CN105825393A - Decision making method and device based on measurement model - Google Patents
Decision making method and device based on measurement model Download PDFInfo
- Publication number
- CN105825393A CN105825393A CN201510013239.7A CN201510013239A CN105825393A CN 105825393 A CN105825393 A CN 105825393A CN 201510013239 A CN201510013239 A CN 201510013239A CN 105825393 A CN105825393 A CN 105825393A
- Authority
- CN
- China
- Prior art keywords
- user
- making device
- data
- group
- observation point
- 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
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The present invention discloses a decision making method and device based on a measurement model. The method comprises the steps that an automatic decision making device determines the data set of each user at a current observation point; the automatic decision making device carries out the fitting of adjustment functions on each user group by using the data set of each user at a current observation point and selects the adjustment function from the fitting result of the adjustment functions; the automatic decision making device obtains the measurement model output result of the current observation point and the current selected adjustment function and adjusts the measurement model output result by using the adjustment function; and the automatic decision making device makes a decision by using the adjusted measurement model output result. In the embodiment of the invention, the automatic decision making device can carry out the adaptive fusion of across-mode output, a comprehensive measurement model is provided for a decision-making system, latest behavioral data can be utilized, and the online adaptive adjustment based on cumulative event generation probability consistency is carried out on the measurement models of different user groups.
Description
Technical field
The application relates to internet arena, particularly relates to a kind of decision method based on metering model and device.
Background technology
Whether metering model, by the mutual relation between equation quantitatively or qualitatively each variable of descriptive system or cause effect relation, thus can make prediction to future event.Metering model can be used in every field.Such as, in marketing domain, whether client can be responded the marketing activity provided by metering model；In e-commerce field, can predict whether client can buy certain commodity within following a period of time by metering model.The application of metering model, particularly at big data age, has been proven to enough provide good reference frame to enterprise or individual in operational decision making.
In metering model development process, developer can set up a model for each custom partitioning (client of stand-alone product), and disposes in systems.Under such a scenario, the strategy that decision system needs corresponding some sets independent processes different customer groups respectively.Such as, on marketing access strategy, it is less than 10% that the client of product A organizes access probability, and it is 20% that the client of product B organizes access probability.
When generating strategy respectively for different customers, although the policy goals of optimum can be reached single customer group, but global optimum can not be reached, and in such as product access strategy, also cannot ensure the consistent of overall event occurrence rate level between different access customer groups.Further, in order to improve metering model performance in commercial applications, all clients would generally first be hived off, and then develops a set of metering model in respectively Mei Gezi customers.And at strategy end, the predicament how carrying out integrated decision-making based on many sub-metering models outputs will be faced.Therefore, how to carry out cross-module type output adaptive and merge, provide a comprehensive metering model to be a great problem that industry runs into for decision system.
Summary of the invention
The embodiment of the present application provides a kind of decision method based on metering model and device, to carry out the self adaptation fusion of cross-module type output, provides a comprehensive metering model for decision system.
The embodiment of the present application provides a kind of decision method based on metering model, said method comprising the steps of: automated decision-making device determines the data set at Current observation point of each user in each user's group；
Described automated decision-making device utilizes described each user that each user's group is adjusted the matching of function at the data set of Current observation point, and chooses Tuning function from the fitting result of Tuning function；
Described automated decision-making device obtains the metering model of Current observation point and exports result and the Tuning function currently chosen, and utilizes described Tuning function to be adjusted metering model output result；
Described automated decision-making device utilizes the metering model output result after adjusting to carry out decision-making.
The metering model containing each user in user's group in described data set exports result, time-event occurrence flag, ID.
Described automated decision-making device determines that each user in each user's group, in the process of the data set of Current observation point, specifically includes:
Described automated decision-making device obtains Current observation point, and determines apart from described Current observation point nearest containing the current observation station of complete table, and obtains the metering model output result of all users of the described observation station current containing complete table；Described automated decision-making device obtains the behavioral data in user's at the appointed time section, and the described appointment time period is the described observation station current containing the complete table interval to current point in time；Described automated decision-making device utilizes the behavioral data in user's at the appointed time section to calculate the time-event occurrence flag that user is corresponding, user occurs the time-event occurrence flag of event to be the first mark within the performance phase, and user does not occurs the time-event occurrence flag of event to be the second mark within the performance phase.
Described automated decision-making device utilizes described each user that each user's group is adjusted the matching of function at the data set of Current observation point, and from the fitting result of Tuning function, choose the process of Tuning function, specifically include: based on the function template being pre-configured with, described automated decision-making device is under identical accumulation events incidence, demarcate difference as target with the probability minimizing metering model, according to each user in independent variable and dependent variable and each user's group at the data set of Current observation point, each user's group is adjusted the matching of function, the functional form of Tuning function is drawn with matching, and from the fitting result of Tuning function, choose the Tuning function meeting the true corresponding relation of data.
