A prepaid consumption risk identification method and early warning system

Abstract

The present application relates to a method for identifying risks in prepaid consumption, which includes the steps of basic data collection, data governance, label extraction, image generation, warning model definition and risk identification. The present application also relates to a prepaid consumption risk warning system, including a communication-connected server and client, at least one of which includes a memory, a processor and a computer program stored in the memory, and the processor implements the above method when executing the computer program. The above scheme can accurately identify and provide advance warning for prepaid consumption risks.

Description

Technical Field

[0001] The present invention relates to the technical field of risk warning, and in particular to a prepaid consumption risk identification method and warning system. Background Art

[0002] In actual work, it is found that various risk problems often occur in companies related to prepaid consumption, and for such problems, they currently still rely more on post-event handling. How to grasp the operating characteristics of prepaid consumption companies, capture abnormal situations, discover signs and signs of prepaid consumption risks in advance and issue warnings, and gain advance time for relevant departments to properly handle them, is a topic that urgently needs to be broken through.

[0003] With the development of big data technology, solutions have been proposed in the prior art for problems and risks in the fields of urban operation, business operation, social security, etc., which can realize timely early warning of risks. For example, the prior patent application with publication number CN117273429A discloses an event monitoring method, which obtains event information of multiple events corresponding to the target area, performs risk analysis on the events based on pre-configured service scenario information, obtains the scenario-based risk results of the events in the target area, and generates early warning information corresponding to at least part of the service scenarios in the target area based on the risk results. Further combined with the specific embodiments given in the prior patent application, it can be known that the specific analysis method for event information is to call the risk identification model associated with the target scenario, the risk identification model includes a hot event identification model and a key event identification model, the hot event identification model can include at least one of a group hot spot identification model, a sudden hot spot identification model, a continuous hot spot identification model, a potential hot spot (hot spot trend) identification model and a hot spot (topic of concern) identification model, and the key event identification model can include at least one of a sensitive event identification model, a repeated complaint event identification model and a high-frequency entity event identification model. The specific processing methods of the above identification models are: determining risks based on the number of events and complaints regarding the same claim or the same entity within a specific time period.

[0004] It can be seen that the above scheme only determines the risk based on the number of complaints, the frequency of complaints, etc., and when the complaint event occurs, it is obvious that the incident involved has also occurred, and when the risk is determined to exist based on the above risk judgment, it is obvious that the corresponding risk event has also caused a certain range of impact. As the application scenario exemplified in the embodiment, the above identification method can timely identify urban infrastructure problems such as broken street lights, road waterlogging, manhole covers, environmental sanitation, etc., to avoid them causing continuous risks.

[0005] However, for the application scenarios of prepaid consumption, if risk judgment is only based on post-event information, it can only achieve timely discovery after the event to prevent its impact from further expanding, but it cannot effectively prevent risks in advance. The above solution still lacks the anticipation and timeliness of risk warning for the application scenarios of prepaid consumption. In addition, the identification of prepaid consumption risks only from the perspective of post-event, and the event / data identification foundation relied on is not comprehensive enough. Summary of the invention

[0006] In order to solve the above problems, the purpose of the present invention is to provide a prepaid consumption risk identification method and early warning system, which can accurately identify and provide early warning for prepaid consumption risks.

[0007] In a first aspect, the present application provides a method for identifying prepaid consumption risks, the method comprising the following steps:

[0008] Basic data collection, including enterprise information data and various governance event data. The enterprise information data includes enterprise entity data, enterprise entity change record data and enterprise operation abnormality data. The governance event data includes government service hotline complaint data, consumer supervision platform complaint data, dispute data, and other channel complaint data;

[0009] Data governance: performing data governance on the basic data according to preset data standards, and storing the processed basic data in a preset basic data table in the database;

[0010] Label extraction: processing the basic data table based on the preset data table calculation rules and the pre-trained NLP natural language model to extract the risk labels of the basic data, which are included in the preset label library;

[0011] Profile generation: associate the enterprise subject with the risk tags it hits to generate an enterprise risk profile, which includes the enterprise subject logo and all the risk tags it hits;

[0012] Early warning model definition: After freely combining the risk tags in the tag library and assigning different scores to each risk tag, one or more risk early warning models for specific scenarios are formed;

[0013] Risk identification, determine the risk warning model for the specific scenario corresponding to the corporate entity, input the corporate risk profile information of the corporate entity into the corresponding risk warning model for the specific scenario, obtain the risk value of the corporate entity, and determine whether the corporate entity has prepaid consumption risks based on the risk value of the corporate entity.

