Method, apparatus, device, and storage medium for creating a data prediction model

Abstract

The present invention provides a method, apparatus, device, and storage medium for creating a data prediction model. The method includes: the first device completes sample alignment with the second device to obtain a common first sample set, the first device determines the partitioning features of the first-layer nodes of the initial symmetric tree structure, and partitions the first sample set according to the partitioning features. The first device determines the partitioning features of any node in the second layer, and according to the partitioning features of the any node, partitions the sample sets of each node in the second layer respectively. The first device adopts the partitioning method for the sample sets of the second-layer nodes to sequentially partition the sample sets of each layer of nodes below the second layer of the symmetric tree structure until the preset condition is met to stop sample partitioning, and uses the symmetric tree structure at the stop time as the data prediction model. In the above creation process, starting from the second-layer nodes, after determining the partitioning features of one node in each layer, it is not necessary to determine the partitioning features of other nodes in that layer, improving the creation efficiency of the model.

Description

Technical Field

[0001] The present invention relates to the field of Internet technologies, and in particular, to a method, apparatus, device, and storage medium for creating a data prediction model. Background Art

[0002] With the rapid development of Internet technologies, in data platforms in various fields, the amount of user data has increased significantly. There is a large overlap of users in various fields, and data platforms in various fields can establish a vertical federated learning data prediction model based on multi-dimensional feature data of overlapping users to improve the prediction ability of the model.

[0003] Currently, the establishment of the above data prediction model is dominated by the initiator, assisted by each participant, and usually trained using a common decision tree structure (binary tree structure). The division conditions of each layer of nodes in this decision tree structure are different. For example, if the root node of the decision tree structure is divided according to feature a of the sample in the initiator, the subsequent levels of child nodes under the root node will not be divided according to feature a again. The establishment of the above data prediction model requires the initiator to confirm the sample division conditions of each node in the decision tree structure. For each node, the initiator needs to communicate with the participant to obtain the division result of the node sample in the participant's division.

[0004] Since the data prediction model uses a common decision tree, the division conditions of each node in each layer of the decision tree need to be confirmed by the initiator. Taking a 3-layer decision tree as an example, the initiator needs to confirm whether and how to divide the samples of 7 nodes. The efficiency of creating the above model is relatively low. Summary of the Invention

[0005] The main purpose of the present invention is to provide a method, apparatus, device, and storage medium for creating a data prediction model to improve the creation efficiency of the data prediction model.

[0006] In a first aspect, the present invention provides a method for creating a data prediction model. The method is applied to a federated system including a first device and a second device. The first device is the initiator device of the data prediction model, and the second device is the participant device of the data prediction model. The first device is used to create the data prediction model, and the second device is used to assist the first device in creating the data prediction model. The data prediction model uses a symmetric tree structure. The method includes:

[0007] The first device obtains a first sample set, where the first sample set is a sample set to be divided of the first layer nodes of an initial symmetric tree structure, and the first sample set includes samples shared by the first device and the second device;

[0008] The first device divides the first sample set according to the division characteristics of the first-layer nodes determined by the first device, and obtains the sample set of the second-layer nodes of the symmetric tree structure. The second-layer nodes include two nodes;

[0009] The first device divides the sample set of each node of the second-layer nodes respectively according to the division characteristics of any one node of the second-layer nodes determined by the first device, and obtains the division result of the second-layer nodes;

[0010] The first device uses the division method of the sample set of the second-layer nodes to divide the sample sets of each layer of nodes below the second layer of the symmetric tree structure respectively until the preset condition is met and the sample division stops. The symmetric tree structure at the stop is used as the data prediction model.

[0011] In an alternative embodiment of the first aspect, the first device divides the first sample set according to the division characteristics of the first-layer nodes determined by the first device, including:

[0012] The first device obtains the first sample purity value of the first device dividing the first sample set using the first feature, and the second sample purity value of the second device dividing the first sample set using the second feature; the first feature includes at least one feature of the sample in the first device, and the second feature includes at least one feature of the sample in the second device;

[0013] The first device determines the division characteristics of the first-layer nodes according to the first sample purity value and the second sample purity value;

[0014] The first device divides the first sample set according to the division characteristics of the first-layer nodes.

[0015] In an alternative embodiment of the first aspect, the first device obtains the first sample purity value of the first device dividing the first sample set using the first feature, including:

[0016] The first device determines the first sample purity value according to the first division result and the annotation result of the first sample set; the first division result is obtained by the first device dividing the first sample set using the first feature.

[0017] In an alternative embodiment of the first aspect, the first device obtains the second sample purity value of the second device dividing the first sample set using the second feature, including:

[0018] The first device obtains a second partitioning result from the second device, where the second partitioning result is obtained by the second device partitioning the first sample set using second features, and the second partitioning result is used by the first device to determine the partitioning features of the first-layer nodes;

[0019] The first device determines the second sample purity value according to the second partitioning result and the annotation result of the first sample set.

[0020] In an optional embodiment of the first aspect, the first device determines the partitioning features of the first-layer nodes according to the first sample purity value and the second sample purity value, including:

[0021] The first device uses the partitioning features corresponding to the maximum sample purity value as the partitioning features of the first-layer nodes.

[0022] In an optional embodiment of the first aspect, each layer of nodes below the second layer includes third-layer nodes, and the third-layer nodes include four nodes; the first device partitions the sample sets of each layer of nodes below the second layer of the symmetric tree structure according to the partitioning method of the sample set of the second-layer nodes by the first device, including:

[0023] The first device partitions the sample sets of each node of the third-layer nodes according to the partitioning features of any one node of the third-layer nodes determined by the first device.

[0024] In an optional embodiment of the first aspect, the preset condition includes any one of the following:

[0025] The number of samples in the sample sets of all nodes to be partitioned is less than or equal to a preset minimum number of samples;

[0026] Reaching the maximum number of layers of the preset symmetric tree structure;

[0027] Reaching the maximum number of leaf nodes of the preset symmetric tree structure;

[0028] The sample purity values of the sample sets of all nodes to be partitioned are greater than a preset threshold.

