Fee return processing method and device based on medical information and automatic fee return machine

Abstract

The invention discloses a medical information-based fee return processing method and device and an automatic fee return machine, and the method comprises the steps: obtaining medical information after a fee return request of a user is obtained, and the medical information is medicine, diagnosis and payment information obtained in real time from different medical systems through a unified data center; according to the medical information, after it is determined that the user meets a fee refund condition, anomaly detection is carried out based on the medical information, and the anomaly detection is detection processing that whether the fee refund behavior of the user is abnormal or not is determined through the medical information; and if the detection result of the anomaly detection is normal, performing fee refund processing according to the medical information. According to the invention, automatic refund processing is carried out by integrating the information of different systems, so that the processing time can be shortened, the processing efficiency can be improved, the situation of manual auditing errors can be avoided, and the refund processing precision can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of information processing, and in particular to a refund processing method, device and automatic refund machine based on medical information. Background Art

[0002] With the development of technology, third-party payment platforms (such as WeChat and Alipay) are becoming increasingly popular, and more and more users are using them to pay for medical expenses. For example, in the medical field, paying through third-party payment platforms can not only improve user experience but also increase payment efficiency.

[0003] After payment, users may need a refund for various reasons (such as not picking up medicine, not seeing a doctor, or canceling registration, etc.). In order to process various expense information in a timely manner to avoid errors in fee settlement, the currently commonly used method is: when it is determined that the user requests a refund, the user's payment information is obtained, and then the payment records and payment items are compared one by one. The refund will be processed only when the payment items and payment records meet the requirements.

[0004] However, the currently commonly used methods have the following technical problems: there are many refund items, and each time they are processed, they must be checked item by item, which not only takes a long time to process and has low processing efficiency, but also uses a single payment record to compare with the paid items, which is prone to omissions and low refund processing accuracy. Summary of the Invention

[0005] The present invention provides a refund processing method, device and automatic refund machine based on medical information, which can solve the technical problems of the prior art that refund processing is time-consuming, inefficient and prone to errors.

[0006] A first aspect of an embodiment of the present invention provides a refund processing method based on medical information, the method comprising:

[0007] After receiving the user's refund request, obtain medical information, which includes medication, diagnosis, and payment information obtained in real time from different medical systems through a unified data center;

[0008] After determining that the user meets the refund conditions based on the medical information, performing anomaly detection based on the medical information, wherein the anomaly detection is a detection process that uses the medical information to determine whether the user's refund behavior is abnormal;

[0009] If the abnormality detection result is normal, a refund will be processed based on the medical information.

[0010] The present invention automatically processes refunds by integrating information from different systems, which not only shortens processing time and improves processing efficiency, but also avoids errors in manual review and improves the accuracy of refund processing.

[0011] In conjunction with the first aspect, in one implementation, performing abnormality detection based on the medical information includes:

[0012] Calling a preset LSTM model to calculate an abnormality probability value based on the medical information and using an isolation forest algorithm to calculate an abnormality score value based on the medical information;

[0013] Whether the user's refund behavior is abnormal is determined based on the abnormal probability value and the abnormal score value.

[0014] In conjunction with the first aspect, in one implementation, determining whether the user's refund behavior is abnormal based on the abnormal probability value and the abnormal score value includes:

[0015] If the abnormal probability value is less than the first preset probability value and the abnormal score value is less than the first preset score value, it is determined that the user's refund behavior is normal;

[0016] If the abnormal probability value is greater than the first preset probability value and the abnormal score value is greater than the first preset score value, it is determined that the user's refund behavior is abnormal.

[0017] In conjunction with the first aspect, in one implementation, after the step of performing abnormality detection based on the medical information, the method further includes:

[0018] If the detection result of the anomaly detection is abnormal, determining the risk category of the user's refund behavior according to the abnormal probability value and the abnormal score value;

[0019] Refund processing, termination processing or warning processing is performed based on the risk category.

[0020] In conjunction with the first aspect, in one implementation, the refund processing, termination processing, or warning processing based on the risk category includes:

[0021] If the risk category is high risk, an alarm process is triggered after intercepting and recording the medical information;

[0022] If the risk category is medium risk, manual review will be conducted and a refund will be made if the manual review is passed, or the application will be terminated if the manual review is not passed;

[0023] If the risk category is low risk, a refund will be processed.

[0024] In conjunction with the first aspect, in one implementation, determining the risk category of the user's refund behavior based on the abnormal probability value and the abnormal score value includes:

[0025] If the abnormal probability value is greater than the first preset probability value, the abnormal probability value is less than the second preset probability value, the abnormal score value is greater than the first preset score value, and the abnormal score value is less than the second preset score value, then the risk category of the user's refund behavior is determined to be low risk;

[0026] If the abnormal probability value is greater than the second preset probability value, the abnormal probability value is less than the third preset probability value, the abnormal score value is greater than the second preset score value, and the abnormal score value is less than the third preset score value, then the risk category of the user's refund behavior is determined to be medium risk;

[0027] If the abnormal probability value is greater than the third preset probability value and the abnormal score value is greater than the third preset score value, it is determined that the risk category of the user's refund behavior is high risk.

[0028] In conjunction with the first aspect, in one implementation, before the step of performing abnormality detection based on the medical information, the method further includes:

[0029] Generate a refund list using the medical information and display the refund list to the user through a preset digital person;

[0030] After receiving the user's confirmation information through the preset digital human, abnormality detection is performed based on the medical information.

[0031] In conjunction with the first aspect, in one implementation, after the step of obtaining medical information, the method further includes:

[0032] After determining that the user does not meet the refund conditions based on the medical information, a refund rejection prompt and refund rejection items are displayed to the user.

[0033] A second aspect of an embodiment of the present invention provides a refund processing device based on medical information, the device comprising:

[0034] An acquisition module is used to obtain medical information after receiving a user's refund request. The medical information is medication, diagnosis, and payment information obtained in real time from different medical systems through a unified data center.

[0035] a detection module, configured to, after determining that the user meets the refund conditions based on the medical information, perform anomaly detection based on the medical information, wherein the anomaly detection is a detection process that uses the medical information to determine whether the user's refund behavior is abnormal;

[0036] The refund processing module is used to perform a refund process based on the medical information if the detection result of the abnormality detection is normal.

[0037] A third aspect of an embodiment of the present invention provides an automatic refund machine, which includes: an automatic refund system and a device body. The automatic refund system is arranged in the device body, and the automatic refund system is applicable to the refund processing method based on medical information as described above.

[0038] Compared to existing technologies, the embodiments of the present invention provide a refund processing method, device, and automatic refund machine based on medical information. The beneficial effects of the present invention are as follows: after obtaining a user's refund request, the present invention can obtain medication, diagnosis, and payment information in real time from different medical systems; after determining that the user meets the refund conditions based on the information from different systems, determine whether the user's refund behavior is abnormal based on the information from different systems; and process the refund if the user's refund behavior is normal. By integrating information from different systems to automatically process refunds, processing time can be shortened and efficiency can be improved. It can also avoid manual review errors and improve the accuracy of refund processing. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 This is a flowchart of a refund processing method based on medical information provided by one embodiment of the present invention;

[0040] Figure 2 This is an operational flow chart of a refund processing method based on medical information provided by one embodiment of the present invention;

[0041] Figure 3 This is a schematic structural diagram of a medical information-based refund processing device provided by one embodiment of the present invention;

[0042] Figure 4 It is a structural diagram of an automatic refund machine provided by one embodiment of the present invention. DETAILED DESCRIPTION

[0043] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0044] With the development of technology, third-party payment platforms (such as WeChat and Alipay) are becoming increasingly popular, and more and more users are using them to pay for medical expenses. For example, in the medical field, paying through third-party payment platforms can not only improve user experience but also increase payment efficiency.

