Medical insurance charging information intelligent checking method and device, electronic equipment and storage medium

Abstract

The application provides a medical insurance charging information intelligent verification method and device, electronic equipment and a storage medium. The method comprises the following steps: according to each charging item in the to-be-audited medical record, respectively extracting a medical order feature of a medical order name in the to-be-audited medical record and a charging item feature of each charging item, the medical order feature comprises a medical order sub-feature corresponding to each charging item, and a weight of a target feature component related to the charging item corresponding to the medical order sub-feature in the medical order sub-feature is greater than a weight of a non-target feature component. Similarities between each medical order sub-feature of the medical order name and the charging item feature of the corresponding charging item are calculated, and matching degrees between the medical order name and each charging item are determined. By using the technical scheme, the weight of the feature component related to the charging item that needs to be matched and analyzed in the medical order feature can be improved, so that the accuracy of measuring the similarity between the medical order name and the charging item, that is, the accuracy of information matching, is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of natural language processing, and particularly relates to a medical insurance charging information intelligent verification method and device, an electronic device and a storage medium. BACKGROUND

[0002] Medical insurance is one of the five major insurance types of social security in China, and is an important measure to benefit the people proposed by the state to ensure the basic medical treatment of the people. However, with the popularization of medical insurance, the phenomenon of medical insurance reimbursement fees not being consistent with medical orders has appeared, that is, there is no corresponding medical order for medical insurance reimbursement fees, so the audit of medical insurance reimbursement fees is particularly important.

[0003] The audit of medical insurance charging items in medical records is usually performed by querying medical insurance charging items without matching medical orders from all medical insurance charging items in medical records according to the matching relationship between each medical insurance charging item and each medical order in the medical record, thereby realizing the audit of medical insurance charging items. In the prior art, a text matching method is usually used to determine the matching relationship between each medical insurance charging item and each medical order in the medical record. However, the existing text matching method only calculates the features of the medical insurance charging item and the features of the medical order respectively, and determines whether the two features match according to the similarity between the two features. However, in the medical field, one medical order can be a mixed item, for example, the medical order is "abdominal and chest color Doppler ultrasound", which includes two items, abdominal color Doppler ultrasound and chest color Doppler ultrasound, and the medical order corresponds to at least two medical insurance charging items, resulting in a situation where one medical order corresponds to multiple medical insurance charging items. For a mixed item medical order, only using the text of the medical order to determine the features of the medical order will affect the accuracy of the matching between the medical order and the medical insurance charging item, thereby affecting the accuracy of the audit of the medical insurance charging item. SUMMARY

[0004] Based on the defects and deficiencies of the prior art, the present application provides a medical insurance charging information intelligent verification method, device, electronic device and storage medium, which can improve the accuracy of information matching.

[0005] The first aspect of the present application provides a medical insurance charging information intelligent verification method, comprising:

[0006] According to each charging item in the to-be-audited medical record, the medical order feature of the medical order name in the to-be-audited medical record is extracted, and the charging item feature of each charging item in the to-be-audited medical record is extracted respectively, the medical order feature includes a medical order sub-feature corresponding to each charging item, the weight of a target feature component in the medical order sub-feature is greater than the weight of a non-target feature component, and the target feature component is a feature component related to the charging item corresponding to the medical order sub-feature;

[0007] The similarity between each medical order sub-feature of the medical order name and the charge item feature of the corresponding charge item is calculated, and the matching degree between the medical order name and each charge item is determined.

[0008] The second aspect of the present application provides a medical insurance charge information intelligent verification device, comprising:

[0009] The feature extraction module is configured to extract a medical order feature of a medical order name in the to-be-audited medical record according to each charge item in the to-be-audited medical record, and extract a charge item feature of each charge item in the to-be-audited medical record respectively, the medical order feature comprises a medical order sub-feature corresponding to each charge item, and a weight of a target feature component in the medical order sub-feature is greater than a weight of a non-target feature component, the target feature component is a feature component related to the charge item corresponding to the medical order sub-feature;

[0010] The matching module is configured to calculate the similarity between each medical order sub-feature of the medical order name and the charge item feature of the corresponding charge item, and determine the matching degree between the medical order name and each charge item.

[0011] The third aspect of the present application provides an electronic device, comprising a memory and a processor;

[0012] The memory is connected with the processor, and is configured to store a program.

[0013] The processor is configured to realize the medical insurance charge information intelligent verification method by running the program in the memory.

[0014] The fourth aspect of the present application provides a storage medium, and the storage medium stores a computer program, and the computer program is executed by a processor to realize the medical insurance charge information intelligent verification method.

[0015] The medical insurance charging information intelligent verification method provided in the application extracts the medical order characteristics of the medical order name in the to-be-audited medical record and the charging item characteristics of each charging item in the to-be-audited medical record according to each charging item in the to-be-audited medical record. The medical order characteristics include the medical order sub-features corresponding to each charging item, and the weight of the target feature component corresponding to the medical order sub-feature in the medical order sub-feature is greater than the weight of the non-target feature component. The similarity between each medical order sub-feature of the medical order name and the charging item characteristics of the corresponding charging item is calculated to determine the matching degree between the medical order name and each charging item. When the medical order name and a certain charging item are matched and analyzed, the medical order characteristics are extracted by using the medical order name and the charging item, so as to improve the weight of the feature component related to the charging item that needs to be matched and analyzed in the medical order characteristics, which can make the extracted medical order characteristics more conducive to measuring the similarity between the medical order name and the charging item, thereby improving the accuracy of measuring the similarity between the medical order name and the charging item, that is, the accuracy of information matching can be improved. BRIEF DESCRIPTION OF DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or prior art description will be briefly introduced below. Obviously, the drawings in the following description are only embodiments of the present application, and other drawings can also be obtained according to the provided drawings without creative labor for those skilled in the art.

[0017] Figure 1 is a flowchart of a medical insurance charging information intelligent verification method provided by an embodiment of the present application;

[0018] Figure 2 is a processing flowchart of extracting medical order characteristics of a medical order name provided by an embodiment of the present application;

[0019] Figure 3 is a structure diagram of a semantic feature extraction model provided by an embodiment of the present application;

[0020] Figure 4 is a structure diagram of a self-attention layer in a semantic feature extraction model provided by an embodiment of the present application;

[0021] Figure 5 is a structure diagram of a cross-attention layer in a semantic feature extraction model provided by an embodiment of the present application;

[0022] Figure 6 is a flowchart of another medical insurance charging information intelligent verification method provided by an embodiment of the present application;

[0023] Figure 7is a processing flow schematic diagram provided by an embodiment of the present application for determining a matching degree threshold corresponding to a medical record charging item;

[0024] Figure 8 is a processing flow schematic diagram provided by an embodiment of the present application for training a semantic feature extraction model;

[0025] Figure 9 is a processing flow schematic diagram provided by an embodiment of the present application for constructing a sample training set;

[0026] Figure 10 is a structural schematic diagram of an intelligent medical insurance charging information verification device provided by an embodiment of the present application;

[0027] Figure 11 is a structural schematic diagram of an electronic device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0028] The technical solution of the embodiment of the present application is applicable to an application scenario of information matching, and is particularly applicable to information matching between medical insurance charging items and medical orders. By using the technical solution of the embodiment of the present application, the accuracy of measuring the similarity between a medical order name and a charging item can be improved, that is, the accuracy of information matching can be improved.

[0029] In order to avoid the occurrence of a bad event of cheating medical insurance funds, it is necessary to audit medical insurance reimbursement fees, that is, to analyze whether the medical insurance charging items in a medical record and the medical orders in the medical record are matched with each other. If the medical insurance charging items in the medical record do not have medical orders matched therewith, it indicates that the medical insurance charging items do not belong to the reimbursable items corresponding to the medical record. Therefore, for the audit of medical insurance reimbursement fees, the matching analysis between the medical insurance charging items in the medical record and the medical orders in the medical record is particularly important.

[0030] In the prior art, for matching analysis of two fields, the features of the two fields are usually extracted respectively, and then the similarity between the features of the two fields is used to determine whether the two fields are matched. For example, the features of a charging item and the features of a medical order are extracted respectively, and then the similarity between the features of the charging item and the features of the medical order is calculated to determine whether the charging item and the medical order are matched. However, in the medical field, one medical order can be a mixed item, for example, the medical order is "abdominal and chest color Doppler ultrasound", so the medical order includes two items of abdominal color Doppler ultrasound and chest color Doppler ultrasound, and the medical order corresponds to at least two medical insurance charging items, resulting in a situation that one medical order corresponds to multiple medical insurance charging items. For a mixed item medical order, only using the text of the medical order to determine the features of the medical order will affect the accuracy of measuring the similarity between the medical order and the medical insurance charging item, that is, the accuracy of information matching is low, thereby affecting the accuracy of the audit of the medical insurance charging item.

[0031] In view of the above deficiencies of the prior art and the fact that the accuracy of information matching between medical orders and medical insurance charging items is low, the present inventors have conducted research and experiments and propose an intelligent medical insurance charging information verification method which can improve the accuracy of measuring the similarity between medical order names and charging items, i.e., can improve the accuracy of information matching.

[0032] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the scope of protection of the present application.

