Financial reimbursement user portrait optimization method, device, equipment and storage medium

Abstract

The present invention discloses a method, device, equipment and storage medium for optimizing user portraits of financial reimbursements. The method obtains basic portraits and user historical reimbursement records related to user financial reimbursement, creates user historical pseudo-labels according to the user historical reimbursement records; obtains sample data of user financial reimbursement, pre-processes the sample data, obtains user feature vectors, and trains a user portrait optimization algorithm model based on a deep neural network according to the user historical pseudo-labels and the user feature vectors; optimizes the basic portrait according to the trained user portrait optimization algorithm model to obtain an optimized target user portrait, which can narrow the collection scope of user sensitive information, reduce the amount of data processing, improve the performance of the user portrait optimization model, realize the automatic processing of user portraits, improve the accuracy of user portraits, and improve the speed and efficiency of optimizing financial reimbursement user portraits.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a method, device, equipment and storage medium for optimizing a financial reimbursement user portrait. Background Art

[0002] User portraits, as an effective tool for portraying target users and linking user demands with design directions, have been widely used in various fields. In actual operations, user portraits are often manually labeled, rule-based, or model-based, with users labeled with simple, easy-to-understand, or life-like phrases or phrases, and these labels are used as virtual representatives of users, thereby associating user attributes, behaviors, and other characteristics with the expected results.

[0003] Reimbursement business is an important part of the financial management of an enterprise, with the characteristics of high frequency of business occurrence, large number of documents involved, and wide coverage of employees. The use of user portraits by enterprises to conduct refined management of employee financial reimbursement is of great significance in reducing costs and increasing efficiency. However, the user portrait construction method for general application scenarios has problems in the field of financial reimbursement, such as low correlation between portraits and business, low effective information density, and poor real-time application capabilities.

[0004] The existing solution is to build a professional database, confirm the portrait based on a large amount of original information, and ask the user to confirm. However, this solution involves a lot of professional background knowledge in the process of building a professional database, and requires the assistance of professionals to complete the construction of the user portrait. In addition, the user's structured and unstructured data in the original information are not fully integrated, and it is difficult to discover the high-dimensional feature information that intersects between different labels. There may be inaccurate results in the overall prediction results. Another solution obtains user data from multiple public service industries to build a topological structure framework of the user portrait, and constructs the user portrait based on the framework and business information. The user portrait constructed by this solution only realizes information aggregation, which is not intuitive enough in the actual application process. In addition, the design data volume is large, and the burden of running in real-time computing load is heavy, and there is a risk of user privacy leakage. Summary of the invention

[0005] The main purpose of the present invention is to provide a financial reimbursement user portrait optimization method, device, equipment and storage medium, aiming to solve the technical problems in the prior art that user portrait construction fails to fully utilize the high-dimensional cross-features of structured and unstructured original data, the user portrait accuracy is difficult to support business applications, the portrait construction is highly complex, the calculation amount is too large, and there is a risk of user privacy leakage.

[0006] In a first aspect, the present invention provides a method for optimizing a user portrait of financial reimbursement, the method comprising the following steps:

[0007] Obtaining a basic profile of the user related to financial reimbursement and the user's historical reimbursement records, and creating a user historical pseudo-label based on the user's historical reimbursement records;

[0008] Obtaining sample data of user financial reimbursement, preprocessing the sample data to obtain a user feature vector, and training a user portrait optimization algorithm model based on a deep neural network according to the user historical pseudo-label and the user feature vector to obtain a trained user portrait optimization algorithm model;

[0009] The basic portrait is optimized according to the trained user portrait optimization algorithm model to obtain an optimized target user portrait.

[0010] Optionally, the obtaining of a basic profile related to financial reimbursement of the user and a historical reimbursement record of the user, and creating a historical pseudo-label of the user according to the historical reimbursement record of the user, includes:

[0011] Obtaining user basic information, reimbursement document records and reimbursement approval records, obtaining the declaration manual number from the user basic information, and obtaining the corresponding document identifier from the reimbursement document records and the reimbursement approval records;

[0012] The reimbursement document record and the reimbursement approval record are associated according to the document identifier, and the user basic information is associated with the reimbursement document record and the reimbursement approval record according to the declaration manual number to generate a basic profile related to the user's financial reimbursement;

[0013] Obtain the user's historical reimbursement records, and create a user's historical pseudo-label according to the user's historical reimbursement records.

[0014] Optionally, the acquiring of the user's historical reimbursement records and the creation of the user's historical pseudo-labels according to the user's historical reimbursement records include:

[0015] Obtain the user's historical reimbursement records, and obtain the user's historical reimbursement one-time pass rate and the average time required for the reimbursement process based on the user's historical reimbursement records;

[0016] Obtaining the reimbursement amount and the time taken for the reimbursement process of the sample document, and determining the user type score of the sample document according to the reimbursement amount and the time taken for the reimbursement process;

[0017] The sample documents are sorted according to the user type score, the user's historical reimbursement one-time pass rate and the average time consumption of the reimbursement process, and the user history pseudo-label is created according to the sorting result.

[0018] Optionally, obtaining the reimbursement amount and the time taken for the reimbursement process of the sample document, and determining the user type score of the sample document according to the reimbursement amount and the time taken for the reimbursement process, includes:

[0019] The reimbursement amount and the time taken for the reimbursement process of the sample document are obtained, and the user type score of the sample document is obtained by calculating the reimbursement amount and the time taken for the reimbursement process through the following formula:

[0020]

[0021] Among them, M is the reimbursement amount of the sample document, T is the time consumption of the reimbursement process of the sample document, and S is the user type score.

[0022] Optionally, the acquiring sample data of user financial reimbursement, preprocessing the sample data to obtain a user feature vector, and training a user portrait optimization algorithm model based on a deep neural network according to the user historical pseudo-label and the user feature vector to obtain a trained user portrait optimization algorithm model includes:

[0023] Obtain sample data of user financial reimbursement, perform data preprocessing in different ways on numerical data and label data in the sample data, and obtain processed user portrait data;

[0024] splicing the user portrait data to obtain a user feature vector;

[0025] Importing the user feature vector into the input layer of the user portrait optimization algorithm model based on a deep neural network to obtain the optimized label output by the output layer;

[0026] Compare the historical pseudo labels and the optimized labels corresponding to the original portrait samples of the same user in the sample data, and store the original portrait samples with inconsistent label comparison results in a difficult example database;

[0027] The user portrait optimization algorithm model is trained and optimized according to the difficult example database until the user portrait optimization algorithm model meets the preset performance conditions, and a trained user portrait optimization algorithm model is generated.

[0028] Optionally, the step of concatenating the user portrait data to obtain a user feature vector includes:

[0029] The label data in the user portrait data that has been processed by one-hot encoding are concatenated head to tail in sequence, the concatenated one-hot encoding is reduced in dimension using embedded encoding, and the reduced-dimensional label features are combined with the numerical features and other label features in the user portrait data to obtain a user feature vector.

[0030] Optionally, the optimizing the basic portrait according to the trained user portrait optimization algorithm model to obtain an optimized target user portrait includes:

[0031] The data not containing pseudo labels in the basic portrait is optimized according to the trained user portrait optimization algorithm model to obtain an optimized target user portrait.

