Method, device and equipment for automatically mining strategy based on multidimensional data driving and medium

Abstract

The application discloses a multi-dimensional data driving-based strategy automatic mining method and device, equipment and medium, and relates to the financial strategy mining field.The application can improve the automation degree of strategy mining.

Description

Technical Field

[0001] This invention relates to the field of financial strategy mining, and in particular to a method, apparatus, device, and medium for automatic strategy mining based on multidimensional data-driven approaches. Background Technology

[0002] Financial strategy mining refers to the process of identifying and extracting patterns, rules, or strategies from large amounts of financial data and information using technologies such as data analysis, machine learning, and natural language processing. These strategies can help investors or financial institutions make better decisions. Strategies can be used to guide investment behavior, optimize risk management, improve business processes, and enhance customer service.

[0003] Traditional data mining methods typically require manual parameter setting and feature selection. This process not only consumes significant human and time resources but is also highly susceptible to bias due to subjective differences in human perception. Furthermore, the efficiency and accuracy of traditional methods are often severely limited when dealing with complex, multidimensional data. Therefore, a new strategy for automated data mining is urgently needed. Summary of the Invention

[0004] This invention solves the technical problem of low efficiency in strategy mining in the prior art by providing a method, apparatus, device and medium for automatic strategy mining based on multidimensional data driving, and achieves the technical effect of improving the efficiency of strategy mining.

[0005] In a first aspect, the present invention provides a method for automatic strategy mining based on multidimensional data-driven approaches, the method comprising:

[0006] Acquire feature datasets, label datasets, evaluation indicator datasets, and actual decision result datasets in the financial field within a target time period;

[0007] The feature dataset, label dataset, evaluation index dataset, and actual decision result dataset are concatenated to obtain a concatenated dataset, which is then divided into a training set, a test set, and a validation set.

[0008] Determine the objective function, which is used to determine the optimization direction and optimization objective;

[0009] Construct several decision trees based on the training set, and determine the hit rate and recall rate of each branch in each decision tree, where each branch corresponds to a strategy.

[0010] Based on the preset hit rate threshold, the hit rate and recall rate corresponding to each branch, the corresponding branches are filtered to obtain several optimal branches, and the data of the optimal branches in the training set are removed to obtain the target policy set.

[0011] The target policy set is evaluated and filtered based on the test set and validation set;

[0012] Based on the spliced ​​dataset, determine the weights of several pre-set evaluation indicators;

[0013] Based on several preset evaluation indicators and the corresponding weights of each preset evaluation indicator, the score of each strategy in the target strategy set is determined, and the optimal strategy is obtained.

[0014] Furthermore, based on a preset hit rate threshold, the hit rate and recall rate corresponding to each branch, the corresponding branches are filtered to obtain several optimal branches, and the data of the optimal branches in the training set are removed to obtain the target policy set, including steps S151-S152:

[0015] Step S151: Based on the preset hit rate threshold, the hit rate and recall rate corresponding to each branch, filter the corresponding branches and determine whether the branch is the optimal branch.

[0016] Step S152: If the branch is the optimal branch, remove the data of the branch in the training set and include the branch in the target policy set; if it is not the optimal branch, repeat step S151 with other branches.

[0017] Furthermore, based on a preset hit rate threshold and the hit rate and recall rate corresponding to each branch, the corresponding branches are filtered, including:

[0018]

[0019] In this function, SelectBranches(T,θ,N) is the selection function, T is the decision tree, b and b′ are branches in the decision tree T, HitRate(b) is the hit rate of branch (b), Recall(b) is the recall rate of branch (b), Recall(b′) is the recall rate of branch (b′), θ is the preset hit rate threshold, and N is the number of branches.

[0020] Furthermore, the objective function includes:

[0021] ObjectiveFunction=maxR(Parameters)

[0022] HitRate<θ

[0023] Where ObjectiveFunction is the objective function, R(Parameters) is the recall rate as a function of the strategy, Parameters is the strategy set, HitRate is the hit rate, and θ is the preset hit rate threshold.