[0045] After payment, users may need a refund for various reasons (such as not picking up medicine, not seeing a doctor, or canceling registration, etc.). In order to process various expense information in a timely manner to avoid errors in fee settlement, the currently commonly used method is: when it is determined that the user requests a refund, the user's payment information is obtained, and then the payment records and payment items are compared one by one. The refund will be processed only when the payment items and payment records meet the requirements.

[0046] However, the currently commonly used methods have the following technical problems: there are many refund items, and each time they are processed, they must be checked item by item, which not only takes a long time to process and has low processing efficiency, but also uses a single payment record to compare with the paid items, which is prone to omissions and low refund processing accuracy.

[0047] In order to solve the above problems, the following specific embodiments will be used to introduce and explain in detail a refund processing method, device and automatic refund machine based on medical information provided in the embodiments of the present application.

[0048] In order to solve the technical problems of the existing refund process that is time-consuming, inefficient and prone to errors, Figure 1 , which shows a flow chart of a refund processing method based on medical information provided by an embodiment of the present invention.

[0049] In one embodiment, the refund processing method based on medical information is applicable to the automatic refund system installed in the automatic refund machine. The automatic refund system can be a control unit, which automatically reviews the information to simplify the processing flow and improve processing efficiency.

[0050] As an example, the refund processing method based on medical information may include:

[0051] S11. After receiving the user's refund request, obtain medical information, which is the drug, diagnosis and payment information obtained in real time from different medical systems through a unified data center.

[0052] In one embodiment, a user can apply for a refund from an automatic refund machine. The automatic refund system can receive the user's refund request. After receiving the refund request, the system can obtain medical information to comprehensively determine whether the user meets the refund requirements and protect the interests of both the user and the hospital. This medical information can include medication, diagnosis, and payment information obtained in real time from different medical systems through a unified data center.

[0053] In one operation mode, the automatic refund system can be connected to the data middle platform, and the data middle platform can be connected to various systems of the hospital. Different information can be obtained from different systems through the data middle platform, and then a comprehensive judgment can be made based on different information to determine whether the user meets the refund requirements.

[0054] For example, it can connect in real time to the hospital's HIS system, electronic medical order system, pharmacy management system, and medical technology department execution system. For example, it can obtain user payment records and medical order status information from the HIS system; obtain drug information from the pharmacy system, including drug status and drug collection status (whether the drug has been collected); and obtain user examination and inspection execution status (whether the patient has signed in or completed) from the medical technology system.

[0055] In one implementation, real-time data can be obtained through the hospital's unified data center (ESB enterprise service bus), for example: obtaining patient payment records (including prescription number, payment time, and payment method) from the HIS system; obtaining drug issuance status from the pharmacy system (associated through prescription number, field: is_delivered, Boolean value); obtaining inspection execution status from the medical technology system (field: execution_time, timestamp, null if not executed).

[0056] Optionally, DeepSeek-ESB can be used to synchronize data across multiple systems within seconds, thereby shortening the refund review time and greatly improving efficiency.

[0057] Specifically, you can use RESTful APIs or message queues (such as RabbitMQ) to achieve real-time data synchronization between systems to ensure that conditional judgments are based on the latest status.

[0058] For data storage and indexing in each system, key status fields can be stored in the Redis cache (such as patient payment records within 30 days) to improve query efficiency; a temporary table temp_refund_conditions can also be created to store the preliminary screening results of the current user's refundable items (such as unexecuted examination order numbers and unissued prescription numbers).

[0059] After obtaining the above medical information, it can be determined whether the user meets the conditions based on the above medical information. In one operation mode, different rules can be pre-set to determine whether the above medical information meets the preset rules to avoid refunds for users violating regulations.

[0060] For example, predefined refund rules and conditions may include: "whether the medicine has been collected" and "whether the inspection has been carried out". These different rules can be converted into logical expressions that can be recognized by the system. For example:

[0061] Whether the drug has been received: Pharmacy Inventory System. Drug Status [Prescription Number] == "Not Issued";

[0062] Check whether the execution is carried out: Medical Technology System. Inspection Record [Inspection Order Number]. Execution Time == null and Current Time - Payment Time ≤ Refund Time Limit (e.g. within 3 days).

[0063] After the user initiates a refund application at the self-service refund machine, the automatic refund system can retrieve payment records that have not been executed within 3 days and verify them one by one according to predefined rules.

[0064] If the conditions are met (such as the medicine has not been collected and opened, the examination has not been performed and no appointment has been made), a list of refundable items will be generated and pushed to the patient for confirmation.

[0065] If the conditions are not met (e.g. the medicine has been collected or the inspection has been performed), the refund will be refused.

[0066] In one embodiment, rules are edited through the DeepSeek-Adaptive rule engine. The DeepSeek-Adaptive rule engine supports rapid iteration, and the deployment time of new rules is short. It can cope with changes in medical insurance policies, thereby improving the response speed of the system.

[0067] Specifically, the business rules for refund conditions predefined by the user can be shown in the following table:

[0068]

[0069]

[0070] You can also define the priorities of different rules, such as mandatory rejection conditions (such as "drugs have been collected" and "examinations have been performed") > refundable conditions (such as "no examination appointment") > special processing conditions (such as "refunds across years require manual review").

[0071] In an optional operation mode, the above rules can be digitally modeled and conditionally parameterized. Users can configure rule parameters (such as refund time threshold and amount limit) through the backend management interface, which are stored in the database table refund_rules_config and support dynamic modification (such as extending the refund time limit to 7 days during the epidemic). The specific configuration can be shown in the following table:

[0072] Rule Name Parameter name default value describe Check refund time limit max_days 3 Maximum number of days after payment for refund Drug distribution status valid_status Not issued Refundable drug status

[0073] In an optional embodiment, when determining the refund conditions, collaborative verification of multiple systems may also be performed to determine whether the user meets the refund conditions.

[0074] Specifically, the collaborative verification of multiple systems may include the following steps:

[0075] Step 1: Preliminary Screening: After a patient requests a refund at the self-service refund machine, the automated refund system filters out non-refundable items based on the user's selected payment_type (e.g., invoiced cash payments require manual processing). The automated refund system then calls the get_unexecuted_orders() function to retrieve orders from the HIS system with a status of "unexecuted" or "uncompleted" within three days (filtered by the status field).

[0076] Step 2: Rule engine matching: traverse the preliminarily screened orders and perform rule verification on each order: Drug orders: query the pharmacy system prescription.is_delivered. If it is false, the drug refund rule is triggered; Inspection orders: check whether the medical technology system exam.execution_time is null and does not exceed max_days (obtained from the configuration table).

[0077] Step 3: Multi-system collaborative verification: If the order involves medical insurance payment, call the medical insurance platform interface check_insurance_status(order_id) to confirm whether it has been reimbursed (if reimbursed, it must be canceled first); if the refund amount exceeds 5,000 yuan, trigger the financial system require_manual_audit(order_id) to mark it for manual review.

