A method for building a customer manager risk model for small sample scenarios

Abstract

The present application relates to a method for constructing a customer manager risk model for a small sample scenario. The method includes: selecting customer managers who have a record of filing a case before a risk point and meet a performance period; then, extracting sample information of these customer managers to form a training sample set; then, extracting customer manager features from the sample set, performing distinguishable enhancement processing on the intrinsic features, obtaining an enhanced feature set, and training a basic risk model for the customer manager based on the enhanced feature set; finally, identifying abnormal sample points from the training sample set, and retraining the basic risk model in combination with these sample points and their features to obtain an enhanced risk model. In this way, it is possible to achieve the effect of accurately constructing and optimizing the customer manager risk assessment model in a small sample data environment.

Description

Technical Field

[0001] The present application relates to the field of credit risk control, and in particular, to a method for constructing a customer manager risk model for a small sample scenario. Background Art

[0002] In the financial services industry, account managers, as an important role connecting financial institutions and customers, play a key role in business promotion, customer service, and risk management. However, due to the complexity and uncertainty of the financial market, the behavior of account managers may bring certain risks, such as operational risk and moral risk. In order to effectively manage and prevent these risks, it is crucial to build a model that can assess the risk level of account managers. However, for small sample scenarios, that is, when the amount of available data is limited, traditional statistical methods or machine learning algorithms may not provide sufficient predictive performance because they usually require a large amount of data to train a reliable model. In addition, the small sample problem may also lead to overfitting, resulting in poor generalization of the model on new data. Not only that, compared with the model construction of the customer dimension, the account manager risk model is different in the selection of observation points and the construction of features in the label definition. In addition, the target samples are mainly manually labeled, so the target samples are missing and small.

[0003] Therefore, an optimized solution for building a client manager risk model for small sample scenarios is desired. Summary of the invention

[0004] This summary is provided to introduce concepts in a brief form that will be described in detail in the detailed description below. This summary is not intended to identify key features or essential features of the claimed technical solution, nor is it intended to limit the scope of the claimed technical solution.

[0005] In a first aspect, the present application provides a method for constructing a customer manager risk model for a small sample scenario, the method comprising:

[0006] Select a time point involving risk, and select target account managers who have received applications within a predetermined period of time before the time point involving risk and who meet the performance period;

[0007] Extracting sample information of the target account manager to obtain a training sample set;

[0008] Extracting the account manager features of the sample information of each target account manager in the training sample set to obtain a set of account manager features;

[0009] Performing intrinsic feature distinguishable enhancement processing on the set of customer manager features to obtain a set of enhanced customer manager features;

[0010] Based on the set of enhanced account manager features, the account manager risk model is trained to obtain the account manager basic risk model:

[0011] Identifying abnormal sample points from the training sample set;

[0012] The account manager basic risk model is trained based on the abnormal sample points and the account manager characteristics of the abnormal sample points to obtain the account manager enhanced risk model.

[0013] Optionally, the sample information of the target account manager includes basic attribute information, business location abnormality information, customer relationship information and historical transaction behavior record information.

[0014] Optionally, extracting account manager features of sample information of each target account manager in the training sample set to obtain a set of account manager features, including: using a account manager sample embedding matrix to respectively embed code the sample information of each target account manager in the training sample set to obtain a set of target account manager embedded semantic coding vectors as the set of account manager features.

[0015] Optionally, the set of account manager features is subjected to intrinsic feature distinguishable enhancement processing to obtain a set of enhanced account manager features, including: using a field mapper to perform field mapping on each target account manager embedded semantic coding vector in the set of target account manager embedded semantic coding vectors to obtain a set of entry target account manager embedded semantic coding vectors; calculating the essential features of the set of entry target account manager embedded semantic coding vectors to obtain a target account manager semantic field essential feature vector; based on the target account manager semantic field essential feature vector, semantically enhancing the set of target account manager embedded semantic coding vectors to obtain a set of enhanced target account manager embedded semantic coding vectors as the set of enhanced account manager features.

[0016] Optionally, calculating the essential characteristics of the set of embedded semantic coding vectors of the entry target account managers to obtain the target account manager semantic field essential characteristic vector, including: calculating the field depth factor of each entry target account manager embedded semantic coding vector in the set of embedded semantic coding vectors of the entry target account managers to obtain a set of target account manager semantic field depth factors; based on the set of target account manager semantic field depth factors, calculating the essential characteristics of the set of embedded semantic coding vectors of the entry target account managers to obtain the target account manager semantic field essential characteristic vector.

[0017] Optionally, the field depth factor of each entry target account manager embedded semantic coding vector in the set of entry target account manager embedded semantic coding vectors is calculated to obtain a set of target account manager semantic field depth factors, including: calculating the square of the one-norm of the entry target account manager embedded semantic coding vector to obtain the one-norm square representation of the entry target account manager embedded semantics; calculating the sum and difference between a constant one and the one-norm square representation of the entry target account manager embedded semantics to obtain the entry target account manager embedded semantics sum representation and the entry target account manager embedded semantics difference representation; calculating the division between the entry target account manager embedded semantics sum representation and the entry target account manager embedded semantics difference representation, and then calculating the logarithmic function value with base 2 of the obtained division value to obtain the target account manager semantic field depth factor.

[0018] Optionally, based on the target account manager semantic field essential feature vector, the set of target account manager embedded semantic coding vectors is semantically enhanced to obtain a set of enhanced target account manager embedded semantic coding vectors as the set of enhanced account manager features, including: taking the target account manager semantic field essential feature vector as a conditional feature vector, and performing distinguishable enhancement on each target account manager embedded semantic coding vector in the set of target account manager embedded semantic coding vectors to obtain the set of enhanced target account manager embedded semantic coding vectors.

