CN112785005B

CN112785005B - Multi-objective task assistant decision-making method and device, computer equipment and medium

Info

Publication number: CN112785005B
Application number: CN202110090631.7A
Authority: CN
Inventors: 张跃; 张浩然
Original assignee: Ping An Life Insurance Company of China Ltd
Current assignee: Ping An Life Insurance Company of China Ltd
Priority date: 2021-01-22
Filing date: 2021-01-22
Publication date: 2023-02-03
Anticipated expiration: 2041-01-22
Also published as: CN112785005A

Abstract

The invention relates to the technical field of computers, and discloses an auxiliary decision-making method, an auxiliary decision-making device, computer equipment and a medium for multi-objective tasks, wherein the method comprises the following steps: aiming at each basic task, respectively establishing a plurality of initial xgboost models for training to obtain a plurality of target xgboost models corresponding to each basic task, inputting sample data into the target xgboost models for training to obtain a plurality of groups of training data, splicing the training data to obtain an evaluation data set, generating a weight set of the evaluation data set by adopting a genetic algorithm, fusing AUCs (central authority) of all basic tasks according to the weight set to obtain a plurality of weight vectors, generating a visual pareto chart according to the weight vectors, scoring the sample data by adopting the target weights as target weights to obtain target scores, and carrying out multi-task decision based on the target scores.

Description

Multi-target task assistant decision-making method and device, computer equipment and medium

Technical Field

The invention relates to the technical field of computers, in particular to an auxiliary decision-making method, an auxiliary decision-making device, computer equipment and a medium for multi-objective tasks.

Background

In the fields of finance, e-commerce and HR management, an AI model is usually used for carrying out prediction scoring and hierarchical management and control on customers and staff. For example, in the situation of increasing members of insurance agents, models are established according to basic information of quasi-agents and various expressions before posts, and the agents submitting applications of the insurance agents are subjected to model scoring, so that the models are expected to have strong prediction capability on various expressions (3 turns, 6 stays, 9 stays, 13 stays, FYP premium, low performance and the like) after posts of the agents, and therefore after the models are scored and layered, policy encouragement (talent and talent dive) can be performed on the high-scoring crowds, the low-scoring crowds (terminal 1% -5% of crowds) are eliminated, the overall level of the entering-department crowds is improved, the training cost is saved for companies, and the income is increased. Due to the fact that problems of model online, threshold value division mechanism demarcation and the like are involved, it is expected that one score of an entrance person can meet the requirement that the score has better performance on a plurality of business targets, and therefore the score is a multi-objective decision problem.

One existing solution is: many learning algorithms are less effective by balancing limited resources in a single learning algorithm to satisfy multi-task learning, and balancing multiple tasks. For example, in the learning process, the reward values of some tasks are large, so that the algorithm focuses on the tasks with prominent reward values at the cost of sacrificing generality, and other tasks cannot achieve good effects; there are algorithms that unify the value of the rewards of the individual tasks by way of reward reduction, which may change the optimization objective, and if the reward values are large non-negative values, the reduction becomes to optimize the frequency of awards rather than accumulating the expected rewards. And the balance of the algorithm among tasks depends not only on the size of the reward value, but also on the reward density, the reward reduction still causes the imbalance of the algorithm among different tasks.

Another solution is called distillation-based learning: mainly, a student network is constructed through an expert network which has supervision and learns a plurality of specific tasks, the learning algorithm provides a result of multi-task strategy compromise, and each expert network needs to be obtained by large-scale training in advance. Although the learning algorithm avoids the problem of unbalanced reward values, the learning algorithm is still balanced among a plurality of tasks, the learning effect is not ideal, and the performance of the learning algorithm is limited by the expert network and cannot be further improved.

Therefore, a multitask aided decision method with high accuracy is needed.

Disclosure of Invention

The embodiment of the invention provides an assistant decision-making method and device for multi-objective tasks, computer equipment and a storage medium, and aims to improve the accuracy of assistant decision-making.

In order to solve the foregoing technical problem, an embodiment of the present application provides a multi-objective task aided decision method, including:

respectively establishing a plurality of initial xgboost models for each basic task, acquiring data characteristics of sample data, and training the initial xgboost models by adopting the data characteristics to obtain a plurality of target xgboost models corresponding to each basic task;

acquiring M sample data aiming at any basic task, inputting the sample data into the target xgboost model for training to obtain a plurality of groups of training data, wherein each group of training data is a training output result corresponding to the M sample data, and each sample data is trained to obtain a training output result;

performing data splicing on the plurality of groups of training data of each basic task to obtain an evaluation data set;

generating an initial weight by adopting a preset generation mode, performing fusion scoring on the evaluation data set through the initial weight to obtain a scoring result, and updating and iterating the initial weight according to the scoring result and a genetic algorithm to generate a weight set of the evaluation data set;

fusing AUCs of all basic tasks according to the weight set to obtain a plurality of weight vectors;

generating a visual pareto chart according to the weight vector, and acquiring a pareto boundary from the visual pareto chart;

and taking the weight corresponding to the pareto boundary as a target weight, scoring the sample data by adopting the target weight to obtain a target score, and performing multi-task decision based on the target score.

Optionally, the obtaining data characteristics of the sample data includes:

obtaining sample data, and classifying the basic data according to a preset label type to obtain initial class information;

carrying out missing value processing on each initial category information to obtain basic category information, wherein each basic category information at least comprises one basic feature;

and calculating the stability of the basic features of each basic category information, and screening out the basic features with the stability exceeding a preset stability threshold value as the data features.