[0078] It should be noted that for exception handling and fault-tolerant operations, an exception can be returned at a certain system interface (such as a pharmacy system timeout), the latest status of the local cache can be enabled (with the mark "data may not be the latest"), and the patient can be prompted "the system is verifying, please wait"; for ambiguous conditions (such as "unopened medicines" because the system does not record the packaging status), the default treatment is "non-refundable", and the patient is guided to the manual window for verification.

[0079] Optionally, for complex scenarios (such as partial refunds and medical insurance splits), the financial system and medical insurance settlement platform can be linked to verify the amount split. For example, the self-paid portion is refunded through the original route, while the medical insurance portion triggers the medical insurance reimbursement process to ensure that the refund amount matches the payment method.

[0080] By using the equipment to perform condition detection based on medical information, there is no need for manual processing, replacing the refund processing method of manually checking each item one by one. This can greatly shorten the processing time and improve the processing accuracy while achieving millisecond-level refund condition verification.

[0081] In addition, to facilitate rule configuration by different technicians, the automatic refund system can feature a visual rule management interface. This interface allows hospital administrators to visually configure refund conditions and supports: adding / modifying rules (drag-and-drop logical expression editing); allowing technicians to view rule hit logs (e.g., if the "Check Not Executed" rule was hit 120 times today); and testing rule validity (by inputting simulated data to verify whether a refund is triggered).

[0082] In one embodiment, when it is determined that the user does not meet the conditions, in order to promptly prompt the user with the reason for rejecting the refund and avoid the user from applying repeatedly, as an example, after the step of obtaining medical information, the method may further include the following steps:

[0083] S21. After determining that the user does not meet the refund conditions based on the medical information, display a refund rejection prompt and refund rejection items to the user.

[0084] In one embodiment, if it is determined based on medical information that the user does not meet the refund conditions, a refund rejection prompt can be displayed to the user to indicate that "the refund conditions are not met" and the reason can be noted (such as "the medicine has been picked up and no refund can be made").

[0085] S12. After determining that the user meets the refund conditions based on the medical information, perform anomaly detection based on the medical information, wherein the anomaly detection is a detection process that uses the medical information to determine whether the user's refund behavior is abnormal.

[0086] In one embodiment, if it is determined based on medical information that the user meets the refund conditions, anomaly detection can be performed based on the medical information. The anomaly detection is a detection process that uses the medical information to determine whether the user's refund behavior is abnormal.

[0087] It is possible that the user meets the refund conditions, but the user makes multiple refunds or the refund amount deviates.

[0088] For example, if a user issues 3 or more refunds in a single day, this indicates that normal patients typically issue a single refund due to reasons such as missed examinations. Multiple refunds in a single day may indicate malicious activity exploiting loopholes in the rules. If a user issues 5 or more refunds within 30 days, this indicates that the number exceeds the reasonable range for normal patients and may indicate repeated payment and refund arbitrage.

[0089] For example, a single refund amount of 5,000 yuan or more: Based on the hospital department's fee standards, if ordinary outpatient services rarely exceed this amount, it may involve fraudulent payment. Repeated refunds for the same drug: If a drug with the same prescription number is refunded multiple times (but only paid once), there may be document forgery.

[0090] Another example is refunds outside of medical consultation hours: for example, requests for refunds concentrated between midnight and 6 a.m. do not conform to normal medical consultation hours. Immediate refunds after payment: requests for refunds within 5 minutes of payment (excluding the reasonable scenario of immediate cancellation of the examination) may be a test system bug.

[0091] Another example is frequent refunds across departments: patients frequently request refunds from multiple departments (e.g., internal medicine, surgery, and laboratory) without a clear logical connection between diagnosis and treatment. Abnormal medical insurance splits: an abnormal ratio between medical insurance reimbursement and out-of-pocket expenses (e.g., out-of-pocket expenses less than 10% and multiple refunds), which may indicate medical insurance fraud.

[0092] Through anomaly detection, erroneous refunds can be avoided to protect the rights and interests of the hospital and improve the accuracy and efficiency of refund processing.

[0093] In an optional embodiment, in order to accurately determine whether the user's behavior is abnormal, the abnormality detection based on the medical information may include the following sub-steps, for example:

[0094] S121. Calling a preset LSTM model to calculate an abnormality probability value based on the medical information and using an isolation forest algorithm to calculate an abnormality score value based on the medical information.

[0095] S122: Determine whether the user's refund behavior is abnormal based on the abnormal probability value and the abnormal score value.

[0096] In one embodiment, the medical information can be input into a preset LSTM model so that the preset LSTM model calculates an abnormality probability value based on the medical information. Similarly, an isolation forest algorithm can be used to calculate an abnormality score value based on the medical information.

[0097] The preset LSTM model can be a long short-term memory network (LSTM). The user's historical medical data (including visit time, payment items, number of refunds, refund amount, diagnosis department, etc.) can be obtained in advance and converted into a time series vector (such as refund frequency and mean amount divided by week / month), which is used as the user's historical behavior sequence. Next, the LSTM is used to learn the temporal patterns of normal refund behavior, such as "the same patient can only receive a maximum of one refund per month" and "unexecuted refunds are concentrated within 24 hours after payment", to obtain an LSTM model. The subsequent LSTM model can output a probability score for abnormal behavior to obtain an abnormal probability value.

[0098] Long short-term memory (LSTM) networks are a special type of recurrent neural network (RNN) that can learn and memorize dependencies in long time series. In the analysis of medical behavior time series, LSTMs build temporal patterns of normal medical behavior by learning from a large amount of normal medical behavior data.

[0099] For example, it learns patterns such as "a patient can only make one refund per month," "refunds that have not been processed are concentrated within 24 hours of payment," and "patients typically make refunds during weekday appointment times." When new medical behavior data is input, the LSTM model calculates the probability that the behavior is normal based on the learned patterns and outputs an abnormality probability score (0-1, with ≥0.8 considered abnormal). If a patient originally made zero or one refund per month, but suddenly made three refunds within a month, the LSTM model would recognize this change as inconsistent with the historical behavior pattern and output a higher abnormality probability score.

[0100] Similarly, the isolation forest algorithm can be used to detect outlier data points or information in medical information and identify short-term extreme abnormal behaviors (such as outliers in amount or number of times).

[0101] The Isolation Forest algorithm is a tree-based anomaly detection method. In time-series analysis of medical behavior, the computer first calculates outlier characteristics, such as the number of refunds per day, the deviation of the refund amount from the historical mean (such as the Z-score), and the concentration of refund items (such as the proportion of multiple drug refunds).

[0102] The Isolation Forest algorithm detects outliers by calculating outlier characteristics such as the number of refunds per day, the deviation of the refund amount from the historical mean (e.g., Z-score), and the concentration of refund items (e.g., the proportion of multiple refunds). The Isolation Forest algorithm then constructs an "isolation tree" representing normal behavior. The path length from the sample to the root node is tested, and samples with too short a path are identified as outliers (e.g., ≥3 refunds per day, ≥3 times the historical mean for a single refund amount).

[0103] For each refund request, an outlier score is calculated in real time to generate an anomaly score. Finally, the anomaly probability value and anomaly score from the LSTM model are combined to comprehensively determine whether the user's refund behavior is abnormal.

