Automatic approval system for intelligent reimbursement process

Abstract

The invention discloses an intelligent reimbursement process automatic examination and approval system, and relates to the technical field of data processing. Historical reimbursement examination and approval data are collected to establish a reimbursement examination and approval data standard, a sample set is established, the historical reimbursement examination and approval data in the sample set are processed at the same time, and after processing is completed, the reimbursement process automatic examination and approval system is established. Constructing a reimbursement approval analysis model based on the processed historical reimbursement approval data in the sample set, and determining a processed historical reimbursement approval data auditing template based on the constructed reimbursement approval analysis model; meanwhile, reimbursement approval data is collected in real time, feature extraction and storage are conducted on the processed historical reimbursement approval data auditing template and the reimbursement approval data collected in real time through the convolutional neural network, and finally the features extracted in real time are compared with the features stored in the sample set in a semantic comparison mode. And the reimbursement approval data corresponding to the features extracted in real time is processed based on the comparison result, so that the intelligence of automatic approval of the reimbursement process is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, in particular to an intelligent reimbursement process automatic approval system. BACKGROUND

[0002] With the continuous expansion of the scale of enterprises, the difficulty of financial management is also increasing; at the same time, since the current financial reimbursement approval is mainly processed through manual audit, the problem of low efficiency of manual audit due to the increase of approval data is highlighted.

[0003] The prior art such as the invention patent application disclosed in the announcement No. CN106600218A provides a processing method and system for data in an approval process, which method comprises: obtaining an approval task in an approval process, and approving the approval task; establishing a dialogue identification ID for each dialogue, obtaining a plurality of dialogue identification IDs, establishing a corresponding task ID for each approval task and resource reservation task, and obtaining a plurality of task IDs; establishing an ID corresponding relationship between the plurality of dialogue identification IDs and the plurality of task IDs; in a session interface, initiating a new task or processing an initiated task; and in a task interface, realizing instant dialogue communication according to the ID corresponding relationship. The present application adopts a mode of combining the approval task with the session communication, and through the ID corresponding relationship, the session interface in the approval process can be used for approval, and the corresponding session interface can be switched to in the approval interface for timely communication, thereby improving the efficiency of approval and communication.

[0004] According to the above-mentioned scheme, the current approval process is still processed in a mode of manual dialogue connection, the approval efficiency is low, and the increasing approval data cannot be met. SUMMARY

[0005] The present application aims to provide an intelligent reimbursement process automatic approval system, which solves the problem of low efficiency of manual audit due to the increase of approval data in the background art.

[0006] In order to solve the above-mentioned technical problem of low efficiency of manual audit due to the increase of approval data, the present application adopts the following technical scheme: the present application provides an intelligent reimbursement process automatic approval system, which specifically comprises the following steps: S1, collecting historical reimbursement approval data to establish a reimbursement approval data standard, and establishing a sample set based on the standardized reimbursement approval data; S2, processing the historical reimbursement approval data in the sample set to obtain processed historical reimbursement approval data; S21, filtering the historical reimbursement approval data in the sample set to obtain filtered historical reimbursement approval data; S22, clustering the filtered historical reimbursement approval data through a data clustering method to obtain clustered historical reimbursement approval data; S23, setting the clustered historical reimbursement approval data as processed historical reimbursement approval data; S3, constructing a reimbursement approval analysis model based on the processed historical reimbursement approval data in the sample set, and determining a processed historical reimbursement approval data review template based on the constructed reimbursement approval analysis model; S4, extracting features of the processed historical reimbursement approval data review template through a convolutional neural network to obtain extracted features of the historical reimbursement approval data review template and save them to the sample set; S5, collecting real-time reimbursement approval data and extracting features of the real-time collected reimbursement approval data through a convolutional neural network; S6, comparing the real-time extracted features with the saved features in the sample set through semantic comparison, and processing the reimbursement approval data corresponding to the real-time extracted features based on the comparison result.