Optionally, the calculating the stability of the basic feature of each basic category information includes:

calculating an information value IV of each basic characteristic, and performing characteristic screening according to the information value IV to obtain a key characteristic;

and calculating the stability index PSI of the key features, and taking the key features of which the stability index PSI exceeds a preset threshold value as the stability features.

Optionally, the generating an initial weight by using a preset generating manner, and performing fusion scoring on the evaluation data set through the initial weight to obtain a scoring result includes:

generating N groups of initial weights in a normal random number generation mode, wherein each group of initial weights comprises N initial weights, and N is a preset positive integer;

and weighting the evaluation data set by adopting each group of initial weights, counting the accuracy of the weighted evaluation data set, and taking the accuracy as the scoring result.

Optionally, the performing update iteration on the initial weights according to the scoring result and a genetic algorithm, and generating the weight set of the evaluation data set includes:

according to the accuracy of the scoring result, selecting an initial weight set with the accuracy reaching a preset threshold value from N sets of initial weights as a weight set to be selected;

performing cross variation on the weight set to be selected to obtain an updated initial weight;

and weighting and scoring the evaluation data set by adopting the updated initial weight, counting the accuracy of the evaluation data set after weighting to be used as a scoring result, returning to the accuracy of the scoring result, selecting an initial weight set with the accuracy reaching a preset threshold value from N groups of initial weights, continuously performing iteration as a step of a weight set to be selected until the iteration times reach the preset times, and using the weight set to be selected obtained at the moment as the weight set.

Optionally, the generating a visualized pareto map according to the weight vector includes:

traversing and combining the weight vectors in a mode of self-defining a granularity threshold to obtain at least two groups of combined weights;

and generating a visualized pareto chart by adopting the combined weight.

In order to solve the above technical problem, an embodiment of the present application further provides a multi-objective task assistant decision device, including:

the model training module is used for respectively establishing a plurality of initial xgboost models for each basic task, acquiring data characteristics of sample data, and training the initial xgboost models by adopting the data characteristics to obtain a plurality of target xgboost models corresponding to each basic task;

the sample training module is used for acquiring M sample data aiming at any basic task, inputting the sample data into the target xgboost model for training to obtain a plurality of groups of training data, wherein each group of training data is a training output result corresponding to the M sample data, and each sample data is trained to obtain a training output result;

the data splicing module is used for carrying out data splicing on the plurality of groups of training data of each basic task to obtain an evaluation data set;

the weight alternation module is used for generating initial weights by adopting a preset generation mode, performing fusion scoring on the evaluation data set through the initial weights to obtain scoring results, and updating and iterating the initial weights according to the scoring results and a genetic algorithm to generate a weight set of the evaluation data set;

the vector generation module is used for fusing AUCs of all basic tasks according to the weight set to obtain a plurality of weight vectors;

the boundary determining module is used for generating a visual pareto chart according to the weight vector and acquiring a pareto boundary from the visual pareto chart;

and the auxiliary decision module is used for taking the weight corresponding to the pareto boundary as a target weight, adopting the target weight to score the sample data to obtain a target score, and carrying out multi-task decision based on the target score.

Optionally, the model training module comprises:

the data classification unit is used for acquiring sample data and classifying the basic data according to a preset label type to obtain initial class information;

the information processing unit is used for carrying out missing value processing on each initial category information to obtain basic category information, and each basic category information at least comprises a basic feature;

and the characteristic screening unit is used for calculating the stability of the basic characteristic of each basic category information and screening out the basic characteristic of which the stability exceeds a preset stability threshold value as the data characteristic.

Optionally, the feature filtering unit includes:

the key feature extraction subunit is used for calculating an information value IV of each basic feature and performing feature screening according to the information value IV to obtain key features;

and the stable characteristic determining subunit is used for calculating the stability index PSI of the key characteristic, and taking the key characteristic of which the stability index PSI exceeds a preset threshold value as the stable characteristic.

Optionally, the weight alternation module comprises:

the initial weight generating unit is used for generating N groups of initial weights in a normal random number generating mode, wherein each group of initial weights comprises N initial weights, and N is a preset positive integer;

and the scoring unit is used for weighting the evaluation data set by adopting each group of initial weights, counting the accuracy of the weighted evaluation data set, and taking the accuracy as the scoring result.

Optionally, the weight alternation module further comprises:

the weight screening unit is used for selecting an initial weight set with the accuracy reaching a preset threshold value from N groups of initial weights according to the accuracy of the scoring result, and taking the initial weight set as a weight set to be selected;

a cross mutation unit, configured to perform cross mutation on the to-be-selected weight set to obtain an updated initial weight;

and the weight iteration unit is used for weighting and scoring the evaluation data set by adopting the updated initial weight, counting the accuracy of the weighted evaluation data set as a scoring result, returning to the step of selecting an initial weight set with the accuracy reaching a preset threshold from N groups of initial weights, continuously performing iteration as the step of selecting the weight set to be selected until the iteration times reach the preset times, and using the weight set to be selected obtained at the moment as the weight set.

Optionally, the boundary determining module includes:

the traversing combination unit is used for traversing and combining the weight vectors in a mode of self-defining a granularity threshold value to obtain at least two groups of combination weights;

and the pareto map generation unit is used for generating a visualized pareto map by adopting the combined weight.

