Intelligent auditing method and device based on multi-dimensional features and medium

Abstract

The invention discloses an intelligent auditing method and device based on multi-dimensional features and a medium, and belongs to the technical field of intelligent auditing. The method comprises the steps of constructing a data approval model, and obtaining to-be-approved data based on the data approval model; wherein the data approval model comprises a data cleaning model, a feature extraction model, a classification identification model and a result approval model; cleaning and sorting the to-be-approved data based on the data cleaning model to generate standardized data; extracting features of the standardized data based on a feature extraction model to generate collection features; processing the collected features based on a classification identification model, and classifying the collected features to obtain a classification result; and processing the classification result according to a preset weight ratio based on a result approval model to generate an approval result. Through the method, the to-be-approved data can be automatically approved, the approval efficiency, accuracy and consistency are improved, and the operation cost is reduced.

Description

Technical Field

[0001] The present application relates to the technical field of intelligent auditing, and in particular to an intelligent auditing method, device and medium based on multi-dimensional features. Background Art

[0002] With the rapid development of information technology and the advent of the digital age, enterprises and organizations generate and process massive amounts of data to be reviewed every day. These data to be reviewed include but are not limited to various application forms, application documents, application images and application videos. These data to be reviewed need to go through the approval process before they can be further processed or implemented. However, the traditional approval process usually relies on manual operation, which is not only inefficient but also prone to errors, and cannot meet the needs of modern enterprises and organizations for efficient and accurate approval.

[0003] More specifically, manual approval has the following significant problems: first, approvers need to spend a lot of time browsing and evaluating the data to be approved, which will increase their workload; second, manual approval is limited by the approvers' professional knowledge, experience and judgment, and may be subjective and inconsistent; finally, the long approval cycle not only affects business efficiency, but may also lead to missed business opportunities or increased operational risks.

[0004] Therefore, how to automatically review and approve data to improve approval efficiency, accuracy, and consistency while reducing operating costs has become a technical problem that needs to be solved urgently. Summary of the invention

[0005] The embodiments of the present application provide an intelligent audit method, device and medium based on multi-dimensional features to solve the following technical problem: how to automatically approve approval data to improve approval efficiency, accuracy and consistency while reducing operating costs.

[0006] In a first aspect, an embodiment of the present application provides an intelligent audit method based on multi-dimensional features, the method comprising: constructing a data approval model, and obtaining data to be approved based on the data approval model; wherein the data approval model comprises a data cleaning model, a feature extraction model, a classification recognition model and a result approval model; cleaning and organizing the data to be approved based on the data cleaning model to generate standardized data; extracting features of the standardized data based on the feature extraction model to generate collection features; processing the collection features based on the classification recognition model to classify the collection features to obtain classification results; wherein the classification recognition model comprises multiple sub-classification recognition models, and the classification results comprise multiple sub-classification results; processing the classification results according to a preset weight ratio based on the result approval model to generate an approval result.

[0007] In one implementation of the present application, the data to be approved is cleaned and organized based on a data cleaning model to generate standardized data, specifically including: identifying and removing abnormal data in the data to be approved based on the data cleaning model; wherein the abnormal data includes at least one of the following: noise, outliers, duplicates and invalid information; processing the data to be approved based on a preset format processing algorithm to format and standardize the data to be approved to generate a standardized format.

[0008] In one implementation of the present application, features of standardized data are extracted based on a feature extraction model to generate collection features, specifically including: determining feature extraction standards, and analyzing standardized data based on the feature extraction standards; wherein the feature extraction standards include at least one of the following: text content, numerical data, and timestamp; processing text content based on a preset natural language algorithm to extract semantic features of standardized data; processing numerical data based on a preset statistical analysis algorithm to extract numerical features of standardized data; and generating collection features based on semantic features, numerical features, and timestamps.

[0009] In one implementation of the present application, the collected features are processed based on a classification recognition model to classify the collected features to obtain classification results, specifically including: inputting the collected features into the classification recognition model to determine sub-collection features and corresponding feature output models; wherein the collected features include multiple sub-collection features; processing the sub-collection features based on the corresponding feature output model to determine multiple sub-classification results; and integrating multiple sub-classification results to generate a classification result.

