Splitting Method and Device of Decision Tree Model, Storage Medium and Electronic Device

Abstract

The present invention discloses a splitting method and apparatus, a storage medium, and an electronic device for a decision tree model. The method includes: obtaining a first encrypted calculation result of a target operator and a second encrypted calculation result of other operators. The first calculation result includes: multiple first feature weights corresponding to multiple first data features in the target operator, and multiple first bin boundaries corresponding to the multiple first feature weights; determining multiple third bin boundaries according to the multiple first feature weights, multiple second feature weights, multiple first bin boundaries, and multiple second bin boundaries, where the multiple third bin boundaries are encrypted boundaries; binning the multiple first data features according to the multiple third bin boundaries to obtain a binning result, and determining a splitting scheme for the target decision tree model according to the binning result, so as to split the leaf nodes in the target decision tree model according to the splitting scheme.

Description

Technical Field

[0001] The present invention relates to the field of communications, and in particular, to a method and device for splitting a decision tree model, a storage medium, and an electronic device. Background Art

[0002] Today, with the booming development of the Internet, network and information security issues have become the top priorities for individuals and enterprises.

[0003] In order to protect the personal information and financial security of the public, operators actively explore diversified technical means based on big data and machine learning to build a preventive technical system for anti-telecom network fraud, so as to achieve rapid discovery and timely interception and control of network fraud events and safeguard network security. However, the anti-fraud modeling of operators faces problems such as a small sample size, insufficient behavior, and poor model effects.

[0004] In view of the problems in the related art, such as the small sample size and insufficient behavior of the existing decision tree model of operators, resulting in poor model effects, no effective solution has been proposed yet. Summary of the Invention

[0005] Embodiments of the present invention provide a method and device for splitting a decision tree model, a storage medium, and an electronic device, so as to at least solve the problems in the related art, such as the small sample size and insufficient behavior of the existing decision tree model of operators, resulting in poor model effects.

[0006] According to an embodiment of the embodiments of the present invention, a method for splitting a decision tree model is provided, including: obtaining a first encrypted calculation result of a target operator and a second encrypted calculation result of other operators, where the first calculation result includes: multiple first feature weights respectively corresponding to multiple first data features in the target operator, and multiple first bin boundary values respectively corresponding to the multiple first feature weights, and the second calculation result includes: multiple second feature weights respectively corresponding to multiple second data features in the other operators, and multiple second bin boundary values respectively corresponding to the multiple second feature weights; determining multiple third bin boundary values according to the multiple first feature weights, the multiple second feature weights, the multiple first bin boundary values, and the multiple second bin boundary values, where the multiple third bin boundary values are encrypted boundary values; binning the multiple first data features according to the multiple third bin boundary values to obtain a binning result, and determining a splitting scheme for the target decision tree model according to the binning result, so as to split leaf nodes in the target decision tree model according to the splitting scheme.

[0007] In an exemplary embodiment, re-binning the data features according to the third bin boundary values includes: sending the third bin boundary values to a key management party, where the other operators include: the key management party; receiving, when the key management party decrypts the third bin boundary values by a target key, the fourth bin boundary values sent by the key management party, where the fourth bin boundary values are the boundary values after decrypting the third bin boundary values; and re-binning the data features according to the fourth bin boundary values.

[0008] In an exemplary embodiment, determining a plurality of third bin boundary values according to the plurality of first feature weights, the plurality of second feature weights, the plurality of first bin boundary values, and the plurality of second bin boundary values includes at least one of the following: obtaining a third feature weight and third bin boundary values of a target data feature in the target operator, and obtaining a fourth feature weight and fourth bin boundary values of the target feature in the other operator, where the plurality of first data features include: the target data feature, the plurality of first feature weights include: the third feature weight, the plurality of second feature weights include: the fourth feature weight, the plurality of first bin boundary values include: the third bin boundary values, and the plurality of second bin boundary values include: the fourth bin boundary values; calculating a weighted sum of the third feature weight, the third bin boundary values, the fourth feature weight, and the fourth bin boundary values, and a weight sum of the third feature weight and the fourth feature weight; and determining the third bin boundary values according to the weighted sum and the weight sum.

[0009] In an exemplary embodiment, after splitting a leaf node in the target decision tree model according to the splitting scheme, the method further includes: calculating a loss function of the split target decision tree model; determining whether a convergence degree of the loss function reaches a preset convergence condition; and obtaining an encrypted third calculation result of the target operator and an encrypted fourth calculation result of the other operator when the convergence degree of the loss function does not reach the preset convergence condition.

[0010] In an exemplary embodiment, determining a splitting scheme of a target decision tree model according to the binning results includes: determining a first feature histogram corresponding to the plurality of first data features according to the binning results; aggregating the first feature histogram and a second feature histogram corresponding to the plurality of second data features when receiving the second feature histogram sent by the other operator to obtain an aggregation result; and determining the splitting scheme of the target decision tree model according to the aggregation result.

