Probability information determination method and device, storage medium and electronic equipment

Abstract

The embodiment of the invention provides a probability information determination method and device, a storage medium and electronic equipment, and the method comprises the steps: determining the type of each piece of first information of a target object, and determining the preprocessing mode of each piece of first information according to the type of each piece of first information; preprocessing each piece of first information according to the preprocessing mode of each piece of first information to obtain preprocessed second information; inputting a plurality of pieces of second information into a plurality of prediction models, so that the plurality of prediction models determine an evaluation value of the target object in the target state based on the plurality of pieces of second information; calling a calculation function, and calculating the plurality of evaluation values through the calculation function to determine a target evaluation value of the target object in the target state; and determining probability information that the target object is in the target state according to the target evaluation value.

Description

Technical Field

[0001] This application relates to the field of data processing, and more specifically, to a method, apparatus, storage medium, and electronic device for determining probability information. Background Technology

[0002] Driven by both policy guidance and market development, financial institutions are facing a significant increase in the demand for loans from micro and small enterprises. This requires us not only to respond quickly, but also to accurately identify entities that may be in a certain state (e.g., default or risk) from massive amounts of data, so as to optimize resource allocation and improve the robustness and efficiency of the financial system.

[0003] Currently, most banks and financial institutions rely on expert experience when assessing the probability of small and micro enterprises being in a certain state, which leads to a low accuracy rate.

[0004] There is currently no effective solution to the problem that the accuracy of assessing the probability of an object being in a certain state often relies on expert experience.

[0005] Therefore, it is necessary to improve the relevant technology to overcome the aforementioned defects. Summary of the Invention

[0006] This application provides a method, apparatus, storage medium, and electronic device for determining probability information, in order to at least solve the problem in the related art that when assessing the probability of an object being in a certain state, the reliance on expert experience often leads to low accuracy.

[0007] According to one embodiment of this application, a method for determining probability information is provided, applied to a computer device, comprising: determining the type of each piece of first information of a target object, and determining a preprocessing method for each piece of first information according to the type of each piece of first information, wherein the first information is information used to determine the probability that the target object is in a target state, and the preprocessing method includes at least one of the following: binning, aggregation, standardization, encoding, compression, and derivation; preprocessing each piece of first information according to the preprocessing method of each piece of first information to obtain preprocessed second information; inputting multiple pieces of second information into multiple prediction models, so that the multiple prediction models determine an evaluation value of the target object being in the target state based on the multiple pieces of second information; calling a calculation function and calculating multiple evaluation values ​​through the calculation function to determine a target evaluation value of the target object being in the target state; and determining the probability information of the target object being in the target state based on the target evaluation value.

[0008] In an exemplary embodiment, preprocessing each piece of first information according to a preprocessing method to obtain preprocessed second information includes: when the type of the first information is a continuous number type, performing binning processing on the first information to determine the binning category of the first information; aggregating according to the binning category of the first information to obtain aggregated first information; encoding the first information based on a first encoding method to obtain second information, and encoding the first information based on a second encoding method to obtain third information; and compressing the second information and the third information to obtain compressed second information and compressed third information, wherein the preprocessed second information includes at least the compressed second information and the compressed third information.

[0009] In one exemplary embodiment, preprocessing each piece of first information according to a preprocessing method to obtain preprocessed second information includes: aggregating the first information if the type of the first information is a discrete number type or a category type to obtain aggregated fourth information; encoding the fourth information based on a first encoding method to obtain fifth information, and encoding the fourth information based on a second encoding method to obtain sixth information; and compressing the fifth information and the sixth information to obtain compressed fifth information and compressed sixth information, wherein the preprocessed second information includes at least the compressed fifth information and the compressed sixth information.

[0010] In an exemplary embodiment, preprocessing each piece of first information according to a preprocessing method to obtain preprocessed second information includes: determining the associated features of each piece of first information; and determining the feature value corresponding to the associated features according to each piece of first information, wherein the preprocessed second information includes the feature value corresponding to the associated features.

[0011] In one exemplary embodiment, before inputting multiple sets of second information into multiple prediction models, the method further includes: acquiring a dataset of multiple category labels, and adjusting the original category labels of the datasets of multiple category labels to a first category label or a second category label, respectively; training multiple initial prediction models based on a first dataset of the first category label and a second dataset of the second category label to obtain multiple intermediate prediction models; and determining the multiple prediction models based on the multiple intermediate prediction models.

[0012] In an exemplary embodiment, before training multiple initial prediction models based on a first dataset of the first category label and a second dataset of the second category label to obtain multiple intermediate prediction models, the method further includes: dividing the first dataset of the first category label into X first subsets and dividing the second dataset of the second category label into X second subsets, where X is a positive integer; training the multiple initial prediction models based on X-1 first subsets and X-1 second subsets, and validating the trained prediction models based on other first subsets and other second subsets to obtain a validation result, wherein the X-1 first subsets and X-1 second subsets used to train different initial prediction models are different, the other first subsets are first datasets other than the X-1 first subsets among the X first subsets, and the other second subsets are second datasets other than the X-1 second subsets among the X second subsets; and determining whether the trained prediction model is the intermediate prediction model based on the validation result.

[0013] In one exemplary embodiment, the method for determining the plurality of prediction models based on the plurality of intermediate prediction models further includes: obtaining a third dataset with the original category label being a first category label and a fourth dataset with the original category label being a second category label; and training the intermediate prediction models based on the third dataset and the fourth dataset to determine the plurality of prediction models.

[0014] According to another embodiment of this application, a probability information determination device is provided, applied to a computer device, comprising: a first determination module, configured to determine the type of each first piece of information of a target object, and determine a preprocessing method for each first piece of information according to the type of each first piece of information, wherein the first information is information used to determine the probability of the target object being in a target state, and the preprocessing method includes at least one of the following: binning, aggregation, standardization, encoding, compression, and derivation; a preprocessing module, configured to preprocess each first piece of information according to the preprocessing method of each first piece of information to obtain preprocessed second information; a second determination module, configured to input multiple pieces of second information into multiple prediction models, so that the multiple prediction models determine an evaluation value of the target object being in the target state based on the multiple pieces of second information; a calculation module, configured to call a calculation function and calculate multiple evaluation values ​​through the calculation function to determine a target evaluation value of the target object being in the target state; and a third determination module, configured to determine the probability information of the target object being in the target state based on the target evaluation value.

[0015] According to yet another embodiment of this application, a computer-readable storage medium is also provided, wherein a computer program is stored therein, and the computer program is configured to perform the steps in any of the above method embodiments when it is run.

[0016] According to yet another embodiment of this application, an electronic device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to perform the steps in any of the above method embodiments.

[0017] According to yet another embodiment of this application, a computer program product is also provided, including a computer program that, when executed by a processor, implements the steps in any of the above method embodiments.