[0033] The embodiments of the present application propose an intelligent medical insurance charging information verification method, as shown in Figure 1 The method comprises the following steps.

[0034] S101, according to each charging item in the to-be-audited medical record, extracting the medical order features of the medical order names in the to-be-audited medical record, and respectively extracting the charging item features of each charging item in the to-be-audited medical record.

[0035] Specifically, the medical record that needs to be audited for medical insurance charging items is taken as the to-be-audited medical record, and the to-be-audited medical record records the medical order details issued by the doctor for the patient, i.e., the medical order list, and the expense details generated when each charging item is executed, i.e., the charging list. In this embodiment, the medical order names and charging items in the to-be-audited medical record need to be determined according to all the medical order details and all the expense details in the to-be-audited medical record.

[0036] For each charge item, the charge item features of each charge item need to be extracted respectively; for the order features of each order name, the order features need to be extracted according to each charge item, one order name needs to be combined with each charge item to extract the order sub-features of the order name corresponding to each charge item, for example, the order name combined with the charge item A, the weight of the feature component related to the charge item A in the features corresponding to the order name is increased, so as to obtain the order sub-feature of the order name corresponding to the charge item A, and the order sub-features of the order name corresponding to each charge item are taken as the order features of the order name. That is, the order features of the order name include the order sub-features corresponding to each charge item, and the weight of the target feature component in the order sub-feature is greater than the weight of the non-target feature component, wherein the target feature component is the feature component related to the charge item corresponding to the order sub-feature. According to the above-mentioned extraction mode of one order feature, the order features of each order name can be extracted. In this embodiment, by increasing the weight of the feature component related to the charge item to be matched and analyzed in the order feature, the order feature extracted can be more conducive to measuring the similarity between the order name and the charge item, thereby improving the accuracy of measuring the similarity between the order name and the charge item, that is, the accuracy of information matching can be improved.

[0037] In addition, the order details in the to-be-audited medical record include the order names issued by the doctor for the patient, and the execution of each order; the expense details in the to-be-audited medical record include the recorded expenses, the expense codes carried by each expense, and whether each expense belongs to a payment expense or a refund expense. Among them, the order details can include multiple orders, and the expense details can include multiple expenses.

[0038] In this embodiment, the expense code carried by each expense is the standard three-directory code corresponding to the expense, and the expense standard name corresponding to the expense code can be queried from the pre-stored three-directory file according to the expense code of each expense. Then, according to whether each expense belongs to a payment expense or a refund expense, a de-interference operation is performed on all expense standard names corresponding to the expense details, that is, the expense standard names corresponding to the expenses paid and then refunded in the expense details are excluded, and finally a deduplication operation is performed on all de-interference expense standard names, that is, the repeated expense standard names are recorded only once, and all deduplicated expense standard names are taken as the charge items corresponding to the to-be-audited medical record.

[0039] For the order details, all order names are extracted, and according to the execution of each order name, all order names are deinterference, that is, all unexecuted order names are excluded, and finally all deinterference order names are de-duplicated, that is, repeated order names are recorded only once, so as to obtain the order names corresponding to the audit medical record.

[0040] Further, the embodiment can extract the charge item features of each charge item and the order features of each order name by using a pre-trained semantic feature extraction model. The semantic feature extraction model is obtained by extracting features of sample charge items and sample order names in a sample medical record generated by using a medical corpus, and training the sample charge items and the sample order names based on the extracted features. The model is trained by using a large-scale medical corpus, so that the trained model can output feature vectors effectively representing semantic segments in a medical context, for example, the similarity between the feature vector of "potassium determination" and the feature vector of "renal function" is high, and the similarity between the feature vector of "eighteen-channel electrocardiogram" and the feature vector of "twelve-channel electrocardiogram" is low. Compared with a simple rule-based text matching method, the feature extraction accuracy of the semantic feature extraction model trained by using the sample medical record generated by using the medical corpus is higher.

[0041] S102, calculate the similarity between each order sub-feature in the order feature of the order name and the charge item feature of the corresponding charge item, and determine the matching degree between the order name and each charge item.

[0042] Specifically, after determining the charge item features of each charge item in the audit medical record and the order features of each order name, the order name needs to be matched and analyzed with each charge item to determine the matching degree between the order name and each charge item. When matching and analyzing the order name with each charge item, if the order name is matched and analyzed with charge item A, the order sub-feature corresponding to charge item A is extracted from the order feature of the order name, and then the similarity between the order sub-feature and the charge item feature of charge item A is calculated, which is taken as the matching degree between the order name and charge item A. In this way, the matching degree between the order name and each charge item can be determined, so that the matching degree between each order name and each charge item can be determined.

[0043] By determining the matching degree between the medical order name and each charge item, the charge item matching the medical order name can be queried from all the charge items, so that the matching charge item corresponding to each medical order name can be determined according to the matching degree between each medical order name and each charge item, and the matching analysis between the charge item and the medical order name in the to-be-audited medical record is realized, so as to analyze whether each charge item in the to-be-audited medical record has a matching medical order name according to the matching relationship between the analyzed charge item and the medical order name. For the charge item without a matching medical order name, it does not meet the medical insurance reimbursement rules, and therefore, the charge item that does not meet the medical insurance reimbursement rules in the to-be-audited medical record can be queried according to the matching relationship between the analyzed charge item and the medical order name.

[0044] As can be seen from the above introduction, the medical insurance charge information intelligent verification method provided by the embodiments of the present application extracts the medical order characteristics of the medical order name and the charge item characteristics of each charge item in the to-be-audited medical record according to each charge item in the to-be-audited medical record. The medical order characteristics include medical order sub-features corresponding to each charge item, and the weight of the target feature component corresponding to the medical order sub-feature in the medical order sub-feature is greater than the weight of the non-target feature component. The similarity between each medical order sub-feature of the medical order name and the charge item characteristics of the corresponding charge item is calculated to determine the matching degree between the medical order name and each charge item. When the medical order name and a certain charge item are matched and analyzed, the medical order characteristics are extracted using the medical order name and the charge item to improve the weight of the feature component related to the charge item that needs to be matched and analyzed in the medical order characteristics, so that the extracted medical order characteristics are more conducive to measuring the similarity between the medical order name and the charge item, thereby improving the accuracy of measuring the similarity between the medical order name and the charge item, i.e., the accuracy of information matching can be improved.

[0045] As an optional implementation, another embodiment of the present application discloses that after step S102, the following steps can also be performed:

[0046] According to the matching degree between the medical order name and each charge item, the charge item with a matching degree greater than a preset matching degree threshold value is screened from each charge item in the to-be-audited medical record as the charge item matching the medical order name.

[0047] Specifically, the matching degree threshold value can be set in advance, and then the sizes of each matching degree and the matching degree threshold value are compared. The charge item with a matching degree greater than the matching degree threshold value in each charge item in the to-be-audited medical record is taken as the charge item matching the medical order name. The matching degree threshold value can be a fixed threshold value set in advance, and all matching degrees are compared with the threshold value.

[0048] However, in the charge items, there are multiple medical order names that match one charge item, such as the special charge item "ultrasound computer graphic report", and some color ultrasound or ultrasound examinations involve the charge item "ultrasound computer graphic report". Therefore, for special charge items, a lower matching degree threshold needs to be set to include more matchable medical order names, and for some one-to-one matching charge items, a higher matching degree threshold needs to be set to improve the matching accuracy between the charge items and the medical order names. Therefore, different charge items can set different matching degree thresholds, and when the matching degree and the matching degree threshold are compared, the matching degree threshold can also be automatically adjusted according to the charge item corresponding to the matching degree compared with it, so as to improve the matching accuracy between the medical order name and the charge item.

[0049] As an optional implementation, referring to Figure 2 As shown in the figure, another embodiment of the present application discloses that in step S101, according to the charge item in the to-be-audited medical record, the medical order features of the medical order name in the to-be-audited medical record are extracted, including the following steps:

[0050] S201, obtaining the charge item coding features of the charge item, and obtaining the medical order coding features of the medical order name.

[0051] Specifically, the present embodiment can perform coding operation on each charge item and each medical order name, so as to obtain the charge item coding features corresponding to each charge item and the medical order coding features corresponding to each medical order name. The present embodiment can use a coder to perform text coding on the charge item and the medical order name, or use a pre-trained semantic feature extraction model to input each charge item and each medical order name into the semantic feature extraction model for coding. Referring to the structure diagram of the semantic feature extraction model shown in the figure, x1-x5 are text fields input into the semantic feature extraction model, and Segment Embedding, Position Embedding and Token Embedding in the semantic feature extraction model can perform coding operation on the text fields input into the semantic feature extraction model. Therefore, inputting the charge item into the semantic feature extraction model can obtain the charge item coding features, and inputting the medical order name into the semantic feature extraction model can obtain the medical order coding features. Figure 3

[0052] S202, based on the charge item coding features, increasing the weight of the feature component related to the charge item coding features in the medical order coding features.