[0029] In an optional embodiment of the first aspect, the partitioning result of the second-layer nodes includes the sample sets of the child nodes of each node of the second-layer nodes, and the method further includes:

[0030] The first device sends the partitioning result of the second-layer nodes to the second device.

[0031] In a second aspect, the present invention provides a device for creating a data prediction model, and the device includes:

[0032] An acquisition module, configured to acquire a first sample set, where the first sample set is a sample set to be partitioned of the first-layer nodes of an initial symmetric tree structure, and the first sample set includes samples common to a first device and a second device;

[0033] A processing module, configured to partition the first sample set according to the partitioning features of the first-layer nodes determined by the first device, to obtain a sample set of the second-layer nodes of the symmetric tree structure, where the second-layer nodes include two nodes;

[0034] According to the partitioning features of any one of the second-layer nodes determined by the first device, partition the sample sets of each node of the second-layer nodes respectively, to obtain the partitioning result of the second-layer nodes;

[0035] Adopt the partitioning method for the sample set of the second-layer nodes, and partition the sample sets of each layer of nodes below the second layer of the symmetric tree structure respectively, until the preset condition is met to stop sample partitioning, and use the symmetric tree structure at the stop time as the data prediction model.

[0036] In an optional embodiment of the second aspect, the acquisition module is configured to:

[0037] Acquire a first sample purity value obtained by the first device partitioning the first sample set using a first feature, and a second sample purity value obtained by the second device partitioning the first sample set using a second feature; the first feature includes at least one feature of the sample in the first device, and the second feature includes at least one feature of the sample in the second device;

[0038] The processing module is configured to determine the partitioning features of the first-layer nodes according to the first sample purity value and the second sample purity value;

[0039] Partition the first sample set according to the partitioning features of the first-layer nodes.

[0040] In an optional embodiment of the second aspect, the processing module is configured to:

[0041] Determine the first sample purity value according to the first partitioning result and the annotation result of the first sample set; the first partitioning result is obtained by the first device partitioning the first sample set using the first feature.

[0042] In an optional embodiment of the second aspect, the acquisition module is configured to acquire a second partitioning result from the second device, where the second partitioning result is obtained by the second device partitioning the first sample set using the second feature, and the second partitioning result is used for the first device to determine the partitioning features of the first-layer nodes;

[0043] The processing module is configured to determine the second sample purity value according to the second partitioning result and the annotation result of the first sample set.

[0044] In an optional embodiment of the second aspect, the processing module is configured to:

[0045] Use the partitioning feature corresponding to the maximum sample purity value as the partitioning feature of the first-layer node.

[0046] In an optional embodiment of the second aspect, each layer of nodes below the second layer includes third-layer nodes, and each third-layer node includes four nodes; the processing module is configured to:

[0047] Partition the sample sets of each node of the third-layer nodes respectively according to the partitioning feature of any one node of the third-layer nodes determined by the first device.

[0048] In an optional embodiment of the second aspect, the preset condition includes any one of the following:

[0049] The number of samples in the sample sets of all nodes to be partitioned is less than or equal to a preset minimum number of samples;

[0050] Reach the maximum number of layers of a preset symmetric tree structure;

[0051] Reach the maximum number of leaf nodes of a preset symmetric tree structure;

[0052] The sample purity values of the sample sets of all nodes to be partitioned are greater than a preset threshold.

[0053] In an optional embodiment of the second aspect, the partitioning result of the second-layer nodes includes the sample sets of the child nodes of each node of the second-layer nodes, and the apparatus further includes: a sending module;

[0054] The sending module is configured to send the partitioning result of the second-layer nodes to the second device.

[0055] In a third aspect, the present invention provides an electronic device, including: a memory, a processor, and a computer program stored on the memory and executable on the processor, where the computer program, when executed by the processor, implements the method according to any one of the first aspects.

[0056] In a fourth aspect, the present invention provides a computer-readable storage medium, on which a computer program is stored, and the computer program, when executed by a processor, implements the method according to any one of the first aspects.

[0057] Fifth aspect, the present invention provides a computer program product, including a computer program, which when executed by a processor implements the method described in any one of the first aspect.

[0058] An embodiment of the present invention provides a method, apparatus, device, and storage medium for creating a data prediction model. The method includes: a first device and a second device complete sample alignment to obtain a common first sample set. The first device determines the division features of the first-layer nodes of the initial symmetric tree structure and divides the first sample set according to the division features. The first device determines the division features of any one node of the second-layer nodes and divides the sample sets of each node of the second-layer nodes respectively according to the division features of any one node. The first device uses the division method of the sample sets of the second-layer nodes to sequentially divide the sample sets of each layer of nodes below the second layer of the symmetric tree structure until the preset condition is met to stop sample division, and uses the symmetric tree structure at the stop as the data prediction model. During the above creation process, starting from the second-layer nodes, after determining the division features of one of the nodes of each layer, there is no need to determine the division features of the other nodes of that layer, which improves the division efficiency of each layer of nodes and thus improves the model creation efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 It is a schematic diagram of sample data for vertical federated learning provided by an embodiment of the present invention;

[0060] Figure 2 It is a schematic diagram of the system architecture of vertical federated learning provided by an embodiment of the present invention;

[0061] Figure 3 It is a schematic diagram of the structure of an existing data prediction model;

[0062] Figure 4 It is a schematic diagram of the structure of a data prediction model provided by an embodiment of the present invention;

[0063] Figure 5 It is a schematic diagram of the method for creating a data prediction model provided by an embodiment of the present invention Figure 1 ;

[0064] Figure 6 It is a schematic diagram of the method for creating a data prediction model provided by an embodiment of the present invention Figure 2 ;

[0065] Figure 7 It is a schematic diagram of the structure of the apparatus for creating a data prediction model provided by an embodiment of the present invention;

[0066] Figure 8 It is a schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0067] Exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present invention are shown in the drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided so that the present invention can be more thoroughly understood and the scope of the present invention can be fully conveyed to those skilled in the art.