[0011] In an exemplary embodiment, before obtaining the first calculation result of the target operator and the encrypted second calculation results of other operators, the method further includes: receiving the target key sent by the key management party; when the target operator calculates multiple first feature weights corresponding to the multiple first data features respectively, and multiple first bin boundary values corresponding to the multiple first feature weights respectively, encrypting the multiple first feature weights and the multiple first bin boundary values by using the target key.

[0012] According to another embodiment of the embodiments of the present invention, there is also provided a splitting device for a decision tree model, including: an obtaining module, configured to obtain the encrypted first calculation result of the target operator and the encrypted second calculation results of other operators, where the first calculation result includes: multiple first feature weights corresponding to multiple first data features in the target operator respectively, and multiple first bin boundary values corresponding to the multiple first feature weights respectively, and the second calculation result includes: multiple second feature weights corresponding to multiple second data features in the other operators respectively, and multiple second bin boundary values corresponding to the multiple second feature weights respectively; a first determining module, configured to determine multiple third bin boundary values according to the multiple first feature weights, the multiple second feature weights, the multiple first bin boundary values and the multiple second bin boundary values, where the multiple third bin boundary values are encrypted boundary values; a second determining module, configured to perform binning on the multiple first data features according to the multiple third bin boundary values to obtain a binning result, and determine a splitting scheme for the target decision tree model according to the binning result, so as to split a leaf node in the target decision tree model according to the splitting scheme.

[0013] In an exemplary embodiment, the device further includes: a sending module, configured to send the third bin boundary values to the key management party, where the other operators include: the key management party; a receiving module, configured to receive the fourth bin boundary values sent by the key management party when the key management party decrypts the third bin boundary values by using the target key, where the fourth bin boundary values are the boundary values after the third bin boundary values are decrypted; the first determining module is further configured to perform binning on the data features again according to the fourth bin boundary values.

[0014] According to yet another aspect of the embodiments of the present invention, there is also provided a computer-readable storage medium, in which a computer program is stored, where the computer program is configured to execute the above-mentioned splitting method of the decision tree model when running.

[0015] According to another aspect of the embodiments of the present invention, an electronic device is further provided, including a memory, a processor, and a computer program stored on the memory and executable on the processor. Wherein, the processor executes the splitting method of the above decision tree model through the computer program.

[0016] In the embodiments of the present invention, the encrypted first calculation result of the target operator and the encrypted second calculation result of other operators are obtained. Wherein, the first calculation result includes: multiple first feature weights respectively corresponding to multiple first data features in the target operator, and multiple first bin boundaries respectively corresponding to the multiple first feature weights; the second calculation result includes: multiple second feature weights respectively corresponding to multiple second data features in other operators, and multiple second bin boundaries respectively corresponding to the multiple second feature weights; multiple third bin boundaries are determined according to the multiple first feature weights, the multiple second feature weights, the multiple first bin boundaries, and the multiple second bin boundaries. Wherein, the multiple third bin boundaries are encrypted boundaries; the multiple first data features are binned according to the multiple third bin boundaries to obtain a binning result, and a splitting scheme of the target decision tree model is determined according to the binning result, so as to split the leaf nodes in the target decision tree model according to the splitting scheme; By adopting the above technical solution, problems such as the small sample size and insufficient behavior of the decision tree model of existing operators, resulting in poor model effects, are solved. Furthermore, on the premise of ensuring that the original data does not leave the domain, private data is not exchanged, and plaintext data is not exchanged, through secure aggregation calculation, joint anti-fraud modeling of cross-operator data is realized, and the effect of the decision tree model is improved by fusing the sample data and behavior characteristics of multiple operators. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The drawings described herein are used to provide a further understanding of the present invention, and constitute a part of this application. The schematic embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation to the present invention. In the drawings:

[0018] Figure 1 It is a hardware structure block diagram of a computer terminal for a splitting method of a decision tree model according to an embodiment of the present invention;

[0019] Figure 2 It is a flowchart of a splitting method of a decision tree model according to an embodiment of the present invention;

[0020] Figure 3 It is a structural diagram of a splitting device of a decision tree model according to an embodiment of the present invention;

[0021] Figure 4 It is a flowchart of a splitting method of a decision tree model according to an embodiment of the present invention;

[0022] Figure 5 It is a structural block diagram of a splitting device of a decision tree model according to an embodiment of the present invention. Specific embodiments

[0023] In order to enable those skilled in the art to better understand the solution of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without making creative efforts shall fall within the protection scope of the present invention.