[0007] Preferably, the collection of historical reimbursement approval data establishes a reimbursement approval data standard, and the establishment of a sample set based on the standardized reimbursement approval data includes the following steps: The reimbursement approval data includes: reimbursement vouchers, reimbursement amounts, application times, and reimbursement contents; A set of standard reimbursement approval data must include four data: reimbursement vouchers, reimbursement personnel, reimbursement amounts, and reimbursement contents; The collected historical reimbursement approval data is saved through four-tuple ; Among them, represents the reimbursement vouchers in the historical reimbursement approval data, represents the reimbursement amount in the historical reimbursement approval data, represents the application time in the historical reimbursement approval data, represents the reimbursement content in the historical reimbursement approval data; The saved four-tuple is summarized to construct a sample set.

[0008] Preferably, the filtering of the historical reimbursement approval data in the sample set to obtain filtered historical reimbursement approval data includes the following steps: Iterate through the historical reimbursement approval data in the sample set. When detecting a repetition in the historical reimbursement approval data during the iteration process, delete the repeated historical reimbursement approval data and summarize to obtain processed historical reimbursement approval data; Traverse the processed historical reimbursement approval data, locate the processed historical reimbursement approval data with missing content, and delete the historical reimbursement approval data with missing content; Set the historical reimbursement approval data after deletion as filtered historical reimbursement approval data.

[0009] Preferably, the clustering of the filtered historical reimbursement approval data by the data clustering method comprises the following steps: S221, selecting R groups of historical reimbursement approval data in the filtered historical reimbursement approval data as initial clustering centers; S222, according to the distance formula Calculate the distance of each group of historical reimbursement approval data to each clustering center, and divide the historical reimbursement approval data into the category corresponding to the nearest clustering center; Wherein, represents the distance between the th group of historical reimbursement approval data and the th clustering center, represents the th group of historical reimbursement approval data, represents the th clustering center; S223, according to all historical reimbursement approval data in each category, recalculate the clustering center, and update the clustering center; S224, repeat steps S222-S223 until the clustering center converges, and output the clustered historical reimbursement approval data.

[0010] Preferably, the historical reimbursement approval data in the sample set is processed to construct a reimbursement approval analysis model, and the processed historical reimbursement approval data is determined based on the constructed reimbursement approval analysis model. The reimbursement approval audit template comprises the following steps: S31, aggregate and output the processed historical reimbursement approval data, randomly select a group of processed historical reimbursement approval data from the processed historical reimbursement approval data, extract all approval keywords, and construct an approval keyword set ; S32, for each approval keyword in the approval keyword set , calculate the probability of the approval keyword becoming a historical reimbursement approval data audit template.

[0011] Preferably, the probability of each approval keyword in the approval keyword set becoming a historical reimbursement approval data audit template comprises the following steps: The probability formula of the approval keyword becoming a historical reimbursement approval data audit template is as follows: ; wherein, denotes the approval keywords in the set of approval keywords under examination become the first group of processed historical reimbursement approval data subject probability, denotes the number of times the approval keywords appear in the first group of processed historical reimbursement approval data, denotes the frequency parameter, denotes the position parameter of the approval keywords .

[0012] Preferably, the feature extraction of the processed historical reimbursement approval data audit template by the convolutional neural network includes the following steps: S41, the processed historical reimbursement approval data audit template is input into the convolutional neural network as input; The convolutional neural network comprises: Squeeze operations, operations and operations; S42, the input historical reimbursement approval data audit template is compressed by the Squeeze operation of the convolutional neural network, and the feature channel is output; S43, after compression, the channel weight is generated by the operation of the convolutional neural network; The operation formula is as follows: ; wherein, denotes the compressed historical reimbursement approval data audit template matrix, denotes the generated channel weight, denotes the operation; S44, the operation of the convolutional neural network is multiplied by the channel weight generated by the operation to realize the re-labeling of the channel features; The operation formula is as follows: ; wherein, denotes the channel re-labeling result corresponding to the c-th two-dimensional matrix in the matrix composed of the historical reimbursement approval data audit template, denotes the Operation function, represents the channel weight of the cth two-dimensional matrix corresponding to the channel. S45, the features output by each channel after re-calibration are integrated by full connection mode to obtain the extraction features of the historical reimbursement approval data review template and save.