[0019] Optionally, taking the target account manager's semantic field essential feature vector as a conditional feature vector, each target account manager embedded semantic coding vector in the set of the target account manager embedded semantic coding vector is distinguishably enhanced to obtain the set of enhanced target account manager embedded semantic coding vectors, including: calculating the multiplication between the target account manager's semantic field essential feature vector and the target account manager's semantic essential weight matrix to obtain the target account manager's semantic essential feature modulation vector; calculating the multiplication between the target account manager's embedded semantic coding vector and the local target account manager's semantic weight matrix to obtain the target account manager's embedded semantic feature modulation vector; calculating the positional sum of the target account manager's semantic essential feature modulation vector and the target account manager's embedded semantic feature modulation vector, and then performing positional summation with a bias vector to obtain a target account manager's semantic comprehensive modulation representation vector; after probabilistically modulating the target account manager's semantic comprehensive modulation representation vector using a sigmoid function, positionally point multiplying the modulated target account manager's semantic comprehensive modulation weight vector with the target account manager's embedded semantic coding vector to obtain an enhanced target account manager embedded semantic coding vector.

[0020] Optionally, the account manager basic risk model is trained based on the abnormal sample points and the account manager characteristics of the abnormal sample points to obtain the account manager enhanced risk model, including: based on the abnormal sample points and the account manager characteristics of the abnormal sample points, the account manager basic risk model is incrementally learned and trained to obtain the account manager enhanced risk model.

[0021] The above technical solution is adopted, by selecting the target account managers who have submitted applications and meet the performance period before and before the risk point, and then extracting the sample information of the target account manager as the training sample set, wherein the sample information contains basic attribute information, abnormal information of business locations, customer relationship information and historical transaction behavior record information, and extracting the account manager features in these sample information, and then the internal feature distinguishable reinforcement processing is aimed at improving the quantity and quality of effective features extracted from limited data, so as to train the account manager basic risk model. This can enhance the relevant information features of those account managers and the features that have stronger explanatory power for risks, thereby improving the learning effect of the model. In addition, the solution also includes an abnormal sample point identification step to further optimize the account manager basic risk model. By using incremental learning to learn abnormal situations, the model can better understand the behavior patterns of account managers under abnormal circumstances, and adjust its risk assessment strategy accordingly, and finally form a more robust account manager reinforcement risk model for small sample scenarios.

[0022] Other features and advantages of the present application will be described in detail in the subsequent detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] The above and other features, advantages and aspects of the embodiments of the present application will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings. Throughout the drawings, the same or similar reference numerals represent the same or similar elements. It should be understood that the drawings are schematic and the originals and elements are not necessarily drawn to scale. In the drawings:

[0024] Figure 1 It is a flowchart of a method for constructing a customer manager risk model applied to a small sample scenario according to an exemplary embodiment.

[0025] Figure 2 It is a flowchart of step S140 of a method for building a customer manager risk model applied to a small sample scenario according to an exemplary embodiment.

[0026] Figure 3 It is a flowchart of step S142 of a method for building a client manager risk model applied to a small sample scenario according to an exemplary embodiment.

[0027] Figure 4 It is a block diagram of a system for building a customer manager risk model applied to a small sample scenario according to an exemplary embodiment. DETAILED DESCRIPTION

[0028] The embodiments of the present application will be described in more detail below with reference to the accompanying drawings. Although certain embodiments of the present application are shown in the accompanying drawings, it should be understood that the present application can be implemented in various forms and should not be construed as being limited to the embodiments described herein. Instead, these embodiments are provided to provide a more thorough and complete understanding of the present application. It should be understood that the drawings and embodiments of the present application are only for exemplary purposes and are not intended to limit the scope of protection of the present application.

[0029] It should be understood that the various steps described in the method implementation of the present application can be performed in different orders and / or performed in parallel. In addition, the method implementation may include additional steps and / or omit the steps shown. The scope of the present application is not limited in this respect.

[0030] The term "including" and its variations used herein are open inclusions, i.e., "including but not limited to". The term "based on" means "based at least in part on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments". The relevant definitions of other terms will be given in the following description.

[0031] It should be noted that the concepts such as "first" and "second" mentioned in this application are only used to distinguish different devices, modules or units, and are not used to limit the order or interdependence of the functions performed by these devices, modules or units.

[0032] It should be noted that the modifications of "one" and "plurality" mentioned in the present application are illustrative rather than restrictive, and those skilled in the art should understand that unless otherwise clearly indicated in the context, it should be understood as "one or more".

[0033] The names of the messages or information exchanged between multiple devices in the embodiments of the present application are only used for illustrative purposes and are not used to limit the scope of these messages or information.

[0034] In response to the above technical problems, in the technical solution of this application, a new method is proposed for the construction of a customer manager risk model in a small sample scenario. This method not only takes into account traditional risk factors (such as basic attribute information, abnormal information of business locations, customer relationship information, and historical transaction behavior record information), but also introduces unsupervised learning and incremental learning methods. It can fully utilize data to identify target samples, thereby helping to increase the model effect in small sample scenarios. Specifically, by using a combination of unsupervised and supervised methods to build a model, first use a supervised method to build a basic model, then use unsupervised and manually verified methods to find abnormal points to increase target samples, and finally use incremental learning to train the model after sample and feature supplementation. This method ensures the capture of existing black samples (such as manually labeled intermediary samples), and can also discover black samples of new patterns, making the model more effective.

[0035] The specific implementation of the present application is described in detail below with reference to the accompanying drawings.

