Ordered classification label determination method, device, electronic device and storage medium

Abstract

The present application provides a method, device, electronic device and storage medium for determining ordered classification labels, the method comprising: obtaining characteristic information of a user; inputting the characteristic information into an ordered classification neural network model, the ordered classification neural network model is used to determine the level probability of the user in the ordered classification label according to the level relationship of the ordered classification label and the hierarchical inclusion relationship derived from the level relationship of the ordered classification label; and determining the ordered classification label of the user using the level probability of the user in the ordered classification label. The hierarchical relationship in the characteristic information of the user is fully utilized, and the accuracy of the ordered classification label of the user is improved.

Description

Technical Field

[0001] The present application relates to the field of science and technology finance, and in particular to a method, device, electronic device and storage medium for determining ordered classification labels. Background Art

[0002] With the development of computer technology, more and more technologies are applied in the financial field. The traditional financial industry is gradually transforming into Fintech. Ordered classification technology is no exception. However, due to the security and real-time requirements of the financial industry, higher requirements are also placed on technology. In the financial field, personalized precision marketing and intelligent risk control are usually carried out for users through user profiling and machine learning. In the process of user profiling and machine learning, users usually need to be classified to determine user labels. At present, the classification of labels is usually divided into three types: binary classification labels, unordered classification labels, and ordered classification labels. Among them, the labels of binary classification labels and unordered classification labels are unordered. Binary classification labels divide users into two categories, such as whether the user is married, the user's gender, etc., and unordered classification labels divide users into multiple categories, such as the user's province, the user's blood type, etc. In contrast, ordered classification labels are multi-classification labels with certain hierarchical relationships such as size, height, strength, etc., such as academic qualifications are divided into university, middle school, elementary school, etc., positions are divided into senior executives, middle-level, grassroots employees, etc., and income is divided into high income, upper middle income, middle income, lower middle income, low income, etc. In actual applications, ordered classification labels account for a large proportion and have important business value.

[0003] In the prior art, the inference of ordered classification labels of users is usually carried out by treating ordered classification labels as unordered classification labels, using a multi-classification algorithm or a one-vs-rest binary classification algorithm to model and determine the ordered classification labels of users. For example, when inferring that a user's wealth level is high, medium or low, the hierarchical relationship of wealth levels of "high > medium > low" is ignored, and the wealth level is used as an unordered classification label to classify the user.

[0004] However, in the prior art, since there is a hierarchical relationship between the label values ​​of the ordered classification labels, the hierarchical relationship of the ordered classification labels cannot be fully utilized by modeling the ordered classification labels as unordered classification labels, resulting in low accuracy in determining the user's ordered classification labels. Summary of the invention

[0005] The present application provides a method, device, electronic device and storage medium for determining ordered classification labels, so as to solve the technical problem of low accuracy in determining user ordered classification labels in the prior art.

[0006] In a first aspect, the present application provides a method for determining ordered classification labels, comprising:

[0007] Acquire the user's characteristic information, which is used to determine the user's ordered classification label; input the characteristic information into the ordered classification neural network model, which is used to determine the user's level probability in the ordered classification label based on the level relationship of the ordered classification labels and the hierarchical inclusion relationship derived from the level relationship of the ordered classification labels; determine the user's ordered classification label using the user's level probability in the ordered classification label.

[0008] In an embodiment of the present application, the hierarchical probability of the user in the ordered classification tags is determined based on the hierarchical relationship of the ordered classification tags and the hierarchical inclusion relationship derived from the hierarchical relationship of the ordered classification tags, and then the ordered classification tags of the user are determined, which fully utilizes the hierarchical relationship in the ordered classification tags and improves the accuracy of the user's ordered classification tags.

[0009] In a possible implementation, before inputting the feature information into the ordered classification neural network model, the method further includes:

[0010] Utilize the hierarchical relationship of ordered classification labels to derive the hierarchical inclusion relationship of ordered classification labels; analyze the probabilistic operation logic of the hierarchical relationship and the hierarchical inclusion relationship, and use the probabilistic operation logic to build a neural network structure; train the neural network structure to generate an ordered classification neural network model.

[0011] In the embodiment of the present application, the reliability of the ordered classification neural network model is improved by analyzing the probability operation logic of the hierarchical relationship and the hierarchical inclusion relationship and using the probability operation logic to build a neural network structure.

[0012] In one possible implementation, the neural network structure includes a feature part and a prediction part. The feature part is used to determine the individual features between each first task, and the prediction part is used to determine the level probability of the ordered classification labels by utilizing the probability operation logical relationship between the individual features and each second task. The first task is used to determine the hierarchical inclusion probability of the ordered classification labels, and the second task is used to determine the level probability of the ordered classification labels.

[0013] In an embodiment of the present application, a feature unit is set in the neural network structure to determine personalized features unique to each level of inclusion probability of the ordered classification labels, so as to better learn the differences between the calculation methods of the inclusion probability of each level, and then through the prediction unit in the neural network structure, the personalized features are used to predict the level probability of the ordered classification labels, thereby improving the learning accuracy of the neural network structure.

[0014] In a possible implementation, the feature part includes a shared feature layer and a personalized feature layer, and uses probabilistic operation logic to build a neural network structure, including:

[0015] The shared feature layer of the neural network structure is constructed by utilizing the shared features between the first tasks; the individual feature layer of the neural network structure is constructed by utilizing the individual features between the first tasks; the prediction part of the neural network structure is constructed by utilizing the probability operation logic relationship between the first task and the second task to obtain the neural network structure.

[0016] In the embodiment of the present application, by taking into account the shared features between the hierarchical inclusion probabilities of ordered classification labels and the individual features between the hierarchical inclusion probabilities, not only the learning accuracy of the neural network structure can be improved, but also the working efficiency of the neural network structure can be improved.

[0017] In one possible implementation, the shared feature layer includes an embedding vector layer, a feature screening layer, and a splicing layer. The embedding vector layer is used to generate an embedding vector of feature information; the feature screening layer is used to perform a linear transformation on the embedding vector, determine the linear transformation value of the embedding vector, and use a nonlinear function to map the linear transformation value, and filter the embedding vector with a negative linear transformation value; the splicing layer is used to splice the filtered embedding vectors to obtain shared features between the first tasks.

[0018] In the embodiment of the present application, by representing the feature information as an embedded vector and screening the embedded vector to filter out redundant information, the accuracy of the ordered classification label inference is further improved.

[0019] In a possible implementation, training a neural network structure to generate an ordered classification neural network model includes:

[0020] Acquire training samples; based on the training samples, use the first loss function to adjust the network parameters of the neural network structure to obtain an ordered classification neural network model.