[0018] This application first identifies and classifies all raw information (i.e., first information). Based on the information attributes, it preprocesses the data using methods such as binning, aggregation, standardization, encoding, compression, or derivation to generate structured features (i.e., second information). Subsequently, the second information is input into a set of multivariate prediction models to calculate the assessment value of the default state. Finally, the assessment values ​​from multiple models are fused to obtain the probability of being in a certain state. This application's embodiments improve data quality through different preprocessing methods, utilize multi-model assessment and result fusion, and examine risk from different perspectives, significantly improving the accuracy of risk assessment. Therefore, it can solve the problem in related technologies where assessing the probability of an object being in a certain state often relies on expert experience, leading to low accuracy. Attached Figure Description

[0019] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0020] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a hardware structure block diagram of a computer device for a method of determining probability information according to an embodiment of this application; Figure 2 This is a flowchart of a method for determining probability information according to an embodiment of this application; Figure 3 This is a structural block diagram of a probability information determination device according to an embodiment of this application. Detailed Implementation

[0022] The embodiments of this application will be described in detail below with reference to the accompanying drawings and examples.

[0023] It should be noted that the terms "first," "second," etc., in the specification, claims, and drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence.

[0024] It should be noted that in the embodiments of this application, certain software, components, models and other existing solutions in the industry may be mentioned. These should be regarded as exemplary and are only intended to illustrate the feasibility of implementing the technical solution of this application. However, it does not mean that the applicant has used or necessarily used the solution.

[0025] Furthermore, the collection, storage, use, processing, transmission, provision, and disclosure of financial data or user data involved in the technical solution of this application all comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0026] The methods and embodiments provided in this application can be executed in a computer device or similar computing device. Taking running on a computer device as an example, Figure 1 This is a hardware structure block diagram of a computer device for a method of determining probability information according to an embodiment of this application. For example... Figure 1 As shown, a computer device may include one or more ( Figure 1 Only one is shown in the diagram. A processor 102 (which may include, but is not limited to, a microprocessor or programmable logic device) and a memory 104 for storing data are also shown. The computer device may further include a transmission device 106 for communication functions and an input / output device 108. Those skilled in the art will understand that... Figure 1 The structure shown is for illustrative purposes only and does not limit the structure of the computer device described above. For example, the computer device may also include components that are more... Figure 1 The more or fewer components shown, or having the same Figure 1 The different configurations shown.

[0027] The memory 104 can be used to store computer programs, such as application software programs and modules, like the computer program corresponding to the probability information determination method in this embodiment. The processor 102 executes various functional applications and data processing by running the computer program stored in the memory 104, thus implementing the above-described method. The memory 104 may include high-speed random access memory and non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include memory remotely located relative to the processor 102, and these remote memories can be connected to computer devices via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0028] The transmission device 106 is used to receive or send data via a network. Specific examples of the network described above may include a wireless network provided by a communication provider for the computer equipment. In one example, the transmission device 106 includes a Network Interface Controller (NIC), which can connect to other network devices via a base station to communicate with the Internet. In another example, the transmission device 106 may be a Radio Frequency (RF) module used for wireless communication with the Internet.

[0029] This embodiment provides a method for determining probability information, applied to the aforementioned computer device. Figure 2 This is a flowchart of a method for determining probability information according to an embodiment of this application, such as... Figure 2 As shown, the process includes the following steps: Step S202: Determine the type of each piece of first information of the target object, and determine the preprocessing method of each piece of first information according to the type of each piece of first information, wherein the first information is information used to determine the probability that the target object is in a target state, and the preprocessing method includes at least one of the following: binning, aggregation, standardization, encoding, compression, and derivation. It should be noted that the above-mentioned target object can be understood as an enterprise, and the above-mentioned target status can be understood as a risk status, a default status, etc.

[0030] In step S202, the type of each data point in the first information is identified. The type of each data point includes, but is not limited to, numerical and categorical types. Once the information type is confirmed, the preprocessing method for each data point is determined. Based on the determined information type and preprocessing method, specific preprocessing measures are implemented for each data point. For example, for continuous numerical features, binning is performed according to their distribution, converting them into interval categories.

[0031] Step S204: Preprocess each piece of first information according to the preprocessing method of each piece of first information to obtain preprocessed second information; Step S206: Input multiple pieces of second information into multiple prediction models so that the multiple prediction models determine the evaluation value of the target object in the target state based on the multiple pieces of second information; Step S208: Invoke the calculation function and calculate multiple evaluation values ​​through the calculation function to determine the target evaluation value of the target object in the target state; It should be noted that the above calculation of the multiple evaluation values ​​using the calculation function includes, but is not limited to, weighted summation of the multiple evaluation values ​​using the calculation function.

[0032] Step S210: Determine the probability information of the target object being in the target state based on the target evaluation value.

[0033] Through the above steps, firstly, all raw information (i.e., first information) is identified and classified. Based on the information attributes, it is preprocessed using methods such as binning, aggregation, standardization, encoding, compression, or derivation to generate structured features (i.e., second information). Subsequently, the second information is input into a set of multivariate prediction models to calculate the assessment value of the default state. Finally, the assessment values ​​from multiple models are fused to obtain the probability of being in a certain state. This embodiment improves data quality through different preprocessing methods, utilizes multi-model assessment and result fusion, and examines risk from different perspectives, significantly improving the accuracy of risk assessment. Therefore, it can solve the problem in related technologies where assessing the probability of an object being in a certain state often relies on expert experience, leading to low accuracy.

[0034] Optionally, preprocessing each piece of first information according to a preprocessing method to obtain preprocessed second information includes: when the type of the first information is a continuous number type, binning the first information to determine the binning category of the first information; aggregating according to the binning category of the first information to obtain aggregated first information; encoding the first information based on a first encoding method to obtain second information, and encoding the first information based on a second encoding method to obtain third information; and compressing the second information and the third information to obtain compressed second information and compressed third information, wherein the preprocessed second information includes at least the compressed second information and the compressed third information.

[0035] First, for information that is continuous numerical in the initial information, binning will be performed. Binning is a method to convert continuous features into categorical features. By dividing the numerical range into multiple bins (or categories), the dimensionality of the data can be reduced, and continuous numerical information can be transformed into categorical information that is easier for the model to process. For example, the continuous numerical feature of "average monthly transaction amount" can be divided into different bins according to specific ranges, such as "0-1000", "1000-5000", "above 5000", etc., to reduce the model's sensitivity to numerical values ​​and improve its generalization ability.

[0036] After binning, the first information will be aggregated according to the bin category to obtain the aggregated first information. Aggregation can employ various strategies, such as calculating the mean, median, mode, or other statistics for each bin, thereby transforming the feature value of each bin into a representative value and further simplifying the data structure. The aggregated first information will be encoded based on a first encoding method to obtain the second information; simultaneously, the same information will be encoded based on a second encoding method to obtain the third information. The first and second encoding methods include, but are not limited to: TfidfVectorizer encoding: It is usually used for text features, but in some cases it can also be applied to classification features. It can reflect the importance of information in different categories, and thus identify which bins have a greater impact on the overall evaluation.

