A predictive method, device, and medium for enterprise label interpretability

Abstract

This application provides an interpretable prediction method, device, and storage medium for enterprise tags, including: acquiring public and private enterprise data; constructing an enterprise knowledge graph based on the public and private enterprise data; training a target prediction model based on the private, public, and knowledge graphs, and obtaining enterprise tag prediction results; determining a first global contribution value for a first enterprise tag and a second global contribution value for a second enterprise tag using the target prediction model; obtaining a joint contribution value based on the first and second global contribution values; obtaining a first weighted contribution value for the first enterprise tag based on the first global contribution value and the joint contribution value; determining second weighted contribution values ​​for other enterprise tags among several enterprise tags; and interpreting the enterprise tag prediction results based on the first and second weighted contribution values. This application aims to improve the accuracy of tag prediction and achieve interpretability of the results.

Description

Technical Field

[0001] This application relates to the field of computer technology, and more particularly to a method for predicting enterprise tags that can be interpreted, a computer device, and a computer storage medium. Background Technology

[0002] Small and micro-sized enterprises (SMEs) within the industrial park are the primary entities supporting the park's sustainable development and enhanced competitiveness. Real-time monitoring of their development and operational status is crucial for ensuring the park's sustainability. Therefore, to improve the park's competitiveness and protect the interests of SMEs, it is necessary to predict their development status and identifying relevant characteristics to provide development suggestions and support business decision-making.

[0003] To improve the accuracy of label prediction, existing technologies can fully learn the features of data and integrate information from various dimensions to achieve relatively accurate predictions of various labels for micro and small enterprises. However, the complexity of deep learning models and the lack of interpretability of their results limit their widespread application in risk prediction industries such as healthcare and finance. In addition, some technologies process data using logistic regression, random forests, XGBoost, and deep learning, and then average the prediction results from multiple groups to make the prediction results somewhat interpretable. However, the model fitting ability of these methods is weak, and they cannot integrate the complex feature interactions of the data to achieve more accurate label predictions. Therefore, it is necessary to propose an interpretable prediction method for enterprise labels that can integrate information from various dimensions to improve the accuracy of enterprise label predictions and achieve interpretability of the prediction results, thereby promoting the credible application of deep learning methods in risk prediction in enterprise decision-making, finance, and other fields. Summary of the Invention

[0004] This application provides a method, computer device, and computer storage medium for predicting enterprise labels that can be interpreted, which can improve the accuracy of enterprise label prediction while achieving interpretability of the prediction results.

[0005] Firstly, this application provides a predictive method for enterprise-specific labels that can be interpreted, the method comprising:

[0006] Acquire public and private enterprise data, and construct an enterprise knowledge graph based on the public and private enterprise data. The enterprise knowledge graph includes several enterprise tags and dynamic data corresponding to each enterprise tag.

[0007] A target prediction model is trained based on the enterprise's private data, the enterprise's public data, and the enterprise's knowledge graph. The enterprise's knowledge graph is then input into the target prediction model to obtain the enterprise tag prediction result.

[0008] Using the target prediction model, the first global contribution value of the first enterprise label and the second global contribution value of the second enterprise label are determined respectively, and a joint contribution value is obtained based on the first global contribution value and the second global contribution value.

[0009] Based on the first global contribution value and the joint contribution value, the first weighted contribution value of the first enterprise label is obtained, and the second weighted contribution value of other enterprise labels among the plurality of enterprise labels is determined;

[0010] The enterprise prediction results are interpreted based on the first weighted contribution value and the second weighted contribution value.

[0011] Secondly, this application also provides a computer device, the computer device comprising:

[0012] Memory and processor;

[0013] The memory is connected to the processor and is used to store programs;

[0014] The processor is configured to implement the steps of the enterprise tag interpretable prediction method as described in any of the embodiments of this application by running a program stored in the memory.

[0015] Thirdly, this application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, causes the processor to implement the steps of the enterprise tag interpretable prediction method as described in any of the embodiments of this application.

[0016] The enterprise tag interpretation prediction method, computer equipment, and storage medium disclosed in this application can improve the prediction accuracy of enterprise tags by integrating information from various dimensions based on both public and private enterprise data. Furthermore, by employing a model-independent post-interpretation method to calculate the weighted contribution value of each enterprise tag, the prediction results can be interpreted, thereby promoting the reliable application of deep learning methods in risk prediction in areas such as corporate decision-making and finance.

