CN115034529A - Training method, device and equipment of state recognition model and readable storage medium - Google Patents

Abstract

Embodiments of the present disclosure disclose a training method, apparatus, device, and readable storage medium for a state recognition model. The training method of the state recognition model includes: obtaining feature data related to a subject and state data related to the subject; determining a specific attribute value matching the feature data according to a preset mapping relationship; The state data related to the subject and the specific attribute value are used to determine the standardized state index of the subject; the state recognition model is trained according to the standardized state index, so that the state data can be converted into a standardized state index, and based on the standardized state index The trained state recognition model can more accurately identify the state of the subject.

Description

Training method, apparatus, device and readable storage medium for state recognition model

technical field

The present disclosure relates to the technical field of data processing, and in particular, to a training method, apparatus, electronic device and readable storage medium for a state recognition model.

Background technique

At present, when identifying a certain state of a subject, attention may be paid to the characteristic data of the subject itself and the state data related to the subject. However, the inventors found that due to the complex actual situation, the above content cannot well reflect the subject's state. The real state is prone to inaccurate state recognition results.

For example, when doing enterprise risk identification, we usually pay attention to the litigation risk of the enterprise. However, for larger enterprises, litigation cases are more likely to occur, but it does not mean that they are enterprises with high litigation risk. On the contrary, smaller enterprises may be involved in not many litigation cases, but compared with similar enterprises Said, may already represent that the business has a higher risk of litigation. Therefore, it is difficult for users to intuitively judge the litigation risk of the enterprise only from the number of defendants' records.

SUMMARY OF THE INVENTION

In order to solve the problems in the related art, the embodiments of the present disclosure provide a training method and apparatus for a state recognition model, a state recognition method and apparatus, equipment, and a medium.

In a first aspect, an embodiment of the present disclosure provides a method for training a state recognition model.

Specifically, the training method of the state recognition model includes:

obtain subject-related characteristic data, and state data related to said subject;

Determine a specific attribute value matching the feature data according to a preset mapping relationship;

determining a normalized state indicator for the subject based on the subject-related state data and the specific attribute value;

The state recognition model is trained according to the standardized state indicators.

In conjunction with the first aspect, in a first implementation manner of the first aspect of the present disclosure, the mapping relationship at least includes:

Correspondence between feature data and specific attribute values; wherein, feature data corresponding to different value ranges have different feature attribute values.

With reference to the first aspect, in a second implementation manner of the first aspect of the present disclosure, the determining of the standardized state index of the subject based on the state data related to the subject and the specific attribute value includes:

determining an original state indicator based on the state data and the specific attribute value;

According to the maximum value and the minimum value of the original state index of different subjects, the original state index is mapped to a predetermined value interval to obtain a standardized state index.

With reference to the first aspect, in a third implementation manner of the first aspect of the present disclosure, the feature data includes graph features determined based on the association relationship of the subject in the subject relationship graph.

With reference to the third implementation manner of the first aspect, in a fourth implementation manner of the first aspect of the present disclosure, the training of the state recognition model according to the standardized state index includes:

determining a first vector representation of characteristic data of the subject;

Converting the first vector representation from a high-dimensional sparse vector representation to a low-dimensional dense second vector representation;

The second vector representation is spliced with the third vector representation of the feature data of the subject in the subject relationship diagram adjacent to the subject to obtain the fourth vector representation of the subject's feature data;

encoding the fourth vector representation as a fifth vector representation of the feature data of the subject having a predetermined dimension as training data for the state recognition model;

Using the standardized state index as a label, the state recognition model is trained.

With reference to the fourth implementation manner of the first aspect, in a fifth implementation manner of the first aspect of the present disclosure, the first vector representation includes a one-hot feature representation, and the first vector representation is composed of high-dimensional sparse The vector representation of is transformed into a low-dimensional dense second vector representation consisting of:

converting the one-hot feature representation into a sixth vector representation through an embedding operation;

A weighted summation of the sixth vector representation yields the low-dimensional dense second vector representation.

With reference to the first aspect and any one of the first to fifth implementation manners of the first aspect, in a sixth implementation manner of the first aspect, the method further includes: graphically presenting the Standardized status indicators.

In a second aspect, an embodiment of the present disclosure provides a state identification method.

Specifically, the state identification method includes:

Obtain the characteristic data of the subject to be identified;

The feature data is input into the state identification model according to any one of the first aspect and the first to sixth aspects of the first aspect, so as to obtain a standardized state index of the subject to be identified.

In a third aspect, an embodiment of the present disclosure provides a training method for a risk identification model.

Specifically, the training method of the risk identification model includes:

Obtain characteristic data of the enterprise and the number of contract litigation cases in which the enterprise is a defendant, wherein the characteristic data includes turnover;

According to the mapping relationship between the turnover and the contract quantity, determine the contract quantity of the enterprise;

determining a standardized risk indicator for the enterprise based on the number of contract litigation cases and the number of contracts for the enterprise;

The training data is determined based on the characteristic data of the enterprise, and the risk identification model is trained with the standardized risk index as a label.

In a fourth aspect, an embodiment of the present disclosure provides an apparatus for training a state recognition model.

Specifically, the training device for the state recognition model includes:

a first acquiring module, configured to acquire feature data related to the subject and state data related to the subject;

a first determining module, configured to determine a specific attribute value matching the feature data according to a preset mapping relationship;

a second determination module configured to determine a standardized state index of the subject based on the subject-related state data and the specific attribute value;

A training module configured to train the state recognition model according to the standardized state index.

With reference to the fourth aspect, the present disclosure is in a first implementation manner of the fourth aspect, wherein the mapping relationship at least includes:

Correspondence between feature data and specific attribute values, wherein the feature attribute values corresponding to feature data with different value ranges are different.

With reference to the fourth aspect, in a second implementation manner of the fourth aspect of the present disclosure, the determining of the standardized state index of the subject based on the state data related to the subject and the specific attribute value includes:

determining an original state indicator based on the state data and the specific attribute value;

According to the maximum and minimum values of the original state indexes of different subjects, the original state indexes are mapped to a predetermined value interval to obtain a standardized state index.

With reference to the fourth aspect, in a third implementation manner of the fourth aspect of the present disclosure, the feature data includes graph features determined based on the association relationship of the subject in the subject relationship graph.

With reference to the third implementation manner of the fourth aspect, in a fourth implementation manner of the fourth aspect of the present disclosure, the training of the state recognition model according to the standardized state index includes:

determining a first vector representation of characteristic data of the subject;

Converting the first vector representation from a high-dimensional sparse vector representation to a low-dimensional dense second vector representation;

The second vector representation is spliced with the third vector representation of the feature data of the subject in the subject relationship diagram adjacent to the subject to obtain the fourth vector representation of the subject's feature data;

encoding the fourth vector representation as a fifth vector representation of the feature data of the subject having a predetermined dimension as training data for the state recognition model;

Using the standardized state index as a label, the state recognition model is trained.

With reference to the fourth implementation manner of the fourth aspect, in a fifth implementation manner of the fourth aspect of the present disclosure, the first vector representation includes a one-hot feature representation, and the first vector representation is composed of high-dimensional sparse The vector representation of is transformed into a low-dimensional dense second vector representation consisting of:

converting the one-hot feature representation into a sixth vector representation through an embedding operation;

A weighted summation of the sixth vector representation yields the low-dimensional dense second vector representation.