[0010] In one implementation of the present application, the classification results are processed according to a preset weight ratio based on the result approval model to generate an approval result, specifically including: obtaining business needs and historical approval data, and setting a weight ratio for the classification results based on the business needs and historical approval data; processing multiple sub-classification results based on the weight ratio to generate an approval result; wherein the approval results include passed, pending review and returned.

[0011] In one implementation of the present application, after processing multiple sub-classification results based on the weight ratio to generate an approval result, the method also includes: if the approval result is pending review, starting a preset manual review process, and processing the pending approval data based on the manual review process to generate a manual approval result; inputting the manual approval result into the data approval model to optimize the data approval model; if the approval result is a return, comparing the preset rejection database based on multiple sub-classification results to determine the reason for rejection.

[0012] In one implementation of the present application, building a data approval model specifically includes: collecting historical approval data to build a training set; wherein the historical approval data includes successfully approved data and unsuccessfully approved data; training a preset preliminary data approval model based on the training set to build a data cleaning model, a feature extraction model, a classification recognition model and a result approval model; integrating the data cleaning model, the feature extraction model, the classification recognition model and the result approval model to generate a result approval model.

[0013] In one implementation of the present application, a preset preliminary data approval model is trained based on a training set to construct a data cleaning model, a feature extraction model, a classification recognition model and a result approval model, specifically including: determining a preliminary data approval model, wherein the preliminary data approval model includes a preliminary data cleaning model, a preliminary feature extraction model, a preliminary classification recognition model and a preliminary result approval model; training a preliminary data cleaning model based on a training set, identifying and removing abnormal data in the training set based on the preliminary data cleaning model to generate data results, and determining the data cleaning model when the accuracy of the output result is greater than a preset cleaning accuracy threshold; training a preliminary feature extraction model based on a training set to generate an extraction result, and determining the feature extraction model when the accuracy of the extraction result is greater than a preset extraction accuracy threshold; training a preliminary classification recognition model based on a training set to generate a classification result, and determining the classification recognition model when the accuracy of the classification result is greater than a preset classification accuracy threshold; setting the weight ratio of each sub-classification result in the result approval model according to business needs and the analysis results of historical approval data to generate a result approval model.

[0014] In the second aspect, the embodiment of the present application also provides an intelligent audit device based on multi-dimensional features, characterized in that the device includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions that can be executed by at least one processor, and the instructions are executed by at least one processor so that the at least one processor can: build a data approval model, and obtain data to be approved based on the data approval model; wherein the data approval model includes a data cleaning model, a feature extraction model, a classification recognition model and a result approval model; based on the data cleaning model, the data to be approved is cleaned and sorted to generate standardized data; based on the feature extraction model, the features of the standardized data are extracted to generate collection features; based on the classification recognition model, the collection features are processed to classify the collection features to obtain classification results; wherein the classification recognition model includes multiple sub-classification recognition models, and the classification results include multiple sub-classification results; based on the result approval model, the classification results are processed according to a preset weight ratio to generate an approval result.

[0015] In the third aspect, the embodiment of the present application also provides a non-volatile computer storage medium for intelligent audit based on multi-dimensional features, which stores computer executable instructions, characterized in that the computer executable instructions are set to: construct a data approval model, and obtain data to be approved based on the data approval model; wherein the data approval model includes a data cleaning model, a feature extraction model, a classification recognition model and a result approval model; based on the data cleaning model, clean and organize the data to be approved to generate standardized data; based on the feature extraction model, extract the features of the standardized data to generate collection features; based on the classification recognition model, process the collection features to classify the collection features to obtain classification results; wherein the classification recognition model includes multiple sub-classification recognition models, and the classification results include multiple sub-classification results; based on the result approval model, process the classification results according to a preset weight ratio to generate an approval result.

[0016] The embodiments of the present application provide a multi-dimensional feature-based intelligent audit method, device, and medium, which have at least the following technical effects:

[0017] 1. It can clean and organize the original data to be approved, remove redundant, erroneous or incomplete data, and generate digital standardized data. It can ensure the quality of subsequent processing data to a certain extent and improve the accuracy of data approval.

[0018] 2. It can classify data through extracted features. Through multiple sub-classification recognition models, each sub-model focuses on different classification standards or data characteristics. The classification recognition model can classify data, which improves the accuracy and flexibility of classification to a certain extent, and can handle complex and diverse data approval needs.