[0014] In a second aspect, the present application provides a prepaid consumption risk warning system, comprising a server and a client connected in communication, at least one of the server and the client comprising a memory, a processor and a computer program stored in the memory, and the processor implements any of the above methods when executing the computer program.

[0015] Preferably, at least some of the steps in the above method are executed on the server side.

[0016] Due to the adoption of the above solution, the present application has the following advantages over the prior art:

[0017] This application uses enterprise information data and governance event data as the basis for risk analysis. According to the business activities of the corporate entities involved in the enterprise information data and combined with the governance event data, a multi-dimensional analysis is conducted to determine whether the corporate entities have characteristics that meet the risks of prepaid consumption. This enables early and comprehensive identification of prepaid consumption risks.

[0018] Basic data includes numerical values ​​and events. For numerical values ​​and simple event information that can be obtained through logical judgment, risk labels can be extracted directly based on the preset data table calculation rules. For other event information, risk labels are extracted based on the NLP natural language model. The above label extraction method can improve the efficiency of label extraction while ensuring the accuracy of label recognition.

[0019] Associate corporate entities with risk labels, and build multi-dimensional corporate risk profiles through their associated risk labels, so as to comprehensively and accurately evaluate the business stability of corporate entities.

[0020] Construct multiple different specific scenario risk warning models for different scenarios. Different risk warning models use different combinations of risk labels, and the risk values ​​corresponding to the same risk label in different risk warning models are different, so that the characteristics of prepaid consumption risks in different scenarios can be reflected in a targeted manner, making risk identification for specific scenarios more accurate. Different specific scenarios are divided for different corporate entities, and corresponding specific scenario risk warning models are used for risk identification, which is conducive to improving the accuracy of risk identification of corporate entities. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 A schematic diagram of the method flow of this application;

[0022] Figure 2 This is a schematic diagram of the label extraction process of the NLP natural language model;

[0023] Figure 3 This is a schematic diagram of enterprise risk profiling. DETAILED DESCRIPTION

[0024] To facilitate understanding of the technical solutions of the embodiments of the present application, the following describes the related technologies of the embodiments of the present application. The following related technologies can be combined with the technical solutions of the embodiments of the present application as optional solutions, and they all belong to the protection scope of the embodiments of the present application.

[0025] The present application embodiment provides a method for identifying prepaid consumption risks, such as Figure 1 As shown, the method includes: basic data collection, the basic data includes enterprise information data and governance event data, the enterprise information data includes enterprise subject data, enterprise subject change record data and enterprise operation abnormality data, and the governance event data includes government service hotline complaint data, consumer supervision platform complaint data, dispute data, and other channel complaint data; data governance, data governance of the basic data according to preset data standards, and storing the processed basic data in a preset basic data table in the database; label extraction, processing the basic data table based on preset data table calculation rules and a pre-trained NLP natural language model to extract risk labels of the basic data, The risk labels are included in a preset label library; portrait generation, establishing an association between the enterprise entity and the risk labels it hits, and generating an enterprise risk portrait, which includes the enterprise entity logo and all the risk labels it hits; early warning model definition, freely combining the risk labels in the label library, and assigning different scores to each risk label to form one or more risk early warning models for specific scenarios; risk identification, determining the risk early warning model for the specific scenario corresponding to the enterprise entity, inputting the enterprise risk portrait information of the enterprise entity into the risk early warning model of the corresponding specific scenario, obtaining the risk value of the enterprise entity, and determining whether the enterprise entity has prepaid consumption risks based on the risk value of the enterprise entity.

[0026] Furthermore, in the case of enterprise management by region, when basic data is collected, only the basic data within the area of ​​interest can be obtained and processed according to the above method.

[0027] In the basic data collection step, the basic data comes from multiple open and reliable channels such as consumer complaint data and public social data, and the corresponding data is automatically obtained through the data interface provided by the corresponding platform.

[0028] As for the timing of obtaining basic data, the full amount of data from each platform is obtained when the platform is first accessed. Since the data of these platforms are updated in real time, the data update frequency / time interval is set in the embodiment, and the updated data is obtained according to the preset data update frequency / time interval. In a specific embodiment, the same data update frequency / time interval can be used for each platform from which the basic data comes, or different data update frequencies / time intervals can be set. Similarly, the basic data from each platform can be obtained at the same time, or the basic data of the basic data from each platform can be obtained at different times.