With reference to the fourth aspect and any one of the first to fifth implementation manners of the fourth aspect, the present disclosure is in a sixth implementation manner of the fourth aspect, and the device further includes:

A presentation module configured to graphically present the standardized status indicators.

In a fifth aspect, an embodiment of the present disclosure provides a state identification device.

Specifically, the state identification device includes:

a second acquisition module, configured to acquire characteristic data of the subject to be identified;

The identification module is configured to input the feature data into the state identification model according to any one of the first aspect and the first to sixth implementation manners of the first aspect, so as to obtain the identification of the subject to be identified. Standardized status indicators.

In a sixth aspect, embodiments of the present disclosure provide an electronic device, including a memory and a processor, wherein the memory is used to store one or more computer instructions, wherein the one or more computer instructions are processed by the The device executes to implement the method as described in any one of the first aspect, the first to sixth implementations of the first aspect, the second aspect or the third aspect.

In a seventh aspect, an embodiment of the present disclosure provides a computer-readable storage medium on which computer instructions are stored, and when the computer instructions are executed by a processor, implement the first aspect and the first to sixth aspects of the first aspect The method of any one of the implementations, the second aspect, or the third aspect.

According to the technical solutions provided by the embodiments of the present disclosure, the characteristic data related to the subject and the state data related to the subject are obtained; the specific attribute value matching the characteristic data is determined according to the preset mapping relationship; The state data related to the subject and the specific attribute value are determined, and the standardized state index of the subject is determined; the state recognition model is trained according to the standardized state index, so that the state data can be converted into a standardized state index, based on the standardized state index. The state recognition model trained by the state index can more accurately identify the state of the subject.

According to the technical solutions provided by the embodiments of the present disclosure, the mapping relationship includes at least the corresponding relationship between the feature data and the specific attribute value, wherein the feature attribute values corresponding to the feature data with different value ranges are different, so that the mapping relationship can be based on the mapping relationship. The number of specific events is converted into a standardized state index, and the state recognition model trained based on the standardized state index can more accurately identify the state of the subject.

According to the technical solutions provided by the embodiments of the present disclosure, determining the standardized state index of the subject based on the state data related to the subject and the specific attribute value includes: based on the state data and the specific attribute value, determining The original state index; according to the maximum and minimum values of the original state index of different subjects, the original state index is mapped to a predetermined value interval to obtain a standardized state index, so that the obtained standardized state index is more effective, and training based on the standardized state index The state recognition model can more accurately identify the state of the subject.

According to the technical solutions provided by the embodiments of the present disclosure, the feature data includes graph features determined based on the association relationship of the subject in the subject relationship graph, so that the state of the subject can be better identified in combination with the graph features.

According to the technical solutions provided by the embodiments of the present disclosure, the first vector representation of the feature data of the subject is determined; the first vector representation is converted from a high-dimensional sparse vector representation to a low-dimensional dense second vector representation; The second vector representation is spliced with the third vector representation of the feature data of the subject's adjacent subjects in the subject relationship diagram to obtain a fourth vector representation of the subject's feature data; the fourth vector representation is Representing the fifth vector representation encoded as the feature data of the subject with a predetermined dimension, as the training data of the state recognition model; using the standardized state index as a label, training the state recognition model, so as to be able to better Identify the state of the subject using graph features.

According to the technical solutions provided by the embodiments of the present disclosure, the one-hot feature representation is converted into a sixth vector representation through an embedding operation; the weighted summation of the sixth vector representation obtains the low-dimensional dense second vector representation, Thus, the problem of excessive resource occupation caused by sparse features is solved.

According to the technical solutions provided by the embodiments of the present disclosure, by presenting the standardized state indicators in a graphical manner, the state of the subject can be displayed more intuitively.

According to the technical solutions provided by the embodiments of the present disclosure, the standardized state of the subject to be identified is obtained by acquiring the characteristic data of the subject to be identified; and inputting the characteristic data into the state identification model trained by the above-mentioned training method of the state identification model indicators, so that the state of the subject can be more accurately identified by standardized state indicators.

According to the technical solutions provided by the embodiments of the present disclosure, the characteristic data of an enterprise and the number of contract litigation cases in which the enterprise is a defendant are obtained, wherein the characteristic data includes turnover; according to the relationship between the turnover and the number of contracts Mapping relationship, determine the number of contracts of the enterprise; determine the standardized risk index of the enterprise based on the number of contract litigation cases and the number of contracts of the enterprise; determine the training data based on the characteristic data of the enterprise, and use the The standardized risk index is used as a label, and the risk identification model is trained, so that the number of contract litigation cases can be converted into standardized risk index, and the risk identification model trained based on the standardized risk index can more accurately identify the risk of the enterprise.

According to the technical solutions provided by the embodiments of the present disclosure, the first obtaining module is configured to obtain the characteristic data related to the subject and the state data related to the subject; the first determining module is configured to obtain according to the preset mapping relationship , determine a specific attribute value that matches the feature data; a second determining module is configured to determine a standardized state index of the subject based on the subject-related state data and the specific attribute value; a training module, The state recognition model is configured to train the state identification model according to the standardized state index, so that the number of specific events can be converted into a standardized state index, and the state identification model trained based on the standardized state index can more accurately identify the state of the subject.

According to the technical solutions provided by the embodiments of the present disclosure, the mapping relationship includes at least the corresponding relationship between the feature data and the specific attribute value, wherein the feature attribute values corresponding to the feature data with different value ranges are different, so that the feature data can be based on the corresponding relationship. The mapping relationship converts the number of specific events into standardized state indicators, and the state recognition model trained based on the standardized state indicators can more accurately identify the state of the subject.

According to the technical solutions provided by the embodiments of the present disclosure, the original state index is determined based on the state data and the specific attribute value; the original state index is mapped to a predetermined value according to the maximum value and the minimum value of the original state index of different subjects According to the value interval, a standardized state index is obtained, so that the obtained standardized state index is more effective, and the state recognition model trained based on the standardized state index can more accurately identify the state of the subject.

According to the technical solutions provided by the embodiments of the present disclosure, the state of the subject can be better identified in combination with the graph features by using the graph features determined based on the association relationship of the subjects in the subject relationship graph as feature data.

According to the technical solutions provided by the embodiments of the present disclosure, the first vector representation of the feature data of the subject is determined; the first vector representation is converted from a high-dimensional sparse vector representation to a low-dimensional dense second vector representation; The second vector representation is spliced with the third vector representation of the feature data of the subject's adjacent subjects in the subject relationship diagram to obtain a fourth vector representation of the subject's feature data; the fourth vector representation is Representing the fifth vector representation encoded as the feature data of the subject with a predetermined dimension, as the training data of the state recognition model; using the standardized state index as a label, training the state recognition model, so as to be able to better Identify the state of the subject using graph features.

According to the technical solutions provided by the embodiments of the present disclosure, the first vector representation represented by the one-hot feature is converted into a sixth vector representation through an embedding operation; and the low-dimensional dense representation is obtained by weighted summation of the sixth vector representation. The second vector representation solves the problem of excessive resource occupation caused by sparse features.