Each user in each user's group that described automated decision-making device determines is at the data set M={M of Current observation point1, M2..., Mk, M1It is the data set at described Current observation point of each user in the 1st user's group, M2It is the data set at described Current observation point of each user in the 2nd user's group ..., MkFor each user in kth user's group at the data set of described Current observation point；Described automated decision-making device utilizes described each user that each user's group is adjusted the matching of function at the data set of Current observation point, and chooses the process of Tuning function from the fitting result of Tuning function, specifically includes:
For each M in Mk, described automated decision-making device is ranked up from low to high according to metering model output result；Described automated decision-making device exports result order from low to high according to metering model, the metering model output probability border of accumulation events incidence search correspondence at certain intervals, obtain accumulation events incidence list, and under each accumulation events incidence, search the model output mark of correspondence；
Described automated decision-making device utilizes described accumulation events incidence list and described model output mark to obtain N group data point Pi；Wherein, 1≤i≤N, described N are default value；
Described automated decision-making device obtains k strong point set P={P1, P2..., Pk}；
Described automated decision-making device chooses a set of data points in described set of data points P, for other k-1 set of data points P remaining in set of data points Pk, under identical accumulation events incidence list, other k-1 set of data points PkIn model output mark as independent variable, the model output mark in the set of data points currently chosen as dependent variable, is target by obtaining minimizing prediction standard deviation, and online fitting goes out other k-1 set of data points PkIn model output mark be mapped in the set of data points currently chosen model output mark Tuning function, calculate Tuning function the goodness of fit.
The embodiment of the present application provides a kind of automated decision-making device, and described automated decision-making device specifically includes:
Determine module, for determining the data set at Current observation point of each user in each user's group；
Select module, for utilizing described each user that each user's group is adjusted at the data set of Current observation point the matching of function, and from the fitting result of Tuning function, choose Tuning function；
Adjusting module, for the metering model output result obtaining Current observation point and the Tuning function currently chosen, and utilizes described Tuning function to be adjusted metering model output result；
Decision-making module, the metering model output result after utilizing adjustment carries out decision-making.
The metering model containing each user in user's group in described data set exports result, time-event occurrence flag, ID.
Described determine module, specifically for obtaining Current observation point, and determine apart from described Current observation point nearest containing the current observation station of complete table, and obtain the metering model output result of all users of the described observation station current containing complete table；Obtaining the behavioral data in user's at the appointed time section, the described appointment time period is the described observation station current containing the complete table interval to current point in time；
The behavioral data in user's at the appointed time section is utilized to calculate the time-event occurrence flag that user is corresponding, user occurs the time-event occurrence flag of event to be the first mark within the performance phase, and user does not occurs the time-event occurrence flag of event to be the second mark within the performance phase.
Described selection module, specifically for based on the function template being pre-configured with, under identical accumulation events incidence, demarcate difference as target with the probability minimizing metering model, according to each user in independent variable and dependent variable and each user's group at the data set of Current observation point, each user's group is adjusted the matching of function, draws the functional form of Tuning function with matching, and from the fitting result of Tuning function, choose the Tuning function meeting the true corresponding relation of data.
The described each user determined in each user's group that module determines is at the data set M={M of Current observation point1, M2..., Mk, M1It is the data set at described Current observation point of each user in the 1st user's group, M2It is the data set at described Current observation point of each user in the 2nd user's group ..., MkFor each user in kth user's group at the data set of described Current observation point；
Described selection module, specifically for for each M in Mk, it is ranked up from low to high according to metering model output result；Result order from low to high is exported according to metering model, the metering model output probability border of accumulation events incidence search correspondence at certain intervals, obtain accumulation events incidence list, and under each accumulation events incidence, search the model output mark of correspondence；Described accumulation events incidence list and described model output mark is utilized to obtain N group data point Pi；Wherein, 1≤i≤N, described N are default value；Obtain k strong point set P={P1, P2..., Pk}；A set of data points is chosen, for other k-1 set of data points P remaining in set of data points P in described set of data points Pk, under identical accumulation events incidence list, other k-1 set of data points PkIn model output mark as independent variable, the model output mark in the set of data points currently chosen as dependent variable, is target by obtaining minimizing prediction standard deviation, and online fitting goes out other k-1 set of data points PkIn model output mark be mapped in the set of data points currently chosen model output mark Tuning function, calculate Tuning function the goodness of fit.
nullCompared with prior art，The embodiment of the present application at least has the advantage that in the embodiment of the present application，Automated decision-making device can carry out the self adaptation of cross-module type output and merge，A comprehensive metering model is provided for decision system，Up-to-date behavioral data can be utilized，The online adaptive carrying out the metering model of different user colony based on accumulation event occurrence rate is consistent adjusts，Output between metering model well can be integrated，Comprehensive metering model result is provided for decision system，Up-to-date performance data can be utilized，Carry out real-time fitting Tuning function on line，Reach effect sample persistently followed the tracks of and metering model integration persistently adjusts，There is good versatility，Commercial undertaking's decision-making can be applied to、The systems such as customer account management，Carry out integrating on cross-module molded line to the metering model with prediction similar events probability of happening function.
Accompanying drawing explanation
In order to the technical scheme of the embodiment of the present application is clearly described, in describing the embodiment of the present application below, the required accompanying drawing used is briefly described, apparently, accompanying drawing in describing below is only some embodiments of the application, for those of ordinary skill in the art, on the premise of not paying creative work, it is also possible to obtain other accompanying drawing according to these accompanying drawings of the embodiment of the present application.