[0036] Figure 1 is a flowchart of a method for constructing a customer manager risk model for a small sample scenario according to an exemplary embodiment. Figure 1 As shown, according to the method for constructing a customer manager risk model for a small sample scenario according to an embodiment of the present application, the method includes:

[0037] Step S110: Select a time point involving risk, and select a target account manager who has received a case within a predetermined period of time before the time point involving risk and who meets the performance period;

[0038] Step S120, extracting sample information of the target account manager to obtain a training sample set;

[0039] Step S130, extracting the account manager features of the sample information of each target account manager in the training sample set to obtain a set of account manager features;

[0040] Step S140, performing an internal feature distinguishable enhancement process on the set of customer manager features to obtain a set of enhanced customer manager features;

[0041] Step S150: Based on the set of enhanced account manager features, the account manager risk model is trained to obtain an account manager basic risk model:

[0042] Step S160, identifying abnormal sample points from the training sample set;

[0043] Step S170: training the account manager basic risk model based on the abnormal sample points and the account manager characteristics of the abnormal sample points to obtain an account manager enhanced risk model.

[0044] In the above-mentioned method for building a customer manager risk model applied to a small sample scenario, the step S110 selects a time point involving risk, and selects a target customer manager who has submitted applications within a predetermined time period before the time point involving risk and meets the performance period. The step S120 extracts the sample information of the target customer manager to obtain a training sample set. Specifically, first, select a time point involving risk, and select a target customer manager who has submitted applications within a predetermined time period before the time point involving risk and meets the performance period. In addition, extract the sample information of the target customer manager to obtain a training sample set, wherein the sample information of the target customer manager includes basic attribute information, business location abnormality information, customer relationship information, and historical transaction behavior record information.

[0045] Specifically, first, we identify risky time points through in-depth analysis of historical data. These time points usually correspond to specific risk events, such as financial fraud, default or abnormal transactions. These events are marked because they represent risky behaviors that account managers may be involved in. In order to identify these risk events, we collect historical data including transaction records, customer feedback and market changes, and analyze these data by using natural language processing technology and statistical methods to mark risk events on the timeline. This is the basis for building a risk model and helps focus on those moments that are most likely to affect risk assessment.

[0046] Then, in the predetermined time period before these risk time points, account managers with application records and meeting the performance period are selected as target account managers. The predetermined time period is set based on business needs and historical data analysis. For example, if risk events usually occur within one month after the transaction, the predetermined time period is set to one month. At the same time, verify whether these account managers meet the performance period requirements, that is, whether their service time in the financial institution is long enough to collect enough data to assess their risk level. In this way, it can ensure that the selected account manager sample is representative and relevant, providing a solid foundation for subsequent risk assessment.

[0047] In the process of selecting target account managers, their sample information is collected, including basic attribute information, business location abnormality information, customer relationship information and historical transaction behavior record information, to build a training sample set. Basic attribute information involves personal characteristics of account managers, such as name, age, gender, education background and professional qualifications, which can be obtained through the internal human resources system or customer management system. This information provides background information of account managers and helps to understand their basic situation and potential risk factors.

[0048] Abnormal information at business promotion points is identified by analyzing the business promotion points of account managers, such as frequent large transactions or unusual transaction patterns. These abnormal behaviors may indicate potential risks. Therefore, using data analysis techniques, such as cluster analysis and anomaly detection algorithms, to identify and mark these abnormal points is crucial to discovering and preventing potential risk behaviors.

[0049] Customer relationship information collects the relationship between the account manager and his or her customers, including the diversity of the customer group and the frequency of transactions, which can be achieved by analyzing transaction records and customer feedback. Understanding the customer network of the account manager helps to assess their risk transmission channels and influence. For example, if an account manager has frequent transactions with multiple high-risk customers, his or her risk level may be increased accordingly.

[0050] Historical transaction behavior records include transaction amounts, transaction types, and transaction times, which are crucial for understanding the behavior patterns of account managers. By analyzing these historical transaction data, it is possible to identify account managers' trading habits and potential risk points. For example, if an account manager often makes a large number of transactions in a short period of time, this may indicate that he has a high operational risk.

[0051] After collecting the sample information, data preprocessing is performed to ensure the quality and availability of the data. The data cleaning step removes incomplete, erroneous or inconsistent data records, involving techniques such as data filling and outlier processing. The data conversion step converts the data into a format suitable for model training, such as converting text data into numeric data, or converting timestamps into dates and times.

[0052] Through these steps, a training sample set containing rich information can be constructed, which provides a solid foundation for subsequent risk model training. This sample set not only includes the basic attribute information of account managers, but also covers data in multiple dimensions such as their business behavior, customer relationships, and transaction history. This collection of data provides a comprehensive perspective to assess the risk level of account managers, allowing the model to capture risk factors from different angles.

[0053] It should be understood that the account manager risk model construction method provides financial institutions with an effective tool to assess and manage account manager-related risks by accurately selecting risk-related time points and target account managers and extracting key sample information from them. It not only improves the accuracy of risk assessment, but also provides a solid foundation for financial institutions' risk management and decision-making. In this way, financial institutions can better understand and prevent potential risks, thereby maintaining sound operations in the highly competitive financial market.