According to the technical solutions provided by the embodiments of the present disclosure, the display module is configured to graphically present the standardized state indicator, so that the state of the subject can be displayed more intuitively.

According to the technical solutions provided by the embodiments of the present disclosure, the second acquisition module is configured to acquire feature data of the subject to be recognized; the recognition module is configured to input the feature data into the training of the state recognition model as described above. The state recognition model trained by the method is used to obtain the standardized state index of the subject to be identified, so that the state of the subject can be more accurately identified through the standardized state index.

It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure.

Description of drawings

Other features, objects and advantages of the present disclosure will become more apparent from the following detailed description of non-limiting embodiments, taken in conjunction with the accompanying drawings. In the attached image:

1 shows a flowchart of a training method for a state recognition model according to an embodiment of the present disclosure;

2 shows a flowchart of determining a mapping relationship between feature data and specific attribute values according to an embodiment of the present disclosure;

3 shows a schematic diagram of a piecewise function according to an embodiment of the present disclosure;

4 shows a flowchart of determining a standardized state indicator according to an embodiment of the present disclosure;

FIG. 5 shows a schematic diagram of predicting a state based on a graph feature according to an embodiment of the present disclosure;

6 shows a schematic diagram of a subject relationship diagram according to an embodiment of the present disclosure;

7 shows a flowchart of training the state recognition model according to the standardized state index according to an embodiment of the present disclosure;

8 shows a schematic diagram of feature data processing according to an embodiment of the present disclosure;

FIG. 9 shows a flowchart of determining a second vector representation according to an embodiment of the present disclosure;

10 shows a schematic diagram of graphically presenting standardized status indicators according to an embodiment of the present disclosure;

11 shows a flowchart of a state identification method according to an embodiment of the present disclosure;

12 shows a flowchart of an enterprise risk identification method according to an embodiment of the present disclosure;

13 shows a block diagram of a training apparatus for a state recognition model according to an embodiment of the present disclosure;

14 shows a block diagram of a state identification device according to an embodiment of the present disclosure;

15 shows a block diagram of an electronic device according to an embodiment of the present disclosure;

FIG. 16 shows a schematic structural diagram of a computer system suitable for implementing the methods and apparatuses of embodiments of the present disclosure.

Detailed ways

Hereinafter, exemplary embodiments of the present disclosure will be described in detail with reference to the accompanying drawings so that those skilled in the art can easily implement them. Also, for the sake of clarity, parts unrelated to describing the exemplary embodiments are omitted from the drawings.

In the present disclosure, it should be understood that terms such as "comprising" or "having" are intended to indicate the presence of features, numbers, steps, acts, components, parts, or combinations thereof disclosed in this specification, and are not intended to exclude a or multiple other features, numbers, steps, acts, components, parts, or combinations thereof may exist or be added.

In addition, it should be noted that the embodiments of the present disclosure and the features of the embodiments may be combined with each other under the condition of no conflict. The present disclosure will be described in detail below with reference to the accompanying drawings and in conjunction with embodiments.

As mentioned above, the characteristic data of the subject itself and the events related to the subject cannot reflect the real state of the subject well, and the problem of inaccurate state recognition results is prone to occur.

For example, when doing enterprise risk identification, we usually pay attention to the litigation risk of the enterprise. However, for larger enterprises, lawsuits are more likely to occur, but this does not mean that larger enterprises are necessarily high litigation risk enterprises. On the contrary, smaller enterprises may not be involved in many lawsuits. However, compared with similar companies, it may already represent that the company has a higher risk of litigation. Therefore, it is difficult for users to intuitively judge the litigation risk of the enterprise only from the number of defendants' records.

According to the technical solutions provided by the embodiments of the present disclosure, the characteristic data related to the subject and the state data related to the subject are obtained; the specific attribute value matching the characteristic data is determined according to the preset mapping relationship; The state data related to the subject and the specific attribute value are determined, and the standardized state index of the subject is determined; the state recognition model is trained according to the standardized state index, so that the state data can be converted into a standardized state index, based on the standardized state index. The state recognition model trained by the state index can more accurately identify the state of the subject.

FIG. 1 shows a flowchart of a state recognition model training method according to an embodiment of the present disclosure.

As shown in FIG. 1 , the state recognition model training method includes operations S110 to S140.

In operation S110, characteristic data related to the subject and state data related to the subject are obtained;

In operation S120, according to a preset mapping relationship, a specific attribute value matching the feature data is determined;

In operation S130, based on the subject-related state data and the specific attribute value, determining a standardized state index of the subject;

In operation S140, the state recognition model is trained according to the standardized state index.

According to an embodiment of the present disclosure, taking the identification of the defendant risk of an enterprise contract as an example, the subject may be an enterprise, the characteristic data may be various enterprise information including information such as turnover, and the status data related to the subject may be, for example, related to The number of specific events related to the enterprise, for example, the specific event can be a lawsuit, a contract lawsuit, or a lawsuit or contract lawsuit in which the company is the defendant. That is, operation S110 may be to acquire characteristic data of the enterprise and the number of contract litigation cases as the defendant.

According to the embodiment of the present disclosure, the specific attribute value is information that is not easy to obtain. For example, in the embodiment of the present disclosure, the specific attribute value may be the number of contracts signed by the enterprise each year. The preset mapping relationship may be a mapping relationship between the feature data and specific attribute values established in advance, so that after acquiring the feature data of the subject, the specific attribute value of the subject can be determined according to the mapping relationship. For example, the inventor found that the number of contracts signed by an enterprise each year has a certain positive correlation with the annual turnover of the enterprise, and a mapping relationship between the annual turnover and the number of contracts signed by the enterprise can be established based on the available data Therefore, when the annual turnover data in the characteristic data is obtained, the estimated value of the number of contracts signed by the enterprise each year can be determined according to the mapping relationship. In the embodiment of the present disclosure, a function is fitted by using the relationship between the average number of contracts signed annually by enterprises of different scales and the scale factor (such as average turnover) of the enterprise. Converted to normalized contract litigation risk for subsequent judgment and modeling.

According to the embodiments of the present disclosure, by comprehensively considering state data and specific attribute values, a standardized state index of the subject can be determined, which is used to better evaluate the state of the subject. For example, since enterprises of different sizes are involved in different litigation cases, large-scale enterprises are more likely to be involved in litigation cases than small and micro enterprises. Standardized status indicators can be determined based on the ratio of contract litigation cases in which the enterprise is the defendant to the contracts signed each year.

According to the embodiment of the present disclosure, based on the state recognition model, all the features of the subject in the previous period can be used to predict the state of the subject in the next period. For example, all the characteristics of each enterprise in the previous period can be used to predict the enterprise Contract litigation risk over time. These characteristics are characteristics that may be related to the risk of contract defendants, such as the industry, region, size characteristics, financial status and so on of the enterprise. In this way, not only can we grasp the company's past contract litigation risks, but also predict the future contract litigation risks that are more likely to be closer to the real ones.

For example, the contract litigation risk of each enterprise in 2019 can be counted according to the records in the obtained judgment documents in 2019 as a label, and various historical characteristics of the enterprise in 2018 can be used, such as the number of defendants in 2018, the number of administrative penalties in 2018, etc. wait to predict the label. Of course, some fundamental information of the company, such as registered capital, usually does not change. You can choose the data of 2018 or 2019. In this way, using the trained model, on the test set, when a new enterprise needs to be analyzed, the standardized contract litigation risk of the current enterprise can be predicted by using the historical characteristics of the enterprise.