[0017] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0018] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a schematic diagram illustrating the steps of an interpretable prediction method for enterprise tags provided in an embodiment of this application;

[0020] Figure 2 This is a schematic diagram illustrating the specific steps of step S11 provided in an embodiment of this application;

[0021] Figure 3 This is a schematic diagram illustrating a specific step of step S13 provided in an embodiment of this application;

[0022] Figure 4 This is a schematic diagram illustrating another specific step of step S13 provided in an embodiment of this application;

[0023] Figure 5 This is a schematic block diagram of a computer device provided in an embodiment of this application;

[0024] Figure 6 This is a schematic diagram of the structure of a computer-readable storage medium provided in this application.

[0025] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Detailed Implementation

[0026] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0027] The flowchart shown in the attached diagram is for illustrative purposes only and does not necessarily include all content and operations / steps, nor does it necessarily have to be performed in the order described. For example, some operations / steps can be broken down, combined, or partially merged, so the actual execution order may change depending on the actual situation.

[0028] It should be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the scope of the application. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0029] It should be understood that, in order to clearly describe the technical solutions of the embodiments of this application, the terms "first" and "second" are used in the embodiments of this application to distinguish identical or similar items with essentially the same function and effect. For example, the first identification model and the second identification model are only used to distinguish different callback functions and do not limit their order. Those skilled in the art will understand that the terms "first" and "second" do not limit the quantity or execution order, and the terms "first" and "second" do not necessarily mean they are different.

[0030] It should also be understood that the term "and / or" as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0031] The following detailed description of some embodiments of this application is provided in conjunction with the accompanying drawings. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0032] Please see Figure 1 , Figure 1 This is a schematic flowchart illustrating the steps of an interpretable prediction method for enterprise tags provided in an embodiment of this application. This method can be applied to computer devices to implement enterprise tag prediction based on interpretable deep learning.

[0033] like Figure 1 As shown, the predictive method that can be interpreted by the enterprise label includes steps S11 to S15.

[0034] Step S11: Obtain enterprise public data and enterprise private data, and construct an enterprise knowledge graph based on the enterprise public data and enterprise private data. The enterprise knowledge graph includes several enterprise tags and dynamic data corresponding to each enterprise tag.

[0035] Among them, public enterprise data refers to data that enterprises publicly disclose, including but not limited to enterprise news, enterprise operating data, and industry policies; private enterprise data refers to data that is private to the enterprise and does not involve the enterprise's confidentiality scope, including but not limited to water and electricity data, property data, company factory area data, and personnel data.

[0036] Furthermore, this application does not limit the method of obtaining public and private corporate data. For example, it can be obtained through news outlets, newspapers, the company's official website, or other trustworthy websites. This application uses the company's official website as an example. Specifically, distributed data collection technology can be used to collect information from the company's official website at regular intervals every day and store the data. In this way, both public and private corporate data can be obtained.

[0037] After obtaining both public and private enterprise data, an enterprise knowledge graph can be constructed based on these data. This enterprise knowledge graph includes several enterprise tags and dynamic data corresponding to each tag.

[0038] In this context, an enterprise knowledge graph is a data structure composed of entities, relationships, and attributes. In this embodiment, enterprises correspond to entities; enterprise tags and their corresponding dynamic data correspond to attributes and belong to the enterprises. Thus, an enterprise knowledge graph can be constructed.

[0039] Optionally, please refer to Figure 2 , Figure 2 This is a schematic diagram illustrating the specific steps of step S11 provided in an embodiment of this application. After obtaining the enterprise's public and private data, a knowledge graph of the enterprise can be constructed based on the enterprise's public and private data through steps S101 to S104.

[0040] Step S101: Perform preprocessing operations on the enterprise's public data and private data respectively.

[0041] The preprocessing operations include, but are not limited to, data cleaning and transformation operations, variable filtering operations, and dataset partitioning operations, which are not limited in this application.

[0042] Because the data collected from websites is often disorganized, containing a large amount of irrelevant data, filtering is necessary. Furthermore, the data formats from different data sources vary significantly and need to be standardized, requiring data cleaning and conversion. Specifically, HTML tags and special characters can be removed from the collected data, and data from different data sources with varying formats can be unified, thus achieving data cleaning and conversion. Further, suitable data can be filtered based on the needs of tag prediction, thus achieving variable filtering. Additionally, the collected data can be categorized and organized according to different types to achieve data normalization.

[0043] Step S102: Based on the preprocessed public enterprise data and private enterprise data, merge them to obtain the enterprise dataset.