[0021] In a possible implementation, the first loss function is:

[0022]

[0023] Among them, loss1 represents the first loss function, M represents the maximum level of the ordered classification label, k represents the level of the ordered classification label, and W k Indicates the weight of the training sample at level k, Y k Indicates whether the actual level of the training sample is k, Y k The value is 0 or 1; P k It is expressed as the predicted probability of the training sample with level k.

[0024] In one possible implementation, a neural network structure is trained to generate an ordered classification neural network model, including: obtaining training samples, the training samples including the user's level label and hierarchical inclusion label; based on the hierarchical inclusion label, using a second loss function to train the neural network structure to determine the shared feature parameters of the shared feature layer; based on the hierarchical inclusion label, using a third loss function to train the neural network structure to determine the individual feature parameters of the individual feature layer; based on the level label, using a first loss function to train the neural network structure to determine the prediction parameters of the prediction unit.

[0025] In the embodiment of the present application, a step-by-step learning strategy is adopted to focus on learning the network parameters in steps, and the similarities and differences between different tasks are utilized to improve the accuracy of the ordered classification neural network model.

[0026] In a possible implementation, the second loss function is:

[0027]

[0028] The third loss function is:

[0029] loss3=[-Y k+ log(P k+ )-(1-Y k+ )log(1-P k+ )];

[0030] Among them, loss2 represents the second loss function, loss3 represents the third loss function, M represents the maximum level of the ordered classification label, k represents the level of the ordered classification label, and Y k+ Indicates whether the actual level of the training sample is greater than or equal to k, Y k+ The value is 0 or 1; P k+ It is expressed as the predicted probability that the rank of a training sample is greater than or equal to k.

[0031] The following introduces the device, electronic device, computer-readable storage medium and computer program product provided in the embodiments of the present application. Their content and effects can refer to the ordered classification label determination method provided in the embodiments of the present application and will not be repeated here.

[0032] In a second aspect, an embodiment of the present application provides an ordered classification label determination device, comprising:

[0033] The acquisition module is used to obtain the user's feature information, and the feature information is used to determine the user's ordered classification label.

[0034] A processing module, used for inputting feature information into an ordered classification neural network model, wherein the ordered classification neural network model is used for determining the level probability of the user in the ordered classification label according to the level relationship of the ordered classification label and the hierarchical inclusion relationship derived from the level relationship of the ordered classification label;

[0035] The processing module is also used to determine the ordered classification label of the user by using the level probability of the user in the ordered classification label.

[0036] In a possible implementation manner, the device for determining ordered classification labels provided in the embodiment of the present application further includes:

[0037] The derivation module is used to utilize the hierarchical relationship of the ordered classification labels to derive the hierarchical inclusion relationship of the ordered classification labels; the construction module is used to parse the probabilistic operation logic of the hierarchical relationship and the hierarchical inclusion relationship, and use the probabilistic operation logic to build the neural network structure; the training module is used to train the neural network structure and generate an ordered classification neural network model.

[0038] In one possible implementation, the neural network structure includes a feature part and a prediction part. The feature part is used to determine the individual features between each first task, and the prediction part is used to determine the level probability of the ordered classification labels by utilizing the probability operation logical relationship between the individual features and each second task. The first task is used to determine the hierarchical inclusion probability of the ordered classification labels, and the second task is used to determine the level probability of the ordered classification labels.

[0039] In a possible implementation, the feature part includes a shared feature layer and a personalized feature layer, and a building module is specifically used for:

[0040] The shared feature layer of the neural network structure is constructed by utilizing the shared features between the first tasks; the individual feature layer of the neural network structure is constructed by utilizing the individual features between the first tasks; the prediction part of the neural network structure is constructed by utilizing the probability operation logic relationship between the first task and the second task to obtain the neural network structure.

[0041] In one possible implementation, the shared feature layer includes an embedding vector layer, a feature screening layer, and a splicing layer. The embedding vector layer is used to generate an embedding vector of feature information; the feature screening layer is used to perform a linear transformation on the embedding vector, determine the linear transformation value of the embedding vector, and use a nonlinear function to map the linear transformation value, and filter the embedding vector with a negative linear transformation value; the splicing layer is used to splice the filtered embedding vectors to obtain shared features between the first tasks.

[0042] In a possible implementation, the training module is specifically used to:

[0043] Acquire training samples; based on the training samples, use the first loss function to adjust the network parameters of the neural network structure to obtain an ordered classification neural network model.

[0044] In a possible implementation, the first loss function is:

[0045]

[0046] Among them, loss1 represents the first loss function, M represents the maximum level of the ordered classification label, k represents the level of the ordered classification label, and W k Indicates the weight of the training sample at level k, Y k Indicates whether the actual level of the training sample is k, Y k The value is 0 or 1; P k It is expressed as the predicted probability of the training sample with level k.

[0047] In a possible implementation, the training module is specifically used to:

[0048] Acquire training samples, which include the user's level label and hierarchical inclusion label; based on the hierarchical inclusion label, use the second loss function to train the neural network structure to determine the shared feature parameters of the shared feature layer; based on the hierarchical inclusion label, use the third loss function to train the neural network structure to determine the individual feature parameters of the individual feature layer; based on the level label, use the first loss function to train the neural network structure to determine the prediction parameters of the prediction unit.

[0049] In a possible implementation, the second loss function is:

[0050]

[0051] The third loss function is:

[0052] loss3=[-Y k+ log(P k+ )-(1-Y k+ )log(1-P k+ )];

[0053] Among them, loss2 represents the second loss function, loss3 represents the third loss function, M represents the maximum level of the ordered classification label, k represents the level of the ordered classification label, and Y k+ Indicates whether the actual level of the training sample is greater than or equal to k, Y k+ The value is 0 or 1; P k+ It is expressed as the predicted probability that the rank of a training sample is greater than or equal to k.

[0054] In a third aspect, an embodiment of the present application provides an electronic device, including: a processor, and a memory communicatively connected to the processor;

[0055] Memory stores computer-executable instructions;

[0056] The processor executes the computer-executable instructions stored in the memory to implement the first aspect or provide an ordered classification label determination method in an implementable manner of the first aspect.

[0057] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, in which computer execution instructions are stored. When the computer execution instructions are executed by a processor, they are used to implement the ordered classification label determination method provided in the first aspect or the implementable method of the first aspect.

[0058] In a fifth aspect, an embodiment of the present application provides a computer program product, comprising computer execution instructions, which, when executed by a processor, are used to implement the ordered classification label determination method provided in the first aspect or the implementable manner of the first aspect.