In order to solve the above technical problem, an embodiment of the present application further provides a computer device, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor, where the processor implements the steps of the multi-objective task assistant decision method when executing the computer program.

In order to solve the above technical problem, an embodiment of the present application further provides a computer-readable storage medium, where a computer program is stored, and the computer program, when executed by a processor, implements the steps of the multi-objective task assistant decision method.

The embodiment of the invention provides an auxiliary decision method, an auxiliary decision device, computer equipment and a storage medium for multi-objective tasks, wherein a plurality of initial xgboost models are respectively established for each basic task to be trained, a plurality of target xgboost models corresponding to each basic task are obtained, M sample data are obtained for any basic task, the sample data are input into the target xgboost models to be trained, a plurality of groups of training data are obtained, a plurality of groups of training data of each basic task are subjected to data splicing to obtain an evaluation data set, an initial weight is generated in a preset generation mode, the evaluation data set is subjected to fusion scoring through the initial weight to obtain a scoring result, the initial weight is subjected to updating iteration according to the scoring result and a genetic algorithm to generate a weight set of the evaluation data set, AUC of all basic tasks are fused according to the weight set to obtain a plurality of weight vectors, a visual cumulative weight graph is generated according to the weight vectors, a cumulative weight is visually obtained from the cumulative weight graph, the weight corresponding to the target weight, the target weights are further subjected to the multiple target scores to obtain a plurality of auxiliary decision, and a plurality of target sample data are obtained based on the auxiliary decision, and the task.

Drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings needed to be used in the description of the embodiments of the present invention will be briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art that other drawings can be obtained according to these drawings without inventive labor.

FIG. 1 is an exemplary system architecture diagram in which the present application may be applied;

FIG. 2 is a flow diagram of one embodiment of a multi-objective task aided decision method of the present application;

FIG. 3 is a block diagram illustrating an embodiment of a multi-objective task aid decision device according to the present application;

FIG. 4 is a schematic block diagram of one embodiment of a computer device according to the present application.

Detailed Description

Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs; the terminology used in the description of the application herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the application; the terms "including" and "having," and any variations thereof, in the description and claims of this application and the description of the above figures are intended to cover non-exclusive inclusions. The terms "first," "second," and the like in the description and claims of this application or in the above-described drawings are used for distinguishing between different objects and not for describing a particular order.

Reference herein to "an embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment can be included in at least one embodiment of the application. The appearances of the phrase in various places in the specification are not necessarily all referring to the same embodiment, nor are separate or alternative embodiments mutually exclusive of other embodiments. It is explicitly and implicitly understood by one skilled in the art that the embodiments described herein can be combined with other embodiments.

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, not all, embodiments of the present invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

Referring to fig. 1, as shown in fig. 1, a system architecture 100 may include

terminal devices

101, 102, 103, a network 104, and a server 105. The network 104 serves as a medium for providing communication links between the

terminal devices

101, 102, 103 and the server 105. Network 104 may include various connection types, such as wired, wireless communication links, or fiber optic cables, to name a few.

A user may use

terminal devices

101, 102, 103 to interact with a server 105 over a network 104 to receive or send messages or the like.

The

terminal devices

101, 102, 103 may be various electronic devices having display screens and supporting web browsing, including but not limited to smart phones, tablet computers, E-book readers, MP3 players (Moving Picture E interface displays the properties Group Audio Layer III, mpeg compression standard Audio Layer 3), MP4 players (Moving Picture E interface displays the properties Group Audio Layer IV, mpeg compression standard Audio Layer 4), laptop portable computers, desktop computers, and the like.

The server 105 may be a server providing various services, such as a background server providing support for pages displayed on the

terminal devices

101, 102, 103.

It should be noted that, the multi-objective task assistant decision method provided in the embodiments of the present application is executed by a server, and accordingly, an assistant decision device for a multi-objective task is disposed in the server.

It should be understood that the number of terminal devices, networks, and servers in fig. 1 is merely illustrative. Any number of terminal devices, networks and servers may be provided according to implementation needs, and the

terminal devices

101, 102 and 103 in this embodiment may specifically correspond to an application system in actual production.

Referring to fig. 2, fig. 2 shows an assistant decision method for multi-objective tasks according to an embodiment of the present invention, which is described by taking the application of the method to the server in fig. 1 as an example, and is detailed as follows:

s201: and aiming at each basic task, respectively establishing a plurality of initial xgboost models, acquiring data characteristics of sample data, and training the initial xgboost models by adopting the data characteristics to obtain a plurality of target xgboost models corresponding to each basic task.

Specifically, for each basic task, according to the self requirement of the basic task, a plurality of initial xgboost models are constructed, data features are extracted from sample data, the extracted data features are input into the initial xgboost models, the initial xgboost models are trained, and a plurality of target xgboost models corresponding to each basic task are obtained.

It should be noted that different basic tasks have different initial xgboost models according to their requirements, for example, in an agent adding scene, the basic task is three turns, 5 initial xgboost models are constructed according to the requirements, the basic task is thirteen reservations, 7 initial xgboost models are constructed according to the requirements, and the initial xgboost models are the xgboost models with default parameters.

It should be noted that, in the present embodiment, each sample involves multiple basic tasks, and thus a decision problem of multiple basic tasks needs to be implemented.