[0054] In the above-mentioned method for building a customer manager risk model for a small sample scenario, the step S130 extracts the customer manager features of the sample information of each target customer manager in the training sample set to obtain a set of customer manager features. Further, since the training sample set contains sample information of different target customer managers, these sample information may contain various types of unstructured or semi-structured data (such as text descriptions, geographic locations, association networks, etc.). Traditional numerical feature extraction methods are difficult to effectively process such complex data types. Embedded coding can map these complex sample information into a continuous low-dimensional space through the learned vector space model, so as to better capture the implicit relationships and patterns therein. Therefore, in order to effectively extract the customer manager features of the sample information of each target customer manager in the training sample set to obtain a set of customer manager features, and comprehensively consider the multiple dimensional semantic information in the sample information of each target customer manager, so that the features of each customer manager are more abundant and diverse, in the technical solution of the present application, the sample information of each target customer manager in the training sample set is further embedded and encoded using the customer manager sample embedding matrix to obtain a set of target customer manager embedded semantic coding vectors as the set of customer manager features. Through embedded coding, the basic attribute information, business location abnormality information, customer association information and historical transaction behavior record information in the sample information of each target customer experience can be mapped to a common embedding space, so that this information can be converted into an embedded semantic coding vector, so that these heterogeneous data can be effectively utilized under the same framework, and these embedded semantic coding vectors can be fused to obtain the account manager characteristics of each account manager. This helps to build a multi-angle and multi-level risk assessment model and improve the model's predictive ability and robustness. In addition, through embedded coding, the relevant information of each account manager (such as basic attributes, business location abnormalities, associations and historical transaction behaviors, etc.) can be analyzed and extracted to determine the relevant information semantics and characteristics of each account manager, which is convenient for the subsequent training of the account manager basic risk model and the analysis of which factors may affect the risk level of the account manager.

[0055] It should be understood that there are certain challenges in processing unstructured or semi-structured data contained in the training sample set, such as text descriptions, geographic locations, and association networks. These complex data types are difficult to effectively process using traditional numerical feature extraction methods. To overcome this challenge, an advanced data representation method, embedded coding technology, is used to map complex sample information into a continuous low-dimensional space. In this way, the implicit relationships and patterns in the data can be better captured, thereby extracting richer and more diverse account manager features.

[0056] In this application, the sample information of each target account manager in the training sample set is embedded and encoded using the account manager sample embedding matrix to obtain a set of target account manager embedded semantic coding vectors. This process maps heterogeneous data such as the basic attribute information of the account manager, abnormal information of the business location, customer relationship information, and historical transaction behavior record information into a common embedding space. In this way, this information can be converted into an embedded semantic coding vector, so that the originally incompatible data can be effectively utilized under the same framework. By fusing the embedded semantic coding vector, the comprehensive characteristics of each account manager can be obtained, laying a solid foundation for building a multi-angle and multi-level risk assessment model. The above steps can not only provide a multi-dimensional feature set for the risk model, but also enhance the model's ability to identify and predict the risk levels of different account managers, and make the constructed risk assessment model more accurate and robust, and able to adapt to the ever-changing financial environment and risk conditions.

[0057] The process can be expressed as follows:

[0058]

[0059] in, The first one in the set of semantic encoding vectors for the target account manager is The target account manager embedding semantic encoding vector, and are the first weight matrix and the second weight matrix respectively, yes The corresponding entry target account manager embedding semantic encoding vector.

[0060] In the above-mentioned method for building a customer manager risk model for a small sample scenario, the step S140 performs an intrinsic feature distinguishable enhancement process on the customer manager feature set to obtain an enhanced customer manager feature set. Figure 2 FIG. 1 is a flowchart of step S140 of a method for building a customer manager risk model for a small sample scenario according to an exemplary embodiment. Figure 2As shown, the step S140 includes: step S141, using a field mapper to perform field mapping on each target account manager embedded semantic coding vector in the set of target account manager embedded semantic coding vectors to obtain a set of entry target account manager embedded semantic coding vectors; step S142, calculating the essential features of the set of entry target account manager embedded semantic coding vectors to obtain the target account manager semantic field essential feature vector; step S143, based on the target account manager semantic field essential feature vector, semantically enhancing the set of target account manager embedded semantic coding vectors to obtain a set of enhanced target account manager embedded semantic coding vectors as the set of enhanced account manager features.

[0061] It should be understood that since each target account manager embedded semantic coding vector in the set of target account manager embedded semantic coding vectors contains account manager features related to different target account managers, and the relevant behaviors and patterns between different account manager features may be similar, it is difficult to distinguish them effectively. That is to say, although each target account manager embedded semantic coding vector represents the relevant semantics and behavioral features of each target account manager, the uniqueness and differences between different account managers have not been fully considered. Therefore, in order to more accurately reflect the relevant behavioral semantics and real risk status of each account manager, it is necessary to extract the intrinsic correlation relationship and key information of these account manager features, so as to more effectively highlight the relevant behavioral differences between different account manager features and deeply explore the intrinsic connection between account manager features, so as to provide a basis for the subsequent construction of account manager risk model. Based on this, in the technical solution of the present application, the set of account manager features is further subjected to intrinsic feature distinguishable enhancement processing to obtain a set of enhanced account manager features. In particular, through the intrinsic feature distinguishable enhancement processing, the deep information structure and intrinsic distinguishability features between different account manager features can be obtained through a multi-stage processing mechanism, so as to significantly improve the quality of account manager feature representation. This makes it easier to distinguish the different relevant behavior semantics and unique behavior patterns between different account managers, which helps to build the account manager basic risk model and identify potential risk factors. In addition, the guidance provided by the field essential characteristics between different account manager characteristics can specifically amplify those characteristics that are considered to be highly relevant to risk, while suppressing those less important parts. This can mine more valuable information and improve the performance of the account manager basic risk model.

[0062] Specifically, in step S141, a field mapper is used to perform field mapping on each target account manager embedded semantic coding vector in the set of target account manager embedded semantic coding vectors to obtain a set of entry target account manager embedded semantic coding vectors. In the data processing flow of the intrinsic feature distinguishable enhancement processing, it first uses a field mapper to perform field mapping on each target account manager embedded semantic coding vector in the set of target account manager embedded semantic coding vectors to obtain a set of entry target account manager embedded semantic coding vectors. That is, a field mapping operation is performed on each vector in the original target account manager embedded semantic coding vector sequence, with the aim of converting these features into a form that is more suitable for the current model or task requirements, that is, obtaining a set of entry target account manager embedded semantic coding vectors. This stage essentially performs a feature space transformation, involving nonlinear transformation or other forms of preprocessing technology, so that the features can better reflect the intrinsic relationship between samples or are more easily captured by the algorithm. Key information.