[0024] It should be noted that the terms "first", "second", etc. in the description and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and do not necessarily need to be used to describe a specific order or sequence. It should be understood that such data can be interchanged under appropriate circumstances so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device comprising a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0025] The method embodiments provided in the embodiments of the present application can be executed on a mobile terminal, a computer terminal or a similar computing device. Taking running on a computer terminal as an example, Figure 1 It is a hardware structural block diagram of a computer terminal of a splitting method of a decision tree model according to an embodiment of the present invention. As Figure 1 shown, the computer terminal may include one or more ( Figure 1 only one is shown in Figure 1 a processor 102 (the processor 102 may include, but is not limited to, a processing device such as a microprocessor MCU or a programmable logic device FPGA) and a memory 104 for storing data. In an exemplary embodiment, the above computer terminal may further include a transmission device 106 for communication functions and an input / output device 108. Those of ordinary skill in the art can understand that Figure 1 the structure shown in Figure 1 is only schematic and does not limit the structure of the above computer terminal. For example, the computer terminal may further include more or fewer components than those shown in Figure 1 or have an equivalent function to that shown in

[0026] The memory 104 can be used to store computer programs, for example, software programs and modules of application software, such as the computer program corresponding to the splitting method of the decision tree model in the embodiments of the present invention. The processor 102 executes various functional applications and data processing by running the computer programs stored in the memory 104, that is, the above-mentioned method is implemented. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memories, or other non-volatile solid-state memories. In some instances, the memory 104 may further include a memory remotely disposed relative to the processor 102, and these remote memories can be connected to the computer terminal through a network. Examples of the above-mentioned network include but are not limited to the Internet, enterprise intranet, local area network, mobile communication network, and combinations thereof.

[0027] The transmission device 106 is used to receive or send data via a network. Specific examples of the above-mentioned network may include a wireless network provided by a communication provider of the computer terminal. In one instance, the transmission device 106 includes a network adapter (Network Interface Controller, abbreviated as NIC), which can be connected to other network devices through a base station and thus communicate with the Internet. In one instance, the transmission device 106 can be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0028] In this embodiment, a splitting method of a decision tree model is provided, which is applied to the above-mentioned computer terminal. Figure 2 It is a flowchart of the splitting method of the decision tree model according to the embodiments of the present invention, and this process includes the following steps:

[0029] Step S202, obtain the first encrypted calculation result of the target operator and the second encrypted calculation result of other operators. Among them, the first calculation result includes: multiple first feature weights respectively corresponding to multiple first data features in the target operator, and multiple first bin boundary values respectively corresponding to the multiple first feature weights; the second calculation result includes: multiple second feature weights respectively corresponding to multiple second data features in other operators, and multiple second bin boundary values respectively corresponding to the multiple second feature weights.

[0030] Step S204, determine multiple third bin boundary values according to the multiple first feature weights, the multiple second feature weights, the multiple first bin boundary values, and the multiple second bin boundary values. Among them, the multiple third bin boundary values are encrypted boundary values.

[0031] Step S206: Bin the multiple first data features according to the multiple third bin boundary values to obtain a binning result, and determine a splitting scheme for the target decision tree model according to the binning result, so as to split the leaf nodes in the target decision tree model according to the splitting scheme.

[0032] Through the above steps, the first calculation result encrypted by the target operator and the second calculation result encrypted by other operators are obtained. The first calculation result includes: multiple first feature weights respectively corresponding to multiple first data features in the target operator, and multiple first bin boundary values respectively corresponding to the multiple first feature weights. The second calculation result includes: multiple second feature weights respectively corresponding to multiple second data features in other operators, and multiple second bin boundary values respectively corresponding to the multiple second feature weights. Determine multiple third bin boundary values according to the multiple first feature weights, the multiple second feature weights, the multiple first bin boundary values and the multiple second bin boundary values, where the multiple third bin boundary values are encrypted boundary values. Bin the multiple first data features according to the multiple third bin boundary values to obtain a binning result, and determine a splitting scheme for the target decision tree model according to the binning result, so as to split the leaf nodes in the target decision tree model according to the splitting scheme. By adopting the above technical solution, problems such as the small sample size and insufficient behavior of the decision tree model of the existing operator, resulting in poor model effects, are solved. Furthermore, on the premise of ensuring that the original data does not leave the domain, does not exchange privacy data, and does not exchange plaintext data, through secure aggregation calculation, cross-operator data joint anti-fraud modeling is realized. By fusing the sample data and behavior characteristics of multiple operators, the effect of the decision tree model is improved.

[0033] In an exemplary embodiment, binning the data features again according to the third bin boundary values includes: sending the third bin boundary values to a key management party, where the other operators include: the key management party; receiving the fourth bin boundary values sent by the key management party when the key management party decrypts the third bin boundary values through a target key, where the fourth bin boundary values are the boundary values after decrypting the third bin boundary values; binning the data features again according to the fourth bin boundary values.

[0034] It should be noted that the target operator and the key management party are randomly negotiated and specified among multiple operators. The key management party is used to generate keys, and the target operator is used to determine the splitting scheme. After the target operator completes the weighted average of features in the encryption domain, it synchronizes the encrypted third bin boundary value to the key management party. After the key management party decrypts the third bin boundary value, it obtains the fourth bin boundary value and distributes it to the target operator and other operators, so that all operators can adjust the bin boundary to re-bin the data features.