[0019] 3. It can process the classification results according to the preset weight ratio and generate the final approval results. By adjusting the weights, it can flexibly respond to different approval standards and emphases, making the approval process more in line with actual needs. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0021] Figure 1 A flow chart of an intelligent audit method based on multi-dimensional features provided in an embodiment of the present application;

[0022] Figure 2 A schematic diagram of the internal structure of an intelligent audit device based on multi-dimensional features provided in an embodiment of the present application. DETAILED DESCRIPTION

[0023] In order to make the purpose, technical solution and advantages of the present application clearer, the technical solution of the present application will be clearly and completely described below in conjunction with the specific embodiments of the present application and the corresponding drawings. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present application.

[0024] The embodiments of the present application provide an intelligent audit method, device and medium based on multi-dimensional features to solve the following technical problem: how to automatically approve approval data to improve approval efficiency, accuracy and consistency while reducing operating costs.

[0025] The technical solution proposed in the embodiments of the present application is described in detail below with reference to the accompanying drawings.

[0026] Figure 1 The present invention provides an intelligent audit flow chart based on multi-dimensional features. Figure 1 As shown, an intelligent audit method based on multi-dimensional features provided in an embodiment of the present application specifically includes the following steps:

[0027] Step 1: Build a data approval model and obtain the data to be approved based on the data approval model; the data approval model includes a data cleaning model, a feature extraction model, a classification recognition model and a result approval model.

[0028] The data approval model covers the entire process from data preprocessing to final result output. The data approval model consists of four main parts: data cleaning model, feature extraction model, classification recognition model and result approval model.

[0029] First, historical approval data is collected to construct a training set; the historical approval data includes successfully approved data and unsuccessful approval data.

[0030] It is understandable that when collecting historical approval data, it is necessary to ensure the comprehensiveness and diversity of the data to a certain extent. Therefore, it is necessary to collect data on successful approvals and data on unsuccessful approvals. This is because the data on successful approvals can provide information on the characteristics and conditions under which an application is likely to be approved, while the data on unsuccessful approvals can provide information on the characteristics and conditions under which an application may be rejected.

[0031] For example, for an application for sick leave, the most important basis (feature) for the application is the leave form issued by the hospital. When a leave form exists, its features, such as the hospital issuing the leave form, the recommended rest time, and the cause of the illness, can be used to determine whether the sick leave standard is met and whether the leave time is appropriate.

[0032] In specific implementation, these historical approval data can be extracted from the enterprise's approval system, database or archives. The data should include the applicant's basic information, the specific content of the application, the approval process record, the approval result (success or failure) and the reason for rejection.

[0033] Further, the preset preliminary data approval model is trained based on the training set to build a data cleaning model, a feature extraction model, a classification recognition model, and a result approval model. In the process of building a data approval model, training the preset preliminary data approval model based on the training set is a key step. The following is a detailed explanation and revised steps.

[0034] A1. Determine the preliminary data approval model, which includes a preliminary data cleaning model, a preliminary feature extraction model, a preliminary classification and recognition model, and a preliminary result approval model. First, it is necessary to determine the preliminary data approval model framework, which includes four main parts: a preliminary data cleaning model, a preliminary feature extraction model, a preliminary classification and recognition model, and a preliminary result approval model. The above models all use existing technologies and will not be described in detail here.

[0035] A2. Train a preliminary data cleaning model based on the training set, identify and remove abnormal data in the training set based on the preliminary data cleaning model to generate data results, and determine the data cleaning model when the accuracy of the output results is greater than the preset cleaning accuracy threshold. Use the training set to train the preliminary data cleaning model. The goal is to identify and remove abnormal data in the training set, such as data with format errors, logical inconsistencies, or obviously unreasonable data. During the training process, by comparing the differences between the model output results and the actual data, the model parameters are continuously adjusted to improve the accuracy of data cleaning. When the accuracy of the model output results exceeds the preset cleaning accuracy threshold, the data cleaning model is considered to be sufficiently accurate, and the data cleaning model is determined.

[0036] A3. Train a preliminary feature extraction model based on the training set to generate extraction results, and determine the feature extraction model when the accuracy of the extraction results is greater than the preset extraction accuracy threshold. Use the cleaned training set data to train the preliminary feature extraction model. Extract key features that affect the approval results from the data. During the training process, evaluate the accuracy of the model's feature extraction, and determine the feature extraction model when the accuracy of the extraction results exceeds the preset extraction accuracy threshold.