[0032] In a second aspect, in order to achieve the above-mentioned purpose, the present invention further proposes a financial reimbursement user portrait optimization device, the financial reimbursement user portrait optimization device comprising:

[0033] A pseudo-label creation module is used to obtain a basic profile of a user related to financial reimbursement and a historical reimbursement record of the user, and to create a historical pseudo-label of the user according to the historical reimbursement record of the user;

[0034] A model training module is used to obtain sample data of user financial reimbursement, pre-process the sample data to obtain a user feature vector, and train a user portrait optimization algorithm model based on a deep neural network according to the user historical pseudo-label and the user feature vector to obtain a trained user portrait optimization algorithm model;

[0035] The portrait optimization module is used to optimize the basic portrait according to the trained user portrait optimization algorithm model to obtain an optimized target user portrait.

[0036] In the third aspect, to achieve the above-mentioned purpose, the present invention also proposes a financial reimbursement user portrait optimization device, the financial reimbursement user portrait optimization device comprising: a memory, a processor, and a financial reimbursement user portrait optimization program stored on the memory and executable on the processor, the financial reimbursement user portrait optimization program being configured to implement the steps of the financial reimbursement user portrait optimization method as described above.

[0037] In a fourth aspect, in order to achieve the above-mentioned purpose, the present invention also proposes a storage medium, on which a financial reimbursement user portrait optimization program is stored, and when the financial reimbursement user portrait optimization program is executed by a processor, the steps of the financial reimbursement user portrait optimization method described above are implemented.

[0038] The financial reimbursement user portrait optimization method proposed in the present invention obtains the basic portrait related to the user's financial reimbursement and the user's historical reimbursement record, creates the user's historical pseudo-label according to the user's historical reimbursement record; obtains sample data of the user's financial reimbursement, pre-processes the sample data to obtain the user's feature vector, and trains the user portrait optimization algorithm model based on the deep neural network according to the user's historical pseudo-label and the user's feature vector to obtain the trained user portrait optimization algorithm model; optimizes the basic portrait according to the trained user portrait optimization algorithm model to obtain the optimized target user portrait, which can reduce the The scope of collection of user sensitive information reduces the amount of data transmitted in the system, reduces server storage and computing burdens, improves the performance of the user portrait optimization model, comprehensively evaluates the category score of the document in the financial reimbursement process based on the reimbursement amount and reimbursement time, determines the training and evaluation targets of the subsequent optimization model, and pre-processes the sample data, which is conducive to the rapid convergence of the algorithm model. At the same time, it also desensitizes the user's privacy data to a certain extent, realizes the automated processing of user portraits, improves the accuracy of user portraits, facilitates the online call of user portraits by other business modules, and improves the speed and efficiency of financial reimbursement user portrait optimization. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 A schematic diagram of the device structure of the hardware operating environment involved in the embodiment of the present invention;

[0040] Figure 2 A schematic diagram of the process of the first embodiment of the financial reimbursement user portrait optimization method of the present invention;

[0041] Figure 3 A schematic diagram of the flow chart of the second embodiment of the method for optimizing the user portrait of financial reimbursement of the present invention;

[0042] Figure 4 A schematic diagram of the flow chart of the third embodiment of the method for optimizing the user portrait of financial reimbursement of the present invention;

[0043] Figure 5 A schematic diagram of a flow chart of a fourth embodiment of a method for optimizing a user profile for financial reimbursement according to the present invention;

[0044] Figure 6 A schematic diagram of generating feature vectors in the financial reimbursement user portrait optimization method of the present invention;

[0045] Figure 7 A flowchart of a fifth embodiment of the method for optimizing user portraits for financial reimbursement according to the present invention;

[0046] Figure 8 This is a functional module diagram of the first embodiment of the financial reimbursement user portrait optimization device of the present invention.

[0047] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0048] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0049] The solution of the embodiment of the present invention is mainly: by obtaining the basic portrait related to the user's financial reimbursement and the user's historical reimbursement records, the user's historical pseudo-label is created according to the user's historical reimbursement records; the sample data of the user's financial reimbursement is obtained, the sample data is pre-processed to obtain the user feature vector, and the user portrait optimization algorithm model based on the deep neural network is trained according to the user's historical pseudo-label and the user feature vector to obtain the trained user portrait optimization algorithm model; the basic portrait is optimized according to the trained user portrait optimization algorithm model to obtain the optimized target user portrait, which can narrow the scope of collection of user sensitive information, reduce the amount of data transmission in the system, reduce the server storage and computing burden, and improve the user The performance of the portrait optimization model is evaluated by comprehensively evaluating the category score of the document in the financial reimbursement process based on the reimbursement amount and reimbursement time, determining the training and evaluation targets of the subsequent optimization model, and preprocessing the sample data, which is conducive to the rapid convergence of the algorithm model. At the same time, the user privacy data is desensitized to a certain extent, realizing the automated processing of user portraits, improving the accuracy of user portraits, facilitating the online call of user portraits by other business modules, and improving the speed and efficiency of financial reimbursement user portrait optimization. It solves the technical problems in the existing technology that user portrait construction fails to fully utilize the high-dimensional cross-features of structured and unstructured raw data, the user portrait accuracy is difficult to support business applications, the portrait construction is complex, the calculation amount is too large, and there is a risk of user privacy leakage.

[0050] Reference Figure 1 , Figure 1 The figure is a schematic diagram of the device structure of the hardware operating environment involved in the embodiment of the present invention.

[0051] like Figure 1As shown, the device may include: a processor 1001, such as a CPU, a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. Among them, the communication bus 1002 is used to realize the connection and communication between these components. The user interface 1003 may include a display screen (Display), an input unit such as a keyboard (Keyboard), and the optional user interface 1003 may also include a standard wired interface and a wireless interface. The network interface 1004 may optionally include a standard wired interface and a wireless interface (such as a Wi-Fi interface). The memory 1005 may be a high-speed RAM memory, or it may be a stable memory (Non-Volatile Memory), such as a disk memory. The memory 1005 may optionally be a storage device independent of the aforementioned processor 1001.

[0052] Those skilled in the art will understand that Figure 1 The device structure shown in the figure does not constitute a limitation of the device, and may include more or less components than shown in the figure, or combine certain components, or arrange the components differently.

[0053] like Figure 1 As shown, the memory 1005 as a storage medium may include an operating device, a network communication module, a user interface module, and a financial reimbursement user portrait optimization program.

[0054] The device of the present invention calls the financial reimbursement user portrait optimization program stored in the memory 1005 through the processor 1001, and performs the following operations:

[0055] Obtaining a basic profile of the user related to financial reimbursement and the user's historical reimbursement records, and creating a user historical pseudo-label based on the user's historical reimbursement records;

[0056] Obtaining sample data of user financial reimbursement, preprocessing the sample data to obtain a user feature vector, and training a user portrait optimization algorithm model based on a deep neural network according to the user historical pseudo-label and the user feature vector to obtain a trained user portrait optimization algorithm model;

[0057] The basic portrait is optimized according to the trained user portrait optimization algorithm model to obtain an optimized target user portrait.

