Summary of the invention
In view of above-mentioned, present disclose provides a kind of method and devices for decision optimization.Utilize this method and device, energyEnough personalized decision optimization is realized for different users.
According to one aspect of the disclosure, a kind of method for decision optimization is provided, comprising: determine in prediction modelPrediction result contribution degree of each decision characteristic variable under user characteristic data, the prediction model is based on decision objectiveCreation, the prediction model includes decision characteristic variable and non-decision characteristic variable；It is special based on each decision determinedThe prediction result contribution degree for levying variable constructs decision characteristic variable combination to be optimized；To constructed decision characteristic variable groupThe variable-value of each decision characteristic variable in conjunction carries out optimizing processing, so that the correspondence prediction result of the prediction modelMost preferably；And it is taken according to the variable of each decision characteristic variable in the decision characteristic variable combination obtained after optimizing is handledValue carries out decision optimization processing.
Optionally, in an example of above-mentioned aspect, determine each decision characteristic variable in prediction model in userPrediction result contribution degree under characteristic includes: each decision characteristic variable determined in prediction model using interpretation modelPrediction result contribution degree under user characteristic data.
Optionally, in an example of above-mentioned aspect, the interpretation model includes one of following interpretation models:Shap value model, LIME model and DeepLift model.
Optionally, in an example of above-mentioned aspect, the prediction knot based on each decision characteristic variable determinedFruit contribution degree, constructing decision characteristic variable combination to be optimized may include: based on each decision characteristic variable determinedPrediction result contribution degree, each decision characteristic variable is ranked up；And it is special from each decision after the sequenceIt levies and selects the biggish predetermined number decision characteristic variable of contribution degree in variable, combined as decision characteristic variable to be optimized.
Optionally, in an example of above-mentioned aspect, to each decision in constructed decision characteristic variable combinationIt may include: using one of following optimizing algorithms optimizing algorithm come to institute that the variable-value of characteristic variable, which carries out optimizing processing,The variable-value progress optimizing processing of each decision characteristic variable in the decision characteristic variable combination of building: particle swarm algorithm,Genetic algorithm, annealing algorithm.
Optionally, in an example of above-mentioned aspect, to each decision in constructed decision characteristic variable combinationIt may include: to determine in predetermined decision variable value range to constructed that the variable-value of characteristic variable, which carries out optimizing processing,The variable-value of each decision characteristic variable in the combination of plan characteristic variable carries out optimizing processing.
According to another aspect of the present disclosure, a kind of device for decision optimization is provided, comprising: contribution degree determination unit,Prediction result contribution degree of each decision characteristic variable under user characteristic data being configured to determine that in prediction model, it is describedPrediction model is created based on decision objective, and the prediction model includes decision characteristic variable and non-decision characteristic variable；CertainlyPlan characteristic variable combines construction unit, is configured as the prediction result contribution based on each decision characteristic variable determinedDegree constructs decision characteristic variable combination to be optimized；Optimizing processing unit is configured as to constructed decision characteristic variable groupThe variable-value of each decision characteristic variable in conjunction carries out optimizing processing, so that the correspondence prediction result of the prediction modelMost preferably；And decision optimization unit, it is configured as according to each in the decision characteristic variable combination obtained after optimizing is handledThe variable-value of a decision characteristic variable carries out decision optimization processing.
Optionally, in an example of above-mentioned aspect, the contribution degree determination unit is configured as: using interpretation modelTo determine prediction result contribution degree of each decision characteristic variable under user characteristic data in prediction model.
Optionally, in an example of above-mentioned aspect, the interpretation model includes one of following interpretation models:Shap value model, LIME model and DeepLift model.
Optionally, in an example of above-mentioned aspect, the decision characteristic variable combination construction unit includes: sequence mouldBlock is configured as the prediction result contribution degree based on each decision characteristic variable determined, to each decision featureVariable is ranked up；And feature selection module, it is configured as selecting tribute from each decision characteristic variable after the sequenceThe biggish predetermined number decision characteristic variable of degree of offering is combined as decision characteristic variable to be optimized.
