can according to need, and increase or replace other component.
Calculating equipment 500 can be any kind of static or mobile computing device, including mobile computer or mobile meter
Calculate equipment (for example, tablet computer, personal digital assistant, laptop computer, notebook computer, net book etc.), movement
Phone (for example, smart phone), wearable calculating equipment (for example, smartwatch, intelligent glasses etc.) or other kinds of shifting
Dynamic equipment, or the static calculating equipment of such as desktop computer or PC.Calculating equipment 500 can also be mobile or state type
Server.
The processor 520 of the calculating equipment 500 of the present embodiment performs the steps of when executing described instruction
Load prediction model, wherein the prediction model carries the model parameter value that training obtains；
The first user identifier and the user to be predicted for obtaining user to be predicted patronized the first of primary brand
Brand identity；
First user identifier is searched from database according to first user identifier and number one brand mark
Corresponding user characteristics collection, the number one brand identify corresponding brand identity collection and first user identifier and described the
The corresponding linked character collection of one brand identity, and save as fisrt feature set；
The fisrt feature set is input to the prediction model, and obtains the prediction model and is joined according to the model
Whether the user to be predicted that numerical value is calculated can become the prediction result of the frequent customer of the brand.
One embodiment of the application also provides a kind of computer readable storage medium, is stored with computer instruction, the instruction
The step of prediction technique of frequent customer as previously described is realized when being executed by processor.
A kind of exemplary scheme of above-mentioned computer readable storage medium for the present embodiment.It should be noted that this is deposited
The technical solution of the technical solution of storage media and the prediction technique of above-mentioned frequent customer belongs to same design, the technology of storage medium
The detail content that scheme is not described in detail may refer to the description of the technical solution of the prediction technique of above-mentioned frequent customer.
The computer instruction includes computer program code, the computer program code can for source code form,
Object identification code form, executable file or certain intermediate forms etc..The computer-readable medium may include: that can carry institute
State any entity or device, recording medium, USB flash disk, mobile hard disk, magnetic disk, CD, the computer storage of computer program code
Device, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory),
Electric carrier signal, telecommunication signal and software distribution medium etc..It should be noted that the computer-readable medium include it is interior
Increase and decrease appropriate can be carried out according to the requirement made laws in jurisdiction with patent practice by holding, such as in certain jurisdictions of courts
Area does not include electric carrier signal and telecommunication signal according to legislation and patent practice, computer-readable medium.
It should be noted that for the various method embodiments described above, describing for simplicity, therefore, it is stated as a series of
Combination of actions, but those skilled in the art should understand that, the application is not limited by the described action sequence because
According to the application, certain steps can use other sequences or carry out simultaneously.Secondly, those skilled in the art should also know
It knows, the embodiments described in the specification are all preferred embodiments, and related actions and modules might not all be this Shen
It please be necessary.
In the above-described embodiments, it all emphasizes particularly on different fields to the description of each embodiment, there is no the portion being described in detail in some embodiment
Point, it may refer to the associated description of other embodiments.
The application preferred embodiment disclosed above is only intended to help to illustrate the application.There is no detailed for alternative embodiment
All details are described, are not limited the invention to the specific embodiments described.Obviously, according to the content of this specification,
It can make many modifications and variations.These embodiments are chosen and specifically described to this specification, is in order to preferably explain the application
Principle and practical application, so that skilled artisan be enable to better understand and utilize the application.The application is only
It is limited by claims and its full scope and equivalent.
Claims (10)
1. a kind of prediction technique of frequent customer, which is characterized in that the described method includes:
Load prediction model, wherein the prediction model carries the model parameter value that training obtains；
The first user identifier and the user to be predicted that obtain user to be predicted patronized the number one brand of primary brand
Mark；
It is corresponding that first user identifier is searched from database according to first user identifier and number one brand mark
User characteristics collection, the number one brand identify corresponding brand identity collection and first user identifier and first product
Board identifies corresponding linked character collection, and saves as fisrt feature set；
The fisrt feature set is input to the prediction model, and obtains the prediction model according to the model parameter value
Whether the user to be predicted being calculated can become the prediction result of the frequent customer of the brand.