Fig. 1 is a kind of based on metering model the decision method schematic flow sheet that the embodiment of the present application one provides；
Fig. 2 is the structural representation of the automated decision-making device proposed in the embodiment of the present application one；
Fig. 3 is the sample observation station proposed in the embodiment of the present application one and performance phase schematic diagram；
Fig. 4 is the structural representation of the automated decision-making device proposed in the embodiment of the present application one；
Fig. 5 is the flow chart merging Adaptable System across model score proposed in the embodiment of the present application one；
Fig. 6 is the mapping relations of Model B mark and the model A mark proposed in the embodiment of the present application one；
Fig. 7 be the embodiment of the present application one proposes in test data set, the result of three models is at the schematic diagram of the comparing result merged by mark before and after Adaptable System；
Fig. 8 is the structural representation of a kind of automated decision-making device that the embodiment of the present application two provides.
Detailed description of the invention
Below in conjunction with the accompanying drawing in the embodiment of the present application, the technical scheme in the embodiment of the present application is clearly and completely described, it is clear that described embodiment is only a part of embodiment of the application rather than whole embodiments.Based on the embodiment in the application, the every other embodiment that those of ordinary skill in the art are obtained under not making creative work premise, broadly fall into the scope of the application protection.
Embodiment one
For problems of the prior art, the embodiment of the present application one provides a kind of decision method based on metering model, as it is shown in figure 1, should decision method based on metering model specifically may comprise steps of:
Step 101, automated decision-making device determines the data set at Current observation point of each user in each user's group.Wherein, each user in each user's group contains metering model output result, time-event occurrence flag, the ID of each user in user's group in the data set of Current observation point.
In the embodiment of the present application, automated decision-making device determines the process at the data set of Current observation point of each user in each user's group, it is specifically including but not limited to following manner: automated decision-making device obtains Current observation point, and determine apart from this current observation station nearest containing the current observation station of complete table, and obtain the metering model output result of all users of the observation station current containing complete table；Further, automated decision-making device obtains the behavioral data in user's at the appointed time section, and this appointment time period is the interval containing the current observation station of complete table to current point in time；Further, the behavioral data in automated decision-making device utilizes user's at the appointed time section calculates the time-event occurrence flag that user is corresponding；Wherein, user occurs the time-event occurrence flag of event to be the first mark (such as 0) within the performance phase, and user does not occurs the time-event occurrence flag of event to be the second mark (such as 1) within the performance phase.
Step 102, automated decision-making device utilizes each user that each user's group is adjusted at the data set of Current observation point the matching of function, and chooses Tuning function from the fitting result of Tuning function.
In the embodiment of the present application, automated decision-making device utilizes each user that each user's group is adjusted the matching of function at the data set of Current observation point, and from the fitting result of Tuning function, choose the process of Tuning function, specifically can include but not limited to following manner: based on the function template being pre-configured with, automated decision-making device is under identical accumulation events incidence, demarcate difference as target with the probability minimizing metering model, according to each user in independent variable and dependent variable and each user's group at the data set of Current observation point, each user's group is adjusted the matching of function, the functional form of Tuning function is drawn with matching, and from the fitting result of Tuning function, choose the Tuning function meeting the true corresponding relation of data.
In the embodiment of the present application, each user in each user's group that automated decision-making device determines is at the data set M={M of Current observation point1, M2..., Mk, M1It is the data set at described Current observation point of each user in the 1st user's group, M2It is the data set at described Current observation point of each user in the 2nd user's group ..., MkFor each user in kth user's group at the data set of described Current observation point.Based on this, automated decision-making device utilizes each user that each user's group is adjusted the matching of function at the data set of Current observation point, and chooses the process of Tuning function from the fitting result of Tuning function, is specifically including but not limited to: for each M in Mk, automated decision-making device is ranked up from low to high according to metering model output result；Automated decision-making device exports result order from low to high according to metering model, the metering model output probability border of accumulation events incidence search correspondence at certain intervals, obtain accumulation events incidence list, and under each accumulation events incidence, search the model output mark of correspondence；Automated decision-making device utilizes accumulation events incidence list and model output mark to obtain N group data point Pi；Wherein, 1≤i≤N, N are default value；Automated decision-making device obtains k strong point set P={P1, P2..., Pk}；Automated decision-making device chooses a set of data points in set of data points P, for other k-1 set of data points P remaining in set of data points Pk, under identical accumulation events incidence list, other k-1 set of data points PkIn model output mark make as independent variable, the model output mark in the set of data points currently chosen as dependent variable, is target by obtaining minimizing prediction standard deviation, and online fitting goes out other k-1 set of data points PkIn model output mark be mapped in the set of data points currently chosen model output mark make Tuning function, calculating Tuning function the goodness of fit.
In the embodiment of the present invention, accumulation events incidence list can be represented by cum_event_rate (i), and the model output mark searching correspondence under each accumulation events incidence can be represented by predict_edge (i).1≤i≤N, N are default value, and cum_event_rate (N) is the overall incidence rate value of user's group that the overall events incidence in k user's group is minimum.Based on this, N group data point P that automated decision-making device obtainsi=(cum_event_rate (i), predict_edge (i)).Further, for other k-1 set of data points P remaining in set of data points Pk, under identical cum_event_rate (i), other k-1 set of data points PkIn predict_edge (i) as independent variable, the predict_edge (i) in the set of data points currently chosen as dependent variable, is target by obtaining minimizing prediction standard deviation, and online fitting goes out other k-1 set of data points PkIn predict_edge be mapped to the Tuning function of the predict_edge in the set of data points currently chosen, calculate the goodness of fit of Tuning function.