[0013] Preferably, the operation of the convolutional neural network Squeeze The operation compresses the input historical reimbursement approval data review template and outputs the feature channel, including the following steps: Squeeze The operation formula is as follows: ; ; wherein, represents the matrix composed of the input historical reimbursement approval data review template, represents the cth two-dimensional matrix in the matrix composed of the historical reimbursement approval data review template, represents the compression result of the cth two-dimensional matrix, represents Squeeze Operation function, respectively represent the height and width of the channel, represents the row and column of the matrix.

[0014] Preferably, the operation of comparing the real-time extracted features with the features saved in the sample set by the semantic comparison method, and processing the reimbursement approval data corresponding to the real-time extracted features based on the comparison result includes the following steps: By calculating the similarity of the two groups of features, it is judged whether the reimbursement approval data meets the standard; ; wherein, represents the similarity value of the two groups of features, represents the real-time extracted features, represents the features saved in the sample set; When the similarity Sim is [0, 1], it means that the two groups of features are similar, and when the similarity Sim is [0, -1], it means that the two groups of features are not similar; If the two feature vectors are similar, it means that the real-time collected reimbursement approval data meets the reimbursement approval standard; if the two feature vectors are not similar, it means that the real-time collected reimbursement approval data does not meet the reimbursement approval standard.

[0015] Preferably, the intelligent reimbursement process automatic approval system comprises a data collection module, a data processing module, a reimbursement data analysis module, a reimbursement data feature extraction module, and a semantic comparison module. The data collection module is used for collecting reimbursement approval data and historical reimbursement approval data in real time. The data processing module is used for processing the collected reimbursement approval data to obtain processed reimbursement approval data. The reimbursement data analysis module is used for analyzing the processed reimbursement approval data and constructing an analysis model, and determining a reimbursement approval data audit template. The reimbursement data feature extraction module is used for extracting features of the determined reimbursement approval data audit template and the real-time collected reimbursement approval data. The semantic comparison module is used for judging whether the real-time collected reimbursement approval data meets the standard through feature comparison.

[0016] The beneficial effects of the present application are as follows: (1) The present application establishes a reimbursement approval data standard by collecting historical reimbursement approval data, and establishes a sample set based on the standardized reimbursement approval data, processes the historical reimbursement approval data in the sample set, and after the processing is completed, constructs a reimbursement approval analysis model based on the processed historical reimbursement approval data in the sample set, and determines a processed historical reimbursement approval data audit template based on the constructed reimbursement approval analysis model; at the same time, real-time collection of reimbursement approval data, feature extraction and saving of the processed historical reimbursement approval data audit template and the real-time collected reimbursement approval data through a convolutional neural network, and finally comparing the real-time extracted features with the saved features in the sample set through a semantic comparison method, and processing the reimbursement approval data corresponding to the real-time extracted features based on the comparison result, improving the intelligence of the reimbursement process automatic approval.

[0017] (2) The present application collects historical reimbursement approval data, and establishes a reimbursement approval data standard based on the content of the historical reimbursement approval data, and establishes a sample set based on the standardized reimbursement approval data, improving the reliability of the reimbursement approval data.

[0018] (3) The present application traverses the historical reimbursement approval data in the sample set, and deletes the existing duplicate data, removes the duplicate data, traverses the processed historical reimbursement approval data after removing the duplicate data, locates the historical reimbursement approval data with missing content and filters, after filtering, clusters the filtered historical reimbursement approval data through a clustering algorithm, and guarantees the accuracy of the reimbursement approval data.

[0019] (4) The present application outputs the processed historical reimbursement approval data, extracts the approval keywords, and calculates the probability of the extracted approval keywords becoming a historical reimbursement approval data audit template, improving the intelligence of the reimbursement approval data approval.

[0020] (5) This invention uses a convolutional neural network to extract features from the processed historical reimbursement approval data template and the real-time collected reimbursement approval data, thereby further improving the intelligence and real-time performance of the reimbursement approval data approval.