[0068] The terms used in the embodiments of the present invention are only used to explain the specific embodiments of the present invention and are not intended to limit the present invention. Terms such as "first", "second", "third", and "fourth" in the specification and claims of the present invention and the accompanying drawings are used to distinguish different objects and not to describe a specific order. In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion.

[0069] To facilitate the understanding of the technical solutions of the embodiments of the present invention, the relevant concepts and principles of the embodiments of the present invention will be described first.

[0070] A symmetric tree belongs to a type of binary tree. If the structures of the left and right subtrees of a binary tree are symmetric, the binary tree is a symmetric tree.

[0071] Federated learning refers to a method of machine learning by combining different participants (participants, or parties, also known as data owners, or clients). In federated learning, participants do not need to expose the data they own to other participants and coordinators (coordinators, also known as parameter servers, or aggregation servers), so federated learning can well protect user privacy and ensure data security.

[0072] Federated learning can be divided into horizontal federated learning and vertical federated learning. For the convenience of understanding, horizontal federated learning and vertical federated learning will be described separately below.

[0073] Horizontal federated learning is to take out the part of the data with the same data features but not completely the same users among the participants when there are many overlapping data features and few overlapping users among the participants. For example, there are two banks in different regions. Their user groups come from their respective regions and have little intersection. However, their businesses are very similar, and a large part of the user data features they record are the same. Horizontal federated learning can be used to help the two banks build a joint model to predict the behavior of their customers.

[0074] Vertical federated learning is to take out the part of users and data with the same users but different user data characteristics for joint machine learning training when the data characteristics overlap of participants is small and the user overlap is large. For example, there are two participants A and B in the same region. Participant A is a bank and participant B is an e-commerce platform. Participants A and B have many same users in the same region, but the businesses of A and B are different, and the recorded user data characteristics are different. In particular, the user data characteristics recorded by A and B may be complementary. In such a scenario, vertical federated learning can be used to help A and B build a joint machine learning prediction model to help A and B provide better services to customers.

[0075] An embodiment of the present invention relates to vertical federated learning.

[0076] Figure 1 It is a schematic diagram of sample data for vertical federated learning provided by an embodiment of the present invention. As Figure 1 shown, taking the sample object as a user as an example, assume that the user groups of three enterprises A, B, and C completely overlap or partially overlap. The sample data of enterprise A includes characteristics such as the age, gender, and marital status of users, as well as the class label y of users (for example, the label y indicates whether the user will pay on time). The sample data of enterprise B includes characteristics such as the payment ability and education level of users. The sample data of enterprise C includes characteristic data such as the credit amount of users, etc.

[0077] The above three enterprises can create a data prediction model through vertical federated learning, which is used to predict data such as whether a certain user will pay on time.

[0078] The following Figure 2 is an explanation of the existing model creation process.

[0079] Figure 2 It is a schematic diagram of the system architecture of vertical federated learning provided by an embodiment of the present invention. As Figure 2 shown, the system architecture includes a first device 11, a second device 12, and a third device 13. The first device 11 is respectively communicatively connected to the second device 12 and the third device 13. Assume that the first device 11 is the initiator of vertical federated learning, and the second device 12 and the third device 13 are the participants of vertical federated learning. For example, the first device 11 can be the server of the above enterprise A, the second device 12 can be the server of the above enterprise B, and the third device 13 can be the server of the above enterprise C.

[0080] The first device 11 stores multiple first training samples (i.e., the samples of enterprise A) and the marking results corresponding to each first training sample (for example, Figure 1in category y), each first training sample includes at least one feature (e.g., Figure 1 age, gender, marital status in Figure 1 ). Multiple second training samples (i.e., samples of enterprise B) are stored in the second device 12, and each second training sample includes at least one feature (e.g., Figure 1 payment ability, education level in

[0081] ). Multiple third training samples (i.e., samples of enterprise C) are stored in the third device 13, and each third training sample includes at least one feature (e.g.,

[0082] credit amount in

[0083] The first device 11 exchanges part of the data with the second device 12. For example, the first device 11 sends the user identification ID stored in the first device 11 to the second device 12, and the second device 12 sends the user identification stored in the second device 12 to the first device 11. Through the data exchange, the first device 11 and the second device 12 determine which users exist in both parties, thereby completing the alignment of user samples between the first device 11 and the second device 12. Similarly, the first device 11 exchanges part of the data with the third device 13 to complete the alignment of user samples. Finally, the alignment of user samples among the three devices is achieved.

[0084] After the alignment of the three-party samples is completed, with the first device 11 as the initiator as the dominant party and the second device 12 and the third device 13 as the participating parties as the auxiliary parties, a data prediction model is created. The first device 11 initializes the tree structure, and based on the aligned training sample set, starting from the root node of the tree structure, the division conditions (e.g., division features) of each node in the tree structure are determined in turn. When the first device 11 determines the division conditions of each node in the tree structure, it needs to interact with the second device 12 and the third device 13 to obtain the division situation of each participating party for each node, so as to determine which party's feature is used for the division of each node. Based on the above division method, the division conditions of each node are determined until the preset stop condition is met, and the tree structure at the stop time is used as the final data prediction model, and the division conditions of each node of this model are determined.

[0085] Figure 3 It is a schematic structural diagram of an existing data prediction model, Figure 4 and is a schematic structural diagram of the data prediction model provided by the embodiments of the present invention.

[0086] Figure 3 The data prediction model shown adopts an ordinary decision tree structure. The root node of the decision tree structure is Figure 3 node 1, and the division feature corresponding to node 1 is feature a. The left and right child nodes of node 1 are node 2 and node 3 respectively. The division feature corresponding to node 2 is feature b, and the division feature corresponding to node 3 is node c. The child nodes of node 2 are node 4 and node 5, and the child nodes of node 3 are node 6 and node 7. The division features corresponding to nodes 4 to 7 are features d, e, f, and g respectively. It can be seen that the division features of each node in the decision tree structure are different, and features a - g are all sample features, and some of them may belong to the features of the sample at the initiating party, and some may belong to the features of the sample at the participating party.