Step 103, automated decision-making device obtains the metering model of Current observation point and exports result and the Tuning function currently chosen, and utilizes Tuning function to be adjusted metering model output result.
Step 104, automated decision-making device utilizes the metering model output result after adjusting to carry out decision-making.
nullCompared with prior art，The embodiment of the present application at least has the advantage that in the embodiment of the present application，Automated decision-making device can realize the Adaptable System merged across model score，The self adaptation that can carry out cross-module type output merges，A comprehensive metering model is provided for decision system，Up-to-date behavioral data can be utilized，The online adaptive carrying out the metering model of different user colony based on accumulation event occurrence rate is consistent adjusts，Output between metering model well can be integrated，Comprehensive metering model result is provided for decision system，Up-to-date performance data can be utilized，Carry out real-time fitting Tuning function on line，Reach effect sample persistently followed the tracks of and metering model integration persistently adjusts，There is good versatility，Commercial undertaking's decision-making can be applied to、The systems such as customer account management，Carry out integrating on cross-module molded line to the metering model with prediction similar events probability of happening function.Wherein, merge across model score, refer to that the some submodels output to having the same event occurrence rate of prediction carries out comprehensive integration so that integrated results can be used in for the strategy in all clients colony, meets simultaneously and keeps some Attribute consistency requirement between sub-customer group.Adaptable System, finger system is according to the change of environment, adjusting and himself make its behavior new or in the environment of having changed, reach characteristic that is best or that at least allow and function, this system to environmental change with adaptive ability is referred to as Adaptable System.
Below in conjunction with concrete application scenarios, the technical scheme of the embodiment of the present application is described in detail.
In the embodiment of the present application, the Adaptable System (i.e. automated decision-making device) merged across model score goes for having been deployed in all kinds of metering models that automated decision-making system (on line) is run.Owing to the cycle of operation of metering model can be day, week, the moon etc., it is also possible to being real time execution, in the embodiment of the present application, it is assumed that metering model is monthly to run once, this is also modal cycle of operation in metering model.
As in figure 2 it is shown, be the structural representation of automated decision-making device, this automated decision-making device specifically includes: data memory module, model module and decision-making module.As it is shown on figure 3, be sample observation station and the schematic diagram of performance phase.Data memory module is used for storing on basic data (such as transaction details data etc.), line the base values needed for metering model, all clients in the model score result etc. of each observation station.Model module is in set time (referred to as observation station monthly, run as shown in Figure 3), and model module is operationally, the most up-to-date base values that metering model is corresponding is obtained from data memory module, then metering model or rule are called, give a mark to all clients, and marking result is stored data memory module.Decision-making module, when needs make a policy, extracts, from data memory module, the metering model scoring that client is corresponding, makes certain decision-making to client, such as refusal user's access or calculating user's loan limit etc..
On the basis of the automated decision-making apparatus structure shown in Fig. 2, as shown in Figure 4, for the structural representation based on the automated decision-making device merging Adaptable System across model score proposed in the embodiment of the present application.In the embodiment of the present application, by disposing the automatic preparation module of data, function adaptive training module, mark automatic regulating module before decision-making module, the mechanism of metering model result is directly obtained with cut-out decision-making module, but after the metering model scoring obtaining client, by the automatic preparation module of data, function adaptive training module, mark automatic regulating module, adjust in real time with the mark to metering model, after event occurrence rate value after being adjusted, just it is supplied to decision-making module and carries out follow-up decision.
The function of the automatic preparation module of data is data pick-up and calculating.The data of enterprise, such as transaction data, customer action data etc., all can store at data memory module and manage.The automatic preparation module of data is connected with data memory module, is the Data entries of self-adapted adjustment system.The automatic preparation module of data is extracted by the metering model output result run in decision system from data memory module.The automatic preparation module of data also needs to extract the behavioral data of user in nearest one period of set time from data memory module.After extracting the behavioral data of user, the automatic preparation module of data starts to calculate in user's dimension, if there occurs that certain needs the event paid close attention to.Such as: whether client has responded pushed notice, if there occurs that credit is exceeded the time limit.In the event occurrence flag obtained in metering model output result and user's dimension, these two parts data are integrated, and are transferred to function adaptive training module by the automatic preparation module of data.
Function adaptive training module is the core of self-adapted adjustment system, and its critical function is for automatically training Tuning function, and provides the goodness index training the Tuning function obtained.Function adaptive training module stores has multiple function template, existing linear function, also has nonlinear function.During training, function adaptive training module can remove the multiple Tuning function of matching according to independent variable and dependent variable, and therefrom chooses the Tuning function best suiting the true corresponding relation of data, and stores the parameter of Tuning function.It addition, function adaptive training module additionally provides different Function Fitting methods, including method of least square, EM (ExpectationMaximuzation, it is desirable to maximize) algorithm etc..For linear function, method of least square can be called and be trained, for nonlinear function, EM algorithm can be called and be trained.
Mark automatic regulating module is to merge the implementation unit of Adaptable System across model score, mark automatic regulating module obtains the metering model output result needing to be adjusted from data memory module, and obtain form and the parameter of the Tuning function of function adaptive training module, then form and the parameter of the Tuning function of function adaptive training module are utilized, to need be adjusted metering model output result be adjusted, and will adjust after result be supplied to decision-making module.Certain decision-making to client is made by decision-making module.