According to the technical solutions provided by the embodiments of the present disclosure, the characteristic data related to the subject and the state data related to the subject are obtained; the specific attribute value matching the characteristic data is determined according to the preset mapping relationship; The state data related to the subject and the specific attribute value are determined, and the standardized state index of the subject is determined; the state recognition model is trained according to the standardized state index, so that the state data can be converted into a standardized state index, based on the standardized state index. The state recognition model trained by the state index can more accurately identify the state of the subject.

According to an embodiment of the present disclosure, the mapping relationship includes at least a corresponding relationship between feature data and feature attribute values, wherein feature data corresponding to different value ranges have different feature attribute values.

FIG. 2 shows a flowchart of determining a mapping relationship between feature data and specific attribute values according to an embodiment of the present disclosure.

As shown in FIG. 2, the method may further include operations S210-S230.

In operation S210, the subject is divided into a plurality of categories based on the feature data, and each category corresponds to a different value interval of the feature data;

In operation S220, an average value of specific attribute values of subjects in each category is determined;

In operation S230, based on the value interval of the feature data and the average value of the specific attribute value, a piecewise function is constructed as a mapping relationship between the feature data and the specific attribute value.

Take corporate contract litigation risk as an example. Cases related to enterprise contracts can be extracted from the judgment documents first, and from this batch of cases, the records that the defendant is an enterprise can be extracted. After the extraction is completed, the number of contract-related defendants of each enterprise is counted separately.

As mentioned above, the number of defendants alone does not directly reflect the contract defendant risk of an enterprise, and it needs to be standardized according to certain rules. The standardized state index can be defined as, for example, the average risk of each contract signed with the current enterprise R _contract =C/N _contract , where C is the historical number of defendants of the current enterprise, and N _contract is the number of contracts signed by the enterprise in history. The relationship between the annual average number of contracts signed by domestic enterprises of different scales and the average turnover of the enterprise is obtained through the sample survey as shown in the following table.

Table 1

On each enterprise-level interval, assuming linear growth in individual intervals, we can construct a linear function based on the upper and lower turnover bounds of the current interval and the average number of contracts in that interval, As an estimate of the actual situation, as shown in Figure 3. Therefore, for the subject to be identified, the specific attribute value of the subject can be estimated by using the feature data through the piecewise function as shown in Figure 3, so as to obtain the subject's label, such as the enterprise's contract defendant risk value R _contract .

According to the technical solutions provided by the embodiments of the present disclosure, the mapping relationship includes at least the corresponding relationship between the feature data and the specific attribute value; wherein, the feature attribute values corresponding to the feature data in different value ranges are different, so that the mapping relationship can be based on the mapping relationship. The number of specific events is converted into a standardized state index, and the state recognition model trained based on the standardized state index can more accurately identify the state of the subject.

FIG. 4 shows a flow diagram of determining a normalized state indicator according to an embodiment of the present disclosure.

As shown in FIG. 4 , the above operation S130, that is, determining the normalized state index of the subject based on the state data and the specific attribute value may include operations S410 and S420.

In operation S410, based on the state data and the specific attribute value, determine an original state indicator;

In operation S420, according to the maximum value and the minimum value of the original state index of different subjects, the original state index is mapped to a predetermined value interval to obtain a standardized state index.

According to the embodiment of the present disclosure, the R _contracts of all enterprises can also be used to perform scaling normalization of the maximum value and the minimum value, so that the overall score range is within a predetermined interval, for example, 0 to 10 points, and the R′ _contract obtained after normalization is used as the standardization Status indicator.

According to the technical solutions provided by the embodiments of the present disclosure, the original state index is determined based on the quantitative state data of the specific event and the specific attribute value of the subject; the maximum and minimum values of the original state index of different subjects are based on different subjects. , the original state index is mapped to a predetermined value interval to obtain a standardized state index, so that the obtained standardized state index is more effective, and the state recognition model trained based on the standardized state index can more accurately identify the state of the subject.

The current related art is more concerned with the influence of the current subject's historical information on the state of the subject. The inventors found that, in actual situations, the information of the subject that has an associated relationship with the current subject is also very useful for predicting the state of the current subject. Great help. Various data and records of such associated nodes are not well utilized in previous schemes.

According to an embodiment of the present disclosure, the feature data includes graph features determined based on the association relationship of the subject in the subject relationship graph.

For example, still taking the prediction of the defendant risk of a corporate contract as an example, the associated relationship may include, for example, an investment relationship, and may also include an equity structure, a partnership relationship, a competitive relationship, and other potential relationships. The relationship between these enterprises will also be of great help in predicting the risk of contractual defendants. There may be accidental factors in the litigation records of a single enterprise, and if a more comprehensive contract litigation prediction model is trained with full data (especially combining the attributes of neighbor nodes and the graph features of records), the results will be more accurate.

FIG. 5 shows a schematic diagram of predicting a state based on a graph feature according to an embodiment of the present disclosure.

As shown in FIG. 5 , the left and right images are an exemplary schematic diagram of the subject relationship diagram. Among them, the nodes marked with 1 represent the samples whose subject state is the first state, such as the enterprise whose contract defendant risk label is high risk, and the node marked 0 represents the samples whose subject state is the second state, such as the contract defendant risk label is low risky business. Unlabeled nodes represent subjects to be identified. The states of other nodes in the surrounding environment are also helpful for judging the state of the current subject, and the state of the subject to be identified can be predicted based on the attributes of the subject itself and the states of surrounding nodes.

For example, if a company invests in subsidiaries that are frequently accused, the company's risk of being accused may also increase. The association relationship here is an example of an investment relationship. If there are 2 or more nodes around a node in the subject relationship graph with high risk of contract defendant records, then the company's defendant risk increases. Then it can be assumed that a feature calculated is the feature of the current company "the number of records of contract defendants in neighbor nodes". In other words, only when the number of contract defendants of a node exceeds 5 times, it is considered a high-risk node, then such a feature is "the number of contract defendant records in neighbor nodes exceeds 5 times". Even each neighbor node can have its own attributes, and it is also possible to set such a feature as "there is a contract defendant record in the neighbor node and it is a private company". There can be many graph features like this, and the model can automatically learn the features related to these graphs, thereby improving the accuracy of state recognition.

According to the technical solutions provided by the embodiments of the present disclosure, the state of the subject can be better identified in combination with the graph features by using the graph features determined based on the association relationship of the subjects in the subject relationship graph as feature data.

According to an embodiment of the present disclosure, the utilization of graph features may, for example, adopt the principle of GraphSage. The basic idea of GraphSage is to learn how the information of a node is aggregated through the features of its neighbor nodes. Compared with the shortcomings of graph convolutional neural network (GCN) and other methods, the graph features of each node are fixed. That is to say, even if it is an old node, if some new connection relationships are established, its corresponding graph features will also change, and it is also easy to learn. The principle of GraphSage is briefly described below with reference to FIG. 6 .

FIG. 6 shows a schematic diagram of a subject relationship diagram according to an embodiment of the present disclosure.