[0053] ​Specifically, the embodiment can improve the weight of the feature component in the order feature related to the charge item order feature based on the attention mechanism, so as to obtain the order feature corresponding to the charge item. The embodiment can realize the adjustment of the weight of the feature component in the order feature by setting the attention structure in the semantic feature extraction model. The attention structure can be composed of 12 layers of attention layers, or 4 or 6 layers of attention layers can be used according to the needs. In the attention structure, the fewer the network layers, the faster the model operation speed, the less the memory occupation, but the performance will also decrease accordingly. For example, Figure 3 Transformer Block 1~Transformer Block L in the above formula are all attention layers, wherein, Transformer Block 1~Transformer Block L-1 are self-attention layers, and a multi-head self-attention mechanism is adopted. The structure diagram of each self-attention layer is shown in Figure 4 The last layer of attention layer Transformer Block L in the attention structure is a cross-attention layer, and a multi-head cross-attention mechanism is adopted. The structure diagram of the cross-attention layer is shown in Figure 5 .

[0054] The formula of the attention mechanism adopted by the attention structure is as follows:

[0055]

[0056] In the semantic feature extraction model, K and V represent order features, and Q represents a charge item order feature.

[0057] S203, generating order sub-features corresponding to the charge item according to the order feature with improved weight.

[0058] Specifically, based on the charge item order feature, the order feature corresponding to the charge item is obtained, and the order feature is taken as the order sub-feature of the order name corresponding to the charge item. Through the above manner, after obtaining the order sub-features of the order name corresponding to each charge item, all the order sub-features of the order name are combined to form the order feature of the order name.

[0059] For example, for the order name a and the charge item A, the order name a and the charge item A are input into the semantic feature extraction model, the order encoding feature a1 of the order name a and the charge item encoding feature A1 of the charge item A are obtained through the Segment Embedding layer, the Position Embedding layer and the Token Embedding layer in the semantic feature extraction model, and the order encoding feature a1 and the charge item encoding feature A1 are input into the attention structure. The weight of the feature component related to the charge item encoding feature A1 in the order encoding feature a1 can be increased, so that the order sub-feature a-A corresponding to the charge item A is obtained and output. Through the above manner, the order sub-feature a-B corresponding to the charge item B, the order sub-feature a-C corresponding to the charge item C and the like can also be obtained, and the order sub-feature a-A, the order sub-feature a-B, the order sub-feature a-C and the like are taken as the order features of the order name a. The order sub-feature b-A corresponding to the charge item A, the order sub-feature b-B corresponding to the charge item B, the order sub-feature b-C corresponding to the charge item C and the like of the order name b can also be obtained, and the order sub-feature b-A, the order sub-feature b-B, the order sub-feature b-C and the like are taken as the order features of the order name b.

[0060] As an optional implementation, referring to FIG. 13, another embodiment of the present application discloses that before the step of "selecting, from the charge items of the to-be-audited medical record, the charge item with the matching degree greater than the preset matching degree threshold as the charge item matched with the order name, according to the matching degree between the order name and each charge item", the following step is included: Figure 6

[0061] S603, determining the matching degree threshold corresponding to each charge item of the to-be-audited medical record according to the matching degree threshold corresponding to the medical record charge item.

[0062] Specifically, the present embodiment needs to collect sample medical records with the matching relationship between the medical record charge item and the order name labeled in advance, and determine the matching degree threshold corresponding to each medical record charge item according to the matching relationship between the medical record charge item and the order name in all sample medical records. For each charge item in the to-be-audited medical record, the matching degree threshold corresponding to each charge item can be directly queried from the matching degree threshold corresponding to the medical record charge item determined in advance.

[0063] S604, selecting, from the charge items of the to-be-audited medical record, the charge item with the matching degree greater than the corresponding matching degree threshold as the charge item matched with the order name, according to the matching degree between the order name and each charge item and the matching degree threshold corresponding to each charge item.

[0064] ​Specifically, the matching degree between the medical order name and the charge item is compared with the matching degree threshold corresponding to the charge item. If the matching degree is greater than the matching degree threshold, it is determined that the charge item is the charge item matched with the medical order name. For example, if the matching degree between the medical order name a and the charge item A is greater than the matching degree threshold corresponding to the charge item A, it is determined that the charge item A is matched with the medical order name a. If the matching degree between the medical order name a and the charge item B is greater than the matching degree threshold corresponding to the charge item B, it is determined that the charge item B is matched with the medical order name a. If the matching degree between the medical order name b and the charge item A is greater than the matching degree threshold corresponding to the charge item A, it is determined that the charge item A is matched with the medical order name b.

[0065] Figure 6 Steps S601-S602 in the embodiment are the same as steps S101-S102 in the embodiment shown in Figure 1 The embodiment does not specifically describe steps S601-S602.

[0066] As an optional implementation, referring to FIG. 7, another embodiment of the present application discloses that the matching degree threshold corresponding to the medical record charge item is determined by the following steps: Figure 7

[0067] S701. Extract the medical record charge item and all medical order names matched with the medical record charge item from all sample medical records in the sample training set.

[0068] Specifically, the embodiment can pre-collect the sample training set, wherein the sample training set stores at least one sample medical record, and all sample medical records in the sample training set are labeled with the matching relationship between each medical record charge item and each medical order name. The embodiment needs to extract the medical record charge item from all sample medical records, and extract all medical order names matched with the medical record charge item from all sample medical records for each medical record charge item.

[0069] S702. According to the medical record charge item, extract the medical order sub-features of all medical order names matched with the medical record charge item, and extract the medical record charge item features of the medical record charge item.

[0070] ​Specifically, the embodiment needs to extract the medical record charge item features of the medical record charge item, and extract the medical order sub-features of all medical order names matched with the medical record charge item according to the medical record charge item features of the medical record charge item. Among the medical order sub-features of the medical order names matched with the medical record charge item, the weight of the feature components related to the medical record charge item is greater than the weight of other feature components. The extraction of the medical record charge item features and the extraction of the medical order sub-features of the medical order names corresponding to the medical record charge item can be performed by using a pre-trained semantic feature extraction model. The specific extraction method is as described in the above embodiment, and this step will not be specifically described.

[0071] S703, determining a matching degree threshold corresponding to the medical record charge item according to the similarity between the medical record charge item features and each medical order sub-feature matched with the medical record charge item.

[0072] Specifically, the embodiment can calculate the similarity between the medical record charge item features and the medical order sub-features of each medical order name matched with the medical record charge item and corresponding to the medical record charge item. The embodiment can determine the similarity between the features by calculating the cosine distance between the feature vectors or the Euclidean distance between the feature vectors, and the embodiment does not limit the calculation method of the similarity.

[0073] For example, for the medical record charge item A, the medical order names a, b, and c matched with the medical record charge item A, the medical order sub-feature a-A of the medical order name a corresponding to the medical record charge item A, the medical order sub-feature b-A of the medical order name b corresponding to the medical record charge item A, and the medical order sub-feature c-A of the medical order name c corresponding to the medical record charge item A, the embodiment needs to calculate the similarity between the medical record charge item features of the medical record charge item A and the medical order sub-feature a-A, the similarity between the medical record charge item features of the medical record charge item A and the medical order sub-feature b-A, and the similarity between the medical record charge item features of the medical record charge item A and the medical order sub-feature c-A, and then determine the matching degree threshold corresponding to the medical record item A according to the above similarities.

[0074] Then, for the medical record charge item B, the medical order names a, b, and c matched with the medical record charge item B, the medical order sub-feature a-B of the medical order name a corresponding to the medical record charge item B, the medical order sub-feature b-B of the medical order name b corresponding to the medical record charge item B, and the medical order sub-feature c-B of the medical order name c corresponding to the medical record charge item B, the embodiment needs to calculate the similarity between the medical record charge item features of the medical record charge item B and the medical order sub-feature a-B, the similarity between the medical record charge item features of the medical record charge item B and the medical order sub-feature b-B, and the similarity between the medical record charge item features of the medical record charge item B and the medical order sub-feature c-B, and then determine the matching degree threshold corresponding to the medical record item B according to the above similarities.

[0075] In this way, the matching degree threshold corresponding to each medical record charge item in all sample medical records can be calculated.

[0076] Further, the step specifically includes:

[0077] First, the similarity between each medical order sub-feature matched with the medical record charge item feature and the medical record charge item feature is combined into a similarity set of the medical record charge item feature.

[0078] For any medical record charge item, the similarity between the medical order sub-feature of each medical order name matched with the medical record charge item and the medical record charge item feature has been calculated, and each similarity is combined into a similarity set of the medical record charge item feature. For example, for medical record charge item A, medical order names a, b, and c matched with medical record charge item A, medical order sub-feature a-A of medical order name a corresponding to medical record charge item A, medical order sub-feature b-A of medical order name b corresponding to medical record charge item A, medical order sub-feature c-A of medical order name c corresponding to medical record charge item A, the similarity set of the medical record charge item feature of medical record charge item A includes the similarity between the medical record charge item feature of medical record charge item A and medical order sub-feature a-A, the similarity between the medical record charge item feature of medical record charge item A and medical order sub-feature b-A, and the similarity between the medical record charge item feature of medical record charge item A and medical order sub-feature c-A.

[0079] In this way, the similarity set of the medical record charge item feature of each medical record charge item can be determined.

[0080] Second, the average value and the standard deviation of all similarities in the similarity set are used to calculate the similarity score of the medical record charge item feature.