Optionally, in an example of above-mentioned aspect, the optimizing processing unit is configured as: being calculated using following optimizingOne of method to carry out optimizing to the variable-value of each decision characteristic variable in constructed decision characteristic variable combinationProcessing: particle swarm algorithm, genetic algorithm, annealing algorithm.
Optionally, in an example of above-mentioned aspect, the optimizing processing unit is configured as: in predetermined decision variableIn value range, the variable-value of each decision characteristic variable in constructed decision characteristic variable combination is carried out at optimizingReason.
According to another aspect of the present disclosure, a kind of calculating equipment is provided, comprising: at least one processor, and with it is describedThe memory of at least one processor coupling, the memory store instruction, when described instruction is by least one described processorWhen execution, so that at least one described processor executes the method for being used for decision optimization as described above.
According to another aspect of the present disclosure, a kind of non-transitory machinable medium is provided, is stored with executableInstruction, described instruction make the machine execute the method for being used for decision optimization as described above upon being performed.
Specific embodiment
Theme described herein is discussed referring now to example embodiment.It should be understood that discussing these embodiments onlyIt is in order to enable those skilled in the art can better understand that being not to claim to realize theme described hereinProtection scope, applicability or the exemplary limitation illustrated in book.It can be in the protection scope for not departing from present disclosureIn the case of, the function and arrangement of the element discussed are changed.Each example can according to need, omit, substitute orAdd various processes or component.For example, described method can be executed according to described order in a different order, withAnd each step can be added, omits or combine.In addition, feature described in relatively some examples is in other examplesIt can be combined.
As used in this article, term " includes " and its modification indicate open term, are meant that " including but not limited to ".Term "based" indicates " being based at least partially on ".Term " one embodiment " and " embodiment " expression " at least one implementationExample ".Term " another embodiment " expression " at least one other embodiment ".Term " first ", " second " etc. may refer to notSame or identical object.Here may include other definition, either specific or implicit.Unless bright in contextIt really indicates, otherwise the definition of a term is consistent throughout the specification.
Decision optimization method and device according to an embodiment of the present disclosure is described in detail below in conjunction with attached drawing.
Fig. 1 shows the flow chart of decision optimization method according to an embodiment of the present disclosure.
As shown in Figure 1, determining each decision characteristic variable in prediction model under user characteristic data in block 110Prediction result contribution degree.Here, the prediction model is to be created based on decision objective, and the prediction model includes decisionCharacteristic variable and non-decision characteristic variable.The user characteristic data is corresponding with decision characteristic variable and non-decision characteristic variableUser characteristic data.
For example, it is assumed that decision objective is user's registration conversion number, in transaction risk control scene in marketing sceneIn, decision objective is amount of money summation of the user because swindling trade loss.For this decision objective, suitable characteristic variable group is createdPrediction model is established, thus based on user characteristic data come forecast and decision target, for example, whether prediction user converts, or pre-Survey the amount of money in customer transaction because of fraud loss.In the disclosure, term " non-decision characteristic variable " refers to that decision-making party can not be doneThe characteristic variable related to, for example, the age of user, personality, behavior over history etc..Term " decision characteristic variable " refers to decisionSide can interfere the characteristic variable of change, for example, in marketing scene, the case where decision objective is user's registration conversion numberUnder the characteristic variables such as the preferential and equity for being issued to user, or in transaction risk control scene, export to the wind of userThe characteristic variables such as danger prompting.
In addition, in the disclosure, the prediction model can have previously been based on decision objective to utilize training data to create,It is also possible to utilize training data to create in real time after getting decision objective.Equally, the decision objective can be in advanceInput or real-time input.In addition, being used to carry out prediction model training to improve the accuracy rate of prediction modelTraining data should cover the user group with multiple features as far as possible, and attempt decision variable value as much as possible, fromAnd ensure the rich of training data dimension.