分别表示节点A、B、C、D的初始属性。在第一轮的时候，对于B节点来说，先将邻居节点A和C的属性

和

加权聚合之后得到的向量与B当前的属性向量

进行拼接，得到

同理可以得到

然后，在第二轮的时候，将

进行加权聚合，得到A节点的邻居节点汇聚向量

再将其与当前A节点本身向量

拼接得到节点A的新向量

以此类推可以得到任意阶数的特征向量，具体阶数的选择可以根据实际需要进行设定。通过Graphsage可以较好地利用邻居节点的属性来预测当前节点的状态。As shown in Figure 6, A's neighbors are B, C, and D. Take one of the neighbors B as an example, the node also has two neighbors A and C. by

Represent the initial properties of nodes A, B, C, and D, respectively. In the first round, for node B, the attributes of neighbor nodes A and C are first

and

The vector obtained after weighted aggregation and the current attribute vector of B

splicing to get

The same can be obtained

Then, in the second round, the

Perform weighted aggregation to get the aggregation vector of neighbor nodes of node A

Then compare it with the current A node itself vector

Splicing to get the new vector of node A

By analogy, eigenvectors of any order can be obtained, and the selection of the specific order can be set according to actual needs. Through Graphsage, the properties of neighbor nodes can be better used to predict the state of the current node.

However, the inventors found that when the traditional Graphsage encounters many enumerated features, the model effect will be deteriorated due to the sparsity of the one-hot (one-hot) variables of the enumerated features. For example, if all enterprises involve 1000 industries, the one-hot vector of industry attributes will have 1000 dimensions, and the features will be very sparse. If there are multiple similar enumerated attribute features, the attribute needs to be one-hot processed and then spliced to other attribute vectors, which will be more sparse. Also, many nodes on the graph may not have attributes. In this way, ordinary Graphsage cannot deal with the sparsity of this attribute feature vector.

The embodiment of the present disclosure proposes an improved Graphsage method based on enumeration attribute embedding compression. By learning the average pooling of attribute embedding of each node, information compression of multiple enumeration attributes is realized, which helps the model to learn better feature.

FIG. 7 shows a flowchart of training the state recognition model according to the standardized state index according to an embodiment of the present disclosure.

As shown in FIG. 7 , the method includes operations S710˜S750.

In operation S710, determining a first vector representation of the characteristic data of the subject;

In operation S720, the first vector representation is converted from a high-dimensional sparse vector representation to a low-dimensional dense second vector representation;

In operation S730, the second vector representation is spliced with the third vector representation of the feature data of the subject's adjacent subjects in the subject relationship diagram to obtain a fourth vector representation of the subject's feature data;

In operation S740, encoding the fourth vector representation as a fifth vector representation of the feature data of the subject having a predetermined dimension as training data for the state recognition model;

In operation S750, the state recognition model is trained using the standardized state index as a label.

通过将该节点的特征向量与相邻节点的第三特征向量拼接得到第四特征向量，并重新编码为预定维度的第五特征向量，可以在原有的属性特征中融入图特征。According to an embodiment of the present disclosure, the first vector representation may be, for example, various enumerated one-hot feature representations. By converting these high-dimensional and sparse vector representations into low-dimensional and dense second vector representations, the solution is at least partially solved. The problem of vector sparsity. Through the idea of Graphsage, use the transformed second vector to represent the initial attribute of the replacement node X

By splicing the feature vector of the node with the third feature vector of the adjacent node to obtain a fourth feature vector, and re-encoding it into a fifth feature vector of a predetermined dimension, the graph features can be integrated into the original attribute features.

Among them, the node embedding information can contain two kinds of information at the same time: the aggregation of the attribute information of the neighbor node to the current node, and the information of the location context between the nodes. The first kind of information can be learned through supervised loss, that is, the state label of the current node, the attributes of its neighbor nodes are aggregated into the current node as the feature of the current node, and the label is used to reversely optimize the embedding of the neighbor node. The role of this information can be understood as if a certain feature (such as trustworthiness) has a greater effect on the state, then the state of the current node can be better judged by aggregating the feature in the neighbor nodes. The second kind of information can be optimized by unsupervised loss. The main idea is to make the embedding distance of adjacent nodes closer. Specifically, a method similar to negative sampling in node2vec can be used. The role of this information can be understood that if the number of state labels in the neighbor nodes is the first state, the probability that the state of the current node is the first state is also high.

The method of FIG. 7 will be described below with reference to the embodiment shown in FIG. 8 .

FIG. 8 shows a schematic diagram of feature data processing according to an embodiment of the present disclosure.

As shown in Figure 8, for nodes A, B, and C, they are initially represented by one-hot feature vectors with different attributes, that is, nodes A, B, and C are represented by their respective first vectors. Still taking the enterprise as an example, the enterprise itself can have a Nodeone-hot, and then each attribute i of the enterprise has a Feat_{i}one-hot, and these one-hot vectors are converted into dense vectors of predetermined dimensions, such as embedding vectors, After the weighted summation (or average) of the attribute's Feat_{i}embedding and the enterprise's own Node embedding, the final embedding vector of the enterprise is obtained, that is, the second vector representation. Among them, if an attribute of an enterprise is empty, the attribute may not be considered when calculating the second feature vector of the enterprise. Nodes A, B, C can compute their respective second vector representations, respectively, as shown in FIG. 8 . By splicing and encoding the second vector representations of node B and node C, the third vector representation of the neighbor nodes of node A can be obtained, and then spliced with the second vector representation of node A to obtain the fourth vector representation of node A , by encoding to a predetermined dimension, a fifth vector representation is obtained, which is the vector representation of node A's graph features fused with first-order neighbor nodes. Through the above method, a vector representation of each node's graph features fused with neighbor nodes of a given order can be obtained. As the training data of the state recognition model, the state recognition model can be trained by using the standardized state index as a label.

According to the technical solutions provided by the embodiments of the present disclosure, the first vector representation of the feature data of the subject is determined; the first vector representation is converted from a high-dimensional sparse vector representation to a low-dimensional dense second vector representation; The second vector representation is spliced with the third vector representation of the feature data of the subject's adjacent subjects in the subject relationship diagram to obtain a fourth vector representation of the subject's feature data; the fourth vector representation is Representing the fifth vector representation encoded as the feature data of the subject with a predetermined dimension, as the training data of the state recognition model; using the standardized state index as a label, training the state recognition model, so as to be able to better Identify the state of the subject using graph features.

FIG. 9 shows a flowchart of determining a second vector representation according to an embodiment of the present disclosure.

According to an embodiment of the present disclosure, the first vector representation includes a one-hot feature representation. As shown in FIG. 9 , the operation S720 is to convert the first vector representation from a high-dimensional sparse vector representation to a low-dimensional dense second representation The vector representation may include operations S910 and S920.

In operation S910, converting the one-hot feature representation into a sixth vector representation through an embedding operation;

In operation S920, the sixth vector representation is weighted and summed to obtain the low-dimensional dense second vector representation.

According to an embodiment of the present disclosure, the one-hot feature representation is a one-hot feature representation. For example, for a subject with 50 non-null attributes, the 50 attributes can be represented as 50 one-hot feature representations respectively, but this representation is too sparse, which is not conducive to subsequent processing. The method of the embodiment of the present disclosure can convert the 50 one-hot vectors into 50 low-dimensional dense vector representations through an embedding operation (embedding), that is, the so-called embedding vector representation, and then aggregate the 50 embedding vectors, For example, weights can be set for 50 attributes respectively, and a vector representation of the subject, that is, a second vector representation, can be obtained by means of weighted summation.