As it is shown in figure 5, be the flow chart merging Adaptable System across model score, idiographic flow includes:
Step 1, in Current observation point (observation station 4 as shown in Figure 3), distance the most nearest can be found containing the current observation station (observation station 3 in Fig. 3) of complete table.The automatic preparation module of data obtains model output result (predict) of all clients of observation station 3 from data memory module.
The automatic preparation module of step 2, data extracts the behavioral data in user's up-to-date a period of time from data memory module, and this time period is a nearest interval arriving current point in time containing the observation station (such as observation station 3) that complete table is current.Afterwards, the automatic preparation module of data can calculate whether each client occurs certain event (event) within the correspondence performance phase, and event occurs then event=1, otherwise event=0.
Step 3, data step 1 and step 2 obtained merge becomes final data collection, and this data set contains model output result predict and time-event occurrence flag event, and comprises subscriber identification field, and this final data collection is designated as data set M1={ ID, predict, event}.In like manner, repeat step 1-step 3, can obtain needing to carry out the data set of other k-1 client's group model of mark fusion, therefore, M={M1, M2..., Mk, M1It is the data set at described Current observation point of each user in the 1st user's group, M2It is the data set at described Current observation point of each user in the 2nd user's group ..., MkFor each user in kth user's group at the data set of described Current observation point.Data set M={M1, M2..., MkIt is sent to function adaptive training module by the automatic preparation module of data.
Step 4, function adaptive training module, after obtaining data set, proceed by Tuning function matching.For each M in Mk, data sample is ranked up from low to high by function adaptive training module according to model result (predict).Afterwards, function adaptive training module according to model result direction from low to high, model output probability (predict) border that accumulation events incidence search at certain intervals is corresponding.Assume that accumulation events incidence is from the beginning of 0.1%, it is spaced apart 0.1%, then obtain accumulation events incidence list cum_event_rate=(0.1%, 0.2%, 0.3% ..., cum_event_rate (i)), 1≤i≤N, wherein, cum_event_rate (N) is the overall incidence rate value of that user group that the overall events incidence in k user's group is minimum.Accordingly, a model output mark of correspondence can be searched under each accumulation events incidence, be designated as cum_event_rate (i) and predict_edge (i) respectively.As such, it is possible to obtain N group data point P1=(cum_event_rate (i), predict_edge (i)).It is said that in general, the needs that arrange of N determine a suitable numerical value according to customer group, the least N can make the data of matching on the low side, fitting result bigger error.In like manner, can obtain other k-1 set of data points, based on this, function adaptive training module can obtain k strong point set P={P1, P2..., Pk}。
Function adaptive training module chooses a set of data points in set of data points P, it is assumed that choose P1As standard, for any one P in other k-1 set of data points remaining in set of data points Pk, under identical cum_event_rate (i), other k-1 set of data points PkIn predict_edge (i) as independent variable, the P currently chosen1In predict_edge (i) be dependent variable, be target by obtaining minimizing prediction standard deviation, online fitting goes out other k-1 set of data points PkIn predict_edge be mapped to the P currently chosen1In the Tuning function of predict_edge, and calculate the goodness of fit of Tuning function, finally, corresponding to k model, have k-1 Tuning function to need matching, and by Tuning function form with estimate that parameter stores.So far, the model score of Current observation point merges that self-adaptative adjustment function is the most trained completes.
Step 5, when decision model need extract client metering model scoring carry out decision-making time, can trigger mark merge Adaptable System.Mark automatic regulating module model output result when data memory module obtains Current observation point (point of observation 4 such as Fig. 3).Afterwards, the form of the corresponding Tuning function that mark automatic regulating module obtaining step 4 obtains and parameter.Further, the adjustment by Tuning function of the model score of client, the mark of the events incidence after being just adjusted, and it is supplied to decision-making module use.
As a example by the PD model whether the prediction loan customer run in certain commercial undertaking's decision system below can exceed the time limit in showing the phase half a year in future, the effect merging Adaptable System across model score is illustrated.Measure of effectiveness uses following index: assume that loan customer is divided into K Ge Zi customers, the entirety of Mei Gezi customers rate of exceeding the time limit is respectively event_rate (i), 1≤i≤K, the most corresponding has K submodel, each individual consumers is predict by the prediction probability value that model score obtains, then accumulative rate deviation df of exceeding the time limit of definition is: df (score)=max (| event_rate_p (i, score)-event_rate_p (j, score) |), 1≤i, j≤K；Event_rate_p (i, score) is the rate of exceeding the time limit meeting predict≤score crowd in sub-customers i, score be (0, min (event_rate (i))] in arbitrary value.Accumulation exceeds the time limit rate maximum deviation df_max and accumulation exceeds the time limit, and rate average deviation df_avg is: df_max=max (df (score))；Df_avg=avg (df (score)).Df value is the biggest, then illustrate in the allowed band of score, and the overall risk difference between colony is the biggest, otherwise, then explanation risk is the most consistent.Based on this, the PD model in system runs once beginning of the month every month, and 6 months futures that every month is run are performance phase window, and the data using 2013/09 to 2013/11 are as training dataset, and 2013/12 month to 2014/01 month as test data set.In the embodiment of the present application, as a example by 3 client crowds, experiment relates to the integration between 3 PD models, and model is called model A, Model B and MODEL C.Data set basic condition is as shown in table 1.Table 2 gives the result example after in function adaptive training module processing sample data.