[0021] (6) This invention compares the features extracted in real time with the features stored in the sample collection through semantic comparison, and quantifies the comparison results through similarity value, thereby improving the accuracy of reimbursement approval data approval. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 This is a schematic diagram of the intelligent reimbursement process automated approval method of the present invention. Detailed Implementation

[0024] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0025] In a specific embodiment of the present invention, Reference Figure 1 As shown, the present invention provides an intelligent expense reimbursement process automated approval system, comprising the following steps: S1. Collect historical reimbursement approval data to establish reimbursement approval data standards, and establish a sample set based on the standardized reimbursement approval data; S2. Process the historical reimbursement approval data in the sample set to obtain the processed historical reimbursement approval data. S21. Filter the historical reimbursement approval data in the sample set to obtain the filtered historical reimbursement approval data. S22. Cluster the filtered historical reimbursement approval data using data clustering methods to obtain clustered historical reimbursement approval data. S23. Set the clustered historical reimbursement approval data as the processed historical reimbursement approval data; S3. Based on the historical reimbursement approval data processed from the sample set, construct a reimbursement approval analysis model, and determine the review template for the processed historical reimbursement approval data based on the constructed reimbursement approval analysis model. S4. Use a convolutional neural network to extract features from the processed historical reimbursement approval data review template, obtain the extracted features of the historical reimbursement approval data review template, and save them to the sample set. S5. Collect reimbursement approval data in real time and extract features from the real-time collected reimbursement approval data through a convolutional neural network; S6. Compare the features extracted in real time with the features stored in the sample collection through semantic comparison, and process the reimbursement approval data corresponding to the features extracted in real time based on the comparison results.

[0026] Furthermore, referring to Figure 1 As shown, the process of collecting historical reimbursement approval data to establish reimbursement approval data standards, and then building a sample set based on the standardized reimbursement approval data, includes the following steps: The reimbursement approval data includes: reimbursement voucher, reimbursement amount, application time, and reimbursement details; A set of standard reimbursement approval data must include four items: reimbursement voucher, reimbursement personnel, reimbursement amount, and reimbursement details. Furthermore, the collected historical expense reimbursement approval data will be processed through a quaternion. Save; in, This refers to reimbursement vouchers from historical reimbursement approval data. This represents the reimbursement amount from historical reimbursement approval data. This indicates the application time in historical reimbursement approval data. This indicates the reimbursement details in historical reimbursement approval data; The saved quadruples are used to construct a sample set; Furthermore, referring to Figure 1 As shown, filtering the historical reimbursement approval data in the sample set to obtain the filtered historical reimbursement approval data includes the following steps: The historical reimbursement approval data in the sample set is traversed. When duplicates are detected in the historical reimbursement approval data during the traversal, the duplicate historical reimbursement approval data is deleted and summarized to obtain the processed historical reimbursement approval data. Furthermore, the processed historical reimbursement approval data is traversed to locate reimbursement approval data with missing content, and the historical reimbursement approval data with missing content is deleted. Set the deleted historical reimbursement approval data to the filtered historical reimbursement approval data; Furthermore, referring toFigure 1 As shown, the filtered historical expense reimbursement approval data is clustered using a data clustering method to obtain the clustered historical expense reimbursement approval data, which includes the following steps: S221. Select group R of historical reimbursement approval data as the initial cluster center from the filtered historical reimbursement approval data. S222, According to the distance formula Calculate the distance from each group of historical reimbursement approval data to each cluster center, and classify the historical reimbursement approval data into the category corresponding to the nearest cluster center; in, Indicates the first Group historical reimbursement approval data up to the first Clustering of cluster centers Indicates the first Collect historical expense reimbursement approval data. Indicates the first Cluster centers; S223. Based on all historical reimbursement approval data in each category, recalculate the cluster centers and update them. S224. Repeat steps S222-S223 until the cluster centers converge, and output the clustered historical reimbursement approval data.