[0104] For example, for the number of refunds per day, if a patient makes ≥3 refunds per day, or the single refund amount is ≥3 times the historical average, the isolation forest algorithm will determine it as an outlier because these behaviors clearly deviate from the normal behavior pattern.

[0105] As an example, determining whether the user's refund behavior is abnormal based on the abnormal probability value and the abnormal score value may include the following sub-steps:

[0106] S1221. If the abnormal probability value is less than the first preset probability value and the abnormal score value is less than the first preset score value, determine that the user's refund behavior is normal.

[0107] S1222: If the abnormal probability value is greater than the first preset probability value and the abnormal score value is greater than the first preset score value, determine that the user's refund behavior is abnormal.

[0108] In one embodiment, if the abnormal probability value is less than a first preset probability value (eg, 0.2) and the abnormal score value is also less than the first preset score value (eg, 0.2), the user's refund behavior is determined to be normal.

[0109] On the contrary, if the abnormal probability value is greater than the first preset probability value (for example, 0.2) and the abnormal score value is also greater than the first preset score value (for example, 0.2), the user's refund behavior is determined to be abnormal.

[0110] For example, if a user applies for refunds between 0:00 and 6:00 in the morning (refunds during non-medical consultation hours), and the LSTM model determines that this behavior does not conform to their previous time patterns, and the isolation forest algorithm also determines it as an outlier, it can be determined that the refund behavior is abnormal.

[0111] There are many types of users who need to make refunds, including young users and elderly users. In order to assist users in performing refund operations on the automatic refund device, as an example, before the step of performing abnormality detection based on the medical information, the method further includes the following steps:

[0112] S31. Generate a refund list using the medical information and display the refund list to the user through a preset digital person.

[0113] S32. After receiving the user's confirmation information through the preset digital human, perform abnormality detection based on the medical information.

[0114] In one embodiment, the pre-configured digital human is interactive and can provide users with audio guides, including guidance on the refund process, hospital department navigation, and an introduction to the hospital's history and culture. It supports bilingual switching between Cantonese and Mandarin, catering to the needs of both local and out-of-town patients. It can also provide intelligent Q&A, leveraging natural language processing (NLP), to answer patients' questions about refund policies, medical insurance reimbursement, and device operation.

[0115] The preset digital human can have emotional perception capabilities, recognize the patient's expression through the camera, and actively ask whether help is needed (such as "You look a little confused, do you need my assistance?").

[0116] Based on natural language processing (NLP) dialogues, data integration and knowledge base construction can be carried out in advance.

[0117] The first step is to obtain input data, which can include: integrating hospital refund policy documents (such as the "Outpatient Refund Management Measures"), medical insurance settlement rules (such as medical insurance split refund clauses), and historical patient consultation records (high-frequency issues such as "drug refund time limit" and "inspection failure to implement the refund process").

[0118] The second step is to build a knowledge graph. Specifically, you can use entity relationship extraction technology (such as SpaCy or StanfordNLP) to associate key entities in the text (such as "refund time limit" and "unissued medicine") with rule logic to form a structured knowledge base. For example:

[0119] {"Refund conditions":{"Drug category":["Prescription status = Not issued","Packaging unopened"],"Inspection category":["Execution time = null","No appointment"]},"Time limit":"Self-service refund within 3 days after payment"}.

[0120] For the processing flow of natural language processing (NLP) dialogue, intent recognition can be performed first. Specifically, patient questions can be classified based on pre-trained models (such as BERT) to distinguish intent types (such as "refund process query" and "medical insurance refund rules").

[0121] Next, perform entity extraction. Specifically, the BiLSTM+CRF model can be used to extract key entities in the question (such as "prescription number" and "check order number") and match them with the knowledge base.

[0122] Finally, answers can be generated dynamically. Specifically, template matching can be performed (for example, predefined answer templates can be directly called for common questions (such as "How do I get a refund?"), and then generative answers can be generated (for example, GPT-3 can be used to generate personalized answers for complex questions (such as "How do I get a refund across departments?"), and real-time data (such as the current status of drug distribution) can be combined to enhance accuracy).

[0123] For the preset digital human emotion perception function, the patient's facial image (resolution ≥720p) can be captured through the self-service machine's high-definition camera and performed using OpenCV; then face detection can be performed, specifically using the Haar cascade classifier to locate the face area.

[0124] Then, key point positioning is performed. For example, 68 facial feature points (such as mouth corners and eyebrow corners) are extracted through the Dlib library and standardized into a 128×128 pixel image.

[0125] When calling the emotion classification model to determine the user's emotion, the emotion classification model can be a convolutional neural network (CNN) based on ResNet-50, with a standardized facial image as input and output of 7 categories of emotion probabilities (anger, confusion, calmness, etc.).

[0126] Training data for the emotion classification model can be FER-2013 (containing 35,887 annotated images). Alternatively, you can create your own dataset, for example, by collecting real patient expressions during a hospital refund request (e.g., confusion after a failed refund, relief after a successful one) to enhance adaptability to specific scenarios.

[0127] When the emotion classification model performs real-time reasoning, if "confusion" is detected (probability > 0.7), the digital human will be triggered to actively ask: "Do you need help?"; if "anger" is detected (probability > 0.8), the interface will be automatically simplified and the call will be transferred to manual customer service.

[0128] In order to further enhance the interactive effect of the digital human, different background music and 3D models of iconic buildings can be played when displaying the digital human to enhance recognition.

[0129] When constructing a digital human, you can use the character xx as the core prototype, referencing historical photos and documents of the character, and designing a refined image of middle-aged elegance (e.g., Mao suits, glasses, etc.). The digital human's clothing and color scheme: The clothing incorporates the blue and white color scheme of the school emblem, and the collar / cuffs feature the school emblem (e.g., shields, crosses, books, etc.), reflecting the school motto of "university love and respect for virtue."

[0130] When doing 3D modeling, you can use tools such as Blender / 3ds Max to build high-precision models, refine facial expressions (such as smiling, focusing), and body movements (such as hand gestures) to ensure that the image is friendly and professional.

[0131] When dynamically rendering digital humans, cloud rendering technology, the same as that used in the hospital's digital virtual medical museum, can be used to reduce device power consumption and support smooth display on multiple terminals (self-service machine screens, mobile apps, and PCs).

[0132] The digital human's voice can include bilingual audio libraries in Cantonese and Mandarin, and through speech synthesis technology (TTS) it can achieve natural and smooth tour guide broadcasting, and support real-time response to patient instructions (such as "Please repeat in Cantonese").

[0133] At the same time, hardware such as cameras and microphones can be integrated into the device, face detection and expression recognition can be achieved through OpenCV, and the interactive interface can be developed in combination with the Unity / UE engine to ensure that the digital human's movements and voice are synchronized (such as matching lip movements when speaking).

[0134] After receiving the user's refund request, the user can be presented with a digital human through which the user can interact. A refund list can also be displayed to the user, containing the refund items requested by the user. Specifically, the items requiring refund and the amount of each item can be extracted from the medical information.

[0135] After the user checks the refund items and corresponding amounts in the refund list and confirms them, the user's confirmation information can be obtained, which can be a confirmation voice or click. Then, an anomaly detection is performed on the user's refund behavior to determine whether the user's refund behavior is normal.