[0035] In an exemplary embodiment, determining multiple third bin boundary values according to the multiple first feature weights, the multiple second feature weights, the multiple first bin boundary values, and the multiple second bin boundary values includes at least one of the following: obtaining the third feature weight and the third bin boundary value of the target data feature in the target operator, and obtaining the fourth feature weight and the fourth bin boundary value of the target feature in other operators, where the multiple first data features include: the target data feature, the multiple first feature weights include: the third feature weight, the multiple second feature weights include: the fourth feature weight, the multiple first bin boundary values include: the third bin boundary value, and the multiple second bin boundary values include: the fourth bin boundary value; calculating the weighted sum of the third feature weight, the third bin boundary value, the fourth feature weight, and the fourth bin boundary value, and the weight sum of the third feature weight and the fourth feature weight; and determining the third bin boundary value according to the weighted sum and the weight sum.

[0036] That is to say, all operators complete the calculation and normalization of the feature weight values (IV or WOE) of the feature data locally, and bin each feature data, and encrypt and send the calculation results: feature number, feature weight value (IV), and the bin boundary value corresponding to the feature weight value to the target operator; the target operator completes the weighted average of the feature data in the encryption domain:

[0037] Binavg = (IVai * Binai + IVbi * Binbi + ……) / (IVai + IVbi + ……); where Binavg is the third bin boundary value, IVai is the first feature weight, IVbi…… is the second feature weight, Binai is the first bin boundary value, and Binbi is the second bin boundary value.

[0038] In an exemplary embodiment, after splitting the leaf nodes in the target decision tree model according to the splitting scheme, calculate the loss function of the split target decision tree model; determine whether the convergence degree of the loss function reaches a preset convergence condition; in the case where the convergence degree of the loss function does not reach the preset convergence condition, obtain the third calculation result encrypted by the target operator and the fourth calculation result encrypted by the other operators.

[0039] That is, all operators continue to split the leaf nodes of the next layer of the decision tree model until a complete decision tree model is obtained; all operators calculate the loss function according to the obtained decision tree model and determine whether the loss reaches the convergence condition; if the preset convergence condition is reached, end the training of the decision tree model; if the preset convergence condition is not reached, continue to train the decision tree model.

[0040] In an exemplary embodiment, determining the splitting scheme of the target decision tree model according to the binning result includes: determining the first feature histogram corresponding to the multiple first data features according to the binning result; in the case of receiving the second feature histogram corresponding to the multiple second data features sent by other operators, aggregating the first feature histogram and the second feature histogram to obtain an aggregation result; determining the splitting scheme of the target decision tree model according to the aggregation result.

[0041] Specifically, the key management party distributes the post-fourth bin boundary values to other operators, all operators adjust the bin boundaries for re-binning, and locally calculate the feature histograms based on their own data features, send the calculation results to the target operator, and the target operator aggregates the histograms to calculate the optimal splitting scheme and distributes it to other operators.

[0042] In an exemplary embodiment, before obtaining the first calculation result of the target operator and the second calculation result encrypted by other operators, receive the target key sent by the key management party; in the case where the target operator calculates the multiple first feature weights corresponding to the multiple first data features and the multiple first bin boundary values corresponding to the multiple first feature weights, encrypt the multiple first feature weights and the multiple first bin boundary values with the target key.

[0043] That is to say, all operators uniformly align the fraud scenario samples and related data features participating in the model calculation, uniformly map the static feature codes, perform unified normalization processing on the numerical features, and eliminate the influence of data imbalance; all operators import the preprocessed data into the secure computing system, start decision tree modeling according to the parameter settings, and in each round of decision tree model calculation, all operators locally complete the calculation and normalization of feature values and the binning of each data feature.

[0044] To better understand the process of the splitting method of the above decision tree model, the following will further illustrate the implementation method flow of the splitting of the above decision tree model in combination with optional embodiments, but it is not used to limit the technical solutions of the embodiments of the present invention.

[0045] In this embodiment, a splitting method of a decision tree model is provided. Figure 3 It is a structural diagram of a splitting device of a decision tree model according to an embodiment of the present invention, as Figure 3 shown. The splitting device of the decision tree model includes: a data management module, a secure computing module, and a secure aggregation module; the specific functions are as follows:

[0046] 1. Data management module:

[0047] a) Import the locally participating calculation data into the secure model calculation system;

[0048] b) Manage the local data in the secure model calculation system, including data dictionaries, etc., to perform necessary alignment operations on the data participated in the calculation by multiple parties;

[0049] 2. Secure computing module: mainly used for calculations of local feature engineering, model training, and model inference.

[0050] a) Feature engineering: preprocessing of local features (K values), including mapping conversion of discrete features, normalization processing of numerical features, etc.; local feature calculation, calculating feature weight values such as IV and WOE.