Xgboost is a distributed efficient gradient lifting algorithm based on a decision tree (CART), a boosting idea is adopted, a next base classifier is learned by fitting a negative gradient of an error between a previous base Classifier (CART) and a target value, XGboost is a serial generation CART tree, but the XGboost can perform parallel processing when processing features, the XGboost parallel principle is embodied in selection of target dividing points, the initial data of a certain sub-model is assumed to have L monitoring features, and in the construction process of a certain round of CART tree: firstly, sequencing training data (initial data participating in model training) according to the characteristic value of each monitoring characteristic, and storing the training data into block structures (for example, L block structures, wherein the number of the block structures is the same as that of the monitoring characteristics); then, selecting an optimal feature segmentation point for each block structure, wherein the node segmentation standard is that the point with the maximum target function drop is selected as a segmentation point (the optimal segmentation point of each monitoring feature is selected) according to the drop degree of the target function; and finally, comparing the target function reduction gains of the optimal characteristic segmentation points of each block structure, and selecting the optimal segmentation points.

The dividing point algorithm of each block structure adopts a greedy algorithm, starting from a root node, one monitoring feature and a corresponding feature value (feature data) are selected each time, so that loss functions are reduced most, and the selected monitoring features are used as dividing nodes. And sequencing the initial data according to the characteristic values, segmenting the initial data from small to large according to the characteristic values of the monitoring characteristics, comparing the size of the target function segmented each time, and selecting the node with the largest reduction as a target segmentation point of the monitoring characteristics. And finally, comparing the target function reduction values of the target segmentation points with different block structures, and selecting the characteristic value with the maximum reduction as the target segmentation point. The XGboost enables the learned model to be simpler, overfitting is prevented, the application capability and the interpretation capability of the model are improved, in addition, the XGBoost preprocesses data before training, the features are sorted in advance and stored as a block structure, and the structure is repeatedly used in the following iteration, so that the calculation amount is greatly reduced.

The termination condition of the iterative training may be a preset number of times, or a loss lower than a preset threshold, and may be specifically set according to an actual situation, which is not limited herein.

S202: and acquiring M sample data aiming at any basic task, inputting the sample data into a target xgboost model for training to obtain a plurality of groups of training data, wherein each group of training data is a training output result corresponding to the M sample data, and each sample data is trained to obtain a training output result.

Specifically, for any one basic task, M sample data are obtained, and the M sample data are input into each target xgboost model corresponding to the basic task for training to obtain a training result input by the target xgboost model.

Where M is the number of preset sample data, in this embodiment, as a preferable mode, the value of M is 1000.

In this embodiment, each basic task is trained by using sample data to obtain a training output result, and the training output results are subsequently adopted to participate in fusion scoring, so that each simulated training result is reflected, that is, each basic task involved is considered in scoring, and the accuracy of multi-objective decision is improved.

S203: and carrying out data splicing on a plurality of groups of training data of each basic task to obtain an evaluation data set.

Specifically, for each basic task, training data of the basic task are spliced to obtain an evaluation data set.

In a specific embodiment, a basic task a takes sample data of 8, 9 and 10 months in 2019 as a training set, predicts a business result of 1 month in 2020, a basic task B takes 9, 10 and 11 months in 2019 as a training set, predicts a business result of 2 months in 2020, a basic task C takes 10, 11 and 12 months in 2019 as a training set, predicts a business result of 3 months in 2020, establishes 7 sub-models for the basic task a, scores the sample data of 1 month in 2020 by using each sub-model, and assumes that the sample data of each month is 10 ten thousand individual data and the size of the 7 scored data sets is 10 ten thousand 7 columns. Similarly, for base task B and base task C, two data sets of 10 ten thousand by 7 columns were also obtained. Splicing the three data sets by rows yields a data set of 30 ten thousand by 7 columns, with the column names: [ s1, s2, s3, s4, s5, s6, s7].

S204: and generating an initial weight by adopting a preset generation mode, performing fusion scoring on the evaluation data set through the initial weight to obtain a scoring result, updating and iterating the initial weight according to the scoring result and a genetic algorithm, and generating a weight set of the evaluation data set.

Among them, genetic Algorithm (GA) is a method of searching a target solution by simulating a natural evolution process. The algorithm converts the solving process of the problem into the processes of crossover, variation and the like of chromosome genes in the similar biological evolution by a mathematical mode and by utilizing computer simulation operation. When a complex combined optimization problem is solved, a better optimization result can be obtained faster compared with some conventional optimization algorithms. In this embodiment, a genetic algorithm is used to perform cross variation on the initial weight with higher accuracy to obtain a new population, and then iterative updating is performed to obtain a weight set with higher scoring accuracy.

Specifically, according to the scoring result and the genetic algorithm, the initial weights are updated and iterated to generate the weight set of the evaluation data set, and reference may be made to the description of the subsequent embodiments, and details are not repeated here in order to avoid repetition.

S205: and fusing AUCs of all basic tasks according to the weight set to obtain a plurality of weight vectors.

Specifically, AUC fusion is performed on all basic tasks by adopting each group of weights in the weight set to obtain weight vectors.