[0037] CountVectorizer encoding: Primarily used for text features, but in classification features, especially when the feature values ​​have a certain distribution density, it can also effectively reflect the frequency of feature values.

[0038] Finally, the second and third information processed by the above two encoding methods are compressed to obtain compressed second and third information. The compression processing in this application embodiment includes, but is not limited to, Principal Component Analysis (PCA), Singular Value Decomposition (SVD), or Truncated Singular Value Decomposition (tSVD).

[0039] Through a preprocessing flow of binning, aggregation, encoding, and compression, continuous numerical information is transformed into second and third information. This information is then compressed, further transforming into compressed second and third information. This embodiment not only simplifies the structure of the original data and improves the model's processing efficiency, but also ensures that the diversity and importance of the information are preserved through encoding and compression.

[0040] Optionally, preprocessing each piece of first information according to a preprocessing method to obtain preprocessed second information includes: aggregating the first information if the type of the first information is discrete number type or category type to obtain aggregated fourth information; encoding the fourth information based on a first encoding method to obtain fifth information, and encoding the fourth information based on a second encoding method to obtain sixth information; and compressing the fifth information and the sixth information to obtain compressed fifth information and compressed sixth information, wherein the preprocessed second information includes at least the compressed fifth information and the compressed sixth information.

[0041] First, for the first piece of information, which is either a discrete number type or a category type, an aggregation operation will be performed to obtain the aggregated fourth piece of information. The purpose of aggregation is to reduce the dimensionality of the data by grouping similar data points, while preserving the structure and key features of the information. For example, for category type information, such as "customer type code", the information can be aggregated by calculating the number of times each code appears, or by merging similar customer types into one category; for discrete number type information, such as "number of transactions", it can be grouped or classified according to the distribution range of the number of transactions, and then the sum or average number of transactions in each category can be calculated.

[0042] The aggregated fourth information will be further encoded to obtain the fifth and sixth information. One encoding method is used, for example, TfidfVectorizer encoding. Another encoding method is used, for example, CountVectorizer encoding.

[0043] The encoded fifth and sixth information are compressed to obtain compressed fifth and sixth information. The compression methods in this application include PCA, SVD, or TruncatedSVD.

[0044] In this embodiment of the application, information of discrete number type or category type is converted into a format that is easier for the model to understand and analyze through aggregation, encoding and compression. At the same time, the processing process ensures that the dimensionality of the data is reduced.

[0045] Optionally, each piece of first information is preprocessed according to the preprocessing method of each piece of first information to obtain preprocessed second information, including: determining the association feature of each piece of first information; determining the feature value corresponding to the association feature according to each piece of first information, wherein the preprocessed second information includes: the feature value corresponding to the association feature.

[0046] Optionally, based on the associated features, corresponding feature values ​​can be calculated or derived, including but not limited to: average value, cumulative value, category statistical features (such as the frequency of occurrence of each category), and text information features (such as keyword frequency and topic model results).

[0047] For example, when the first information includes "cumulative monthly debit transaction amount" and "cumulative monthly debit transaction count," the associated feature can include: an average value, which is obtained by dividing "cumulative monthly debit transaction amount" by "cumulative monthly debit transaction count" to obtain the "monthly average debit transaction amount." Similarly, when the first information includes "cumulative monthly credit transaction amount" and "cumulative monthly credit transaction count," the "monthly average credit transaction amount" is obtained by dividing "cumulative monthly credit transaction amount" by "cumulative monthly credit transaction count."

[0048] Optionally, before inputting the second information into multiple prediction models, the method further includes: obtaining a dataset of multiple category labels, and adjusting the original category labels of the dataset of multiple category labels to a first category label or a second category label respectively; training multiple initial prediction models based on a first dataset of the first category label and a second dataset of the second category label to obtain multiple intermediate prediction models; and determining the multiple prediction models based on the multiple intermediate prediction models.

[0049] In this embodiment, a dataset containing multiple category labels is obtained from available data. These category labels reflect different probabilities or credit statuses of being in a target state. Next, the original category labels in the dataset are adjusted, transforming them into two simplified category labels: a first category label and a second category label. These two labels can typically be considered as binary classification labels such as "normal" and "abnormal," "probability of being in a target state less than or equal to a first preset probability," and "probability of being in a target state greater than a second preset probability." For example, the original level label "probability of being in a target state less than or equal to a first preset probability" (which can be understood as low risk) is adjusted to "probability of being in a target state less than or equal to a first preset probability," and the two main level labels "probability of being in a target state greater than a first preset probability but less than or equal to a second preset probability" (which can be understood as medium risk) and "probability of being in a target state greater than a second preset probability" (which can be understood as high risk) are adjusted to "probability of being in a target state greater than a second preset probability."

[0050] The dataset is split into a first dataset and a second dataset based on the adjusted first and second class labels. The first dataset contains all samples with the first class label, and the second dataset contains all samples with the second class label. Multiple initial prediction models are then trained using both datasets.

[0051] After obtaining several intermediate prediction models, it is necessary to further evaluate the performance of these models and select the most suitable model combination. The area under the curve (AUC) of the model is evaluated using a validation set.

[0052] Optionally, before training multiple initial prediction models based on a first dataset of the first category label and a second dataset of the second category label to obtain multiple intermediate prediction models, the method further includes: dividing the first dataset of the first category label into X first subsets and dividing the second dataset of the second category label into X second subsets; training the multiple initial prediction models based on X-1 first subsets and X-1 second subsets, and validating the trained prediction models based on other first subsets and other second subsets to obtain a validation result, wherein the X-1 first subsets and X-1 second subsets used to train different initial prediction models are different, the other first subsets are the first datasets other than the X-1 first subsets among the X first subsets, and the other second subsets are the second datasets other than the X-1 second subsets among the X second subsets; and determining whether the trained prediction model is the intermediate prediction model based on the validation result.

[0053] In this embodiment of the application, the first dataset with the first category label (e.g., a dataset labeled "the probability of being in the target state is less than or equal to the first preset probability") is divided into X first sub-datasets. Similarly, the second dataset with the second category label (e.g., a dataset labeled "the probability of being in the target state is greater than the second preset probability") is divided into X second sub-datasets.

[0054] Next, multiple initial prediction models are trained based on X-1 first subsets and X-1 second subsets (i.e., all subsets except for a reserved validation set). Each model may be assigned to different combinations of subsets for training. Simultaneously, the trained models are validated based on the remaining first subset (the other first subsets) and the remaining second subset (the other second subsets) to obtain validation results. Based on the validation results, it is determined whether the trained prediction model is suitable as an intermediate prediction model.