[0029] For the content of obtaining basic data, only the data of the type of interest is obtained from the corresponding platform each time. In a specific embodiment, those skilled in the art pre-set the data type to be obtained from each platform, or perform data screening operations after obtaining the full amount of data from each platform. Furthermore, a custom operation can be provided in the human-computer interaction interface for the operator to perform data screening within a certain range. Preferably, only newly added and changed data is obtained each time to reduce the pressure of data processing.

[0030] In this embodiment, the determination of the type of basic data used as the basis for analysis is one of the key points of this solution. The enterprise subject data is the basic information of the enterprise subject, and its content includes the registered enterprise subject name, unified social credit code, residence, legal person and other basic information. The enterprise subject change record includes the change content and change time of the enterprise subject. According to the business experience summary, the change of the enterprise subject can reflect the business stability of the enterprise to a certain extent. For example, the recent change of legal person and cross-regional change of address are operations that frequently occur before the enterprise encounters prepaid consumption risks. The abnormal business data of the enterprise comes from the public abnormal business situation of the enterprise. In the embodiment, it can be obtained from the public information such as the abnormal business list of the enterprise, the enterprise's dishonesty and execution. These aspects can also reflect the business stability of the enterprise to a certain extent. The complaint data of the government service hotline includes consumer complaints and reporting data from the telephone and network of the government service hotline. This part is an important part of the basic data. In the subsequent processing process, the complaint and reporting data are processed by semantic recognition technology. The consumer supervision platform will also feedback the complaint data, so the complaint data from the consumer supervision platform is also obtained. Dispute data can monitor the previous public contract litigation of the enterprise subject to determine its business risk. Through the above-mentioned multi-channel and multi-type data, a comprehensive and multi-faceted judgment can be made on whether the corporate entity has prepaid consumption risks, thereby achieving prior prevention and timely handling afterwards.

[0031] In the above basic data, various types of governance event data are stored uniformly. However, this method only focuses on prepaid consumption events in the governance event data. This setting can reduce the operations of reclassification and screening from massive governance event data, avoid the problem of incomplete and inaccurate data due to classification and screening errors, and make the data processing method in the data processing stage simpler and more efficient, so as to reduce the pressure of data processing.

[0032] For the collected basic data, further data governance is required. The collected data must go through the big data standard cleaning process, remove dirty data, supplement missing data, supplement missing fields, translate enumeration values, etc., and the cleaned data can enter the business end for use. In this embodiment, the processed basic data is stored in the basic data table preset in the database. The basic data table in this embodiment includes the basic information table of the enterprise, the enterprise branch information table and the enterprise shareholder information table involving the enterprise subject information, the enterprise change record table involving the enterprise subject change record data, the enterprise operation abnormality table involving the enterprise operation abnormality, the enterprise limit consumption table, the enterprise negative list table, the enterprise dishonesty record table, the event information table involving complaints and disputes on the consumer supervision platform, and the complaint event information table involving the complaint data of the government service convenience hotline. It should be further explained that the above tables and data types are not completely one-to-one corresponding, and there can also be cross-calls. For example, in the enterprise negative list table, according to the contract dispute data, specific negative items can be divided into types such as enterprise operation abnormality, illegal dishonesty, etc., and the data divided into enterprise operation abnormality belongs to the enterprise operation abnormality data. Based on this, preferably, the data governance includes classifying and processing the collected basic data.

[0033] Data governance includes adjusting the data in specific fields to meet the required format. For example, for fields involving addresses, it is required to be stored in a complete format of No. XX, XX Street, XX District, XX City, XX Province, but there may be multiple representations of the same address in the addresses obtained from various external interfaces, such as incomplete hierarchy, or No. XX, XX Street, XX Building, and XX Unit all represent the same address. In this case, the corresponding fields in the table can be unified through the association matching in each table. For example, the business address of the enterprise in the enterprise basic information table comes directly from the enterprise's registration information, which is stored in a standard format. The address field can be unified by matching the enterprise basic information table. In actual applications, for governance event data, especially complaint data, there are usually cases where information is omitted during identification due to the non-standard expression of the complaint content. For example, for complaint data, there is a situation in which the enterprise entity is referred to by the brand name in its content. If it is identified only by matching information through the enterprise name, unified credit code, etc., it is very easy to miss. Or, for address information, it is described in a way such as 50 meters to the left of the XX intersection, which cannot be directly associated with the registered address of the enterprise entity. In this regard, specific fields can be uniformly adjusted to the required format for storage by calling a third-party application. Taking the map application as an example, relevant features, such as brand name, XX building or XX unit, XX intersection, can be searched in the map application to locate the detailed address and unify it into the required format. The above solution can improve the accuracy and comprehensiveness of information identification in the data governance stage and prevent information omissions. Storing in a unified format facilitates subsequent data applications.

