Model processing method, device, storage medium and electronic device

Abstract

The present specification discloses a model processing method, device, storage medium and electronic device, wherein the method comprises: obtaining user transaction training data for an initial behavior recognition model, extracting transaction training features in the user transaction training data, and then performing causal inference processing on the transaction training features through the initial behavior recognition model to obtain causal inference features, and then performing model training on the initial behavior recognition model based on the transaction training features and the causal inference features, so as to obtain a behavior recognition model after model training.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a model processing method, device, storage medium and electronic device. Background Art

[0002] In daily transaction scenarios, it involves predicting the transaction behavior of online users. In such online transaction scenarios, recognition models based on machine learning are often used to implement this. Usually, machine learning is used to collect user data over a period of time on the current transaction platform to train the corresponding recognition model, and then the trained model is put online and applied to daily transaction scenarios. Summary of the invention

[0003] This specification provides a model processing method, device, storage medium and electronic device, and the technical solution is as follows:

[0004] In a first aspect, this specification provides a model processing method, the method comprising:

[0005] Acquire user transaction training data for the initial behavior recognition model, and extract transaction training features from the user transaction training data;

[0006] Performing causal inference processing on the transaction training features through the initial behavior recognition model to obtain causal inference features;

[0007] Model training is performed on the initial behavior recognition model based on the transaction training features and the causal inference features to obtain a behavior recognition model after model training.

[0008] In a second aspect, this specification provides a model processing device, the device comprising:

[0009] A feature extraction module, used to obtain user transaction training data for the initial behavior recognition model, and extract transaction training features from the user transaction training data;

[0010] A causal inference module, used to perform causal inference processing on the transaction training features through the initial behavior recognition model to obtain causal inference features;

[0011] The model processing module is used to perform model training on the initial behavior recognition model based on the transaction training features and the causal inference features to obtain a behavior recognition model after model training.

[0012] In a third aspect, the present specification provides a computer storage medium, wherein the computer storage medium stores a plurality of instructions, wherein the instructions are suitable for being loaded by a processor and executing the above-mentioned method steps.

[0013] In a fourth aspect, the present specification provides an electronic device, which may include: a processor and a memory; wherein the memory stores a computer program, and the computer program is suitable for being loaded by the processor and executing the above-mentioned method steps.

[0014] The beneficial effects brought by the technical solutions provided by some embodiments of this specification include at least:

[0015] In one or more embodiments of the present specification, by obtaining user transaction training data for the initial behavior recognition model, extracting transaction training features from the user transaction training data, and then performing causal inference processing on the transaction training features through the initial behavior recognition model to obtain causal inference features, and then training the initial behavior recognition model based on the transaction training features and causal inference features, a behavior recognition model after model training can be obtained. By deeply mining the potential relationships between discrete transaction training features and the causal explanation characteristics between transaction training features, the trained behavior recognition model not only ensures the strong interpretability of the behavior recognition model itself, but also ensures the accuracy and operating efficiency of the behavior recognition model itself, and improves the generalization ability of the behavior recognition model after it goes online. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in this specification or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0017] Figure 1 It is a scenario diagram of a model processing system provided in this manual;

[0018] Figure 2 It is an example schematic diagram of a model processing method provided in this manual;

[0019] Figure 3 is an example schematic diagram of another model processing method provided in this manual;

[0020] Figure 4 It is a schematic diagram of a scenario processed by a model provided in this manual;

[0021] Figure 5 It is an example schematic diagram of a model processing method provided in this manual;

[0022] Figure 6 This is a training diagram of a behavior recognition model provided in this manual;

[0023] Figure 7It is a scenario diagram of a behavior classification network provided in this specification;

[0024] Figure 8 is a scenario diagram of another behavior classification network provided in this specification;

[0025] Fig. 9 It is a structural schematic diagram of a model processing device provided in this specification;

[0026] Fig.10 It is a structural diagram of a causal inference module provided in this specification;

[0027] Fig.11 is a structural diagram of a relationship inference unit provided in this specification;

[0028] Fig.12 It is a structural schematic diagram of an electronic device provided in this manual;

[0029] Fig.13 It is a schematic diagram of the structure of the operating system and user space provided in this manual;

[0030] Fig.14 yes Fig.13 The architecture diagram of the Android operating system;

[0031] Fig.15 yes Fig.13 Architecture diagram of the IOS operating system. DETAILED DESCRIPTION

[0032] The following will be combined with the drawings in this specification to clearly and completely describe the technical solutions in this specification. Obviously, the described embodiments are only part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0033] In the description of the present application, it should be understood that the terms "first", "second", etc. are only used for descriptive purposes and cannot be understood as indicating or implying relative importance. In the description of the present application, it should be noted that, unless otherwise clearly specified and limited, "including" and "having" and any of their variations are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not limited to the listed steps or units, but optionally also includes steps or units that are not listed, or optionally also includes other steps or units inherent to these processes, methods, products or devices. For those of ordinary skill in the art, the specific meanings of the above terms in the present application can be understood in specific circumstances. In addition, in the description of the present application, unless otherwise specified, "multiple" refers to two or more. "And / or" describes the association relationship of associated objects, indicating that there can be three relationships, for example, A and / or B, which can represent: A exists alone, A and B exist at the same time, and B exists alone. The character " / " generally indicates that the associated objects before and after are an "or" relationship.

[0034] In related technologies, in some actual transaction scenarios (such as consumer finance scenarios), the user training data collected on the current transaction platform over a period of time may face situations such as only a very small amount of historical fuzzy positive samples and no clear negative samples. In this way, the user transaction training data before model training usually only has a certain proportion of a small amount of sample data labeled (and the labeled labels are usually positive labels), and the remaining sample data usually does not contain labels. Therefore, user behavior predictions such as estimating the transaction behavior status of users on similar platforms and estimating the transaction loss behavior of users under corresponding transactions are difficult, and the black box model trained in the conventional way does not have good interpretability and has a low accuracy rate after going online.

[0035] The present application is described in detail below with reference to specific embodiments.

[0036] See also Figure 1 , is a schematic diagram of a scenario of a model processing system provided in this specification. Figure 1 As shown, the model processing system may include at least a client cluster and a service platform 100 .

[0037] The client cluster may include at least one client, such as Figure 1 As shown, it specifically includes client 1 corresponding to user 1, client 2 corresponding to user 2, ..., client n corresponding to user n, where n is an integer greater than 0.

[0038] Each client in the client cluster may be an electronic device with communication function, including but not limited to: wearable device, handheld device, personal computer, tablet computer, vehicle-mounted device, smart phone, computing device or other processing device connected to a wireless modem, etc. Electronic devices may be called different names in different networks, such as: user equipment, access terminal, user unit, user station, mobile station, mobile station, remote station, remote terminal, mobile device, user terminal, terminal, wireless communication device, user agent or user device, cellular phone, cordless phone, personal digital assistant (PDA), electronic device in 5G network or future evolution network, etc.

[0039] The service platform 100 can be a separate server device, such as a rack-mounted, blade, tower, or cabinet-mounted server device, or a workstation, mainframe computer, or other hardware device with strong computing power; it can also be a server cluster composed of multiple servers, and the servers in the service cluster can be composed in a symmetrical manner, wherein each server has equivalent functions and status in the transaction link, and each server can provide services to the outside independently, and the independent service can be understood as not requiring the assistance of other servers.

[0040] In one or more embodiments of the present specification, the service platform 100 may establish a communication connection with at least one client in the client cluster, and complete data interaction during the model processing based on the communication connection, such as online transaction data interaction. For example, the service platform 100 may recognize the user behavior of online users based on the behavior recognition model obtained by the model processing method of the present specification; for example, the service platform 100 may obtain user transaction training data under actual transactions from the client.

[0041] It should be noted that the service platform 100 and at least one client in the client cluster establish a communication connection through a network for interactive communication, wherein the network can be a wireless network or a wired network, the wireless network includes but is not limited to a cellular network, a wireless local area network, an infrared network or a Bluetooth network, and the wired network includes but is not limited to Ethernet, a universal serial bus (USB) or a controller local area network. In one or more embodiments of the specification, technologies and / or formats including Hyper Text Mark-up Language (HTML), Extensible Markup Language (XML), etc. are used to represent data (such as a target compressed package) exchanged through a network. In addition, conventional encryption technologies such as Secure Socket Layer (SSL), Transport Layer Security (TLS), Virtual Private Network (VPN), Internet Protocol Security (IPsec), etc. can also be used to encrypt all or some links. In other embodiments, customized and / or dedicated data communication technologies can also be used to replace or supplement the above data communication technologies.