Where AUC (Area Under Curve) is defined as the Area enclosed by the coordinate axes Under the ROC Curve, it is obvious that the value of this Area is not larger than 1. Since ROC curves are generally located above the line y = x, the AUC ranges between 0.5 and 1. The closer the AUC is to 1.0, the higher the authenticity of the detection method is; and when the value is equal to 0.5, the authenticity is lowest, and the application value is not high.

For example, in a specific embodiment, there are two basic tasks, the labels of the basic tasks are respectively three turns (whether the turn is positive after 3 months) and 13 remains (whether the turn is retained after 13 months), the three turn target can be subdivided into 5 sub-targets, the 13 remain target can be subdivided into 7 sub-targets, 5+7=12 sub-targets need to be modeled separately, what needs to be optimized by the genetic algorithm is a weight vector with 12 dimensions, and for the processing of the genetic algorithm adaptability score: the AUCs of the two targets need to be fused with a certain weight, for example, in a [0.4,0.6] manner: (0.4 × three turns AUC +0.6 × 13 left AUC) defines the fitness score, so a target weight vector (12 dimensions) of 12 sub-models is obtained under a weight of [0.4,0.6 ].

S206: and generating a visual pareto chart according to the weight vector, and acquiring a pareto boundary from the visual pareto chart.

Specifically, any basic task is split into a plurality of sub-targets and is modeled independently, and then the basic task performance relative to baseline can be greatly improved through a fusion strategy. The basic task is single, and if a plurality of business targets exist, multi-target pareto optimization can be performed on the basis of multi-target balance analysis. For example, in an agent scene, if the score of the agent is expected to have a good prediction effect on three turns and 13, the adaptive score can be combined with a plurality of basic tasks in the process of evaluating the weight of the sub-model in the genetic algorithm, the genetic algorithm is guided to rapidly generate pareto boundaries of a plurality of target effects by adjusting the weights of the plurality of basic tasks, and then the pareto boundaries are generated. And selecting a weight combination on the pareto boundary as a final online scheme.

Optionally, the proposal can traverse ownership recombination of multiple basic tasks in a mode of self-defining the granularity threshold to obtain multiple groups of weights, and then generate a visual pareto chart through the obtained multiple groups of weights.

For example, in a business agent distillation scenario, for three turns and 13 to leave two basic tasks, traversing all combinations of two main targets with a granularity of 0.05 would result in about 20 sets of weights, and based on the 20 sets of weights to represent three turns and 13 to leave AUC, a pareto graph can be made, finding pareto boundaries, and then selecting decision points on the boundaries based on the basic tasks. It should be appreciated that in the case of computational resource abundance, if the primary target weight granule is adjusted to 0.01, then 100 weight vectors are obtained, and the pareto boundary is smoother.

S207: and taking the weight corresponding to the pareto boundary as a target weight, scoring the sample data by adopting the target weight to obtain a target score, and performing multi-task decision based on the target score.

Specifically, the weight corresponding to the pareto boundary is used as a target weight, the target weight is adopted to score the sample data to obtain a target score corresponding to each sample data, and multi-task decision is performed based on the target score, for example, which sample data are selected as target data meeting requirements.

In the embodiment, aiming at each basic task, a plurality of initial xgboost models are respectively established for training to obtain a plurality of target xgboost models corresponding to each basic task, M sample data are obtained aiming at any basic task, the sample data are input into the target xgboost models for training to obtain a plurality of groups of training data, the groups of training data of each basic task are subjected to data splicing to obtain an evaluation data set, an initial weight is generated in a preset generation mode, the evaluation data set is subjected to fusion scoring through the initial weight to obtain a scoring result, the initial weight is subjected to update iteration according to the scoring result and a genetic algorithm to generate a weight set of the evaluation data set, AUC (AUC) scoring is performed on all basic tasks according to the weight set to obtain a plurality of weight vectors, a visual pareto graph is generated according to the weight vectors, a pareto obtain pareto border from the visual pareto border, the weight corresponding to the pareto border is used as a target weight, the target weight is used for scoring the sample data to obtain a target score, a plurality of tasks based on the target score, a plurality of tasks is obtained, a plurality of tasks, a plurality of auxiliary decisions are quickly obtained, and the tasks have an auxiliary decision making effect on each basic task, and the auxiliary decision.

In some optional implementation manners of this embodiment, in step S201, acquiring data characteristics of the sample data includes:

and calculating the stability of the basic characteristics of each basic category information, and screening out the basic characteristics of which the stability exceeds a preset stability threshold value as data characteristics.

The preset tag type may be set according to an actual situation, for example, in a scenario of adding members to a service agent, the preset tag includes but is not limited to: professional experience, educational background, identity information, and personality bias, among others.

In order to ensure the stability of the features, the missing values are counted, the initial information categories with the missing values exceeding a preset proportion are removed, the initial information categories meeting the requirements are used as basic category information, and then the stability of each basic feature contained in the basic information categories is evaluated subsequently.

The stability of the basic features of the basic category information is mainly evaluated through the stability indicator PSI, which may specifically refer to the description of the subsequent embodiments, and is not repeated here to avoid repetition.

In the embodiment, basic category information is obtained by classifying and processing missing values of the sample data, and then stable data characteristics are obtained by screening the stability for training of a subsequent model, so that the accuracy of the obtained data characteristics is improved, and the efficiency and the accuracy of model training are improved.

In some optional implementations of the present embodiment, calculating the stability of the basic feature of each basic category information includes:

calculating an information value IV of each basic characteristic, and performing characteristic screening according to the information value IV to obtain key characteristics;

and calculating a stability index PSI of the key features, and taking the key features with the stability index PSI exceeding a preset threshold value as the stability features.