[0055] For example, using 5-fold cross-validation, the data is divided into five equal parts: the first dataset and the second dataset. In each training iteration, four parts are used as the training set, and the remaining part is used as the validation set. Through this process, five intermediate prediction models with different parameters are obtained.

[0056] In this embodiment, cross-validation is used to improve the model's generalization ability, ensuring that the model not only fits the training data well but also reliably predicts future or unseen data. Dividing the dataset into subsets and repeatedly training and validating the model helps identify its shortcomings, and through feedback loops, continuously improves the model, ultimately selecting the best-performing intermediate prediction model.

[0057] Optionally, the method further includes determining the plurality of prediction models based on the plurality of intermediate prediction models: obtaining a third dataset with the original category label as the first category label and a fourth dataset with the original category label as the second category label; and training the intermediate prediction models based on the third dataset and the fourth dataset to determine the plurality of prediction models.

[0058] After obtaining several intermediate prediction models through initial training, a third dataset with the first category label and a fourth dataset with the second category label are further acquired. The first category label typically refers to a state where the probability of being in the target state is less than or equal to a first preset probability, while the second category label corresponds to a state where the probability of being in the target state is greater than a second preset probability. Obtaining these datasets allows for retraining of the model, enabling it to more accurately identify and classify specific categories.

[0059] The intermediate prediction model is retrained using the third and fourth datasets. This can improve the model's performance on specific classes, especially in cases of class imbalance or poor model identification of a particular class. In this way, the model can better learn the subtle discriminative features between different samples, thus enabling more accurate evaluation in practical applications.

[0060] After the refinement training phase, based on the model's performance on the third and fourth datasets, multiple intermediate prediction models are further evaluated and optimized to ultimately determine the multiple prediction models.

[0061] To better understand the process of determining the above probability information, the implementation flow of the above probability information determination method will be described below in conjunction with optional embodiments, but this is not intended to limit the technical solution of the embodiments of this application.

[0062] This embodiment provides a method for determining probability information, which includes the following steps: Step 1, Obtain customer information (equivalent to the first information in the above embodiment); The customer data includes: customer corporate deposit account balance information (as shown in Table 1), customer corporate deposit account transaction information, customer basic information, customer business information, customer credit summary information, customer housing provident fund payment record basic information, customer legal information, customer tax payment information, customer master table list and tags.

[0063] Table 1

[0064] Step 2: Data processing and feature construction; 1) Processing customer corporate deposit account balance information, including: 1.1 The "Data Date" field in the corporate deposit account balance information is converted to a unified "datetime64" date and time format to ensure data consistency and calculation accuracy. Next, the year and month information are extracted from the "Data Date" to lay the foundation for subsequent time series analysis and data trend insights. Time difference features are generated by calculating the difference between each customer's "Data Date" and a preset baseline time point.

[0065] 1.2. The account's "deposit balance" and "monthly accumulated amount" are binned according to a finely defined numerical range, specifically [0, 1e-2, 1e-1, ..., 1e13]. Converting continuous monetary data into categorical features facilitates understanding and processing by machine learning algorithms. Furthermore, binning allows for the capture of the potential impact of different monetary levels on customer default risk, ultimately resulting in two new binned features.

[0066] 1.3. Group the corporate deposit account balance information and all previously derived features according to "customer number". For each group, aggregate according to "number of unique values", "last value", and "first value". The aggregation operation not only simplifies the data structure, but also preserves the information of account balance changes over time.

[0067] 1.4 For “Deposit Balance”, “Monthly Accumulation”, and date-related features (“Data Date_Year”, “Data Date_Month”, etc.), group them by “Customer Number” and then perform numerical statistics, including “Maximum Value”, “Minimum Value”, “Median”, “Mean”, “Standard Deviation”, “Mean Absolute Deviation”, “Skewing”, “Quarter 1 / 2”, “Third Quarter 1 / 2”, and “Mode”.

[0068] 1.5. For category features such as "protocol identifier," "protocol modifier," "account identifier," and "currency code," as well as the aforementioned time difference and binning features, grouping and "listing" are performed based on "customer number." The aggregated dataset is encoded using TfidfVectorizer and CountVectorizer, which transform the data from the perspectives of information importance and frequency, respectively. Finally, based on TruncatedSVD, the encoded features are compressed to two dimensions, thereby significantly reducing the data dimensionality and improving model efficiency while maintaining information integrity.

[0069] 1.6 Integrate all corporate deposit account balance information that has undergone the above processing and feature construction into a comprehensive feature table, using "customer number" as the key.

[0070] 2) Processing information on transactions in customers' corporate deposit accounts, including: 2.1 The "monthly average debit transaction amount" is obtained by dividing the "monthly cumulative debit transaction amount" by the "monthly cumulative number of debit transactions". Similarly, the "monthly average credit transaction amount" is obtained by dividing the "monthly cumulative credit transaction amount" by the "monthly cumulative number of credit transactions".

[0071] The "Monthly Cumulative Transaction Amount" is obtained by merging the "Monthly Cumulative Credit Transaction Amount" and the "Monthly Cumulative Debit Transaction Amount". The "Monthly Cumulative Transaction Count" is obtained by adding the "Monthly Cumulative Credit Transaction Count" and the "Monthly Cumulative Debit Transaction Count".

[0072] The "Monthly Cumulative Credit / Debit Transaction Difference" is calculated by subtracting the "Monthly Cumulative Credit Transaction Count" from the "Monthly Cumulative Debit Transaction Count". The "Monthly Average Transaction Amount" is then calculated by dividing the "Monthly Cumulative Transaction Amount Difference" by the "Monthly Cumulative Transaction Count". Finally, the "Monthly Average Transaction Amount Difference_1" is obtained by dividing the "Monthly Cumulative Transaction Amount Difference" by the "Monthly Cumulative Credit / Debit Transaction Difference" to obtain the "Monthly Average Transaction Amount Difference_2".

[0073] 2.2 Convert the "Data Date" to "datetime64" type and extract the year and month information to form time-dimensional features. Then, by calculating the difference with the selected time point, generate new features that reflect the time characteristics of transaction behavior.

[0074] 2.3. The features constructed above are binned, and a specific numerical range [-1e13, -1e12, -1e11, ..., 1e13] is used as the bin boundaries to convert continuous transaction amounts and transaction counts into discrete feature categories.

[0075] 2.4 Group the transaction information and derivative features of corporate deposit accounts according to "customer number" and aggregate customer-level transaction information by calculating the "number of unique values", "last value" and "first value" for each customer.

[0076] 2.5 For the transaction information of corporate deposit accounts and the numerical indicators in the above-constructed features, further group and aggregate them according to "customer number". The aggregation strategy includes calculating the "maximum", "minimum", "median", "mean", "standard deviation", "mean absolute deviation", "skewness", "quarter quantile", "third quartile", and "mode" of each group of features to determine the statistical characteristics of customer transaction behavior.