[0059] The method, device, electronic device and storage medium for determining ordered classification labels provided by the present application obtain the characteristic information of the user; input the characteristic information into the ordered classification neural network model to classify the user in an ordered manner, and the ordered classification neural network model is used to determine the user's level probability in the ordered classification label according to the level relationship of the ordered classification label and the hierarchical inclusion relationship derived from the level relationship of the ordered classification label; finally, the user's ordered classification label is determined using the level probability. Since the user's level probability in the ordered classification label is determined according to the level relationship of the ordered classification label and the hierarchical inclusion relationship derived from the level relationship of the ordered classification label, and then the user's ordered classification label is determined, the hierarchical relationship in the user's characteristic information is fully utilized, and the accuracy of the user's ordered classification label is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0061] Figure 1 This is an exemplary application scenario architecture diagram provided by an embodiment of the present application;

[0062] Figure 2 It is a flowchart of a method for determining ordered classification labels provided in an embodiment of the present application;

[0063] Figure 3 It is a schematic diagram of the workflow of the ordered classification neural network model provided in one embodiment of the present application;

[0064] Figure 4It is a schematic diagram of hierarchical inclusion relationship provided by an embodiment of the present application;

[0065] Figure 5 is a flowchart of a method for determining ordered classification labels provided by another embodiment of the present application;

[0066] Figure 6 is a structural diagram of a neural network structure provided by an embodiment of the present application;

[0067] Figure 7 It is a structural schematic diagram of an ordered classification label determination device provided in an embodiment of the present application;

[0068] Figure 8 is a schematic diagram of the structure of an ordered classification label determination device provided by another embodiment of the present application;

[0069] Fig. 9 It is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application.

[0070] The above drawings have shown clear embodiments of the present application, which will be described in more detail later. These drawings and text descriptions are not intended to limit the scope of the present application in any way, but to illustrate the concept of the present application to those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION

[0071] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application. Instead, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the appended claims.

[0072] The terms "first" and "second" in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein, for example. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0073] With the development of computer technology, more and more technologies are applied in the financial field. The traditional financial industry is gradually transforming into financial technology (Fintech), and ordered classification technology is no exception. However, due to the security and real-time requirements of the financial industry, higher requirements are also placed on the technology. In the prior art, the calculation of the ordered classification labels of users usually adopts the method of treating the ordered classification labels as unordered classification labels, and using a multi-classification algorithm or a one-vs-rest two-classification algorithm to model and determine the ordered classification labels of users. Since there is a hierarchical relationship between the category values ​​of the ordered classification labels, the hierarchical relationship of the ordered classification labels cannot be fully utilized by modeling the ordered classification labels as unordered classification labels, resulting in a low accuracy in determining the ordered classification labels.

[0074] The inventive concept of the ordered classification label determination method, device, electronic device and storage medium provided in the embodiments of the present application is to construct an ordered classification neural network model by utilizing the hierarchical relationship of the ordered classification labels and the hierarchical inclusion relationship derived from the hierarchical relationship of the ordered classification labels, and then to implement the ordered classification of users through the ordered classification neural network model, thereby achieving the determination of the ordered classification labels of users. The hierarchical relationship in the user's feature information is fully utilized to improve the accuracy of the user's ordered classification labels, and while fully utilizing the hierarchical information of the ordered classification labels, no subjective problems are introduced and no additional assumptions are required, further ensuring the accuracy of determining the ordered classification labels.

[0075] The following is an introduction to exemplary application scenarios of the embodiments of the present application.

[0076] The ordered classification label determination method provided in the embodiment of the present application can be performed by the ordered classification label determination device provided in the embodiment of the present application. The ordered classification label determination device provided in the embodiment of the present application can be integrated in the terminal device, or the ordered classification label determination device can be the terminal device itself. The embodiment of the present application does not limit the specific type of the terminal device. For example, the terminal device can be a personal computer, a tablet computer, a wearable device, a vehicle-mounted terminal, a monitoring device, etc.

[0077] Figure 1 is an exemplary application scenario architecture diagram provided by an embodiment of the present application, such as Figure 1 As shown, the architecture mainly includes: terminal device 01 (personal computer). In one possible implementation, the ordered classification label determination method provided in the embodiment of the present application can be applied to the terminal device 01. The terminal device 01 can obtain the user's feature information, and then classify the user in an ordered manner through the ordered classification neural network model in the terminal device 01 to determine the user's ordered classification label. In another possible implementation, Figure 1As shown, the application scenario architecture diagram provided by the embodiment of the present application may also include a server 02, the terminal device 01 obtains the characteristic information of the user, and then processes the characteristic information through the server 02, determines the ordered classification label of the user, and sends the processing result to the terminal device 01, and the terminal device 01 receives the processing result sent by the server 02. The embodiment of the present application is only taken as an example and is not limited to this.

[0078] The technical solution of the present application and how the technical solution of the present application solves the above-mentioned technical problems are described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of the present application will be described below in conjunction with the accompanying drawings.

[0079] Figure 2 1 is a flow chart of a method for determining an ordered classification label provided by an embodiment of the present application. The method can be executed by an ordered classification label determination device, which can be implemented by software and / or hardware. The following describes the method for determining an ordered classification label with a terminal device as the execution subject. Figure 2 As shown, the method for determining ordered classification labels provided in the embodiment of the present application may include:

[0080] Step S101: Acquire user characteristic information.

[0081] The user's characteristic information is used to determine the user's ordered classification label. The user's characteristic information may include characteristic information such as the user's income, the user's credit card consumption amount, and the user's work industry. For different ordered classification labels, one or more different characteristic information of the user may be required for calculation. The embodiment of the present application does not limit the specific characteristic information of the user, and the user's characteristic information can be obtained according to demand. For example, if the ordered classification label is the user's wealth level, the user's characteristic information may include the user's account information. For another example, if the ordered classification label is the user's consumption level, the user's characteristic information may include the user's consumption amount, etc. The embodiment of the present application only takes this as an example and is not limited to this.

[0082] The embodiment of the present application does not limit the specific implementation method of obtaining the characteristic information of the user. For example, the characteristic information of the user can be directly read from the database of the server, or the characteristic information of the user can be screened out from a database including various information of the user. The embodiment of the present application takes this as an example only and is not limited thereto.

[0083] Step S102: inputting feature information into an ordered classification neural network model, which is used to determine the level probability of the user in the ordered classification labels according to the level relationship of the ordered classification labels and the hierarchical inclusion relationship derived from the level relationship of the ordered classification labels.

[0084] Figure 3 : is a schematic diagram of the workflow of the ordered classification neural network model provided in one embodiment of the present application, such as Figure 3 As shown, the user's feature information is input into the ordered classification neural network model, and the ordered classification neural network model analyzes and processes the feature information and outputs the user's level probability in the ordered classification label.