According to the embodiment of the present disclosure, if the attribute is not an enumeration type one-hot but a continuous numerical type, two solutions can be used for processing. The type of reference; the other is to normalize the value and splicing it onto the Mixture embedding as one of the bits.

According to the technical solutions provided by the embodiments of the present disclosure, the one-hot feature representation is converted into a sixth vector representation through an embedding operation; the weighted summation of the sixth vector representation obtains the low-dimensional dense second vector representation, Thus, the problem of excessive resource occupation caused by sparse features is solved.

According to an embodiment of the present disclosure, the method may further include graphically presenting the normalized state indicator. For example, FIG. 10 shows a schematic diagram of presenting standardized status indicators in a graphical manner according to an embodiment of the present disclosure. As shown in FIG. 10 , the standardized status indicators can be presented on a user interface in a dashboard style, for example, so that the user can be intuitively prompted A normalized state indicator for the current subject.

According to the technical solutions provided by the embodiments of the present disclosure, by presenting the standardized state indicators in a graphical manner, the state of the subject can be displayed more intuitively.

FIG. 11 shows a flowchart of a state identification method according to an embodiment of the present disclosure.

As shown in FIG. 11 , the method includes operations S1110 and S1120.

In operation S1110, acquiring characteristic data of the subject to be identified;

In operation S1120, the feature data is input into a state recognition model trained by any one of the state recognition model training methods described in Figs. 1 to 10 to obtain a standardized state index of the subject to be recognized.

According to the technical solutions provided by the embodiments of the present disclosure, by acquiring the characteristic data of the subject to be identified; and inputting the characteristic data into the state identification model trained by the above state identification model training method, to obtain the standardized state index of the subject to be identified , so that the state of the subject can be more accurately identified through the standardized state index.

FIG. 12 shows a flowchart of an enterprise risk identification method according to an embodiment of the present disclosure.

As shown in FIG. 12 , the method includes operations S1210˜S1240.

In operation S1210, characteristic data of the enterprise and the number of contract litigation cases in which the enterprise is a defendant are acquired, wherein the characteristic data includes turnover;

In operation S1220, the contract quantity of the enterprise is determined according to the mapping relationship between the turnover and the contract quantity;

In operation S1230, a standardized risk index of the enterprise is determined based on the number of contract litigation cases and the number of contracts of the enterprise;

In operation S1240, training data is determined based on the characteristic data of the enterprise, and a risk identification model is trained with the standardized risk index as a label.

According to the technical solutions provided by the embodiments of the present disclosure, the characteristic data of an enterprise and the number of contract litigation cases in which the enterprise is a defendant are obtained, wherein the characteristic data includes turnover; according to the relationship between the turnover and the number of contracts Mapping relationship, determine the number of contracts of the enterprise; determine the standardized risk index of the enterprise based on the number of contract litigation cases and the number of contracts of the enterprise; determine the training data based on the characteristic data of the enterprise, and use the The standardized risk index is used as a label, and the risk identification model is trained, so that the number of contract litigation cases can be converted into standardized risk index, and the risk identification model trained based on the standardized risk index can more accurately identify the risk of the enterprise.

For example, before an enterprise intends to sign a contract with another subject, the corporate legal affairs of the enterprise may apply the method of the embodiments of the present disclosure to assess the risk status of the other subject as a basis for decision-making.

Those skilled in the art can understand that the above solutions in the embodiments of the present disclosure can not only be used to determine the risks of the enterprise to be identified, but also can be used to identify the status or risks of organizations and personnel inside or outside the enterprise, and can also be used to identify the natural environment. The biological or non-biological status or risk in the data can also be used to identify the status or risk of the subject to be identified in the form of data.

For example, the subject of this embodiment of the present disclosure may be a merchant in an e-commerce platform, and a standardized state indicator of the merchant is judged according to the method of this embodiment of the present disclosure, and based on the standardized state indicator, it is used to manage the merchant, or to process the information for the merchant. complaints, etc. For another example, a lawyer can determine a standardized state index for a certain subject by using the method of the embodiment of the present disclosure, which can be used for the point of view of a mooting court as an evaluation of the subject. For another example, the procuratorate can apply the method of the real-time example of the present disclosure to make predictions for forming a plan. In addition, in the handling of various other civil, administrative, or civil and administrative disputes, the people's courts, arbitration committees, governments, and government departments can all apply the standardized state indicators generated by the methods provided in the embodiments of the present disclosure. Reference. The method of an embodiment of the present disclosure can also be used to check whether the focus of the dispute has changed.

FIG. 13 shows a block diagram of a state recognition model training apparatus according to an embodiment of the present disclosure. Wherein, the apparatus may be realized by software, hardware or a combination of the two to become part or all of the electronic device.

As shown in FIG. 13 , the state recognition model training apparatus 1300 includes a first acquisition module 1310 , a first determination module 1320 , a second determination module 1330 and a training module 1340 .

a first obtaining module 1310, configured to obtain feature data related to the subject and state data related to the subject;

The first determining module 1320 is configured to determine a specific attribute value matching the feature data according to a preset mapping relationship;

The second determination module 1330 is configured to determine the standardized state index of the subject based on the state data related to the subject and the specific attribute value;

The training module 1340 is configured to train the state recognition model according to the standardized state index.

According to the technical solutions provided by the embodiments of the present disclosure, the first obtaining module is configured to obtain the characteristic data related to the subject and the state data related to the subject; the first determining module is configured to obtain according to the preset mapping relationship , determine a specific attribute value that matches the feature data; a second determining module is configured to determine a standardized state index of the subject based on the subject-related state data and the specific attribute value; a training module, The state recognition model is configured to train the state identification model according to the standardized state index, so that the number of specific events can be converted into a standardized state index, and the state identification model trained based on the standardized state index can more accurately identify the state of the subject.

According to an embodiment of the present disclosure, the mapping relationship includes at least a corresponding relationship between the feature data and a specific attribute value, wherein the feature attribute values corresponding to the feature data in different value ranges are different.

According to the technical solutions provided by the embodiments of the present disclosure, the mapping relationship includes at least the corresponding relationship between the feature data and the specific attribute value, wherein the feature attribute values corresponding to the feature data with different value ranges are different, so that the mapping relationship can be based on the mapping relationship. The number of specific events is converted into a standardized state index, and the state recognition model trained based on the standardized state index can more accurately identify the state of the subject.

According to an embodiment of the present disclosure, the determining of the standardized state index of the subject based on the state data related to the subject and the specific attribute value includes:

determining an original state indicator based on the state data and the specific attribute value;

According to the maximum and minimum values of the original state indexes of different subjects, the original state indexes are mapped to a predetermined value interval to obtain a standardized state index.

According to the technical solutions provided by the embodiments of the present disclosure, the original state index is determined based on the state data and the specific attribute value; the original state index is mapped to a predetermined value according to the maximum value and the minimum value of the original state index of different subjects According to the value interval, a standardized state index is obtained, so that the obtained standardized state index is more effective, and the state recognition model trained based on the standardized state index can more accurately identify the state of the subject.