2. the prediction technique of frequent customer according to claim 1, which is characterized in that the model of the prediction model is joined
Numerical value is prepared by the following:
The second user mark and target brand of the frequent customer of screening target brand from the historical data that database prestores
The third user identifier of non-frequent customer；
It is identified according to the second user, the second user is searched from database and identifies corresponding user characteristics collection；
According to the third user identifier, the corresponding user characteristics collection of the third user identifier is searched from database；
According to the second brand identity of the target brand, it is special that the corresponding brand of second brand identity is searched from database
Collection；
According to second user mark and second brand identity, the second user mark and institute are searched from database
State the corresponding linked character collection of the second brand identity；
According to the third user identifier and second brand identity, the third user identifier and institute are searched from database
State the corresponding linked character collection of the second brand identity；
The second user is identified into corresponding user characteristics collection, the corresponding user characteristics collection of the third user identifier, described
The corresponding brand identity collection of second brand identity, the second user identify linked character corresponding with second brand identity
Collection and third user identifier linked character collection corresponding with second brand identity save as second feature set；
The second feature set is input to prediction model to be trained, obtains and save the model parameter of the prediction model
Value.
3. the prediction technique of frequent customer according to claim 2, which is characterized in that the model parameter value passes through following public affairs
Formula calculates:
Wherein,The transmission right of j-th of node for l layers in the prediction model (W, b) of i-th of node to l+1 layers
Weight；
For the output biasing of l layers of i-th of node in the prediction model (W, b)；
Alpha is learning rate, controls the renewal speed of every wheel study model parameter value；
S is second feature set；
L (W, b；It S) is the loss function of the prediction model (W, b) on the second feature set S.
4. the prediction technique of frequent customer according to claim 1, which is characterized in that the user to be predicted is being calculated
After the prediction result for the frequent customer that whether can become the brand, further includes:
The actual result for the frequent customer whether prediction result and the user to be predicted can become the brand be carried out pair
Than obtaining predictablity rate.
5. the prediction technique of frequent customer according to claim 1, which is characterized in that
The user characteristics collection includes: that have dinner record, user of customer attribute information, user has dinner geography information, user with regard to food product
Board information, user are had dinner shops's information, user's time for eating meals information, one or more in user's taste information；
The brand identity collection includes: one in brand evaluation information, brand ranking information, brand taste and brand temporal information
It is a or multiple；
The linked character collection includes: first time dining information of the user in brand.
6. the prediction technique of frequent customer according to claim 1, which is characterized in that the prediction model include: input layer,
Output layer and at least one layer of hidden layer between the input layer and the output layer；
The input layer includes at least one node；
The hidden layer includes at least one node；
The output layer includes a node；
Every layer of each node is connected with each node of adjacent layer respectively.
7. a kind of prediction meanss of frequent customer, which is characterized in that described device includes:
Prediction model loading module, for loading prediction model, wherein the prediction model carries the model ginseng that training obtains
Numerical value；
Identifier acquisition module, the first user identifier and the user to be predicted for obtaining user to be predicted were patronized once
Brand number one brand mark；
Fisrt feature set generation module, for being identified from database according to first user identifier and the number one brand
Search the corresponding user characteristics collection of first user identifier, the number one brand identifies corresponding brand identity collection and described
First user identifier linked character collection corresponding with number one brand mark, and save as fisrt feature set；
Prediction result generation module for the fisrt feature set to be input to the prediction model, and obtains the prediction
Whether the user to be predicted that model is calculated according to the model parameter value can become the pre- of the frequent customer of the brand
Survey result.
8. the prediction meanss of frequent customer according to claim 7, which is characterized in that described device further include:
Predictablity rate generation module, for whether the prediction result and the user to be predicted can be become the brand
The actual result of frequent customer compares, and obtains predictablity rate.
9. a kind of calculating equipment including memory, processor and stores the calculating that can be run on a memory and on a processor
Machine instruction, which is characterized in that the processor performs the steps of when executing described instruction
Load prediction model, wherein the prediction model carries the model parameter value that training obtains；
The first user identifier and the user to be predicted that obtain user to be predicted patronized the number one brand of primary brand
Mark；
It is corresponding that first user identifier is searched from database according to first user identifier and number one brand mark
User characteristics collection, the number one brand identify corresponding brand identity collection and first user identifier and first product
Board identifies corresponding linked character collection, and saves as fisrt feature set；
The fisrt feature set is input to the prediction model, and obtains the prediction model according to the model parameter value
Whether the user to be predicted being calculated can become the prediction result of the frequent customer of the brand.