[0044] Optionally, based on the preprocessed public and private enterprise data, an enterprise dataset is obtained by merging the data, including: determining several enterprise labels, and obtaining corresponding enterprise data information from the preprocessed public and private enterprise data based on each enterprise label; performing a data mapping operation on each enterprise label and its corresponding enterprise data information to obtain several fifth samples; determining the correlation between the enterprise label in each fifth sample and its corresponding enterprise data information; and merging multiple fifth samples with a correlation greater than a preset threshold to obtain the enterprise dataset.

[0045] Specifically, enterprise labels can be predetermined. For example, enterprise labels may include basic enterprise attribute labels, such as business scope labels, industry labels, company tenure labels, regional labels, and scale labels; enterprise operating status labels, such as enterprise innovation labels, enterprise competitiveness labels, enterprise development potential labels, and enterprise development momentum labels; and enterprise risk information labels, such as enterprise self-risk labels, enterprise surrounding risk labels, early warning and reminder labels, and operational risk labels. This application does not limit these.

[0046] Furthermore, based on pre-defined tags, corresponding enterprise data information can be obtained from pre-processed public and private enterprise data. That is, related data can be obtained from public and private enterprise data based on tags. For example, since the size tag is related to company factory area data and personnel data, data such as company factory area data and personnel data can be obtained from pre-processed public and private enterprise data based on the size tag.

[0047] After obtaining the enterprise tags and their corresponding enterprise data information, a data mapping operation can be performed on each enterprise tag and its corresponding enterprise data information to obtain several fifth samples.

[0048] The fifth sample is a sample after performing a data mapping operation on each enterprise label and its corresponding enterprise data information, which includes the relevance information of each enterprise label and its corresponding enterprise data information.

[0049] Specifically, the relevance between each enterprise tag and its corresponding enterprise data information can be calculated through data mapping operations. This ensures that the fifth sample includes the relevance information of each enterprise tag and its corresponding enterprise data information.

[0050] It should be noted that data mapping refers to the process of establishing a correspondence between data elements based on two given data models. Furthermore, this application does not limit the method of data mapping operations; for example, it can be done through manual coding or visualization. Manual coding involves directly defining the data correspondence using programming languages ​​such as XSLT, JAVA, or C++. Visualization typically allows users to draw a line between data items to define the correspondence between them.

[0051] Furthermore, multiple fifth samples with a relevance greater than a preset threshold can be selected and fused to obtain an enterprise dataset. This yields an enterprise dataset with high relevance, which includes enterprise tags and their corresponding enterprise data.

[0052] It should be noted that this application does not limit the above-mentioned preset threshold, for example, it can be 80%. Based on this, if the relevance between the above-obtained enterprise tags and their corresponding enterprise data information is greater than 80%, they can be merged to obtain an enterprise dataset.

[0053] Step S103: Extract enterprise entity and enterprise label information from the enterprise dataset, and construct a Euclidean space model of the two based on the enterprise entity and enterprise label information.

[0054] The enterprise entity is the enterprise itself, and the enterprise tag information includes the enterprise tag and several data information corresponding to each enterprise tag.

[0055] It's important to note that Euclidean space, also known as Euclidean space, is a generalization in mathematics of the 2- and 3-dimensional spaces studied by Euclidean. Specifically, it transforms distances, and related concepts like length and angle, into coordinate systems of arbitrary dimensions. Euclidean space is a special metric space that allows us to investigate its topological properties, such as compactness. Inner product spaces are a generalization of Euclidean space.

[0056] A three-dimensional Euclidean space model can be constructed based on enterprise entity and enterprise tag information. Specifically, a linear space is constructed based on the enterprise tags of the enterprises in the enterprise tag information and the data information corresponding to each enterprise tag. Then, the inner product is defined based on the enterprise entities and assigned to the linear space, thus obtaining the Euclidean space model of enterprise entities and enterprise tag information.

[0057] Step S104: Construct an enterprise knowledge graph based on the Euclidean space model.

[0058] Because both public and private enterprise data are constantly being updated, a dynamic spatial model needs to be built to enable data updates when constructing an enterprise knowledge graph.

[0059] Optionally, since vectors have translation invariance, each vector in the Euclidean space model can be mapped to a vector in the negative curvature hyperbolic space, thus obtaining a negative curvature hyperbolic space model of enterprise entities and enterprise tag information, which is also the enterprise knowledge graph.

[0060] It should be noted that negative curvature hyperbolic space is an open, infinite space, possessing the ability to express hierarchical structures and infinite spatial capacity. It can not only restore the hierarchical structure of the data itself but also express parameters with the same capacity as Euclidean space using fewer parameters. Furthermore, negative curvature hyperbolic space allows for real-time data updates, enabling the updating of enterprise knowledge graphs.