[0037] A4. Train a preliminary classification recognition model based on the training set to generate classification results, and determine the classification recognition model when the accuracy of the classification results is greater than a preset classification accuracy threshold. Use the extracted features to train the preliminary classification recognition model, and evaluate the performance of the model by comparing the accuracy of the model classification results with the preset classification accuracy threshold. When the accuracy of the classification results exceeds the threshold, the classification recognition model is determined.

[0038] A5. According to the business requirements and the analysis results of historical approval data, set the weight ratio of each sub-classification result in the result approval model to generate the result approval model. According to the specific business requirements and the analysis results of historical approval data, set the weight ratio for the different sub-classification results output by the classification recognition model. These weights reflect the degree of influence of different classification results on the final approval decision. Combining these weights with business rules, the final result approval model can be constructed.

[0039] Further, the data cleaning model, feature extraction model, classification recognition model and result approval model are integrated to generate a result approval model. The data cleaning model, feature extraction model, classification recognition model and result approval model are integrated in a logical order in the approval process. It can be understood that the input and output formats of each model should be guaranteed to be compatible with adjacent models so that data can be smoothly transferred between the models.

[0040] Step 2: Clean and organize the data to be approved based on the data cleaning model to generate standardized data.

[0041] Here is a detailed explanation of step 2:

[0042] First, based on the data cleaning model, identify and remove abnormal data in the data to be approved; abnormal data includes at least one of the following: noise, outliers, duplicates, and invalid information. Use the constructed data cleaning model to perform preliminary processing on the data to be approved. The data cleaning model can identify and remove outliers in the data. Abnormal data may include noise, outliers, duplicates, and invalid information. If abnormal data is not processed, it may interfere with subsequent data analysis and model training.

[0043] In a specific example, take the bank loan approval application form as an example:

[0044] In the age field, some applicants' ages were incorrectly entered as 0 or as values ​​greater than 100 (outliers).

[0045] In the income field, some data is filled in as text (such as "confidential" or "unknown") instead of numbers (invalid information).

[0046] In some application forms, the name and occupation fields were filled in twice (duplicate entries).

[0047] For the age field, the data cleaning model identifies 0 and values ​​greater than 100 as outliers and removes or replaces them with reasonable default values ​​(such as average age). In the income field, non-numeric text is identified and treated as invalid information, removed and determined to be unsuccessful. For duplicate names and occupation information, duplicate records are identified and removed to ensure that each applicant has only one unique record.

[0048] Some of the loan amount fields contain non-numeric characters such as commas or currency symbols (noise).

[0049] Further, the data to be approved is processed based on a preset format processing algorithm to format and standardize the data to be approved to generate a standardized format. After removing the abnormal data, in order to ensure that the data has a unified format and structure to facilitate subsequent analysis and model training, the remaining data needs to be formatted and standardized.

[0050] In a specific example, take an employee asking for leave as an example:

[0051] When an employee of an enterprise takes sick leave, he or she usually needs to fill out a leave application form. The leave application form includes but is not limited to the following fields: employee name, employee ID, department, position, leave type (such as sick leave, personal leave, etc.), leave start time, leave end time, leave reason, emergency contact information, etc.

[0052] Arrange the above information in a preset order for standardized processing.

[0053] Step 3: Extract the features of the standardized data based on the feature extraction model to generate acquisition features.

[0054] First, determine the feature extraction criteria, and analyze the standardized data based on the feature extraction criteria; wherein the feature extraction criteria include at least one of the following: text content, numerical data, and timestamp. Before performing feature extraction, it is first necessary to determine what types of features to extract. The feature extraction criteria may include text content, numerical data, and timestamp. According to the determined feature extraction criteria, feature extraction is performed on the cleaned and organized standardized data based on the feature extraction criteria.

[0055] For example, text content may include reasons for leave, awards, study abroad experience and other data. The above text information needs to understand its semantic meaning and be converted into digital information or short text information.

[0056] Numerical data may include price, sales volume, and ratings, which require statistical analysis.

[0057] Timestamp data can provide specific time information, such as leave time, working time, etc.

[0058] Furthermore, the text content is processed based on a preset natural language algorithm to extract semantic features of the standardized data.

[0059] Use the preset natural language processing (NLP) algorithm to process the text content. The NLP algorithm adopts the bag-of-words model, TF-IDF (term frequency-inverse document frequency), word2vec or BERT model to extract semantic features from the text.