[0063] Specifically, the step S142 is to calculate the essential features of the set of embedded semantic coding vectors of the target account manager to obtain the target account manager semantic field essential feature vector. Figure 3 is a flowchart of step S142 of a method for building a customer manager risk model for a small sample scenario according to an exemplary embodiment. Figure 3 As shown, the step S142 includes: step S1421, calculating the field depth factor of each entry target account manager embedded semantic coding vector in the set of entry target account manager embedded semantic coding vectors to obtain a set of target account manager semantic field depth factors; step S1422, based on the set of target account manager semantic field depth factors, calculating the essential features of the set of entry target account manager embedded semantic coding vectors to obtain the target account manager semantic field essential feature vector.

[0064] Specifically, the step S1421 calculates the field depth factor of each entry target account manager embedded semantic coding vector in the set of entry target account manager embedded semantic coding vectors to obtain a set of target account manager semantic field depth factors. That is, for each entry target account manager embedded semantic coding vector after field mapping, the corresponding field depth factor is further calculated. The target account manager semantic field depth factor can be understood as an indicator to measure the relative importance or complexity of the feature point in the area to which it belongs. By obtaining such factors one by one for all entry target account manager embedded semantic coding vectors in the entire set, a new sequence is formed, namely, a field depth factor sequence. Furthermore, using the set of target account manager semantic field depth factors obtained above as a basis, the field effect normalization mechanism is used to extract a feature distribution that comprehensively reflects the core attributes of the entire batch of entry target account manager embedded semantic coding vectors, namely, the target account manager semantic field essential feature vector.

[0065] In a specific embodiment of the present application, step S1421, calculates the field depth factor of each entry target account manager embedded semantic coding vector in the set of entry target account manager embedded semantic coding vectors to obtain a set of target account manager semantic field depth factors, including: calculating the square of the one-norm of the entry target account manager embedded semantic coding vector to obtain the one-norm square representation of the entry target account manager embedded semantics; calculating the sum and difference between a constant one and the one-norm square representation of the entry target account manager embedded semantics to obtain the entry target account manager embedded semantics sum representation and the entry target account manager embedded semantics difference representation; calculating the division between the entry target account manager embedded semantics sum representation and the entry target account manager embedded semantics difference representation, and then calculating the logarithmic function value with base 2 of the obtained division value to obtain the target account manager semantic field depth factor.

[0066] The process can be expressed as follows:

[0067]

[0068] in, is the target account manager embedding semantic coding vector corresponding to the target account manager embedding semantic coding vector, To calculate the square of the vector-norm, represents the logarithmic function value with base 2, For calculation The depth of field factor, for The corresponding target account manager semantic field depth factor.

[0069] Specifically, in step S1422, based on the set of semantic field depth factors of the target account manager, the essential features of the set of embedded semantic coding vectors of the target account manager are calculated to obtain the semantic field essential feature vector of the target account manager. This step not only involves in-depth analysis of the embedded vectors of each account manager, but also can dig out deeper information and connections behind these vectors. Specifically, key semantic information is extracted from the set of embedded semantic coding vectors of the target account manager, which covers multiple dimensions such as the basic attributes, business development behavior, and customer relationships of the account manager. By comprehensively considering the semantic field depth factors of these dimensions, the importance and influence of each dimension in the overall risk assessment can be evaluated. On this basis, advanced data analysis techniques, such as principal component analysis, cluster analysis, or neural networks, are used to identify and extract the attributes that best represent the risk characteristics of the account manager. Finally, these features are integrated into a semantic field essential feature vector of the target account manager, which not only captures the key risk characteristics of the account manager, but also reveals the intrinsic connection and interaction between different features.

[0070] The process can be expressed as follows:

[0071]

[0072] in, for The corresponding target account manager semantic field depth factor, It means to calculate the value of the natural exponential function with the natural constant e as the base. for The corresponding standardized target account manager semantic field depth factor, The first one in the set of semantic encoding vectors for the target account manager is The target account manager embedding semantic encoding vector, It is the essential feature vector of the semantic field of the target account manager.

[0073] Specifically, in step S143, based on the target account manager semantic field essential feature vector, the set of target account manager embedded semantic coding vectors is semantically enhanced to obtain a set of enhanced target account manager embedded semantic coding vectors as the set of enhanced account manager features. Using the target account manager semantic field essential feature vector as the conditional feature vector, each target account manager embedded semantic coding vector in the set of target account manager embedded semantic coding vectors is distinguishably enhanced to obtain the set of enhanced target account manager embedded semantic coding vectors. That is, using the target account manager semantic field essential feature vector as additional conditional information, the initial target account manager embedded semantic coding vector sequence is subjected to element-by-element distinguishable enhancement processing, so as to specifically amplify important features in certain aspects according to specific rules or weight adjustment schemes, while suppressing other less important parts, so as to achieve the purpose of optimizing feature expression.

[0074] Among them, based on the target account manager semantic field essential feature vector, the set of target account manager embedded semantic coding vectors is semantically enhanced to obtain a set of enhanced target account manager embedded semantic coding vectors as the set of enhanced account manager features, including: taking the target account manager semantic field essential feature vector as the conditional feature vector, and respectively performing distinguishable enhancement on each target account manager embedded semantic coding vector in the set of target account manager embedded semantic coding vectors to obtain the set of enhanced target account manager embedded semantic coding vectors.