[0087] Figure 4 The data prediction model shown adopts a symmetric tree structure. The symmetric tree structure also has three layers. The root node of the symmetric tree structure (i.e., Figure 4 node 1) corresponds to the division feature of feature a. The left and right child nodes of node 1 (i.e., Figure 4 node 2 and node 3) correspond to the same division feature, such as feature b. The left and right child nodes of node 2 (i.e., Figure 4 node 4 and node 5) and the left and right child nodes of node 3 (i.e., Figure 4 nodes 6 and 7 in) correspond to the same division feature, such as feature c. Among them, feature a, feature b, and feature c are all sample features, and some of them may belong to the features of the sample at the initiating party, and some may belong to the features of the sample at the participating party.

[0088] Based on the above two tree structures, if the number of layers of the tree structure is set to be the same, when establishing a data prediction model with the same number of layers, since the division features of the nodes in the same layer of the symmetric tree structure are the same, the creation speed of the data prediction model based on the symmetric tree structure is faster.

[0089] It should be noted that for the convenience of description, the embodiments of the present invention take the interaction between two devices as an example to describe the creation process of the data prediction model. In actual applications, the creation of a vertical federated learning model may be based on three or more devices, and its creation principle is similar to that based on two devices.

[0090] The following will specifically describe the creation process of the data prediction model in combination with specific embodiments. These specific embodiments below can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments.

[0091] Figure 5 Schematic of the method for creating a data prediction model provided by an embodiment of the present invention Figure 1 This method is applied to a federated system including a first device and a second device. The first device is the initiating device of the data prediction model, and the second device is the participating device of the data prediction model. The first device is used to create the data prediction model, and the second device is used to assist the first device in creating the data prediction model. The data prediction model in this embodiment adopts a symmetric tree structure, and its creation process involves the interaction between the first device and the second device.

[0092] As Figure 5 shown, the model creation method in this embodiment includes the following steps:

[0093] Step 101: The first device aligns samples with the second device to obtain a first sample set.

[0094] The first sample set is the sample set to be divided for the first-layer nodes of the initial symmetric tree structure, and the first sample set includes the samples common to the first device and the second device.

[0095] In a possible implementation manner, the first device and the second device determine the samples common to the devices by interacting with the sample identification IDs (for example, Figure 1 user ID in) in their respective devices. Optionally, the first device may use an encryption algorithm such as a public key encryption algorithm or a hash algorithm to encrypt its sample ID and send it to the second device, and the second device may also use the same encryption algorithm to encrypt its sample ID and send it to the first device. By interacting with the sample IDs between the devices, the samples common to the devices are obtained.

[0096] Optionally, the first sample set includes all the samples common to the first device and the second device.

[0097] Optionally, the first sample set includes some samples common to the first device and the second device, and these some samples may be randomly selected by the first device. It should be noted that after the first device randomly selects some samples, it needs to synchronize them to the second device.

[0098] The first device and the second device create a data prediction model based on the samples in the first sample set determined in this step.

[0099] Step 102: The first device divides the first sample set according to the division characteristics of the first-layer nodes determined by the first device to obtain the sample set of the second-layer nodes of the symmetric tree structure.

[0100] Among them, the first-layer nodes are the root nodes of the initial symmetric tree structure, and the number of first-layer nodes is 1. The second-layer nodes include two nodes, namely the left child node and the right child node of the first-layer node.

[0101] In a possible implementation, the partitioning feature of the first-level nodes determined by the first device is randomly selected from the sample features, and the sample features include the features of the sample on the first device and the features of the sample on the second device.

[0102] If the randomly selected partitioning feature belongs to the features of the first device, the first device partitions the first sample set according to the partitioning feature of the first device to obtain the sample set of the second-level nodes.

[0103] If the randomly selected partitioning feature belongs to the features of the second device, the first device interacts with the second device to obtain the sample set of the second-level nodes from the second device. Among them, the second device partitions the first sample set according to the partitioning result of the partitioning feature of the second device (i.e., the sample set of the second-level nodes), and sends the partitioning result to the first device.

[0104] In another possible implementation, the first device determines the partitioning feature of the first-level nodes by obtaining the first partitioning result and the second partitioning result. The first partitioning result is obtained by the first device partitioning the first sample set according to at least one feature of the sample on the first device, and the second partitioning result is obtained by the second device partitioning the first sample set according to at least one feature of the sample on the second device.

[0105] Optionally, the first device selects the feature corresponding to the partitioning result with the highest partitioning accuracy from multiple partitioning results, and uses this feature as the partitioning feature of the first-level nodes.

[0106] Step 103: The first device sends the sample set of the second-level nodes to the second device.

[0107] After the first device determines the partitioning feature of the first-level nodes and partitions the first sample set according to this partitioning feature, it obtains the sample set of the second-level nodes, and this sample set is the sample set of the left and right child nodes of the first-level nodes.

[0108] Step 104: The first device partitions the sample set of each node of the second-level nodes respectively according to the partitioning feature of any one node of the second-level nodes determined by the first device to obtain the partitioning result of the second-level nodes.

[0109] In a possible implementation, the first device randomly selects a node from the second-level nodes. For example, it selects the left child node of the first-level node. The first device first determines the partitioning feature of this left child node, and then partitions the sample set of the left child node and the sample set of the right child node of the first-level node according to the partitioning feature of this left child node, and obtains the sample set of the child nodes of this left child node and the sample set of the child nodes of this right child node respectively. The partitioning result of the second-level nodes includes the sample set of the child nodes of this left child node and the sample set of the child nodes of this right child node.

[0110] In this embodiment, the first device only determines the partitioning feature of one of the nodes in the second layer of nodes, and partitions the sample set of each node in the second layer of nodes according to the partitioning feature of this node, obtaining the partitioning result of the second layer of nodes, which improves the partitioning efficiency of the second layer of nodes.

[0111] Under normal circumstances, the first device needs to sequentially determine the partitioning feature of each node in the second layer of nodes, and the determination process involves multiple interactions with the second device, resulting in a large amount of communication. By comparison, this embodiment can reduce the number of interactions between the first device and the second device, reduce the communication volume of data transmission to a certain extent, improve the sample partitioning efficiency, and thus improve the creation efficiency of the data prediction model.