[0085] Exemplarily, the characteristic information of the user is the income information of the user, the ordered classification label is the wealth level, and the wealth levels include high income, middle income and low income. The ordered classification neural network model can analyze and process the income information of the user to generate the probability that the user belongs to the high-income level, the middle-income level and the low-income level.

[0086] Among them, the ordered classification neural network model is used to determine the user's level probability in the ordered classification label according to the level relationship of the ordered classification label and the hierarchical inclusion relationship derived from the level relationship of the ordered classification label. For the sake of ease of introduction, the embodiment of the present application takes the ordered classification label as the wealth level as an example for introduction, but is not limited to this. For example, the ordered classification label can also include the user's education, user position, user consumption ability, etc., which can be set according to actual needs and will not be repeated.

[0087] In a possible implementation, the wealth level may include five levels: 0, 1, 2, 3, and 4, and the level relationship of the wealth levels is 4>3>2>1>0. Then, the level relationship of the wealth levels may be used to derive the hierarchical inclusion relationship of the wealth levels: wealth level ≥1→wealth level ≥2→wealth level ≥3→wealth level = 4. For ease of understanding, Figure 4 Schematic diagram of hierarchical inclusion relationship provided by an embodiment of the present application. Figure 4 As shown, the outermost ellipse represents the probability of wealth level ≥ 0, and its probability value is 1. The second outermost ellipse represents the probability of wealth level ≥ 1. The third ellipse represents the probability of wealth level ≥ 2. The second innermost ellipse represents the probability of wealth level ≥ 3. The innermost ellipse represents the probability of wealth level = 4.

[0088] After deriving the hierarchical inclusion relationship of wealth levels from the hierarchical relationship of wealth levels, the probability of a user's level in wealth levels can be determined based on the hierarchical relationship of wealth levels and the hierarchical inclusion relationship of wealth levels. Its working principle is described as follows:

[0089] Let Y represent wealth level and X represent input feature information, then P(Y=k|X) represents the probability that the user’s wealth level is k given X, where the possible values ​​of k are 0, 1, 2, 3 or 4. Similarly, P(Y≥k|X) represents the probability that the user’s wealth level is greater than or equal to k given X.

[0090] but,

[0091] P(Y=k-1|X)=P(Y≥k-1|X)-P(Y≥k|X); (1)

[0092] P(Y≥k|X)=P(Y≥k-1|X)×P(Y≥k|Y≥k-1,X); (2)

[0093] Among them, k=1, 2, 3, 4.

[0094] Let P k represents P(Y=k|X), P k+ represents P(Y≥k|X), Q k+ It represents P(Y≥k|Y≥k-1,X).

[0095] According to formula (1), we can get:

[0096] P1=P(Y=1|X)=P(Y≥1|X)-P(Y≥2|X)=P 1+ -P 2+ ;

[0097] P2=P(Y=2|X)=P(Y≥2|X)-P(Y≥3|X)=P 2+ -P 3+ ;

[0098] P3=P(Y=3|X)=P(Y≥3|X)-P(Y≥4|X)=P 3+ -P 4+ ; (3)

[0099] According to formula (2), we can get:

[0100] P 2+ =P(Y≥2|X)=P(Y≥1|X)×P(Y≥2|Y≥1,X)=P 1+ Q 2+ ;

[0101] P 3+ =P(Y≥3|X)=P(Y≥2|X)×P(Y≥3|Y≥2,X)=P 2+ Q 3+ ;

[0102] P 4+=P(Y≥4|X)=P(Y≥3|X)×P(Y≥4|Y≥3,X)=P 3+ Q 4+ ; (4)

[0103] In summary, P0, P1, P2, P3 and P4 can be expressed as:

[0104] P0=1-P 1+ ;

[0105] P1=P 1+ -P 2+ =P 1+ -P 1+ Q 2+ ;

[0106] P2=P 2+ -P 3+ =P 1+ Q 2+ -P 1+ Q 2+ Q 3+ ;

[0107] P3=P 3+ -P 4+ =P 1+ Q 2+ Q 3+ -P 1+ Q 2+ Q 3+ Q 4+ ;

[0108] P4=P 4+ =P 1+ Q 2+ Q 3+ Q 4+ ; (5)

[0109] The final output of each wealth level probability is P0, P1, P2, P3 and P4, and the level inclusion probability of each wealth level is P 1+ , P 2+ , P 3+ , the conditional probability of the intermediate node is Q 2+ , Q 3+ and Q 4+ .

[0110] By determining the level probability of the user in the ordered classification labels according to the level relationship of the ordered classification labels and the hierarchical inclusion relationship derived from the level relationship of the ordered classification labels, the level relationship in the user's feature information is fully utilized. Compared with the prior art, the ordered classification labels are treated as unordered classification labels, the level relationship of the ordered classification labels is not considered, and the level probability of the ordered classification labels is directly determined, which can improve the accuracy of the level probability of the ordered classification labels.

[0111] Step S103: Determine the ordered classification label of the user by using the level probability of the user in the ordered classification label.

[0112] After determining the user's level probability in the ordered classification label, the user's level probability in the ordered classification label is used to determine the user's ordered classification label. Taking the above-mentioned ordered classification label as wealth level as an example, the user's wealth level can be determined by setting the wealth level with the highest level probability as the user's wealth level, or by setting the weight of each level probability. The embodiment of the present application does not limit this.

[0113] In an embodiment of the present application, the hierarchical probability of the user in the ordered classification labels is determined based on the hierarchical relationship of the ordered classification labels and the hierarchical inclusion relationship derived from the hierarchical relationship of the ordered classification labels, and then the ordered classification labels of the user are determined, which fully utilizes the hierarchical relationship in the user's feature information and improves the accuracy of the user's ordered classification labels.

[0114] Moreover, compared with the prior art, the numerical encoding of ordered classification labels is used to model the regression problem. For example, the wealth levels in the ordered classification labels are encoded as 3, 2, and 1, and the default distance size of each wealth level is 1. This method requires subjective judgment of the distance size of each wealth level in order to perform numerical encoding, but if the encoding is inaccurate, noise will be introduced; then the regression algorithm is used to output continuous values, and thresholds need to be defined to divide each wealth level, which again introduces subjective problems. The method for determining ordered classification labels provided in the embodiment of the present application avoids subjective judgment of numerical encoding and threshold division, and ensures the accuracy of the user's ordered classification labels.

[0115] In addition, compared with the prior art, the Ordinal Logistic algorithm in the statistical method is used, and the assumption that the distances between the levels in the ordered classification labels are the same is used. However, the actual situation often cannot meet the assumption, or it is difficult to determine whether the assumption is met. The method for determining the ordered classification labels provided in the embodiment of the present application does not require the above assumptions, avoids the limitations brought by additional assumptions, and ensures the accuracy of determining the ordered classification labels.