[0075] In a specific embodiment of the present application, the target account manager semantic field essential feature vector is used as a conditional feature vector, and each target account manager embedded semantic coding vector in the set of the target account manager embedded semantic coding vector is distinguishably enhanced to obtain the set of enhanced target account manager embedded semantic coding vectors, including: calculating the multiplication between the target account manager semantic field essential feature vector and the target account manager semantic essential weight matrix to obtain the target account manager semantic essential feature modulation vector; calculating the multiplication between the target account manager embedded semantic coding vector and the local target account manager semantic weight matrix to obtain the target account manager embedded semantic feature modulation vector; calculating the positional sum between the target account manager semantic essential feature modulation vector and the target account manager embedded semantic feature modulation vector, and then performing positional summation with the bias vector to obtain the target account manager semantic comprehensive modulation representation vector; using After the function probabilistically modulates the target account manager's semantic comprehensive modulation representation vector, the modulated target account manager's semantic comprehensive modulation weight vector is point-multiplied by the target account manager's embedded semantic coding vector to obtain an enhanced target account manager embedded semantic coding vector.

[0076] The process can be expressed as follows:

[0077]

[0078] in, The first one in the set of semantic encoding vectors for the target account manager is The target account manager embedding semantic encoding vector, is the essential feature vector of the semantic field of the target account manager, is the semantic essence weight matrix of the target account manager, is the semantic weight matrix of local target account managers, is the bias vector, yes function, It is the point multiplication by position. is the first in the set of the embedded semantic encoding vectors of the enhanced target account manager A reinforced target account manager is embedded into a semantic encoding vector.

[0079] In the above-mentioned method for constructing a customer manager risk model applied to a small sample scenario, the step S150 trains the customer manager risk model based on the set of enhanced customer manager features to obtain a customer manager basic risk model. Furthermore, based on the set of enhanced customer manager features, the customer manager risk model is trained to obtain a customer manager basic risk model. In other words, by training the model using a set of optimized and enhanced customer manager features, it can be ensured that the model learns the key information that best reflects the customer manager's behavior patterns and their potential risks. These enhanced features are usually carefully designed to better capture the subtle differences and complex relationships between different customer manager features, thereby helping to improve the performance and reliability of the customer manager basic risk model. This not only enhances the practical value of the model, but also provides a solid foundation for subsequent risk management and decision-making.

[0080] In the above-mentioned method for building a customer manager risk model for a small sample scenario, the step S160 identifies abnormal sample points from the training sample set. It should be understood that the traditional supervised learning method relies on labeled data (such as black samples and white samples) to train the model. However, in a small sample scenario, not all risk patterns may be fully covered by existing labels, especially those new or rare risk behaviors. Therefore, abnormal sample points are further identified from the training sample set, and the customer manager basic risk model is trained based on the abnormal sample points and the customer manager characteristics of the abnormal sample points to obtain the customer manager enhanced risk model. In particular, by identifying abnormal sample points through an unsupervised method, these new patterns or abnormal situations that are not covered by the existing label system can be captured, thereby expanding the risk identification ability of the model. At the same time, since incremental learning is a method that can gradually absorb new information without retraining the entire model. For the field of financial risk control, the market environment and business rules are constantly changing, so it is very important to be able to quickly adapt to new trends and challenges. Through incremental learning, the newly identified abnormal sample points and their customer manager characteristics can be effectively integrated into the existing basic model without the need for comprehensive retraining every time. This approach not only improves efficiency, but also ensures that account managers' risk models can continue to evolve to adapt to the new risk environment.

[0081] In the above-mentioned method for constructing a customer manager risk model for a small sample scenario, the step S170 trains the customer manager basic risk model based on the abnormal sample points and the customer manager characteristics of the abnormal sample points to obtain the customer manager enhanced risk model. Specifically, the step S170 includes: based on the abnormal sample points and the customer manager characteristics of the abnormal sample points, performing incremental learning training on the customer manager basic risk model to obtain the customer manager enhanced risk model.

[0082] In a preferred example, considering that each customer manager feature in the set of customer manager features represents the embedded coding features of the sample information of the corresponding target customer manager, when the features described based on the sequence conditions are distinguishably enhanced, due to the differences in the descriptive properties of the embedded coding sequence conditions under the local sample domain, the set of enhanced customer manager features will also have probability convergence and divergence based on different distinguishable enhancements, thereby affecting the training speed of the customer manager risk model.

[0083] Therefore, when the applicant of the present application trains the account manager risk model based on the set of enhanced account manager features to obtain the account manager basic risk model, the applicant optimizes the set of enhanced account manager features in each training iteration, including the following steps:

[0084] Determine the median eigenvalue, the maximum eigenvalue and the minimum eigenvalue in the set of enhanced account manager features, and divide the median eigenvalue by the difference between the maximum eigenvalue and the minimum eigenvalue to obtain the enhanced account manager feature distribution probability value, that is, ;in, , and are respectively the median eigenvalue, the maximum eigenvalue and the minimum eigenvalue in the set of enhanced account manager features, is the probability value of the enhanced account manager feature distribution;

[0085] Each feature value in the set of enhanced customer manager features is normalized to the maximum value to obtain the enhanced customer manager probability feature value ;

[0086] Subtract the probability value of the enhanced account manager feature distribution from one Calculate the probability characteristic value of each enhanced account manager for the index The power function of the first enhanced account manager probability convergence eigenvalue ;in, is the probability value of the enhanced account manager feature distribution, is the probability characteristic value of the enhanced account manager, is the first enhanced account manager probability convergence eigenvalue;

