An evaluation method and system for reimbursement matters

Abstract

The present invention relates to the technology of reimbursement item management, and discloses an evaluation method and system for reimbursement items. The reimbursement item information is submitted through a reimbursement system, and the reimbursement item information includes at least two reimbursement item participants; the reimbursement information is sent to all reimbursement item participants through the reimbursement system, and all reimbursement item participants receive a credit request; all reimbursement item participants feedback the reimbursement item credit result information to the reimbursement system; the reimbursement system processes the received reimbursement item credit result information through a credit evaluation model to obtain a reimbursement item credit score. The evaluation method for reimbursement items designed by the present invention can effectively reduce the management cost of enterprises and save a lot of manpower and material resources in the process of reimbursing item approval.

Description

Technical Field

[0001] The present invention relates to the technology of reimbursement matter management, and particularly to an evaluation method and system for reimbursement matters. Background Art

[0002] Reimbursement management is a common operation in the business process of an enterprise. The review process of reimbursement is mainly to determine the authenticity and compliance of invoices and business matters. Among them, the authenticity of invoices can be verified through the verification service or interface provided by the tax system; compliance can be achieved by setting detailed reimbursement rules and checking them one by one during the approval process; the determination of the authenticity of business matters is to confirm whether the reimbursed expenses are reasonable expenses incurred for carrying out the company's business and prevent dishonest behaviors during the reimbursement process.

[0003] The existing methods for determining the authenticity of reimbursement matters mainly include: providing more detailed supporting information: such as the starting and ending points of taxi tickets; the consumption details in hotel invoices; the list of participants in catering invoices, etc.; manually checking the authenticity of business: for important reimbursements, a special person conducts full - scale or sampling verification with relevant participants, such as colleagues, customers, suppliers or other relevant persons.

[0004] For example, the existing technology CN202011381995.2 only evaluates bills, which consumes a lot of manpower and material resources for other information, thus increasing the enterprise management cost, and at the same time lengthening the reimbursement cycle and affecting the user experience. Summary of the Invention

[0005] Aiming at the problems in the existing reimbursement, which has high operating costs for enterprises and is time - consuming and labor - intensive, the present invention provides an evaluation method and system for reimbursement matters. To solve the above - mentioned technical problems, the present invention is solved by the following technical solutions:

[0006] An evaluation method for reimbursement matters, including a reimbursement system, and the method includes:

[0007] Step 1, submission of reimbursement matters: Submit reimbursement matter information through the reimbursement system, and the reimbursement matter information includes at least two participants in the reimbursement matter;

[0008] Step 2, credit request for reimbursement matters: Send reimbursement information to all participants in the reimbursement matter through the reimbursement system, and all participants in the reimbursement matter receive the credit request;

[0009] Step 3, feedback of reimbursement matter credit information: All participants in the reimbursement matter feedback the reimbursement matter credit result information to the reimbursement system; the reimbursement matter credit result information includes true reimbursement matter information, to - be - verified reimbursement matter information, and untrue reimbursement matter information;

[0010] Step 4: Generation of the credit score for reimbursement items. The reimbursement system processes the received credit result information of reimbursement items through a credit assessment model for reimbursement items to obtain the credit score for reimbursement items.

[0011] Preferably, it also includes the approval of reimbursement items. Receive the credit score of the reimbursement items from the reimbursement system, analyze the credit score of the reimbursement items, and determine the approval status of the reimbursement items. The approval status of the reimbursement items includes reimbursement items approved, reimbursement items rejected for approval, and reimbursement items pending approval;

[0012] For the reimbursement items approved, directly conduct the approval;

[0013] For the reimbursement items rejected for approval, obtain the information rejected for approval through metric learning for feedback;

[0014] For the reimbursement items pending approval, hand them over to manual approval. After the approval is completed, they enter the sample pool of the information to be evaluated as supplementary samples. When the data volume of the sample pool of the information to be evaluated reaches the sample threshold, conduct active learning to screen for high-information samples, and select the screened samples into the dataset for model training.