[0027] Furthermore, referring to Figure 1 As shown, based on the processed historical reimbursement approval data of the sample set, a reimbursement approval analysis model is constructed, and the review template for the processed historical reimbursement approval data is determined based on the constructed reimbursement approval analysis model, including the following steps: S31. Summarize and output the processed historical reimbursement approval data. Randomly select a set of processed historical reimbursement approval data, extract all approval keywords, and construct a set of approval keywords. ; S32. For the set of approval keywords For each approval keyword in the data, calculate the probability that the approval keyword becomes the review template for historical reimbursement approval data; The probability formula for approval keywords becoming audit templates in historical reimbursement approval data is shown below: ; in, Indicates the set of approval keywords Keywords for approval Become the The probability of the subject of the historical reimbursement approval data after group processing. Indicates the first Approval keywords in historical reimbursement approval data after group processing Number of times it appears Indicates frequency parameters, Keywords indicating approval Position parameters; Furthermore, referring to Figure 1 As shown, the process of extracting features from the processed historical reimbursement approval data template using a convolutional neural network, and saving the extracted features of the historical reimbursement approval data template to the sample set includes the following steps: S41. Input the processed historical reimbursement approval data template into the convolutional neural network. The convolutional neural network includes: Squeeze operate, Operation and operate; S42, through a convolutional neural network Squeeze The operation compresses the input historical reimbursement approval data template and outputs a feature channel; Squeeze The operating formula is as follows: ; ; in, This represents a matrix composed of historical expense reimbursement approval data and review templates. This represents the c-th two-dimensional matrix in the matrix composed of historical expense reimbursement approval data templates. This represents the compression result of the c-th two-dimensional matrix. express Squeeze Operation functions, These represent the height and width of the channel, respectively. Represents the rows and columns of a matrix; S43. After compression, the data is processed through a convolutional neural network. The operation generates channel weights; The operating formula is as follows: ; in, This represents the compressed historical expense reimbursement approval data review template matrix. This represents the generated channel weights. express operate; S44, via convolutional neural networks Operation pair The generated channel weights are multiplied and weighted to achieve recalibration of channel features; The operating formula is as follows: ; in, This represents the channel recalibration result corresponding to the c-th two-dimensional matrix in the matrix composed of historical reimbursement approval data audit templates. express Operation functions, This represents the channel weight corresponding to the c-th two-dimensional matrix; S45. Integrate the features output by each channel after recalibration through a fully connected method to obtain and save the extracted features of the historical reimbursement approval data review template. Furthermore, referring to Figure 1 As shown, the real-time extracted features are compared with the features stored in the sample set using a semantic comparison method, and the reimbursement approval data corresponding to the real-time extracted features are processed based on the comparison results, including the following steps: The similarity between two sets of features is used to determine whether the reimbursement approval data meets the standards. ; in, This represents the similarity value between two sets of features. This indicates features extracted in real time. This indicates the characteristics of the samples stored in a centralized manner; When the similarity Sim is [0,1], it means that the two sets of features are similar; when the similarity Sim is [0,-1], it means that the two sets of features are not similar. If two feature vectors are similar, it means that the real-time collected reimbursement approval data meets the reimbursement approval standards; if two feature vectors are not similar, it means that the real-time collected reimbursement approval data does not meet the reimbursement approval standards.

[0028] In one specific embodiment, the intelligent reimbursement process automated approval system includes: a data collection module, a data processing module, a reimbursement data analysis module, a reimbursement data feature extraction module, and a semantic comparison module; The data collection module is used to collect reimbursement approval data and historical reimbursement approval data in real time. The data processing module is used to process the collected reimbursement approval data to obtain processed reimbursement approval data. The reimbursement data analysis module is used to analyze the processed reimbursement approval data, build an analysis model, and determine the reimbursement approval data review template. The reimbursement data feature extraction module is used to extract features from the determined reimbursement approval data review template and the reimbursement approval data collected in real time. The semantic comparison module is used to determine whether the reimbursement approval data collected in real time meets the standards by comparing features.

[0029] It should be noted that The above content is merely an example and illustration of the concept of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the concept of the invention or exceed the scope defined by the present invention, and all such modifications and additions should fall within the protection scope of the present invention.