[0112] Step 105: The first device sends the partitioning result of the second layer of nodes to the second device.

[0113] The second device receives the partitioning result of the second layer of nodes, and can obtain the sample set of the child nodes of each node in the second layer of nodes (i.e., the sample set of the third layer of nodes).

[0114] In some embodiments, the second device can perform sample partitioning on the sample set of the third layer of nodes according to the partitioning result of the second layer of nodes. That is, the partitioning result of the second layer of nodes can be used by the second device to perform sample partitioning on the sample set of the child nodes of each node in the second layer of nodes.

[0115] Optionally, the partitioning result of the second layer of nodes further includes the partitioning feature of the second layer of nodes. The partitioning feature of each node in the second layer of nodes is the same.

[0116] Step 106: The first device uses the partitioning method for the sample set of the second layer of nodes to partition the sample set of each layer of nodes below the second layer of the symmetric tree structure respectively until the preset condition is met and the sample partitioning stops.

[0117] In this embodiment, the partitioning method for the sample set of each layer of nodes below the second layer by the first device is similar to the partitioning method for the sample set of the second layer of nodes, that is, the first device only needs to determine the partitioning feature of any one node in each layer of nodes, and then uses the partitioning feature of this node to complete the sample partitioning of all nodes in this layer.

[0118] Taking the third layer of nodes as an example, the first device partitions the sample set of each node in the third layer of nodes according to the partitioning feature of any one node in the third layer of nodes determined by the first device. For example, Figure 4 in, the first device determines that the partitioning feature of the third layer of node 4 is feature c, then the first device partitions the sample sets of the third layer of nodes 4-7 according to feature 4 respectively.

[0119] In this embodiment, the preset conditions include any one of the following:

[0120] Condition 1: The number of samples in the sample set of all nodes to be partitioned is less than or equal to a preset minimum number of samples;

[0121] Condition 2: Reaching the maximum number of layers of a preset symmetric tree structure;

[0122] Condition 3: Reaching the maximum number of leaf nodes of a preset symmetric tree structure;

[0123] Condition 4: The sample purity value of the sample set of all nodes to be partitioned is greater than a preset threshold.

[0124] The above preset conditions are described below through specific examples.

[0125] Exemplarily, taking Condition 1 as an example, assume that the current layer is the third layer of the symmetric tree structure, and the number of samples in the sample set of a certain node in the third layer nodes is less than or equal to the preset minimum number of samples (for example, the minimum number of samples is 2), then the partitioning of this node is stopped. It should be understood that when the number of samples in the sample sets of all nodes to be partitioned is less than or equal to the minimum number of samples, the creation of the symmetric tree structure is completed.

[0126] Exemplarily, taking Condition 2 as an example, assume that the maximum number of layers of the preset symmetric tree structure is 3 layers. If the current layer is the third layer, then the samples of the third layer nodes are no longer partitioned.

[0127] Exemplarily, taking Condition 3 as an example, assume that the maximum number of leaf nodes of the preset symmetric tree structure is 4. If the current layer is the second layer and the second layer includes 2 nodes, then the second layer nodes are further partitioned until the number of leaf nodes reaches 4 and the partitioning stops.

[0128] Exemplarily, taking Condition 4 as an example, assume that the current layer is the third layer of the symmetric tree structure, and the sample purity value of the sample set of a certain node in the third layer nodes is greater than the preset threshold, indicating that the sample purity (or the feature similarity of the samples) of this node is high, then the partitioning of this node is stopped. It should be understood that when the sample purity values of the sample sets of all nodes to be partitioned are greater than the preset threshold, the creation of the symmetric tree structure is completed.

[0129] Step 107: The first device and the second device use the symmetric tree structure at the stop time as the data prediction model.

[0130] It should be noted that in this embodiment, the main body of sample partitioning is the first device, that is, the initiating device. Each time the first device completes the partitioning of a node, it needs to synchronize the partitioning result to the second device, and the process of transmitting the partitioning result also needs to be encrypted to ensure the security of data transmission.

[0131] The method for creating the data prediction model shown in this embodiment is that the first device and the second device complete sample alignment to obtain a common first sample set. The first device determines the division features of the first-layer nodes of the initial symmetric tree structure and divides the first sample set according to the division features. The first device determines the division features of any node in the second layer and divides the sample sets of each node in the second layer according to the division features of any node. The first device uses the division method of the sample sets of the second-layer nodes to sequentially divide the sample sets of each layer of nodes below the second layer of the symmetric tree structure until the preset condition is met to stop sample division, and uses the symmetric tree structure at this time as the data prediction model. During the above creation process, starting from the second-layer nodes, after determining the division features of one of the nodes in each layer, it is no longer necessary to determine the division features of the other nodes in this layer, which improves the division efficiency of each layer of nodes and thus improves the model creation efficiency.

[0132] The following uses a specific embodiment to elaborate in detail how the above first device divides the sample set of the first-layer nodes (i.e., the root nodes).

[0133] Figure 6 Schematic of the method for creating the data prediction model provided by the embodiment of the present invention Figure 2 . After the first device and the second device complete sample alignment, the second device obtains the first sample set (i.e., step 101 of the above embodiment). The model creation method includes the following steps:

[0134] Step 201: The second device divides the first sample set using the second feature to obtain a second division result. The second feature includes at least one feature of the sample in the second device.

[0135] Step 202: The second device sends the second division result to the first device.

[0136] Exemplarily, taking the sample as a user, if the second feature includes feature d (such as Figure 1 the payment ability of the user in the middle). The second device divides the first sample set using feature d to obtain a second division result. For example, if the payment ability of the user is greater than or equal to 5000 yuan, the user is divided into the left child node of the first-layer node; if the payment ability of the user is less than 5000 yuan, the user is divided into the right child node of the first-layer node. The second division result includes the IDs of the users in the left and right child nodes of the first-layer node when dividing the users using feature d. For example, the second division result includes feature d (after encryption processing), the user IDs in the left child node of the first-layer node, and the user IDs in the right child node of the first-layer node. The second device sends the second division result to the first device so that the first device can determine whether the samples of the first-layer node are divided using the local features of the second device or the local features of the samples in the first device.