[0081] This embodiment needs to calculate the average value and the standard deviation of all similarities in the similarity set of the medical record charge item feature, and then according to the three-sigma rule, the difference between the average value and three times the standard deviation is taken as the similarity score of the medical record charge item feature. Then, according to this way, the similarity score of each medical record charge item feature can be calculated according to the similarity set of each medical record charge item feature. The calculation formula of the similarity score is as follows:

[0082]

[0083]

[0084] thread mean = mean(g) - 3std(g)

[0085] Where g represents the set of similarities of the features of medical record charging items, g i Let represent the i-th similarity in the similarity set, n represent the number of similarities in the similarity set, mean(g) represent the average of all similarities in the similarity set g, and std(g) represent the standard deviation of all similarities in the similarity set g. mean Similarity score representing the characteristics of medical record billing items.

[0086] Third, the maximum value between the minimum similarity and the maximum similarity score in the similarity set of medical record billing item features is used as the matching threshold corresponding to the medical record billing item features.

[0087] In practical applications, the calculated similarity score may be negative. Therefore, an additional constraint is needed to improve the accuracy of the matching threshold. Specifically, the maximum value between the minimum similarity score and the similarity score in the similarity set of the medical record billing item features is used as the matching threshold for that feature. For different medical record billing items, the matching threshold for each item is calculated based on the similarity set and similarity score of its features. The specific formula is as follows:

[0088] thread = max(min(g), thread mean )

[0089] Here, thread represents the matching threshold corresponding to the medical record charging item.

[0090] As an optional implementation method, see [link to implementation details]. Figure 8 As shown in another embodiment of this application, the training process of the semantic feature extraction model is disclosed in the following steps:

[0091] S801. Input the sample fee items and sample medical order names from the pre-constructed sample training set into the semantic feature extraction model to obtain the sample fee item features corresponding to each sample fee item and the sample medical order features corresponding to each sample medical order name.

[0092] Specifically, this embodiment requires pre-constructing a sample training set and a semantic feature extraction model. The sample training set includes sample billing items and sample medical order names, and labels the matching relationships between each sample billing item and each sample medical order name. The structure of the semantic feature extraction model is as follows: Figure 3The embodiment can input the sample charge item and the sample medical order name into the semantic feature extraction model, and the semantic feature extraction model can extract the sample charge item feature of the sample charge item and the sample medical order feature corresponding to the sample medical order name. The sample medical order feature includes a sample medical order sub-feature corresponding to each sample charge item, and the weight of a target feature component in the sample medical order sub-feature is greater than the weight of a non-target feature component. The target feature component is a feature component related to the sample charge item corresponding to the sample medical order sub-feature. For example, the sample medical order sub-feature a-A of the sample medical order name a corresponding to the sample charge item A is a sample charge item coding feature based on the sample charge item A, and the weight of the feature component related to the sample charge item coding feature of the sample charge item A in the sample medical order coding feature of the sample medical order name a is increased to obtain.

[0093] S802, calculating the similarity between each sample medical order sub-feature of the sample medical order name and the sample charge item feature corresponding to the sample medical order sub-feature, and determining the matching degree between the sample medical order name and each sample charge item.

[0094] Specifically, after calculating each sample medical order sub-feature of the sample medical order name, the similarity between the sample medical order sub-feature and the sample charge item feature corresponding to the sample medical order sub-feature is calculated, and the similarity is taken as the matching degree between the sample medical order and the sample charge item, so as to determine the matching degree between the sample medical order name and each sample charge item.

[0095] For example, the sample medical order sub-feature a-A of the sample medical order name a corresponding to the sample charge item A, the sample medical order sub-feature a-B of the sample medical order name a corresponding to the sample charge item B, the sample medical order sub-feature a-C of the sample medical order name a corresponding to the sample charge item C, the matching degree between the sample medical order a and the sample charge item A is the similarity between the sample medical order sub-feature a-A and the sample charge item feature of the sample charge item A, the matching degree between the sample medical order a and the sample charge item B is the similarity between the sample medical order sub-feature a-B and the sample charge item feature of the sample charge item B, and the matching degree between the sample medical order a and the sample charge item C is the similarity between the sample medical order sub-feature a-C and the sample charge item feature of the sample charge item C.

[0096] S803, adjusting the parameters of the semantic feature extraction model by using the matching degree between the sample medical order name and each sample charge item and the matching relationship between the sample charge item and the sample medical order name marked in advance.

[0097] Specifically, according to the matching degree between the sample medical order name and the sample charging item and the pre-set matching degree threshold, the matching relationship between the sample medical order name and the sample charging item is determined, and the matching relationship between the sample medical order name and the sample charging item currently determined is compared with the pre-labeled matching relationship between the sample charging item and the sample medical order name. According to the comparison result, the parameters of the semantic feature extraction model are adjusted to make the matching relationship between the sample medical order name and the sample charging item currently determined gradually consistent with the pre-labeled matching relationship between the sample charging item and the sample medical order name, so as to realize the training of the semantic feature extraction model.

[0098] As an optional implementation, referring to FIG. 8, Figure 9 As shown in FIG. 8, another embodiment of the present application discloses that, when the semantic feature extraction model is trained, the construction process of the sample training set used is as follows:

[0099] S901, based on the pre-acquired original labeled sample medical record, the importance information of the keywords corresponding to each sample charging item in the original labeled sample medical record is determined.

[0100] In the constructed data set, as many cases as possible should be included, such as various aliases of medical orders, colloquial abbreviations or common summary items. Such data construction needs the assistance of professionals, but if it completely relies on manual annotation, it is too time-consuming and labor-intensive, affecting the construction efficiency of the sample training set, and it is difficult to obtain a satisfactory amount of data.

[0101] Therefore, the present embodiment can automatically annotate the matching relationship of the unlabeled sample medical record according to the sample medical record labeled with the matching relationship between the sample charging item and the sample medical order name, improve the construction efficiency of the sample training set, and increase the data amount of the sample training set.

[0102] Specifically, the doctor can label the matching relationship between the charging item and the medical order name in a small amount of actual medical records, and use these actual medical records as original labeled sample medical records, the charging item in the medical record as the sample charging item, and the medical order name as the sample medical order name. The present embodiment needs to extract the keywords corresponding to the sample charging item from the sample medical order name matched with the sample charging item in the original labeled sample medical record, and then determine the importance information of the keywords corresponding to the sample charging item, and determine the importance of each keyword corresponding to the sample charging item.

[0103] Among them, the extraction granularity of the keywords is different, and the extracted keywords are different. In the present embodiment, the keyword extraction granularity includes word granularity, word granularity and medical entity granularity, wherein the keyword extraction of the medical entity granularity can be extracted by using the medical NER (Named Entity Recognition, named entity recognition) tool.

[0104] If the keyword extraction granularity is character granularity, the extracted keywords corresponding to the sample charge item are characters, and the keyword importance information corresponding to the sample charge item includes key character importance information. If the keyword extraction granularity is word granularity, the extracted keywords corresponding to the sample charge item are words, and the keyword importance information corresponding to the sample charge item includes key word importance information. If the keyword extraction granularity is medical entity granularity, the extracted keywords corresponding to the sample charge item are medical entities, and the keyword importance information corresponding to the sample charge item includes key medical entity importance information. In this embodiment, the keywords can be extracted according to only one granularity, so as to calculate the keyword importance information in one granularity, or the keywords can be extracted according to multiple granularities, so as to calculate the keyword importance information in multiple granularities. For example, if the keywords are extracted according to character granularity and word granularity, the keyword importance information includes key character importance information and key word importance information.

[0105] Further, the step specifically includes:

[0106] First, the keywords corresponding to the sample charge item are extracted from the sample order name in the original annotated sample medical record and matching the sample charge item.

[0107] For a sample charge item in the original annotated sample medical record, the keywords corresponding to the sample charge item need to be extracted from the sample order name in the original annotated sample medical record and matching the sample charge item. If the keyword extraction granularity is character granularity, characters are extracted from the sample order name as the keywords corresponding to the sample charge item. If the keyword extraction granularity is word granularity, words are extracted from the sample order name as the keywords corresponding to the sample charge item. If the keyword extraction granularity is medical entity granularity, medical entities are extracted from the sample order name as the keywords corresponding to the sample charge item.

[0108] Second, the importance of each keyword corresponding to the sample charge item is calculated by using the number of occurrences of each keyword, the number of sample charge items corresponding to each keyword, and the total number of sample charge items in the original annotated sample medical record.

[0109] After extracting the keywords corresponding to the sample charge item, the importance of each keyword needs to be calculated. For a keyword, the importance of the keyword is calculated according to the number of occurrences of the keyword (i.e., the number of times the keyword appears in all sample medical order names matched with the sample charge item), the number of sample charge items corresponding to the keyword (i.e., the number of sample charge items in the original labeled sample medical record containing the keyword), and the total number of sample charge items in the original labeled sample medical record. Thus, the importance of each keyword corresponding to the sample charge item can be calculated, and the importance of all keywords corresponding to the sample charge item is collected as the keyword importance information corresponding to the sample charge item.