[0058] The device of the present invention calls the financial reimbursement user portrait optimization program stored in the memory 1005 through the processor 1001, and further performs the following operations:

[0059] Obtaining user basic information, reimbursement document records and reimbursement approval records, obtaining the declaration manual number from the user basic information, and obtaining the corresponding document identifier from the reimbursement document records and the reimbursement approval records;

[0060] The reimbursement document record and the reimbursement approval record are associated according to the document identifier, and the user basic information is associated with the reimbursement document record and the reimbursement approval record according to the declaration manual number to generate a basic profile related to the user's financial reimbursement;

[0061] Obtain the user's historical reimbursement records, and create a user's historical pseudo-label according to the user's historical reimbursement records.

[0062] The device of the present invention calls the financial reimbursement user portrait optimization program stored in the memory 1005 through the processor 1001, and further performs the following operations:

[0063] Obtain the user's historical reimbursement records, and obtain the user's historical reimbursement one-time pass rate and the average time required for the reimbursement process based on the user's historical reimbursement records;

[0064] Obtaining the reimbursement amount and the time taken for the reimbursement process of the sample document, and determining the user type score of the sample document according to the reimbursement amount and the time taken for the reimbursement process;

[0065] The sample documents are sorted according to the user type score, the user's historical reimbursement one-time pass rate and the average time consumption of the reimbursement process, and the user history pseudo-label is created according to the sorting result.

[0066] The device of the present invention calls the financial reimbursement user portrait optimization program stored in the memory 1005 through the processor 1001, and further performs the following operations:

[0067] The reimbursement amount and the time taken for the reimbursement process of the sample document are obtained, and the user type score of the sample document is obtained by calculating the reimbursement amount and the time taken for the reimbursement process through the following formula:

[0068]

[0069] Among them, M is the reimbursement amount of the sample document, T is the time consumption of the reimbursement process of the sample document, and S is the user type score.

[0070] The device of the present invention calls the financial reimbursement user portrait optimization program stored in the memory 1005 through the processor 1001, and further performs the following operations:

[0071] Obtain sample data of user financial reimbursement, perform data preprocessing in different ways on numerical data and label data in the sample data, and obtain processed user portrait data;

[0072] splicing the user portrait data to obtain a user feature vector;

[0073] Importing the user feature vector into the input layer of the user portrait optimization algorithm model based on a deep neural network to obtain the optimized label output by the output layer;

[0074] Compare the historical pseudo labels and the optimized labels corresponding to the original portrait samples of the same user in the sample data, and store the original portrait samples with inconsistent label comparison results in a difficult example database;

[0075] The user portrait optimization algorithm model is trained and optimized according to the difficult example database until the user portrait optimization algorithm model meets the preset performance conditions, and a trained user portrait optimization algorithm model is generated.

[0076] The device of the present invention calls the financial reimbursement user portrait optimization program stored in the memory 1005 through the processor 1001, and further performs the following operations:

[0077] The label data in the user portrait data that has been processed by one-hot encoding are concatenated head to tail in sequence, the concatenated one-hot encoding is reduced in dimension using embedded encoding, and the reduced-dimensional label features are combined with the numerical features and other label features in the user portrait data to obtain a user feature vector.

[0078] The device of the present invention calls the financial reimbursement user portrait optimization program stored in the memory 1005 through the processor 1001, and further performs the following operations:

[0079] The data not containing pseudo labels in the basic portrait is optimized according to the trained user portrait optimization algorithm model to obtain an optimized target user portrait.

[0080] Through the above scheme, this embodiment obtains basic portraits and user historical reimbursement records related to user financial reimbursement, creates user historical pseudo labels according to the user historical reimbursement records; obtains sample data of user financial reimbursement, pre-processes the sample data, obtains user feature vectors, and trains a user portrait optimization algorithm model based on a deep neural network according to the user historical pseudo labels and the user feature vectors to obtain a trained user portrait optimization algorithm model; optimizes the basic portrait according to the trained user portrait optimization algorithm model to obtain an optimized target user portrait, which can narrow the collection scope of user sensitive information, reduce the amount of data transmitted in the system, reduce server storage and computing burden, and improve the performance of the user portrait optimization model. The category score of the document in the financial reimbursement process is comprehensively evaluated from the reimbursement amount and reimbursement time, and the training and evaluation targets of the subsequent optimization model are determined. The sample data is pre-processed, which is conducive to the rapid convergence of the algorithm model. At the same time, the user privacy data is desensitized to a certain extent, and the automatic processing of the user portrait is realized, the accuracy of the user portrait is improved, and it is convenient for other business modules to call the user portrait online, and the speed and efficiency of the optimization of the financial reimbursement user portrait are improved.

[0081] Based on the above hardware structure, an embodiment of the financial reimbursement user portrait optimization method of the present invention is proposed.

[0082] Reference Figure 2 , Figure 2 This is a flow chart of the first embodiment of the financial reimbursement user portrait optimization method of the present invention.

[0083] In a first embodiment, the financial reimbursement user portrait optimization method includes the following steps:

[0084] Step S10: Obtain a basic profile related to the user's financial reimbursement and the user's historical reimbursement records, and create a user historical pseudo-label based on the user's historical reimbursement records.

[0085] It should be noted that the basic portrait related to the user's financial reimbursement includes but is not limited to information such as user basic information, reimbursement document records and reimbursement approval records. The user's historical reimbursement records are the reimbursement records of the target user in the past period of time. The user's historical reimbursement records can be used to create a user's historical pseudo-label.

[0086] Step S20: Obtain sample data of user financial reimbursement, preprocess the sample data to obtain a user feature vector, and train a user portrait optimization algorithm model based on a deep neural network according to the user historical pseudo-label and the user feature vector to obtain a trained user portrait optimization algorithm model.

[0087] It can be understood that the sample data of user financial reimbursement is the training data for training the user portrait optimization algorithm model based on deep neural network. By preprocessing the sample data, mutual interference between data of different orders of magnitude and different measurement units can be avoided, thereby obtaining the user feature vector, and the user portrait optimization algorithm model based on deep neural network is trained according to the user historical pseudo-label and the user feature vector to obtain the trained user portrait optimization algorithm model.

[0088] Step S30: Optimize the basic portrait according to the trained user portrait optimization algorithm model to obtain an optimized target user portrait.

[0089] It should be understood that the basic portrait is optimized according to the trained user portrait optimization algorithm model to obtain the optimized target user portrait, thereby realizing the automated processing of the user portrait, improving the accuracy of the user portrait, and facilitating the online calling of the user portrait by other business modules.