According to an embodiment of the present disclosure, the feature data includes graph features determined based on the association relationship of the subject in the subject relationship graph.

According to the technical solutions provided by the embodiments of the present disclosure, the state of the subject can be better identified in combination with the graph features by using the graph features determined based on the association relationship of the subjects in the subject relationship graph as feature data.

According to an embodiment of the present disclosure, the training of the state recognition model according to the standardized state index includes:

determining a first vector representation of characteristic data of the subject;

Converting the first vector representation from a high-dimensional sparse vector representation to a low-dimensional dense second vector representation;

The second vector representation is spliced with the third vector representation of the feature data of the subject in the subject relationship diagram adjacent to the subject to obtain the fourth vector representation of the subject's feature data;

encoding the fourth vector representation as a fifth vector representation of the feature data of the subject having a predetermined dimension as training data for the state recognition model;

Using the standardized state index as a label, the state recognition model is trained.

According to the technical solutions provided by the embodiments of the present disclosure, the first vector representation of the feature data of the subject is determined; the first vector representation is converted from a high-dimensional sparse vector representation to a low-dimensional dense second vector representation; The second vector representation is spliced with the third vector representation of the feature data of the subject's adjacent subjects in the subject relationship diagram to obtain a fourth vector representation of the subject's feature data; the fourth vector representation is Representing the fifth vector representation encoded as the feature data of the subject with a predetermined dimension, as the training data of the state recognition model; using the standardized state index as a label, training the state recognition model, so as to be able to better Identify the state of the subject using graph features.

According to an embodiment of the present disclosure, the first vector representation includes a one-hot feature representation, and the converting the first vector representation from a high-dimensional sparse vector representation to a low-dimensional dense second vector representation includes:

converting the one-hot feature representation into a sixth vector representation through an embedding operation;

A weighted summation of the sixth vector representation yields the low-dimensional dense second vector representation.

According to the technical solutions provided by the embodiments of the present disclosure, the first vector representation represented by the one-hot feature is converted into a sixth vector representation through an embedding operation; and the low-dimensional dense representation is obtained by weighted summation of the sixth vector representation. The second vector representation solves the problem of excessive resource occupation caused by sparse features.

According to an embodiment of the present disclosure, the apparatus may further include a presentation module configured to present the standardized status indicator in a graphical manner, so that the status of the subject can be displayed more intuitively.

FIG. 14 shows a block diagram of a state recognition apparatus according to an embodiment of the present disclosure. Wherein, the apparatus may be realized by software, hardware or a combination of the two to become part or all of the electronic device.

As shown in FIG. 14 , the state identification device 1400 includes a second acquisition module 1410 and an identification module 1420 .

The second obtaining module 1410 is configured to obtain characteristic data of the subject to be identified;

The identification module 1420 is configured to input the feature data into the state identification model as described above, so as to obtain the standardized state index of the subject to be identified.

According to the technical solutions provided by the embodiments of the present disclosure, the second acquisition module is configured to acquire feature data of the subject to be recognized; the recognition module is configured to input the feature data into the state recognition model training method described above The trained state identification model is used to obtain the standardized state index of the subject to be identified, so that the state of the subject can be more accurately identified through the standardized state index.

The present disclosure also discloses an electronic device, and FIG. 15 shows a block diagram of the electronic device according to an embodiment of the present disclosure.

As shown in FIG. 15 , the electronic device 1500 includes a memory 1501 and a processor 1502 , wherein the memory 1501 is used to store one or more computer instructions, wherein the one or more computer instructions are executed by the processor 1502 to do the following:

obtain subject-related characteristic data, and state data related to said subject;

Determine a specific attribute value matching the feature data according to a preset mapping relationship;

determining a normalized state indicator for the subject based on the subject-related state data and the specific attribute value;

The state recognition model is trained according to the standardized state indicators.

According to an embodiment of the present disclosure, the mapping relationship includes at least: a correspondence relationship between the feature data and a specific attribute value; wherein the feature attribute values corresponding to the feature data in different value ranges are different.

According to an embodiment of the present disclosure, the determining of the standardized state index of the subject based on the state data related to the subject and the specific attribute value includes:

determining an original state indicator based on the state data and the specific attribute value;

According to the maximum and minimum values of the original state indexes of different subjects, the original state indexes are mapped to a predetermined value interval to obtain a standardized state index.

According to an embodiment of the present disclosure, the feature data includes graph features determined based on the association relationship of the subject in the subject relationship graph.

According to an embodiment of the present disclosure, the training of the state recognition model according to the standardized state index includes:

determining a first vector representation of characteristic data of the subject;

Converting the first vector representation from a high-dimensional sparse vector representation to a low-dimensional dense second vector representation;

The second vector representation is spliced with the third vector representation of the feature data of the subject in the subject relationship diagram adjacent to the subject to obtain the fourth vector representation of the subject's feature data;

encoding the fourth vector representation as a fifth vector representation of the feature data of the subject having a predetermined dimension as training data for the state recognition model;

Using the standardized state index as a label, the state recognition model is trained.

According to an embodiment of the present disclosure, the first vector representation includes a one-hot feature representation, and the converting the first vector representation from a high-dimensional sparse vector representation to a low-dimensional dense second vector representation includes:

converting the one-hot feature representation into a sixth vector representation through an embedding operation;

A weighted summation of the sixth vector representation yields the low-dimensional dense second vector representation.

According to the embodiment of the present disclosure, the processor 1502 is further configured to perform: presenting the standardized state indicator in a graphical manner.

According to an embodiment of the present disclosure, the execution of the one or more computer instructions by the processor 1502 may implement the following operations:

Obtain the characteristic data of the subject to be identified;

The feature data is input into the state recognition model as described above to obtain standardized state indicators of the subject to be recognized.

According to an embodiment of the present disclosure, the execution of the one or more computer instructions by the processor 1502 may implement the following operations:

Obtain characteristic data of the enterprise and the number of contract litigation cases in which the enterprise is a defendant, wherein the characteristic data includes turnover;

According to the mapping relationship between the turnover and the contract quantity, determine the contract quantity of the enterprise;

determining a standardized risk indicator for the enterprise based on the number of contract litigation cases and the number of contracts for the enterprise;

The training data is determined based on the characteristic data of the enterprise, and the risk identification model is trained with the standardized risk index as a label.

FIG. 16 shows a schematic structural diagram of a computer system suitable for implementing the methods and apparatuses of embodiments of the present disclosure.

As shown in FIG. 16, a computer system 1600 includes a processing unit 1601 that can perform the above-described implementation according to a program stored in a read only memory (ROM) 1602 or a program loaded from a storage section 1608 into a random access memory (RAM) 1603 various treatments in the example. In the RAM 1603, various programs and data necessary for the operation of the system 1600 are also stored. The processing unit 1601 , the ROM 1602 , and the RAM 1603 are connected to each other through a bus 1604 . Input/output (I/O) interface 1605 is also connected to bus 1604 .

The following components are connected to the I/O interface 1605: an input section 1606 including a keyboard, a mouse, etc.; an output section 1607 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 1608 including a hard disk, etc. ; and a communication section 1609 including a network interface card such as a LAN card, a modem, and the like. The communication section 1609 performs communication processing via a network such as the Internet. Drivers 1610 are also connected to I/O interface 1605 as needed. A removable medium 1611, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is mounted on the drive 1610 as needed so that a computer program read therefrom is installed into the storage section 1608 as needed. The processing unit 1601 may be implemented as a processing unit such as a CPU, a GPU, a TPU, an FPGA, and an NPU.