[0110] In this embodiment, the importance of the keyword can be evaluated by using the statistical method of TF-IDF (term frequency-inverse document frequency). The calculation formula of the importance of the keyword is as follows:

[0111]

[0112] wherein n ij represents the number of occurrences of the keyword (i.e., the number of times the keyword appears in all sample medical order names matched with the sample charge item), ∑ k n kj represents the total number of occurrences of all keywords corresponding to the sample charge item, D represents the total number of sample charge items in the original labeled sample medical record, and {j: t i ∈d j} represents the number of sample charge items corresponding to the keyword (i.e., the number of sample charge items in the original labeled sample medical record containing the keyword).

[0113] In addition, if the keywords corresponding to the sample charge item are extracted by using the three keyword extraction granularities of word granularity, term granularity and medical entity granularity, the importance of each keyword character corresponding to the sample charge item needs to be calculated, the importance of all keyword characters corresponding to the sample charge item is collected as the keyword character importance information corresponding to the sample charge item; the importance of each keyword vocabulary corresponding to the sample charge item needs to be calculated, the importance of all keyword vocabularies corresponding to the sample charge item is collected as the keyword vocabulary importance information corresponding to the sample charge item; and the importance of each medical entity corresponding to the sample charge item needs to be calculated, the importance of all medical entities corresponding to the sample charge item is collected as the medical entity importance information corresponding to the sample charge item.

[0114] S902, use the keyword importance degree information corresponding to the sample charge item to calculate the sample matching degree between the sample charge item in the pre-collected unlabeled sample medical record and the sample medical order name.

[0115] Specifically, after determining the keyword importance degree information corresponding to each sample charge item in the original labeled sample medical record in the above manner, for example, for the calculation of the sample matching degree between the sample charge item A in the unlabeled sample medical record and the sample medical order name a, first, find the keyword importance degree information corresponding to the sample charge item A from the keyword importance degree information corresponding to each sample charge item in the original labeled sample medical record, then find the importance degree corresponding to the keyword from the keyword importance degree information corresponding to the sample charge item A according to the sample charge item A corresponding to the keyword contained in the sample medical order name a, and then calculate the sample matching degree between the sample charge item A in the unlabeled sample medical record and the sample medical order name a. The sample matching degree between any sample charge item and any sample medical order name in the unlabeled sample medical record is calculated in the above manner.

[0116] Further, this step specifically includes:

[0117] First, according to the keyword extraction granularity corresponding to the sample charge item in the original labeled sample medical record, the sample medical order name in the pre-collected unlabeled sample medical record is segmented, and the segmentation result corresponding to the sample medical order name is obtained.

[0118] This embodiment needs to segment the sample medical order name in the pre-collected unlabeled sample medical record according to the keyword extraction granularity corresponding to the sample charge item in the original labeled sample medical record. If the keyword extraction granularity corresponding to the sample charge item in the original labeled sample medical record includes three kinds of word granularity, word granularity and medical entity granularity, then the sample medical order name in the unlabeled sample medical record also needs to be segmented with these three granularities to obtain the word granularity corresponding to the segmentation, the word granularity corresponding to the segmentation and the medical entity granularity corresponding to the segmentation. Among them, the segmentation extraction of medical entity granularity can use medical NER (Named Entity Recognition, named entity recognition) tool extraction.

[0119] Second, from the keyword importance degree information corresponding to the sample charge item, the importance degree of each segmentation corresponding to the sample medical order name in the unlabeled sample medical record is queried, and the average importance degree of each segmentation is calculated, which is used as the sample matching degree of the sample charge item and the sample medical order name in the unlabeled sample medical record under the keyword extraction granularity.

[0120] The embodiment can query the importance of each segmented word corresponding to the sample medical order name a in the unannotated sample medical record from the keyword importance information corresponding to the sample charging item A. For the segmented word corresponding to the character granularity, the importance of the segmented word corresponding to the character granularity needs to be queried from the keyword importance information corresponding to the sample charging item A. For the segmented word corresponding to the word granularity, the importance of the segmented word corresponding to the word granularity needs to be queried from the keyword importance information corresponding to the sample charging item A. For the segmented word corresponding to the medical entity granularity, the importance of the segmented word corresponding to the medical entity granularity needs to be queried from the keyword importance information corresponding to the sample charging item A. For the segmented word corresponding to the character granularity, the average value of the importance of all segmented words corresponding to the character granularity is taken as the sample matching degree of the sample charging item A and the sample medical order name a in the unannotated sample medical record at the character granularity. For the segmented word corresponding to the word granularity, the average value of the importance of all segmented words corresponding to the word granularity is taken as the sample matching degree of the sample charging item A and the sample medical order name a in the unannotated sample medical record at the word granularity. For the segmented word corresponding to the medical entity granularity, the average value of the importance of all segmented words corresponding to the medical entity granularity is taken as the sample matching degree of the sample charging item A and the sample medical order name a in the unannotated sample medical record at the medical entity granularity.

[0121] In the above manner, the sample matching degree of any sample charging item and any sample medical order name in the unannotated sample medical record at the keyword extraction granularity can be calculated.

[0122] Thirdly, according to the sample matching degree of the sample charging item and the sample medical order name in the unannotated sample medical record at the keyword extraction granularity, the sample matching degree between the sample charging item and the sample medical order name in the unannotated sample medical record is determined.

[0123] If the keyword importance information corresponding to each sample charging item in the original annotated sample medical record only contains keyword importance information corresponding to one granularity, the sample matching degree of the sample charging item and the sample medical order name in the unannotated sample medical record at the keyword extraction granularity is taken as the sample matching degree between the sample charging item and the sample medical order name in the unannotated sample medical record.

[0124] If the keyword importance information corresponding to each sample charge item in the original labeled sample medical record contains keyword importance information of multiple granularities, the embodiment needs to take the average value between the sample charge item in the unlabeled sample medical record and the sample medical order name in the sample matching degree of multiple keyword extraction granularities as the sample matching degree between the sample charge item and the sample medical order name in the unlabeled sample medical record. For example, if the keyword importance information corresponding to each sample charge item in the original labeled sample medical record contains keyword importance information, keyword importance information, and keyword medical entity importance information, then the embodiment needs to take the average value between the sample charge item in the unlabeled sample medical record and the sample medical order name in the sample matching degree of the word granularity, the sample matching degree of the word granularity, and the sample matching degree of the medical entity granularity as the final sample matching degree.

[0125] S903, match and label the sample charge item and the sample medical order name in the unlabeled sample medical record whose sample matching degree reaches the preset threshold to obtain a labeled sample medical record corresponding to the unlabeled sample medical record.

[0126] Through the above steps, after calculating the sample matching degree between each sample charge item and each sample medical order name in the unlabeled sample medical record, the sample charge item and the sample medical order name whose sample matching degree reaches the preset threshold are matched and labeled, and after analyzing the sample matching degree of all sample charge items and sample medical order names in the unlabeled sample medical record, the final labeled sample medical record is obtained as the labeled sample medical record corresponding to the unlabeled sample medical record.

[0127] S904, take the set of the original labeled sample medical record and the labeled sample medical record corresponding to the unlabeled sample medical record as a sample training set.

[0128] According to the original labeled sample medical record, the embodiment matches and labels the unlabeled sample medical record, and then takes the set of the labeled sample medical record corresponding to the unlabeled sample medical record and the original labeled sample medical record as the sample training set of the semantic feature extraction model. According to the sample medical record labeled with the matching relationship between the sample charge item and the sample medical order name, the embodiment automatically labels the matching relationship of the unlabeled sample medical record, improves the construction efficiency of the sample training set, increases the data amount of the sample training set, and ensures the quality of the training sample.

[0129] As an optional implementation, another embodiment of the application discloses that when the semantic feature extraction model is trained, the sample training set used includes: a first sample training set and a second sample training set. The first sample training set includes: a positive sample set and a random negative sample set; and the second sample training set includes: the positive sample set and a similar negative sample set.

[0130] The positive sample set includes sample medical records pre-labeled with correct matching relationships; the random negative sample set and the similar negative sample set include sample medical records pre-labeled with incorrect matching relationships.

[0131] The random negative sample set can be sample charge items and sample medical order names after random error matching of sample medical records.

[0132] For the similar negative sample, in the sample medical records in the similar negative sample set, the sample charge item having an incorrect matching relationship with the sample medical order name is a charge item similar in semantics to the charge item correctly matched with the sample medical order name. For example, each standard charge name has a medical insurance three-category code, and the arrangement of these codes has certain regularity, which is specifically manifested in that items with similar meanings are generally adjacent, such as “chest color Doppler ultrasound examination”, “abdominal color Doppler ultrasound examination”, and “urinary system color Doppler ultrasound examination”. Therefore, for the sample charge item, a medical order name that can be matched with an adjacent item but cannot be matched with itself can be selected as a negative sample, so that similar negative samples to the positive sample can be found.

[0133] When training the semantic feature extraction model, the semantic feature extraction model is first trained using the first sample training set to adapt the semantic feature extraction model to a general difficulty matching task. After the model converges, the semantic feature extraction model is trained using the second sample training set, which will impose higher requirements on the semantic feature extraction model, so that more refined classification standards can be learned, thereby improving the accuracy of the semantic feature extraction in feature extraction.