[0090] Through the above scheme, this embodiment obtains basic portraits and user historical reimbursement records related to user financial reimbursement, creates user historical pseudo labels according to the user historical reimbursement records; obtains sample data of user financial reimbursement, pre-processes the sample data, obtains user feature vectors, and trains a user portrait optimization algorithm model based on a deep neural network according to the user historical pseudo labels and the user feature vectors to obtain a trained user portrait optimization algorithm model; optimizes the basic portrait according to the trained user portrait optimization algorithm model to obtain an optimized target user portrait, which can narrow the collection scope of user sensitive information, reduce the amount of data transmitted in the system, reduce server storage and computing burden, and improve the performance of the user portrait optimization model. The category score of the document in the financial reimbursement process is comprehensively evaluated from the reimbursement amount and reimbursement time, and the training and evaluation targets of the subsequent optimization model are determined. The sample data is pre-processed, which is conducive to the rapid convergence of the algorithm model. At the same time, the user privacy data is desensitized to a certain extent, and the automatic processing of the user portrait is realized, the accuracy of the user portrait is improved, and it is convenient for other business modules to call the user portrait online, and the speed and efficiency of the optimization of the financial reimbursement user portrait are improved.

[0091] Further, Figure 3 Schematic diagram of the flow of the second embodiment of the method for optimizing the user portrait of financial reimbursement of the present invention. Figure 3 As shown, a second embodiment of the financial reimbursement user portrait optimization method of the present invention is proposed based on the first embodiment. In this embodiment, step S10 specifically includes the following steps:

[0092] Step S11, obtaining user basic information, reimbursement document records and reimbursement approval records, obtaining the declaration manual number from the user basic information, and obtaining the corresponding document identifier from the reimbursement document records and the reimbursement approval records.

[0093] It should be noted that after obtaining the user's basic information, reimbursement document records and reimbursement approval records, the declaration number can be obtained from the user's basic information, and the corresponding document identifier can be obtained from the reimbursement document records and the reimbursement approval records for subsequent generation of basic portraits.

[0094] Step S12: associate the reimbursement document record and the reimbursement approval record according to the document identifier, and associate the user basic information with the reimbursement document record and the reimbursement approval record according to the reporting manual number to generate a basic portrait related to the user's financial reimbursement.

[0095] It is understandable that the user's work number, user gender, date of birth, education level, current position, employment time and department are obtained from the user's basic information. Table 1 is an illustration of the user's basic information.

[0096] Table 1

[0097] Work No. gender Date of Birth Education Current rank Date of employment department 12348181234 male 1990.08.31 Undergraduate Level 12 2010.06.01 RJ 12348181235 female 1990.07.30 master Level 10 2015.09.01 ZX 12348181236 male 1990.06.29 PhD Level 9 2020.03.01 CW 12348181237 female 1990.05.28 Undergraduate Level 11 2010.06.01 DQ 12348181238 female 1990.04.27 PhD Level 9 2020.09.01 RS

[0098] The document ID, reporting personnel number, business type, business trip date, destination, business trip duration, reimbursement amount and other information are obtained from the reimbursement document record. Table 2 is a schematic diagram of the user's previously completed reimbursement document record information.

[0099] Table 2

[0100] Document identification Report a manual number Business Type Travel Date destination Business trip duration Reimbursement amount 20200001 12348181234 Management 2020.01.01 Xiangyang 2 days 1500 20200002 12348181235 Training 2020.01.01 Shanghai 3 days 3000 20200003 12348181236 Project Class 2020.01.01 Mohe 7 days 10000 20200004 12348181237 Project Class 2020.01.01 Xiangyang 2 days 1500 20200005 12348181238 Training 2020.01.01 Shanghai 3 days 3000

[0101] From the reimbursement approval record, obtain the document ID, reporting number, reimbursement process start date, reimbursement process end date, reimbursement process time, whether rejection has occurred, and other information. Table 3 shows an example of the reimbursement approval record information previously completed by the relevant user.

[0102] Table 3

[0103]

[0104] The time consumption of the reimbursement process in Table 3 is calculated by the following formula:

[0105] t cost =T end -T start

[0106] Among them, t costRepresents the time taken for the reimbursement process in the record, in days, T end Represents the end date of the reimbursement process in the record, T start Represents the date the reimbursement process started in the record.

[0107] The document identification can be used to link the information of the reimbursement document record and the reimbursement approval record. The work number can be used to link the user's basic information and document-related information. After completing the above-mentioned related work, the collection and construction of the user's basic portrait information is completed.

[0108] Step S13: Obtain the user's historical reimbursement records, and create a user history pseudo-label based on the user's historical reimbursement records.

[0109] It should be understood that after obtaining the user's historical reimbursement records, a user history pseudo-label may be created based on the relevant information of the user's historical reimbursement records.

[0110] It should be noted that in order to train deep neural network algorithm models, it is often necessary to obtain the true value labels corresponding to the samples. However, in actual scenarios, only virtual labels that represent the real situation to a certain extent can be obtained based on historical statistical information. Therefore, when training the algorithm, such labels are called pseudo labels.

[0111] This embodiment adopts the above scheme, by obtaining user basic information, reimbursement document records and reimbursement approval records, obtaining the declaration manual number from the user basic information, and obtaining the corresponding document identification from the reimbursement document record and the reimbursement approval record; associating the reimbursement document record and the reimbursement approval record according to the document identification, and associating the user basic information with the reimbursement document record and the reimbursement approval record according to the declaration manual number, so as to generate a basic portrait related to the user's financial reimbursement; obtaining the user's historical reimbursement records, and creating a user's historical pseudo-label according to the user's historical reimbursement records, which can narrow the scope of collection of user sensitive information, reduce the amount of data transmitted in the system, reduce the server storage and computing burden, and improve the speed and efficiency of optimizing the financial reimbursement user portrait.

[0112] Further, Figure 4 FIG. 1 is a flow chart of a third embodiment of the method for optimizing the user portrait of financial reimbursement according to the present invention. Figure 4 As shown, a third embodiment of the financial reimbursement user portrait optimization method of the present invention is proposed based on the second embodiment. In this embodiment, step S13 specifically includes the following steps:

[0113] Step S131, obtaining the user's historical reimbursement records, and obtaining the user's historical reimbursement one-time pass rate and the average time consumed in the reimbursement process according to the user's historical reimbursement records.

[0114] It should be noted that in order to obtain relevant information on the quality of user reimbursement document filling and process efficiency, the user's historical reimbursement one-time pass rate and the average time required for the reimbursement process can be calculated based on the relevant information in the user's historical reimbursement records.

[0115] The specific calculation can be obtained through the following formula:

[0116]

[0117]

[0118] Among them, p once represents the one-time pass rate of the user's historical reimbursement, n represents the total number of historical reimbursement documents (in order to ensure the timeliness of user data, only the last 10 records of the user are often selected as the total number, and all records are taken if there are less than 10 records), n pass Represents the number of historical reimbursement documents that have not been rejected, that is, the total number of records in which "Has rejection occurred" is "No"; Represents the average time taken by the user's historical reimbursement process, in days, n represents the total number of historical reimbursement documents, t i Represents the time taken for the reimbursement process of each reimbursement document. The above calculation results are used as feature values ​​for algorithm model training. The algorithm model will learn the characteristics of users in the financial reimbursement process more quickly, speeding up the model convergence speed while improving the accuracy of the optimized user portrait.

[0119] Step S132: Obtain the reimbursement amount and the time required for the reimbursement process of the sample document, and determine the user type score of the sample document based on the reimbursement amount and the time required for the reimbursement process.