[0034] Data processing also includes associating governance event data in basic data with enterprise information data. For the government service hotline complaint data in the governance event data, there are cases where brand names are used to refer to corporate entities in its content. Therefore, it is necessary to identify the corporate entity information involved in the governance event data. If the corporate entity information involved in the complaint data is directly contained in the corporate information data, then the association is established directly based on the corporate entity information. For example, if the information in the business license of the enterprise, such as the name of the enterprise and the unified credit code, appears in the complaint data, the corporate entity can be identified directly through keyword matching, and the association between the complaint data and the corporate information data is established. If the corporate entity information involved in the complaint data is represented by a brand name, the store address is searched from a third-party map tool based on the brand name, and the corporate registration address contained in the corporate information data is matched based on the store address; and / or, the corporate name corresponding to the brand name is obtained from a third-party review website based on the brand name, and the corporate registration name contained in the corporate information data is matched based on the corporate name. The above-mentioned third-party map and third-party review website can cooperate to support each other's search results and improve the accuracy of the search results.

[0035] The basic data processed by data governance is input into the business end for label extraction. In this method, the prepaid consumption risk events in the past few years are combed and reviewed in full cycle in advance, the risk characteristics of prepaid consumption are summarized, and a risk label library is constructed. In this embodiment, the risk labels include abnormal enterprise labels, high-frequency complaint labels, legal person changes within the first preset time reaching the first preset value label, equity change labels within the second preset time, cross-regional address changes, labels inconsistent between the registered place and the actual place of business, abnormal recharge discount labels, complaints to the government service hotline within the third preset time exceeding the second preset value label, and complaints to the consumer supervision platform within the fourth preset time exceeding the third preset value label. One or more of the above. Preferably, this method includes providing a risk label customization module, and adjusting the risk labels in the label library through the risk label customization module. The specific adjustment operations include adding, deleting and modifying risk labels.

[0036] In this embodiment, a multi-level label system is adopted for risk labels. As shown in Table 1, risk labels include first-level labels and second-level labels, and a third-level label is also set in the preferred embodiment. Among them, the first-level label is a type label of the risk label. There is no need to assign a risk value to the first-level label, and it does not participate in the subsequent calculation of the enterprise risk value; the second-level label is a business label, and the third-level label is a further subdivision of the second-level label from the time dimension and the quantity dimension. The second-level label and the third-level label are respectively assigned risk values ​​and participate in the subsequent calculation of the enterprise risk value. It should be noted that in the embodiment, whether to further set a subdivided third-level label for the second-level label and the specific number of subdivided third-level labels are optional. In this embodiment, a three-level label system is adopted to further subdivide the risk labels of the same type, so that the risk value assigned to the risk label can more accurately reflect whether the enterprise entity has a prepaid consumption risk.

[0037]

[0038] Table 1 Schematic diagram of multi-level labeling system

[0039] The label extraction includes processing the basic data table based on the preset data table calculation rules and the pre-trained NLP natural language model to extract the risk labels of the basic data, and the risk labels are contained in the preset label library. The input data of the NLP natural language model is the basic data after data governance processing, and the output data of the NLP natural language model is the risk label data in the label library. In this embodiment, different risk label extraction methods are used for different basic data. For enterprise information data, the data table calculation rules are directly executed based on its corresponding basic data table to extract the corresponding risk labels; for all or part of the governance event data, label extraction is performed based on the pre-trained NLP natural language model. For other data in the governance event data, the above method is selected for processing according to the specific content of concern.

[0040] As mentioned above, for the abnormal business operation labels in the label library, such as restricting high consumption, etc., they can be identified only through the combination of single or multiple fields in the corresponding basic data table to achieve label extraction. Taking the label of the legal person change reaching the first preset value within the first preset time as an example, in this embodiment, the first preset time is within 3 months, and the first preset value is 1 time. The content fields before and after each change and the change time are recorded in the enterprise change record table. Based on the above information, it can be directly determined whether the legal person has been changed within three months, thereby achieving label extraction. Therefore, for this type of information, label extraction can be achieved based on the preset data table calculation rules, and the calculation method is simple.

[0041] For risk labels in the label library, such as abnormal recharge discounts, recharge amounts reaching preset alarm values, and abnormal salary payments, which can only be determined based on semantic recognition of governance events such as complaint data, it is necessary to extract labels based on the pre-trained NLP natural language model. The label extraction process of the NLP natural language model is as follows: Figure 2 As shown. It includes establishing an NLP natural language model, establishing a training data set, and inputting the training data set to perform model training on the NLP natural language model. In this embodiment, the model training includes performing joint training of a masked language model and next sentence prediction based on the training data set; the masked language model specifically randomly masks some words or characters in the input text and predicts the masked content; the next sentence prediction specifically determines whether the second sentence follows the first sentence in a given article.