[0015] Preferably, the credit assessment model is a machine learning classification model based on Xgboost. The method for establishing the machine learning classification model of Xgboost includes:

[0016] Determination of the reimbursement feature vector. The feature vector of the reimbursement item includes the feature vector of the reimbursement person, the time vector feature of the reimbursement, and the type vector feature of the reimbursement;

[0017] Binning of the reimbursement feature values. Conduct chi-square binning according to the type of each feature vector in the feature vector. Among them, the feature vectors with fewer categories do not need to be binned, and the feature vectors with more categories are binned through ordered features, thus ensuring the orderliness of the binning;

[0018] Calculation of the IV value for reimbursement evaluation, through calculate the IV value; where WOE i is the evaluation index value, p yi is the proportion of positive samples in this category in the binning, p ni is the proportion of negative samples in this category in the binning;

[0019]

[0020] Among them, y i is the data volume of positive samples in this category; n i is the data volume of negative samples in this category; y T is the total data volume of positive samples; n T is the total data volume of negative samples;

[0021] Feature data within a reasonable range are selected through the IV value of each feature. The feature data includes the training set, validation set, and test set.

[0022] Preferably, the method of metric learning includes:

[0023] SI: Input the dataset D, where

[0024] D = (x1, y1), (x2, y2),...(x i , y i )...(x n , y n ), where any sample x i is an n-dimensional vector, and y i ∈C1, C2…C k ;

[0025] S2: Calculate the sample neighbor distribution p ij ;

[0026]

[0027] S3: Predict the i-th sample as y k , and the probability of correctly predicting the i-th sample is p i , where

[0028] S4: Optimize the objective score f(A);

[0029]

[0030] S5: Update A using the gradient and judge the updated A. When A is the smallest, reject the reimbursement item as a similar sample in the reimbursement system; otherwise, return to step S2;

[0031]

[0032] Preferably, the method for actively learning to screen high-information samples includes

[0033] (1) Classify the data of reimbursement items through a classifier. The data of reimbursement items is divided into the training set of reimbursement items, the validation set of reimbursement items, and the dataset of unlabeled reimbursement items;

[0034] (2) Initialize the approval item model. When the approval data is greater than the approval threshold, initialize the approval item model;

[0035] (3) Add the newly added approval data of reimbursement items to the dataset of unlabeled approval items;

[0036] (4) Prediction of the dataset without marked approval items. When the dataset without marked approval items accumulates to the unlabeled data threshold, use the approval item model to predict each sample in the unlabeled dataset one by one to obtain the prediction results of each sample;

[0037] (5) Prediction of the value of labeled samples. Measure the prediction of the value of labeled samples according to the uncertainty sampling strategy;

[0038] (6) Update of the approval item model. Select the labeled samples after prediction and combine the approval status results to join the approval item model in (2) for training to obtain a new approval item model;

[0039] (7) Verification of approval item data. Verify the approval item data through the approval item model in (6). If the performance of the approval item model has reached the target, end the iterative process; otherwise, execute (3).

[0040] To solve the above technical problems, the present application also provides an evaluation system for reimbursement items, which includes:

[0041] A submission module for reimbursement items, which submits reimbursement item information through the reimbursement system. The reimbursement item information includes at least two reimbursement item participants;

[0042] A credit request module for reimbursement items, which sends reimbursement information to all reimbursement item participants through the reimbursement system, and all reimbursement item participants receive the credit request;

[0043] A feedback module for reimbursement item credit information. All reimbursement item participants feedback the reimbursement item credit result information to the reimbursement system; the reimbursement item credit result information includes true reimbursement item information, to-be-verified reimbursement item information, and untrue reimbursement item information;

[0044] A generation module for reimbursement item credit scores. The reimbursement system processes the received reimbursement item credit result information through a credit assessment model to obtain the reimbursement item credit scores.

[0045] Preferably, it further includes: an approval module for reimbursement items. The approval module for reimbursement items is used to receive the reimbursement item credit scores of the reimbursement system, analyze the reimbursement item credit scores, and determine the approval status of the reimbursement items. The approval status of the reimbursement items includes reimbursement items approved, reimbursement items rejected for approval, and reimbursement items pending approval.

[0046] To solve the above technical problems, the present application also provides an electronic device, which is an electronic device implemented by an evaluation method for reimbursement items.

[0047] To solve the above technical problems, the present application also provides a storage medium, which is implemented by an evaluation method for reimbursement items.

[0048] Due to the adoption of the above technical solutions, the present invention has remarkable technical effects:

[0049] Through the evaluation method for reimbursement items designed by the present invention, the management cost of enterprises can be effectively reduced, and a large amount of manpower and material resources can be saved for the approval of reimbursement items. Description of the Drawings

[0050] Figure 1 It is the flowchart of Embodiment 1 of the present invention.