Table 1
| Colony | Number of users (DEV) | Rate of exceeding the time limit (DEV) | Number of users (OOT) | Rate of exceeding the time limit (OOT) |
| Colony one | 56,000 | 7.8% | 36,000 | 7.7% |
| Colony two | 90,000 | 6.3% | 50,000 | 6.8% |
| Colony three | 45,000 | 3.2% | 25,000 | 3.0% |
Table 2
Function adaptive training module is according to the accumulation events incidence obtained and corresponding model score result, export as standard using model A, then using model A probabilistic margins as dependent variable, Model B probabilistic margins, MODEL C probabilistic margins are as independent variable, 2 Tuning function of auto-adapted fitting, and calculate the R-square judging quota as the goodness of fit.Functional relationship for Model B mark Yu model A mark, system gives an exponential function after fitting function, contrast, as shown in Figure 6, functional form can be drawn the mapping relations of Model B mark and model A mark by canbe used on line method of least square adaptive training.
Afterwards, the Tuning function obtained being applied to test data set, as it is shown in fig. 7, be in test data set, the result of three models is merging the comparing result schematic diagram before and after Adaptable System by mark.It can be seen from figure 7 that before and after integrating, be positioned at three curves overlapped parts, the accumulation that PD mark is corresponding in three colonies rate of exceeding the time limit has obtained good integration.Through above-mentioned adjustment, under identical PD mark, the accumulation of the 3 groups of samples rate average deviation that exceeds the time limit drops to 0.3% from 1.6%, and maximum deviation also falls below 0.4% from 2%.On the whole, merge self-adaptative adjustment across model score and efficiently utilize up-to-date customer action data, to some, there is identical function, the model output set up on different customer groups has carried out self-adaptative adjustment, so that final model output can be by wider application, ensure that the decision-making made on overall client, it is possible between Ge Zi customers, obtain consistent overall risk simultaneously.
Embodiment two
Conceiving based on the application as said method, additionally provide a kind of automated decision-making device in the embodiment of the present application, as shown in Figure 8, described automated decision-making device specifically includes:
Determine module 11, for determining the data set at Current observation point of each user in each user's group；Wherein, the metering model containing each user in user's group in described data set exports result, time-event occurrence flag, ID；
Select module 12, for utilizing described each user that each user's group is adjusted at the data set of Current observation point the matching of function, and from the fitting result of Tuning function, choose Tuning function；
Adjusting module 13, for the metering model output result obtaining Current observation point and the Tuning function currently chosen, and utilizes described Tuning function to be adjusted metering model output result；
Decision-making module 14, the metering model output result after utilizing adjustment carries out decision-making.
Described determine module 11, specifically for obtaining Current observation point, and determine apart from described Current observation point nearest containing the current observation station of complete table, and obtain the metering model output result of all users of the described observation station current containing complete table；Obtaining the behavioral data in user's at the appointed time section, the described appointment time period is the described observation station current containing the complete table interval to current point in time；
The behavioral data in user's at the appointed time section is utilized to calculate the time-event occurrence flag that user is corresponding, user occurs the time-event occurrence flag of event to be the first mark within the performance phase, and user does not occurs the time-event occurrence flag of event to be the second mark within the performance phase.
Described selection module 12, specifically for based on the function template being pre-configured with, under identical accumulation events incidence, demarcate difference as target with the probability minimizing metering model, according to each user in independent variable and dependent variable and each user's group at the data set of Current observation point, each user's group is adjusted the matching of function, draws the functional form of Tuning function with matching, and from the fitting result of Tuning function, choose the Tuning function meeting the true corresponding relation of data.
The described each user determined in each user's group that module 11 determines is at the data set M={M of Current observation point1, M2..., Mk, M1It is the data set at described Current observation point of each user in the 1st user's group, M2It is the data set at described Current observation point of each user in the 2nd user's group ..., MkFor each user in kth user's group at the data set of described Current observation point；
Described selection module 12, specifically for for each M in Mk, it is ranked up from low to high according to metering model output result；Result order from low to high is exported according to metering model, the metering model output probability border of accumulation events incidence search correspondence at certain intervals, obtain accumulation events incidence list, and under each accumulation events incidence, search the model output mark of correspondence；Described accumulation events incidence list and described model output mark is utilized to obtain N group data point Pi；Wherein, 1≤i≤N, described N are default value；Obtain k strong point set P={P1, P2..., Pk}；A set of data points is chosen, for other k-1 set of data points P remaining in set of data points P in described set of data points Pk, under identical accumulation events incidence list, other k-1 set of data points PkIn model output mark as independent variable, the model output mark in the set of data points currently chosen as dependent variable, is target by obtaining minimizing prediction standard deviation, and online fitting goes out other k-1 set of data points PkIn model output mark be mapped in the set of data points currently chosen model output mark Tuning function, calculate Tuning function the goodness of fit.
Wherein, the modules of the application device can be integrated in one, it is also possible to separates and disposes.Above-mentioned module can merge into a module, it is also possible to is further split into multiple submodule.