[0087] The probability value of the enhanced account manager feature distribution Calculate a probability characteristic value for each of the enhanced account managers for the index The power function of the difference is used to obtain the second enhanced account manager probability convergence eigenvalue ;in, is the probability value of the enhanced account manager feature distribution, is the probability characteristic value of the enhanced account manager, is the second enhanced account manager probability convergence eigenvalue;

[0088] The probability characteristic value of each enhanced account manager minus the probability value of the enhanced account manager feature distribution The difference is multiplied to obtain the first enhanced account manager probability qualification eigenvalue ;in, is the probability value of the enhanced account manager feature distribution, is the probability characteristic value of the enhanced account manager, is the first enhanced account manager probability limiting feature value;

[0089] The probability characteristic value of each enhanced account manager The difference between the value of and the probability value of the enhanced account manager feature distribution Multiply to obtain the second enhanced account manager probability qualified eigenvalue ;in, is the probability value of the enhanced account manager feature distribution, is the probability characteristic value of the enhanced account manager, is the second enhanced account manager probability limiting feature value;

[0090] After multiplying the first enhanced account manager probability convergence characteristic value and the second enhanced account manager probability convergence characteristic value, further add the first enhanced account manager probability limit characteristic value and the second enhanced account manager probability limit characteristic value to obtain the optimized characteristic value of the set of enhanced account manager characteristics ;in, is the first enhanced account manager probability convergence eigenvalue, is the second enhanced account manager probability convergence eigenvalue, is the first enhanced account manager probability-limited feature value, is the second enhanced account manager probability-qualified eigenvalue, It is the optimized feature value of the set of enhanced customer manager features.

[0091] Based on this, by using the cross-entropy form power series of the distribution analytic probability of the set of enhanced customer manager features as the probability distribution convergence limit, the regression probability convergence approximation is performed on the basis of the combination of the feature set distribution and the probability density distribution of the set of enhanced customer manager features, so as to avoid the conservative convergence control of the fatal potential divergence of the set of enhanced customer manager features, and further use the probability distribution cross entropy as the objective function to guide the limited recovery strategy, so as to achieve the common agility of the convergence of the feature set with probability convergence and divergence to the probability density distribution space, and improve the training speed when training the customer manager risk model based on the set of enhanced customer manager features to obtain the customer manager basic risk model.

[0092] In summary, the above scheme is adopted to select account managers who have application records before and at the risk time point and meet the performance period; then, extract the sample information of these account managers to form a training sample set; then, extract the account manager features from the sample set, perform distinguishable enhancement processing on the intrinsic features, obtain the enhanced feature set, and train the account manager basic risk model based on the enhanced feature set; finally, identify abnormal sample points from the training sample set, and retrain the basic risk model in combination with these sample points and their features to obtain the enhanced risk model. In this way, it is possible to achieve the effect of accurately constructing and optimizing the account manager risk assessment model in a small sample data environment.

[0093] Figure 4 FIG. 1 is a block diagram of a system for building a customer manager risk model for a small sample scenario according to an exemplary embodiment. Figure 4 As shown, according to the embodiment of the present application, the customer manager risk model construction system 200 applied to the small sample scenario includes: a target customer manager selection module 210, which is used to select a time point involving risk, and select a target customer manager who has submitted a case within a predetermined time period before the time point involving risk and meets the performance period; a training sample set extraction module 220, which is used to extract the sample information of the target customer manager to obtain a training sample set; a customer manager feature mining module 230, which is used to extract the customer manager features of the sample information of each target customer manager in the training sample set to obtain a set of customer manager features; a distinguishable enhancement processing module 240, which is used to perform intrinsic feature distinguishable enhancement processing on the set of customer manager features to obtain a set of enhanced customer manager features; a training module 250, which is used to train the customer manager risk model based on the set of enhanced customer manager features to obtain a customer manager basic risk model: an abnormal sample point identification module 260, which is used to identify abnormal sample points from the training sample set; a customer manager enhanced risk model generation module 270, which is used to train the customer manager basic risk model based on the abnormal sample points and the customer manager features of the abnormal sample points to obtain a customer manager enhanced risk model.

[0094] Here, those skilled in the art can understand that the specific operations of each module in the above-mentioned account manager risk model construction based on the small sample scenario have been referred to above. Figures 1 to 3 The description of the client manager risk model construction method applied to small sample scenarios has been introduced in detail, and therefore, its repeated description will be omitted.

[0095] The above description is only a preferred embodiment of the present application and an explanation of the technical principles used. Those skilled in the art should understand that the scope of disclosure involved in the present application is not limited to the technical solution formed by a specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the above disclosed concept. For example, the above features are replaced with the technical features with similar functions disclosed in this application (but not limited to) by each other to form a technical solution.

[0096] In addition, although each operation is described in a specific order, this should not be construed as requiring these operations to be performed in the specific order shown or to be performed in a sequential order. Under certain circumstances, multitasking and parallel processing may be advantageous. Similarly, although some specific implementation details are included in the above discussion, these should not be construed as limiting the scope of the application. Some features described in the context of a separate embodiment can also be implemented in a single embodiment in combination. On the contrary, the various features described in the context of a single embodiment can also be implemented in multiple embodiments individually or in any suitable sub-combination mode.

[0097] Although the subject matter has been described in language specific to structural features and / or method logic actions, it should be understood that the subject matter defined is not necessarily limited to the specific features or actions described above. On the contrary, the specific features and actions described above are merely exemplary forms of implementation. With respect to the apparatus in the above-mentioned embodiments, the specific manner in which each module performs the operation has been described in detail in the embodiments related to the method, and will not be described in detail here.