[0024] Furthermore, the concatenated dataset is divided into a training set, a test set, and a validation set, including:

[0025] The target time period is divided into a first time period and a second time period, wherein the length of the first time period is longer than the length of the second time period and the first time period precedes the second time period.

[0026] The spliced ​​data in the first time period is divided into training set and test set according to a preset ratio;

[0027] The spliced ​​data in the second time period is divided into a validation set.

[0028] Furthermore, the optimal strategy is obtained, including:

[0029] The strategy corresponding to the highest score is taken as the optimal strategy.

[0030] Furthermore, the score for each policy in the target policy set is determined, including:

[0031]

[0032] Wherein, F(s) k ) is strategy s k The score, w i Let x be the weight of the i-th evaluation indicator. ki For strategy s k The i-th evaluation index.

[0033] Secondly, the present invention provides a strategy automatic mining device based on multidimensional data-driven methods, the device comprising:

[0034] The acquisition module is used to acquire feature datasets, label datasets, evaluation indicator datasets, and actual decision result datasets in the financial field within a target time period.

[0035] The partitioning module is used to concatenate the feature dataset, label dataset, evaluation index dataset, and actual decision result dataset to obtain a concatenated dataset, and then divide the concatenated dataset into a training set, a test set, and a validation set.

[0036] The function determination module is used to determine the objective function, which in turn determines the optimization direction and optimization objective.

[0037] The decision module is used to construct several decision trees based on the training set and determine the hit rate and recall rate of each branch in each decision tree, where each branch corresponds to a strategy.

[0038] The optimization module is used to filter the corresponding branches based on the preset hit rate threshold, the hit rate and recall rate corresponding to each branch, to obtain several optimal branches, and remove the data of the optimal branches in the training set to obtain the target policy set.

[0039] The filtering module is used to evaluate and filter the set of target strategies based on the test set and the validation set;

[0040] The weighting module is used to determine the weights of several preset evaluation indicators based on the concatenated dataset.

[0041] The optimal strategy module is used to determine the score of each strategy in the target strategy set based on several preset evaluation indicators and the weights corresponding to each preset evaluation indicator, and to obtain the optimal strategy.

[0042] Thirdly, the present invention provides an electronic device, comprising:

[0043] processor;

[0044] Memory used to store processor-executable instructions;

[0045] The processor is configured to execute an automated strategy mining method based on multidimensional data-driven approaches, as provided in the first aspect.

[0046] Fourthly, the present invention provides a non-transitory computer-readable storage medium that, when the instructions in the storage medium are executed by the processor of an electronic device, enables the electronic device to execute the strategy automatic mining method based on multidimensional data-driven methods as provided in the first aspect.

[0047] One or more technical solutions provided in this invention have at least the following technical effects or advantages:

[0048] This invention provides an automatic strategy mining method based on multidimensional data-driven approaches. The method includes: acquiring feature datasets, label datasets, evaluation index datasets, and actual decision result datasets from the financial field within a target time period; concatenating the feature datasets, label datasets, evaluation index datasets, and actual decision result datasets to obtain a concatenated dataset, and dividing the concatenated dataset into a training set, a test set, and a validation set; determining an objective function, which is used to determine the optimization direction and objective; constructing several decision trees based on the training set, and determining the hit rate and recall rate of each branch in each decision tree, where each branch corresponds to a strategy; filtering the corresponding branches based on a preset hit rate threshold and the hit rate and recall rate corresponding to each branch to obtain several optimal branches, and removing the data of the optimal branches from the training set to obtain a target strategy set; evaluating and filtering the target strategy set based on the test set and validation set; determining the weights of several preset evaluation indexes based on the concatenated dataset; determining the score of each strategy in the target strategy set based on the several preset evaluation indexes and their corresponding weights, and obtaining the optimal strategy. This invention constructs an automated strategy mining mechanism that can automatically adjust the strategy set based on multidimensional data, achieving a globally optimal recall rate under a set hit rate requirement, without manual intervention, thus improving the automation level of strategy mining. Based on business experience, the strategy set is pruned to remove parts that do not meet actual business needs. Multiple evaluation dimensions are constructed based on evaluation metrics, and the strategy set that performs best across multiple evaluation metrics is selected from numerous strategy sets. By predicting, measuring, and comparing the actual impact of different strategy sets on production, the most effective strategy for the actual goal is selected. This invention, based on a decision tree, has interpretability, improves the efficiency and accuracy of strategy updates, and enables strategies to adapt to rapidly changing data environments. Attached Figure Description

[0049] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0050] Figure 1 This is a flowchart illustrating the automatic strategy mining method based on multidimensional data-driven approach provided by the present invention. Detailed Implementation

[0051] This invention provides an automatic strategy mining method based on multidimensional data, which solves the technical problem of low efficiency in strategy mining in the prior art.