[0061] In this embodiment, public and private enterprise data can be fused to obtain multi-dimensional data, thereby improving the accuracy of enterprise tag prediction. Furthermore, the enterprise knowledge graph can be updated based on a negative curvature hyperbolic space, ensuring timely updates to the enterprise tag prediction results.

[0062] Step S12: Train the target prediction model based on the enterprise's private data, public data, and knowledge graph, and input the enterprise knowledge graph into the target prediction model to obtain the enterprise label prediction result.

[0063] Optionally, a target prediction model is trained based on enterprise private data, enterprise public data, and enterprise knowledge graph, including: constructing an initial prediction model of enterprise tags based on enterprise private data and enterprise public data; and optimizing the loss function of the initial prediction model using the enterprise knowledge graph to obtain a target prediction model of enterprise tags.

[0064] Specifically, a deep learning network model for label prediction can be established first. Then, a training sample set can be constructed based on the enterprise's private and public data. The weights can be adaptively adjusted based on the data in the training samples to train the deep learning network model for label prediction. When the weights in the samples reach the maximum number of iterations, the training stops, and the initial prediction model for the enterprise labels is obtained.

[0065] Optionally, in order to make the prediction results of enterprise tags closer to the true values, the loss function of the initial prediction model can be optimized using the enterprise knowledge graph, thereby obtaining the optimized target prediction model for enterprise tags.

[0066] Furthermore, after obtaining the target prediction model, the enterprise knowledge graph can be input into the target prediction model to obtain the enterprise tag prediction results. The enterprise tag prediction results include several enterprise tags and the weighted contribution value corresponding to each enterprise tag, that is, the importance level.

[0067] In this embodiment, a dynamic enterprise knowledge graph can be constructed based on enterprise private data and enterprise public data to achieve real-time data updates. Furthermore, after inputting the enterprise knowledge graph into the target prediction, enterprise tag prediction results can be obtained. Since the data in the enterprise knowledge graph is dynamic, the timely updating of enterprise tag prediction results can also be ensured based on the enterprise knowledge graph.

[0068] Step S13: Using the target prediction model, determine the first global contribution value of the first enterprise label and the second global contribution value of the second enterprise label among several enterprise labels, and obtain the joint contribution value based on the first global contribution value and the second global contribution value.

[0069] Wherein, the first enterprise label is a randomly selected enterprise label, and this application does not limit it; the first global contribution is the global contribution corresponding to the first enterprise label; the second enterprise label is a randomly selected enterprise label that is different from the first enterprise label; the second global contribution is the global contribution corresponding to the second enterprise label.

[0070] It is understandable that after obtaining the enterprise label prediction results, which include several enterprise labels and the contribution value corresponding to each enterprise label, through the target prediction model based on the above embodiments, enterprise labels and their corresponding weighted contribution values ​​can also be obtained through other methods to interpret the enterprise label prediction results.

[0071] Optional, please refer to Figure 3 , Figure 3 This is a schematic diagram illustrating a specific step of step S13 provided in an embodiment of this application. For example... Figure 3 As shown, steps S131 to S134 can be used to determine the first global contribution value of the first enterprise label among several enterprise labels through the target prediction model.

[0072] Step S131: Obtain the data corresponding to each enterprise label in the first moment from the several enterprise labels.

[0073] Step S132: Based on each enterprise label and its corresponding data at the first time, construct a first sample and a second sample, wherein the first sample contains each enterprise label and its corresponding data at the first time, and the second sample does not contain the first enterprise label and its corresponding data at the first time.

[0074] Since the data corresponding to the enterprise tags is dynamic and changes over time, a first moment can be determined, and the data corresponding to each enterprise tag among several enterprise tags at that first moment can be obtained. The first moment is a randomly selected time, such as 3 PM on November 1st; this application does not limit this selection.

[0075] Furthermore, based on the data obtained above, a first sample and a second sample can be constructed. The first sample, also known as the original sample, includes each enterprise label and its corresponding data at the first time point; the second sample is a sample that does not include the first enterprise label and its corresponding data at the first time point.

[0076] Step S133: The first sample and the second sample are respectively input into the target prediction model to obtain the first prediction result and the second prediction result, and the first marginal contribution value of the first enterprise label is obtained based on the first prediction result and the second prediction result.