[0051] b) Model training: local modeling calculations of each participating party (equivalent to all operators in the above embodiments) on their own data, including histogram information of their respective statistical attribute data, calculating loss functions, etc.; encrypt and send the information to be interacted to the secure aggregation module; receive the aggregation result returned by the secure aggregation module, and determine whether the model converges. If it does not converge, continue with the next round of model calculations.

[0052] c) Model inference: storage and management of the final aggregated model; call the aggregated model for model inference calculations.

[0053] 3. Secure aggregation module: mainly used for key management, secure interaction, and aggregation calculations.

[0054] a) Generate and manage keys;

[0055] b) Establish a secure connection and interact with the other party's system (equivalent to other operators in the above embodiments);

[0056] c) Receive the encrypted information sent by the participating party and perform secure aggregation calculations such as weighted average;

[0057] d) Return the aggregation result to the calculation module.

[0058] In this embodiment, a splitting method for a decision tree model is also provided. Figure 4 It is a flowchart of the splitting method for the decision tree model according to the embodiment of the present invention, as Figure 4 shown, and the specific steps are as follows:

[0059] Step S401: Each operator uniformly aligns the fraud scenario samples and related data features participating in model calculation, uniformly maps the static feature codes, and performs unified normalization processing on the numerical features to eliminate the influence of data imbalance.

[0060] Step S402: Each operator imports the preprocessed data into the secure computing system and starts the decision tree modeling and calculation of the tree model in each round according to the parameter settings.

[0061] Step S403: The system randomly designates a model aggregator (equivalent to the target operator in the above embodiment) and a key management party among the participants.

[0062] Step S404: According to the characteristics of the fraud sample data, each participant selects all positive sample users and randomly extracts negative sample data for training.

[0063] Step S405: Each party completes the calculation and normalization of the feature IV (or WOE) value and the binning of each feature locally, and encrypts and sends the calculation results (feature serial number, feature weight (IV), binning boundary list) to the aggregator.

[0064] Step S406: The aggregator completes the weighted average of the features in the encrypted domain: Binavg = (IVai * Binai + IVbi * Binbi + ……) / (IVai + IVbi + ……); and synchronizes the averaged binning boundaries to the key management party. After decrypting the results, the key management party distributes them to other participants. Other participants adjust the binning boundaries for re-binning, and calculate the feature histograms Gi,k and Hi,k locally based on their own data, and send the calculation results to the aggregator.

[0065] Step S407: The aggregator converges the histogram to calculate the optimal splitting scheme and distributes it to all parties.

[0066] Step S408: All parties continue to split the next-level leaf nodes until a complete tree model is obtained.

[0067] Step S409: All parties calculate the loss function according to the obtained model and determine whether the loss reaches the convergence condition; if the pre-set convergence condition is reached, proceed to the next step; if the pre-set convergence condition is not reached, repeat steps S404 - S409.

[0068] Step S410: The loss reaches the convergence condition, and the anti-fraud model is completed.

[0069] Step S411: Each operator obtains the model after fusion calculation and can perform local data inference to output target customers.

[0070] Currently, in the field of telecom anti-fraud, existing machine learning-based patents create rules or train models based on data samples of a single operator. In the case of insufficient data samples or limited fraud types of a single operator, it is difficult to further improve the accuracy of the model. This patent proposes a secure joint calculation method that can jointly use data from multiple operators for more sufficient training to further improve the model accuracy. Comparing the existing decision tree model with distributed modeling and technologies such as federated learning in other fields, regardless of whether a client-server or peer-to-peer network distributed architecture is adopted, in the case of only two cooperating parties without a third-party coordination, the following problems will inevitably occur: Only one of the two parties can serve as the aggregator, and the other cooperating party inevitably needs to let the aggregator know the specific binning situation of its own data and understand its own data distribution. Although this data does not involve individual privacy, it is also information that the data party does not want to expose; without a trusted third party, homomorphic encryption, scrambling, etc. methods in the client-server architecture are also not feasible in the case of only two parties.

[0071] The embodiment of the present invention improves the above problems through a decentralized calculation method: The system randomly designates the aggregator. In the case of only two parties, the two parties take turns serving as the model aggregator; according to the characteristics of the anti-fraud model, the negative samples are randomly resampled in each round of training, and the single-round model only aggregates the sampled data; in the binning aggregation process, the results of feature calculation and normalization are innovatively added as weights to perform weighted averaging on the binning of both parties; the non-aggregator generates the homomorphic key, and the aggregator performs weighted average calculation in the encrypted domain, and the non-aggregator decrypts the calculation result; the aggregator aggregates the encrypted results and cannot decrypt them, so it does not know the original situation of the data; the non-aggregator synchronizes the decrypted weighted average result to each party after decryption; each party does not know the feature weights of the other party, so it cannot infer the original value of the other party based on the weighted average result; this solution uses a secure encryption algorithm to comprehensively utilize the data and features of multiple operators while protecting user privacy, taking into account both accuracy and security.