[0120] It is understandable that after obtaining the reimbursement amount of the sample document and the time taken for the reimbursement process, the user type rating of the sample document can be performed.

[0121] Furthermore, the step S132 specifically includes the following steps:

[0122] The reimbursement amount and the time taken for the reimbursement process of the sample document are obtained, and the user type score of the sample document is obtained by calculating the reimbursement amount and the time taken for the reimbursement process through the following formula:

[0123]

[0124] Among them, M is the reimbursement amount of the sample document, T is the time consumption of the reimbursement process of the sample document, and S is the user type score.

[0125] It can be understood that this embodiment provides an example of assigning a user type pseudo-label to each group of data samples based on the amount of the user's historical reimbursement documents and the time required for the reimbursement process. The relevant calculation formula and method are shown above.

[0126] Step S133: sort the sample documents according to the user type score, the user's historical reimbursement one-time pass rate and the average time consumption of the reimbursement process, and create a user history pseudo label according to the sorting result.

[0127] It should be understood that after obtaining the user type score through the above calculation formula, it is arranged in ascending order according to the score, and the top 20%, 20%-40%, 40%-60%, 60%-80%, 80%-100% of the samples are assigned "Level I", "Level II", "Level III", "Level IV", "Level V" and other 5 types of pseudo labels, thereby completing the creation of sample pseudo labels. Of course, it can also be set to other levels or categories corresponding to different quantities and ranges, that is, the 5 in the formula can be other values, and this embodiment is not limited to this.

[0128] Furthermore, the user type score can be further sorted and adjusted according to the user's historical one-time reimbursement pass rate and the average time consumed in the reimbursement process, that is, samples with higher pass rates and less time consumption are placed in front, and the pseudo-labels for sample classification are adjusted. Of course, other sorting and adjustment methods are also possible, and this embodiment does not limit this. By calculating the user's historical one-time pass rate and average reimbursement time, it is helpful to extract the user's reporting quality and reimbursement efficiency information in the financial reimbursement business, which is conducive to accelerating model convergence and improving the performance of the user portrait optimization model.

[0129] In a specific implementation, optionally, there are other possible pseudo-label creation methods, as described below:

[0130] Alternative 1: Pseudo-label creation based on time-consuming reimbursement process

[0131] Another possible example of pseudo-label creation method is to directly use 7 days as a unit and mark the user documents whose reimbursement process takes "less than 7 days", "greater than or equal to 7 days and less than 14 days", "greater than or equal to 14 days and less than 21 days", "greater than or equal to 21 days and less than 28 days", and "greater than or equal to 28 days" as "Level I" to "Level V" respectively.

[0132] Alternative 2: Pseudo-label creation solution based on machine learning unsupervised clustering algorithm

[0133] Another possible example of a pseudo-label creation method is to use known relevant data information of the sample other than the pseudo-label data to form a feature vector after preprocessing, and use but not limited to principal component analysis (PCA), linear discriminant analysis (LDA), singular value decomposition (SVD) and other dimensionality reduction algorithms to reduce the dimension of the feature vector, and then use but not limited to K-means (K means), density-based spatial clustering of applications with noise (DBSCAN), cluster generative adversarial nets (ClusterGAN) and other clustering algorithms to cluster and label the samples, so as to complete the creation of pseudo-labels for the original samples.

[0134] This embodiment adopts the above scheme to obtain the user's historical reimbursement records, and obtain the user's historical reimbursement one-time pass rate and the average time of the reimbursement process based on the user's historical reimbursement records; obtain the reimbursement amount and reimbursement process time of the sample document, and determine the user type score of the sample document according to the reimbursement amount and the reimbursement process time; sort the sample documents according to the user type score, the user's historical reimbursement one-time pass rate and the average time of the reimbursement process, and create user history pseudo-labels based on the sorting results, which can narrow the scope of collection of user sensitive information, reduce the amount of data transmitted in the system, reduce the server storage and computing burden, and improve the speed and efficiency of financial reimbursement user portrait optimization.

[0135] Further, Figure 5 Schematic diagram of the fourth embodiment of the method for optimizing the user portrait of financial reimbursement according to the present invention. Figure 5 As shown, a fourth embodiment of the financial reimbursement user portrait optimization method of the present invention is proposed based on the first embodiment. In this embodiment, step S20 specifically includes the following steps:

[0136] Step S21: Obtain sample data of user financial reimbursement, perform data preprocessing in different ways on the numerical data and label data in the sample data, and obtain processed user portrait data.

[0137] It should be noted that sample data cannot be directly used as sample features for training algorithm models. In order to avoid mutual interference between data of different orders of magnitude and different measurement units, and to solve the problem that non-numerical data is difficult to input into the model as features, it is necessary to perform corresponding preprocessing according to the data type. By performing different methods of data preprocessing on the numerical data and label data in the sample data, the processed user portrait data can be obtained.

[0138] In the specific implementation, in the example of the previous step, the duration of the business trip, the reimbursement amount, the time taken for the reimbursement process, etc. are all numerical features, and the numerical calculation result information such as the user's age obtained by subtracting the date of birth from the current date is also a numerical feature; the preprocessing methods that can be used for numerical features usually include but are not limited to standardization, normalization, scaling, binning, etc. Generally, one or more of the above preprocessing methods can be selected for each numerical feature for processing; it is worth noting that the features after the numerical features have completed the binning preprocessing can also be regarded as label features for processing.

[0139] It is understandable that in the examples described above, gender, education, current job level, department, business type, destination, etc. are all label features; some features may be recorded in numbers, but they are only used to mark categories and the numerical value has no actual meaning. This is also a label feature. Label features are difficult to use for model training before encoding. Usually, the preprocessing of encoding label features includes but is not limited to dummy variables, one-hot, ordinal, frequency, embedding, etc. Generally, only one of the above preprocessing methods is selected for each label feature, and a combination of them can be used according to actual needs to improve the utilization rate and information density of label features; it is worth noting that the encoding method of some label features includes numerical attributes, and the encoded label features can also be processed and calculated as numerical features.

[0140] It should be understood that, using the above data preprocessing method, corresponding processing is performed on the acquired user basic portrait data respectively; as shown in Table 4, it is an example of a preprocessing solution for the basic portrait data of financial reimbursement users.

[0141] Table 4

[0142] Feature Name Feature Category Preprocessing method Work No. Label Delete Document identification Label Delete gender Label One-hot encoding age Numeric Equal frequency binning Education Label One-hot encoding Current position Label Ordinal encoding Length of employment Numeric Decision Tree Binning department Label One-hot encoding Business Type Label One-hot encoding Business trip duration Numeric Decision Tree Binning destination Label Frequency Coding Reimbursement amount Numeric Box-Cox transformation Time-consuming reimbursement Numeric Equal frequency binning User one-time pass rate Numeric Max-Min Transform Average time spent on user reimbursement Numeric Box-Cox transformation

[0143] It should be noted that the user data processed by the above solution has been desensitized to a certain extent. In order to further protect user privacy, the user's relevant information can be further offset and desensitized on the basis of the above, and the relevant data can be decentralized, anonymous and encrypted using blockchain means.

[0144] Step S22: splicing the user portrait data to obtain a user feature vector.