[0077] The first prediction result is the prediction result obtained based on the first sample input target prediction model, which includes each enterprise label and the marginal contribution value corresponding to each enterprise label at the first time point; the second prediction result is the prediction result obtained based on the second sample input target prediction model, which includes other labels except the first enterprise label at the first time point and their corresponding marginal contribution values.

[0078] Furthermore, the difference between the first prediction result and the second prediction result can be calculated to obtain the first marginal contribution value of the first enterprise label, wherein the first marginal contribution value is the marginal contribution value of the first enterprise label at the first time.

[0079] Step S134: Determine the second marginal contribution value of the first enterprise label at the second time point, and obtain the first global contribution value based on the first marginal contribution value and the second marginal contribution value.

[0080] The second time point is a random time point different from the first time point, such as 3 PM on November 2nd, which is not limited in this application; the second marginal contribution value is the marginal contribution value of the first enterprise label at the second time point; the first global contribution value is the global contribution value corresponding to the first enterprise label.

[0081] Specifically, the second marginal contribution value of the first enterprise label at the second time point can be obtained through the above steps, and the average value of the second marginal contribution value and the first marginal contribution value can be calculated to obtain the first global contribution value of the first enterprise label.

[0082] Optionally, the above steps can be repeated at different times to obtain multiple marginal contribution values ​​for the first enterprise label, and the average of these multiple marginal contribution values ​​can be calculated to obtain the final global contribution value for the first enterprise label. By sampling multiple times, the obtained global contribution value can be made more accurate.

[0083] Understandably, steps S131 to S134 can also be used to determine the second global contribution value of the second enterprise label among several enterprise labels using the target prediction model. To avoid repetition, this will not be elaborated upon here. Here, the second enterprise label is an enterprise label different from the first enterprise label, and the second global contribution value is the global contribution value corresponding to the second enterprise label.

[0084] Alternatively, the enterprise knowledge graph can be input into a deep learning model, and the global contribution value of the enterprise label can be obtained through the operation of the global average pooling layer. This application does not limit this.

[0085] For further details, please refer to Figure 4 , Figure 4 This is a schematic diagram illustrating another specific step of step S13 provided in an embodiment of this application. For example... Figure 4 As shown, the joint contribution value can be obtained based on the first global contribution value and the second global contribution value through steps S135 to S138.

[0086] Step S135: Obtain the data corresponding to each enterprise label in the third time step.

[0087] Step S136: Based on each enterprise label and its corresponding data at the third time point, construct a third sample and a fourth sample. The third sample includes each enterprise label and its corresponding data at the third time point, while the fourth sample does not include the first enterprise label, the second enterprise label, and their corresponding data at the third time point.

[0088] Step S137: Input the third sample and the fourth sample into the target prediction model respectively to obtain the third prediction result and the fourth prediction result, and obtain the global contribution value of the first enterprise label and the second enterprise label based on the third prediction result and the fourth prediction result.

[0089] The third time point is a random time point different from the first and second time points, such as 3 PM on November 3rd, which is not limited in this application; the third sample includes each enterprise label and its corresponding data at the third time point; the fourth sample is a sample excluding the first and second enterprise labels and their corresponding data at the third time point; the third prediction result is the prediction result obtained by inputting the third sample into the target prediction model, wherein the third prediction result is each enterprise label and its corresponding marginal contribution value at the third time point; the fourth prediction result is the prediction result obtained by inputting the fourth sample into the target prediction model, wherein the fourth prediction result is the other labels besides the first and second enterprise labels and their corresponding marginal contribution values ​​at the third time point.

[0090] Specifically, the global contribution values ​​of the first enterprise label and the second enterprise label can be obtained by referring to steps S132 to S133. To avoid duplication, they will not be elaborated here.

[0091] Step S138: Obtain the joint contribution value based on the first global contribution value, the second global contribution value, and the global contribution values ​​of the first enterprise label and the second enterprise label.

[0092] Specifically, the combined contribution value can be obtained by subtracting the global contribution value of the first enterprise label and the second enterprise label from the sum of the first global contribution value and the second global contribution value.

[0093] Step S14: Based on the first global contribution value and the joint contribution value, obtain the first weighted contribution value of the first enterprise label, and determine the second weighted contribution value of other enterprise labels among several enterprise labels.

[0094] Among them, the first weighted contribution value is the weighted contribution value of the first enterprise label; the second weighted contribution value is the weighted contribution value of the other enterprise labels.

[0095] Optionally, the above-mentioned method of obtaining the first weighted contribution value of the first enterprise label based on the first global contribution value and the joint contribution value, and determining the second weighted contribution value of other enterprise labels among several enterprise labels, includes: determining the single weight of the first enterprise label based on the first global contribution value, determining the joint weight of the first enterprise label based on the joint contribution value, and determining the weighted contribution value of the first enterprise label based on the single weight and the joint weight.