[0072] Through the description of the above embodiments, those skilled in the art can clearly understand that the method according to the above embodiments can be implemented by means of software plus a necessary general hardware platform. Of course, it can also be implemented by hardware, but in many cases the former is a better implementation. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes several instructions for causing a terminal device (which can be a mobile phone, a computer, a server, or a network device, etc.) to execute the methods of the various embodiments of the present invention.

[0073] In this embodiment, a splitting device for a decision tree model is further provided. The splitting device for the decision tree model is used to implement the above embodiments and preferred embodiments, and those that have been described will not be repeated. As used below, the term "module" can be a combination of software and / or hardware that can achieve a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, implementation in hardware, or a combination of software and hardware is also possible and contemplated.

[0074] Figure 5 is a structural block diagram of a splitting device for a decision tree model according to an embodiment of the present invention; as Figure 5 shown, it includes:

[0075] An acquisition module 52, configured to acquire a first calculation result encrypted by a target operator and a second calculation result encrypted by other operators. Among them, the first calculation result includes: multiple first feature weights corresponding to multiple first data features in the target operator, and multiple first bin boundary values corresponding to the multiple first feature weights; the second calculation result includes: multiple second feature weights corresponding to multiple second data features in the other operators, and multiple second bin boundary values corresponding to the multiple second feature weights;

[0076] A first determination module 54, configured to determine multiple third bin boundary values according to the multiple first feature weights, the multiple second feature weights, the multiple first bin boundary values, and the multiple second bin boundary values, where the multiple third bin boundary values are encrypted boundary values;

[0077] A second determination module 56, configured to bin the multiple first data features according to the multiple third bin boundary values to obtain a binning result, and determine a splitting scheme for the target decision tree model according to the binning result, so as to split the leaf nodes in the target decision tree model according to the splitting scheme.

[0078] Through the above device, obtain the first encrypted calculation result of the target operator and the second encrypted calculation result of other operators. The first calculation result includes: multiple first feature weights respectively corresponding to multiple first data features in the target operator, and multiple first bin boundary values respectively corresponding to the multiple first feature weights. The second calculation result includes: multiple second feature weights respectively corresponding to multiple second data features in other operators, and multiple second bin boundary values respectively corresponding to the multiple second feature weights. Determine multiple third bin boundary values according to the multiple first feature weights, the multiple second feature weights, the multiple first bin boundary values, and the multiple second bin boundary values. The multiple third bin boundary values are encrypted boundary values. Bin the multiple first data features according to the multiple third bin boundary values to obtain a binning result, and determine a splitting scheme for the target decision tree model according to the binning result, so as to split the leaf nodes in the target decision tree model according to the splitting scheme. On the premise of ensuring that the original data does not go out of the domain, private data is not exchanged, and plaintext data is not exchanged, through secure aggregation calculation, realize the joint anti-fraud modeling of cross-operator data, and improve the effect of the decision tree model by fusing the sample data and behavior characteristics of multiple operators.

[0079] In an exemplary embodiment, the device further includes: a sending module, configured to send the third bin boundary value to a key management party, where the other operators include: the key management party; a receiving module, configured to receive the fourth bin boundary value sent by the key management party when the key management party decrypts the third bin boundary value through a target key, where the fourth bin boundary value is the boundary value after decrypting the third bin boundary value; the first determining module is further configured to bin the data features again according to the fourth bin boundary value.

[0080] In an exemplary embodiment, the first determining module is further configured to obtain the third feature weight and the third bin boundary value of the target data feature in the target operator, and obtain the fourth feature weight and the fourth bin boundary value of the target feature in other operators, where the multiple first data features include: the target data feature, the multiple first feature weights include: the third feature weight, the multiple second feature weights include: the fourth feature weight, the multiple first bin boundary values include: the third bin boundary value, and the multiple second bin boundary values include: the fourth bin boundary value; calculate the weighted sum of the third feature weight, the third bin boundary value, the fourth feature weight, and the fourth bin boundary value, and the weight sum of the third feature weight and the fourth feature weight; determine the third bin boundary value according to the weighted sum and the weight sum.

[0081] In an exemplary embodiment, the obtaining module is further configured to calculate the loss function of the split target decision tree model; determine whether the convergence degree of the loss function reaches a preset convergence condition; and in the case where the convergence degree of the loss function does not reach the preset convergence condition, obtain the third calculation result encrypted by the target operator and the fourth calculation result encrypted by the other operator.

[0082] In an exemplary embodiment, the second determination module is further configured to determine a first feature histogram corresponding to the plurality of first data features according to the binning result; in the case of receiving the second feature histogram corresponding to the plurality of second data features sent by the other operator, aggregate the first feature histogram and the second feature histogram to obtain an aggregation result; and determine a splitting scheme of the target decision tree model according to the aggregation result.