[0052] The technical solution of this invention is to solve the above-mentioned technical problems, and the overall idea is as follows:

[0053] This paper presents an automatic strategy mining method based on multidimensional data-driven approaches. The method includes: acquiring feature datasets, label datasets, evaluation index datasets, and actual decision result datasets from the financial field within a target time period; concatenating these datasets to obtain a concatenated dataset, which is then divided into training, testing, and validation sets; determining an objective function to define the optimization direction and objective; constructing several decision trees based on the training set and determining the hit rate and recall rate of each branch in each tree, where each branch corresponds to a strategy; filtering the branches based on a preset hit rate threshold and the hit rate and recall rate of each branch to obtain several optimal branches, and removing data from the training set for the optimal branches to obtain a target strategy set; evaluating and filtering the target strategy set based on the testing and validation sets; determining the weights of several preset evaluation indicators based on the concatenated dataset; and determining the score of each strategy in the target strategy set based on the preset evaluation indicators and their corresponding weights to obtain the optimal strategy.

[0054] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific implementation methods.

[0055] First, it should be clarified that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.

[0056] This invention provides, for example Figure 1 The illustrated method for automatic strategy mining based on multidimensional data-driven approaches includes steps S11-S18:

[0057] Step S11: Obtain the feature dataset, label dataset, evaluation index dataset, and actual decision result dataset of the financial field within the target time period.

[0058] The target time period can be selected according to the actual situation. This invention uses data from the financial field from January 1, 2022 to August 1, 2024 as an example for illustration.

[0059] A feature dataset is a collection of data containing various types of features, used to describe the attributes and characteristics of a sample. In financial scenarios, feature datasets may include, but are not limited to, GPS location feature data, personal information feature data, aggregated feature data, and image feature data. By constructing feature datasets, the attributes and characteristics of samples can be comprehensively described, providing a reliable data foundation for fraud detection and risk assessment in financial scenarios.

[0060] A labeled dataset is used to define a set of labels to identify whether a sample in the dataset falls under a certain fraud or overdue scenario. For example, in an anti-fraud model, a labeled dataset might contain a binary label indicating whether a sample has been defrauded offline. Furthermore, a labeled dataset can also use custom combinations of multiple labels to more accurately describe the fraud risk of a sample.

[0061] An evaluation metric dataset is a dataset containing various types of metrics used to describe risks and fraud in financial scenarios. In financial scenarios, evaluation metrics may include indicators such as delinquency rate, fraud delinquency rate, and case investigation anomaly rate.

[0062] The actual decision outcome dataset is used to describe decision outcomes in financial scenarios. In financial scenarios, actual decision outcomes can include fraud-prevention rejections, manual approvals, and authorizations.

[0063] Understandably, within the target time period, various data in the financial field are compiled into corresponding feature datasets, label datasets, evaluation indicator datasets, and actual decision result datasets.

[0064] Step S12: The feature dataset, label dataset, evaluation index dataset, and actual decision result dataset are concatenated to obtain a concatenated dataset, and the concatenated dataset is divided into a training set, a test set, and a validation set.

[0065] Specifically, this includes: dividing the target time period into a first time period and a second time period, wherein the length of the first time period is longer than the length of the second time period and the first time period precedes the second time period; dividing the spliced ​​data in the first time period into a training set and a test set according to a preset ratio; and dividing the spliced ​​data in the second time period into a validation set.

[0066] The purpose of dividing the time period in this invention is to ensure the stability of the strategy in different time intervals. The data in the second time period is more relevant to the current time, so the validation set obtained based on the data in the second time period can make the validation results more accurate.