Through the above description of the embodiments, those skilled in the art is it can be understood that can add the mode of required general hardware platform by software to the application and realize, naturally it is also possible to by hardware, but a lot of in the case of the former is more preferably embodiment.Based on such understanding, the part that prior art is contributed by the technical scheme of the application the most in other words can embody with the form of software product, this computer software product is stored in a storage medium, including some instructions with so that a computer equipment (can be personal computer, server, or the network equipment etc.) perform the method described in each embodiment of the application.It will be appreciated by those skilled in the art that accompanying drawing is the schematic diagram of a preferred embodiment, module or flow process in accompanying drawing are not necessarily implemented necessary to the application.It will be appreciated by those skilled in the art that the module in the device in embodiment can describe according to embodiment to carry out being distributed in the device of embodiment, it is also possible to carry out respective change and be disposed other than in one or more devices of the present embodiment.The module of above-described embodiment can merge into a module, it is also possible to is further split into multiple submodule.Above-mentioned the embodiment of the present application sequence number, just to describing, does not represent the quality of embodiment.The several specific embodiments being only the application disclosed above, but, the application is not limited to this, and the changes that any person skilled in the art can think of all should fall into the protection domain of the application.
Claims (10)
1. a decision method based on metering model, it is characterised in that said method comprising the steps of:
Automated decision-making device determines the data set at Current observation point of each user in each user's group；
Described automated decision-making device utilizes described each user that each user's group is adjusted the matching of function at the data set of Current observation point, and chooses Tuning function from the fitting result of Tuning function；
Described automated decision-making device obtains the metering model of Current observation point and exports result and the Tuning function currently chosen, and utilizes described Tuning function to be adjusted metering model output result；
Described automated decision-making device utilizes the metering model output result after adjusting to carry out decision-making.
2. the method for claim 1, it is characterised in that the metering model containing each user in user's group in described data set exports result, time-event occurrence flag, ID.
3. method as claimed in claim 2, it is characterised in that described automated decision-making device determines that each user in each user's group, in the process of the data set of Current observation point, specifically includes:
Described automated decision-making device obtains Current observation point, and determines apart from described Current observation point nearest containing the current observation station of complete table, and obtains the metering model output result of all users of the described observation station current containing complete table；Described automated decision-making device obtains the behavioral data in user's at the appointed time section, and the described appointment time period is the described observation station current containing the complete table interval to current point in time；Described automated decision-making device utilizes the behavioral data in user's at the appointed time section to calculate the time-event occurrence flag that user is corresponding, user occurs the time-event occurrence flag of event to be the first mark within the performance phase, and user does not occurs the time-event occurrence flag of event to be the second mark within the performance phase.
4. the method as described in any one of claim 1-3, it is characterized in that, described automated decision-making device utilizes described each user that each user's group is adjusted the matching of function at the data set of Current observation point, and chooses the process of Tuning function from the fitting result of Tuning function, specifically includes:
Based on the function template being pre-configured with, described automated decision-making device is under identical accumulation events incidence, demarcate difference as target with the probability minimizing metering model, according to each user in independent variable and dependent variable and each user's group at the data set of Current observation point, each user's group is adjusted the matching of function, draw the functional form of Tuning function with matching, and from the fitting result of Tuning function, choose the Tuning function meeting the true corresponding relation of data.
5. method as claimed in claim 2 or claim 3, it is characterised in that each user in each user's group that described automated decision-making device determines is at the data set M={M of Current observation point1, M2..., Mk, M1It is the data set at described Current observation point of each user in the 1st user's group, M2It is the data set at described Current observation point of each user in the 2nd user's group ..., MkFor each user in kth user's group at the data set of described Current observation point；Described automated decision-making device utilizes described each user that each user's group is adjusted the matching of function at the data set of Current observation point, and chooses the process of Tuning function from the fitting result of Tuning function, specifically includes:
For each M in Mk, described automated decision-making device is ranked up from low to high according to metering model output result；Described automated decision-making device exports result order from low to high according to metering model, the metering model output probability border of accumulation events incidence search correspondence at certain intervals, obtain accumulation events incidence list, and under each accumulation events incidence, search the model output mark of correspondence；
Described automated decision-making device utilizes described accumulation events incidence list and described model output mark to obtain N group data point Pi；Wherein, 1≤i≤N, described N are default value；
Described automated decision-making device obtains k strong point set P={P1, P2..., Pk}；
Described automated decision-making device chooses a set of data points in described set of data points P, for other k-1 set of data points P remaining in set of data points Pk, under identical accumulation events incidence list, other k-1 set of data points PkIn model output mark as independent variable, the model output mark in the set of data points currently chosen as dependent variable, is target by obtaining minimizing prediction standard deviation, and online fitting goes out other k-1 set of data points PkIn model output mark be mapped in the set of data points currently chosen model output mark Tuning function, calculate Tuning function the goodness of fit.
6. an automated decision-making device, it is characterised in that described automated decision-making device specifically includes:
Determine module, for determining the data set at Current observation point of each user in each user's group；
Select module, for utilizing described each user that each user's group is adjusted at the data set of Current observation point the matching of function, and from the fitting result of Tuning function, choose Tuning function；
Adjusting module, for the metering model output result obtaining Current observation point and the Tuning function currently chosen, and utilizes described Tuning function to be adjusted metering model output result；
Decision-making module, the metering model output result after utilizing adjustment carries out decision-making.
7. automated decision-making device as claimed in claim 6, it is characterised in that the metering model containing each user in user's group in described data set exports result, time-event occurrence flag, ID.