[0116] exist Figure 2Based on the embodiment shown, in a possible implementation, Figure 5 1 is a flow chart of a method for determining ordered classification labels provided by another embodiment of the present application. The method can be executed by an ordered classification label determination device, which can be implemented by software and / or hardware. The method for determining ordered classification labels is described below with a terminal device as the execution subject. Figure 5 As shown, the ordered classification label determination method provided in the embodiment of the present application may further include, before inputting the feature information into the ordered classification neural network model in step S102:

[0117] Step S201: utilizing the hierarchical relationship of the ordered classification labels to derive the hierarchical inclusion relationship of the ordered classification labels.

[0118] Taking the above-mentioned ordered classification label as wealth level, the wealth level includes five levels of 0, 1, 2, 3, and 4, and the hierarchical relationship of wealth level is 4>3>2>1>0 as an example, in one possible implementation, the hierarchical inclusion relationship of the derived wealth level can be: wealth level ≥1→wealth level ≥2→wealth level ≥3→wealth level = 4. In another possible implementation, the hierarchical inclusion relationship of the derived wealth level can also be: wealth level ≤4→wealth level ≤3→wealth level ≤2→wealth level = 1. The embodiment of the present application only introduces the hierarchical inclusion relationship of wealth level as: wealth level ≥1→wealth level ≥2→wealth level ≥3→wealth level = 4 as an example.

[0119] Step S202: Analyze the probability operation logic of the hierarchical relationship and the hierarchical inclusion relationship, and use the probability operation logic to build a neural network structure.

[0120] Under different representations of hierarchical inclusion relations, the probability operation logic of hierarchical relations and hierarchical inclusion relations will be different, and the principles are similar, so the embodiments of this application will not be repeated. Taking the above as an example, the probability operation logic between the hierarchical relationship of wealth levels and the hierarchical inclusion relationship of wealth levels is analyzed as formula (5), and the neural network structure is built using the probability operation logic. The neural network structure is used to calculate the inclusion probability of each level and the conditional probability of the intermediate nodes, and to determine the probability of each level of the ordered classification label using the probability operation logic.

[0121] The embodiments of the present application do not limit the specific structure of the neural network structure, as long as it can generate the hierarchical probability of the ordered classification labels by utilizing the hierarchical relationship and hierarchical inclusion relationship of the ordered classification labels.

[0122] In one possible implementation, the neural network structure includes a feature part and a prediction part. The feature part is used to determine the individual features between each first task, and the prediction part is used to determine the level probability of the ordered classification labels by utilizing the probability operation logical relationship between the individual features and each second task. The first task is used to determine the hierarchical inclusion probability of the ordered classification labels, and the second task is used to determine the level probability of the ordered classification labels.

[0123] Taking the above example, the first task is to predict P 1+ , P 2+ , P 3+ The second task is to predict P0, P1, P2, P3, and P4. The feature part is used to utilize each P k+ The prediction unit is used to use the probability operation logic relationship between the personality characteristics and the second task, and can refer to formula (5) to achieve the prediction of the level probability of the most ordered classification label.

[0124] The embodiment of the present application does not limit the specific network structure construction method of the feature unit and the prediction unit. In a possible implementation method, Figure 6 is a schematic diagram of a neural network structure provided in an embodiment of the present application, such as Figure 6 As shown, the neural network structure includes a feature unit 21 and a prediction unit 22. The feature unit 21 may include a shared feature layer 211 and a personalized feature layer 212, and the neural network structure is constructed using probability operation logic, including:

[0125] The shared feature layer 211 of the neural network structure is constructed by utilizing the shared features between the first tasks; the individual feature layer 212 of the neural network structure is constructed by utilizing the individual features between the first tasks; and the prediction part 22 of the neural network structure is constructed by utilizing the probability operation logic relationship between the first task and the second task to obtain the neural network structure.

[0126] In the embodiment of the present application, by taking into account the shared features between the hierarchical inclusion probabilities of ordered classification labels and the individual features between the hierarchical inclusion probabilities, not only the learning accuracy of the neural network structure can be improved, but also the working efficiency of the neural network structure can be improved.

[0127] In a possible implementation, the shared feature layer 211 may include a normalization layer, a concatenation layer, and a fully connected layer. The user's feature information is normalized by the normalization layer, and then the normalized feature information is concatenated and passed through the fully connected layer to obtain the shared features.

[0128] In another possible implementation, in the ordered classification label determination method provided in the embodiment of the present application, the shared feature layer 211 may include an embedding vector layer, a feature screening layer and a splicing layer, the embedding vector layer is used to generate an embedding vector (Embedding) of feature information; the feature screening layer is used to perform a linear transformation on the embedding vector, determine the linear transformation value of the embedding vector, and map the linear transformation value using a nonlinear function, and filter the embedding vector with a negative linear transformation value; the splicing layer is used to splice the filtered embedding vectors to obtain shared features between the first tasks.

[0129] The embedding vector layer is used to generate the embedding vector of feature information. For example, the dimension of the embedding vector can be 10-20. If the feature information is a categorical feature, the embedding vector layer randomly generates an embedding vector for each enumeration value. For example, if the feature information is the work industry, the enumeration values ​​of the work industry can be represented as a 10-dimensional vector: the financial industry is represented as [0.56, 0.61, ..., 0.93, 0.33]; the manufacturing industry is represented as [0.80, 0.14, ..., 0.30, 0.82]; the catering industry is represented as [0.15, 0.98, ..., 0.46, 0.01]. The randomly generated embedding vector is trained through the neural network structure to learn the final embedding vector.

[0130] If the feature information is a numerical feature, the numerical value can be first binned and converted into a categorical feature, and then an embedding vector is randomly generated. For example, if the feature information is the amount of credit card consumption in the current month, the amount of credit card consumption in the current month can be first divided into four intervals: "≥100,000", "[10,000, 100,000)", "[5,000, 10,000)", and "<5,000", and then the embedding vectors are randomly generated for each of the four intervals. For example, the interval of ≥100,000 is represented as [0.46, 0.55, ..., 0.15, 0.32]; the interval of [10,000, 100,000) is represented as [0.52, 0.25, ..., 0.96, 0.88]; the interval of [5,000, 10,000) is represented as [0.23, 0.26, ..., 0.12, 0.31]; the interval of <5,000 is represented as [0.63, 0.28, ..., 0.75, 0.74]. The randomly generated embedding vector is trained by the neural network structure to learn the final embedding vector.