[0077] 2.6. After grouping categorical features such as "protocol identifier" and "protocol modifier" and time-dimensional information according to "customer number," a "list" aggregation method is used to convert each group of features into a list format. Subsequently, TfidfVectorizer and CountVectorizer encoding techniques are used to encode the aggregated data. The encoded features are then dimensionality-reduced using the TruncatedSVD model, compressing them into two-dimensional features.

[0078] 2.7. All the processed and constructed characteristics of customer corporate deposit accounts are integrated according to "customer number" to form a summary table that comprehensively reflects the customer's transaction behavior and time characteristics.

[0079] 3) Processing basic customer information, including: 3.1 The "Data Date" field in the company's customer basic information was converted to the "datetime64" date and time format to ensure consistency and ease of calculation. Next, the specific year and month information was extracted from this standardized date. A time difference feature was constructed by calculating the difference between the "Data Date" and a predefined point in time.

[0080] Extract the first four digits of "Business License Expiry Date" and "Inception Date" to use as year information, then calculate the difference between the two to generate the feature "Business License Validity Period_Years".

[0081] 3.2 Only the last six characters of the "Customer Type Code" should be retained.

[0082] 3.3. Adopt an encoding strategy to map the "Tax Registration Status" feature to numbers, convert "KH111152" to the number 2, "KH111151" to the number 1, and encode the remaining statuses as 0.

[0083] 3.4 Based on the construction rules of the business license number, the "business license number" was subjected to feature derivation, from which multi-dimensional features such as number length, issuing agency category, administrative division code, province and city information, and registration management department were extracted.

[0084] 3.5 For numerical features such as "registered capital", "number of employees", "total assets at the end of last year", and "(three-year average) annual sales", a binning strategy is adopted to transform continuous values ​​into hierarchical categories. This process generates new binned features by defining a series of boundary values ​​[-1e13, -1e12, -1e11, ..., 1e13].

[0085] 3.6. Group the company's basic information and all the aforementioned derived features according to "customer number", and then integrate the information using the "last value" aggregation principle.

[0086] 4) Processing customer business registration information, including: 4.1 The difference between "Registered Capital Amount" and "Paid-in Capital Amount" in the basic information of the enterprise's business license is calculated, and the feature "Registered_Paid-in_Difference" is generated.

[0087] Text analysis was performed on the "Description of Business Scope of the Organization," and non-entity content such as periods and parentheses was removed to extract the core description of the business scope. Missing values ​​in the "Industry Type Code" were filled with 0.

[0088] 4.2 Outlier and missing values ​​were handled for the "Administrative Division Code of the Unit". By analyzing the composition of the administrative division code, the first two and first four digits of the "Administrative Division Code of the Unit" were extracted, thereby generating two new features: "Business License_Administrative Division Code_Province" and "Business License_Administrative Division Code_City" respectively.

[0089] 4.3. Convert the format of the four time fields, namely "Start Date (P9 control field)," "Organization Establishment Date," "End Date (P9 control field)," and "End Date of Operating Period," to a unified "datetime64" type. Extract the year information for each date.

[0090] Based on the extracted year information, the time difference between each enterprise and a certain benchmark time point is further calculated to generate time difference features. At the same time, by subtracting each pair of times, six derived features are constructed: "operating year", "operating year already completed", "remaining operating year", and the corresponding "operating days", "operating days already completed", and "remaining operating days".

[0091] 4.4 A binary coding strategy was adopted for the "old business license number". The specific code "qyhm0d58c94ec918e6f366e6db99c190533a7e" was mapped to 0, and all other numbers were mapped to 1, thereby determining whether the company used the old business license.

[0092] 4.5 For key monetary features such as "Registered Capital Amount" and "Paid-in Capital Amount", as well as the date difference features generated above ("Date Difference", "Business Days", "Business Days Already Operated", "Remaining Business Days"), a binning strategy is adopted to divide these feature values ​​into a series of predefined intervals, generating new binning features with intervals from [-1e13, -1e12, -1e11, ..., 1e13].

[0093] 5) Process customer credit summary information, including: 5.1. Group the "Data Date (P9 control field)" in the enterprise information summary information according to the "Enterprise Information Identifier". Then, perform an aggregation operation on the grouped "Data Dates", record the "Number of Unique Values" of "Data Date" in each enterprise's information to determine the update frequency of the credit report, and save the "First Value" to determine the earliest credit record time of the enterprise.

[0094] 5.2. For the "Data Date (P9 control field)" and its first occurrence date "Data Date (P9 control field)_first", convert them to a unified "datetime64" format. Next, calculate the difference between the "Data Date (P9 control field)" and the preset base time point to generate a time difference feature.

[0095] Subtracting “Data Date (P9 Control Field)” from “Data Date (P9 Control Field)_first” yields “Data Date (P9 Difference)”. In addition, the “First Repayment Year” feature is generated by calculating the difference between “Year of First Repayment Responsibility” and “Year of First Credit Transaction”.

[0096] 5.3. Conduct a comprehensive statistical analysis of the records of violations, and add the "Number of Administrative Penalty Records" and the "Number of Enforcement Records" together to obtain the "Total Number of Administrative Penalty Records - Enforcement Records".

[0097] 6) Processing basic information on customers' housing provident fund contribution records, including: 6.1 The "Housing Provident Fund Contribution Base", "Number of Employees", and "Accumulated Amount of Arrears" in the second-generation enterprise housing provident fund contribution record basic information table are processed. Through a binning strategy, these continuous numerical features are divided into an interval defined by a series of boundary values ​​[-1e13, -1e12, -1e11, ..., 1e13].

[0098] 6.2. Convert the "Data Date (P9 control field)" and "Most Recent Payment Date" into a consistent "datetime64" date format; then, subtract them from a specific date to obtain a new feature. By calculating the time difference between the "Start Date (P9 control field)" and the "Most Recent Payment Date," the feature "Start_Most Recent Payment_Difference" is derived. The time attributes such as "Initial Payment Month / Year," "Housing Provident Fund Payment Completed Month / Year," and "Statistical Month / Year" are truncated, retaining only the year. By calculating the year difference between "Housing Provident Fund Payment Completed Month / Year" and "Initial Payment Month / Year," the feature "Housing Provident Fund Payment Completed Month / Year" is determined.

[0099] 7) Processing client legal information, including: 7.1 Convert the "Data Date (P9 control field)" and "Case Closure Date" into a unified "datetime64" format, and then extract the year information of these dates. Further, calculate the year difference ("Data Date (P9 control field)_Case Closure_Date Difference_year") and day difference ("Data Date (P9 control field)_Case Closure_Date Difference_days") between the "Data Date (P9 control field)" and the "Case Closure" by comparing them with the selected baseline.