[0042] The model processing system embodiment provided in this specification and the model processing method in one or more embodiments belong to the same concept. The execution subject corresponding to the model processing method involved in one or more embodiments of the specification can be the electronic device corresponding to the above-mentioned service platform 100; the execution subject corresponding to the model processing method involved in one or more embodiments of the specification can also be the electronic device corresponding to the client, which is determined based on the actual application environment. The implementation process of the model training system embodiment can be detailed in the following method embodiment, which will not be repeated here.

[0043] based on Figure 1 The scene diagram is shown, and the model processing method provided by one or more embodiments of this specification is introduced in detail below.

[0044] See also Figure 2 , is a flow chart of a model processing method for one or more embodiments of this specification, which can be implemented by a computer program and can be run on a background investigation device based on the von Neumann system. The computer program can be integrated into an application or run as an independent tool application. The model processing device can be an electronic device.

[0045] Specifically, the model processing method includes:

[0046] S102: Acquire user transaction training data for an initial behavior recognition model, and extract transaction training features from the user transaction training data;

[0047] It can be understood that the user transaction training data can be the user feature data of a large number of sample users in a certain transaction scenario, and the user feature data of a large number of users can be collected in this transaction scenario; there are common features between these user transaction training data in the user behavior prediction dimension, and these common features are of reference significance for the prediction of the corresponding user behavior status of the user in the transaction scenario; for example, it can be to predict the transaction behavior status of the current platform user on other platforms: that is, in the actual transaction scenario, based on the user feature data of a certain user that can be collected by the current platform, to predict the user transaction behavior status of the user on at least one similar platform (such as the active status type of a certain transaction behavior of the user on other platforms); for another example, it can be to predict the transaction behavior churn state type of the current user for a certain transaction behavior (such as financial transaction behavior, loan transaction behavior, savings transaction behavior): the churn state type can be characterized by a certain churn parameter to belong to a specific type.

[0048] In one or more embodiments of the present specification, the user transaction training data includes a number of user feature data, and the user transaction training data may be derived from user feature data in a corresponding transaction scenario.

[0049] It can be understood that the data feature types of the extracted transaction training features may be different based on different transaction scenarios. For example, in a financial transaction scenario, they may be user age features, credit performance features, savings behavior features, borrowing behavior features, user basic attribute features, etc.

[0050] The initial behavior recognition model is a pre-created machine learning model, and the machine learning model can be based on the fitting of one or more network models such as a neural network, a decision tree network, a classifier network, a Bayesian network, and a support vector machine network.

[0051] In a feasible implementation, after acquiring the user transaction training data, the user transaction training data is input into an initial behavior recognition model. The initial behavior recognition model can at least extract the portion of interest in the user transaction training data, that is, the transaction training features.

[0052] Optionally, the extracted transaction training features may be in the form of feature vectors, that is, feature vectors of corresponding feature dimensions are extracted.

[0053] S104: performing causal inference processing on the transaction training features through an initial behavior recognition model to obtain causal inference features;

[0054] The causal inference processing is used to deeply mine the causal relationship and correlation relationship between transaction training features. Usually, transaction training features correspond to features of each dimension and are discretized and have high dimensions (such as hundreds or thousands of dimensions). By further performing causal inference processing on the extracted transaction training features, the feature causal characteristics and feature correlation characteristics between features of each dimension can be deeply mined. The potential relationship between several discrete dimensional features can be used to construct and mine the causal explanation characteristics between several discrete dimensional features, so as to obtain the processed causal inference features.

[0055] In the related art, during the training phase of the machine learning model in the corresponding transaction scenario, the model training is performed only based on the extracted transaction features, and the model training process is deeply dependent on the annotation of the sample data. In the model processing method involved in this specification, after the transaction training features are extracted, the causal relationship and correlation relationship between the features of each dimension are further explored to optimize the model training effect, improve the recognition accuracy of the model after it is launched, and make the model more interpretable.

[0056] In a feasible implementation, the transaction training features are causally inferred through an initial behavior recognition model, and the feature causal characteristics and feature correlation characteristics between several dimensional features in the transaction training features are deeply mined. Based on the aforementioned characteristics of the several dimensional features, a causal feature structure (transaction) composed of feature nodes of several dimensions (one dimensional feature corresponds to at least one feature node) can be constructed. The causal feature structure can be represented in the form of a causal (structure) tree, a causal linked list, a causal (structure) graph, etc. The causal feature structure is usually a directed acyclic feature structure.

[0057] It is understandable that by decomposing the causal path based on the causal feature structure, several causal paths can be obtained. It is understandable that several feature path nodes corresponding to a causal path correspond to corresponding dimensional features (which can be understood as features of the corresponding transaction feature type). Based on several causal paths, several causal inference features can be obtained. Usually, a causal path is vectorized to obtain a causal inference feature corresponding to one dimension. In other words, the causal inference feature is composed of features of several dimensions.

[0058] S106: Performing model training on the initial behavior recognition model based on the transaction training features and the causal inference features to obtain a behavior recognition model after model training.

[0059] The behavior recognition model is a machine learning model after the initial behavior recognition model has completed training.

[0060] It can be understood that after the transaction training features are extracted and causal inference is processed, feature combination or feature fusion based on the transaction training features and causal inference features is performed to realize white-box interpretation in the feature processing stage. Based on the interpretable attribution of causal structure and causal path and the characteristics of reusable feature sample representation, the causal inference features are fused with the discrete transaction training features, and the target fusion features are input into the initial behavior recognition model for black-box model training until the preset training end conditions for the initial behavior recognition model are met. The behavior recognition model after model training can be obtained. The training end conditions may include, for example, the value of the loss function is less than or equal to the preset loss function threshold, the number of iterations reaches the preset number threshold, etc. The specific training end conditions can be determined based on the actual transaction scenario and will not be repeated here.

[0061] In this specification, by obtaining user transaction training data for the initial behavior recognition model, extracting transaction training features from the user transaction training data, and then performing causal inference processing on the transaction training features through the initial behavior recognition model to obtain causal inference features, and then training the initial behavior recognition model based on the transaction training features and causal inference features, the behavior recognition model after model training can be obtained. By deeply mining the potential relationships between discrete transaction training features and the causal explanation characteristics between transaction training features, the trained behavior recognition model not only ensures the strong interpretability of the behavior recognition model itself, but also ensures the accuracy and operating efficiency of the behavior recognition model itself, and improves the generalization ability of the behavior recognition model after it goes online.

[0062] See also Figure 3 , Figure 3 This is a flow chart of another embodiment of a model processing method proposed in one or more embodiments of this specification. Specifically:

[0063] S202: Acquire user transaction training data for the initial behavior recognition model, and extract transaction training features from the user transaction training data;

[0064] S204: Determine at least one transaction cause feature and at least one transaction result feature using the initial behavior recognition model based on the transaction training feature;

[0065] In one or more embodiments of the present specification, causal inference is performed on transaction training features through an initial behavior recognition model, and the feature causal characteristics and feature correlation characteristics between several dimensional features in the transaction training features are deeply mined. Specifically, several dimensional features in the transaction training features have feature logical relationships represented by cause characteristics and result characteristics (or effect characteristics), or feature relationships that can be formally expressed by causal logical relationships.

[0066] It can be understood that the feature corresponding to the cause feature in the transaction training feature is the transaction cause feature, and the feature corresponding to the result feature in the transaction training feature is the transaction result feature.

[0067] Specifically, the initial behavior model is used to calculate and analyze the characteristic causal characteristics and characteristic correlation characteristics between several dimensional characteristics in the transaction training characteristics, and then it can be determined whether there are directed connections between several dimensional (transaction training) characteristics in the transaction training characteristics for representing the causal relationship; it can be understood that there may be multiple directed connections for a certain dimensional (transaction training) feature. The transaction cause feature or transaction result feature is used to describe the causal (or correlation) characteristic relationship of at least one dimensional feature relative to another dimensional (transaction training) feature.

[0068] For example, when there is a direct causal relationship between transaction training feature A and transaction training feature B, and transaction training feature A affects transaction training feature B, then transaction training feature A can be regarded as the transaction cause feature relative to transaction training feature B, and conversely, transaction training feature B can be regarded as the transaction result feature. Based on this, it can be determined that there is a directed connection structure between transaction training feature A and transaction training feature B (such as a directed connection relationship from transaction training feature A to transaction training feature B).