In an example of the disclosure, interpretation model can be used to determine that each decision feature in prediction model becomesMeasure the prediction result contribution degree under user characteristic data.The interpretation model includes one of following interpretation models: ShapValue model, LIME model and DeepLift model.On how to use interpretation model determine in prediction model it is each certainlyPrediction result contribution degree of the plan characteristic variable under user characteristic data, will be illustrated below in reference to Fig. 2.
After the prediction result contribution degree for as above determining each decision characteristic variable, in block 120, based on what is determinedThe prediction result contribution degree of each decision characteristic variable constructs decision characteristic variable combination to be optimized.On how to based on instituteThe prediction result contribution degree for each decision characteristic variable determined combines to construct decision characteristic variable to be optimized, will be underFace combines example shown in Fig. 3 to be illustrated.
After constructing decision characteristic variable combination to be optimized, in block 130, constructed decision characteristic variable is combinedIn each decision characteristic variable variable-value carry out optimizing processing so that the correspondence prediction result of the prediction model is mostIt is good.
For example, in one example, the change to each decision characteristic variable in constructed decision characteristic variable combinationIt may include: using one of following optimizing algorithms optimizing algorithm come special to constructed decision that measurement value, which carries out optimizing processing,The variable-value for levying each decision characteristic variable in variable combination carries out optimizing processing: particle swarm algorithm, genetic algorithm, annealingAlgorithm.Here it is possible to determine used optimizing algorithm based on the type of the decision characteristic variable of pending optimizing processing.ThanSuch as, in the case where the decision characteristic variable handled to optimizing is classifying type variable, optimizing algorithm it is preferable to use particle swarm algorithm,For example, discrete particle cluster algorithm.
In addition, in another example of the disclosure, to each decision feature in constructed decision characteristic variable combinationIt may include: in predetermined decision variable value range, to constructed decision spy that the variable-value of variable, which carries out optimizing processing,The variable-value for levying each decision characteristic variable in variable combination carries out optimizing processing.In addition, in order to enable optimizing effect moreIt is good, it can carry out properly to adjust the size of decision variable value range according to the actual situation.
After the variable-value for determining each decision characteristic variable in the combination of decision characteristic variable, in block 140, according toThe variable-value of each decision characteristic variable in the combination of decision characteristic variable obtained after optimizing is handled carries out decisionOptimization processing.For example, using the variable-value of each decision characteristic variable come to decision-making mechanism used in decision engineOptimize adjustment.
Fig. 2 shows the one of the prediction result contribution degree determination process of decision characteristic variable according to an embodiment of the present disclosureA exemplary flow chart.
As shown in Fig. 2, for each decision characteristic variable in prediction model, firstly, selecting a decision in block 210Characteristic variable is as initial current decision characteristic variable.For example, one can be randomly choosed from each decision characteristic variableDecision characteristic variable is as initial current decision characteristic variable.
Then, for the current decision characteristic variable, perfoming block 220 arrives the operation of block 270, until special for all decisionsSign variable all determines corresponding prediction result contribution degree.Specifically, in block 220, determine in the combination of decision characteristic variable in addition toPrediction result corresponding to any combination of remaining decision characteristic variable except the current decision characteristic variable.For example, it is assumed thatPrediction model is current decision characteristic variable there are 4 decision characteristic variables V1, V2, V3, V4 and V1, then using prediction mouldType is V2 the input that calculates in prediction model, in the case where the corresponding user characteristic data of the arbitrary characteristics combination of V3, V4Model prediction result.Then, in block 230, appointing comprising the current decision characteristic variable in the combination of decision characteristic variable is determinedThe corresponding prediction result of what characteristic variable combination.Characteristic variable combination described here is for every kind of feature in block 220Combination all increases the obtained characteristic variable combination of characteristic variable V1.Then, the model determined in block 240, calculation block 220Corresponding model prediction prediction result (model prediction result when the no current decision characteristic variable) and determined in block 230As a result the difference for (having the model prediction result when current decision characteristic variable).