[0100] 7.2. Using the number 2 as a separator, the "Remarks Description" field is divided into two parts. Based on the composition of "Name of Administrative Penalty Authority," it is separated into provincial, municipal, and autonomous region levels, from which the feature "Name of Administrative Penalty Authority_Province" is extracted.

[0101] 7.3 Divide “Case Amount”, “P9_Case Closure_Date Difference_days”, “Data Date (P9 Control Field)”, and “Case Closure Date” into multiple categories based on a specific range [-1e13, -1e12, -1e11, ..., 1e13].

[0102] 7.4. Based on the “Participant Name”, group the “Judgment Execution Case Number”, “KEYID”, “Administrative Penalty Authority Name” and their derived features, and then aggregate them according to the “Number of Unique Values” in each group.

[0103] 7.5. For the "Case Amount" and the time difference features derived from the date (e.g., "Data Date (P9 control field)", "Case Amount", "Case Closure Date", "Data Date (P9 control field)_year", "Case Closure Date_year", "P9_Case Closure_Date Difference_year", "P9_Case Closure_Date Difference_days", "Case Amount_cut", "P9_Case Closure_Date Difference_days_cut", "Data Date (P9 control field)_cut", "Case Closure Date_cut"), group them according to "Participant Name" and further perform statistical aggregation, including calculating "Maximum", "Minimum", "Median", "Mean", "Standard Deviation", "Mean Absolute Deviation", "Skew", "Quarter 1 / 2", "Third 1 / 2", and "Mode").

[0104] 7.6. For textual features in judicial information (e.g., "judgment execution case number", "KEYID", "name of administrative penalty authority", "description of main illegal facts of administrative penalty", "case closure date", "description of litigation subject matter", "remarks description", "data date (P9 control field)_year", "case closure date_year", "P9_case closure_date difference_year", "P9_case closure_date difference_days", "name of administrative penalty authority_province", "case subject matter amount_cut", "P9_case closure_date difference_days_cut", "data date (P9 control field)_cut", "case closure date_cut"), the data is first grouped according to "participant name", and then the information is structured using a "list" aggregation method. Then, the aggregated textual data is encoded using TfidfVectorizer and CountVectorizer encoding techniques. The encoded features are further dimensionality-reduced using the TruncatedSVD model, compressed into two-dimensional features.

[0105] 7.7 Integrate all processed and constructed civil and commercial judgment documents according to the "name of the participant" to form a summary table reflecting the client's judicial information.

[0106] 8) Process customer tax payment information, including: 8.1 The "Total Output Value Difference" feature is generated by calculating the difference between "Total Output Value Amount of the Previous Year" and "Total Output Value Amount of the Year Before Last Year".

[0107] 8.2. Convert the format of the timestamp in the "Data Date (P9 control field)" field to ensure all date information is uniformly of type "datetime64". Then, extract the year and month information. Generate a time difference feature by comparing the difference between the "Data Date" and the preset base date.

[0108] 8.3 For monetary features such as "total output value of the previous year" and "total output value of the year before last", a binning strategy is adopted to divide the continuous numerical range into a series of discrete bins. The binning range is from [-1e13, -1e12, -1e11, ..., 1e13].

[0109] 8.4. Based on the "Unified Social Credit Code", group the external tax details and previously derived characteristics of micro and small enterprises, and aggregate the statistics of "number of unique values" and "first value".

[0110] 8.5 For “sales revenue amount”, a numerical statistical strategy was adopted, including calculating “maximum value”, “minimum value”, “median”, “mean”, and “standard deviation”.

[0111] 8.6. For the category feature of "tax rate status code", perform mapping encoding, for example, map "27001" to the number 1, "28001" to 2, and "28003" to the number 3.

[0112] 9) Process the customer master table list and risk tags, including: 9.1 Integrate the contract-related information of the micro-loan quick loan (with Y-tag) with the data from the eight previously processed topics. Perform left join or inner join operations based on their respective primary keys (such as customer number, contract number, etc.) to finally form a master table containing all necessary information and derived features.

[0113] 9.2 For the "Account Opening Date" and "Data Date" fields in the master table, convert these two date fields to the "datetime64" type. Next, extract the monthly dimension information, and then generate the "Account Opening_Data_Date Difference" feature by calculating the time difference between "Data Date" and "Account Opening Date".

[0114] 9.3 In the data cleaning stage, several features in the master table were filtered. For example, features containing only a single value or with a large number of missing values ​​were deleted. For instance, features such as "cust_tax registration status" and "doc_judgment execution case number_cv_1_word" were deleted because they were redundant or incomplete and unlikely to contribute to model training.

[0115] 9.4. Processing categorical features, such as "cust_customer type code", "cust_department code", and "cust_industry category code". This is achieved through digitization using TfidfVectorizer and CountVectorizer encoding techniques, and dimensionality reduction using TruncatedSVD technology, compressing the features to two dimensions.

[0116] 9.5. Recombine the above features into a complete feature table.

[0117] 9.6. In the total feature set, identify sparse features with high dimensionality and low frequency of occurrence. By applying the TruncatedSVD model, the sparse features are compressed to retain only the four principal components.

[0118] Step 3: Initial model building and label preprocessing; Step 31: Preprocess the three types of labels contained in the main customer table. In the original labeling system, target=1 refers to defaulting customers at level M2 (i.e., bad customers), target=0 indicates normal loan status (good customers), and target=2 marks minor default situations between M0 and M1 (gray customers). To adapt to the model training requirements, a label relabeling strategy is adopted to adjust all gray customers (target=2) to bad customers (target=1).

[0119] Step 32: The overall dataset was divided using 5-fold cross-validation, resulting in 5 balanced subsets.

[0120] Step 33: In each modeling process, randomly select 4 sets as the training set and the remaining set as the validation set.

[0121] In this embodiment, XGBoost is selected as the model framework, and the AUC metric is used to evaluate the model. During model training, binary_logloss is used as the loss function.

[0122] Through the above steps, five XGBoost models with different parameters are constructed.

[0123] Step 4: Transfer learning.

[0124] Data previously labeled as "gray market customers" (originally target=2) was removed, leaving only data with two clearly labeled classes as training data. Using this filtered training dataset, the five initially constructed XGBoost models were retrained, with fine-tuning of the model parameters. The five XGBoost models, fine-tuned through transfer learning, were then saved, forming a complete prediction system.

[0125] Step 5: When there is an actual forecasting need, the new data is preprocessed and features are derived through the feature engineering process described above.

[0126] Step 6: Input the preprocessed prediction data into the five XGBoost models saved in Step 4. Each model independently completes the prediction and outputs the predicted value of default risk. Finally, the prediction results of these models are averaged.

[0127] This application embodiment constructs an efficient and accurate default risk assessment system for micro and small enterprise quick loans through label preprocessing, fine division of the dataset, introduction of the XGBoost model, application of the AUC index, personalized model tuning, feature adaptation of the predicted data, and fusion of prediction results from multiple models, effectively managing default risk.