[0069] In a feasible implementation, in actual transaction scenarios (such as consumer finance transaction scenarios), the transaction training features contained in the user transaction training data have many feature dimensions, and not all transaction training features of all feature dimensions have high interpretability. In some transaction scenarios, the "feature causal characteristics and feature correlation characteristics" deeply mined from transaction training features with high interpretability have a certain reference value. Based on this, the transaction training features of several dimensions can also be filtered through the initial behavior recognition model to filter out transaction training features of some dimensions with low interpretability, and the "feature causal characteristics and feature correlation characteristics" between each other can be deeply mined based on the transaction training features of the remaining dimensions after filtering.

[0070] Optionally, feature labels can be set in advance for transaction training features of several dimensions that can be included in the reference. It can be understood that the transaction training features corresponding to several feature labels are highly interpretable to each other and have a high feature causal reference effect. For example, n feature labels are set in advance, and after all transaction training features are extracted, the transaction training features that can be included in the reference corresponding to these n feature labels are selected, and the transaction training features that do not belong to these n feature labels are filtered out. In some scenarios, such feature labels can be called feature reference labels.

[0071] Optionally, feature filtering labels can be set in advance for transaction training features of several dimensions that are avoided from being included in the reference. It is understandable that the transaction training features corresponding to several feature filtering labels have low interpretability to each other and do not have a high feature causal reference effect. For example, y feature labels are set in advance, and after all transaction training features are extracted, the transaction training features that are avoided from being included in the reference corresponding to these y feature labels are selected for filtering. Filtering means that the filtered transaction training features do not use the initial behavior recognition model to determine at least one transaction cause feature and at least one transaction result feature, rather than deleting these "filtered transaction training features".

[0072] Furthermore, the setting of the feature reference label and / or the feature filtering label can be set by calling the expert end service. After the setting is completed, the feature reference label and / or the feature filtering label can instruct the initial behavior recognition model to perform causal recognition on the transaction training features that can be included in the reference, so as to determine the transaction cause features and transaction result features between the several transaction training features.

[0073] S206: Perform causal relationship inference processing based on the transaction cause characteristics and the transaction result characteristics to obtain at least one transaction causal chain;

[0074] It can be understood that after determining at least one transaction cause feature and at least one transaction result feature based on the transaction training feature using the initial behavior recognition model, what is obtained at this time is: based on the transaction training feature, the transaction cause feature and the transaction result feature between at least two dimensions (transaction training) features are determined, and a directed connection relationship is constructed between at least two dimensions (transaction training) features, from the transaction cause feature to the transaction result feature. For example, feature A, feature B, feature C, feature A and feature B are in a state that affects feature C, then feature A and feature B have a directed connection relationship pointing to feature C.

[0075] It can be understood that for each transaction cause feature and transaction result feature between at least two dimensional (transaction training) features, a directed connection relationship between the transaction cause feature and the transaction result feature of the aforementioned "at least two dimensional (transaction training) features" is sequentially constructed. After all directed connection relationships are established, what is usually obtained is a causal feature structure composed of several dimensional (transaction training) feature nodes (one (transaction training) dimensional feature corresponds to at least one feature node) after deep mining of the feature causal characteristics and feature correlation characteristics between several (transaction training) dimensional features in the transaction training features;

[0076] Optionally, the causal feature structure may be represented in the form of a causal (structure) tree, a causal chain table, a causal (structure) graph, etc. The causal feature structure is usually a directed acyclic feature structure.

[0077] Furthermore, the initial behavior recognition model will disassemble the causal path chain based on the causal feature structure, and obtain several transaction causal chains corresponding to the causal paths, that is, several transaction causal chains can be obtained. It can be understood that several feature path nodes corresponding to a causal path correspond to corresponding dimensional features (which can be understood as features of the corresponding transaction feature type). Based on several causal paths, several causal inference features can be obtained. Usually, a causal path is subjected to feature vectorization processing to obtain a causal inference feature corresponding to one dimension. In other words, the causal inference feature is composed of features of several dimensions.

[0078] Optionally, the number of directed connection relationships of a single feature node (corresponding to a transaction training feature of one dimension) in each transaction causal chain can be customized, such as setting the number of directed connection relationships to be less than a certain threshold value; schematically, the certain threshold value can be 1, which can be understood as the feature node in each disassembled transaction causal chain usually has one and only one directed path pointing to the next node.

[0079] In a specific implementation scenario, causal relationship inference processing is performed based on the transaction cause feature and the transaction result feature to obtain at least one transaction causal chain, which may be:

[0080] 1. The electronic device first determines the transaction causal relationship between the transaction cause feature and the transaction result feature, and constructs a transaction causal graph based on the transaction causal relationship;

[0081] It can be understood that the transaction cause feature and the transaction result feature between at least two dimensional (transaction training) features are determined based on the transaction training feature. At this time, the transaction causal relationship between at least two dimensional (transaction training) features is also determined accordingly. The transaction causal relationship can be understood as the causal (or correlation) characteristic relationship between at least one dimensional feature and another dimensional (transaction training) feature described based on the transaction cause feature or the transaction result feature.

[0082] It can be understood that after determining the transaction causal relationship between at least two dimensional (transaction training) features, the directed connection edges from the transaction cause feature to the transaction result feature of the aforementioned "at least two dimensional (transaction training) features" are sequentially constructed. After completing the establishment of all directed connection edges, what is usually obtained is a transaction causal graph composed of several dimensional (transaction training) feature nodes (one (transaction training) dimensional feature corresponds to at least one feature node) after deep mining of the feature causal characteristics and feature correlation characteristics between several (transaction training) dimensional features in the transaction training features;

[0083] Indicatively, Figure 4 As shown, Figure 4This is a schematic diagram of a scenario of model processing involved in this specification. Taking the transaction scenario as a consumer finance transaction scenario as an example, by obtaining transaction training data such as user age, credit performance, savings behavior, borrowing behavior, (platform) user basic attributes, etc., transaction training features of several dimensions in the transaction training data can be extracted using methods such as feature engineering (such as Figure 4 Then, the initial behavior recognition model is used to perform causal inference on the transaction training features of several dimensions, determine the transaction cause features and transaction result features between the transaction training features of several dimensions, and construct the transaction causal relationship between the transaction cause features and the transaction result features. Figure 4 The transaction causal graph shown in FIG. The transaction causal graph can provide feedback such as the topological relationship between transaction cause features and transaction result features.

[0084] In a feasible implementation, during the model training process for the initial behavior recognition model, transaction training data will be used in the feature fusion process before each round of model training to extract transaction training features and determine the transaction cause features and the transaction result features; and causal relationship inference processing will be performed based on the transaction cause features and the transaction result features to improve the transaction causal graph corresponding to each round. It can be understood that after the transaction causal graph is created in the feature fusion process before the first round of model training, the feature fusion process before each subsequent round of model training will perform causal graph learning, graph structure iteration, graph structure update and other causal graph learning processing on the transaction graph structure of the current round of transaction causal graph, until a transaction causal graph with a better graph structure is determined in the pre-feature fusion process as the model training process is completed.

[0085] 2. Then, the transaction causal graph is subjected to path decomposition processing to obtain at least one transaction causal chain.

[0086] Furthermore, the initial behavior recognition model will decompose the causal path chain based on the causal feature structure, and obtain transaction causal chains corresponding to several causal paths. In other words, several transaction causal chains can be obtained. It can be understood that several feature path nodes corresponding to a causal path correspond to corresponding dimensional features (which can be understood as features of the corresponding transaction feature type).

[0087] One path decomposition processing method is to traverse the directed connection relationship between each feature path node of the transaction causal graph and determine the terminal feature path node. The transaction training feature corresponding to the terminal feature path node does not affect other feature path nodes. That is, the node that does not contain a directed connection edge based on the feature path node is the terminal feature path node. Figure 4 As shown, the terminal feature path node is Figure 4The "Y node" of the transaction causal graph in the figure is determined by determining the terminal characteristic path node and then decomposing it from the terminal node from back to front. The decomposition requirement is: the number of branches of any node in each transaction causal chain is less than or equal to the preset decomposition number (such as the decomposition number is 1).