[0042] In a specific embodiment, the NLP natural language model mainly extracts labels for complaint content in governance event data. The composition of the training data set is shown in Table 2. The complaint data in the governance event data is combined in advance to summarize the purpose of the complaint in the specific complaint content, and associate it with the existing labels to form a complaint content and label association table. Table 2 serves as the NLP natural language model training data set.

[0043]

[0044] Table 2 Training dataset

[0045] In this embodiment, two pre-training methods are used: Masked Language Model and NextSentence Prediction, to pre-train the NLP natural language model so that it can capture the features at the word and sentence levels respectively. In the embodiment, MLM (Masked Language Model) and NSP (NextSentence Prediction) are two different pre-training tasks, and their training can be performed simultaneously. By combining the above two methods for joint training, the NLP natural language model can be helped to learn different aspects of the language, thereby improving the performance of the model.

[0046] Masked Language Model is called Masked Language Model (MLM for short). MLM can be understood as a cloze test. We will randomly mask a certain proportion of words in each sentence and use its context to predict the original words. For example: My name is Li Bai → My [mask], here the "name" is masked, and then the word in the mask position is predicted. During the training process, the NLP natural language model will do the following processing: 80% of the time, [mask] is used, my name is Li Bai -> my [mask] is Li Bai, 10% of the time, a random word is taken to replace the word in [mask], my name is Li Bai -> my idol is Li Bai; 10% of the time, it remains unchanged, my name is Li Bai -> my name is Li Bai. The main reason for this is that the [MASK] tag will not appear in the sentence in the subsequent fine-tuning task, and when predicting a word, the model does not know whether the word in the corresponding position of the input is the correct word, which forces the model to rely more on contextual information to predict words and gives the model a certain error correction ability.

[0047] Next Sentence Prediction is called next sentence prediction (NSP for short), that is, given two sentences in an article, determine whether the second sentence follows the first sentence in the text. In a specific embodiment, 50% of the training set is in accordance with the IsNext (context association) relationship, and the other 50% of the second sentences are randomly extracted from the corpus, and their relationship is NotNext (context not associated), and this relationship is stored. Through these two training methods, a preliminary understanding of the target text can be achieved, and a preliminary association with the training label can be made, and a preliminary result can be output.

[0048] Furthermore, the training of the NLP natural language model includes model fine-tuning. For the label results extracted by the NLP natural language model, the technicians will conduct a preliminary review based on the text semantics and the model extraction results, and fine-tune the label extraction results with low accuracy. For the NSP task, its conditional probability is: P = softmax (CW T ). Fine-tuning means further training the pre-trained model using event data in the government affairs field to adapt to the needs of specific tasks. It is a deep bidirectional pre-trained language model based on the Transformer architecture. It learns rich language representations by pre-training on large-scale unsupervised data. The model fine-tuning method adopts the conventional method in this field. As long as the method can improve the accuracy of the output results of the NLP natural language model, it will not be described in detail here.

[0049] After the label extraction is completed, the enterprise risk profile can be generated by associating the enterprise entity with its hit risk label, such as Figure 3 As shown, the enterprise risk profile includes the enterprise entity mark and all the risk labels it hits. On the one hand, the association between the enterprise entity and the risk label it hits can be directly established based on the association between the basic data tables. On the other hand, the output data of the NLP natural language model in this embodiment also includes enterprise entity data. The NLP natural language model performs entity extraction while executing the label extraction step, that is, extracts brand names, store names, addresses, and company names from the governance event data. The extracted data has both the subject and the risk label. By matching the extracted company name with the company name in the standard enterprise library, a certain risk label can be automatically associated with the enterprise. Through this process, a multi-dimensional enterprise risk profile is carried out for the enterprise of concern, thereby evaluating the business stability of the enterprise.