[0096] Specifically, the global contribution value of each enterprise label can be obtained by referring to steps S131 to S134. The first global contribution value is then divided by the sum of the global contribution values ​​of each enterprise label to obtain the single weight of the first enterprise label. Similarly, steps S131 to S134 can be used to obtain several other joint contribution values ​​that include combinations of two enterprise labels. The joint contribution value of the first enterprise label and the second enterprise label is then divided by the sum of the other joint contribution values ​​to obtain the joint weight of the first enterprise label and the second enterprise label.

[0097] Furthermore, the weighted contribution value of the first enterprise label can be determined based on single weight and joint weight using the following formula:

[0098] overall i =α i φ i +β {ij} φ {ij}

[0099] Among them: overall i The weighted contribution value of the first enterprise label; α i The single weight for the first enterprise label; φ i The first global contribution value; β {ij} The combined weight of the first and second company labels; φ {ij} The combined weight of the first and second enterprise labels.

[0100] Understandably, this method can also be used to determine the second weighted contribution value of other enterprise labels among several enterprise labels. To avoid duplication, this will not be elaborated here.

[0101] Step S15: Interpret the prediction results based on the first weighted contribution value and the second weighted contribution value.

[0102] Since enterprise labels can be predicted using a target prediction model, enterprise label prediction results are obtained, which include several enterprise labels and their corresponding weighted contribution values. Therefore, the interpretation method in this embodiment can be used to interpret the enterprise label prediction results by calculating each enterprise label and its corresponding weighted contribution value. This promotes the reliable application of deep learning methods in risk fields such as enterprise-centric decision-making and finance.

[0103] The enterprise tag interpretation prediction method proposed in this application can improve the prediction accuracy of enterprise tags by integrating information from various dimensions based on both public and private enterprise data. Furthermore, it can obtain enterprise tag prediction results based on a target prediction model, whereby the enterprise tag prediction results include the enterprise tags and their corresponding weighted contributions. In addition, it can calculate the weighted contribution value of each enterprise tag, enabling interpretation of the enterprise tag prediction results, thereby promoting the reliable application of deep learning methods in risk prediction fields such as corporate decision-making and finance.

[0104] The methods and apparatus of this application can be used in a wide variety of general-purpose or special-purpose computing system environments or configurations. For example: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer terminal devices, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, etc.

[0105] For example, the above-described method and apparatus can be implemented as a computer program, which can be used in, for example... Figure 5 It runs on the computer device shown.

[0106] Please see Figure 5 , Figure 5 This is a schematic diagram of a computer device provided in an embodiment of this application. The computer device 300 may be a server. Figure 5 As shown, the computer device 300 includes a processor 301, a memory 302, and a network interface connected via a system bus. The memory 302 may include volatile storage media, non-volatile storage media, and internal memory. The non-volatile storage media may store an operating system and a computer program. The computer program includes program instructions that, when executed, cause the processor 301 to perform any enterprise-label-interpretable prediction method.

[0107] The processor 301 provides computing and control capabilities to support the operation of the entire computer device 300.

[0108] The internal memory provides an environment for the execution of computer programs in non-volatile storage media, which, when executed by the processor 301, enable the processor 301 to perform any enterprise-label-interpretable prediction method.

[0109] This network interface is used for network communication, such as sending assigned tasks. Those skilled in the art will understand that the structure of this computer device 300 is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device 300 to which the present application applies. A specific computer device 300 may include more or fewer components than shown in the figure, or combine certain components, or have different component arrangements.

[0110] It should be understood that processor 301 can be a Central Processing Unit (CPU), but it can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among these, a general-purpose processor can be a microprocessor or any conventional processor.

[0111] In some embodiments, the processor 301 is used to run a computer program stored in the memory 302 to perform the following steps: acquiring public and private enterprise data; constructing an enterprise knowledge graph based on the public and private enterprise data, wherein the enterprise knowledge graph includes several enterprise tags and dynamic data corresponding to each enterprise tag; training a target prediction model based on the private enterprise data, the public enterprise data, and the enterprise knowledge graph, and inputting the enterprise knowledge graph into the target prediction model to obtain enterprise tag prediction results; determining a first global contribution value of a first enterprise tag and a second global contribution value of a second enterprise tag among the several enterprise tags through the target prediction model, and obtaining a joint contribution value based on the first and second global contribution values; obtaining a first weighted contribution value of the first enterprise tag based on the first global contribution value and the joint contribution value, and determining a second weighted contribution value of other enterprise tags among the several enterprise tags; and interpreting the enterprise prediction results based on the first and second weighted contribution values.