[0060] Further, the numerical data is processed based on a preset statistical analysis algorithm to extract the numerical features of the standardized data. The numerical data is processed by applying a preset statistical analysis algorithm, such as mean, median, standard deviation, and correlation analysis.

[0061] Statistical analysis helps in extracting numerical features of data such as trends, distributions, outliers, etc.

[0062] For example, in leave information, the leave deadline can be determined based on the leave time and the number of days of leave.

[0063] Furthermore, acquisition features are generated based on semantic features, numerical features and timestamps. Acquisition features are generated by combining semantic features extracted from text content, numerical features extracted from numerical data, and timestamp data.

[0064] In a specific case, take asking for leave as an example:

[0065] The leave information application form includes: the employee's name, reason for leave, number of days of leave, start time of leave, end time of leave and sick leave certificate issued by the hospital.

[0066] Extract the text content of "reason for leave". Through the feature extraction model, the reason is converted into a numerical vector, which can quantify the text information so that the subsequent model can process and understand it. For example, "feeling unwell and having a fever" can extract the text information of "feeling unwell" and "fever". The above text information belongs to "sick (mild)", so "sick (mild)" can be converted into a specific numerical vector.

[0067] Extract the numerical data feature of "number of days of leave".

[0068] Extract the time features from the two timestamp fields "Leave Start Time" and "Leave End Time". This includes information such as the specific date of the leave and the length of the leave period.

[0069] Similarly, extract the cause of illness, recommended number of days of leave, time of issuance of sick leave certificate, issuing hospital and other information from the sick leave certificate, and convert it into digital information according to preset conversion rules.

[0070] Step 4: Process the collected features based on the classification recognition model to classify the collected features to obtain a classification result; wherein the classification recognition model includes multiple sub-classification recognition models, and the classification result includes multiple sub-classification results.

[0071] First, the acquisition feature is input into the classification recognition model to determine the sub-acquisition feature and the corresponding feature output model; wherein the acquisition feature includes multiple sub-acquisition features. The acquisition feature extracted in step 3 is input into the classification recognition model. Since the classification recognition model includes multiple sub-classification recognition models, it is necessary to determine which sub-acquisition features will be input into which sub-classification recognition model.

[0072] Further, the sub-collection features are processed based on the corresponding feature output model to determine multiple sub-classification results. Once the matching relationship between the sub-collection features and the corresponding sub-classification recognition models is determined, the sub-classification recognition models are used to process the respective sub-collection features. Each sub-classification recognition model will make independent classification judgments based on the features it receives and output a sub-classification result.

[0073] Furthermore, multiple sub-classification results are integrated to generate a classification result.

[0074] Step 5: Based on the result approval model, the classification results are processed according to the preset weight ratio to generate the approval result.

[0075] First, business requirements and historical approval data are obtained, and weight ratios are set for classification results based on business requirements and historical approval data. According to business requirements and historical approval data, corresponding weight ratios are set for each sub-classification result in the classification result, and the weight ratios are set manually.

[0076] Furthermore, multiple sub-classification results are processed based on weight ratios to generate approval results; wherein the approval results include passed, pending review, and returned.

[0077] For example, for bank loan review, the review factors include whether there is a car property, real estate, and whether there is a default record. Among them, the approval result of the car property (non-real estate) in the sub-classification result is not passed, and the approval result of the real estate (real estate) in the sub-classification result is passed. The weight ratio of the car property is 0.1, and the weight ratio of the real estate is 0.4. If the weighted approval result is greater than 0.8, the approval is passed. Otherwise, it is returned.

[0078] Furthermore, if the approval result is "pending review", the preset manual review process is started, and the data to be approved is processed based on the manual review process to generate a manual approval result. If the approval result is "pending review", the preset manual review process is triggered. In the manual review process, professional reviewers will conduct a detailed review of the data to be approved and generate a manual approval result.

[0079] Furthermore, the manual approval results are input into the data approval model to optimize the data approval model. The manual approval results are fed back into the data approval model for model optimization and improvement. Through continuous learning and adjustment, the accuracy and versatility of the data approval model can be improved to a certain extent.