[0067] For example, a spliced ​​dataset from January 1, 2022 to November 1, 2024 can be divided into the period from January 1, 2022 to August 1, 2024 (the first time period) and the period from August 1, 2024 to November 1, 2024 (the second time period).

[0068] The preset ratio can be determined according to the actual situation. For example, a ratio of 7:3 means that 70% of the data in the first time period is used as the training set and 30% of the data is used as the test set.

[0069] Step S13: Determine the objective function. The objective function is used to determine the optimization direction and optimization objective.

[0070] Policy mining using the training set involves iteratively improving the decision tree to discover and extract effective strategies from the training set. Decision tree mining is a tree-structured data classification and regression method. Its core principle is to recursively select the best splitting attribute to split the dataset until a stopping condition is met, at which point leaf nodes are generated. Each branch of the decision tree can serve as a decision rule or strategy. In financial risk assessment, this could involve deciding whether to approve a loan based on a borrower's credit score, income level, and debt ratio. Each node represents a decision point, and each branch represents a different action path based on the attribute value of that decision point.

[0071] Regarding the objective function, it includes:

[0072] ObjectiveFunction=maxR(Parameters)

[0073] HitRate<θ

[0074] Where ObjectiveFunction is the objective function, R(Parameters) is the recall rate as a function of the strategy, Parameters is the strategy set, HitRate is the hit rate, and θ is the preset hit rate threshold.

[0075] Regarding recall rate, including:

[0076]

[0077] Where Recall is the recall rate, TP is the number of samples correctly predicted as positive, and FN is the number of samples incorrectly predicted as negative but actually positive.

[0078]

[0079] Where Precision is the hit rate, TP is the number of samples correctly predicted as positive by the model, and FP is the number of samples that were actually negative but were incorrectly predicted as negative.

[0080] Based on the objective function, we can ensure that the strategy maximizes recall while meeting the hit rate requirement, thereby achieving both optimization efficiency and accuracy of the strategy.

[0081] Step S14: Construct several decision trees based on the training set, and determine the hit rate and recall rate of each branch in each decision tree, where each branch corresponds to a policy.

[0082] Random row and column sampling is performed on the training set to construct multiple decision trees. For each decision tree, the hit rate and recall rate of each branch set can be determined.

[0083] Step S15: Based on the preset hit rate threshold, the hit rate and recall rate corresponding to each branch, the corresponding branches are filtered to obtain several optimal branches, and the data of the optimal branches in the training set are removed to obtain the target policy set.

[0084] Specifically, steps S151-S152 are included: Step S151, based on the preset hit rate threshold, the hit rate and recall rate corresponding to each branch, the corresponding branches are filtered to determine whether the branch is the optimal branch; Step S152, if the branch is the optimal branch, the data of the branch in the training set is removed and the branch is included in the target policy set; if it is not the optimal branch, step S151 is repeated with other branches.

[0085] It should be noted that the data corresponding to the optimal branch (also known as the dataset) is removed, and S151-S152 is repeated until the hit rate is less than the preset hit rate threshold θ and the recall rate no longer increases. After each update, it is still necessary to check whether the hit rate is still less than θ. If it is, continue optimization; otherwise, stop optimization.

[0086] The process of selecting the optimal branch can be formalized as the following function, which takes the decision tree (T), the hit rate threshold (θ), and the number of branches to be selected (N) as input, and returns a list of branches that meet the conditions.

[0087] Based on a preset hit rate threshold, and the hit rate and recall rate corresponding to each branch, the corresponding branches are filtered, including:

[0088]

[0089] In this function, SelectBranches(T,θ,N) is the selection function, T is the decision tree, b and b′ are branches in the decision tree T, HitRate(b) is the hit rate of branch (b), Recall(b) is the recall rate of branch (b), Recall(b′) is the recall rate of branch (b′), θ is the preset hit rate threshold, and N is the number of branches.

[0090] The mathematical expression for obtaining the target policy set can include:

[0091]

[0092] Where s′ is the target policy set, S b U is the set of data hit by branch b, U is the union of the sets, and SelectedBranches are the selected optimal branches.