[0136] S13. If the abnormality detection result is normal, a refund is processed according to the medical information.

[0137] In one embodiment, when the abnormality detection result is normal, it is determined that the user meets the refund requirements. At this time, a refund can be processed based on the refund amount of the medical information.

[0138] The computer performs a preliminary screening of refund requests based on preset threshold rules. These rules are based on the hospital's business experience and refund policies, such as "3 or more refunds per day," "5 or more refunds within 30 days," and "single refund amount 5,000 yuan or more."

[0139] When a user initiates a refund request at a self-service refund machine, the system checks these rules in real time. If a refund action meets any of these conditions, it will be marked as suspicious and enter the subsequent further analysis process.

[0140] In addition, the system not only makes judgments based on a single rule or model, but also conducts a comprehensive analysis based on the patient's multi-dimensional information, including the patient's age, department visited, historical refund reasons, payment method, medical insurance type, etc.

[0141] For example, for patients with chronic diseases, regular medication returns may be a reasonable behavior, and the system will make a judgment based on their medical records; for situations where the ratio of the medical insurance reimbursement amount to the self-paid amount is abnormal (such as the self-paid proportion is less than 10% and there are multiple refunds), the medical insurance reimbursement records and relevant policies will be further checked to determine whether there is suspicion of medical insurance fraud.

[0142] In one embodiment, the user making a refund may be an elderly person, who may have made a misoperation or an error in the operation, resulting in an abnormal refund behavior. It may also be that an illegal user has violated the rules many times, thereby identifying his refund behavior as abnormal. When it is determined that the user's refund behavior is abnormal, in order to conduct a multi-dimensional correlation analysis to determine the risk level or risk category of this abnormal behavior, and in combination with information such as the patient's age, the department visited, and the historical refund reasons, reasonable scenarios are excluded (such as the need for manual review of regular medication returns for chronic patients). Among them, as an example, after the step of performing anomaly detection based on the medical information, the method may also include the following steps:

[0143] S14. If the detection result of the anomaly detection is abnormal, determine the risk category of the user's refund behavior according to the abnormality probability value and the abnormality score value.

[0144] In one embodiment, the risk category of the user's abnormal refund behavior may be determined based on the numerical value of the abnormal probability value and the numerical value of the abnormal score value.

[0145] If it is a high risk, an alarm can be triggered. If it is a low risk, you can then decide whether to refund.

[0146] By analyzing the risk categories of users' abnormal refund behaviors, we can improve the effectiveness of risk prevention and control, thereby enhancing users' refund experience and protecting the interests of users and hospitals.

[0147] In one embodiment, determining the risk category of the user's refund behavior based on the abnormal probability value and the abnormal score value may include the following sub-steps:

[0148] S141. If the abnormal probability value is greater than the first preset probability value, the abnormal probability value is less than the second preset probability value, the abnormal score value is greater than the first preset score value, and the abnormal score value is less than the second preset score value, then determine that the risk category of the user's refund behavior is low risk.

[0149] S142: If the abnormal probability value is greater than the second preset probability value, the abnormal probability value is less than the third preset probability value, the abnormal score value is greater than the second preset score value, and the abnormal score value is less than the third preset score value, then determine that the risk category of the user's refund behavior is medium risk.

[0150] S143: If the abnormal probability value is greater than a third preset probability value and the abnormal score value is greater than a third preset score value, determine that the risk category of the user's refund behavior is high risk.

[0151] In one operation mode, if the abnormal probability value is greater than a first preset probability value (for example, 0.2), the abnormal probability value is less than a second preset probability value (for example, 0.6), the abnormal score value is greater than the first preset score value (for example, 0.2), and the abnormal score value is less than the second preset score value (for example, 0.6), then the risk category of the user's refund behavior is determined to be low risk.

[0152] If the abnormal probability value is greater than the second preset probability value (for example, 0.6), the abnormal probability value is less than the third preset probability value (for example, 0.8), the abnormal score value is greater than the second preset score value (for example, 0.6) and the abnormal score value is less than the third preset score value (for example, 0.8), then the risk category of the user's refund behavior is determined to be medium risk.

[0153] Similarly, if the abnormal probability value is greater than the third preset probability value (eg, 0.8) and the abnormal score value is also greater than the third preset score value (eg, 0.8), the risk category of the user's refund behavior is determined to be high risk.

[0154] It should be noted that the above-mentioned preset values ​​can be adjusted according to actual needs, and this application does not limit them.

[0155] S15. Perform refund processing, termination processing or alarm processing based on the risk category.

[0156] In one embodiment, different operations are performed for different risk categories. For example, behaviors in the low-risk category (such as an abnormal probability value of <0.5 and an isolation forest score of <0.5) may pass the review directly and continue with the refund process; behaviors in the medium-risk category (such as an abnormal probability value of 0.5-0.8 or an isolation forest score of 0.5-0.9) will be marked as "suspicious", and relevant information will be recorded and may trigger further manual review, terminating the refund process; behaviors in the high-risk category (such as LSTM probability >0.8 and isolation forest score >0.9) will be subject to alarm processing, which can directly trigger manual review or reject the refund operation, and prompt the patient "to go to the manual window for processing."

[0157] At the same time, the system will record all abnormal behavior logs, including operation time, patient information, abnormality type, etc., to facilitate subsequent data tracing and security audits. If necessary, it will link with the hospital security system to track IP addresses, equipment numbers and other information to prevent malicious attacks and fraud.

[0158] In one application example, in the dynamic risk control mechanism of the self-service refund system at XXXX Memorial Hospital, the abnormal refund behavior identification rules cover abnormal pattern classification, detection methods, and handling procedures, as follows:

[0159] Abnormal behavior can be classified into the following categories: high-frequency refunds (≥3 refunds per day; ≥5 refunds within 30 days, which exceeds the reasonable range for ordinary patients and may involve malicious use of loopholes in the rules or "repeated payment-refund" arbitrage). Refunds for high or abnormal items (a single refund amount ≥5,000 yuan may involve false payments; repeated refunds for the same drug, that is, multiple refund applications for the same prescription number, may indicate document forgery). Time sequence anomalies (refunds during non-consultation hours, such as concentrated refund applications between 0:00 and 6:00 in the morning, which does not conform to the normal pattern of medical treatment time; refunds immediately after payment, or refunds within 5 minutes after payment (excluding reasonable scenarios of immediate cancellation of examinations), may be testing system vulnerabilities). Abnormal behavior patterns (high frequency refunds across departments, patients frequently apply for refunds in multiple departments (such as internal medicine, surgery, and laboratory), with no clear logical connection between diagnosis and treatment; abnormal medical insurance splitting, abnormal ratio of medical insurance reimbursement and self-paid amount (such as self-paid proportion <10% and multiple refunds), may involve medical insurance fraud).

[0160] In one embodiment, the refund processing, termination processing, or warning processing based on the risk category may include the following sub-steps:

[0161] S151. If the risk category is high risk, an alarm process is triggered after intercepting and recording the medical information.

[0162] S152. If the risk category is medium risk, manual review is performed and a refund is made if the manual review is passed, or the application is terminated if the manual review is not passed.

[0163] S153. If the risk category is low risk, a refund is processed.

[0164] If the risk category is determined to be high risk, an alarm process is triggered after the medical information is intercepted and recorded.