[0080] Furthermore, if the approval result is a rejection, the rejection reason is determined based on the comparison of the multiple sub-classification results with the preset rejection database. If the approval result is "rejected", a comparison analysis is performed based on the multiple sub-classification results and the preset rejection database. Through the above comparison, the specific reasons or rejection reasons leading to the rejection can be determined, providing a reference for the applicant.

[0081] The above is an embodiment of the method proposed in this application. Based on the same inventive concept, the embodiment of this application also provides an intelligent audit device based on multi-dimensional features, whose structure is as follows: Figure 2 shown.

[0082] Figure 2 The following is a schematic diagram of the internal structure of an intelligent audit device based on multi-dimensional features provided in an embodiment of the present application. Figure 2 As shown, the device includes:

[0083] at least one processor 201;

[0084] and, a memory 202 communicatively connected to the at least one processor;

[0085] The memory 202 stores instructions that can be executed by at least one processor, and the instructions are executed by at least one processor 201 to enable at least one processor 201 to:

[0086] Construct a data approval model, and obtain the data to be approved based on the data approval model; wherein the data approval model includes a data cleaning model, a feature extraction model, a classification recognition model and a result approval model; based on the data cleaning model, clean and organize the data to be approved to generate standardized data; based on the feature extraction model, extract the features of the standardized data to generate collection features; based on the classification recognition model, process the collection features to classify the collection features to obtain classification results; based on the result approval model, process the classification results according to the preset weight ratio to generate the approval results.

[0087] Some embodiments of the present application provide corresponding Figure 1 A non-volatile computer storage medium for intelligent audit based on multi-dimensional features stores computer executable instructions, wherein the computer executable instructions are configured to: construct a data approval model, and obtain data to be approved based on the data approval model; wherein the data approval model includes a data cleaning model, a feature extraction model, a classification recognition model, and a result approval model; based on the data cleaning model, the data to be approved is cleaned and sorted to generate standardized data; based on the feature extraction model, features of the standardized data are extracted to generate collection features; based on the classification recognition model, the collection features are processed to classify the collection features to obtain classification results; based on the result approval model, the classification results are processed according to a preset weight ratio to generate approval results.

[0088] Each embodiment in this application is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the IoT device and medium embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiments.

[0089] The system and medium provided in the embodiments of the present application correspond one-to-one to the method. Therefore, the system and medium also have similar beneficial technical effects to the corresponding methods. Since the beneficial technical effects of the methods have been described in detail above, the beneficial technical effects of the system and medium will not be repeated here.

[0090] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0091] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0092] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0093] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0094] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0095] The memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0096] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0097] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity 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, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0098] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included within the scope of the claims of the present application.

Claims

1. An intelligent audit method based on multi-dimensional features, characterized in that: The method comprises: Constructing a data approval model, and acquiring data to be approved based on the data approval model; wherein the data approval model includes a data cleaning model, a feature extraction model, a classification recognition model and a result approval model; Cleaning and arranging the data to be approved based on the data cleaning model to generate standardized data; Extracting features of the standardized data based on the feature extraction model to generate acquisition features; Processing the collected features based on the classification recognition model to classify the collected features to obtain a classification result; wherein the classification recognition model includes a plurality of sub-classification recognition models, and the classification result includes a plurality of sub-classification results; The classification results are processed according to a preset weight ratio based on the result approval model to generate an approval result.

2. According to claim 1, the intelligent audit method based on multidimensional features is characterized in that: Cleaning and arranging the data to be approved based on the data cleaning model to generate standardized data, specifically including: Identify and remove abnormal data in the data to be approved based on the data cleaning model; wherein the abnormal data includes at least one of the following: noise, outliers, duplicates, and invalid information; The data to be approved is processed based on a preset format processing algorithm to format and standardize the data to be approved to generate the standardized format.

3. According to claim 1, the intelligent audit method based on multidimensional features is characterized in that: Extracting features of the standardized data based on the feature extraction model to generate acquisition features specifically includes: Determining a feature extraction criterion, and analyzing the standardized data based on the feature extraction criterion; wherein the feature extraction criterion includes at least one of the following: text content, numerical data, and timestamp; Processing the text content based on a preset natural language algorithm to extract semantic features of the standardized data; Processing the numerical data based on a preset statistical analysis algorithm to extract numerical features of the standardized data; A collection feature is generated based on the semantic feature, the numerical feature and the timestamp.