After the difference for determining each model prediction result, in block 350, all model prediction result differences are carried out asking flat, to obtain the prediction result contribution degree of the current decision characteristic variable.Then, in block 260, it is determined whether there are untreatedDecision characteristic variable.If it is present selecting a decision feature to become from untreated decision characteristic variable in block 270Amount is used as next current decision characteristic variable, is then return to block 220, re-executes the operation that block 220 arrives block 270.
Fig. 3 shows an exemplary process of decision characteristic variable combination building process according to an embodiment of the present disclosureFigure.
As shown in figure 3, firstly, the prediction result based on each decision characteristic variable determined is contributed in block 310Degree, is ranked up each decision characteristic variable.Then, in block 320, from each decision characteristic variable after the sequenceThe middle biggish predetermined number decision characteristic variable of selection contribution degree is combined as decision characteristic variable to be optimized.Here, instituteStating predetermined number can be determined based on practical application scene or other appropraite conditions.In addition, in other examples of the disclosureIn, decision characteristic variable combination to be optimized can also be constructed using other suitable modes.
Come for different user by using the user characteristic data of the user using the decision optimization method of the disclosureThe prediction result contribution degree for determining each decision characteristic variable in prediction model, is constructed based on the prediction result determinedDecision characteristic variable to be optimized combines, and each decision in decision characteristic variable combination is determined using optimizing algorithmThe variable-value of characteristic variable.In the decision optimization method, due to available each for different user characteristic datasThe different prediction result contribution degrees of decision characteristic variable combine, thus needle so as to construct different decision characteristic variablesTo different users, the decision characteristic variable combination optimized also can be different, and then realize personalized decision optimization.
Using the decision optimization method of the disclosure, decision characteristic variable value model is properly adjusted by according to the actual situationThe size enclosed, and in decision characteristic variable value range, to each decision in constructed decision characteristic variable combinationThe variable-value of characteristic variable carries out optimizing processing, it is possible thereby to promote optimizing effect, and then improves decision optimization effect.
Fig. 4 shows an exemplary stream of the online handoff procedure of personalized decision-making mechanism according to an embodiment of the present disclosureCheng Tu.
As shown in figure 4, being determined online after as above determining optimized personalized decision-making mechanism and engineWhen plan, to the customer data to decision of inflow, which is randomly distributed to after former decision-making mechanism or optimizationProperty decision-making mechanism is handled, and then is made a policy, and record corresponding decision-making results.After at one section of such operation,The decision-making results of personalized decision-making mechanism after former decision-making mechanism or optimization is assessed, if the decision effect of former decision-making mechanismFruit is more preferable, then keeps former decision-making mechanism, if the personalized decision-making mechanism after optimization is preferable, is determined using the personalization after optimizationPlan mechanism replaces former decision-making mechanism, thereby guarantees that the stationarity of decision business.
Fig. 5 shows the block diagram of decision optimization device 500 according to an embodiment of the present disclosure.As shown in figure 5, decision is excellentIt includes contribution degree determination unit 510, decision characteristic variable combination construction unit 520,530 and of optimizing processing unit that makeup, which sets 500,Decision optimization unit 540.
Contribution degree determination unit 510 is configured to determine that each decision characteristic variable in prediction model in user characteristics numberPrediction result contribution degree under, the prediction model are created based on decision objective, and the prediction model includes decision spyLevy variable and non-decision characteristic variable.The operation of contribution degree determination unit 510 can be with reference to the block 110 described above with reference to Fig. 1Operation and referring to Fig. 2 describe operation.
Decision characteristic variable combination construction unit 520 is configured as based on the pre- of each decision characteristic variable determinedResult contribution degree is surveyed, decision characteristic variable combination to be optimized is constructed.The operation that decision characteristic variable combines construction unit 520 canWith reference to the operation above with reference to Fig. 1 block 120 described and the operation described referring to Fig. 3.