The preset mode may specifically be that data of the key features of each month is selected according to the dimension of time, and the stability index PSI of the key features is calculated month by month.

The Information Value (IV) is a measure of the Information amount of a variable.

Wherein the stability Index PSI (population stability Index) is used to assess the characteristic stability.

In this embodiment, the key features are obtained by calculating the information value of each piece of basic category information, and then the stability index of each key feature is calculated to obtain the stable features, so that the accuracy of the obtained stable features is improved.

In some optional implementation manners of this embodiment, in step S204, a preset generation manner is adopted to generate an initial weight, and fusion scoring is performed on the evaluation data set through the initial weight, and obtaining a scoring result includes:

and weighting the evaluation data set by adopting each group of initial weights, counting the accuracy of the evaluation data set after weighting, and taking the accuracy as a scoring result.

Specifically, N groups of initial weights are generated in a normal random number generation mode, the number N of the initial weights contained in each group of initial weights is the same as that of a target xgboost model corresponding to a basic task, and then fusion scoring is performed on the evaluation data set by adopting each group of initial weights to obtain a scoring result (adaptive score).

The fusion scoring can be achieved by weighting the evaluation data set by adopting the initial weight according to actual requirements.

In a specific embodiment, taking the example in step S30 as an example, 7 random numbers are generated by means of a normal random number generator for the basic task a: [ r1, r2, \ 8230;, r7], let sum = abs (r 1) + abs (r 2) + \ 8230, abs (r 7), the first generation of individuals was [ abs (r 1)/sum, abs (r 2)/sum, \ 8230;, abs (r 7)/sum ], which resulted in 1000 initial solutions, each solution being a vector of 7 weights, resulting in a weight matrix size of 1000. The evaluation dataset was further weighted to obtain a fusion score of 1000 solutions, i.e. comb _ score = score weight _ T (matrix multiplication). If score size is 30 ten thousand by 7, weight _Tis the transpose of weight, size is 7 by 1000, then comb score size is 30 ten thousand by 1000, i.e., 1 month to 3 months each person who enters the house has 1000 fusion scores. And for the person entering the department for 1 month, calculating TOP1% accuracy acc _ a according to the sorted fusion score, and obtaining acc _ B and acc _ C of the basic task B and the basic task C in the same way. The fitness score is then the mean of the TOP1% accuracy of the three scenarios: mean ([ acc _ a, acc _ b, acc _ c ]).

In the embodiment, by generating the random number and weighting the evaluation data set, the accuracy of the evaluation data set is statistically weighted, the accuracy of weight screening is improved, cross variation is favorably performed according to the screened weights in the follow-up process, and more accurate weight vectors are obtained.

In some optional implementations of this embodiment, in step S204, performing update iteration on the initial weights according to the scoring result and the genetic algorithm, and generating the weight set of the evaluation data set includes:

according to the accuracy of the scoring result, selecting an initial weight set with the accuracy reaching a preset threshold value from the N sets of initial weights as a weight set to be selected;

carrying out cross variation on the weight set to be selected to obtain an updated initial weight;

and adopting the updated initial weight to carry out weighting and scoring on the evaluation data set, counting the accuracy of the evaluation data set after weighting to be used as a scoring result, returning the accuracy according to the scoring result, selecting an initial weight set with the accuracy reaching a preset threshold value from the N groups of initial weights, continuously carrying out iteration execution as a step of a weight set to be selected until the iteration number reaches a preset number, and using the weight set to be selected obtained at the moment as a weight set.

Specifically, the genetic algorithm firstly measures and calculates the adaptability score of the initial solution according to the concept of elimination of superiority and inferiority, eliminates the poor-performance solution, retains the good-performance solution, and enables the good-performance solution to be crossed and varied to generate a new generation of population, and can quickly enter the generation and optimization stage of the high-value solution from the random search state through multiple cycle iterations, so that under the conditions of more sub-targets and large search subspace, the calculation resources are saved, and the model effect is improved. And finally, providing the weight vector for the submodel of the basic task to be scored to obtain a decision scheme compatible with a plurality of basic tasks.

The preset number of times may be set according to actual requirements, and as a preferable preference, the preset number of times in the embodiment is 500 times.

In this embodiment, the initial weights are updated and iterated through a genetic algorithm to generate a weight set of the evaluation data set, which is beneficial to improving the quality of the generated weight set, so that the pareto chart is generated through the weight set subsequently.

In some optional implementations of this embodiment, in step S206, generating a visualized pareto chart according to the weight vector includes:

and generating a visual pareto chart by adopting the combined weight.

Specifically, different weight vectors are subjected to traversal combination in a mode of self-defining a granularity threshold value to obtain a plurality of groups of combination weights, and then a visual pareto chart is generated through the combination weights.

For example, in the retorting scenario of the business agent, for three turns and 13, leaving two basic tasks, traversing all combinations of two main objectives with a granularity of 0.05 would result in about 20 sets of weights, and based on the 20 sets of weights leaving the performance of AUC in three turns and 13, a pareto map can be made, finding pareto boundaries, and then selecting decision points on the boundaries based on the basic tasks. It should be appreciated that in the case of computational resource abundance, if the primary target weight granule is adjusted to 0.01, then 100 weight vectors are obtained, and the pareto boundary is smoother.