[0128] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods according to the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the related technology, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0129] This embodiment also provides a probability information determination device, which is used to implement the above embodiments and preferred embodiments; details already described will not be repeated. As used below, the term "module" can be a combination of software and / or hardware that implements a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0130] Figure 3 This is a structural block diagram of a probability information determination device according to an embodiment of this application, such as... Figure 3 As shown, the device includes: The first determining module 32 is used to determine the type of each piece of first information of the target object, and to determine the preprocessing method of each piece of first information according to the type of each piece of first information, wherein the first information is information used to determine the probability that the target object is in a target state, and the preprocessing method includes at least one of the following: binning, aggregation, standardization, encoding, compression, and derivation. The preprocessing module 34 is used to preprocess each piece of first information according to the preprocessing method of each piece of first information to obtain preprocessed second information; The second determining module 36 is used to input multiple pieces of second information into multiple prediction models, so that the multiple prediction models determine the evaluation value of the target object in the target state based on the multiple pieces of second information; Calculation module 38 is used to call a calculation function and calculate multiple evaluation values ​​through the calculation function to determine the target evaluation value of the target object in the target state; The third determining module 40 is used to determine the probability information of the target object being in the target state based on the target evaluation value.

[0131] Using the aforementioned apparatus, firstly, all raw information (i.e., first information) is identified and classified. Based on information attributes, it is preprocessed using methods such as binning, aggregation, standardization, encoding, compression, or derivation to generate structured features (i.e., second information). Subsequently, the second information is input into a set of multivariate prediction models to calculate the assessment value of the default state. Finally, the assessment values ​​from multiple models are fused to obtain the probability of being in a certain state. This embodiment improves data quality through different preprocessing methods, utilizes multi-model assessment and result fusion, and examines risk from different perspectives, significantly improving the accuracy of risk assessment. Therefore, it can solve the problem in related technologies where assessing the probability of an object being in a certain state often relies on expert experience, leading to low accuracy.

[0132] In an exemplary embodiment, the preprocessing module is configured to: firstly, if the type of the first information is a continuous number type, perform binning processing on the first information to determine the binning category of the first information; aggregate the first information according to the binning category to obtain aggregated first information; encode the first information based on a first encoding method to obtain second information, and encode the first information based on a second encoding method to obtain third information; and compress the second information and the third information to obtain compressed second information and compressed third information, wherein the preprocessed second information includes at least the compressed second information and the compressed third information.

[0133] In an exemplary embodiment, the preprocessing module is configured to aggregate the first information to obtain aggregated fourth information if the type of the first information is a discrete number type or a category type; encode the fourth information based on a first encoding method to obtain fifth information, and encode the fourth information based on a second encoding method to obtain sixth information; and compress the fifth information and the sixth information to obtain compressed fifth information and compressed sixth information, wherein the preprocessed second information includes at least the compressed fifth information and the compressed sixth information.

[0134] In an exemplary embodiment, a preprocessing module is configured to determine the associated features of each piece of first information; and to determine the feature value corresponding to the associated feature based on each piece of first information, wherein the preprocessed second information includes the feature value corresponding to the associated feature.

[0135] In one exemplary embodiment, the apparatus further includes: a training module, configured to acquire a dataset of multiple category labels, and adjust the original category labels of the dataset of multiple category labels to a first category label or a second category label respectively; train multiple initial prediction models based on a first dataset of the first category labels and a second dataset of the second category labels to obtain multiple intermediate prediction models; and determine the multiple prediction models based on the multiple intermediate prediction models.

[0136] In an exemplary embodiment, the training module is configured to divide a first dataset of the first category label into X first subsets and a second dataset of the second category label into X second subsets; train the plurality of initial prediction models based on X-1 first subsets and X-1 second subsets, and validate the trained prediction models based on other first subsets and other second subsets to obtain validation results, wherein the X-1 first subsets and X-1 second subsets used to train different initial prediction models are different, the other first subsets are the first datasets other than the X-1 first subsets among the X first subsets, and the other second subsets are the second datasets other than the X-1 second subsets among the X second subsets; and determine whether the trained prediction model is the intermediate prediction model based on the validation results.

[0137] In one exemplary embodiment, a training module is configured to acquire a third dataset with the original category label as the first category label and a fourth dataset with the original category label as the second category label; and to train the intermediate prediction model based on the third dataset and the fourth dataset to determine the plurality of prediction models.

[0138] It should be noted that the above modules can be implemented by software or hardware. For the latter, they can be implemented in the following ways, but are not limited to: all the above modules are located in the same processor; or, the above modules are located in different processors in any combination.

[0139] Embodiments of this application also provide a computer-readable storage medium storing a computer program, wherein the computer program is configured to execute the steps in any of the above method embodiments when run.

[0140] Optionally, in this embodiment, the storage medium may be configured to store program code for performing the following steps: S1, determine the type of each piece of first information of the target object, and determine the preprocessing method of each piece of first information according to the type of each piece of first information, wherein the first information is information used to determine the probability that the target object is in a target state, and the preprocessing method includes at least one of the following: binning, aggregation, standardization, encoding, compression, and derivation. S2, each piece of first information is preprocessed according to the preprocessing method of each piece of first information to obtain preprocessed second information; S3, inputting multiple pieces of second information into multiple prediction models, so that the multiple prediction models determine the evaluation value of the target object in the target state based on the multiple pieces of second information; S4, Invoke the calculation function, and calculate multiple evaluation values ​​through the calculation function to determine the target evaluation value of the target object in the target state; S5, determine the probability information of the target object being in the target state based on the target evaluation value.

[0141] In one exemplary embodiment, the aforementioned computer-readable storage medium may include, but is not limited to, various media capable of storing computer programs, such as a USB flash drive, read-only memory (ROM), random access memory (RAM), portable hard disk, magnetic disk, or optical disk.

[0142] Embodiments of this application also provide an electronic device, including a memory and a processor, wherein the memory stores a computer program and the processor is configured to run the computer program to perform the steps in any of the above method embodiments.

[0143] In one exemplary embodiment, the electronic device may further include a transmission device and an input / output device, wherein the transmission device is connected to the processor and the input / output device is connected to the processor.

[0144] Optionally, in this embodiment, the processor can be configured to perform the following steps via a computer program: S1, determine the type of each piece of first information of the target object, and determine the preprocessing method of each piece of first information according to the type of each piece of first information, wherein the first information is information used to determine the probability that the target object is in a target state, and the preprocessing method includes at least one of the following: binning, aggregation, standardization, encoding, compression, and derivation. S2, each piece of first information is preprocessed according to the preprocessing method of each piece of first information to obtain preprocessed second information; S3, inputting multiple pieces of second information into multiple prediction models, so that the multiple prediction models determine the evaluation value of the target object in the target state based on the multiple pieces of second information; S4, Invoke the calculation function, and calculate multiple evaluation values ​​through the calculation function to determine the target evaluation value of the target object in the target state; S5, determine the probability information of the target object being in the target state based on the target evaluation value.