[0112] In some embodiments, the processor 301 is further configured to acquire data corresponding to each of the plurality of enterprise tags at a first time; construct a first sample and a second sample based on each enterprise tag and its corresponding data at the first time, wherein the first sample includes each enterprise tag and its corresponding data at the first time, and the second sample does not include the first enterprise tag and its corresponding data at the first time; input the first sample and the second sample into the target prediction model respectively to obtain a first prediction result and a second prediction result, and obtain a first marginal contribution value of the first enterprise tag based on the first prediction result and the second prediction result; determine a second marginal contribution value of the first enterprise tag at a second time, and obtain a first global contribution value based on the first marginal contribution value and the second marginal contribution value.

[0113] In some embodiments, the processor 301 is further configured to acquire data corresponding to each of the plurality of enterprise tags at a third time; construct a third sample and a fourth sample based on each enterprise tag and its corresponding data at the third time, wherein the third sample includes each enterprise tag and its corresponding data at the third time, and the fourth sample does not include the first enterprise tag, the second enterprise tag, and their respective corresponding data at the third time; input the third sample and the fourth sample into the target prediction model respectively to obtain a third prediction result and a fourth prediction result, and obtain the global contribution value of the first enterprise tag and the second enterprise tag based on the third prediction result and the fourth prediction result; and obtain the joint contribution value based on the first global contribution value, the second global contribution value, and the global contribution value of the first enterprise tag and the second enterprise tag.

[0114] In some implementations, the processor 301 is further configured to determine a single weight of the first enterprise label based on the first global contribution value, determine a joint weight of the first enterprise label based on the joint contribution value, and determine a weighted contribution value of the first enterprise label based on the single weight and the joint weight.

[0115] In some embodiments, the processor 301 is further configured to preprocess the enterprise public data and the enterprise private data respectively; obtain an enterprise dataset based on the preprocessed enterprise public data and the enterprise private data; extract enterprise entity and enterprise tag information from the enterprise dataset, and construct a Euclidean space model of the two based on the enterprise entities and the enterprise tag information; and construct the enterprise knowledge graph based on the Euclidean space model.

[0116] In some embodiments, the processor 301 is further configured to determine a plurality of enterprise tags, and based on each enterprise tag, obtain a plurality of corresponding enterprise data information from the preprocessed public enterprise data and the private enterprise data; perform a data mapping operation on each enterprise tag and its corresponding enterprise data information to obtain a plurality of fifth samples; determine the correlation between the enterprise tag in each fifth sample and its corresponding enterprise data information; and fuse a plurality of fifth samples with a correlation greater than a preset threshold to obtain the enterprise dataset.

[0117] In some embodiments, the processor 301 is further configured to construct an initial prediction model for the enterprise tag based on the enterprise's private data and the enterprise's public data; and to optimize the loss function of the initial prediction model using the enterprise knowledge graph to obtain a target prediction model for the enterprise tag.

[0118] In some embodiments, the processor 301 is further configured to perform data cleaning and transformation operations, variable filtering operations, and dataset partitioning operations on the enterprise public data and the enterprise private data respectively.

[0119] This application also provides a computer-readable storage medium storing a computer program, the computer program including program instructions, which, when executed, implement the enterprise tag interpretable prediction method provided in this application.

[0120] The computer-readable storage medium may be the internal storage unit of the computer device 300 described in the foregoing embodiments, such as the hard disk or memory of the computer device.

[0121] Please see Figure 6 , Figure 6 This is a schematic diagram of a computer-readable storage medium provided for this application. The storage medium 40 of this application stores a computer program capable of implementing all the aforementioned enterprise label interpretable prediction methods. This computer program can be stored in the storage medium 40 in the form of a software product, including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute all or part of the steps of the methods of various embodiments of this application. The aforementioned storage devices include various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks, or devices such as computers, servers, mobile phones, and tablets.