[0145] It should be understood that the user portrait data needs to be spliced ​​to form a feature vector that can be used for algorithm model training, that is, after splicing the user portrait data, a user feature vector can be obtained as the input of the training model.

[0146] Furthermore, the step S22 specifically includes the following steps:

[0147] The label data in the user portrait data that has been processed by one-hot encoding are concatenated head to tail in sequence, the concatenated one-hot encoding is reduced in dimension using embedded encoding, and the reduced-dimensional label features are combined with the numerical features and other label features in the user portrait data to obtain a user feature vector.

[0148] In the specific implementation, the target feature vector obtained after splicing can be used for the training of the algorithm model. Figure 6 This is a schematic diagram of feature vector generation in the financial reimbursement user portrait optimization method of the present invention, see Figure 6 , all the label data after one-hot encoding are concatenated head to tail in sequence, and the dimension is reduced by embedding encoding for the one-hot encoding after the concatenation. The label features after dimension reduction are combined with numerical features and other label features to form the final feature vector for model training.

[0149] Step S23: import the user feature vector into the input layer of the user portrait optimization algorithm model based on the deep neural network to obtain the optimization label output by the output layer.

[0150] It can be understood that after the user feature vector is imported into the input layer of the user portrait optimization algorithm model based on the deep neural network, the optimization label can be obtained through the pre-set user portrait optimization algorithm model.

[0151] Step S24: perform label comparison between the historical pseudo-labels and the optimized labels corresponding to the original portrait samples of the same user in the sample data, and store the original portrait samples with inconsistent label comparison results in a difficult example database.

[0152] It should be understood that by performing label matching and comparison between the historical pseudo-labels and the optimized labels corresponding to the original portrait samples of the same user in the sample data, the corresponding label comparison results can be obtained. At this time, the original portrait samples with inconsistent label comparison results can be stored in the difficult example database.

[0153] In the specific implementation, the original portrait samples of the same user are labeled by manual annotation and optimization model, and the historical pseudo-labels are used as the true labels. The optimization model results are compared with the true labels to achieve performance evaluation of the optimization model. At the same time, the original portrait samples when the labels of the two are inconsistent are stored in the difficult example database for tuning and retraining of the optimization model.

[0154] Step S25: train and optimize the user portrait optimization algorithm model according to the difficult example database until the user portrait optimization algorithm model meets the preset performance conditions, and generate a trained user portrait optimization algorithm model.

[0155] It can be understood that the user portrait optimization algorithm model is trained and optimized through the difficult example database until the user portrait optimization algorithm model meets the pre-set preset performance conditions, and a trained user portrait optimization algorithm model can be generated, that is, the optimization model can be deployed in the production system only when the model performance meets the expected online conditions.

[0156] In the specific implementation, in order to evaluate whether the performance of the algorithm model can meet the needs of user portrait optimization, a typical example of a deep neural network model performance evaluation method can use mAP (mean Average Precision) to evaluate the accuracy of the model, and use the Kappa coefficient to characterize the stability of the model. The closer the mAP is to 1, the higher the accuracy of the model, and the closer the Kappa coefficient is to 1, the better the stability of the model.

[0157] It is understandable that the user portrait optimization algorithm model can adopt but is not limited to machine learning frameworks such as TensorFlow, PyTorch, and PaddlePaddle. For the sake of compatibility and convenience, the trained model uses the Open Neural Network Exchange (ONNX) protocol to save the model's architecture and weights as a model.onnx file so that it can be deployed and applied in any ONNX-compatible environment.

[0158] Through the above scheme, this embodiment obtains sample data of user financial reimbursement, performs data preprocessing in different ways on the numerical data and label data in the sample data, and obtains processed user portrait data; splices the user portrait data to obtain user feature vectors; imports the user feature vectors into the input layer of the user portrait optimization algorithm model based on the deep neural network to obtain the optimized labels output by the output layer; compares the historical pseudo-labels and the optimized labels corresponding to the original portrait samples of the same user in the sample data, and stores the original portrait samples with inconsistent label comparison results into a difficult example database; trains and optimizes the user portrait optimization algorithm model according to the difficult example database until the user portrait optimization algorithm model meets the preset performance conditions, and generates a trained user portrait optimization algorithm model; it can be conducive to the rapid convergence of the algorithm model, and at the same time, it also desensitizes the user privacy data to a certain extent, realizes the automatic processing of user portraits, improves the accuracy of user portraits, and improves the speed and efficiency of financial reimbursement user portrait optimization.

[0159] Further, Figure 7 FIG. 5 is a flow chart of the fifth embodiment of the method for optimizing the user portrait of financial reimbursement according to the present invention. Figure 7 As shown, a fifth embodiment of the financial reimbursement user portrait optimization method of the present invention is proposed based on the first embodiment. In this embodiment, the step S30 specifically includes the following steps:

[0160] Step S31: Optimize the data not containing pseudo labels in the basic portrait according to the trained user portrait optimization algorithm model to obtain an optimized target user portrait.

[0161] It should be noted that in actual application, the data that does not contain pseudo-labels in the collected user basic portraits is preprocessed according to the above steps and spliced ​​into feature vectors. Then the optimization model is called to obtain the optimized user portraits. At this time, refined management of relevant users can be implemented according to the portrait labels.

[0162] Through the above scheme, this embodiment optimizes the data that does not contain pseudo-labels in the basic portrait according to the trained user portrait optimization algorithm model to obtain an optimized target user portrait, which can desensitize the user's privacy data to a certain extent, realize the automatic processing of user portraits, improve the accuracy of user portraits, facilitate the online call of user portraits by other business modules, and improve the speed and efficiency of financial reimbursement user portrait optimization.

[0163] Accordingly, the present invention further provides a financial reimbursement user portrait optimization device.

[0164] Reference Figure 8 , Figure 8 This is a functional module diagram of the first embodiment of the financial reimbursement user portrait optimization device of the present invention.

[0165] In a first embodiment of the financial reimbursement user portrait optimization device of the present invention, the financial reimbursement user portrait optimization device comprises:

[0166] The pseudo-label creation module 10 is used to obtain a basic profile related to the user's financial reimbursement and the user's historical reimbursement records, and to create a user's historical pseudo-label based on the user's historical reimbursement records.

[0167] The model training module 20 is used to obtain sample data of user financial reimbursement, preprocess the sample data to obtain user feature vectors, and train a user portrait optimization algorithm model based on a deep neural network according to the user historical pseudo-labels and the user feature vectors to obtain a trained user portrait optimization algorithm model.

[0168] The portrait optimization module 30 is used to optimize the basic portrait according to the trained user portrait optimization algorithm model to obtain an optimized target user portrait.

[0169] The pseudo-label creation module 10 is also used to obtain user basic information, reimbursement document records and reimbursement approval records, obtain the declaration manual number from the user basic information, and obtain the corresponding document identification from the reimbursement document record and the reimbursement approval record; associate the reimbursement document record and the reimbursement approval record according to the document identification, and associate the user basic information with the reimbursement document record and the reimbursement approval record according to the declaration manual number to generate a basic portrait related to the user's financial reimbursement; obtain the user's historical reimbursement records, and create the user's historical pseudo-label according to the user's historical reimbursement records.