[0083] In an exemplary embodiment, the receiving module is further configured to receive a target key sent by the key management party; and in the case where the target operator calculates a plurality of first feature weights respectively corresponding to the plurality of first data features and a plurality of first bin boundaries respectively corresponding to the plurality of first feature weights, encrypt the plurality of first feature weights and the plurality of first bin boundaries by using the target key.

[0084] An embodiment of the present invention further provides a storage medium, which includes a stored program, where the above program executes the method according to any one of the above when running.

[0085] Optionally, in this embodiment, the above storage medium may be set to store program codes for executing the following steps:

[0086] S1. Obtain a first calculation result encrypted by a target operator and a second calculation result encrypted by other operators, where the first calculation result includes: a plurality of first feature weights respectively corresponding to a plurality of first data features in the target operator, and a plurality of first bin boundaries respectively corresponding to the plurality of first feature weights, and the second calculation result includes: a plurality of second feature weights respectively corresponding to a plurality of second data features in the other operator, and a plurality of second bin boundaries respectively corresponding to the plurality of second feature weights;

[0087] S2. Determine a plurality of third bin boundaries according to the plurality of first feature weights, the plurality of second feature weights, the plurality of first bin boundaries, and the plurality of second bin boundaries, where the plurality of third bin boundaries are encrypted boundaries;

[0088] S3. Bin the multiple first data features according to the multiple third bin boundary values to obtain a binning result, and determine a splitting scheme for the target decision tree model according to the binning result, so as to split the leaf nodes in the target decision tree model according to the splitting scheme.

[0089] An embodiment of the present invention further provides an electronic device, including a memory and a processor. A computer program is stored in the memory, and the processor is configured to run the computer program to execute the steps in any one of the above method embodiments.

[0090] Optionally, the above electronic device may further include a transmission device and an input / output device. Among them, the transmission device is connected to the above processor, and the input / output device is connected to the above processor.

[0091] Optionally, in this embodiment, the above processor may be configured to execute the following steps through a computer program:

[0092] S1. Obtain the first encrypted calculation result of the target operator and the second encrypted calculation result of other operators. The first calculation result includes: multiple first feature weights respectively corresponding to multiple first data features in the target operator, and multiple first bin boundary values respectively corresponding to the multiple first feature weights. The second calculation result includes: multiple second feature weights respectively corresponding to multiple second data features in other operators, and multiple second bin boundary values respectively corresponding to the multiple second feature weights.

[0093] S2. Determine multiple third bin boundary values according to the multiple first feature weights, the multiple second feature weights, the multiple first bin boundary values, and the multiple second bin boundary values, where the multiple third bin boundary values are encrypted boundary values.

[0094] S3. Bin the multiple first data features according to the multiple third bin boundary values to obtain a binning result, and determine a splitting scheme for the target decision tree model according to the binning result, so as to split the leaf nodes in the target decision tree model according to the splitting scheme.

[0095] Optionally, in this embodiment, the above storage medium may include but is not limited to: various media such as USB flash drives, read-only memories (ROM for short), random access memories (RAM for short), mobile hard disks, magnetic disks, or optical discs that can store program codes.

[0096] Optionally, specific examples in this embodiment may refer to the examples described in the above embodiments and optional implementation manners, and will not be repeated here.

[0097] Obviously, those skilled in the art should understand that the above-mentioned modules or steps of the present invention can be implemented by a general-purpose computing device. They can be concentrated on a single computing device or distributed on a network composed of multiple computing devices. Optionally, they can be implemented by program codes executable by the computing device. Thus, they can be stored in a storage device and executed by the computing device. And in some cases, the steps shown or described can be executed in a different order from here, or they can be separately fabricated into individual integrated circuit modules, or multiple modules or steps among them can be fabricated into a single integrated circuit module for implementation. In this way, the present invention is not limited to any specific combination of hardware and software.