[0137] In some embodiments, the second feature includes two or more features, and the second partitioning result includes multiple partitioning results of the first sample set partitioned by different second features by the second device.

[0138] Step 203: The first device determines a second sample purity value according to the second partitioning result and the annotation result of the first sample set.

[0139] The annotation result of the first sample set includes the category of each sample in the first sample set. Exemplarily, taking the sample as a user, the user categories include 0 and 1, where 0 means the user will pay on time, and 1 means the user will not pay on time.

[0140] The first device can determine the sample purity value of the left child node and the sample purity value of the right child node of the first-level node according to the sample identifiers of the left and right child nodes of the first-level node in the second partitioning result and the category of each sample in the annotation result. For example, if there are 10 users in the first-level node and after partitioning using the above-mentioned feature d, the left and right child nodes of the first-level node have 6 users and 4 users respectively. Assuming that 5 out of the 6 users in the left child node are of category 0 and all 4 users in the right child node are of category 1, then the sample purity value of the left child node is 5 / 6, and the sample purity value of the right child node is 1.

[0141] The first device determines the second sample purity value of the first-level node partitioned by the second feature according to the sample purity values of the left and right child nodes of the first-level node. Optionally, the first device determines the second sample purity value according to the average value of the sample purity values of the left and right child nodes of the first-level node. For example, based on the sample purity values of the left and right child nodes of the first-level node in the above example, the second sample purity value = (5 / 6 + 1) / 2 = 11 / 12.

[0142] Step 204: The first device obtains the first sample purity value of the first sample set partitioned by the first device using the first feature.

[0143] In this embodiment, the first device determines the first sample purity value according to the first partitioning result and the annotation result of the first sample set. The first partitioning result is obtained by the first device partitioning the first sample set using the first feature. The first feature includes at least one feature of the sample in the first device.

[0144] Exemplarily, taking the sample as a user, if the first feature includes feature a (for example Figure 1The age of the user in the first device. The first device divides the first sample set using feature a to obtain a first division result. For example, if the user's age is greater than or equal to 50 years old, the user is divided into the left child node of the first-level node; if the user's age is less than 50 years old, the user is divided into the right child node of the first-level node. The first division result includes the IDs of the users in the left and right child nodes of the first-level node when dividing the users using feature a.

[0145] In some embodiments, the first feature includes two or more features, and the first division result includes multiple division results of the first device dividing the first sample set using different first features.

[0146] In this embodiment, the principle and steps for the first device to determine the first sample purity value based on the first division result and the annotation result of the first sample set are similar to those for determining the second sample purity value in step 203, which can be referred to above and will not be elaborated here.

[0147] Step 205: The first device determines the division feature of the first-level node according to the first sample purity value and the second sample purity value.

[0148] In a possible implementation, the first device uses the division feature corresponding to the maximum sample purity value as the division feature of the first-level node. The division feature corresponding to the maximum sample purity value may be a feature of the sample in the first device or a feature of the sample in the second device.

[0149] The above implementation uses the feature with the best division effect (i.e., the highest division accuracy) as the division feature of the first-level node to ensure the division effect of this node.

[0150] Step 206: The first device divides the first sample set according to the division feature of the first-level node.

[0151] The first device divides the first sample set to obtain the division result of the first-level node. The first device sends the division result of the first-level node to the second device, and the second device can obtain the sample set of the child nodes of the first-level node (i.e., the sample set of the second-level node).

[0152] In some embodiments, the second device can divide the sample set of the second-level node according to the division result of the first-level node. That is, the division result of the first-level node can be used by the second device to divide the sample set of the child nodes of the first-level node.

[0153] Optionally, the division result of the first-level node further includes the division feature of the first-level node.

[0154] If the division feature of the first-level node belongs to the local feature of the first device, the first device can encrypt the division feature of the first-level node to prevent the second device from obtaining the information of the division feature.

[0155] If the partitioning feature of the first - layer node belongs to the local feature of the second device, the first device directly sends the partitioning feature to the second device. It should be understood that this partitioning feature has already been the feature encrypted by the second device (the first device obtains it from the second partitioning result in step 202).

[0156] It should be noted that in step 105 of the above - mentioned embodiment, the partitioning feature of the second - layer node may be the local feature of the first device or the local feature of the second device. The partitioning feature of the second - layer node transmitted by the first device to the second device is also a feature that has been encrypted, and its implementation principle is the same as that of the partitioning feature of the first - layer node in this embodiment.

[0157] It should be noted that the execution order of each step in this embodiment is only taken as an example and should not constitute any limitation to the present invention.

[0158] It should be understood that the principle of determining the partitioning feature of any node in the symmetric tree structure is similar to that of determining the partitioning feature of the first - layer node. The partitioning feature of any node in the symmetric tree structure can be determined based on the technical solution of this embodiment.

[0159] In the method for creating a data prediction model shown in this embodiment, the first device analyzes the partitioning situation of the root - node sample set by the first device using the first feature and the partitioning situation of the root - node sample set by the second device using the second feature, selects the target feature with the best partitioning effect (i.e., the highest partitioning accuracy) from the first feature and the second feature, and uses this target feature as the partitioning feature of the root node. The first device partitions the root - node sample set based on this partitioning feature. The first feature includes at least one feature of the sample in the first device, and the second feature is at least one feature of the sample in the second device. The above process is also applicable to other nodes in the symmetric tree structure. The data prediction model created by the above method has a high prediction accuracy.

[0160] Embodiments of the present invention can perform functional - module partitioning on the device for creating a data prediction model (the first device or the initiating device of federated learning) according to the above - mentioned method embodiments. For example, each functional module can be corresponding to each function, or two or more functions can be integrated into one processing module. The above - integrated module can be implemented in the form of hardware or in the form of a software functional module. It should be noted that the partitioning of modules in embodiments of the present invention is illustrative, only a logical functional partitioning, and there may be other partitioning methods in actual implementation. The following takes the example of partitioning each functional module corresponding to each function for explanation.