[0122] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A predictive method for enterprise tags that can be interpreted, characterized in that, The method includes: Acquire public and private enterprise data, and construct an enterprise knowledge graph based on the public and private enterprise data. The enterprise knowledge graph includes several enterprise tags and dynamic data corresponding to each enterprise tag. A target prediction model is trained based on the enterprise's private data, the enterprise's public data, and the enterprise's knowledge graph. The enterprise's knowledge graph is then input into the target prediction model to obtain the enterprise tag prediction result. Using the target prediction model, the first global contribution value of the first enterprise label and the second global contribution value of the second enterprise label are determined respectively, and a joint contribution value is obtained based on the first global contribution value and the second global contribution value. Based on the first global contribution value and the joint contribution value, the first weighted contribution value of the first enterprise label is obtained, and the second weighted contribution value of other enterprise labels among the plurality of enterprise labels is determined; The enterprise label prediction results are interpreted based on the first weighted contribution value and the second weighted contribution value; wherein... The step of determining the first global contribution value of the first enterprise label among the plurality of enterprise labels using the target prediction model includes: Obtain the data corresponding to each enterprise label in the plurality of enterprise labels at the first moment; Based on each enterprise tag and its corresponding data at the first time, a first sample and a second sample are constructed, wherein the first sample includes each enterprise tag and its corresponding data at the first time, and the second sample does not include the first enterprise tag and its corresponding data at the first time. The first sample and the second sample are respectively input into the target prediction model to obtain the first prediction result and the second prediction result, and the first marginal contribution value of the first enterprise label is obtained based on the first prediction result and the second prediction result. Determine the second marginal contribution value of the first enterprise tag at the second time point, and obtain the first global contribution value based on the first marginal contribution value and the second marginal contribution value; The step of obtaining a joint contribution value based on the first global contribution value and the second global contribution value includes: Obtain the data corresponding to each of the aforementioned enterprise tags at the third time point; Based on each enterprise tag and its corresponding data at the third time point, a third sample and a fourth sample are constructed. The third sample includes each enterprise tag and its corresponding data at the third time point, while the fourth sample does not include the first enterprise tag, the second enterprise tag, and their corresponding data at the third time point. The third and fourth samples are input into the target prediction model to obtain the third and fourth prediction results, and the global contribution values ​​of the first enterprise label and the second enterprise label are obtained based on the third and fourth prediction results. The joint contribution value is obtained based on the first global contribution value, the second global contribution value, and the global contribution values ​​of the first enterprise label and the second enterprise label.

2. The method according to claim 1, characterized in that, The first weighted contribution value of the first enterprise tag obtained based on the first global contribution value and the joint contribution value includes: The single weight of the first enterprise tag is determined based on the first global contribution value, and the joint weight of the first enterprise tag is determined based on the joint contribution value. The weighted contribution value of the first enterprise label is determined based on the single weight and the joint weight.

3. The method according to claim 1, characterized in that, The construction of the enterprise knowledge graph based on the enterprise's public data and private data includes: Preprocessing operations are performed on the enterprise's public data and the enterprise's private data respectively; The enterprise dataset is obtained by fusing the preprocessed public data and the private data of the enterprise. Extract enterprise entities and enterprise tag information from the enterprise dataset, and construct a Euclidean space model of the two based on the enterprise entities and the enterprise tag information; The enterprise knowledge graph is constructed based on the Euclidean space model.

4. The method according to claim 3, characterized in that, The preprocessed public enterprise data and the private enterprise data are merged to obtain an enterprise dataset, including: A number of enterprise tags are determined, and based on each enterprise tag, corresponding enterprise data information is obtained from the preprocessed public enterprise data and the private enterprise data. Perform a data mapping operation on each of the enterprise tags and its corresponding enterprise data information to obtain several fifth samples; Determine the correlation between the enterprise label in each of the fifth samples and the corresponding enterprise data information; The enterprise dataset is obtained by fusing multiple fifth samples with a relevance greater than a preset threshold.

5. The method according to claim 1, characterized in that, The target prediction model trained based on the enterprise's private data, the enterprise's public data, and the enterprise's knowledge graph includes: An initial prediction model for the enterprise's tags is constructed based on the enterprise's private data and public data. The loss function of the initial prediction model is optimized using the enterprise knowledge graph to obtain the target prediction model for the enterprise tag.

6. The method according to claim 3, characterized in that, The preprocessing operations performed on the enterprise's public data and private data respectively include: Data cleaning and transformation, variable filtering, and dataset partitioning operations are performed sequentially on the enterprise's public data and private data, respectively.

7. A computer device, characterized in that, The computer device includes: Memory and processor; The memory is connected to the processor and is used to store programs; The processor is configured to implement the steps of the enterprise tag interpretable prediction method as described in any one of claims 1-6 by running a program stored in the memory.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, causes the processor to implement the steps of the enterprise tag interpretable prediction method as described in any one of claims 1-6.

Images