[0050] In order to evaluate whether the corresponding enterprise subject has prepaid consumption risk based on the risk label hit by the enterprise subject, in this embodiment, the risk labels in the risk label library are freely combined, and each risk label is assigned a different score to form a risk warning model for more than one specific scenario. Preferably, the specific score of the risk label can be adjusted at any time in combination with business experience during use, and the fine-tuning of the score will affect the number and accuracy of the warning. In a specific embodiment, in combination with the above-mentioned three-level risk label system, the free combination of the risk labels can be a free combination of secondary labels and a free combination of the third-level labels under it. For example, in one embodiment, the A scenario risk warning model and the B scenario risk warning model both include an address change label and all the third-level labels under the address change label; in another embodiment, the A scenario risk warning model and the B scenario risk warning model both include an address change label, but there is a difference in the third-level labels under the address change label included; in another embodiment, the A scenario risk warning model includes an address change label, and the address change label only involves the second-level label of the address change label, and is not refined to the third-level label, and the B scenario risk warning model does not include the address change label. In the further example above, the risk warning model of scenario A and the risk warning model of scenario B both include the address change label and all the third-level labels under the address change label, and the risk values ​​corresponding to the address change label and all the third-level labels under the address change label in scenarios A and B may be the same or different.

[0051] The input of the risk warning model is the enterprise risk portrait, and the output of the risk warning model is the risk value of the enterprise entity. Based on the risk value of the enterprise entity, it is determined whether the enterprise entity has prepaid consumption risks. This method adopts a 1+3+N model system, which specifically includes a general model for prepaid cards, 3 specific scenario models, including: beauty and hairdressing scenario model, fitness scenario model and education and training scenario model, and N scalable scenario models. The risk warning model for a specific scenario is constructed by determining the risk label combination of the risk warning model corresponding to the specific scenario and the risk value of the corresponding risk label.

[0052] In this embodiment, for enterprise scenarios where prepaid consumption risks frequently occur, risk warning models for multiple specific scenarios such as payment card general model, education and training scenario model, fitness scenario model and beauty and hairdressing scenario model are constructed in advance. Since the types and emphases of enterprise risk labels involved in scenarios such as education and training institutions, fitness, and beauty and hairdressing are different, the emphasis is reflected in the embodiment through the risk value corresponding to the risk label. The higher the risk value, the higher its importance / impact. Therefore, for risk warning models of different scenarios, the combination of risk labels involved and the label risk value corresponding to the same risk label may be different; and in risk warning models of different scenarios, some risk labels can be divided into secondary labels, and some risk labels can be subdivided into tertiary labels under secondary labels. In one embodiment, some risk labels and label risk values ​​involved in the beauty and hairdressing scenario model are shown in Table 3:

[0053]

[0054] Table 3 Some risk labels and label risk values ​​involved in the beauty salon scenario model

[0055] After obtaining the risk portrait, a risk identification operation is performed to determine the risk warning model of the specific scenario corresponding to the enterprise entity, input the enterprise risk portrait information of the enterprise entity into the risk warning model of the corresponding specific scenario, obtain the risk value of the enterprise entity, and determine whether the enterprise entity has a prepaid consumption risk based on the risk value of the enterprise entity. In an embodiment, the risk warning model for the specific scenario corresponding to the enterprise entity is determined, including determining the risk warning model for the specific scenario corresponding to the enterprise entity based on the business scope in the enterprise entity data. The business scope can be directly obtained from the basic data table associated with the enterprise entity. The risk warning model for a specific scenario can be calculated according to the preset risk label and the corresponding risk value according to the preset calculation rules to obtain the enterprise risk value of the enterprise entity. In this embodiment, the risk value of the secondary label of the risk label hit and the risk value of the corresponding tertiary label are comprehensively determined to determine the final enterprise risk value. In one embodiment, the final enterprise risk value is the sum of the secondary label and the primary label. For example, in this embodiment, Company A is a beauty salon company, and the risk profile of Company A is: a label for the abnormal number of complaints on the government service hotline, a label for service quality issues, and a label for the abnormal number of complaints on the consumer supervision platform; the risk profile of Company A is input into the beauty salon scenario model, and according to the risk values ​​of each risk label preset in Table 3 above, it is finally determined that the enterprise risk value of Company A is 35.

[0056] According to the comparison between the enterprise risk value and the preset warning value of the risk warning model, if the enterprise risk value reaches the preset warning value, it is determined that the corresponding enterprise entity has a prepaid consumption risk. In one embodiment, the preset risk threshold is 11. If the enterprise risk value reaches 11, it is determined that there is a prepaid consumption risk. The enterprise risk value of the above-mentioned enterprise A is 35, and it is determined that enterprise A has a prepaid consumption risk. Further, the risk level is divided according to the enterprise risk value, specifically, multiple risk levels and the risk value range corresponding to each risk level are preset, and the corresponding enterprise risk level is determined according to the range to which the enterprise risk value belongs. The enterprise risk level indicates the degree of prepaid consumption risk of the enterprise entity, which is convenient for the processor to determine the urgency of the risk of the enterprise entity according to the risk level. In this embodiment, the enterprise risk level is divided into three levels: blue warning, yellow warning and red warning according to the risk level from low to high. In one embodiment, when the enterprise risk value is between 11 (inclusive) and 25 (excluding this number), the risk level is blue warning, when the enterprise risk value is between 25 (inclusive) and 60 (excluding this number), the risk level is yellow warning, and when the enterprise risk value is above 60 (inclusive), the risk level is red warning. The enterprise risk value of the above-mentioned enterprise A is 35, so the risk level of enterprise A is yellow warning.