[0131] After the embedding vector is generated, the embedding vector is linearly transformed through the feature filtering layer, and then mapped through a nonlinear function to filter out the embedding vectors with negative linear transformation values. The nonlinear function can be the relu function, and the calculation expression of the feature filtering layer can be output = ReLU (w Tv+b), v is the input embedding vector, w and b are the parameters of the feature screening layer, w and b can be determined by training the neural network structure. The ReLU function expression is: ReLU(x)=max(0,x), where x represents the linear transformation value of the embedding vector. If the linear transformation value is negative, the output value of the relu function is 0, thereby filtering some features, and the filtering results for each user may be different, thereby achieving individual-level feature screening, solving the problem that traditional neural network structures cannot directly filter features.

[0132] After filtering the embedding vectors, the filtered embedding vectors are concatenated through a concatenation layer (concat) to form a new feature vector, thereby obtaining the shared features between the first tasks.

[0133] In the embodiment of the present application, by representing the feature information as an embedded vector and screening the embedded vector to filter out redundant information, the accuracy of the ordered classification label inference is further improved.

[0134] The embodiment of the present application does not limit the specific network structure of the prediction unit. In a possible implementation, the prediction unit may include three layers of operation nodes, namely, a sigmoid layer 221, a product layer 222, and a subtraction layer 223. The sigmoid layer 221 maps the input feature to a probability value between 0 and 1, and obtains P 1+ , Q 2+ , Q 3+ and Q 4+ The product layer 222 obtains P by multiplying the probability of the output of the previous layer. 2+ , P 3+ and P4, etc.; subtractive layer 223 through 1-P 1+ Get P0 through the adjacent P k+ Subtract them to get P1, P2, and P3.

[0135] In an embodiment of the present application, a feature unit is set in the neural network structure to determine personalized features unique to each level of inclusion probability of the ordered classification labels, so as to better learn the differences between the calculation methods of the inclusion probability of each level, and then through the prediction unit in the neural network structure, the personalized features are used to predict the level probability of the ordered classification labels, thereby improving the learning accuracy of the neural network structure.

[0136] Step S203: training the neural network structure to generate an ordered classification neural network model.

[0137] After the neural network structure is built, an ordered classification neural network model is generated by training the neural network structure. The specific training method can refer to the training method of the neural network structure in the prior art, and the embodiment of the present application does not limit this.

[0138] In a possible implementation, training a neural network structure to generate an ordered classification neural network model includes: obtaining training samples; and adjusting network parameters of the neural network structure using a first loss function based on the training samples to generate an ordered classification neural network model.

[0139] In a possible implementation, the first loss function is:

[0140]

[0141] Among them, loss1 represents the first loss function, M represents the maximum level of the ordered classification label, k represents the level of the ordered classification label, and W k Indicates the weight of the training sample at level k, Y k Indicates whether the actual level of the training sample is k, Y k The value is 0 or 1; P k It is expressed as the predicted probability of the training sample with level k.

[0142] Among them, if the current round of iterative training has not reached a convergence state, the network parameters of the neural network structure are adjusted based on the first loss function, and the next round of iterative training is continued; if the current round of iterative training has reached a convergence state, the training ends. For the specific implementation process of adjusting the network parameters of the neural network structure according to the loss function, please refer to the introduction of the prior art, and this embodiment will not be repeated here.

[0143] In the embodiment of the present application, the reliability of the ordered classification neural network model is improved by analyzing the probability operation logic of the hierarchical relationship and the hierarchical inclusion relationship and using the probability operation logic to build a neural network structure.

[0144] Based on any of the above embodiments, in a possible implementation, a neural network structure is trained to generate an ordered classification neural network model, including: obtaining training samples, the training samples include the user's level label and hierarchical inclusion label; based on the hierarchical inclusion label, using a second loss function to train the neural network structure to determine the shared feature parameters of the shared feature layer; based on the hierarchical inclusion label, using a third loss function to train the neural network structure to determine the individual feature parameters of the individual feature layer; based on the level label, using the first loss function to train the neural network structure to determine the prediction parameters of the prediction unit.

[0145] Each sample in the training sample will be labeled with two sets of prediction labels, namely, level label Y k and the hierarchy contains label Y k+ For example, if the user's wealth level is 3, then its level labels are: Y0 = 0, Y1 = 0, Y2 = 0, Y3 = 1, Y4 = 0, and the corresponding level contains the label: Y 1+=1, Y 2+ =1, Y 3+ =1, Y 4+ =0.

[0146] Based on label Y k+ , the second loss function is used to train the neural network structure. Since each P k+ The tasks are all about learning the possibility of high-level relative to low-level, and there are inherent similarities. The shared feature parameters are learned through the second loss function. The network parameters in the embedding vector layer and the feature screening layer can be trained, and other network parameters can be pre-trained.

[0147] The embodiment of the present application does not limit the specific function formula of the second loss function. In a possible implementation manner, the second loss function is:

[0148]

[0149] Among them, loss2 represents the second loss function, M represents the maximum level of the ordered classification label, k represents the level of the ordered classification label, and Y k+ Indicates whether the actual level of the training sample is greater than or equal to k, Y k+ The value is 0 or 1; P k+ It is expressed as the predicted probability that the rank of a training sample is greater than or equal to k.

[0150] Among them, if the current round of iterative training has not reached a convergence state, the network parameters of the neural network structure are adjusted based on the second loss function, and the next round of iterative training is continued; if the current round of iterative training has reached a convergence state, the training ends. For the specific implementation process of adjusting the network parameters of the neural network structure according to the loss function, please refer to the introduction of the prior art, and this embodiment will not be repeated here.

[0151] After determining the shared feature parameters, fix the shared feature parameters and then k+ , the third loss function is used to train the neural network structure. Through the third loss function, we focus on each P k+ The uniqueness of the task, the personality characteristic parameters of the personality characteristic layer are trained. The strategy of focused learning is adopted, starting from the lowest level, and focusing on only one P at each step. k+ The task is learned and the parameters learned in the previous step are fixed. For example, first focus on P 1+ task, learn its local task parameters and fix them, then focus on P 2+ Task, learn new local task parameters and fix them, and so on, until all P k+ After the task is learned, the personality characteristic parameters are adjusted.

[0152] The embodiment of the present application does not limit the specific function formula of the third loss function. In a possible implementation manner, the third loss function is:

[0153] loss3=[-Y k+ log(P k+ )-(1-Y k+ )log(1-P k+ )];

[0154] Among them, loss3 represents the third loss function, M represents the maximum level of the ordered classification label, k represents the level of the ordered classification label, and Y k+ Indicates whether the actual level of the training sample is greater than or equal to k, Y k+ The value is 0 or 1; P k+ It is expressed as the predicted probability that the rank of a training sample is greater than or equal to k.

[0155] Among them, if the current round of iterative training has not reached a convergence state, the network parameters of the neural network structure are adjusted based on the third loss function, and the next round of iterative training is continued; if the current round of iterative training has reached a convergence state, the training ends. For the specific implementation process of adjusting the network parameters of the neural network structure according to the loss function, please refer to the introduction of the prior art, and this embodiment will not be repeated here.