Claims

1. An intelligent expense reimbursement process automated approval system, characterized in that, Includes the following steps: S1. Collect historical reimbursement approval data to establish reimbursement approval data standards, and establish a sample set based on the standardized reimbursement approval data; S2. Process the historical reimbursement approval data in the sample set to obtain the processed historical reimbursement approval data. S21. Filter the historical reimbursement approval data in the sample set to obtain the filtered historical reimbursement approval data. S22. Cluster the filtered historical reimbursement approval data using data clustering methods to obtain clustered historical reimbursement approval data. S23. Set the clustered historical reimbursement approval data as the processed historical reimbursement approval data; S3. Based on the historical reimbursement approval data processed from the sample set, construct a reimbursement approval analysis model, and determine the review template for the processed historical reimbursement approval data based on the constructed reimbursement approval analysis model. S4. Use a convolutional neural network to extract features from the processed historical reimbursement approval data review template, obtain the extracted features of the historical reimbursement approval data review template, and save them to the sample set. S5. Collect reimbursement approval data in real time and extract features from the real-time collected reimbursement approval data through a convolutional neural network; S6. Compare the features extracted in real time with the features stored in the sample collection through semantic comparison, and process the reimbursement approval data corresponding to the features extracted in real time based on the comparison results.

2. The intelligent expense reimbursement process automated approval system according to claim 1, characterized in that, The process of collecting historical reimbursement approval data, establishing reimbursement approval data standards, and building a sample set based on the standardized reimbursement approval data includes the following steps: The reimbursement approval data includes: reimbursement voucher, reimbursement amount, application time, and reimbursement details; A set of standard reimbursement approval data must include four items: reimbursement voucher, reimbursement personnel, reimbursement amount, and reimbursement details. The collected historical expense reimbursement approval data will be processed through a quadruple. Save; in, This refers to reimbursement vouchers from historical reimbursement approval data. This represents the reimbursement amount from historical reimbursement approval data. This indicates the application time in historical reimbursement approval data. This indicates the reimbursement details in historical reimbursement approval data; The saved quadruplets are used to construct a sample set.

3. The intelligent expense reimbursement process automated approval system according to claim 1, characterized in that, The process of filtering historical reimbursement approval data in the sample set to obtain filtered historical reimbursement approval data includes the following steps: The historical reimbursement approval data in the sample set is traversed. When duplicates are detected in the historical reimbursement approval data during the traversal, the duplicate historical reimbursement approval data is deleted and summarized to obtain the processed historical reimbursement approval data. The system iterates through the processed historical reimbursement approval data, locates the reimbursement approval data with missing content, and deletes the historical reimbursement approval data with missing content. Set the deleted historical reimbursement approval data to the filtered historical reimbursement approval data.

4. The intelligent expense reimbursement process automated approval system according to claim 1, characterized in that, The process of clustering the filtered historical reimbursement approval data using a data clustering method to obtain the clustered historical reimbursement approval data includes the following steps: S221. Select group R of historical reimbursement approval data as the initial cluster center from the filtered historical reimbursement approval data. S222, According to the distance formula Calculate the distance from each group of historical reimbursement approval data to each cluster center, and classify the historical reimbursement approval data into the category corresponding to the nearest cluster center; in, Indicates the first Group historical reimbursement approval data up to the first Clustering of cluster centers Indicates the first Collect historical expense reimbursement approval data. Indicates the first Cluster centers; S223. Based on all historical reimbursement approval data in each category, recalculate the cluster centers and update them. S224. Repeat steps S222-S223 until the cluster centers converge, and output the clustered historical reimbursement approval data.

5. The intelligent expense reimbursement process automated approval system according to claim 1, characterized in that, The process of constructing a reimbursement approval analysis model based on historical reimbursement approval data processed from the sample set, and determining the review template for the processed historical reimbursement approval data based on the constructed reimbursement approval analysis model, includes the following steps: S31. Summarize and output the processed historical reimbursement approval data. Randomly select a set of processed historical reimbursement approval data, extract all approval keywords, and construct a set of approval keywords. ; S32. For the set of approval keywords For each approval keyword in the data, calculate the probability that the approval keyword will become the audit template for historical reimbursement approval data.