[0165] Abnormal behavior categorized as high-risk (e.g., five refunds in a single day) can be directly rejected with a prompt stating "need to go to a manual counter for processing." For example, if a user makes three or more refunds in a single day: Upon identification as high-risk, the refund request will be immediately intercepted, with a prompt stating "Your refund requests are too frequent and need to be processed at a manual counter." The abnormal behavior will be logged and the risk level noted. Upon receiving the notification, the manual counter staff will review the patient's medical records, payment records, and other information to confirm whether there is malicious arbitrage activity.

[0166] For example, if a user's single refund amount is ≥5,000 yuan, the refund process can be stopped and frozen after being classified as high-risk, and the user will be prompted to submit a refund explanation and relevant supporting documents at the manual window. The manual review focuses on verifying the detailed cost and the implementation of medical treatment items to confirm whether there is any fraudulent payment or other violations.

[0167] For example, if a user's medical insurance split is abnormal (the ratio of medical insurance reimbursement to out-of-pocket expenses is abnormal, and there are multiple refunds), the refund process will be immediately stopped after the user is identified as high-risk, and the medical insurance management department will be contacted to verify the reimbursement records and patient medical records. If suspected medical insurance fraud is confirmed, it will be reported to the relevant regulatory authorities as required, and evidence will be retained to assist in the investigation.

[0168] For example, if a user repeatedly requests refunds for the same medication, the system can reject the refund request and place the patient on a high-risk list after determining the patient is at high risk. The system also notifies the pharmacy and finance department to jointly verify the authenticity of the prescription and medication dispensing records. If fraudulent activity, such as forged receipts, is detected, accountability procedures will be initiated.

[0169] When terminating a refund or providing manual prompts, the digital human can also determine the user's mood. For example, if negative emotions are detected during a failed refund, the specific reason for failure (such as "Drugs have been issued, please contact the receptionist") will be prioritized to reduce repeated operations. It can also record the emotional changes of the same patient during multiple refunds, assisting human service personnel in anticipating potential disputes (e.g., three consecutive refunds accompanied by "anger").

[0170] If the risk category is medium risk (such as three refunds within 30 days), the financial staff will be notified to conduct manual review and processing, and the refund will be processed if the manual review is passed, or the processing will be terminated if the manual review is not passed.

[0171] Manual review can be done by the finance department to verify the authenticity of documents. It can also trace data, log abnormal behavior, and link with the hospital security system to track IP addresses and device numbers to prevent malicious attacks.

[0172] For example, if a user requests five or more refunds within 30 days, they will be classified as medium-risk and will be logged as such, triggering a manual review process. Finance staff will conduct a comprehensive review of the patient's refund history and medical records. If no reasonable reason is found, they will deny subsequent refund requests and report the situation to hospital management.

[0173] For example, if a user applies for a refund outside of medical treatment hours (applying for refunds between midnight and 6 a.m.), and is classified as medium-risk, the relevant operation information can be recorded and pushed to the manual review queue. The reviewer will determine whether the request is reasonable based on the patient's historical medical habits and the current medical situation. If there is any doubt, the reviewer will contact the patient to verify the reason for the refund.

[0174] For example, if a user requests a refund immediately after payment (applying for a refund within 5 minutes of payment, excluding the reasonable scenario of immediate cancellation of the examination), and is classified as medium-risk, the system can automatically retrieve detailed information about the payment item to determine whether there is a possibility of a test system vulnerability. If suspicion cannot be ruled out, the refund will be suspended and the patient will be notified to provide a reasonable explanation, and manual review will be conducted if necessary.

[0175] For example, if a user frequently refunds fees across departments and is classified as medium-risk, this abnormal behavior can be reported to the medical and finance departments. The two departments will jointly analyze the patient's diagnosis and treatment logic in different departments. If they find frequent refunds without actual treatment needs, the refund application will be rejected and the patient will be investigated.

[0176] Through the above design, not only can a closed loop of the entire process from rule triggering, intelligent interaction to risk prevention and control be achieved, taking into account both efficiency improvement and financial security, but it can also integrate the cultural characteristics of the hospital and enhance the patient experience.

[0177] In addition, to improve the accuracy of the hospital's self-service refund system's abnormal refund behavior identification rules, we can start from data mining, model optimization, rule refinement and multi-dimensional collaboration. The specific optimization strategies are as follows:

[0178] First, we will deepen data mining and feature engineering, including: Adding behavioral feature dimensions: In addition to existing data, we will include data such as patient visit duration, correlation between examination items, and doctor's order modification records. For example, if a patient has a very short visit but requests a refund for multiple examinations, this may be an anomaly. We will also analyze the logical correlation between examination items (such as the correlation between blood routine results and subsequent diagnoses). If the refund item has no reasonable connection to the diagnosis, it will be marked as suspicious.

[0179] Alternatively, dynamic weights can be set for each recognition rule based on different departments, diseases, and time periods. For example, the weight of refunds due to emergency situations in the emergency department could be appropriately reduced, while the weight of repeated refunds for the same medication in the chronic disease department could be increased. Weight parameters can be determined through historical data training and regularly adjusted based on actual conditions.

[0180] Second, optimize the model algorithm. You can introduce an ensemble learning model to integrate LSTM and isolation forest algorithms with other models (such as random forest and XGBoost), leveraging the complementary advantages of different models to reduce the limitations of a single model. For example, using the stacking ensemble method, the outputs of multiple base models are used as the input of the meta-model to further improve the accuracy of anomaly identification. Alternatively, strengthen the deep learning model: Based on LSTM, try more complex deep learning architectures such as bidirectional LSTM, GRU (gated recurrent unit), or Transformer models to better capture complex patterns in long and short-term time series. At the same time, increase the amount and diversity of model training data to improve the model's generalization ability.

[0181] Third, refine and dynamically update rules. Specifically, scenario-based rule design can be used: Develop more detailed rules for different refund scenarios. For example, for refunds of medical checkup packages, set a rule that states "some items in the package have been completed, and the refund of the remaining items requires manual review." For refunds of inpatients, consider factors such as length of stay and stage of treatment.

[0182] A dynamic rule update mechanism can be set up, specifically a real-time monitoring and feedback mechanism for rules, which can promptly update and supplement rules based on emerging abnormal cases, hospital policy adjustments, changes in medical insurance policies, etc. For example, when a new type of fraud method emerges, its characteristics can be quickly converted into rules and incorporated into the system.

[0183] Fourth, multi-dimensional collaborative analysis can strengthen data interaction with external hospital systems, such as the medical insurance anti-fraud platform and the public security system (for verifying patient identity), to obtain more diverse verification information. If the medical insurance anti-fraud platform indicates that a patient has a suspicious refund record from another medical institution, the system will automatically increase the patient's risk level, thereby achieving cross-system data linkage.

[0184] We not only focus on individual refunds, but also analyze the patient's overall behavior patterns over time. For example, if a patient repeatedly refunds similar items at different hospitals, even if each refund behavior alone does not trigger an anomaly rule, it can still be identified as an anomaly through correlation analysis.

[0185] Fifth, AI-assisted audits can be implemented. Natural language processing (NLP) can be used to perform semantic analysis on the reasoning provided by patients in refund applications to determine its rationale. If the description is vague, inconsistent, or inconsistent with the actual situation, the risk level will be automatically increased. Alternatively, human-machine collaborative audits can be implemented to optimize the manual audit process, providing auditors with detailed risk analysis reports and historical case references. Auditors can also provide feedback from their experience to the system, continuously optimizing recognition rules and model parameters.