[0134] As an optional implementation, another embodiment of the present application discloses that the medical insurance charge information intelligent verification method further comprises:

[0135] determining a charge item in the medical record to be audited that does not match a medical order name as a charge item without a medical order; and / or performing consistency verification on a charge item in the medical record to be audited and a medical order name corresponding to the charge item, to obtain a medical order charge item.

[0136] Specifically, if there is a charge item in the medical record to be audited that does not match a medical order name, the charge item that does not match a medical order name is determined as a charge item without a medical order, and corresponding charge information without a medical order can also be output. The prompt information can include the charge item name or three-category code corresponding to the charge item without a medical order, or the cost identifier corresponding to the charge item without a medical order, i.e., the id of all costs belonging to the charge item without a medical order, so that the user can query.

[0137] For the matched charge item and the medical order name in the to-be-audited medical record, the consistency check can also be performed on the charge item corresponding cost details and the medical order name corresponding to the medical order details matched with the charge item to obtain the medical order-free cost details. The cost details include state information of payment of the charge item corresponding cost, such as payment time; the medical order details include state information of the medical order corresponding to the medical order name, such as the time of issuing the medical order. If there is a certain cost detail in the charge item, and there is no medical order detail corresponding to the cost detail in all medical order details corresponding to the charge item, then the cost detail is a medical order-free cost detail, and the corresponding medical order-free charge prompt can also be output. The prompt information can include the cost identifier corresponding to the medical order-free cost detail, so that the user can query.

[0138] For example, the charge item A includes the T1 time cost detail A1, the T2 time cost detail A2 and the T3 time cost detail A3, but the medical order name a matched with the charge item A only includes the T1 time medical order detail a1 and the T3 time medical order detail a3, and the T2 time does not issue the medical order matched with the charge item A, then the T2 time cost detail A2 is a medical order-free cost detail.

[0139] Corresponding to the above-mentioned medical insurance charge information intelligent checking method, the embodiment of the application also proposes a medical insurance charge information intelligent checking device, as shown in Figure 10 The device comprises:

[0140] The feature extraction module 100 is configured to extract the medical order features of the medical order name in the to-be-audited medical record according to each charge item in the to-be-audited medical record, and extract the charge item features of each charge item in the to-be-audited medical record respectively. The medical order features include medical order sub-features corresponding to each charge item. The weight of the target feature component in the medical order sub-feature is greater than the weight of the non-target feature component. The target feature component is a feature component related to the charge item corresponding to the medical order sub-feature.

[0141] The matching module 110 is configured to calculate the similarity between each medical order sub-feature of the medical order name and the charge item feature of the corresponding charge item, and determine the matching degree between the medical order name and each charge item.

[0142] The medical insurance charging information intelligent verification device provided in the embodiments of the present application, the feature extraction module 100 extracts the order feature of the order name in the to-be-audited medical record and the charging item feature of each charging item in the to-be-audited medical record according to each charging item in the to-be-audited medical record, the order feature includes the order sub-feature corresponding to each charging item, and the weight of the target feature component related to the charging item corresponding to the order sub-feature in the order sub-feature is greater than the weight of the non-target feature component. The matching module 110 calculates the similarity between each order sub-feature of the order name and the charging item feature of the corresponding charging item, and determines the matching degree between the order name and each charging item. When the order name and a certain charging item are matched and analyzed by using the technical solution of the embodiments, the order feature is extracted by using the order name and the charging item, so that the weight of the feature component related to the charging item that needs to be matched and analyzed in the order feature is improved, the order feature extracted is more conducive to measuring the similarity between the order name and the charging item, and therefore the accuracy of measuring the similarity between the order name and the charging item is improved, that is, the accuracy of information matching can be improved.

[0143] As an optional implementation manner, the other embodiments of the present application also disclose that the feature extraction module 100 is specifically used for:

[0144] obtaining the charging item coding feature of the charging item and obtaining the order coding feature of the order name;

[0145] improving the weight of the feature component related to the charging item coding feature in the order coding feature based on the charging item coding feature;

[0146] generating the order sub-feature corresponding to the charging item according to the order coding feature with the improved weight.

[0147] As an optional implementation manner, the other embodiments of the present application also disclose that the feature extraction module 100 is specifically used for:

[0148] inputting the charging item and the order name into the pre-trained semantic feature extraction model, so that the semantic feature extraction model extracts the charging item coding feature of the charging item and the order coding feature of the order name, improves the weight of the feature component related to the charging item coding feature in the order coding feature based on the charging item coding feature, and generates the order sub-feature corresponding to the charging item according to the order coding feature with the improved weight.

[0149] As an optional implementation manner, the other embodiments of the present application also disclose that the medical insurance charging information intelligent verification device in the embodiments further comprises:

[0150] The screening module is configured to screen, from each charge item of the medical record to be audited, a charge item having a matching degree greater than a preset matching degree threshold with respect to the medical order name, as a charge item matching the medical order name.

[0151] As an optional implementation, the medical insurance charge information intelligent verification device according to another embodiment of the present application further comprises a threshold determination module.

[0152] The threshold determination module is configured to determine, according to a matching degree threshold corresponding to a medical record charge item, a matching degree threshold corresponding to each charge item of the medical record to be audited.

[0153] The screening module is specifically configured to screen, from each charge item of the medical record to be audited, a charge item having a matching degree greater than a corresponding matching degree threshold with respect to the medical order name, as a charge item matching the medical order name, according to the matching degree between the medical order name and each charge item and the matching degree threshold corresponding to each charge item.

[0154] As an optional implementation, the medical insurance charge information intelligent verification device according to another embodiment of the present application further comprises a sample extraction module, a sample feature determination module and a threshold calculation module.

[0155] The sample extraction module is configured to extract, from all sample medical records in a sample training set, medical record charge items and all medical order names matching the medical record charge items; the sample training set comprises at least one sample medical record, and the sample medical record is pre-labeled with a matching relationship between each medical record charge item and each medical order name.

[0156] The sample feature determination module is configured to extract, according to a medical record charge item, medical order sub-features of all medical order names matching the medical record charge item, and extract a medical record charge item feature of the medical record charge item.

[0157] The threshold calculation module is configured to determine a matching degree threshold corresponding to a medical record charge item according to a similarity between the medical record charge item feature and each medical order sub-feature matching the medical record charge item.

[0158] As an optional implementation, the threshold calculation module according to another embodiment of the present application is specifically configured to:

[0159] combine the similarity between each medical order sub-feature matching the medical record charge item feature and the medical record charge item feature into a similarity set of the medical record charge item feature;

[0160] calculate a similarity score of the medical record charge item feature by using an average value and a standard deviation of all similarities in the similarity set;

[0161] The minimum similarity in the similarity set of the medical record charging item characteristics and the maximum value between the similarity scores are taken as the matching degree threshold corresponding to the medical record charging item characteristics.

[0162] As an optional implementation, the medical insurance charging information intelligent verification device of the embodiment further includes an input module, a similarity calculation module, and a parameter adjustment module.

[0163] The input module is configured to input sample charging items and sample medical order names in a pre-constructed sample training set into the semantic feature extraction model to obtain sample charging item characteristics corresponding to each sample charging item and sample medical order characteristics corresponding to each sample medical order name. The sample medical order characteristics include sample medical order sub-features corresponding to each sample charging item.

[0164] The similarity calculation module is configured to calculate the similarity between each sample medical order sub-feature of the sample medical order name and the sample charging item characteristics corresponding to each sample charging item, and determine the matching degree between the sample medical order name and each sample charging item.

[0165] The parameter adjustment module is configured to adjust the parameters of the semantic feature extraction model by using the matching degree between the sample medical order name and each sample charging item and the matching relationship between the pre-labeled sample charging items and sample medical order names.

[0166] As an optional implementation, the medical insurance charging information intelligent verification device of the embodiment further includes a keyword importance determination module, a sample matching degree calculation module, a matching labeling module, and a set module.

[0167] The keyword importance determination module is configured to determine keyword importance information corresponding to each sample charging item in a pre-acquired original labeled sample medical record based on the original labeled sample medical record.

[0168] The sample matching degree calculation module is configured to calculate the sample matching degree between sample charging items and sample medical order names in a pre-acquired unlabeled sample medical record by using the keyword importance information corresponding to the sample charging items.

[0169] The matching labeling module is configured to match and label sample charging items and sample medical order names in the unlabeled sample medical record, which have a sample matching degree reaching a preset threshold, to obtain a labeled sample medical record corresponding to the unlabeled sample medical record.

[0170] The set module is configured to take the set of the original labeled sample medical record and the labeled sample medical record corresponding to the unlabeled sample medical record as a sample training set.

[0171] The original labeled sample medical record has a matching relationship between a sample charge item and a sample medical order name.

[0172] As an optional implementation, another embodiment of the present application further discloses a keyword importance degree determination module, which is specifically used for:

[0173] extracting a keyword corresponding to the sample charge item from a sample medical order name matched with the sample charge item in the original labeled sample medical record;

[0174] calculating an importance degree of each keyword corresponding to the sample charge item by using the number of occurrences of each keyword, the number of sample charge items corresponding to each keyword, and the total number of sample charge items in the original labeled sample medical record;

[0175] The number of occurrences of the keyword is the number of occurrences of the keyword in all sample medical order names matched with the sample charge item, and the number of sample charge items corresponding to the keyword is the number of sample charge items containing the keyword in the original labeled sample medical record.