[0088] Taking the behavior recognition model used to predict "the user transaction behavior status of the user on at least one similar platform (such as the active status type of a transaction behavior of the user on other platforms)" as an example, the feature corresponding to the terminal feature path node is usually the active status type of a transaction behavior of the user on other platforms (such as active status types of different levels).

[0089] One path decomposition processing method may be: a target data label may be set, which is a label set for a transaction behavior feature. Usually, the transaction behavior feature corresponding to the target data label is the terminal feature path node in the transaction causal graph or the key transaction result feature node of interest. The target data label may be set in advance. For example, if the behavior recognition model is used to predict the transaction behavior status of users on the same platform, the transaction result feature node of interest is set based on the model effect of the behavior recognition model, that is, the transaction training feature corresponding to the "user transaction behavior status on the same platform" is set as the target data label, and exists in the form of a target data label.

[0090] Specifically, the decomposition processing method may be to obtain a label result node corresponding to a target data label from the transaction causal graph, and perform path decomposition processing on the transaction causal graph based on the label result node, thereby obtaining at least one transaction causal chain.

[0091] In some implementation scenarios, the target data label is also the data label annotated by the transaction training data before the model training.

[0092] Understandably, during the model training process, the corresponding training data is usually labeled. In some business scenarios, semi-supervised learning is used for data labeling, that is, part of the training data is labeled with the target data label, and the other part of the training data is not labeled.

[0093] In one or more embodiments of the present specification, the unlabeled training data during the current round of model training will also be subjected to causal inference processing based on the current round of transaction causal graph, so as to obtain the current transaction causal graph corresponding to the transaction result features corresponding to the target data label calculated on the basis of the network structure corresponding to the current round of transaction causal graph, and based on the label result node corresponding to the target data label, the transaction causal graph is subjected to path decomposition processing based on the label result node, so as to obtain at least one transaction causal chain. The corresponding transaction causal chain measures the uncertainty of the transaction causal chain in the form of a conditional probability.

[0094] It is understandable that the labeled training data during the current round of model training will be causally inferred based on the current round of transaction causal graph to obtain the corresponding transaction causal chain. These transaction causal chains will also correspond to the prediction (or calculation) of a conditional probability, but the confidence level corresponding to the conditional probability is usually higher, such as a confidence level of 1.

[0095] In a specific implementation scenario, "extracting transaction training features from the user transaction training data, determining at least one transaction cause feature and at least one transaction result feature based on the transaction training features, performing causal relationship inference processing based on the transaction cause features and the transaction result features, and obtaining at least one transaction causal chain" can be regarded as a feature fusion process before model training. The feature fusion process needs to perform feature vectorization based on the aforementioned at least one transaction causal connection to obtain causal inference features and perform feature fusion with the original transaction training features to complete the feature fusion process.

[0096] Specifically, during the feature fusion process before model training, at least one transaction causal chain is determined and a corresponding module or network for causal inference may be set in the initial behavior recognition model. For example, a causal inference network may be set in some embodiments.

[0097] Specifically, causal inference network training is performed based on the transaction training data of each round of model training. In each round of causal inference network training, transaction training features of the initial behavior recognition model are extracted and transaction cause features and transaction result features are determined; and causal relationship inference processing is performed based on the transaction cause features and transaction result features to improve the transaction causal graph corresponding to each round of causal inference network;

[0098] It can be understood that in the first round of causal inference network training, a transaction causal graph will be created to obtain the transaction causal chain of the first round of transaction causal graph; each subsequent round of causal inference network training will perform causal graph learning, graph structure iteration, graph structure update and other causal graph learning processes on the transaction graph structure of the transaction causal graph corresponding to the current round of causal inference network, and the transaction causal chain output by the causal inference network will be obtained in the current round; until the causal inference network is trained in the pre-causal inference network training process with the completion of the entire model training process, the graph structure of the transaction causal graph maintained by the causal inference network is determined. In this way, after the actual model is launched, the current transaction causal graph will be predicted based on the actual transaction data through the graph structure of the transaction causal graph maintained by the causal inference network and several transaction causal chains will be decomposed.

[0099] S208: Determine a causal inference feature based on the at least one transaction causal chain.

[0100] It can be understood that after determining at least one transaction causal chain, several transaction causal chains can be used to characterize the feature causal characteristics and feature correlation characteristics between transaction training features with several dimensions of discreteness, and the potential relationship between transaction training features with several dimensions of discreteness can be utilized to explore the causal explanation characteristics between features, so as to achieve accurate prediction of user behavior recognition, for example.

[0101] In a feasible implementation, after determining at least one transaction causal chain, feature vectorization is performed on several transaction causal chains to obtain causal inference features corresponding to the transaction causal chains. Assuming that feature vectorization is required for i (i is an integer) transaction causal chains, causal inference features of i dimensions will be generated. For example, feature vectorization can be performed on all transaction causal chains to obtain causal inference features corresponding to all transaction causal chains.

[0102] In a feasible implementation, at least one transaction causal chain can be evaluated and screened, and important transaction causal chains can be selected to construct causal inference features. On the one hand, it can avoid the interference of many transaction causal chains on model training, save model processing resources, and improve model processing efficiency; on the other hand, selecting several important transaction causal chains can highlight the more representative causal characteristics between features and ensure high interpretability.

[0103] The details are as follows:

[0104] 1. The electronic device can screen and process at least one transaction causal chain to obtain at least one target transaction chain; the target transaction chain is also the transaction causal chain that can be included in the reference after the screening process.

[0105] Optionally, a chain structure score may be performed on at least one transaction causal chain to obtain a target chain score corresponding to each transaction causal chain; and then based on each of the target chain scores, a target transaction causal chain indicated by a target dimension quantity may be determined from each of the at least one transaction causal chain.

[0106] Indicatively, each transaction causal chain can be scored based on the set chain structure scoring algorithm to obtain the target chain score corresponding to each transaction causal chain. Then, a scoring threshold can be pre-set, and the scoring threshold is used to screen the transaction causal chains, and the transaction causal chains with a target chain score greater than or equal to the scoring threshold can be used as the target transaction chain.

[0107] Optionally, the chain structure scoring algorithm can be a fit of one or more of a scoring algorithm based on posterior probability, a scoring algorithm based on Bayesian Information Criterion (BIC), a scoring algorithm based on Akaike Information Criterion (AIC), a scoring algorithm based on neural network scoring optimization, etc.

[0108] Indicatively, a chain structure scoring algorithm may be: determine the transaction training sample data (Data) and the corresponding transaction causal chain, and then use the posterior probability of the transaction causal chain as a scoring and quantification tool for the chain structure quality, which is used to evaluate the causal fit of the structure S of the transaction causal chain relative to the transaction training data (Data) corresponding to the transaction training features. Usually, the posterior probability P(S|Data) is difficult to calculate. The posterior probability of the transaction causal chain can be transformed using the Bayesian method, and converted into a prior probability P(S) based on the transaction causal chain, and a normal term P(Data) based on the sample data volume and the chain structure. The likelihood value P(Data|S) under the transaction causal chain is judged, that is, the structure S of the transaction causal chain is traversed to calculate the likelihood value P(Data|S), so that the value of P(Data|S) is as high as possible, and the value of P(Data|S) can be used as a score. For example, the calculation of the value of P(Data|S) can be calculated based on methods such as the BIC criterion in related technologies.

[0109] 2. After the electronic device screens out the target transaction chain, it then constructs a causal inference feature corresponding to at least one target transaction chain.

[0110] Schematically, after the electronic device screens out the target transaction chains, it performs feature vectorization processing on several target transaction chains to obtain causal inference features corresponding to the target transaction chains.

[0111] S210: Performing model training on the initial behavior recognition model based on the transaction training features and the causal inference features to obtain a behavior recognition model after model training.

[0112] In a feasible implementation, the electronic device may perform feature fusion processing on the transaction training features and the causal inference features based on the initial behavior recognition model to obtain the target fusion features; and then perform model training on the initial behavior recognition model based on the target fusion features.

[0113] Illustratively, the feature fusion processing method can be to directly perform feature splicing processing on the transaction training features and the causal inference features to obtain the target fusion features after splicing processing; based on the target fusion features, the initial behavior recognition model is trained until the training end conditions preset for the initial behavior recognition model are met, and the behavior recognition model after model training can be obtained. The training end conditions may include, for example, the value of the loss function is less than or equal to the preset loss function threshold, the number of iterations reaches the preset number threshold, etc. The specific training end conditions can be determined based on the actual transaction scenario, which will not be repeated here.