In this embodiment, the weight vectors are subjected to traversal combination in a mode of self-defining a granularity threshold, and then the combined weight is adopted to generate a visual pareto chart, so that smoothness of the pareto chart is adjusted according to actual requirements, and the pareto chart is beneficial to improvement of quality of a pareto boundary.

It should be understood that, the sequence numbers of the steps in the foregoing embodiments do not imply an execution sequence, and the execution sequence of each process should be determined by its function and inherent logic, and should not constitute any limitation to the implementation process of the embodiments of the present invention.

Fig. 3 is a schematic block diagram of a multi-objective task assistant decision device corresponding to the multi-objective task assistant decision method of the above embodiment. As shown in fig. 3, the multi-objective task assistant decision device includes a model training module 31, a sample training module 32, a data splicing module 33, a weight alternation module 34, a vector generation module 35, a boundary determination module 36 and an assistant decision module 37. The detailed description of each functional module is as follows:

the model training module 31 is configured to respectively establish a plurality of initial xgboost models for each basic task, acquire data characteristics of sample data, train the initial xgboost models by using the data characteristics, and obtain a plurality of target xgboost models corresponding to each basic task;

the sample training module 32 is configured to acquire M sample data for any basic task, input the sample data into the target xgboost model, and perform training to obtain a plurality of sets of training data, where each set of training data is a training output result corresponding to the M sample data, and each sample data is trained to obtain a training output result;

the data splicing module 33 is configured to perform data splicing on a plurality of sets of training data of each basic task to obtain an evaluation data set;

the weight alternation module 34 is configured to generate an initial weight by using a preset generation manner, perform fusion scoring on the evaluation data set through the initial weight to obtain a scoring result, and perform update iteration on the initial weight according to the scoring result and a genetic algorithm to generate a weight set of the evaluation data set;

the vector generation module 35 is configured to fuse the AUCs of all the basic tasks according to the weight set to obtain a plurality of weight vectors;

a boundary determining module 36, configured to generate a visualized pareto chart according to the weight vector, and obtain a pareto boundary from the visualized pareto chart;

and the auxiliary decision module 37 is configured to take the weight corresponding to the pareto boundary as a target weight, score the sample data by using the target weight to obtain a target score, and perform multi-task decision based on the target score.

Optionally, the model training module 31 includes:

the data classification unit is used for acquiring sample data and classifying the basic data according to the type of a preset label to obtain initial class information;

the information processing unit is used for processing missing values of each initial category information to obtain basic category information, and each basic category information at least comprises one basic feature;

Optionally, the feature filtering unit includes:

and the stable characteristic determining subunit is used for calculating a stability index PSI of the key characteristic, and taking the key characteristic of which the stability index PSI exceeds a preset threshold value as the stable characteristic.

Optionally, the weight alternation module 34 includes:

and the scoring unit is used for weighting the evaluation data set by adopting each group of initial weights, counting the accuracy of the weighted evaluation data set, and taking the accuracy as a scoring result.

Optionally, the weight alternation module 34 further comprises:

the weight screening unit is used for selecting an initial weight set with the accuracy rate reaching a preset threshold value from the N groups of initial weights according to the accuracy rate of the scoring result, and taking the initial weight set as a weight set to be selected;

the cross mutation unit is used for carrying out cross mutation on the weight set to be selected to obtain an updated initial weight;

and the weight iteration unit is used for weighting and scoring the evaluation data set by adopting the updated initial weight, counting the accuracy of the evaluation data set after weighting, taking the accuracy as a scoring result, returning the accuracy according to the scoring result, selecting an initial weight set with the accuracy reaching a preset threshold value from the N groups of initial weights, continuously performing iteration as the step of the weight set to be selected until the iteration number reaches the preset number, and taking the weight set to be selected obtained at the moment as a weight set.

Optionally, the boundary determining module 35 includes:

the traversal combination unit is used for performing traversal combination on the weight vectors in a mode of self-defining the granularity threshold value to obtain at least two groups of combination weights;

and the pareto chart generation unit is used for generating a visual pareto chart by adopting the combined weight.

For specific limitations of the apparatus for assisting decision-making for multi-objective tasks, reference may be made to the above limitations of the method for assisting decision-making for multi-objective tasks, which are not described in detail herein. All or part of each module in the multi-target task assistant decision device can be realized by software, hardware and a combination thereof. The modules can be embedded in a hardware form or independent from a processor in the computer device, and can also be stored in a memory in the computer device in a software form, so that the processor can call and execute operations corresponding to the modules.

In order to solve the technical problem, an embodiment of the present application further provides a computer device. Referring to fig. 4, fig. 4 is a block diagram of a basic structure of a computer device according to the present embodiment.

The computer device 4 comprises a memory 41, a processor 42, a network interface 43 communicatively connected to each other via a system bus. It is noted that only the computer device 4 having the components connection memory 41, processor 42, network interface 43 is shown, but it is understood that not all of the shown components are required to be implemented, and that more or fewer components may be implemented instead. As will be understood by those skilled in the art, the computer device is a device capable of automatically performing numerical calculation and/or information processing according to instructions set or stored in advance, and the hardware thereof includes but is not limited to a microprocessor, an Application Specific Integrated Circuit (ASIC), a Programmable Gate Array (FPGA), a Digital Signal Processor (DSP), an embedded device, and the like.