[0176] As an optional implementation, another embodiment of the present application further discloses a sample matching degree calculation module, which is specifically used for:

[0177] performing word segmentation on sample medical order names in the pre-collected unlabeled sample medical record according to the keyword extraction granularity of the keyword corresponding to the sample charge item in the original labeled sample medical record to obtain a word segmentation result corresponding to the sample medical order name;

[0178] querying the importance degrees of each word segmentation corresponding to the sample medical order name in the unlabeled sample medical record from the keyword importance degree information corresponding to the sample charge item, and calculating an average value of the importance degrees of each word segmentation, taking the average value as a sample matching degree between the sample charge item and the sample medical order name in the unlabeled sample medical record at the keyword extraction granularity;

[0179] determining a sample matching degree between the sample charge item and the sample medical order name in the unlabeled sample medical record according to the sample matching degree between the sample charge item and the sample medical order name in the unlabeled sample medical record at the keyword extraction granularity.

[0180] As an optional implementation, another embodiment of the present application further discloses that the sample training set comprises a first sample training set and a second sample training set.

[0181] The first sample training set comprises a positive sample set and a random negative sample set, and the second sample training set comprises a positive sample set and a similar negative sample set.

[0182] When the semantic feature extraction model is trained, the semantic feature extraction model is trained by using the first sample training set first, and then the semantic feature extraction model is trained by using the second sample training set.

[0183] The positive sample set includes sample medical records pre-labeled with correct matching relationships; the random negative sample set and the similar negative sample set include sample medical records pre-labeled with incorrect matching relationships; in the sample medical records in the similar negative sample set, the sample charge items having incorrect matching relationships with the sample medical order names are charge items having similar semantics to the charge items correctly matched with the sample medical order names.

[0184] As an optional implementation, another embodiment of the present application further discloses that the medical insurance charge information intelligent verification device of the embodiment further includes a determination module.

[0185] The determination module is configured to determine the charge items in the medical record to be audited that do not match the medical order names as charge items without medical orders.

[0186] And / or,

[0187] The determination module is configured to determine the charge items in the medical record to be audited that do not match the medical order names as charge items without medical orders.

[0188] The medical insurance charge information intelligent verification device provided by the embodiment belongs to the same application concept as the medical insurance charge information intelligent verification method provided by the above-mentioned embodiments of the present application, can execute the medical insurance charge information intelligent verification method provided by any of the above-mentioned embodiments of the present application, and has the corresponding functional modules and beneficial effects of executing the medical insurance charge information intelligent verification method. Technical details not described in detail in the embodiment can be referred to the specific processing content of the medical insurance charge information intelligent verification method provided by the above-mentioned embodiments of the present application, which will not be described here.

[0189] Another embodiment of the present application further discloses an electronic device, as shown in Figure 11 The device includes:

[0190] The memory 200 and the processor 210;

[0191] The memory 200 is connected with the processor 210, and is configured to store programs.

[0192] The processor 210 is configured to realize the medical insurance charge information intelligent verification method disclosed in any of the above-mentioned embodiments by running the programs stored in the memory 200.

[0193] In particular, the electronic device described above can further include a bus, a communication interface 220, an input device 230, and an output device 240.

[0194] The processor 210, the memory 200, the communication interface 220, the input device 230, and the output device 240 are connected to each other through the bus. Among them:

[0195] The bus can include a path for transmitting information between the various components of the computer system.

[0196] The processor 210 can be a general-purpose processor, such as a general-purpose central processing unit (CPU), a microprocessor, etc., or an application-specific integrated circuit (ASIC), or one or more integrated circuits for controlling the execution of programs of the present application. It can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a ready-to-use programmable gate array (FPGA), or other programmable logic device, discrete gate or transistor logic device, discrete hardware component.

[0197] The processor 210 can include a main processor and can further include a baseband chip, a modem, etc.

[0198] The memory 200 stores programs for executing the technical solutions of the present application, and can also store operating systems and other key services. Specifically, the program can include program code, and the program code includes computer operation instructions. More specifically, the memory 200 can include read-only memory (ROM), other types of static storage devices that can store static information and instructions, random access memory (RAM), other types of dynamic storage devices that can store information and instructions, disk storage, flash, etc.

[0199] The input device 230 can include a device that receives data and information input by a user, such as a keyboard, a mouse, a camera, a scanner, a light pen, a voice input device, a touch screen, a pedometer, or a gravity sensor, etc.

[0200] The output device 240 can include a device that allows information to be output to a user, such as a display screen, a printer, a speaker, etc.

[0201] The communication interface 220 can include a device using any transceiver to communicate with other devices or communication networks, such as Ethernet, a radio access network (RAN), a wireless local area network (WLAN), etc.

[0202] The processor 210 executes programs stored in the memory 200 and invokes other devices to implement each step of the medical insurance charging information intelligent verification method provided by the embodiments of the present application.

[0203] Another embodiment of the present application further provides a storage medium, which stores a computer program. When the computer program is executed by a processor, each step of the medical insurance charging information intelligent verification method provided by any of the above embodiments is implemented.

[0204] For each method embodiment described above, in order to simply describe, each method embodiment is described as a combination of a series of actions, but those skilled in the art should know that the present application is not limited to the order of actions described, because according to the present application, some steps can be performed in other order or simultaneously. Secondly, those skilled in the art should know that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present application.

[0205] It should be noted that each embodiment in the specification is described in a progressive manner, and each embodiment mainly describes the difference from other embodiments, and the same and similar parts between each embodiment can be known by referring to each other. For the device embodiment, since it is basically similar to the method embodiment, it is described relatively simply, and the relevant parts can be known by referring to the part of the method embodiment.

[0206] The steps in the method of each embodiment of the present application can be adjusted, combined and reduced in sequence according to actual needs.

[0207] The modules and sub-modules in the device and terminal of each embodiment of the present application can be combined, divided and reduced according to actual needs.

[0208] In several embodiments provided by the present application, it should be understood that the disclosed terminal, device and method can be implemented by other ways. For example, the terminal embodiments described above are only schematic, for example, the division of modules or sub-modules is only a logical function division, and there can be another division way in actual implementation, for example, a plurality of sub-modules or modules can be combined or integrated into another module, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the shown or discussed mutual elements can be indirect coupling or communication connection through some interface, device or module, and can be electrical, mechanical or other forms.

[0209] The modules or sub-modules described as separate components may or may not be physically separate, and the components of the modules or sub-modules may or may not be physical modules or sub-modules, i.e., may be located in one place or distributed over multiple network modules or sub-modules. Some or all of the modules or sub-modules can be selected as needed to achieve the purposes of the embodiments.

[0210] In addition, each functional module or sub-module in the various embodiments of the present application can be integrated into one processing module, or each module or sub-module can exist physically alone, or two or more modules or sub-modules can be integrated into one module. The integrated module or sub-module can be realized in the form of hardware or in the form of a software functional module or sub-module.

[0211] The skilled person will further appreciate that the units and algorithm steps of the examples described in connection with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. In order to clearly illustrate the interchangeability of hardware and software, the elements and steps of the examples have been described in general terms above as being primarily associated with various functional means. Whether such functions are implemented in hardware or software depends on the particular application and design constraints imposed on the overall system. Skilled persons can implement the described functions in varying ways for each particular application, but such implementation does not depart from the scope of the present application.

[0212] The steps of the methods or algorithms described in connection with the embodiments disclosed herein can be directly implemented in hardware, executed by a software unit of a processor, or a combination of both. The software unit can be stored in a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.

[0213] Finally, it should be noted that the terms such as first and second, etc., are merely used to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "comprise", "include" or any other variant thereof are intended to cover non-exclusive inclusion, so that processes, methods, articles or devices including a series of elements not only include those elements, but also include other elements not explicitly listed or inherent to such processes, methods, articles or devices. Without more limitations, the element defined by the phrase "comprising a" does not exclude the presence of additional identical elements in the process, method, article or device including the element.

[0214] The foregoing description of the disclosed embodiments enables a person skilled in the art to make or use the application. Modifications of these embodiments will occur to persons of skill in the art, and, while certain embodiments according to the principles set forth herein are shown and described, it is to be understood that the same are not limiting of the scope of the application as it is set forth in the appended claims, and that various modifications are made within the scope of the appended claims. Therefore, it is contemplated to cover the application in its broadest scope, including all features that can be made or used in the embodiments described herein.