[0114] In one or more embodiments of the present specification, by obtaining user transaction training data for the initial behavior recognition model, extracting transaction training features from the user transaction training data, and then performing causal inference processing on the transaction training features through the initial behavior recognition model to obtain causal inference features, and then training the initial behavior recognition model based on the transaction training features and causal inference features, a behavior recognition model after model training can be obtained. By deeply mining the potential relationships between discrete transaction training features and the causal explanation characteristics between transaction training features, the trained behavior recognition model not only ensures the strong interpretability of the behavior recognition model itself, but also ensures the accuracy and operating efficiency of the behavior recognition model itself, and improves the generalization ability of the behavior recognition model after it goes online.

[0115] See also Figure 5 , Figure 5 This is a flow chart of another embodiment of a model processing method proposed in one or more embodiments of this specification. Specifically:

[0116] S302: Acquire user transaction training data for the initial behavior recognition model, and extract transaction training features from the user transaction training data;

[0117] For details, please refer to the execution steps of other method embodiments involved in this specification, which will not be repeated here.

[0118] S304: The initial behavior recognition model at least includes a causal inference network and a behavior classification network. The transaction training features are subjected to causal inference processing by the causal inference network to obtain causal inference features.

[0119] S306: Performing model training on the behavior classification network based on the transaction training features and the causal inference features to obtain a behavior recognition model after model training.

[0120] The initial behavior recognition model includes at least a causal inference network and a behavior classification network. In some embodiments, the initial behavior recognition model may include an extraction network (or module) for feature extraction such as transaction training, a fusion network (or module) for feature fusion, and the like.

[0121] The causal inference network can be constructed based on a Bayesian network (BN), a neural network for causal inference, a dynamic uncertain causal network, or other types of networks.

[0122] According to some embodiments, causal inference network training is performed based on transaction training data of each round of model training, such as Figure 6 As shown, Figure 6 It is a training diagram of the behavior recognition model involved in the present application. In each round of causal inference network training process: transaction training features are extracted based on transaction training data, and then the transaction cause features and the transaction result features are extracted and determined through the causal inference network, and causal relationship inference processing is performed based on the transaction cause features and the transaction result features to update the graph structure of the transaction causal graph corresponding to each round of the causal inference network, and the transaction causal graph of the current round is subjected to path decomposition processing to obtain at least one transaction causal chain, and then the causal inference features are determined through the causal inference network, and the transaction training features of the current round and the causal inference features are feature fused, and the behavior classification network is model trained based on the target fusion features after feature fusion until the model training completion conditions are met to obtain the behavior recognition model after model training.

[0123] It can be understood that in the first round of training, a transaction causal graph will be created based on the causal inference network training, and the first round of training will be obtained: several transaction causal chains corresponding to the transaction causal graph, and the causal inference features will be determined based on several transaction causal chains; each subsequent round of training will perform causal graph learning, graph structure iteration, graph structure update and other causal graph learning processes on the transaction graph structure of the transaction causal graph corresponding to each round of the causal inference network, and the transaction causal chain output by the causal inference network will be obtained in each round, and then the causal inference features will be determined, and the transaction training features of the current round and the causal inference features will be fused, and the behavior classification network will be trained based on the fused target fusion features. Until the causal inference network is trained in the pre-causal inference network training process with the completion of the entire model training process, the graph structure of the transaction causal graph maintained by the causal inference network is determined. In this way, after the actual model is launched, the current transaction causal graph will be predicted based on the actual transaction data through the graph structure of the transaction causal graph maintained by the causal inference network and several transaction causal chains will be decomposed.

[0124] Indicatively, the following performs model training on the behavior classification network based on the fused target fusion features until the model training completion condition is met.

[0125] The behavior classification network can be based on one or more of the network models such as neural network, decision tree network, classifier network, support vector machine network, cost-sensitive PU-Learning machine learning network, etc., and the behavior classification network can realize user behavior prediction based on the target fusion feature after fusing causal inference features and original transaction training features. In actual transactions, the behavior recognition model containing the behavior classification network can realize a variety of user behavior prediction scenarios.

[0126] It can be understood that in actual transaction scenarios, objective factors limit the use of semi-supervised learning to implement the model training process of the initial behavior recognition network. Semi-supervised learning means that among the training samples (that is, user transaction training data) used to train the initial behavior recognition network, only some of the training samples are labeled samples, and the rest of the training samples are unlabeled samples. The unlabeled samples are used to assist the learning process of the labeled samples.

[0127] In some actual transaction scenarios (such as consumer finance scenarios), there are only a very small number of fuzzy positive samples in the historical inventory, and there are no clear negative samples, and there is no user behavior data on similar transaction platforms. That is, before model training, usually only a certain proportion of a small amount of sample data has been labeled (and the labeled labels are usually positive labels), and the remaining sample data usually does not contain labels. Therefore, user behavior predictions such as estimating user flow to transaction behaviors on similar platforms are difficult, and the models trained in conventional ways do not have good interpretability and have low accuracy after going online. In order to better meet the needs and current status of actual transactions, the initial behavior recognition model is created based on the causal inference network and the behavior classification network;

[0128] Furthermore, the causal inference network can use a Bayesian network for causal inference, and the behavior classification network included in the initial behavior recognition model can be created based on the PU-Learning machine learning network, where PU Learning (Positive and Unlabeled Learning), that is, a machine learning network for labeled positive samples and unlabeled samples, is combined to solve the actual defects existing in the above-mentioned actual transaction scenarios, so that the behavior recognition model can have higher interpretability and stronger recognition accuracy after it is launched.

[0129] In a specific implementation scenario, the model training of the behavior classification network based on the fused target fusion features is explained. Specifically, the behavior classification network is a PU-Learning machine learning network. The following will be explained in combination with the two-step method as follows:

[0130] Based on whether the original user transaction training data is marked with positive labels, the target fusion features obtained by feature processing based on the original user transaction training data can be divided into a set of target fusion features corresponding to all marked positive labels as a positive feature sample set (P) and a set of target fusion features corresponding to all unmarked labels as an unlabeled feature sample set (U).

[0131] Step 1: Indicative, e.g. Figure 7 As shown, Figure 7 is a scenario diagram of a behavior classification network involved in this application,

[0132] First: by using the Spy-EM algorithm, a part of positive feature samples S (also called spy feature sample set) is extracted from the positive feature sample set (P), and S is added to U to generate the counter-example feature sample set US;

[0133] Secondly: by removing S from the positive feature sample set (P), a positive feature set PS (which can be expressed as PS=PS) is generated; based on the negative feature sample set US and the positive feature set PS, the created classification network clf1 is trained.

[0134] Schematically, the positive example feature set PS is marked as c1, and the US class is marked as c2. By adopting the Spy-EM (spy sample) algorithm, the threshold tr of the negative sample probability in c2 is determined according to the distribution of S, and the reliable negative example feature sample set RN is found from the negative example feature sample set US, and then the negative example samples are determined.

[0135] In the specific implementation, probability identification is used to determine which feature samples are most likely to be counterexample samples. A threshold tr is used to determine that the feature samples in the counterexample feature sample set US whose probability Pr(1|d) is less than t are considered to be the most likely counterexample feature samples, represented by RN.

[0136] Step 2: Indicatively, Figure 8 As shown, Figure 8 is a scenario diagram of a behavior classification network involved in this application,

[0137] A new classification network clf2 is established using P, RN and Q (Q = U-RN, Q is also called a possible positive proportion set). Depending on the quality and quantity of the data in the RN set, this step can be used once or multiple times based on a related classification network learning algorithm (such as NB or SVM). In the iteration, the feature sample RN_new that appears as a counterexample is divided into the counterexample feature sample set RN and the possible positive proportion set Q is updated. After each iteration, the counterexample samples and Q in the counterexample feature sample set RN will change. The learning algorithm is iteratively run until it converges or reaches a certain stopping condition, such as the stopping condition that no new counterexample feature sample RN_new can be generated, so that the cycle is terminated to obtain the final behavior classification network (that is, Figure 8 classifier clf_final in .

[0138] S308: Obtain user transaction data of the target online transaction, input the user transaction data into a behavior recognition model, and output behavior prediction parameters.

[0139] It can be understood that after completing the model training to obtain the behavior recognition model, the behavior recognition model can be deployed to the target online transaction, such as the target online transaction can be a consumer finance transaction, and then based on the real-time user transaction data of the online platform, the behavior recognition model is used to perform behavior prediction based on the user transaction data.