[0170] The pseudo-label creation module 10 is also used to obtain the user's historical reimbursement records, and obtain the user's historical reimbursement one-time pass rate and the average time of the reimbursement process based on the user's historical reimbursement records; obtain the reimbursement amount and reimbursement process time of the sample document, and determine the user type score of the sample document based on the reimbursement amount and the reimbursement process time; sort the sample documents according to the user type score, the user's historical reimbursement one-time pass rate and the average time of the reimbursement process, and create a user history pseudo-label based on the sorting result.

[0171] The pseudo-label creation module 10 is also used to obtain the reimbursement amount and the time consumption of the reimbursement process of the sample document, and obtain the user type score of the sample document by calculating the following formula according to the reimbursement amount and the time consumption of the reimbursement process:

[0172]

[0173] Among them, M is the reimbursement amount of the sample document, T is the time consumption of the reimbursement process of the sample document, and S is the user type score.

[0174] The model training module 20 is also used to obtain sample data of user financial reimbursement, perform different data preprocessing on the numerical data and label data in the sample data, and obtain processed user portrait data; splice the user portrait data to obtain a user feature vector; import the user feature vector into the input layer of the user portrait optimization algorithm model based on a deep neural network to obtain the optimized label output by the output layer; compare the historical pseudo-labels corresponding to the original portrait samples of the same user in the sample data with the optimized labels, and store the original portrait samples with inconsistent label comparison results in a difficult example database; train and optimize the user portrait optimization algorithm model according to the difficult example database until the user portrait optimization algorithm model meets the preset performance conditions, and generate a trained user portrait optimization algorithm model.

[0175] The model training module 20 is also used to sequentially concatenate the head and tail of the label data in the user portrait data after the one-hot encoding process, reduce the dimension of the concatenated one-hot encoding by using an embedded encoding method, and combine the reduced-dimensional label features with the numerical features and other label features in the user portrait data to obtain a user feature vector.

[0176] The portrait optimization module 30 is further used to optimize the data not containing pseudo labels in the basic portrait according to the trained user portrait optimization algorithm model to obtain an optimized target user portrait.

[0177] Among them, the steps implemented by each functional module of the financial reimbursement user portrait optimization device can refer to the various embodiments of the financial reimbursement user portrait optimization method of the present invention, and will not be repeated here.

[0178] In addition, an embodiment of the present invention further provides a storage medium, on which a financial reimbursement user portrait optimization program is stored. When the financial reimbursement user portrait optimization program is executed by a processor, the following operations are implemented:

[0179] Obtaining a basic profile of the user related to financial reimbursement and the user's historical reimbursement records, and creating a user historical pseudo-label based on the user's historical reimbursement records;

[0180] Obtaining sample data of user financial reimbursement, preprocessing the sample data to obtain a user feature vector, and training a user portrait optimization algorithm model based on a deep neural network according to the user historical pseudo-label and the user feature vector to obtain a trained user portrait optimization algorithm model;

[0181] The basic portrait is optimized according to the trained user portrait optimization algorithm model to obtain an optimized target user portrait.

[0182] Furthermore, when the financial reimbursement user portrait optimization program is executed by the processor, the following operations are also implemented:

[0183] Obtaining user basic information, reimbursement document records and reimbursement approval records, obtaining the declaration manual number from the user basic information, and obtaining the corresponding document identifier from the reimbursement document records and the reimbursement approval records;

[0184] The reimbursement document record and the reimbursement approval record are associated according to the document identifier, and the user basic information is associated with the reimbursement document record and the reimbursement approval record according to the declaration manual number to generate a basic profile related to the user's financial reimbursement;

[0185] Obtain the user's historical reimbursement records, and create a user's historical pseudo-label according to the user's historical reimbursement records.

[0186] Furthermore, when the financial reimbursement user portrait optimization program is executed by the processor, the following operations are also implemented:

[0187] Obtain the user's historical reimbursement records, and obtain the user's historical reimbursement one-time pass rate and the average time required for the reimbursement process based on the user's historical reimbursement records;

[0188] Obtaining the reimbursement amount and the time taken for the reimbursement process of the sample document, and determining the user type score of the sample document according to the reimbursement amount and the time taken for the reimbursement process;

[0189] The sample documents are sorted according to the user type score, the user's historical reimbursement one-time pass rate and the average time consumption of the reimbursement process, and the user history pseudo-label is created according to the sorting result.

[0190] Furthermore, when the financial reimbursement user portrait optimization program is executed by the processor, the following operations are also implemented:

[0191] The reimbursement amount and the time taken for the reimbursement process of the sample document are obtained, and the user type score of the sample document is obtained by calculating the reimbursement amount and the time taken for the reimbursement process through the following formula:

[0192]

[0193] Among them, M is the reimbursement amount of the sample document, T is the time consumption of the reimbursement process of the sample document, and S is the user type score.

[0194] Furthermore, when the financial reimbursement user portrait optimization program is executed by the processor, the following operations are also implemented:

[0195] Obtain sample data of user financial reimbursement, perform data preprocessing in different ways on numerical data and label data in the sample data, and obtain processed user portrait data;

[0196] splicing the user portrait data to obtain a user feature vector;

[0197] Importing the user feature vector into the input layer of the user portrait optimization algorithm model based on a deep neural network to obtain the optimized label output by the output layer;

[0198] Compare the historical pseudo labels and the optimized labels corresponding to the original portrait samples of the same user in the sample data, and store the original portrait samples with inconsistent label comparison results in a difficult example database;

[0199] The user portrait optimization algorithm model is trained and optimized according to the difficult example database until the user portrait optimization algorithm model meets the preset performance conditions, and a trained user portrait optimization algorithm model is generated.

[0200] Furthermore, when the financial reimbursement user portrait optimization program is executed by the processor, the following operations are also implemented:

[0201] The label data in the user portrait data that has been processed by one-hot encoding are concatenated head to tail in sequence, the concatenated one-hot encoding is reduced in dimension using embedded encoding, and the reduced-dimensional label features are combined with the numerical features and other label features in the user portrait data to obtain a user feature vector.

[0202] Furthermore, when the financial reimbursement user portrait optimization program is executed by the processor, the following operations are also implemented:

[0203] The data not containing pseudo labels in the basic portrait is optimized according to the trained user portrait optimization algorithm model to obtain an optimized target user portrait.

[0204] Through the above scheme, this embodiment obtains basic portraits and user historical reimbursement records related to user financial reimbursement, creates user historical pseudo labels according to the user historical reimbursement records; obtains sample data of user financial reimbursement, pre-processes the sample data, obtains user feature vectors, and trains a user portrait optimization algorithm model based on a deep neural network according to the user historical pseudo labels and the user feature vectors to obtain a trained user portrait optimization algorithm model; optimizes the basic portrait according to the trained user portrait optimization algorithm model to obtain an optimized target user portrait, which can narrow the collection scope of user sensitive information, reduce the amount of data transmitted in the system, reduce server storage and computing burden, and improve the performance of the user portrait optimization model. The category score of the document in the financial reimbursement process is comprehensively evaluated from the reimbursement amount and reimbursement time, and the training and evaluation targets of the subsequent optimization model are determined. The sample data is pre-processed, which is conducive to the rapid convergence of the algorithm model. At the same time, the user privacy data is desensitized to a certain extent, and the automatic processing of the user portrait is realized, the accuracy of the user portrait is improved, and it is convenient for other business modules to call the user portrait online, and the speed and efficiency of the optimization of the financial reimbursement user portrait are improved.