Claims

1. A method for constructing a customer manager risk model for a small sample scenario, characterized in that: include: Select a time point involving risk, and select a target account manager who has received a case in a predetermined period of time before the time point involving risk and who meets the performance period condition, wherein the performance period condition refers to whether the target account manager has served in the financial institution for a long enough period of time; Extracting sample information of the target account manager to obtain a training sample set; Extracting the account manager features of the sample information of each target account manager in the training sample set to obtain a set of account manager features; Performing intrinsic feature distinguishable enhancement processing on the set of customer manager features to obtain a set of enhanced customer manager features; Based on the set of enhanced account manager features, the account manager risk model is trained to obtain the account manager basic risk model: Identifying abnormal sample points from the training sample set; The account manager basic risk model is trained based on the abnormal sample points and the account manager characteristics of the abnormal sample points to obtain the account manager enhanced risk model.

2. The method for constructing a customer manager risk model for a small sample scenario according to claim 1, characterized in that: The sample information of the target account manager includes basic attribute information, business location abnormality information, customer relationship information and historical transaction behavior record information.

3. The method for constructing a customer manager risk model for a small sample scenario according to claim 2, characterized in that: Extracting the customer manager features of the sample information of each target customer manager in the training sample set to obtain a set of customer manager features, including: using a customer manager sample embedding matrix to respectively embed code the sample information of each target customer manager in the training sample set to obtain a set of target customer manager embedded semantic coding vectors as the set of customer manager features.

4. The method for constructing a customer manager risk model for a small sample scenario according to claim 3, characterized in that: The set of customer manager features is subjected to an intrinsic feature distinguishable enhancement process to obtain a set of enhanced customer manager features, including: Using a field mapper to perform field mapping on each target account manager embedded semantic coding vector in the set of target account manager embedded semantic coding vectors to obtain a set of entry target account manager embedded semantic coding vectors; Calculating the essential features of the set of embedded semantic coding vectors of the target account manager to obtain the target account manager semantic field essential feature vector; Based on the target account manager's semantic field essential feature vector, the set of target account manager embedded semantic coding vectors is semantically enhanced to obtain a set of enhanced target account manager embedded semantic coding vectors as the set of enhanced account manager features.

5. The method for constructing a customer manager risk model for a small sample scenario according to claim 4, characterized in that: Calculating the essential features of the set of embedded semantic coding vectors of the target account manager to obtain the target account manager semantic field essential feature vector, including: Calculating the field depth factor of each entry target account manager embedded semantic coding vector in the set of entry target account manager embedded semantic coding vectors to obtain a set of target account manager semantic field depth factors; Based on the set of semantic field depth factors of the target account manager, the essential features of the set of embedded semantic coding vectors of the incoming target account manager are calculated to obtain the semantic field essential feature vector of the target account manager.

6. The method for constructing a customer manager risk model for a small sample scenario according to claim 5, characterized in that: Calculating the field depth factor of each entry target account manager embedded semantic coding vector in the set of entry target account manager embedded semantic coding vectors to obtain a set of target account manager semantic field depth factors, including: Calculate the square of the one-norm of the embedding semantic coding vector of the target customer manager to obtain the one-norm square representation value of the embedding semantics of the target customer manager; Calculate the sum and difference between the constant one and the squared representation of the norm of the target customer manager embedded semantics to obtain the sum representation of the target customer manager embedded semantics and the difference representation of the target customer manager embedded semantics; After calculating the division between the embedded semantic sum representation of the entry target account manager and the embedded semantic difference representation of the entry target account manager, the logarithmic function value with base 2 of the obtained division value is calculated to obtain the semantic field depth factor of the target account manager.

7. The method for constructing a customer manager risk model for a small sample scenario according to claim 6, characterized in that: Based on the target account manager's semantic field essential feature vector, the set of target account manager's embedded semantic coding vectors is semantically enhanced to obtain a set of enhanced target account manager embedded semantic coding vectors as the set of enhanced account manager features, including: taking the target account manager's semantic field essential feature vector as a conditional feature vector, and performing distinguishable enhancement on each target account manager embedded semantic coding vector in the set of target account manager embedded semantic coding vectors to obtain the set of enhanced target account manager embedded semantic coding vectors.

8. The method for constructing a customer manager risk model for a small sample scenario according to claim 7, characterized in that: Taking the target account manager semantic field essential feature vector as the conditional feature vector, each target account manager embedded semantic coding vector in the set of target account manager embedded semantic coding vectors is distinguishably enhanced to obtain the set of enhanced target account manager embedded semantic coding vectors, including: Calculate the multiplication between the target account manager semantic field essential feature vector and the target account manager semantic essential weight matrix to obtain the target account manager semantic essential feature modulation vector; Calculate the multiplication between the target account manager embedded semantic coding vector and the local target account manager semantic weight matrix to obtain the target account manager embedded semantic feature modulation vector; Calculate the positional sum of the target account manager's semantic essential feature modulation vector and the target account manager's embedded semantic feature modulation vector, and then perform positional summation with the bias vector to obtain the target account manager's semantic comprehensive modulation representation vector; After probabilistically modulating the target account manager's semantic comprehensive modulation representation vector using the sigmoid function, the modulated target account manager's semantic comprehensive modulation weight vector is point-multiplied by the target account manager's embedded semantic coding vector to obtain an enhanced target account manager embedded semantic coding vector.

9. The method for constructing a customer manager risk model for a small sample scenario according to claim 8, characterized in that: Based on the abnormal sample points and the characteristics of the customer managers of the abnormal sample points, the customer manager basic risk model is trained to obtain the customer manager enhanced risk model, including: based on the abnormal sample points and the characteristics of the customer managers of the abnormal sample points, the customer manager basic risk model is incrementally learned and trained to obtain the customer manager enhanced risk model.

Images