[0140] Optionally, the behavior prediction parameters consist of predicted behavior and behavior prediction probability.

[0141] Indicatively, the behavior recognition model can be applied to: predicting the transaction behavior status of the current platform user on other platforms, that is, the output of the behavior recognition model is the user's behavior status prediction parameter on other platforms (such as behavior status prediction probability), and the output of the model is the user's predicted behavior status on other platforms + behavior status prediction probability. The predicted behavior status can be different active status types of a certain transaction behavior of the user on other platforms (such as financial transaction behavior, loan transaction behavior, savings transaction behavior));

[0142] Indicatively, the behavior recognition model can be applied to: predicting the churn status of the current user under certain transaction behaviors, and the output of the model consists of the churn status type under a certain transaction behavior + the churn status prediction probability.

[0143] It can be understood that after the behavior recognition model is launched, the user transaction data is input into the behavior recognition model, and the user transaction features of the user transaction data are extracted based on the behavior recognition model. On the one hand, the user transaction features are causally inferred based on the causal inference network to obtain the current causal inference features; on the other hand, the current causal inference features are fused with the user transaction features, and the behavior classification network is used to perform behavior recognition based on the fused transaction fusion features to output behavior prediction parameters.

[0144] In actual transaction scenarios, the execution model processing method can adapt to the training data environment with a small number of samples and no clear negative samples in actual transaction scenarios. The highly interpretable causal inference network and the semi-supervised black box model, that is, the behavior classification network, are combined to propose a semi-supervised behavior recognition model based on causal inference, which not only ensures the strong interpretability of the behavior recognition model itself, but also ensures the accuracy and operating efficiency of the behavior recognition model itself. In addition, the effectiveness of the model processing method was verified through various channels such as survey questionnaires and comparative experiments in the actual transaction stage, which improved the generalization ability of the behavior recognition model after it was launched.

[0145] In one or more embodiments of the present specification, by obtaining user transaction training data for the initial behavior recognition model, extracting transaction training features from the user transaction training data, and then performing causal inference processing on the transaction training features through the initial behavior recognition model to obtain causal inference features, and then training the initial behavior recognition model based on the transaction training features and causal inference features, a behavior recognition model after model training can be obtained. By deeply mining the potential relationships between discrete transaction training features and the causal explanation characteristics between transaction training features, the trained behavior recognition model not only ensures the strong interpretability of the behavior recognition model itself, but also ensures the accuracy and operating efficiency of the behavior recognition model itself, and improves the generalization ability of the behavior recognition model after it goes online.

[0146] The following will be combined Fig. 9 , the model processing device provided in this specification is introduced in detail. It should be noted that, Fig. 9 The model processing device shown is used to execute the present application Figures 1 to 8 For the convenience of explanation, only the relevant parts of this specification are shown. For specific technical details not disclosed, please refer to this application. Figures 1 to 8 The embodiment shown.

[0147] See also Fig. 9 , which shows a schematic diagram of the structure of the model processing device of this specification. The model processing device 1 can be implemented as all or part of the user terminal through software, hardware or a combination of both. According to some embodiments, the model processing device 1 includes a feature extraction module 11, a causal inference module 12 and a model processing module 13, which are specifically used to:

[0148] A feature extraction module 11 is used to obtain user transaction training data for the initial behavior recognition model and extract transaction training features from the user transaction training data;

[0149] A causal inference module 12, configured to perform causal inference processing on the transaction training features through the initial behavior recognition model to obtain causal inference features;

[0150] The model processing module 13 is used to perform model training on the initial behavior recognition model based on the transaction training features and the causal inference features to obtain a behavior recognition model after model training.

[0151] Optional, such as Fig.10 As shown, the causal inference module 12 includes:

[0152] A feature parsing unit 121, configured to determine at least one transaction cause feature and at least one transaction result feature based on the transaction training feature and using the initial behavior recognition model;

[0153] A relationship inference unit 122, configured to perform causal relationship inference processing based on the transaction cause feature and the transaction result feature to obtain at least one transaction causal chain;

[0154] The feature generating unit 123 is configured to determine a causal inference feature based on the at least one transaction causal chain.

[0155] Optional, such as Fig.11 As shown, the relationship inference unit 122 includes:

[0156] A graph construction unit 1221, configured to determine a transaction causal relationship between the transaction cause feature and the transaction result feature, and to construct a transaction causal graph based on the transaction causal relationship;

[0157] The chain generation unit 1222 is used to perform path decomposition processing on the transaction causal graph to obtain at least one transaction causal chain.

[0158] Optionally, the chain generating unit 1222 is specifically configured to:

[0159] Obtaining a label result node corresponding to a target data label from the transaction causal graph;

[0160] The transaction causal graph is subjected to path decomposition processing based on the label result node to obtain at least one transaction causal chain.

[0161] Optionally, the feature generating unit 123 is specifically used for:

[0162] Screening the at least one transaction causal chain to obtain at least one target transaction chain;

[0163] Constructing a causal inference feature corresponding to the at least one target transaction chain.

[0164] Optionally, the feature generating unit 123 is specifically used for:

[0165] Performing a chain structure score on the at least one transaction causal chain to obtain a target chain score corresponding to each transaction causal chain;

[0166] Based on each of the target chain scores, a target transaction causal chain indicated by the number of target dimensions is determined from each of the at least one transaction causal chain.

[0167] Optionally, the model processing module 13 is specifically used to:

[0168] Performing feature fusion processing on the transaction training feature and the causal inference feature to obtain a target fusion feature;

[0169] The initial behavior recognition model is trained based on the target fusion features.

[0170] Optionally, the initial behavior recognition model includes a causal inference network and a behavior classification network;

[0171] The device 1 is specifically used for:

[0172] Performing causal inference processing on the transaction training features through the causal inference network to obtain causal inference features;

[0173] Model training is performed on the behavior classification network based on the transaction training features and the causal inference features to obtain a behavior recognition model after model training.

[0174] Optionally, the device 1 is specifically used for:

[0175] Obtain user transaction data of a target online transaction, input the user transaction data into a behavior recognition model, and output behavior prediction parameters.

[0176] It should be noted that the model processing device provided in the above embodiment only uses the division of the above functional modules as an example when executing the model processing method. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. In addition, the model processing device provided in the above embodiment belongs to the same concept as the model processing method embodiment, and its implementation process is detailed in the method embodiment, which will not be repeated here.

[0177] The above serial numbers in this specification are for description only and do not represent the advantages or disadvantages of the embodiments.

[0178] In one or more embodiments of the present specification, by obtaining user transaction training data for the initial behavior recognition model, extracting transaction training features from the user transaction training data, and then performing causal inference processing on the transaction training features through the initial behavior recognition model to obtain causal inference features, and then training the initial behavior recognition model based on the transaction training features and causal inference features, a behavior recognition model after model training can be obtained. By deeply mining the potential relationships between discrete transaction training features and the causal explanation characteristics between transaction training features, the trained behavior recognition model not only ensures the strong interpretability of the behavior recognition model itself, but also ensures the accuracy and operating efficiency of the behavior recognition model itself, and improves the generalization ability of the behavior recognition model after it goes online.

[0179] The present specification also provides a computer storage medium, which can store multiple instructions, which are suitable for being loaded and executed by a processor as described above. Figures 1 to 8 The model processing method of the embodiment shown in the figure can be found in the specific execution process. Figures 1 to 8 The specific description of the illustrated embodiment will not be repeated here.

[0180] The present application also provides a computer program product, which stores at least one instruction, and the at least one instruction is loaded and executed by the processor as described above. Figures 1 to 8 The model processing method of the embodiment shown in the figure can be found in the specific execution process. Figures 1 to 8 The specific description of the illustrated embodiment will not be repeated here.

[0181] Please refer to Fig.12, which shows a block diagram of the structure of an electronic device provided by an exemplary embodiment of the present application. The electronic device in the present application may include one or more of the following components: a processor 110, a memory 120, an input device 130, an output device 140, and a bus 150. The processor 110, the memory 120, the input device 130, and the output device 140 may be connected via the bus 150.