10. a kind of computer readable storage medium, is stored with computer instruction, which is characterized in that the instruction is held by processor
The step of claim 1-6 any one the method is realized when row.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810421976.4A CN109657832A (en) | 2018-05-04 | 2018-05-04 | A kind of prediction technique and device of frequent customer |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810421976.4A CN109657832A (en) | 2018-05-04 | 2018-05-04 | A kind of prediction technique and device of frequent customer |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109657832A true CN109657832A (en) | 2019-04-19 |
Family
ID=66110206
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810421976.4A Pending CN109657832A (en) | 2018-05-04 | 2018-05-04 | A kind of prediction technique and device of frequent customer |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109657832A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110210913A (en) * | 2019-06-14 | 2019-09-06 | 重庆邮电大学 | A kind of businessman frequent customer's prediction technique based on big data |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102156932A (en) * | 2010-02-11 | 2011-08-17 | 阿里巴巴集团控股有限公司 | Prediction method and device for secondary purchase intention of customers |
| CN106909990A (en) * | 2017-03-01 | 2017-06-30 | 腾讯科技（深圳）有限公司 | A kind of Forecasting Methodology and device based on historical data |
| CN107220845A (en) * | 2017-05-09 | 2017-09-29 | 北京小度信息科技有限公司 | User purchases probabilistic forecasting/user quality and determines method, device and electronic equipment again |
-
2018
- 2018-05-04 CN CN201810421976.4A patent/CN109657832A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102156932A (en) * | 2010-02-11 | 2011-08-17 | 阿里巴巴集团控股有限公司 | Prediction method and device for secondary purchase intention of customers |
| CN106909990A (en) * | 2017-03-01 | 2017-06-30 | 腾讯科技（深圳）有限公司 | A kind of Forecasting Methodology and device based on historical data |
| CN107220845A (en) * | 2017-05-09 | 2017-09-29 | 北京小度信息科技有限公司 | User purchases probabilistic forecasting/user quality and determines method, device and electronic equipment again |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110210913A (en) * | 2019-06-14 | 2019-09-06 | 重庆邮电大学 | A kind of businessman frequent customer's prediction technique based on big data |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110008399A (en) | A kind of training method and device, a kind of recommended method and device of recommended models | |
| KR102189065B1 (en) | Interest profile of a user of a mobile application | |
| CN106326345B (en) | Friends method for digging in a kind of social networks based on user behavior | |
| CN110503531A (en) | The dynamic social activity scene recommended method of timing perception | |
| CN106776930B (en) | A kind of location recommendation method incorporating time and geographical location information | |
| CN107800801A (en) | A kind of pushing learning resource method and system for learning preference based on user | |
| Mechant et al. | E–deliberation 2.0 for smart cities: a critical assessment of two'idea generation'cases | |
| CN106503028A (en) | Recommend method and system | |
| CN110020186A (en) | A kind of dining room recommended method and system | |
| CN110033342A (en) | A kind of training method and device, a kind of recommended method and device of recommended models | |
| US11397757B2 (en) | Relevance estimation and actions based thereon | |
| CN111695046B (en) | User portrait inference method and device based on spatio-temporal mobile data representation learning | |
| CN105913159A (en) | Social network event based user's influence prediction method | |
| Hu et al. | Mining shopping patterns for divergent urban regions by incorporating mobility data | |
| Mohammadi et al. | Artificial intelligence-based solution to estimate the spatial accuracy of volunteered geographic data | |
| Kang | Combining grey relationship analysis and neural network to develop attractive automobile booth design | |
| CN106407379A (en) | Hadoop platform based movie recommendation method | |
| CN110276387A (en) | A kind of generation method and device of model | |
| CN110263250A (en) | A kind of generation method and device of recommended models | |
| Xie et al. | We know your preferences in new cities: Mining and modeling the behavior of travelers | |
| CN109657832A (en) | A kind of prediction technique and device of frequent customer | |
| CN109977979A (en) | Position method, apparatus, electronic equipment and the storage medium of seed user | |
| Romero et al. | Two decades of agent-based modeling in marketing: a bibliometric analysis | |
| Sellappan et al. | Catering to caterers: expectations and satisfaction of restaurant partners from food online order and delivery (food) operator | |
| CN110502693A (en) | The method, apparatus used in project recommendation calculates equipment and storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190419 |