[0093] Step S16: Evaluate and filter the target policy set based on the test set and the validation set.

[0094] The performance of the strategy is evaluated on the test set. This ensures that the strategy maximizes both hit rate and recall, and the PSI (Population Stability Index) is calculated over a given time interval. The test set is used to verify whether the strategy achieves stable performance within the same sample set, allowing for the removal of unstable strategy sets.

[0095] The validation set is used to filter out strategies that offer the best stability and performance. The validation set focuses more on stability over time, thus eliminating strategies that are unstable over time. Through these steps, automatic strategy discovery and updating are achieved, enabling the identification of multiple globally optimal strategy sets, thereby improving the accuracy and efficiency of decision-making.

[0096] In addition, during the strategy optimization process, pruning operations can be performed on the strategy set. Specifically, a rule-based approach can be used, setting specific rules based on business experience to eliminate strategy branches that are too costly to implement or conflict with common business sense.

[0097] Step S17: Determine the weights of several preset evaluation indicators based on the spliced ​​dataset.

[0098] Evaluation metrics (such as delinquency rate, fraud delinquency rate, and case investigation anomaly rate) can be used as the basis for evaluation in the validation set. E = e1, e2, ..., e k The weights of each indicator are determined using the analytic hierarchy process (AHP). (This method can also be used when the test and training sets have sufficient performance.)

[0099] Specifically: A hierarchical model is established, with the target layer representing the set of optimal strategies, the criterion layer representing various evaluation indicators, and the solution layer representing different sets of strategies. Pre-set weights are used to compare the importance of each element at the same level with respect to a certain criterion in the previous level, constructing a judgment matrix B = (b... ij ) n×n , where b ijThis represents the importance of indicator i relative to indicator j. The judgment matrix is ​​solved using the eigenvector method to obtain the weights w = (w1, w2, ..., wj) of each evaluation indicator. n ),

[0100] Bw=λ max w

[0101] Where, λ max To determine the largest eigenvalue of a matrix, multiple evaluation dimensions can be constructed based on several weights.

[0102] Step S18: Based on several preset evaluation indicators and the weights corresponding to each preset evaluation indicator, determine the score of each strategy in the target strategy set and obtain the optimal strategy.

[0103] Obtaining the optimal strategy includes selecting the strategy corresponding to the highest score as the optimal strategy.

[0104] Determine the score for each policy in the target policy set, including:

[0105]

[0106] Wherein, F(s) k ) is strategy s k The score, w i Let x be the weight of the i-th evaluation indicator. ki For strategy s k The i-th evaluation index.

[0107] The impact of decisions based on different policy sets can be predicted by establishing a simulation environment and running simulations using historical data. This is achieved through the current decision outcome set (D = d1, d2, ..., d...). p ) Calculate the actual hit rate, anti-fraud rejection rate, approval rate and transfer rate to human agents after going live, measure and compare different strategies, and then carefully select the strategy that is most effective for the actual goal.

[0108] In summary, this invention provides an automatic strategy mining method based on multidimensional data-driven approaches. The method includes: acquiring feature datasets, label datasets, evaluation index datasets, and actual decision result datasets from the financial field within a target time period; concatenating the feature datasets, label datasets, evaluation index datasets, and actual decision result datasets to obtain a concatenated dataset, and dividing the concatenated dataset into training set, test set, and validation set; determining an objective function, which is used to determine the optimization direction and objective; constructing several decision trees based on the training set, and determining the hit rate and recall rate of each branch in each decision tree, where each branch corresponds to a strategy; filtering the corresponding branches based on a preset hit rate threshold and the hit rate and recall rate corresponding to each branch to obtain several optimal branches, and removing the data of the optimal branches from the training set to obtain a target strategy set; evaluating and filtering the target strategy set based on the test set and validation set; determining the weights of several preset evaluation indicators based on the concatenated dataset; and determining the score of each strategy in the target strategy set based on the several preset evaluation indicators and their corresponding weights, thus obtaining the optimal strategy. This invention constructs an automated strategy mining mechanism that can automatically adjust the strategy set based on multidimensional data, achieving a globally optimal recall rate under a set hit rate requirement, without manual intervention, thus improving the automation level of strategy mining. Based on business experience, the strategy set is pruned to remove parts that do not meet actual business needs. Multiple evaluation dimensions are constructed based on evaluation metrics, and the strategy set that performs best across multiple evaluation metrics is selected from numerous strategy sets. By predicting, measuring, and comparing the actual impact of different strategy sets on production, the most effective strategy for the actual goal is selected. This invention, based on a decision tree, has interpretability, improves the efficiency and accuracy of strategy updates, and enables strategies to adapt to rapidly changing data environments.