Optimizing processing unit 530 is configured as to each decision characteristic variable in constructed decision characteristic variable combinationVariable-value carry out optimizing processing so that the correspondence prediction result of the prediction model is best.One in the disclosure is shownIn example, optimizing processing unit 530 is configured as: using one of following optimizing algorithms come to constructed decision characteristic variableThe variable-value of each decision characteristic variable in combination carries out optimizing processing: particle swarm algorithm, genetic algorithm, annealing algorithm.In another example of the disclosure, optimizing processing unit 530 is configured as: in predetermined decision variable value range, to institute's structureThe variable-value for each decision characteristic variable in the combination of decision characteristic variable built carries out optimizing processing.Optimizing processing unit530 operation can be with reference to the operation above with reference to Fig. 1 block 130 described.
Decision optimization unit 540 is configured as according to each in the decision characteristic variable combination obtained after optimizing is handledThe variable-value of a decision characteristic variable carries out decision optimization processing.The operation of decision optimization unit 540 can be with reference to aboveThe operation of the block 140 described referring to Fig.1.
In an example of the disclosure, contribution degree determination unit 510 be can be configured as: be determined using interpretation modelPrediction result contribution degree of each decision characteristic variable under user characteristic data in prediction model.The interpretation model can be withIncluding one of following interpretation models: Shap value model, LIME model and DeepLift model.
Fig. 6 show decision characteristic variable according to an embodiment of the present disclosure combination construction unit 520 one is exemplaryBlock diagram.As shown in fig. 6, decision characteristic variable combination construction unit 520 includes sorting module 521 and feature selection module 523.
Sorting module 521 is configured as the prediction result contribution degree based on each decision characteristic variable determined, rightEach decision characteristic variable is ranked up.
Feature selection module 523 is configured as selecting contribution degree larger from each decision characteristic variable after the sequencePredetermined number decision characteristic variable, combined as decision characteristic variable to be optimized.
Above with reference to Fig. 1 to Fig. 6, the embodiment of the decision optimization method and device according to the disclosure is described.Decision optimization device above can use hardware realization, can also be realized using the combination of software or hardware and software.
Fig. 7 shows the hardware structure diagram of the calculating equipment 700 according to an embodiment of the present disclosure for decision optimization.Such asShown in Fig. 7, calculating equipment 700 may include at least one processor 710, memory 720, memory 730 and communication interface 740,And at least one processor 710, memory 720, memory 730 and communication interface 740 link together via bus 760.ExtremelyA few processor 710 executes at least one computer-readable instruction for storing or encoding in memory (that is, above-mentioned with softwareThe element that form is realized).
In one embodiment, computer executable instructions are stored in memory, make at least one when implementedProcessor 710: prediction result contribution degree of each decision characteristic variable under user characteristic data in prediction model, institute are determinedStating prediction model is created based on decision objective, and the prediction model includes decision characteristic variable and non-decision characteristic variable；Based on the prediction result contribution degree for each decision characteristic variable determined, decision characteristic variable combination to be optimized is constructed；Optimizing processing is carried out to the variable-value of each decision characteristic variable in constructed decision characteristic variable combination, so that instituteThe correspondence prediction result for stating prediction model is best；And according in the decision characteristic variable combination obtained after optimizing is handledThe variable-value of each decision characteristic variable carries out decision optimization processing.
It should be understood that the computer executable instructions stored in memory make at least one processor when implemented710 carry out the above various operations and functions described in conjunction with Fig. 1-6 in each embodiment of the disclosure.
In the disclosure, calculating equipment 700 can include but is not limited to: personal computer, server computer, workIt stands, desktop computer, laptop computer, notebook computer, mobile computing device, smart phone, tablet computer, beeCellular telephone, personal digital assistant (PDA), hand-held device, messaging devices, wearable calculating equipment, consumer-elcetronics devices etc.Deng.