In particular, according to embodiments of the present disclosure, the methods described above may be implemented as computer software programs. For example, embodiments of the present disclosure include a computer program product comprising a computer program tangibly embodied on a readable medium thereof, the computer program containing program code for performing the above-described method. In such an embodiment, the computer program may be downloaded and installed from the network through the communication portion 1609, and/or installed from the removable medium 1611.

The flowchart and block diagrams in the Figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to various embodiments of the present disclosure. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of code that contains one or more functions for implementing the specified logical function(s) executable instructions. It should also be noted that, in some alternative implementations, the functions noted in the blocks may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It is also noted that each block of the block diagrams and/or flowchart illustrations, and combinations of blocks in the block diagrams and/or flowchart illustrations, can be implemented in dedicated hardware-based systems that perform the specified functions or operations , or can be implemented in a combination of dedicated hardware and computer instructions.

The units or modules involved in the embodiments of the present disclosure may be implemented in a software manner, or may be implemented in a programmable hardware manner. The described units or modules may also be provided in the processor, and the names of these units or modules do not constitute a limitation on the units or modules themselves in certain circumstances.

As another aspect, the present disclosure also provides a computer-readable storage medium, and the computer-readable storage medium may be a computer-readable storage medium included in the electronic device or computer system in the above-mentioned embodiments; it may also exist independently , a computer-readable storage medium that does not fit into a device. The computer-readable storage medium stores one or more programs used by one or more processors to perform the methods described in the present disclosure.

The above description is merely a preferred embodiment of the present disclosure and an illustration of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in the present disclosure is not limited to the technical solutions formed by the specific combination of the above-mentioned technical features, and should also cover the above-mentioned technical features without departing from the inventive concept. Other technical solutions formed by any combination of its equivalent features. For example, a technical solution is formed by replacing the above features with the technical features disclosed in the present disclosure (but not limited to) with similar functions.

Claims (17)

1. A training method of a state recognition model comprises the following steps:

obtaining feature data relating to a subject, and status data relating to the subject;

determining a specific attribute value matched with the characteristic data according to a preset mapping relation;

determining a standardized status indicator for the subject based on the status data related to the subject and the specific attribute value;

and training the state recognition model according to the standardized state index.

2. The method of claim 1, the mapping comprising at least: the corresponding relation between the characteristic data and the specific attribute value; and the characteristic attribute values corresponding to the characteristic data in different value ranges are different.

3. The method of claim 1, wherein the determining a normalized status indicator for the subject based on the status data related to the subject and the particular attribute value comprises:

determining an original state index based on the state data and the specific attribute value;

and mapping the original state indexes to a preset value interval according to the maximum value and the minimum value of the original state indexes of different main bodies to obtain the standardized state indexes.

4. The method of claim 1, wherein the feature data comprises a graph feature determined based on an associative relationship of the subject in a subject relationship graph.

5. The method of any of claims 4, wherein said training said state recognition model according to said normalized state metrics comprises:

determining a first vector representation of feature data of the subject;

converting the first vector representation from a high-dimensional sparse vector representation to a low-dimensional dense second vector representation;

splicing the second vector representation with a third vector representation of feature data of an adjacent main body of the main body in the main body relation graph to obtain a fourth vector representation of the feature data of the main body;

encoding the fourth vector representation as a fifth vector representation of feature data of the subject having predetermined dimensions as training data of the state recognition model;

and training the state recognition model by taking the standardized state index as a label.

6. The method of claim 5, wherein the first vector representation comprises a one-hot feature representation, the converting the first vector representation from a high-dimensional sparse vector representation to a low-dimensional dense second vector representation comprises:

converting the one-hot feature representation into a sixth vector representation through an embedding operation;

and carrying out weighted summation on the sixth vector representation to obtain the second vector representation with low dimension and density.

7. A state recognition method, comprising:

acquiring characteristic data of a main body to be identified;

inputting the characteristic data into the state recognition model according to any one of claims 1-6 to obtain a standardized state index of the subject to be recognized.

8. A method of training a risk recognition model, comprising:

acquiring characteristic data of an enterprise and the number of contract litigation cases of the enterprise as being billed, wherein the characteristic data comprises turnover;

determining the number of contracts of the enterprise according to the mapping relation between the turnover and the number of contracts;

determining a normalized risk indicator for the enterprise based on the number of contract litigation cases and the number of contracts for the enterprise;

and determining training data based on the characteristic data of the enterprise, and training a risk identification model by taking the standardized risk index as a label.

9. A state recognition model training apparatus comprising:

a first acquisition module configured to acquire feature data related to a subject, and status data related to the subject;

the first determination module is configured to determine a specific attribute value matched with the feature data according to a preset mapping relation;

a second determination module configured to determine a normalized status indicator for the subject based on the status data related to the subject and the specific attribute value;

a training module configured to train the state recognition model according to the normalized state index.

10. The apparatus of claim 9, wherein the mapping relationship comprises at least:

and the corresponding relation between the characteristic data and the specific attribute value, wherein the characteristic attribute values corresponding to the characteristic data in different value ranges are different.

11. The apparatus of claim 9, wherein the determining a normalized status indicator for the subject based on the status data related to the subject and the particular attribute value comprises:

determining an original state index based on the state data and the specific attribute value;

and mapping the original state indexes to a preset value interval according to the maximum value and the minimum value of the original state indexes of different main bodies to obtain the standardized state indexes.

12. The apparatus of claim 9, wherein the feature data comprises a graph feature determined based on an associative relationship of the subject in a subject relationship graph.

13. The apparatus of claim 12, wherein training the state recognition model according to the normalized state metrics comprises:

determining a first vector representation of feature data of the subject;

converting the first vector representation from a high-dimensional sparse vector representation to a low-dimensional dense second vector representation;

splicing the second vector representation with a third vector representation of feature data of an adjacent main body of the main body in the main body relation graph to obtain a fourth vector representation of the feature data of the main body;

encoding the fourth vector representation as a fifth vector representation of feature data of the subject having predetermined dimensions as training data of the state recognition model;

and training the state recognition model by taking the standardized state index as a label.

14. The apparatus of claim 13, wherein the first vector representation comprises a one-hot feature representation, the converting the first vector representation from a high-dimensional sparse vector representation to a low-dimensional dense second vector representation comprises:

converting the one-hot feature representation into a sixth vector representation through an embedding operation;

and carrying out weighted summation on the sixth vector representation to obtain the second vector representation with low dimension and density.

15. A state recognition device comprising:

the second acquisition module is configured to acquire characteristic data of the subject to be identified;

an identification module configured to input the feature data into the state recognition model according to any one of claims 1 to 7 to obtain a standardized state index of the subject to be recognized.

16. An electronic device comprising a memory and a processor; wherein the memory is configured to store one or more computer instructions, wherein the one or more computer instructions are executed by the processor to implement the method steps of any of claims 1-8.

17. A readable storage medium having stored thereon computer instructions, characterized in that the computer instructions, when executed by a processor, carry out the method steps of any of claims 1 to 8.

Images