[0161] Figure 7 It is a schematic structural diagram of the device for creating a data prediction model provided by an embodiment of the present invention. AsFigure 7 As shown in Figure 7 , the creation device 300 of the data prediction model in this embodiment includes: an acquisition module 301, a processing module 302, and a sending module 303.

[0162] The acquisition module 301 is used to acquire a first sample set, where the first sample set is a sample set to be divided of the first-layer nodes of an initial symmetric tree structure, and the first sample set includes samples common to a first device and a second device;

[0163] The processing module 302 is used to divide the first sample set according to the division features of the first-layer nodes determined by the first device, and obtain a sample set of the second-layer nodes of the symmetric tree structure, where the second-layer nodes include two nodes;

[0164] According to the division features of any one of the second-layer nodes determined by the first device, divide the sample sets of each node of the second-layer nodes respectively, and obtain the division results of the second-layer nodes;

[0165] Adopt the division method of the sample set of the second-layer nodes to divide the sample sets of each layer of nodes below the second layer of the symmetric tree structure respectively, and stop sample division until a preset condition is met. Take the symmetric tree structure at the stop as the data prediction model.

[0166] In an alternative embodiment of this embodiment, the acquisition module 301 is used to:

[0167] Acquire a first sample purity value obtained by the first device dividing the first sample set using a first feature, and a second sample purity value obtained by the second device dividing the first sample set using a second feature; the first feature includes at least one feature of the sample in the first device, and the second feature includes at least one feature of the sample in the second device;

[0168] The processing module 302 is used to determine the division features of the first-layer nodes according to the first sample purity value and the second sample purity value;

[0169] Divide the first sample set according to the division features of the first-layer nodes.

[0170] In an alternative embodiment of this embodiment, the processing module 302 is used to:

[0171] Determine the first sample purity value according to the first division result and the annotation result of the first sample set; the first division result is obtained by the first device dividing the first sample set using the first feature.

[0172] In an alternative embodiment of the present embodiment, the obtaining module 301 is configured to obtain a second partitioning result from the second device, where the second partitioning result is obtained by the second device partitioning the first sample set using a second feature, and the second partitioning result is used by the first device to determine the partitioning feature of the first layer of nodes;

[0173] The processing module 302 is configured to determine the second sample purity value according to the second partitioning result and the annotation result of the first sample set.

[0174] In an alternative embodiment of the present embodiment, the processing module 302 is configured to:

[0175] Use the partitioning feature corresponding to the maximum sample purity value as the partitioning feature of the first layer of nodes.

[0176] In an alternative embodiment of the present embodiment, each layer of nodes below the second layer includes a third layer of nodes, and the third layer of nodes includes four nodes; the processing module 302 is configured to:

[0177] Partition the sample sets of each node of the third layer of nodes according to the partitioning feature of any one node of the third layer of nodes determined by the first device.

[0178] In an alternative embodiment of the present embodiment, the preset condition includes any one of the following:

[0179] The number of samples in the sample sets of all nodes to be partitioned is less than or equal to a preset minimum number of samples;

[0180] Reach the maximum number of layers of the preset symmetric tree structure;

[0181] Reach the maximum number of leaf nodes of the preset symmetric tree structure;

[0182] The sample purity values of the sample sets of all nodes to be partitioned are greater than a preset threshold.

[0183] In an alternative embodiment of the present embodiment, the partitioning result of the second layer of nodes includes the sample sets of the child nodes of each node of the second layer of nodes, and the apparatus further includes: a sending module 303;

[0184] The sending module 303 is configured to send the partitioning result of the second layer of nodes to the second device.

[0185] The data prediction model creation apparatus provided by the embodiments of the present invention is used to execute the technical solutions of the first device in any of the foregoing method embodiments, and its implementation principles and technical effects are similar and will not be elaborated here.

[0186] Figure 8The structural schematic diagram of the electronic device provided by the embodiment of the present invention. As Figure 8 shown, the electronic device 400 of this embodiment may include:

[0187] At least one processor 401 ( Figure 8 only one processor is shown in

[0188] it); and

[0189] A memory 402 communicatively connected to the at least one processor; wherein,

[0190] The memory 402 stores a computer program executable by the at least one processor 401. When the computer program is executed by the at least one processor 401, the electronic device 400 can execute the technical solution of the first device in any of the foregoing method embodiments. The implementation principle and technical effect are similar and will not be elaborated here.

[0191] Optionally, the memory 402 can be either independent or integrated with the processor 401.

[0192] When the memory 402 is a device independent of the processor 401, the electronic device 400 further includes: a bus 403 for connecting the memory 402 and the processor 401.

[0193] The embodiment of the present invention also provides a computer-readable storage medium storing a computer program, which is used to implement the technical solution of the first device in any of the foregoing method embodiments when the computer program is executed by a processor.

[0193] The embodiment of the present invention also provides a computer program product including a computer program, which implements the technical solution of the first device in any of the foregoing method embodiments when the computer program is executed by a processor.

[0194] The embodiment of the present invention also provides a chip including a processing module and a communication interface, and the processing module can execute the technical solution of the first device in any of the foregoing method embodiments.

[0195] Optionally, the chip further includes a storage module (such as a memory). The storage module is used to store instructions, and the processing module is used to execute the instructions stored in the storage module. And the execution of the instructions stored in the storage module enables the processing module to execute the technical solution of the first device in any of the foregoing method embodiments.

[0196] It should be understood that the above-mentioned processor may be a Central Processing Unit (CPU), or may also be other general-purpose processors, Digital Signal Processors (DSPs), Application Specific Integrated Circuits (ASICs), etc. The general-purpose processor may be a microprocessor or any conventional processor, etc. The steps of the method disclosed in combination with the invention can be directly embodied as being executed and completed by a hardware processor, or by a combination of hardware and software modules in the processor.

[0197] The memory may include high-speed RAM memory, and may also include non-volatile storage NVM, such as at least one disk memory, and may also be a USB flash drive, a mobile hard disk, a read-only memory, a magnetic disk, or an optical disc, etc.