According to one embodiment, a kind of program product of such as non-transitory machine readable media is provided.Non-transitoryMachine readable media can have instruction (that is, above-mentioned element realized in a software form), which when executed by a machine, makesIt obtains machine and executes the above various operations and functions described in conjunction with Fig. 1-6 in each embodiment of the disclosure.Specifically, Ke YitiFor being furnished with the system or device of readable storage medium storing program for executing, store on the readable storage medium storing program for executing any in realization above-described embodimentThe software program code of the function of embodiment, and read and execute the computer of the system or device or processor and be stored inInstruction in the readable storage medium storing program for executing.
According to one embodiment, a kind of program product of such as non-transitory machine readable media is provided.Non-transitoryMachine readable media can have instruction (that is, above-mentioned element realized in a software form), which when executed by a machine, makesIt obtains machine and executes the above various operations and functions described in conjunction with Fig. 1-6 in each embodiment of the disclosure.Specifically, Ke YitiFor being furnished with the system or device of readable storage medium storing program for executing, store on the readable storage medium storing program for executing any in realization above-described embodimentThe software program code of the function of embodiment, and read and execute the computer of the system or device or processor and be stored inInstruction in the readable storage medium storing program for executing.
In this case, it is real that any one of above-described embodiment can be achieved in the program code itself read from readable mediumThe function of example is applied, therefore the readable storage medium storing program for executing of machine readable code and storage machine readable code constitutes of the invention onePoint.
The embodiment of readable storage medium storing program for executing include floppy disk, hard disk, magneto-optic disk, CD (such as CD-ROM, CD-R, CD-RW,DVD-ROM, DVD-RAM, DVD-RW, DVD-RW), tape, non-volatile memory card and ROM.It selectively, can be by communication networkNetwork download program code from server computer or on cloud.
It will be appreciated by those skilled in the art that each embodiment disclosed above can be in the situation without departing from invention essenceUnder make various changes and modifications.Therefore, protection scope of the present invention should be defined by the appended claims.
It should be noted that step and unit not all in above-mentioned each process and each system construction drawing is all necessary, certain step or units can be ignored according to the actual needs.Each step execution sequence be not it is fixed, can be according to needIt is determined.Apparatus structure described in the various embodiments described above can be physical structure, be also possible to logical construction, that is, haveA little units may be realized by same physical entity, be realized alternatively, some units may divide by multiple physical entities, alternatively, can be withIt is realized jointly by certain components in multiple autonomous devices.
In the above various embodiments, hardware cell or module mechanically or can be realized electrically.For example, oneHardware cell, module or processor may include permanent dedicated circuit or logic (such as special processor, FPGA orASIC) corresponding operating is completed.Hardware cell or processor can also include programmable logic or circuit (such as general processor orOther programmable processors), interim setting can be carried out by software to complete corresponding operating.Concrete implementation mode is (mechanicalMode or dedicated permanent circuit or the circuit being temporarily arranged) it can be determined based on cost and temporal consideration.
The specific embodiment illustrated above in conjunction with attached drawing describes exemplary embodiment, it is not intended that may be implementedOr fall into all embodiments of the protection scope of claims." exemplary " meaning of the term used in entire this specificationTaste " be used as example, example or illustration ", be not meant to than other embodiments " preferably " or " there is advantage ".For offer pairThe purpose of the understanding of described technology, specific embodiment include detail.However, it is possible in these no detailsIn the case of implement these technologies.In some instances, public in order to avoid the concept to described embodiment causes indigestionThe construction and device known is shown in block diagram form.
The foregoing description of present disclosure is provided so that any those of ordinary skill in this field can be realized or makeUse present disclosure.To those skilled in the art, the various modifications carried out to present disclosure are apparent, also, can also answer generic principles defined herein in the case where not departing from the protection scope of present disclosureFor other modifications.Therefore, present disclosure is not limited to examples described herein and design, but disclosed herein with meetingPrinciple and novel features widest scope it is consistent.