[0051] Figure 2 It is the flowchart of Embodiment 2 of the present invention.

[0052] Figure 3 It is the flowchart of the method for actively learning and screening high-information samples of the present invention. Detailed Embodiments

[0053] The present invention will be further described in detail below with reference to the drawings and embodiments.

[0054] Embodiment 1

[0055] An evaluation method for reimbursement items, including a reimbursement system, and the method includes:

[0056] Step 1: Submission of reimbursement items. Submit reimbursement item information through the reimbursement system. The reimbursement item information includes at least two reimbursement item participants. Here, there are two reimbursement item participants, one is a direct participant, and the other is an indirect participant;

[0057] Step 2: Credit request for reimbursement items. Send reimbursement information to all reimbursement item participants through the reimbursement system, and all participants receive the credit request;

[0058] Step 3: Feedback of reimbursement item credit information. All reimbursement item participants feedback the reimbursement item credit result information to the reimbursement system; the reimbursement item credit result information includes true reimbursement item information, to-be-verified reimbursement item information, and untrue reimbursement item information;

[0059] Step 4: Generation of reimbursement item credit scores. The reimbursement system processes the received reimbursement item credit result information through a credit evaluation model to obtain the reimbursement item credit scores.

[0060] The credit evaluation model is a machine learning classification model based on Xgboost. The method for establishing the machine learning classification model of Xgboost includes:

[0061] Determination of reimbursement feature vectors. The feature vectors of reimbursement items include reimbursement personnel feature vectors, reimbursement time vector features, and reimbursement type vector features;

[0062] Binning of reimbursement feature values. Chi-square binning is performed according to the type of each feature vector in the feature vectors. Among them, feature vectors with fewer categories do not need to be binned, and feature vectors with more categories are binned through ordered features, thus ensuring the orderliness of binning;

[0063] Calculation of the IV value for reimbursement evaluation. Through calculate the IV value; among them, WOE i is the evaluation index value, p yi is the proportion of positive samples in this category in the binning, p ni is the proportion of negative samples in this category in the binning;

[0064]

[0065] Among them, y i is the data volume of positive samples in this category; n i is the data volume of negative samples in this category; y T is the total data volume of positive samples; n T is the total data volume of negative samples;

[0066] Feature data within a reasonable range is selected through the IV value of each feature. The feature data includes training sets, validation sets, and test sets. The separation ratios of the training set, validation set, and test set are 70%, 15%, and 15% respectively. Among them, the training set and validation set are used in training the xgboost model. The test set is used to evaluate the accuracy rate and AUC area index of the model after the model training is completed.

[0067] Embodiment 2

[0068] Based on Embodiment 1, this embodiment further includes the approval of reimbursement items. Receive the credit score of reimbursement items in the reimbursement system, and analyze the credit score of reimbursement items to determine the approval status of reimbursement items. The approval status of reimbursement items includes reimbursement items approved through, reimbursement items rejected for approval, and reimbursement items pending approval;

[0069] For reimbursement items approved through, directly conduct the approval;

[0070] For reimbursement items rejected for approval, obtain the information rejected for approval through metric learning for feedback;

[0071] The reimbursement items to be approved are submitted for manual approval. After the approval is completed, they enter the sample pool of the information volume to be evaluated as supplementary samples. When the data volume in the sample pool of the information volume to be evaluated reaches the sample threshold, and the set sample threshold range is 100 - 150, active learning is performed to screen for high-information samples. The selected samples are included in the dataset for model training.

[0072] The credit assessment model is a machine learning classification model based on Xgboost. The method for establishing the machine learning classification model of Xgboost includes:

[0073] Determination of the reimbursement feature vector. The feature vector of the reimbursement item includes the feature vector of the reimbursement personnel, the feature vector of the reimbursement time, and the feature vector of the reimbursement type;

[0074] Binning of the reimbursement feature values. For the determined reimbursement feature vector, binning is performed according to the discrete attributes of the feature vector;

[0075] A chi-square threshold is preset. Here, the set threshold is 0.95. The instances are sorted according to the attribute to be discretized, and each instance belongs to an interval merging interval; the chi-square value of each pair of adjacent intervals is calculated, and the pair of intervals with the smallest chi-square value is merged to determine whether the stop condition is satisfied. If not, continue; otherwise, stop; The stop conditions are:

[0076] The number of bins reaches the limit condition and the chi-square value of the smallest adjacent bins is greater than the threshold. The threshold here is set to

[0077] Calculation of the reimbursement evaluation IV value, through the IV value is calculated; where WOE i is the evaluation index value, p yi is the proportion of positive samples in the category in the bin, p ni is the proportion of negative samples in the category in the bin;

[0078]

[0079] where y i is the data volume of positive samples in the category; n i is the data volume of negative samples in the category; y T is the total data volume of positive samples; n T is the total data volume of negative samples;

[0080] Determination of the evaluation score of the reimbursement item. The evaluation score of the reimbursement item is determined based on the AUC area; the AUC area is calculated by screening the IV value.