[0182] The processor 110 may include one or more processing cores. The processor 110 uses various interfaces and lines to connect various parts of the entire electronic device, and executes various functions and processes data of the electronic device 100 by running or executing instructions, programs, code sets or instruction sets stored in the memory 120, and calling data stored in the memory 120. Optionally, the processor 110 can be implemented in at least one hardware form of digital signal processing (DSP), field-programmable gate array (FPGA), and programmable logic array (PLA). The processor 110 can integrate one or a combination of a central processing unit (CPU), a graphics processing unit (GPU), and a modem. Among them, the CPU mainly processes the operating system, user interface, and application programs; the GPU is responsible for rendering and drawing display content; and the modem is used to process wireless communications. It can be understood that the above-mentioned modem may not be integrated into the processor 110, but may be implemented separately through a communication chip.

[0183] The memory 120 may include a random access memory (RAM) or a read-only memory (ROM). Optionally, the memory 120 includes a non-transitory computer-readable storage medium. The memory 120 may be used to store instructions, programs, codes, code sets or instruction sets. The memory 120 may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for implementing at least one function (such as a touch function, a sound playback function, an image playback function, etc.), instructions for implementing the following various method embodiments, etc. The operating system may be an Android system, including a system deeply developed based on the Android system, an IOS system developed by Apple, including a system deeply developed based on the IOS system or other systems. The data storage area may also store data created by the electronic device during use, such as a phone book, audio and video data, chat record data, etc.

[0184] See also Fig.13 As shown, the memory 120 can be divided into an operating system space and a user space. The operating system runs in the operating system space, and native and third-party applications run in the user space. In order to ensure that different third-party applications can achieve good operating results, the operating system allocates corresponding system resources to different third-party applications. However, different application scenarios in the same third-party application also have different requirements for system resources. For example, in the local resource loading scenario, the third-party application has higher requirements for disk reading speed; in the animation rendering scenario, the third-party application has higher requirements for GPU performance. The operating system and third-party applications are independent of each other, and the operating system often cannot perceive the current application scenario of the third-party application in a timely manner, resulting in the operating system being unable to perform targeted system resource adaptation according to the specific application scenario of the third-party application.

[0185] In order for the operating system to distinguish the specific application scenarios of third-party applications, it is necessary to open up data communication between third-party applications and the operating system so that the operating system can obtain the current scenario information of third-party applications at any time, and then perform targeted system resource adaptation based on the current scenario.

[0186] Taking the Android operating system as an example, the programs and data stored in the memory 120 are as follows: Fig.14As shown, the memory 120 may store a Linux kernel layer 320, a system runtime library layer 340, an application framework layer 360 and an application layer 380, wherein the Linux kernel layer 320, the system runtime library layer 340 and the application framework layer 360 belong to the operating system space, and the application layer 380 belongs to the user space. The Linux kernel layer 320 provides underlying drivers for various hardware of electronic devices, such as display drivers, audio drivers, camera drivers, Bluetooth drivers, Wi-Fi drivers, power management, etc. The system runtime library layer 340 provides the main feature support for the Android system through some C / C++ libraries. For example, the SQLite library provides database support, the OpenGL / ES library provides 3D drawing support, and the Webkit library provides browser kernel support, etc. The Android runtime library (Android runtime) is also provided in the system runtime library layer 340, which mainly provides some core libraries that allow developers to use the Java language to write Android applications. The application framework layer 360 provides various APIs that may be used when building applications. Developers can also use these APIs to build their own applications, such as activity management, window management, view management, notification management, content provider, package management, call management, resource management, and location management. At least one application runs in the application layer 380. These applications can be native applications that come with the operating system, such as contact applications, text messaging applications, clock applications, camera applications, etc.; they can also be third-party applications developed by third-party developers, such as game applications, instant messaging applications, photo beautification applications, etc.

[0187] Taking the operating system as an IOS system as an example, the programs and data stored in the memory 120 are as follows: Fig.15As shown, the IOS system includes: a core operating system layer 420 (Core OS layer), a core service layer 440 (Core Services layer), a media layer 460 (Media layer), and a touchable layer 480 (Cocoa Touch Layer). The core operating system layer 420 includes the operating system kernel, drivers, and underlying program frameworks, which provide functions closer to the hardware for use by the program framework located in the core service layer 440. The core service layer 440 provides system services and / or program frameworks required by the application, such as the foundation framework, account framework, advertising framework, data storage framework, network connection framework, geographic location framework, motion framework, etc. The media layer 460 provides audio-visual interfaces for the application, such as graphics and image related interfaces, audio technology related interfaces, video technology related interfaces, and wireless playback (AirPlay) interfaces for audio and video transmission technologies. The touchable layer 480 provides various commonly used interface-related frameworks for application development, and the touchable layer 480 is responsible for the user's touch interaction operations on the electronic device. For example, local notification service, remote push service, advertising framework, game tool framework, message user interface (UI) framework, user interface UIKit framework, map framework, etc.

[0188] exist Fig.15 Among the frameworks shown, the frameworks related to most applications include but are not limited to: the basic framework in the core service layer 440 and the UIKit framework in the touchable layer 480. The basic framework provides many basic object classes and data types, provides the most basic system services for all applications, and has nothing to do with UI. The classes provided by the UIKit framework are basic UI class libraries for creating touch-based user interfaces. iOS applications can provide UIs based on the UIKit framework, so it provides the basic architecture of applications for building user interfaces, drawing, processing and user interaction events, responding to gestures, etc.

[0189] Among them, the method and principle of implementing data communication between third-party applications and the operating system in the IOS system can be referred to the Android system, and this application will not go into details here.

[0190] Among them, the input device 130 is used to receive input instructions or data, and the input device 130 includes but is not limited to a keyboard, a mouse, a camera, a microphone or a touch device. The output device 140 is used to output instructions or data, and the output device 140 includes but is not limited to a display device and a speaker. In one example, the input device 130 and the output device 140 can be combined, and the input device 130 and the output device 140 are touch screens, which are used to receive touch operations on or near them by users using any suitable objects such as fingers and touch pens, and to display the user interfaces of various applications. The touch screen is usually set on the front panel of the electronic device. The touch screen can be designed as a full screen, a curved screen or a special-shaped screen. The touch screen can also be designed as a combination of a full screen and a curved screen, or a combination of a special-shaped screen and a curved screen, which is not limited in this specification.

[0191] In addition, those skilled in the art will appreciate that the structure of the electronic device shown in the above drawings does not constitute a limitation on the electronic device, and the electronic device may include more or fewer components than shown, or combine certain components, or arrange the components differently. For example, the electronic device also includes a radio frequency circuit, an input unit, a sensor, an audio circuit, a wireless fidelity (WiFi) module, a power supply, a Bluetooth module and other components, which will not be described in detail here.

[0192] In this specification, the execution subject of each step may be the electronic device described above. Optionally, the execution subject of each step is the operating system of the electronic device. The operating system may be an Android system, an IOS system, or other operating systems, which is not limited in this specification.

[0193] The electronic device of this specification may also be equipped with a display device, which may be any device capable of realizing a display function, such as a cathode ray tube display (CR), a light-emitting diode display (LED), an electronic ink screen, a liquid crystal display (LCD), a plasma display panel (PDP), etc. The user may use the display device on the electronic device 101 to view displayed text, images, videos and other information. The electronic device may be a smart phone, a tablet computer, a gaming device, an AR (Augmented Reality) device, a car, a data storage device, an audio playback device, a video playback device, a notebook, a desktop computing device, a wearable device such as an electronic watch, an electronic glasses, an electronic helmet, an electronic bracelet, an electronic necklace, electronic clothing and other devices.

[0194] exist Fig.11 In the electronic device shown, where the electronic device may be a terminal, the processor 110 may be used to call an application stored in the memory 120 and specifically perform the following operations:

[0195] Acquire user transaction training data for the initial behavior recognition model, and extract transaction training features from the user transaction training data;

[0196] Performing causal inference processing on the transaction training features through the initial behavior recognition model to obtain causal inference features;

[0197] Model training is performed on the initial behavior recognition model based on the transaction training features and the causal inference features to obtain a behavior recognition model after model training.

[0198] In one embodiment, when the processor 110 performs the causal inference processing on the transaction training features through the initial behavior recognition model to obtain the causal inference features, the processor 110 performs the following operations:

[0199] Determine at least one transaction cause feature and at least one transaction result feature using the initial behavior recognition model based on the transaction training feature;

[0200] Perform causal relationship inference processing based on the transaction cause characteristics and the transaction result characteristics to obtain at least one transaction causal chain;

[0201] A causal inference feature is determined based on the at least one transaction causal chain.