[0156] After determining the shared feature parameters and the individual feature parameters, the shared feature parameters and the individual feature parameters are fixed, based on the label Y k , the first loss function is used to train the neural network and adjust and optimize the network parameters of the prediction part.

[0157] In the embodiment of the present application, a step-by-step learning strategy is adopted to focus on learning the network parameters in steps, and the similarities and differences between different tasks are utilized to improve the accuracy of the ordered classification neural network model.

[0158] The following is an embodiment of the device of the present application, which can be used to execute the embodiment of the method of the present application. For details not disclosed in the embodiment of the device of the present application, please refer to the embodiment of the method of the present application.

[0159] Figure 7 is a schematic diagram of the structure of an ordered classification label determination device provided in an embodiment of the present application. The device can be implemented by software and / or hardware, for example, by a terminal device, such as Figure 7 As shown, the ordered classification label determination device provided in the embodiment of the present application may include: an acquisition module 31 and a processing module 32.

[0160] The acquisition module 31 is used to acquire the characteristic information of the user, and the characteristic information is used to determine the ordered classification label of the user.

[0161] The processing module 32 is used to input the feature information into the ordered classification neural network model, and the ordered classification neural network model is used to determine the level probability of the user in the ordered classification label according to the level relationship of the ordered classification label and the hierarchical inclusion relationship derived from the level relationship of the ordered classification label.

[0162] The processing module 32 is further used to determine the ordered classification label of the user by using the level probability of the user in the ordered classification label.

[0163] The device of this embodiment can perform the above Figure 2 The technical principle and technical effect of the method embodiment shown are similar to those of the above-mentioned embodiment and will not be described in detail here.

[0164] exist Figure 7 On the basis of the embodiment shown, further, Figure 8 is a schematic diagram of the structure of an ordered classification label determination device provided by another embodiment of the present application. The device can be implemented by software and / or hardware, for example, by a terminal device, such as Figure 8 As shown, the ordered classification label determination device provided in the embodiment of the present application may also include a derivation module 33, a construction module 34 and a training module 35.

[0165] The derivation module 33 is used to derive the hierarchical inclusion relationship of the ordered classification labels by utilizing the hierarchical relationship of the ordered classification labels; the construction module 34 is used to parse the probabilistic operation logic of the hierarchical relationship and the hierarchical inclusion relationship, and to construct the neural network structure by utilizing the probabilistic operation logic; the training module 35 is used to train the neural network structure and generate an ordered classification neural network model.

[0166] In one possible implementation, the neural network structure includes a feature part and a prediction part. The feature part is used to determine the individual features between each first task, and the prediction part is used to determine the level probability of the ordered classification labels by utilizing the individual features and the probability operation logical relationship between each second task. The first task is used to determine the hierarchical inclusion probability of the ordered classification labels, and the second task is used to determine the level probability of all ordered classification labels.

[0167] In a possible implementation, the feature part includes a shared feature layer and a personalized feature layer, and the building module 34 is specifically used for:

[0168] The shared feature layer of the neural network structure is constructed by utilizing the shared features between the first tasks; the individual feature layer of the neural network structure is constructed by utilizing the individual features between the first tasks; the prediction part of the neural network structure is constructed by utilizing the probability operation logic relationship between the first task and the second task to obtain the neural network structure.

[0169] In one possible implementation, the shared feature layer includes an embedding vector layer, a feature screening layer, and a splicing layer. The embedding vector layer is used to generate an embedding vector of feature information; the feature screening layer is used to perform a linear transformation on the embedding vector, determine the linear transformation value of the embedding vector, and use a nonlinear function to map the linear transformation value, and filter the embedding vector with a negative linear transformation value; the splicing layer is used to splice the filtered embedding vectors to obtain shared features between the first tasks.

[0170] In a possible implementation, the training module 35 is specifically configured to:

[0171] Acquire training samples; based on the training samples, use the first loss function to adjust the network parameters of the neural network structure to obtain an ordered classification neural network model.

[0172] In a possible implementation, the first loss function is:

[0173]

[0174] Among them, loss1 represents the first loss function, M represents the maximum level of the ordered classification label, k represents the level of the ordered classification label, and W k Indicates the weight of the training sample at level k, Y k Indicates whether the actual level of the training sample is k, Y k The value is 0 or 1; P k It is expressed as the predicted probability of the training sample with level k.

[0175] In a possible implementation, the training module 35 is specifically configured to:

[0176] Acquire training samples, which include the user's level label and hierarchical inclusion label; based on the hierarchical inclusion label, use the second loss function to train the neural network structure to determine the shared feature parameters of the shared feature layer; based on the hierarchical inclusion label, use the third loss function to train the neural network structure to determine the individual feature parameters of the individual feature layer; based on the level label, use the first loss function to train the neural network structure to determine the prediction parameters of the prediction unit.

[0177] In a possible implementation, the second loss function is:

[0178]

[0179] The third loss function is:

[0180] loss3=[-Y k+ log(P k+ )-(1-Y k+ )log(1-P k+ )];

[0181] Among them, loss2 represents the second loss function, loss3 represents the third loss function, M represents the maximum level of the ordered classification label, k represents the level of the ordered classification label, and Y k+ Indicates whether the actual level of the training sample is greater than or equal to k, Y k+ The value is 0 or 1; P k+ It is expressed as the predicted probability that the rank of a training sample is greater than or equal to k.

[0182] The device of this embodiment can perform the above Figure 5 The technical principle and technical effect of the method embodiment shown are similar to those of the above-mentioned embodiment and will not be described in detail here.

[0183] The device embodiments provided in this application are merely illustrative. Figure 7 and Figure 8 The module division in the description is only a logical function division, and there may be other division methods in actual implementation. For example, multiple modules can be combined or integrated into another system. The coupling between the modules can be realized through some interfaces, which are usually electrical communication interfaces, but it is not excluded that they may be mechanical interfaces or other forms of interfaces. Therefore, the modules described as separate components may or may not be physically separated, and may be located in one place or distributed in different locations of the same or different devices.

[0184] Fig. 9 is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application, such as Fig. 9 As shown, the electronic device includes:

[0185] A receiver 40, a transmitter 41, a processor 42, a memory 43 and a computer program; wherein the receiver 40 and the transmitter 41 implement data transmission between other devices, the computer program is stored in the memory 43 and is configured to be executed by the processor 42, and the computer program includes instructions for executing the above-mentioned method for processing transactions in the blockchain. For its content and effects, please refer to the method embodiment.