[0081] The methods of metric learning include:

[0082] S1: Input the dataset D, where

[0083] D = (x1, y1), (x2, y2),...(x i , y i )...(x n , y n ), where any sample x i is an n-dimensional vector, and y i ∈ C1, C2…C k ;

[0084] S2: Calculate the sample neighborhood distribution p ij ;

[0085]

[0086] S3: Predict the i-th sample as y k , and the probability of correctly predicting the i-th sample is p i , where

[0087] S4: Optimize the objective score f(A);

[0088]

[0089] S5: Update A using the gradient and judge the updated A. When A is the smallest, reject the reimbursement item as a similar sample in the reimbursement system; otherwise, return to step S2;

[0090]

[0091] Through metric learning, when the approval of a reimbursement item is rejected, the reimbursement system will automatically give a similar existing reimbursement item in the reimbursement system as evidence for rejecting the approved reimbursement item, determine the similarity of the learning samples, and thus find the closest reimbursement item sample.

[0092] The method for actively learning and screening high-information samples includes:

[0093] (1) Classify the data of reimbursement items through a classifier. The data of reimbursement items are divided into a training set of reimbursement items, a validation set of reimbursement items, and an unlabeled dataset of reimbursement items;

[0094] (2) Initialize the approval item model. When the approval data is greater than the approval threshold, and the maximum approval threshold is greater than 1500, initialize the approval item model;

[0095] (3) Add the newly added approval data of reimbursement items to the dataset of unlabeled approval items;

[0096] (4) Prediction of the dataset without marked approval items. When the dataset without marked approval items accumulates to the unlabeled data approval item threshold, where the threshold range of the approval items is 100 - 150, use the approval item model to predict each sample of the unlabeled dataset one by one to obtain the prediction result of each sample;

[0097] (5) Prediction of the value of the labeled sample. Measure the prediction of the value of the labeled sample according to the uncertainty sampling strategy;

[0098] (6) Update of the approval item model. Select the labeled samples after prediction, and combine the approval status results to join the approval item model in (2) for training to obtain a new approval item model;

[0099] (7) Verification of the approval item data. Use the approval item model in (6) to verify the approval item data. If the performance of the approval item model has reached the target, end the iteration process; otherwise, execute (3).

[0100] Example 3

[0101] Based on Example 1, an evaluation system for reimbursement items includes:

[0102] A submission module for reimbursement items, which submits reimbursement item information through the reimbursement system. The reimbursement item information includes at least 2 participants;

[0103] A credit request module for reimbursement items, which sends reimbursement information to all participants through the reimbursement system, and all participants receive the credit request;

[0104] A feedback module for reimbursement item credit information. All participants feedback the reimbursement item credit result information to the reimbursement system; the reimbursement item credit result information includes true reimbursement item information, to-be-verified reimbursement item information, and untrue reimbursement item information;

[0105] A generation module for reimbursement item credit scores. The reimbursement system processes the received reimbursement item credit result information through a credit assessment model to obtain the reimbursement item credit score.

[0106] Example 4

[0107] Based on the above example, this example further includes: An approval module for reimbursement items. The approval module for reimbursement items is used to receive the reimbursement item credit score of the reimbursement system, analyze the credit score, and determine the approval status of the reimbursement item. The approval status of the reimbursement item includes reimbursement items approved, reimbursement items rejected for approval, and reimbursement items pending approval.

[0108] Example 5

[0109] Based on the above embodiments, this embodiment provides an electronic device.

[0110] Embodiment 6

[0111] Based on the above embodiments, this embodiment provides a storage medium.