[0098] The above are only the preferred embodiments of the present invention and are not used to limit the present invention. For those skilled in the art, the present invention can have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A splitting method for a decision tree model, characterized in that, Including: Obtaining a first encrypted calculation result of a target operator and a second encrypted calculation result of other operators. The first calculation result includes multiple first feature weights corresponding to multiple first data features in the target operator, and multiple first bin boundary values corresponding to the multiple first feature weights. The second calculation result includes multiple second feature weights corresponding to multiple second data features in the other operators, and multiple second bin boundary values corresponding to the multiple second feature weights. Determining multiple third bin boundary values according to the multiple first feature weights, the multiple second feature weights, the multiple first bin boundary values, and the multiple second bin boundary values, where the multiple third bin boundary values are encrypted boundary values. Binning the multiple first data features according to the multiple third bin boundary values to obtain a binning result, and determining a splitting scheme for a target decision tree model according to the binning result, so as to split leaf nodes in the target decision tree model according to the splitting scheme. Among them, determining multiple third bin boundary values according to the multiple first feature weights, the multiple second feature weights, the multiple first bin boundary values, and the multiple second bin boundary values includes: obtaining a first feature weight and a first bin boundary value of a target data feature in the target operator, and obtaining a second feature weight and a second bin boundary value of the target data feature in the other operators. The multiple first data features include: The target data feature, the multiple first feature weights include: a first feature weight, and the multiple second feature weights include: a second feature weight; calculating a weighted sum of the first feature weight, the first bin boundary value, the second feature weight, and the second bin boundary value, and a weight sum of the first feature weight and the second feature weight; determining a third bin boundary value according to the weighted sum and the weight sum. Among them, determining a splitting scheme for a target decision tree model according to the binning result includes: determining a first feature histogram corresponding to the multiple first data features according to the binning result; aggregating the first feature histogram and a second feature histogram corresponding to the multiple second data features sent by other operators to obtain an aggregation result when receiving the second feature histogram; determining a splitting scheme for the target decision tree model according to the aggregation result.

2. The splitting method of the decision tree model according to claim 1, wherein Binning the data features again according to the third bin boundary values includes: Sending the third bin boundary values to a key management party, where the other operators include: the key management party; Receiving fourth bin boundary values sent by the key management party when the key management party decrypts the third bin boundary values through a target key, where the fourth bin boundary values are boundary values after decrypting the third bin boundary values; Binning the data features again according to the fourth bin boundary values.

3. The splitting method of the decision tree model according to claim 1, wherein After splitting the leaf nodes in the target decision tree model according to the splitting scheme, the method further includes: Calculating the loss function of the split target decision tree model; Determining whether the convergence degree of the loss function reaches a preset convergence condition; In the case where the convergence degree of the loss function does not reach the preset convergence condition, obtaining the third calculation result encrypted by the target operator and the fourth calculation result encrypted by the other operators.

4. The splitting method of the decision tree model according to claim 1, characterized in that Before obtaining the first calculation result of the target operator and the second calculation result encrypted by the other operators, the method further includes: Receiving the target key sent by the key management party; When the target operator calculates the multiple first feature weights corresponding to the multiple first data features and the multiple first bin boundaries corresponding to the multiple first feature weights, encrypting the multiple first feature weights and the multiple first bin boundaries by using the target key.

5. A splitting device for a decision tree model, characterized in that Includes: An obtaining module, configured to obtain the first calculation result encrypted by the target operator and the second calculation result encrypted by the other operators, where the first calculation result includes: multiple first feature weights corresponding to multiple first data features in the target operator, and multiple first bin boundaries corresponding to the multiple first feature weights, and the second calculation result includes: multiple second feature weights corresponding to multiple second data features in the other operators, and multiple second bin boundaries corresponding to the multiple second feature weights; A first determining module, configured to determine multiple third bin boundaries according to the multiple first feature weights, the multiple second feature weights, the multiple first bin boundaries, and the multiple second bin boundaries, where the multiple third bin boundaries are encrypted boundaries; A second determining module, configured to bin the multiple first data features according to the multiple third bin boundaries to obtain a binning result, and determine a splitting scheme of the target decision tree model according to the binning result, so as to split the leaf nodes in the target decision tree model according to the splitting scheme; Wherein, the first determining module is further configured to obtain the first feature weight and the first bin boundary of the target data feature in the target operator, and obtain the second feature weight and the second bin boundary of the target data feature in the other operator, where the multiple first data features include: the target data feature, the multiple first feature weights include: the first feature weight, and the multiple second feature weights include: the second feature weight; calculating the weighted sum of the first feature weight, the first bin boundary, the second feature weight, and the second bin boundary, and the weight sum of the first feature weight and the second feature weight; determining the third bin boundary according to the weighted sum and the weight sum; Among them, the second determination module is further configured to determine a first feature histogram corresponding to the multiple first data features according to the binning result; in the case of receiving a second feature histogram corresponding to the multiple second data features sent by other operators, aggregate the first feature histogram and the second feature histogram to obtain an aggregation result; and determine a splitting scheme of the target decision tree model according to the aggregation result.

6. The splitting device of the decision tree model according to claim 5, characterized in that, The apparatus further includes: a sending module, configured to send the third bin boundary value to a key management party, where the other operators include: the key management party; a receiving module, configured to receive a fourth bin boundary value sent by the key management party in the case that the key management party decrypts the third bin boundary value by using a target key, where the fourth bin boundary value is the boundary value after decrypting the third bin boundary value; The first determination module is further configured to perform binning on the data features again according to the fourth bin boundary value.

7. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored program, where the program, when running, executes the method described in any one of claims 1 to 4 above.

8. An electronic device, comprising a memory and a processor, characterized in that, A computer program is stored in the memory, and the processor is configured to execute the method described in any one of claims 1 to 4 through the computer program.

Images