[0057] Furthermore, a manual grading module is included, which allows the person handling the risk to manually modify the risk level of the corresponding enterprise entity according to the actual investigation situation. Furthermore, when building the label library, the events associated with the enterprise risk label are determined in advance, so that the specific risk events of the enterprise entity with risk can be determined.

[0058] In this embodiment, after determining in the risk identification step that the corresponding corporate entity has a prepaid consumption risk, the corresponding corporate entity information and risk information are displayed on the human-computer interaction interface; and / or the corporate entity information and risk information are pushed to a preset processor.

[0059] The enterprise entity information and risk information are pushed to the preset handler, specifically, the contact information of the specific business personnel can be set in advance, including email, phone and other contact methods, and the above information can be directly pushed to the client of the corresponding business personnel. Through this setting, the corresponding business personnel can be reminded to handle the enterprises with prepaid risks in a timely manner.

[0060] The above is the complete execution process of the prepaid consumption risk identification method in one processing cycle in this embodiment. Since the basic data used as the basis for analysis is constantly updated, in this embodiment, the newly added basic data in each time period is obtained according to the preset cycle, and data governance, label extraction, portrait generation and risk identification are performed based on the newly added basic data. In a specific embodiment, one day is a time cycle, and prepaid consumption risk identification is automatically executed as a daily task to timely supervise the enterprises of concern.

[0061] Based on the same technical concept, an embodiment of the present application also provides a prepaid consumption risk warning system, including a server and a client connected in communication, at least one of the server and the client including a memory, a processor and a computer program stored in the memory. When executing the computer program, the processor implements a prepaid consumption risk identification method as described in any of the above embodiments.

[0062] The technical concept of the present application and its corresponding technical solution can target the specific application scenario of prepaid consumption. By summarizing the characteristics of the application scenario, a multi-dimensional analysis can be conducted before and after the event to determine whether the corporate entity has characteristics that meet the risks of prepaid consumption. This can identify prepaid consumption risks in advance and comprehensively, and provide effective auxiliary support for the identification and early warning of prepaid consumption risks.

[0063] The following actual use cases of some prepaid consumption risk warning systems further support the technical effects of this application.

[0064] Case 1:

[0065] Company A is an educational and training institution. At the end of August 2023, the system issued an early warning: Since March 2023, Company A has had operational problems, including arrears in venue fees, abnormal industrial and commercial registration, administrative penalties, and height restrictions on its legal person. In addition, 11 related complaints have been received in succession, and it is determined that the company has a high prepaid consumption risk.

[0066] After receiving the above reminder, the regulatory unit immediately conducted an investigation, classified the enterprise as a "blue" risk warning, and transferred the responsibility unit to carry out disposal. After the risk enterprise has been basically disposed of, after monitoring, no new complaints were found about the enterprise, and the case was closed at the end of December 2023.

[0067] Case 2:

[0068] Company B is a beauty and hairdressing company. In March 2024, the system issued an early warning: Since March 2024, Company B has received three complaints through the government service hotline and two complaints through the consumer supervision platform involving refund issues, and the number of complaints is growing.

[0069] After receiving the above reminder, the regulatory unit quickly understood the situation, confirmed that the store was at risk, and directly rated the risk of Enterprise B as "red" and transferred it for disposal. As of early April 2024, the risks of the enterprise have been basically controlled.

[0070] In addition, after a comparative test of risky enterprises in 2023, the overlap between the online warning enterprises given by the method of this embodiment and the offline daily risk-hit enterprises reached 82%. Since the system involved in the above embodiment was put into operation in October 2023, 346 enterprises that may be at risk have been warned, and 57 risky enterprises have been verified offline, of which 33 enterprises are still operating normally but have prepaid consumption risks, and the early warning rate has reached 57.89%.

[0071] It can be seen that the method and system of the present application have high accuracy and timeliness in prepaid consumption risk identification and early warning, and can greatly improve the efficiency of corresponding law enforcement departments in discovering and handling problems.