Claims

1. An evaluation method for reimbursement matters, including a reimbursement system, the method comprising: Step 1, submission of reimbursement matters: Submit reimbursement matter information through the reimbursement system, and the reimbursement matter information includes at least two participants in the reimbursement matter; Step 2, credit request for reimbursement matters: Send reimbursement information to all participants in the reimbursement matter through the reimbursement system, and all participants in the reimbursement matter receive the credit request; Step 3, feedback of reimbursement matter credit information: All participants in the reimbursement matter feedback the reimbursement matter credit result information to the reimbursement system; the reimbursement matter credit result information includes true reimbursement matter information, reimbursement matter information to be verified, and untrue reimbursement matter information; Step 4, generation of reimbursement matter credit scores: The reimbursement system processes the received reimbursement matter credit result information through a credit evaluation model to obtain a reimbursement matter credit score; it also includes the approval of reimbursement matters, receiving the reimbursement matter credit score of the reimbursement system, and analyzing the credit score to determine the approval status of the reimbursement matter. The approval status of the reimbursement matter includes reimbursement matters approved, reimbursement matters rejected for approval, and reimbursement matters pending approval; For reimbursement matters approved, directly conduct the approval; For reimbursement matters rejected for approval, obtain the reimbursement matter information rejected for approval through metric learning for feedback; For reimbursement matters pending approval, enter the sample pool of the amount of reimbursement matters to be evaluated as supplementary samples of reimbursement matters. When the data volume of the sample pool of the amount of reimbursement matters to be evaluated reaches the reimbursement matter sample threshold, actively learn to screen high-information samples and select the screened samples into the dataset for model training.

2. The evaluation method of a reimbursement item according to claim 1, wherein The reimbursement matter credit evaluation model is a machine learning classification model based on Xgboost. The method for establishing the machine learning classification model of Xgboost includes: Determination of reimbursement matter feature vectors: The feature vectors of reimbursement matters include reimbursement personnel, reimbursement time, and reimbursement type; Binning of reimbursement feature values: Perform chi-square binning according to the type of each feature vector in the reimbursement feature vector; Calculation of the IV value for reimbursement evaluation is performed by calculating the IV value; wherein, WOE i is the evaluation index value, p yi is the proportion of positive samples in the bin for this type, and p ni is the proportion of negative samples in the bin for this type; Among them, y i is the number of positive sample data in the bin; n i is the number of negative sample data in this type; y T is the total number of positive sample data; n T is the total number of negative sample data; Screen out reimbursement matter feature data within a reasonable range through the IV value of each feature. The feature data includes a training set, a validation set, and a test set.

3. The evaluation method of a reimbursement item according to claim 1, characterized in that The method for reimbursement matter metric learning includes: S1: Input the dataset D, D = (x1, y1), (x2, y2), …(x i , y i )…(x n , y n ), where any sample x i is an n-dimensional vector, and y i ∈ C1, C2…C k ; S2: Calculate the sample nearest neighbor distribution p ij , S3: Predict the i-th sample as y k , and the probability of correctly predicting the i-th sample is p i , where S4: Optimize the target score f(A); S5: Update A using the gradient and judge the updated A. When A is the smallest, reject the reimbursement matter as a similar sample in the reimbursement system; otherwise, return to step S2; 4. The evaluation method of a reimbursement item according to claim 1, wherein The method for actively learning to screen high-information samples includes: (1) Classify the data of reimbursement matters. The data of reimbursement matters is divided into a training set of accumulated reimbursement matters, a validation set of reimbursement matters, and an unlabeled reimbursement matter dataset; (2) Initialization of the approval matter model: When the approved and unapproved approval matters reach the approval matter threshold, initialize the approval matter model; (3) Add the newly added reimbursement matter approval data to the unlabeled approval matter dataset; (4) Prediction of the dataset without marked approval items. When the dataset without marked approval items accumulates to the unlabeled data threshold, use the approval item model to predict each sample of the unlabeled dataset one by one to obtain the prediction results of each sample; (5) Prediction of the value of labeled samples. Measure the prediction of the value of labeled samples according to the uncertainty sampling strategy; (6) Update of the approval item model. Select the labeled samples after prediction and combine the approval status results to train the approval item model in (2) to obtain a new approval item model; (7) Verification of approval item data. Verify the approval item data through the approval item model in (6). If the performance of the approval item model has reached the target, end the iteration process; otherwise, execute (3).

5. An electronic device, characterized in that, An electronic device implemented by the evaluation method of a reimbursement item according to any one of claims 1-4.

6. A storage medium, characterized in that, A storage medium implemented by the evaluation method of a reimbursement item according to any one of claims 1-4.

Images