8. automated decision-making device as claimed in claim 7, it is characterised in that
Described determine module, specifically for obtaining Current observation point, and determine apart from described Current observation point nearest containing the current observation station of complete table, and obtain the metering model output result of all users of the described observation station current containing complete table；Obtaining the behavioral data in user's at the appointed time section, the described appointment time period is the described observation station current containing the complete table interval to current point in time；
The behavioral data in user's at the appointed time section is utilized to calculate the time-event occurrence flag that user is corresponding, user occurs the time-event occurrence flag of event to be the first mark within the performance phase, and user does not occurs the time-event occurrence flag of event to be the second mark within the performance phase.
9. the automated decision-making device as described in any one of claim 6-8, it is characterised in that
Described selection module, specifically for based on the function template being pre-configured with, under identical accumulation events incidence, demarcate difference as target with the probability minimizing metering model, according to each user in independent variable and dependent variable and each user's group at the data set of Current observation point, each user's group is adjusted the matching of function, draws the functional form of Tuning function with matching, and from the fitting result of Tuning function, choose the Tuning function meeting the true corresponding relation of data.
10. automated decision-making device as claimed in claim 7 or 8, it is characterised in that the described each user determined in each user's group that module determines is at the data set M={M of Current observation point1, M2..., Mk, M1It is the data set at described Current observation point of each user in the 1st user's group, M2It is the data set at described Current observation point of each user in the 2nd user's group ..., MkFor each user in kth user's group at the data set of described Current observation point；
Described selection module, specifically for for each M in Mk, it is ranked up from low to high according to metering model output result；Result order from low to high is exported according to metering model, the metering model output probability border of accumulation events incidence search correspondence at certain intervals, obtain accumulation events incidence list, and under each accumulation events incidence, search the model output mark of correspondence；Described accumulation events incidence list and described model output mark is utilized to obtain N group data point Pi；Wherein, 1≤i≤N, described N are default value；Obtain k strong point set P={P1, P2..., Pk}；A set of data points is chosen, for other k-1 set of data points P remaining in set of data points P in described set of data points Pk, under identical accumulation events incidence list, other k-1 set of data points PkIn model output mark as independent variable, the model output mark in the set of data points currently chosen as dependent variable, is target by obtaining minimizing prediction standard deviation, and online fitting goes out other k-1 set of data points PkIn model output mark be mapped in the set of data points currently chosen model output mark Tuning function, calculate Tuning function the goodness of fit.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510013239.7A CN105825393A (en) | 2015-01-09 | 2015-01-09 | Decision making method and device based on measurement model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510013239.7A CN105825393A (en) | 2015-01-09 | 2015-01-09 | Decision making method and device based on measurement model |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN105825393A true CN105825393A (en) | 2016-08-03 |
Family
ID=56514272
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510013239.7A Pending CN105825393A (en) | 2015-01-09 | 2015-01-09 | Decision making method and device based on measurement model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105825393A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108022123A (en) * | 2016-11-04 | 2018-05-11 | 苏宁云商集团股份有限公司 | The automatic adjusting method and device of a kind of business model |
-
2015
- 2015-01-09 CN CN201510013239.7A patent/CN105825393A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108022123A (en) * | 2016-11-04 | 2018-05-11 | 苏宁云商集团股份有限公司 | The automatic adjusting method and device of a kind of business model |
| CN108022123B (en) * | 2016-11-04 | 2021-12-07 | 南京星云数字技术有限公司 | Automatic adjustment method and device for business model |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108230007B (en) | User intention identification method and device, electronic equipment and storage medium | |
| US10748193B2 (en) | Assessing probability of winning an in-flight deal for different price points | |
| CN108156204B (en) | Target object pushing system and method | |
| CN111931055B (en) | Object recommendation method, object recommendation device and electronic equipment | |
| CN109858957A (en) | Finance product recommended method, device, computer equipment and storage medium | |
| CN105630827B (en) | A kind of information processing method, system and auxiliary system | |
| CN109933721A (en) | A kind of interpretable recommended method merging user concealed article preference and implicit trust | |
| CN114187125A (en) | Claims case distribution method, device, equipment and storage medium | |
| US20180005248A1 (en) | Product, operating system and topic based | |
| CN111738762A (en) | Method, device, equipment and storage medium for determining recovery price of poor assets | |
| US20170372341A1 (en) | Top-down pricing of a complex service deal | |
| CN115880075A (en) | Financial data management method, device and equipment and readable storage medium | |
| CN105405051A (en) | Financial event prediction method and apparatus | |
| CN113869700A (en) | Performance index prediction method and device, electronic equipment and storage medium | |
| CN105550282A (en) | User interest forecasting method by utilizing multidimensional data | |
| EP1107157A2 (en) | System and method for performing predictive analysis | |
| CN105512298A (en) | Interested content prediction method based on machine learning | |
| CN105825393A (en) | Decision making method and device based on measurement model | |
| WO2015179717A1 (en) | Determination of initial value for automated delivery of news items | |
| CN115689708A (en) | Screening method, risk assessment method, device, equipment and medium of training data | |
| CN110766488A (en) | Method and device for automatically determining theme scene | |
| CN105208033A (en) | Group auxiliary recommendation method and system based on intelligent terminal scenes | |
| US20210357401A1 (en) | Automatic frequency recommendation for time series data | |
| CN113571198A (en) | Conversion rate prediction method, device, equipment and storage medium | |
| CN105574095A (en) | Context-sensitive intelligent estimation method for user behaviors |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20160803 |
|
| RJ01 | Rejection of invention patent application after publication |