[0072] Although the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the claims and their equivalents.

Claims

1. A method for identifying prepaid consumption risks, characterized in that: The method comprises the following steps: Basic data collection, the basic data includes enterprise information data and governance event data in the area of ​​concern, the enterprise information data includes enterprise entity data, enterprise entity change record data and enterprise operation abnormality data, the governance event data includes government service hotline complaint data, consumer supervision platform complaint data, dispute data, and other channels complaint data; Data governance, performing data governance on the basic data according to preset data standards, and associating the governance event data in the basic data with the enterprise information data, and storing the processed basic data in the preset basic data table in the database; associating the governance event data in the basic data with the enterprise information data, including: if the enterprise subject information involved in the governance event data is directly contained in the enterprise information data, then establishing an association directly based on the enterprise subject information; if the enterprise subject information involved in the governance event data is represented by a brand name, then searching for the store address from a third-party map tool based on the brand name, and matching the enterprise registration address contained in the enterprise information data based on the store address; and / or, obtaining the enterprise name corresponding to the brand name from a third-party review website based on the brand name, and matching the enterprise registration name contained in the enterprise information data based on the enterprise name; Label extraction, based on the preset data table calculation rules and the pre-trained NLP natural language model, the basic data table is processed to extract the risk label of the basic data, and the risk label is included in the preset label library; specifically, for enterprise information data, the data table calculation rules are directly executed based on its corresponding basic data table to extract the corresponding risk label; for governance event data, label extraction is performed based on the pre-trained NLP natural language model; the input data of the NLP natural language model is the basic data after data governance processing, and the output data of the NLP natural language model is the risk label data in the label library; risk labels include risk labels including abnormal enterprise labels, high-frequency complaint labels, legal person changes within the first preset time reaching the first preset value label, equity changes within the second preset time, cross-regional address changes, registration place and actual business place inconsistency labels, recharge discount abnormal labels, government service hotline complaints exceeding the second preset value label within the third preset time, consumer supervision platform complaints exceeding the third preset value label within the fourth preset time, refund disputes, service quality issues, and salary payment abnormalities; Profile generation: associate the enterprise subject with the risk tags it hits to generate an enterprise risk profile, which includes the enterprise subject logo and all the risk tags it hits; Early warning model definition: After freely combining the risk labels in the label library and assigning different scores to each risk label, one or more risk early warning models for specific scenarios are formed. The input of the risk early warning model is the enterprise risk profile, and the output of the risk early warning model is the risk value of the enterprise entity. Risk identification: determine the risk warning model for the specific scenario corresponding to the enterprise entity based on the business scope in the enterprise entity data, input the enterprise risk profile information of the enterprise entity into the risk warning model for the corresponding specific scenario, obtain the enterprise risk value of the enterprise entity, and determine whether the enterprise entity has prepaid consumption risks based on the enterprise risk value of the enterprise entity; It also includes establishing an NLP natural language model, inputting training data, wherein the training data includes an association table of complaint content and risk labels, and performing joint training of a masked language model and next sentence prediction based on the training data set; the masked language model specifically randomly masks some words or characters in the input text and predicts the masked content; the next sentence prediction specifically determines whether the second sentence follows the first sentence in a given article; and training the NLP natural language model includes model fine-tuning.

2. A prepaid consumption risk identification method according to claim 1, characterized in that: The risk warning models include a general model for prepaid cards, an education and training scenario model, a fitness scenario model, and a beauty and hairdressing scenario model.

3. A prepaid consumption risk identification method according to claim 1, characterized in that: Acquire new basic data according to the preset time period, and perform data governance, label extraction, profile generation and risk identification based on the new basic data.

4. A prepaid consumption risk identification method according to claim 1, characterized in that: It also includes early warning generation. After the risk identification step determines that the corresponding corporate entity has prepaid consumption risks, the corresponding corporate entity information and risk information are displayed on the human-computer interaction interface; And / or push the enterprise entity information and risk information to the preset processor.

5. According to a prepaid consumption risk identification method described in claim 1, the risk label is divided into three levels: primary label, secondary label and tertiary label, wherein the primary label is the type classification of the risk label; the secondary label is the business label, and the tertiary label is a subdivision of the secondary label, and the secondary label and the tertiary label are respectively assigned label risk values.

6. A prepaid consumption risk warning system, comprising a server and a client connected in communication, at least one of the server and the client comprising a memory, a processor and a computer program stored in the memory, characterized in that: The processor implements the method according to any one of claims 1 to 5 when executing the computer program.

Images