[0109] Based on the same inventive concept, this invention provides a multi-dimensional data-driven strategy automatic mining device, the device comprising:

[0110] The acquisition module is used to acquire feature datasets, label datasets, evaluation indicator datasets, and actual decision result datasets in the financial field within a target time period.

[0111] The partitioning module is used to concatenate the feature dataset, label dataset, evaluation index dataset, and actual decision result dataset to obtain a concatenated dataset, and then divide the concatenated dataset into a training set, a test set, and a validation set.

[0112] The function determination module is used to determine the objective function, which in turn determines the optimization direction and optimization objective.

[0113] The decision module is used to construct several decision trees based on the training set and determine the hit rate and recall rate of each branch in each decision tree, where each branch corresponds to a strategy.

[0114] The optimization module is used to filter the corresponding branches based on the preset hit rate threshold, the hit rate and recall rate corresponding to each branch, to obtain several optimal branches, and remove the data of the optimal branches in the training set to obtain the target policy set.

[0115] The filtering module is used to evaluate and filter the set of target strategies based on the test set and the validation set;

[0116] The weighting module is used to determine the weights of several preset evaluation indicators based on the concatenated dataset.

[0117] The optimal strategy module is used to determine the score of each strategy in the target strategy set based on several preset evaluation indicators and the weights corresponding to each preset evaluation indicator, and to obtain the optimal strategy.

[0118] Based on the same inventive concept, the present invention also provides an electronic device, comprising:

[0119] processor;

[0120] Memory used to store processor-executable instructions;

[0121] The processor is configured to execute an automated strategy mining method based on multidimensional data-driven approaches, as described above.

[0122] Based on the same inventive concept, the present invention also provides a non-transitory computer-readable storage medium, which, when the instructions in the storage medium are executed by the processor of an electronic device, enables the electronic device to execute an automatic strategy mining method based on multidimensional data drive as described above.

[0123] Since the electronic device described in this embodiment is an electronic device used to implement the information processing method in the embodiments of the present invention, those skilled in the art can understand the specific implementation methods and various variations of the electronic device in this embodiment based on the information processing method described in the embodiments of the present invention. Therefore, how the electronic device implements the method in the embodiments of the present invention will not be described in detail here. Any electronic device used by those skilled in the art to implement the information processing method in the embodiments of the present invention falls within the scope of protection of the present invention.

[0124] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0125] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0126] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0127] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0128] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0129] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A strategy automatic mining method based on multidimensional data-driven approach, characterized in that, The method includes: Acquire feature datasets, label datasets, evaluation indicator datasets, and actual decision result datasets in the financial field within a target time period; The feature dataset, the label dataset, the evaluation index dataset, and the actual decision result dataset are concatenated to obtain a concatenated dataset, which is then divided into a training set, a test set, and a validation set. Determine the objective function, which is used to determine the optimization direction and optimization objective; Construct several decision trees based on the training set, and determine the hit rate and recall rate of each branch in each decision tree, wherein each branch corresponds to a strategy. Based on the preset hit rate threshold, the hit rate and recall rate corresponding to each branch, the corresponding branches are filtered to obtain several optimal branches, and the data of the optimal branches in the training set are removed to obtain the target policy set. The target policy set is evaluated and filtered based on the test set and the validation set; Based on the assembled dataset, the weights of several preset evaluation indicators are determined; Based on several preset evaluation indicators and the weights corresponding to each preset evaluation indicator, the score of each strategy in the target strategy set is determined, and the optimal strategy is obtained.