[0186] In addition, we designed test cases for both normal and abnormal scenarios based on predefined refund rules (such as "drugs not collected" and "examination not performed"). For example, we simulated refunds for medications that had been collected, refunds for examinations that had not been performed, and refunds for split medical insurance payments to verify that the system accurately handled the rules.

[0187] At the same time, manual sampling can also be carried out, randomly selecting a certain number of actual refund records (such as 100 records per month), manually checking the consistency between the system's judgment results and the rule execution, and calculating the accuracy of the rule execution (number of correct judgments / total number of samplings).

[0188] To assess the accuracy of anomaly identification using models and algorithms, we can use historical refund data to label true anomalies and normal samples, compare the identification results with the true labels, and calculate precision, recall, and F1 value. For example, if the system identifies 100 anomaly records, 80 of which are true anomalies and 20 are false positives, the precision is 80%. If there are actually 120 anomaly records and the system only identifies 80, the recall is approximately 66.7%.

[0189] For the statistics of missed alarms and false alarms, we can count the real abnormal cases that were not identified (missed alarms) and the normal cases that were mistakenly judged as abnormal (false alarms), and analyze the causes of missed alarms and false alarms, such as rule loopholes, model misjudgment, etc.

[0190] It also assesses whether the system can identify all pre-set abnormal patterns (such as high-frequency refunds, high-value refunds, and timing anomalies). By simulating new abnormal scenarios not covered by the system (such as refunds exploiting loopholes in new policies), the system's generalization capabilities can be tested and improvement suggestions can be made.

[0191] For system operation stability assessment, performance indicator monitoring can be performed, including:

[0192] Response time: In high-concurrency scenarios (such as peak outpatient visits), monitor the system's average response time for processing refund applications to ensure that it does not exceed the set threshold (such as 2 seconds).

[0193] Throughput: Count the maximum number of refund requests that the system can process per unit time (such as per minute) to evaluate the system's carrying capacity.

[0194] During use, the system may malfunction. The number of failures, failure types and recovery time of the system within a certain period (such as 1 month) can be recorded, the causes of the failures (such as interface timeout, server crash, etc.) can be analyzed, and the stability and fault tolerance of the system can be evaluated.

[0195] Furthermore, continuous optimization can be achieved through data-driven processes during use. For example, regular (e.g., monthly) statistical analysis can be performed to analyze trends in indicators such as anomaly recognition rate, false alarm rate, and missed alarm rate to determine whether prevention and control effectiveness has improved. For example, if the false alarm rate drops from 15% to 8% after optimization, this indicates that the improvement measures are effective.

[0196] Reference Figure 2 , shows an operational flow chart of a refund processing method based on medical information provided by an embodiment of the present invention.

[0197] Specifically, the refund processing method based on medical information may include the following steps:

[0198] The first step is to obtain the patient's refund application.

[0199] The second step is to obtain data from different systems through the data middle platform to obtain medical information.

[0200] The third step is to use the rule engine to determine whether the user meets the refund conditions based on the medical information. If the refund conditions are not met, the reason for rejection will be prompted.

[0201] In the fourth step, if the refund conditions are met, a refund list is generated and the user is interacted with through a digital human.

[0202] In the fifth step, after the user's confirmation information is confirmed, it can be determined whether the user's refund behavior is abnormal based on the medical information.

[0203] Step 6: If it is determined that the user's refund behavior is normal, the refund can be processed.

[0204] In the seventh step, if it is determined that the user's refund behavior is abnormal, the user's abnormal behavior can be risk graded to determine the risk category.

[0205] Step 8. If the user's abnormal refund behavior is determined to be low-risk, a refund can be processed.

[0206] In the ninth step, if the user's abnormal refund behavior is determined to be in the medium-risk category, manual review can be conducted and the refund can be processed if the manual review is passed. Otherwise, the refund process will be terminated if the manual review fails.

[0207] In the tenth step, if the user's abnormal refund behavior is determined to be a high-risk category, the refund application can be immediately intercepted and the user's application log can be recorded. At the same time, the alarm system can be linked and the user's data can be traced to conduct anti-fraud processing.

[0208] In this embodiment, a refund processing method based on medical information is provided. The beneficial effects of the present invention are as follows: after obtaining a user's refund request, the present invention can obtain medication, diagnosis, and payment information in real time from different medical systems; after determining that the user meets the refund conditions based on the information from different systems, determine whether the user's refund behavior is abnormal based on the information from different systems; and perform refund processing when the user's refund behavior is determined to be normal. By integrating information from different systems to automatically process refunds, not only can processing time be shortened and efficiency be improved, but manual review errors can also be avoided, thereby improving the accuracy of refund processing.

[0209] The embodiment of the present invention also provides a refund processing device based on medical information, see Figure 3 , showing a structural diagram of a refund processing device based on medical information provided by an embodiment of the present invention.

[0210] As an example, the refund processing device based on medical information may include:

[0211] The acquisition module 201 is used to obtain medical information after obtaining the user's refund request. The medical information is medication, diagnosis and payment information obtained in real time from different medical systems through the unified data center;

[0212] A detection module 202 is configured to, after determining that the user meets the refund conditions based on the medical information, perform anomaly detection based on the medical information, wherein the anomaly detection is a detection process that uses the medical information to determine whether the user's refund behavior is abnormal;

[0213] The refund processing module 203 is used to perform a refund process based on the medical information if the detection result of the abnormality detection is normal.

[0214] Optionally, performing abnormality detection based on the medical information includes:

[0215] Calling a preset LSTM model to calculate an abnormality probability value based on the medical information and using an isolation forest algorithm to calculate an abnormality score value based on the medical information;

[0216] Whether the user's refund behavior is abnormal is determined based on the abnormal probability value and the abnormal score value.

[0217] Optionally, determining whether the user's refund behavior is abnormal based on the abnormal probability value and the abnormal score value includes:

[0218] If the abnormal probability value is less than the first preset probability value and the abnormal score value is less than the first preset score value, it is determined that the user's refund behavior is normal;

[0219] If the abnormal probability value is greater than the first preset probability value and the abnormal score value is greater than the first preset score value, it is determined that the user's refund behavior is abnormal.

[0220] Optionally, the device further comprises:

[0221] a risk classification module, configured to, after the step of performing anomaly detection based on the medical information, determine the risk category of the user's refund behavior based on the anomaly probability value and the anomaly score value if the detection result of the anomaly detection is abnormal;

[0222] The category processing module is used to perform refund processing, termination processing or alarm processing based on the risk category.

[0223] Optionally, the refund processing, termination processing, or warning processing based on the risk category includes:

[0224] If the risk category is high risk, an alarm process is triggered after intercepting and recording the medical information;

[0225] If the risk category is medium risk, manual review will be conducted and a refund will be made if the manual review is passed, or the application will be terminated if the manual review is not passed;

[0226] If the risk category is low risk, a refund will be processed.