[0186] In addition, an embodiment of the present application further provides a computer-readable storage medium, in which computer-executable instructions are stored. When at least one processor of a user device executes the computer-executable instructions, the user device executes the above-mentioned various possible methods.

[0187] Among them, computer-readable media include computer storage media and communication media, wherein the communication media include any media that facilitates the transmission of computer programs from one place to another. The storage medium can be any available medium that can be accessed by a general or special-purpose computer. An exemplary storage medium is coupled to the processor so that the processor can read information from the storage medium and write information to the storage medium. Of course, the storage medium can also be an integral part of the processor. The processor and the storage medium can be located in an ASIC. In addition, the ASIC can be located in a user device. Of course, the processor and the storage medium can also be present in a communication device as discrete components.

[0188] Those skilled in the art can understand that all or part of the steps of implementing the above-mentioned method embodiments can be completed by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, the steps of the above-mentioned method embodiments are executed; and the aforementioned storage medium includes: ROM, RAM, disk or optical disk and other media that can store program codes.

[0189] Those skilled in the art will readily appreciate other embodiments of the present application after considering the specification and practicing the invention disclosed herein. The present application is intended to cover any modification, use or adaptation of the present application, which follows the general principles of the present application and includes common knowledge or customary techniques in the art that are not disclosed in the present application. The specification and examples are intended to be exemplary only, and the true scope and spirit of the present application are indicated by the claims.

[0190] It should be understood that the present application is not limited to the precise structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present application is limited only by the appended claims.

Claims

1. A method for determining ordered classification labels, characterized in that: Utilizing the hierarchical relationship of the ordered classification labels, deriving the hierarchical inclusion relationship of the ordered classification labels; Analyze the probability operation logic of the hierarchical relationship and the hierarchical inclusion relationship, and use the probability operation logic to build a neural network structure; the neural network structure includes a feature part and a prediction part, the feature part is used to determine the individual characteristics between the first tasks, the prediction part is used to use the probability operation logic relationship between the individual characteristics and the second tasks to determine the hierarchical probability of the ordered classification labels, the first task is used to determine the hierarchical inclusion probability of the ordered classification labels, and the second task is used to determine the hierarchical probability of the ordered classification labels; Training the neural network structure to generate an ordered classification neural network model; Acquire characteristic information of the user, where the characteristic information is used to determine an ordered classification label of the user; Inputting the feature information into an ordered classification neural network model, the ordered classification neural network model is used to determine the level probability of the user in the ordered classification label according to the level relationship of the ordered classification label and the hierarchical inclusion relationship derived from the level relationship of the ordered classification label; The ordered classification label of the user is determined by utilizing the rank probability of the user in the ordered classification label.

2. The method according to claim 1, characterized in that The feature part includes a shared feature layer and a personal feature layer, and the neural network structure is constructed by using the probability operation logic, including: The shared feature layer of the neural network structure is constructed by utilizing the shared features between the first tasks; the individual feature layer of the neural network structure is constructed by utilizing the individual features between the first tasks; the prediction part of the neural network structure is constructed by utilizing the probability operation logical relationship between the first task and the second task to obtain the neural network structure.

3. The method according to claim 2, characterized in that The shared feature layer includes an embedding vector layer, a feature screening layer and a splicing layer. The embedding vector layer is used to generate an embedding vector of the feature information; the feature screening layer is used to perform a linear transformation on the embedding vector, determine the linear transformation value of the embedding vector, and map the linear transformation value using a nonlinear function, and filter the embedding vector with a negative linear transformation value; the splicing layer is used to splice the embedded vectors after filtering to obtain shared features between the first tasks.

4. The method according to any one of claims 2 to 3, characterized in that: The step of training the neural network structure to generate the ordered classification neural network model includes: Get training samples; Based on the training samples, the network parameters of the neural network structure are adjusted using a first loss function to obtain the ordered classification neural network model.

5. The method according to claim 4, characterized in that The first loss function is: Among them, loss1 represents the first loss function, M represents the maximum level of the ordered classification label, k represents the level of the ordered classification label, and W k Indicates the weight of the training sample at level k, Y k Indicates whether the actual level of the training sample is k, Y k The value is 0 or 1; P k It is expressed as the predicted probability of the training sample with level k.

6. The method according to claim 5, characterized in that The training of the neural network structure to generate the ordered classification neural network model includes: obtaining training samples, wherein the training samples include user level labels and hierarchical inclusion labels; Based on the hierarchical inclusion labels, the neural network structure is trained using a second loss function to determine shared feature parameters of the shared feature layer; Based on the hierarchical inclusion labels, the neural network structure is trained using a third loss function to determine the personality characteristic parameters of the personality characteristic layer; Based on the level labels, the neural network structure is trained using a first loss function to determine prediction parameters of the prediction unit.

7. The method according to claim 6, characterized in that The second loss function is: ； The third loss function is: Among them, loss2 represents the second loss function, loss3 represents the third loss function, M represents the maximum level of the ordered classification label, and k represents the level of the ordered classification label. Indicates whether the actual level of the training sample is greater than or equal to k, The value is 0 or 1; It is expressed as the predicted probability that the rank of a training sample is greater than or equal to k.

8. A device for determining ordered classification labels, characterized in that: include: A derivation module, used to derive a hierarchical inclusion relationship of the ordered classification labels by utilizing the hierarchical relationship of the ordered classification labels; A building module, used to analyze the probability operation logic of the hierarchical relationship and the hierarchical inclusion relationship, and use the probability operation logic to build a neural network structure; the neural network structure includes a feature part and a prediction part, the feature part is used to determine the individual characteristics between each first task, the prediction part is used to determine the hierarchical probability of the ordered classification label by using the probability operation logic relationship between the individual characteristics and each second task, the first task is used to determine the hierarchical inclusion probability of the ordered classification label, and the second task is used to determine the hierarchical probability of the ordered classification label; A training module, used to train the neural network structure to generate an ordered classification neural network model; An acquisition module, used to acquire characteristic information of a user, wherein the characteristic information is used to determine an ordered classification label of the user; A processing module, used for inputting the feature information into an ordered classification neural network model, wherein the ordered classification neural network model is used for determining the level probability of the user in the ordered classification label according to the level relationship of the ordered classification label and the hierarchical inclusion relationship derived from the level relationship of the ordered classification label; The processing module is further used to determine the ordered classification label of the user by using the rank probability of the user in the ordered classification label.

9. An electronic device, characterized in that: include: A processor, and a memory communicatively connected to the processor; The memory stores computer-executable instructions; The processor executes the computer-executable instructions stored in the memory to implement the ordered classification label determination method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the ordered classification label determination method according to any one of claims 1 to 7.

11. A computer program product, characterized in that The method comprises computer-executable instructions, which implement the method according to any one of claims 1 to 7 when executed by a processor.

Images