The computer device can be a desktop computer, a notebook, a palm computer, a cloud server and other computing devices. The computer equipment can carry out man-machine interaction with a user through a keyboard, a mouse, a remote controller, a touch panel or voice control equipment and the like.

The memory 41 includes at least one type of readable storage medium including a flash memory, a hard disk, a multimedia card, a card-type memory (e.g., SD or D interface display memory, etc.), a Random Access Memory (RAM), a Static Random Access Memory (SRAM), a Read Only Memory (ROM), an Electrically Erasable Programmable Read Only Memory (EEPROM), a Programmable Read Only Memory (PROM), a magnetic memory, a magnetic disk, an optical disk, etc. In some embodiments, the memory 41 may be an internal storage unit of the computer device 4, such as a hard disk or a memory of the computer device 4. In other embodiments, the memory 41 may also be an external storage device of the computer device 4, such as a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) Card, a Flash memory Card (Flash Card), and the like, which are provided on the computer device 4. Of course, the memory 41 may also include both internal and external storage devices of the computer device 4. In this embodiment, the memory 41 is generally used for storing an operating system installed in the computer device 4 and various types of application software, such as program codes for controlling electronic files. Further, the memory 41 may also be used to temporarily store various types of data that have been output or are to be output.

The processor 42 may be a Central Processing Unit (CPU), controller, microcontroller, microprocessor, or other data Processing chip in some embodiments. The processor 42 is typically used to control the overall operation of the computer device 4. In this embodiment, the processor 42 is configured to execute the program code stored in the memory 41 or process data, such as program code for executing control of an electronic file.

The network interface 43 may comprise a wireless network interface or a wired network interface, and the network interface 43 is generally used for establishing communication connection between the computer device 4 and other electronic devices.

The present application further provides another embodiment, which is a computer-readable storage medium storing an interface display program, which is executable by at least one processor to cause the at least one processor to perform the steps of the multi-objective task assistant decision method as described above.

Through the description of the foregoing embodiments, it is clear to those skilled in the art that the method of the foregoing embodiments may be implemented by software plus a necessary general hardware platform, and certainly may also be implemented by hardware, but in many cases, the former is a better implementation. Based on such understanding, the technical solutions of the present application may be embodied in the form of a software product, which is stored in a storage medium (such as ROM/RAM, magnetic disk, optical disk) and includes instructions for enabling a terminal device (such as a mobile phone, a computer, a server, an air conditioner, or a network device) to execute the method according to the embodiments of the present application.

It is to be understood that the above-described embodiments are merely illustrative of some, but not restrictive, of the broad invention, and that the appended drawings illustrate preferred embodiments of the invention and do not limit the scope of the invention. This application is capable of embodiments in many different forms and is provided for the purpose of enabling a thorough understanding of the disclosure of the application. Although the present application has been described in detail with reference to the foregoing embodiments, it will be apparent to one skilled in the art that modifications can be made to the embodiments described in the foregoing detailed description, or equivalents can be substituted for some of the features described therein. All equivalent structures made by using the contents of the specification and the drawings of the present application are directly or indirectly applied to other related technical fields and are within the protection scope of the present application.

Claims

1. An assistant decision-making method for multi-objective tasks is characterized by comprising the following steps:

and taking the weight corresponding to the pareto boundary as a target weight, adopting the target weight to score the sample data to obtain a target score, and carrying out multi-task decision based on the target score.

2. The multi-objective task aided decision making method of claim 1, wherein the obtaining data characteristics of sample data comprises:

obtaining sample data, and classifying the sample data according to a preset label type to obtain initial class information;

carrying out missing value processing on each initial category information to obtain basic category information, wherein each basic category information at least comprises a basic characteristic;

3. The multi-objective task aided decision making method of claim 2, wherein the calculating the stability of the base features of each base category information comprises:

4. The multi-objective task aided decision making method according to claim 1, wherein the generating of the initial weights by using a preset generating manner, and the fusion scoring of the evaluation data sets by the initial weights to obtain scoring results comprises:

5. The multi-objective task aided decision making method of claim 1, wherein the updating iteration of the initial weights according to the scoring results and a genetic algorithm to generate the set of weights of the evaluation dataset comprises:

and weighting and scoring the evaluation data set by adopting the updated initial weight, counting the accuracy of the evaluation data set after weighting to be used as a scoring result, returning the accuracy according to the scoring result, selecting an initial weight set with the accuracy reaching a preset threshold value from the N groups of initial weights, continuously performing iteration as a step of a weight set to be selected until the iteration times reach the preset times, and using the weight set to be selected obtained at the moment as the weight set.

6. An aid decision method for multi-objective tasks according to any one of claims 1 to 5, wherein the generating of visualized pareto maps from the weight vectors comprises:

and generating the visualized pareto chart by adopting the combined weight.

7. An aid-decision device for multi-objective tasks, comprising:

8. The multi-objective task aided decision making apparatus of claim 7, wherein the model training module comprises:

the data classification unit is used for acquiring sample data and classifying the sample data according to a preset label type to obtain initial class information;

9. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements a multi-objective task aid decision method as claimed in any one of claims 1 to 6 when executing the computer program.

10. A computer-readable storage medium, in which a computer program is stored which, when being executed by a processor, carries out a method for multi-objective task aided decision making as claimed in any one of claims 1 to 6.