2. The automatic strategy mining method based on multidimensional data-driven approach as described in claim 1, characterized in that, Based on a preset hit rate threshold, the hit rate and recall rate corresponding to each branch, the corresponding branches are filtered to obtain several optimal branches. The data of the optimal branches in the training set are then removed to obtain the target policy set, including steps S151-S152: Step S151: Based on the preset hit rate threshold, the hit rate and recall rate corresponding to each branch, filter the corresponding branches and determine whether the branch is the optimal branch. Step S152: If the branch is the optimal branch, remove the data of the branch in the training set and include the branch in the target policy set; if it is not the optimal branch, repeat step S151 with other branches.

3. The automatic strategy mining method based on multidimensional data-driven approach as described in claim 2, characterized in that, Based on a preset hit rate threshold, and the hit rate and recall rate corresponding to each branch, the corresponding branches are filtered, including: In this function, SelectBranches(T,θ,N) is the selection function, T is the decision tree, b and b′ are branches in the decision tree T, HitRate(b) is the hit rate of branch (b), Recall(b) is the recall rate of branch (b), Recall(b′) is the recall rate of branch (b′), θ is the preset hit rate threshold, and N is the number of branches.

4. The automatic strategy mining method based on multidimensional data-driven approach as described in claim 1, characterized in that, The objective function includes: Objective Function=maxR(Parameters) HitRate<θ Wherein, Objective Function is the objective function, R(Parameters) is the recall rate as a function of the strategy, Parameters is the strategy set, HitRate is the hit rate, and θ is the preset hit rate threshold.

5. The automatic strategy mining method based on multidimensional data-driven approach as described in claim 1, characterized in that, The concatenated dataset is divided into a training set, a test set, and a validation set, including: The target time period is divided into a first time period and a second time period, wherein the length of the first time period is longer than the length of the second time period and the first time period precedes the second time period. The spliced ​​data in the first time period is divided into training set and test set according to a preset ratio; The spliced ​​data in the second time period is divided into a validation set.

6. The automatic strategy mining method based on multidimensional data-driven approach as described in claim 1, characterized in that, To obtain the optimal strategy, including: The strategy corresponding to the highest score is taken as the optimal strategy.

7. The automatic strategy mining method based on multidimensional data-driven approach as described in claim 1, characterized in that, Determining the score for each policy in the target policy set includes: Wherein, F(s) k ) is strategy s k The score, w i Let x be the weight of the i-th evaluation indicator. ki For strategy s k The i-th evaluation index.

8. A strategy automatic mining device based on multidimensional data-driven approach, characterized in that, The device includes: The acquisition module is used to acquire feature datasets, label datasets, evaluation indicator datasets, and actual decision result datasets in the financial field within a target time period. The partitioning module is used to concatenate the feature dataset, the label dataset, the evaluation index dataset, and the actual decision result dataset to obtain a concatenated dataset, and to partition the concatenated dataset into a training set, a test set, and a validation set. The function determination module is used to determine the objective function, which is used to determine the optimization direction and optimization objective; The decision module is used to construct several decision trees based on the training set and determine the hit rate and recall rate of each branch in each decision tree, wherein each branch corresponds to a strategy. The optimization module is used to filter the corresponding branches based on a preset hit rate threshold, the hit rate and recall rate corresponding to each branch, to obtain several optimal branches, and to remove the data of the optimal branches in the training set to obtain the target policy set. A filtering module is used to evaluate and filter the target policy set based on the test set and the validation set; The weighting module is used to determine the weights of several preset evaluation indicators based on the spliced ​​dataset. The optimal strategy module is used to determine the score of each strategy in the target strategy set based on several preset evaluation indicators and the weights corresponding to each preset evaluation indicator, and to obtain the optimal strategy.

9. An electronic device, characterized in that, include: processor; Memory used to store the processor's executable instructions; The processor is configured to execute the automatic strategy mining method based on multidimensional data-driven methods as described in any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium, characterized in that, When the instructions in the storage medium are executed by the processor of the electronic device, the electronic device is able to perform the strategy automatic mining method based on multidimensional data drive as described in any one of claims 1 to 7.

Images