[0202] In one embodiment, when the processor 1001 performs causal relationship inference processing based on the transaction cause feature and the transaction result feature to obtain at least one transaction causal chain, the processor 1001 specifically performs the following operations:

[0203] Determine a transaction causal relationship between the transaction cause feature and the transaction result feature, and construct a transaction causal graph based on the transaction causal relationship;

[0204] Perform path decomposition processing on the transaction causal graph to obtain at least one transaction causal chain.

[0205] In one embodiment, when the processor 1001 performs the path decomposition processing on the transaction causal graph to obtain at least one transaction causal chain, the processor 1001 specifically performs the following steps:

[0206] Obtaining a label result node corresponding to a target data label from the transaction causal graph;

[0207] The transaction causal graph is subjected to path decomposition processing based on the label result node to obtain at least one transaction causal chain.

[0208] In one embodiment, the processor 1001, when executing the step of determining the causal inference feature based on the at least one transaction causal chain, includes:

[0209] Screening the at least one transaction causal chain to obtain at least one target transaction chain;

[0210] Constructing a causal inference feature corresponding to the at least one target transaction chain.

[0211] In one embodiment, when the processor 1001 performs the screening process on the at least one transaction causal chain to obtain at least one target transaction chain, the processor 1001 specifically performs the following steps:

[0212] Performing a chain structure score on the at least one transaction causal chain to obtain a target chain score corresponding to each transaction causal chain;

[0213] Based on each of the target chain scores, a target transaction causal chain indicated by the number of target dimensions is determined from each of the at least one transaction causal chain.

[0214] In one embodiment, when the processor 1001 performs the model training of the initial behavior recognition model based on the transaction training feature and the causal inference feature, the processor 1001 specifically performs the following steps:

[0215] Performing feature fusion processing on the transaction training feature and the causal inference feature to obtain a target fusion feature;

[0216] The initial behavior recognition model is trained based on the target fusion features.

[0217] In one embodiment, the processor 1001 executes the initial behavior recognition model including a causal inference network and a behavior classification network;

[0218] When performing causal inference processing on the transaction training features through the initial behavior recognition model to obtain causal inference features, and performing model training on the initial behavior recognition model based on the transaction training features and the causal inference features to obtain a behavior recognition model after model training, the following steps are specifically performed:

[0219] Performing causal inference processing on the transaction training features through the causal inference network to obtain causal inference features;

[0220] Model training is performed on the behavior classification network based on the transaction training features and the causal inference features to obtain a behavior recognition model after model training.

[0221] In one embodiment, after executing the behavior recognition model obtained after model training, the processor 1001 further performs the following steps:

[0222] Obtain user transaction data of a target online transaction, input the user transaction data into a behavior recognition model, and output behavior prediction parameters.

[0223] In one or more embodiments of the present specification, by obtaining user transaction training data for the initial behavior recognition model, extracting transaction training features from the user transaction training data, and then performing causal inference processing on the transaction training features through the initial behavior recognition model to obtain causal inference features, and then training the initial behavior recognition model based on the transaction training features and causal inference features, a behavior recognition model after model training can be obtained. By deeply mining the potential relationships between discrete transaction training features and the causal explanation characteristics between transaction training features, the trained behavior recognition model not only ensures the strong interpretability of the behavior recognition model itself, but also ensures the accuracy and operating efficiency of the behavior recognition model itself, and improves the generalization ability of the behavior recognition model after it goes online.

[0224] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing related hardware through a computer program, and the program can be stored in a computer-readable storage medium, and when the program is executed, it can include the processes of the embodiments of the above-mentioned methods. The storage medium can be a disk, an optical disk, a read-only storage memory, or a random access memory, etc.

[0225] The above disclosure is only the preferred embodiment of the present application, which certainly cannot be used to limit the scope of rights of the present application. Therefore, equivalent changes made according to the claims of the present application are still within the scope covered by the present application.

Claims

1. A model processing method, the method comprising: Acquire user transaction training data for the initial behavior recognition model, and extract transaction training features from the user transaction training data; Performing causal inference processing on the transaction training features through the initial behavior recognition model to obtain causal inference features; Performing model training on the initial behavior recognition model based on the transaction training features and the causal inference features to obtain a behavior recognition model after model training; The performing causal inference processing on the transaction training features through the initial behavior recognition model to obtain causal inference features includes: Determine at least one transaction cause feature and at least one transaction result feature using the initial behavior recognition model based on the transaction training feature; Performing causal relationship inference processing based on the transaction cause feature and the transaction result feature to obtain at least one transaction causal chain, wherein performing causal relationship inference processing based on the transaction cause feature and the transaction result feature to obtain at least one transaction causal chain includes: determining a transaction causal relationship between the transaction cause feature and the transaction result feature, constructing a transaction causal graph based on the transaction causal relationship, and performing path decomposition processing on the transaction causal graph to obtain at least one transaction causal chain; A causal inference feature is determined based on the at least one transaction causal chain.

2. According to the method of claim 1, the step of performing path decomposition processing on the transaction causal graph to obtain at least one transaction causal chain comprises: Obtaining a label result node corresponding to a target data label from the transaction causal graph; The transaction causal graph is subjected to path decomposition processing based on the label result node to obtain at least one transaction causal chain.

3. The method according to claim 1, wherein determining the causal inference feature based on the at least one transaction causal chain comprises: Screening the at least one transaction causal chain to obtain at least one target transaction chain; Constructing a causal inference feature corresponding to the at least one target transaction chain.

4. The method according to claim 3, wherein the screening process of the at least one transaction causal chain to obtain at least one target transaction chain comprises: Performing a chain structure score on the at least one transaction causal chain to obtain a target chain score corresponding to each transaction causal chain; Based on each of the target chain scores, a target transaction causal chain indicated by the number of target dimensions is determined from each of the at least one transaction causal chain.

5. According to the method of claim 1, the step of training the initial behavior recognition model based on the transaction training feature and the causal inference feature comprises: Performing feature fusion processing on the transaction training feature and the causal inference feature to obtain a target fusion feature; The initial behavior recognition model is trained based on the target fusion features.

6. The method according to any one of claims 1 to 5, wherein the initial behavior recognition model comprises a causal inference network and a behavior classification network; The process of performing causal inference processing on the transaction training features through the initial behavior recognition model to obtain causal inference features, and performing model training on the initial behavior recognition model based on the transaction training features and the causal inference features to obtain a behavior recognition model after model training includes: Performing causal inference processing on the transaction training features through the causal inference network to obtain causal inference features; Model training is performed on the behavior classification network based on the transaction training features and the causal inference features to obtain a behavior recognition model after model training.

7. The method according to claim 1, after obtaining the behavior recognition model after model training, further comprising: Obtain user transaction data of a target online transaction, input the user transaction data into a behavior recognition model, and output behavior prediction parameters.

8. A model processing device, comprising: A feature extraction module, used to obtain user transaction training data for the initial behavior recognition model, and extract transaction training features from the user transaction training data; A causal inference module, used to perform causal inference processing on the transaction training features through the initial behavior recognition model to obtain causal inference features; A model processing module, used to perform model training on the initial behavior recognition model based on the transaction training features and the causal inference features to obtain a behavior recognition model after model training; The causal inference processing of the transaction training features by the initial behavior recognition model to obtain causal inference features includes: determining at least one transaction cause feature and at least one transaction result feature based on the transaction training features using the initial behavior recognition model, performing causal relationship inference processing based on the transaction cause feature and the transaction result feature to obtain at least one transaction causal chain, the causal relationship inference processing based on the transaction cause feature and the transaction result feature to obtain at least one transaction causal chain includes: determining the transaction causal relationship between the transaction cause feature and the transaction result feature, constructing a transaction causal graph based on the transaction causal relationship, performing path decomposition processing on the transaction causal graph to obtain at least one transaction causal chain, and determining the causal inference features based on the at least one transaction causal chain.

9. A computer storage medium storing a plurality of instructions, wherein the instructions are suitable for being loaded by a processor and executing the method steps as claimed in any one of claims 1 to 7.

10. A computer program product, wherein the computer program product stores at least one instruction, wherein the at least one instruction is loaded by a processor and executes the method steps according to any one of claims 1 to 7.

11. An electronic device, comprising: A processor and a memory; wherein the memory stores a computer program, and the computer program is suitable for being loaded by the processor and executing the method steps as claimed in any one of claims 1 to 7.

Images