[0145] Embodiments of this application also provide a computer program product, which includes a computer program that, when executed by a processor, implements the steps in any of the above method embodiments.

[0146] Embodiments of this application also provide another computer program product, including a non-volatile computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps in any of the above method embodiments.

[0147] Embodiments of this application also provide a computer program that includes computer instructions stored in a computer-readable storage medium; a processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the steps in any of the above method embodiments.

[0148] Optionally, in this embodiment, the processor can be configured to perform the following steps via a computer program: S1, determine the type of each piece of first information of the target object, and determine the preprocessing method of each piece of first information according to the type of each piece of first information, wherein the first information is information used to determine the probability that the target object is in a target state, and the preprocessing method includes at least one of the following: binning, aggregation, standardization, encoding, compression, and derivation. S2, each piece of first information is preprocessed according to the preprocessing method of each piece of first information to obtain preprocessed second information; S3, inputting multiple pieces of second information into multiple prediction models, so that the multiple prediction models determine the evaluation value of the target object in the target state based on the multiple pieces of second information; S4, Invoke the calculation function, and calculate multiple evaluation values ​​through the calculation function to determine the target evaluation value of the target object in the target state; S5, determine the probability information of the target object being in the target state based on the target evaluation value.

[0149] Specific examples in this embodiment can be found in the examples described in the above embodiments and exemplary implementations, and will not be repeated here.

[0150] Obviously, those skilled in the art should understand that the modules or steps of this application described above can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. They can be implemented using computer-executable program code, and thus can be stored in a storage device for execution by a computing device. In some cases, the steps shown or described can be performed in a different order than those presented here, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, this application is not limited to any particular combination of hardware and software.

[0151] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the principles of this application should be included within the protection scope of this application.

Claims

1. A method for determining probability information, characterized in that, Applied to computer equipment, including: The type of each piece of first information of the target object is determined, and the preprocessing method of each piece of first information is determined according to the type of each piece of first information, wherein the first information is information used to determine the probability that the target object is in a target state, and the preprocessing method includes at least one of the following: binning, aggregation, standardization, encoding, compression, and derivation. Each piece of first information is preprocessed according to the preprocessing method of each piece of first information to obtain preprocessed second information; Multiple pieces of second information are input into multiple prediction models, so that the multiple prediction models determine the evaluation value of the target object in the target state based on the multiple pieces of second information; A calculation function is invoked, and multiple evaluation values ​​are calculated using the calculation function to determine the target evaluation value of the target object in the target state; The probability information of the target object being in the target state is determined based on the target evaluation value.

2. The method according to claim 1, characterized in that, Each piece of first information is preprocessed according to the preprocessing method for each piece of first information to obtain preprocessed second information, including: If the type of the first information is a continuous number type, the first information is binned to determine the binning category of the first information; Aggregate the bins according to the bin categories of the first information to obtain the aggregated first information; The first information is encoded using a first encoding method to obtain the second information, and the first information is encoded using a second encoding method to obtain the third information; The second information and the third information are compressed to obtain compressed second information and compressed third information, wherein the preprocessed second information includes at least the compressed second information and the compressed third information.

3. The method according to claim 1, characterized in that, Each piece of first information is preprocessed according to the preprocessing method for each piece of first information to obtain preprocessed second information, including: If the type of the first information is a discrete number type or a category type, the first information is aggregated to obtain the aggregated fourth information; The fourth information is encoded using the first encoding method to obtain the fifth information, and the fourth information is encoded using the second encoding method to obtain the sixth information; The fifth and sixth information are compressed to obtain compressed fifth and sixth information, wherein the preprocessed second information includes at least the compressed fifth and sixth information.

4. The method according to claim 1, characterized in that, Each piece of first information is preprocessed according to the preprocessing method for each piece of first information to obtain preprocessed second information, including: Determine the associated features of each piece of first information; The feature value corresponding to the associated feature is determined based on each piece of first information, wherein the preprocessed second information includes the feature value corresponding to the associated feature.

5. The method according to claim 1, characterized in that, Before inputting multiple pieces of second information into multiple prediction models, the method further includes: Obtain a dataset with multiple category labels, and adjust the original category labels of the datasets with multiple category labels to either the first category label or the second category label; Multiple initial prediction models are trained based on a first dataset with the first category label and a second dataset with the second category label to obtain multiple intermediate prediction models; The plurality of prediction models are determined based on the plurality of intermediate prediction models.

6. The method according to claim 5, characterized in that, Before training multiple initial prediction models based on a first dataset with the first category label and a second dataset with the second category label to obtain multiple intermediate prediction models, the method further includes: The first dataset of the first category label is divided into X first subsets, and the second dataset of the second category label is divided into X second subsets, where X is a positive integer; The multiple initial prediction models are trained based on X-1 first subsets and X-1 second subsets, and the trained prediction models are validated based on other first subsets and other second subsets to obtain validation results. The X-1 first subsets and X-1 second subsets used to train different initial prediction models are different. The other first subsets are the first datasets other than the X-1 first subsets among the X first subsets, and the other second subsets are the second datasets other than the X-1 second subsets among the X second subsets. Based on the verification results, determine whether the trained prediction model is the intermediate prediction model.

7. The method according to claim 5, characterized in that, The method further includes determining the plurality of prediction models based on the plurality of intermediate prediction models: Obtain the third dataset with the original category label as the first category label and the fourth dataset with the original category label as the second category label; The intermediate prediction model is trained based on the third dataset and the fourth dataset to determine the plurality of prediction models.

8. A device for determining probability information, characterized in that, Applied to computer equipment, including: The first determining module is used to determine the type of each piece of first information of the target object, and to determine the preprocessing method of each piece of first information according to the type of each piece of first information, wherein the first information is information used to determine the probability that the target object is in a target state, and the preprocessing method includes at least one of the following: binning, aggregation, standardization, encoding, compression, and derivation. A preprocessing module is used to preprocess each piece of first information according to the preprocessing method of each piece of first information to obtain preprocessed second information; The second determining module is used to input multiple pieces of second information into multiple prediction models, so that the multiple prediction models determine the evaluation value of the target object in the target state based on the multiple pieces of second information; The calculation module is used to call a calculation function and calculate multiple evaluation values ​​through the calculation function to determine the target evaluation value of the target object in the target state; The third determining module is used to determine the probability information of the target object being in the target state based on the target evaluation value.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored program, wherein the program, when executed, performs the method of any one of claims 1 to 7.

10. An electronic device comprising a memory and a processor, characterized in that, The memory stores a computer program, and the processor is configured to execute the method of any one of claims 1 to 7 through the computer program.

Images