Claims

1. A medical insurance charging information intelligent verification method, characterized in that, The method comprises the following steps: According to each charge item in the to-be-audited medical record, the medical order characteristics of the medical order name in the to-be-audited medical record are extracted, and the charge item characteristics of each charge item in the to-be-audited medical record are respectively extracted, the medical order characteristics include medical order sub-features corresponding to each charge item, the weight of a target feature component in the medical order sub-feature is greater than the weight of a non-target feature component, and the target feature component is a feature component related to the charge item corresponding to the medical order sub-feature; Based on the weight of the feature component related to the corresponding charge item in each medical order sub-feature of the medical order name, the similarity between each medical order sub-feature of the medical order name and the charge item characteristics of the corresponding charge item is calculated, and the matching degree between the medical order name and each charge item is determined; According to the charge item in the to-be-audited medical record, the medical order characteristics of the medical order name in the to-be-audited medical record are extracted, which comprises: Obtaining the charge item coding characteristics of the charge item, and obtaining the medical order coding characteristics of the medical order name; Based on the charge item coding characteristics, the weight of the feature component related to the charge item coding characteristics in the medical order coding characteristics is increased; According to the weight-increased medical order coding characteristics, the medical order sub-feature corresponding to the charge item is generated.

2. The method of claim 1, wherein, Obtaining the charge item coding characteristics of the charge item, and obtaining the medical order coding characteristics of the medical order name, based on the charge item coding characteristics, the weight of the feature component related to the charge item coding characteristics in the medical order coding characteristics is increased, and according to the weight-increased medical order coding characteristics, the medical order sub-feature corresponding to the charge item is generated, which comprises: The charge item and the medical order name are input into the pre-trained semantic feature extraction model, so that the semantic feature extraction model obtains the charge item coding characteristics of the charge item, and obtains the medical order coding characteristics of the medical order name, based on the charge item coding characteristics, the weight of the feature component related to the charge item coding characteristics in the medical order coding characteristics is increased, and according to the weight-increased medical order coding characteristics, the medical order sub-feature corresponding to the charge item is generated.

3. The method according to any one of claims 1 to 2, characterized in that, The method further comprises: According to the matching degree between the medical order name and each charge item, the charge items with the matching degree greater than a preset matching degree threshold value are screened from the charge items in the to-be-audited medical record as the charge items matched with the medical order name.

4. The method of claim 3, wherein, The method further comprises: According to the pre-determined matching degree threshold value corresponding to the medical record charge item, the matching degree threshold value corresponding to each charge item of the to-be-audited medical record is determined; According to the matching degree between the medical order name and each charge item, the charge items with the matching degree greater than a preset matching degree threshold value are screened from the charge items in the to-be-audited medical record as the charge items matched with the medical order name, which comprises: According to the matching degree between the medical order name and each charge item, and the matching degree threshold value corresponding to each charge item, the charge items with the matching degree greater than the corresponding matching degree threshold value are screened from the charge items in the to-be-audited medical record as the charge items matched with the medical order name.

5. The method of claim 4, wherein, The matching degree threshold corresponding to the medical record charging item is determined by the following processing: From all sample medical records in a sample training set, extract medical record charging items and all medical order names matched with the medical record charging items; the sample training set contains at least one sample medical record, and the sample medical record is pre-labeled with the matching relationship between each medical record charging item and each medical order name; According to the medical record charging item, extract medical order sub-features of all medical order names matched with the medical record charging item, and extract medical record charging item features of the medical record charging item; According to the similarity between the medical record charging item features and each matched medical order sub-feature, determine the matching degree threshold corresponding to the medical record charging item.

6. The method of claim 5, wherein, According to the similarity between the medical record charging item features and each matched medical order sub-feature, determine the matching degree threshold corresponding to the medical record charging item, comprising: Combine the similarity between each medical order sub-feature matched with the medical record charging item features and the medical record charging item features into a similarity set of the medical record charging item features; Calculate the average value and standard deviation of all similarities in the similarity set to obtain a similarity score of the medical record charging item features; The maximum value between the minimum similarity in the similarity set of the medical record charging item features and the similarity score is taken as the matching degree threshold corresponding to the medical record charging item features.

7. The method of claim 2, wherein, The training process of the semantic feature extraction model comprises: Input the pre-constructed sample charging items and sample medical order names in the sample training set into the semantic feature extraction model to obtain sample charging item features corresponding to each sample charging item and sample medical order features corresponding to each sample medical order name; the sample medical order features include sample medical order sub-features corresponding to each sample charging item; Calculate the similarity between each sample medical order sub-feature of the sample medical order name and the sample charging item feature of the corresponding sample charging item to determine the matching degree between the sample medical order name and each sample charging item; Adjust the parameters of the semantic feature extraction model by using the matching degree between the sample medical order name and each sample charging item and the pre-labeled matching relationship between the sample charging item and the sample medical order name.

8. The method of claim 7, wherein, The construction process of the sample training set comprises: Based on the pre-collected original labeled sample medical records, determine the keyword importance information corresponding to each sample charging item in the original labeled sample medical records; Using the keyword importance information corresponding to the sample charging item, calculate the sample matching degree between the sample charging item and the sample medical order name in the pre-collected unlabeled sample medical records; Match the sample charging item and the sample medical order name in the unlabeled sample medical record with the sample matching degree reaching the preset threshold to obtain the labeled sample medical record corresponding to the unlabeled sample medical record; The set of original labeled sample medical records and labeled sample medical records corresponding to the unlabeled sample medical records is taken as the sample training set; The original annotated sample medical record has a matching relationship between a sample charge item and a sample medical order name.

9. The method of claim 8, wherein, The keyword importance information corresponding to the sample charge item in the original annotated sample medical record is determined based on the pre-collected original annotated sample medical record, including: The keyword corresponding to the sample charge item is extracted from the sample medical order name in the original annotated sample medical record which matches the sample charge item. The importance of each keyword corresponding to the sample charge item is calculated by using the number of occurrences of each keyword, the number of sample charge items corresponding to each keyword, and the total number of sample charge items in the original annotated sample medical record. The number of occurrences of the keyword is the number of occurrences of the keyword in all sample medical order names matching the sample charge item; the number of sample charge items corresponding to the keyword is the number of sample charge items containing the keyword in the original annotated sample medical record.

10. The method of claim 8, wherein, The sample matching degree between the sample charge item and the sample medical order name in the pre-collected unannotated sample medical record is calculated by using the keyword importance information corresponding to the sample charge item, including: The sample medical order name in the pre-collected unannotated sample medical record is segmented according to the keyword extraction granularity corresponding to the sample charge item in the original annotated sample medical record to obtain the segmentation result corresponding to the sample medical order name. The importance of each segmentation of the sample medical order name in the unannotated sample medical record is queried from the keyword importance information corresponding to the sample charge item, and the average importance of each segmentation is calculated, which is used as the sample matching degree of the sample charge item and the sample medical order name in the unannotated sample medical record at the keyword extraction granularity. The sample matching degree between the sample charge item and the sample medical order name in the unannotated sample medical record is determined according to the sample matching degree of the sample charge item and the sample medical order name in the unannotated sample medical record at the keyword extraction granularity.

11. The method of claim 7, wherein, The sample training set includes a first sample training set and a second sample training set. The first sample training set includes a positive sample set and a random negative sample set; the second sample training set includes the positive sample set and a similar negative sample set. When training the semantic feature extraction model, the first sample training set is used to train the semantic feature extraction model, and then the second sample training set is used to train the semantic feature extraction model. The positive sample set includes sample medical records pre-annotated with correct matching relationships; the random negative sample set and the similar negative sample set include sample medical records pre-annotated with incorrect matching relationships; in the similar negative sample set, the sample charge item with incorrect matching relationship with the sample medical order name is a charge item with similar semantics to the charge item correctly matched with the sample medical order name.

12. The method of claim 3, wherein, Also includes: determine a charge item without a medical order in the medical record to be audited as a charge item without a medical order, if the charge item does not match the medical order name; and / or, perform consistency verification on a fee detail corresponding to the charge item in the medical record to be audited and a medical order detail corresponding to the medical order name matched with the charge item, to obtain a medical order fee detail without a medical order; wherein the fee detail includes state information of payment of the fee corresponding to the charge item, and the medical order detail includes state information of issuance of the medical order corresponding to the medical order name.

13. A medical insurance charging information intelligent checking device, characterized in that, comprise: a feature extraction module configured to extract a medical order feature of a medical order name in a medical record to be audited according to each charge item in the medical record to be audited, and extract a charge item feature of each charge item in the medical record to be audited respectively, the medical order feature including a medical order sub-feature corresponding to each charge item, a weight of a target feature component in the medical order sub-feature being greater than a weight of a non-target feature component, the target feature component being a feature component related to the charge item corresponding to the medical order sub-feature; a matching module configured to calculate a similarity between each medical order sub-feature of the medical order name and the charge item feature of the corresponding charge item based on the weight of the feature component related to the corresponding charge item in each medical order sub-feature of the medical order name, and determine a matching degree between the medical order name and each charge item; the feature extraction module is specifically configured to obtain a charge item coding feature of the charge item, and obtain a medical order coding feature of the medical order name; increase the weight of a feature component related to the charge item coding feature in the medical order coding feature based on the charge item coding feature; and generate a medical order sub-feature corresponding to the charge item according to the medical order coding feature with the increased weight.

14. An electronic device, comprising: comprise: a memory and a processor; wherein the memory is connected with the processor, and is configured to store a program; the processor is configured to realize the medical insurance charge information intelligent verification method in any one of claims 1 to 12 by running the program in the memory.

15. A storage medium, characterized by The storage medium has a computer program stored thereon, and the computer program is executed by the processor to realize the medical insurance charge information intelligent verification method in any one of claims 1 to 12.

Images