4. The intelligent audit method based on multidimensional features according to claim 1 is characterized in that: Processing the collected features based on the classification recognition model to classify the collected features to obtain classification results specifically includes: Inputting the acquisition feature into the classification recognition model to determine a sub-acquisition feature and a corresponding feature output model; wherein the acquisition feature includes a plurality of sub-acquisition features; Processing the sub-collected features based on the corresponding feature output model to determine a plurality of the sub-classification results; The plurality of sub-classification results are integrated to generate a classification result.

5. The intelligent audit method based on multidimensional features according to claim 1 is characterized in that: Processing the classification results according to a preset weight ratio based on the result approval model to generate an approval result, specifically including: Obtaining business requirements and historical approval data, and setting weight ratios for the classification results based on the business requirements and historical approval data; The plurality of sub-classification results are processed based on the weight ratio to generate an approval result; wherein the approval result includes passed, pending review and returned.

6. The intelligent audit method based on multidimensional features according to claim 5 is characterized in that: After processing the plurality of sub-classification results based on the weight ratio to generate an approval result, the method further includes: If the approval result is pending review, the preset manual review process is started, and the pending review data is processed based on the manual review process to generate a manual review result; Inputting the manual approval result into the data approval model to optimize the data approval model; If the approval result is a rejection, the rejection reason is determined by comparing the plurality of sub-classification results with a preset rejection database.

7. The intelligent audit method based on multidimensional features according to claim 1 is characterized in that: Build a data approval model, including: Collect historical approval data to construct a training set; wherein the historical approval data includes successfully approved data and unsuccessfully approved data; Training a preset preliminary data approval model based on the training set to construct the data cleaning model, feature extraction model, classification recognition model and result approval model; The data cleaning model, feature extraction model, classification recognition model and result approval model are integrated to generate a result approval model.

8. The intelligent audit method based on multi-dimensional features according to claim 7 is characterized in that: Based on the training set, a preset preliminary data approval model is trained to construct the data cleaning model, feature extraction model, classification recognition model and result approval model, specifically including: Determining the preliminary data approval model, wherein the preliminary data approval model includes a preliminary data cleaning model, a preliminary feature extraction model, a preliminary classification and recognition model, and a preliminary result approval model; Training the preliminary data cleaning model based on the training set, identifying and removing abnormal data in the training set based on the preliminary data cleaning model to generate data results, and determining the data cleaning model when the accuracy of the output result is greater than a preset cleaning accuracy threshold; Training the preliminary feature extraction model based on the training set to generate an extraction result, and determining the feature extraction model when the accuracy of the extraction result is greater than a preset extraction accuracy threshold; Training the preliminary classification recognition model based on the training set to generate a classification result, and determining the classification recognition model when the accuracy of the classification result is greater than a preset classification accuracy threshold; According to the business requirements and the analysis results of the historical approval data, the weight ratio of each sub-classification result in the result approval model is set to generate the result approval model.

9. An intelligent audit device based on multi-dimensional features, characterized in that: The device comprises: at least one processor; and, a memory communicatively coupled to the at least one processor; The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to: Constructing a data approval model, and acquiring data to be approved based on the data approval model; wherein the data approval model includes a data cleaning model, a feature extraction model, a classification recognition model and a result approval model; Cleaning and arranging the data to be approved based on the data cleaning model to generate standardized data; Extracting features of the standardized data based on the feature extraction model to generate acquisition features; Processing the collected features based on the classification recognition model to classify the collected features to obtain a classification result; wherein the classification recognition model includes a plurality of sub-classification recognition models, and the classification result includes a plurality of sub-classification results; The classification results are processed according to a preset weight ratio based on the result approval model to generate an approval result.

10. A non-volatile computer storage medium for intelligent auditing based on multi-dimensional features, storing computer executable instructions, characterized in that: The computer executable instructions are configured to: Constructing a data approval model, and acquiring data to be approved based on the data approval model; wherein the data approval model includes a data cleaning model, a feature extraction model, a classification recognition model and a result approval model; Cleaning and arranging the data to be approved based on the data cleaning model to generate standardized data; Extracting features of the standardized data based on the feature extraction model to generate acquisition features; Processing the collected features based on the classification recognition model to classify the collected features to obtain a classification result; wherein the classification recognition model includes a plurality of sub-classification recognition models, and the classification result includes a plurality of sub-classification results; The classification results are processed according to a preset weight ratio based on the result approval model to generate an approval result.

Images