[0205] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or device including the element.

[0206] The serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages or disadvantages of the embodiments.

[0207] The above are only preferred embodiments of the present invention, and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A method for optimizing a user profile for financial reimbursement, characterized in that: The financial reimbursement user portrait optimization method includes: Obtaining a basic profile of the user related to financial reimbursement and the user's historical reimbursement records, and creating a user historical pseudo-label based on the user's historical reimbursement records; Obtaining sample data of user financial reimbursement, preprocessing the sample data to obtain a user feature vector, and training a user portrait optimization algorithm model based on a deep neural network according to the user historical pseudo-label and the user feature vector to obtain a trained user portrait optimization algorithm model; Optimizing the basic portrait according to the trained user portrait optimization algorithm model to obtain an optimized target user portrait; The step of obtaining a basic profile related to the user's financial reimbursement and the user's historical reimbursement record, and creating a user's historical pseudo-label according to the user's historical reimbursement record, includes: Obtaining user basic information, reimbursement document records and reimbursement approval records, obtaining the declaration manual number from the user basic information, and obtaining the corresponding document identifier from the reimbursement document records and the reimbursement approval records; The reimbursement document record and the reimbursement approval record are associated according to the document identifier, and the user basic information is associated with the reimbursement document record and the reimbursement approval record according to the declaration manual number to generate a basic profile related to the user's financial reimbursement; Obtaining a user's historical reimbursement record, and creating a user's historical pseudo-label according to the user's historical reimbursement record; The step of obtaining the user's historical reimbursement records and creating the user's historical pseudo-labels according to the user's historical reimbursement records includes: Obtain the user's historical reimbursement records, and obtain the user's historical reimbursement one-time pass rate and the average time required for the reimbursement process based on the user's historical reimbursement records; Obtaining the reimbursement amount and the time taken for the reimbursement process of the sample document, and determining the user type score of the sample document according to the reimbursement amount and the time taken for the reimbursement process; The sample documents are sorted according to the user type score, the user's historical reimbursement one-time pass rate and the average time consumption of the reimbursement process, and the user history pseudo-label is created according to the sorting result.

2. The financial reimbursement user portrait optimization method according to claim 1, characterized in that: The obtaining of the reimbursement amount and the time taken for the reimbursement process of the sample document, and determining the user type score of the sample document according to the reimbursement amount and the time taken for the reimbursement process, includes: The reimbursement amount and the time taken for the reimbursement process of the sample document are obtained, and the user type score of the sample document is obtained by calculating the reimbursement amount and the time taken for the reimbursement process through the following formula: Among them, M is the reimbursement amount of the sample document, T is the time consumption of the reimbursement process of the sample document, and S is the user type score.

3. The financial reimbursement user portrait optimization method according to claim 1, characterized in that: The method of obtaining sample data of user financial reimbursement, preprocessing the sample data to obtain a user feature vector, and training a user portrait optimization algorithm model based on a deep neural network according to the user historical pseudo-label and the user feature vector to obtain a trained user portrait optimization algorithm model includes: Obtain sample data of user financial reimbursement, perform data preprocessing in different ways on numerical data and label data in the sample data, and obtain processed user portrait data; splicing the user portrait data to obtain a user feature vector; Importing the user feature vector into the input layer of the user portrait optimization algorithm model based on a deep neural network to obtain the optimized label output by the output layer; Compare the historical pseudo labels and the optimized labels corresponding to the original portrait samples of the same user in the sample data, and store the original portrait samples with inconsistent label comparison results in a difficult example database; The user portrait optimization algorithm model is trained and optimized according to the difficult example database until the user portrait optimization algorithm model meets the preset performance conditions, and a trained user portrait optimization algorithm model is generated.

4. The financial reimbursement user portrait optimization method according to claim 3, characterized in that: The step of splicing the user portrait data to obtain a user feature vector includes: The label data in the user portrait data that has been processed by one-hot encoding are concatenated head to tail in sequence, the concatenated one-hot encoding is reduced in dimension using embedded encoding, and the reduced-dimensional label features are combined with the numerical features and other label features in the user portrait data to obtain a user feature vector.

5. The financial reimbursement user portrait optimization method according to claim 1, characterized in that: The step of optimizing the basic portrait according to the trained user portrait optimization algorithm model to obtain an optimized target user portrait includes: The data not containing pseudo labels in the basic portrait is optimized according to the trained user portrait optimization algorithm model to obtain an optimized target user portrait.

6. A financial reimbursement user portrait optimization device, characterized in that: The financial reimbursement user portrait optimization device comprises: A pseudo-label creation module is used to obtain a basic profile of a user related to financial reimbursement and a historical reimbursement record of the user, and to create a historical pseudo-label of the user according to the historical reimbursement record of the user; A model training module is used to obtain sample data of user financial reimbursement, pre-process the sample data to obtain a user feature vector, and train a user portrait optimization algorithm model based on a deep neural network according to the user historical pseudo-label and the user feature vector to obtain a trained user portrait optimization algorithm model; A portrait optimization module is used to optimize the basic portrait according to the trained user portrait optimization algorithm model to obtain an optimized target user portrait; The pseudo-label creation module is also used to obtain user basic information, reimbursement document records and reimbursement approval records, obtain the declaration manual number from the user basic information, and obtain the corresponding document identifier from the reimbursement document record and the reimbursement approval record; associate the reimbursement document record with the reimbursement approval record according to the document identifier, and associate the user basic information with the reimbursement document record and the reimbursement approval record according to the declaration manual number to generate a basic portrait related to the user's financial reimbursement; obtain the user's historical reimbursement records, and create the user's historical pseudo-label according to the user's historical reimbursement records; The pseudo-label creation module is also used to obtain the user's historical reimbursement records, and obtain the user's historical reimbursement one-time pass rate and the average time of the reimbursement process based on the user's historical reimbursement records; obtain the reimbursement amount and reimbursement process time of the sample document, and determine the user type score of the sample document based on the reimbursement amount and the reimbursement process time; sort the sample documents according to the user type score, the user's historical reimbursement one-time pass rate and the average time of the reimbursement process, and create a user history pseudo-label based on the sorting result.

7. A financial reimbursement user portrait optimization device, characterized in that: The financial reimbursement user portrait optimization device includes: a memory, a processor, and a financial reimbursement user portrait optimization program stored in the memory and executable on the processor, wherein the financial reimbursement user portrait optimization program is configured to implement the steps of the financial reimbursement user portrait optimization method as described in any one of claims 1 to 5.

8. A storage medium, characterized in that: The storage medium stores a financial reimbursement user portrait optimization program, which, when executed by a processor, implements the steps of the financial reimbursement user portrait optimization method according to any one of claims 1 to 5.

Images