[0198] The bus may be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, an Extended Industry Standard Architecture (EISA) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For the sake of convenience in representation, the bus in the drawings of the present invention is not limited to only one bus or one type of bus.

[0199] The above-mentioned storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read-Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), magnetic memory, flash memory, a magnetic disk, or an optical disc. The storage medium can be any available medium that can be accessed by a general-purpose or special-purpose computer.

[0200] An exemplary storage medium is coupled to the processor, so that the processor can read information from the storage medium and write information to the storage medium. Of course, the storage medium can also be a component of the processor. The processor and the storage medium can be located in an Application Specific Integrated Circuit (ASIC). Of course, the processor and the storage medium can also exist as discrete components in an electronic device.

[0201] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for creating a data prediction model, characterized in that, Applied to a federated system including a first device and a second device, where the first device is the initiating device of the data prediction model, the second device is the participating device of the data prediction model, the first device is used to create the data prediction model, the second device is used to assist the first device in creating the data prediction model, and the data prediction model adopts a symmetric tree structure. The method includes: The first device obtains a first sample set, which is the sample set to be divided for the first-layer nodes of the initial symmetric tree structure, and the first sample set includes samples shared by the first device and the second device; The first device divides the first sample set according to the division feature of the first-layer nodes determined by the first device, and obtains the sample set of the second-layer nodes of the symmetric tree structure. The second-layer nodes include two nodes; The first device divides the sample set of each node of the second-layer nodes respectively according to the division feature of any one node of the second-layer nodes determined by the first device, and obtains the division result of the second-layer nodes; The first device uses the division method of the sample set of the second-layer nodes to divide the sample sets of each layer of nodes below the second layer of the symmetric tree structure respectively until the preset condition is met to stop sample division, and takes the symmetric tree structure at the stop as the data prediction model. The first device divides the first sample set according to the division feature of the first-layer nodes determined by the first device, including: The first device obtains the first sample purity value of the first device dividing the first sample set using the first feature, and the second sample purity value of the second device dividing the first sample set using the second feature. The first feature includes at least one feature of the sample in the first device, and the second feature includes at least one feature of the sample in the second device; The first device determines the division feature of the first-layer nodes according to the first sample purity value and the second sample purity value; The first device divides the first sample set according to the division feature of the first-layer nodes.

2. The method according to claim 1, wherein The first device obtains the first sample purity value of the first device dividing the first sample set using the first feature, including: The first device determines the first sample purity value according to the first division result and the annotation result of the first sample set. The first division result is obtained by the first device dividing the first sample set using the first feature.

3. The method according to claim 1, wherein The first device obtains the second sample purity value of the second device dividing the first sample set using the second feature, including: The first device obtains the second division result from the second device. The second division result is obtained by the second device dividing the first sample set using the second feature, and the second division result is used for the first device to determine the division feature of the first-layer nodes; The first device determines the second sample purity value according to the second division result and the annotation result of the first sample set.

4. The method according to claim 1, wherein The first device determines the partitioning characteristics of the first-layer nodes according to the first sample purity value and the second sample purity value, including: The first device uses the partitioning characteristic corresponding to the maximum sample purity value as the partitioning characteristic of the first-layer nodes.

5. The method according to claim 1, wherein Each layer of nodes below the second layer includes third-layer nodes, and the third-layer nodes include four nodes; the first device partitions the sample sets of each layer of nodes below the second layer of the symmetric tree structure according to the partitioning method of the sample set of the second-layer nodes by the first device, including: The first device partitions the sample sets of each node of the third-layer nodes respectively according to the partitioning characteristic of any one node of the third-layer nodes determined by the first device.

6. The method according to claim 1, characterized in that, The preset condition includes any one of the following: The number of samples in the sample sets of all nodes to be partitioned is less than or equal to a preset minimum number of samples; Reaching the maximum number of layers of the preset symmetric tree structure; Reaching the maximum number of leaf nodes of the preset symmetric tree structure; The sample purity values of the sample sets of all nodes to be partitioned are greater than a preset threshold.

7. The method according to any one of claims 1 to 6, characterized in that The partitioning result of the second-layer nodes includes the sample sets of the child nodes of each node of the second-layer nodes, and the method further includes: The first device sends the partitioning result of the second-layer nodes to the second device.

8. An apparatus for creating a data prediction model, characterized in that, The data prediction model adopts a symmetric tree structure, and the device includes: An acquisition module, configured to acquire a first sample set, where the first sample set is a sample set to be partitioned of the first-layer nodes of an initial symmetric tree structure, and the first sample set includes samples shared by the first device and the second device; A processing module, configured to partition the first sample set according to the partitioning characteristics of the first-layer nodes determined by the first device, to obtain the sample sets of the second-layer nodes of the symmetric tree structure, where the second-layer nodes include two nodes; According to the partitioning characteristic of any one node of the second-layer nodes determined by the first device, partition the sample sets of each node of the second-layer nodes respectively, to obtain the partitioning result of the second-layer nodes; Adopt the partitioning method of the sample set of the second-layer nodes to partition the sample sets of each layer of nodes below the second layer of the symmetric tree structure respectively until the preset condition is met and the sample partitioning stops, and use the symmetric tree structure at the stop as the data prediction model; The acquisition module is configured to: Acquire the first sample purity value obtained by the first device partitioning the first sample set using a first feature, and the second sample purity value obtained by the second device partitioning the first sample set using a second feature; the first feature includes at least one feature of the sample in the first device, and the second feature includes at least one feature of the sample in the second device; The processing module is configured to determine the partitioning characteristics of the first-layer nodes according to the first sample purity value and the second sample purity value; Partition the first sample set according to the partitioning characteristics of the first-layer nodes.

9. An electronic device, characterized in that, Include: A memory, a processor, and a computer program stored on the memory and executable on the processor, the computer program, when executed by the processor, implementing the method according to any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, A computer program is stored on the computer-readable storage medium, and when the computer program is executed by a processor, the method according to any one of claims 1-7 is implemented.

11. A computer program product, characterized in that, It includes a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1-7 is implemented.

Images