[0227] Optionally, determining the risk category of the user's refund behavior according to the abnormal probability value and the abnormal score value includes:

[0228] If the abnormal probability value is greater than the first preset probability value, the abnormal probability value is less than the second preset probability value, the abnormal score value is greater than the first preset score value, and the abnormal score value is less than the second preset score value, then the risk category of the user's refund behavior is determined to be low risk;

[0229] If the abnormal probability value is greater than the second preset probability value, the abnormal probability value is less than the third preset probability value, the abnormal score value is greater than the second preset score value, and the abnormal score value is less than the third preset score value, then the risk category of the user's refund behavior is determined to be medium risk;

[0230] If the abnormal probability value is greater than the third preset probability value and the abnormal score value is greater than the third preset score value, it is determined that the risk category of the user's refund behavior is high risk.

[0231] Optionally, the device further comprises:

[0232] A display list module, configured to generate a refund list using the medical information and display the refund list to the user through a preset digital human before the step of performing abnormality detection based on the medical information;

[0233] The confirmation module is used to perform abnormality detection based on the medical information after receiving the user's confirmation information through the preset digital human.

[0234] Optionally, the device further comprises:

[0235] The rejection prompt module is used to display a rejection refund prompt and rejection refund items to the user after determining that the user does not meet the refund conditions based on the medical information after the step of obtaining medical information.

[0236] The present invention also provides an automatic refund machine, see Figure 4 , which shows a structural diagram of an automatic refund machine provided by one embodiment of the present invention.

[0237] As an example, the automatic refund machine may include: an automatic refund system and a device body, wherein the automatic refund system is arranged in the device body, and the automatic refund system is applicable to the refund processing method based on medical information as described in the above embodiment.

[0238] Those skilled in the art can clearly understand that, for the sake of convenience and brevity, the specific working process of the device described above can refer to the corresponding process in the aforementioned method embodiment and will not be repeated here.

[0239] Furthermore, an embodiment of the present application also provides an electronic device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the refund processing method based on medical information as described in the above embodiment is implemented.

[0240] Furthermore, an embodiment of the present application also provides a computer-readable storage medium, which stores a computer-executable program, and the computer-executable program is used to enable a computer to execute the refund processing method based on medical information as described in the above embodiment.

[0241] In the description of the embodiments of the present invention, it should be noted that the orientation or positional relationship indicated by the terms "upper" and "lower" is based on the orientation or positional relationship shown in the accompanying drawings. It is only for the convenience of describing the embodiments of the present invention and simplifying the description, and does not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation. Therefore, it should not be understood as a limitation of the present invention. When an element such as a layer, region or substrate is referred to as being "on" or "above" another element, it can be directly on the other element, or there can be an intermediate element. In contrast, when an element is referred to as being "directly on" or "above" another element, there are no intermediate elements. It should also be understood that when an element is referred to as being "under" or "below" another element, it can be directly under or below the other element, or there can be an intermediate element. In contrast, when an element is referred to as being "directly under" or "below" another element, there are no intermediate elements. Unless otherwise specified or limited, the terms "mounted," "connected," and "connected" should be interpreted broadly. For example, they can refer to fixed, removable, or integral connections; mechanical or electrical connections; direct or indirect connections through an intermediary; and internal communication between two components. Those skilled in the art will understand the specific meanings of these terms in the present invention based on specific circumstances.

[0242] Those skilled in the art will appreciate that the embodiments of the present application may also provide computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0243] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), apparatuses and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0244] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0245] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0246] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A refund processing method based on medical information, characterized in that: The method comprises: After receiving the user's refund request, obtain medical information, which includes medication, diagnosis, and payment information obtained in real time from different medical systems through a unified data center; After determining that the user meets the refund conditions based on the medical information, performing anomaly detection based on the medical information, wherein the anomaly detection is a detection process that uses the medical information to determine whether the user's refund behavior is abnormal; If the abnormality detection result is normal, a refund will be processed based on the medical information.

2. The refund processing method based on medical information according to claim 1, characterized in that: The performing abnormality detection based on the medical information includes: Calling a preset LSTM model to calculate an abnormality probability value based on the medical information and using an isolation forest algorithm to calculate an abnormality score value based on the medical information; Whether the user's refund behavior is abnormal is determined based on the abnormal probability value and the abnormal score value.

3. The refund processing method based on medical information according to claim 2, characterized in that: The determining whether the user's refund behavior is abnormal according to the abnormal probability value and the abnormal score value includes: If the abnormal probability value is less than the first preset probability value and the abnormal score value is less than the first preset score value, it is determined that the user's refund behavior is normal; If the abnormal probability value is greater than the first preset probability value and the abnormal score value is greater than the first preset score value, it is determined that the user's refund behavior is abnormal.

4. The refund processing method based on medical information according to claim 2, characterized in that: After the step of performing abnormality detection based on the medical information, the method further includes: If the detection result of the anomaly detection is abnormal, determining the risk category of the user's refund behavior according to the abnormal probability value and the abnormal score value; Refund processing, termination processing or warning processing is performed based on the risk category.

5. The refund processing method based on medical information according to claim 4, characterized in that: The refund processing, termination processing or warning processing based on the risk category includes: If the risk category is high risk, an alarm process is triggered after intercepting and recording the medical information; If the risk category is medium risk, manual review will be conducted and a refund will be made if the manual review is passed, or the application will be terminated if the manual review is not passed; If the risk category is low risk, a refund will be processed.

6. The refund processing method based on medical information according to claim 4, characterized in that: The determining the risk category of the user's refund behavior according to the abnormal probability value and the abnormal score value includes: If the abnormal probability value is greater than the first preset probability value, the abnormal probability value is less than the second preset probability value, the abnormal score value is greater than the first preset score value, and the abnormal score value is less than the second preset score value, then the risk category of the user's refund behavior is determined to be low risk; If the abnormal probability value is greater than the second preset probability value, the abnormal probability value is less than the third preset probability value, the abnormal score value is greater than the second preset score value, and the abnormal score value is less than the third preset score value, then the risk category of the user's refund behavior is determined to be medium risk; If the abnormal probability value is greater than the third preset probability value and the abnormal score value is greater than the third preset score value, it is determined that the risk category of the user's refund behavior is high risk.

7. The refund processing method based on medical information according to any one of claims 1 to 6, characterized in that: Before the step of performing abnormality detection based on the medical information, the method further includes: Generate a refund list using the medical information and display the refund list to the user through a preset digital person; After receiving the user's confirmation information through the preset digital human, abnormality detection is performed based on the medical information.

8. The refund processing method based on medical information according to any one of claims 1 to 6, characterized in that: After the step of obtaining medical information, the method further includes: After determining that the user does not meet the refund conditions based on the medical information, a refund rejection prompt and refund rejection items are displayed to the user.

9. A refund processing device based on medical information, characterized in that: The device comprises: An acquisition module is used to obtain medical information after receiving a user's refund request. The medical information is medication, diagnosis, and payment information obtained in real time from different medical systems through a unified data center. a detection module, configured to, after determining that the user meets the refund conditions based on the medical information, perform anomaly detection based on the medical information, wherein the anomaly detection is a detection process that uses the medical information to determine whether the user's refund behavior is abnormal; The refund processing module is used to perform a refund process based on the medical information if the detection result of the abnormality detection is normal.

10. An automatic refund machine, characterized in that: The automatic refund machine includes: an automatic refund system and a device body, the automatic refund system is arranged in the device body, and the automatic refund system is applicable to the refund processing method based on medical information as described in any one of claims 1-8.

Images