6. The intelligent expense reimbursement process automated approval system according to claim 5, characterized in that, The set of approval keywords For each approval keyword in the data, the probability of that keyword becoming a historical expense reimbursement approval data review template is calculated by the following steps: The probability formula for approval keywords becoming audit templates in historical reimbursement approval data is shown below: ； in, Indicates the set of approval keywords Keywords for approval Become the The probability of the subject of the historical reimbursement approval data after group processing. Indicates the first Approval keywords in historical reimbursement approval data after group processing Number of times it appears Indicates frequency parameters, Keywords indicating approval Position parameters.

7. The intelligent expense reimbursement process automated approval system according to claim 1, characterized in that, The step of extracting features from the processed historical reimbursement approval data template using a convolutional neural network, and saving the extracted features of the historical reimbursement approval data template to a sample set, includes the following steps: S41. Input the processed historical reimbursement approval data template into the convolutional neural network. The convolutional neural network includes: Squeeze operate, Operation and operate; S42, through a convolutional neural network Squeeze The operation compresses the input historical reimbursement approval data template and outputs a feature channel; S43. After compression, the data is processed through a convolutional neural network. The operation generates channel weights; The operating formula is as follows: ； in, This represents the compressed historical expense reimbursement approval data review template matrix. This represents the generated channel weights. express operate; S44, via convolutional neural networks Operation pair The generated channel weights are multiplied and weighted to achieve recalibration of channel features; The operating formula is as follows: ； in, This represents the channel recalibration result corresponding to the c-th two-dimensional matrix in the matrix composed of historical reimbursement approval data audit templates. express Operation functions, This represents the channel weight corresponding to the c-th two-dimensional matrix; S45. Integrate the features output by each channel after recalibration through a fully connected method to obtain and save the extracted features of the historical reimbursement approval data review template.

8. The intelligent expense reimbursement process automated approval system according to claim 7, characterized in that, The convolutional neural network Squeeze The operation compresses the input historical expense reimbursement approval data template and outputs the feature channel, including the following steps: Squeeze The operating formula is as follows: ； ； in, This represents a matrix composed of historical expense reimbursement approval data and review templates. This represents the c-th two-dimensional matrix in the matrix composed of historical expense reimbursement approval data templates. This represents the compression result of the c-th two-dimensional matrix. express Squeeze Operation functions, These represent the height and width of the channel, respectively. Represents the rows and columns of a matrix.

9. The intelligent expense reimbursement process automated approval system according to claim 1, characterized in that, The process of comparing real-time extracted features with features stored in the sample set using semantic comparison, and then processing the reimbursement approval data corresponding to the real-time extracted features based on the comparison results, includes the following steps: The similarity between two sets of features is used to determine whether the reimbursement approval data meets the standards. ； in, This represents the similarity value between two sets of features. This indicates features extracted in real time. This indicates the characteristics of the samples stored in a centralized manner; When the similarity Sim is [0,1], it means that the two sets of features are similar; when the similarity Sim is [0,-1], it means that the two sets of features are not similar. If two feature vectors are similar, it means that the real-time collected reimbursement approval data meets the reimbursement approval standards; if two feature vectors are not similar, it means that the real-time collected reimbursement approval data does not meet the reimbursement approval standards.

10. The intelligent expense reimbursement process automated approval system according to claim 1, characterized in that, include: The system includes a data collection module, a data processing module, a reimbursement data analysis module, a reimbursement data feature extraction module, and a semantic comparison module. The data collection module is used to collect reimbursement approval data and historical reimbursement approval data in real time. The data processing module is used to process the collected reimbursement approval data to obtain processed reimbursement approval data. The reimbursement data analysis module is used to analyze the processed reimbursement approval data, build an analysis model, and determine the reimbursement approval data review template. The reimbursement data feature extraction module is used to extract features from the determined reimbursement approval data review template and the reimbursement approval data collected in real time. The semantic comparison module is used to determine whether the reimbursement approval data collected in real time meets the standards by comparing features.

Images