Fine health follow-up visit real-time evidence storage system and method based on robot process automation

Abstract

The invention discloses a precise follow-up visit real-time evidence storage system and method based on robot process automation, and belongs to the technical field of information processing, and the method comprises the steps: carrying out the calculation based on first original data to obtain a scheduling score, sorting patients based on the scheduling score to obtain a priority sequence, and outputting a task structure based on the priority sequence; performing multi-modal data acquisition on the patient in the follow-up visit process, performing hash encryption on second original data to obtain a first encrypted fingerprint, verifying the second original data, and constructing a semantic consistency score and a behavior rationality score; reporting the priority based on the semantic consistency score and the behavior rationality score, and performing task simulation execution and data entry; performing behavior compliance verification and data consistency judgment, and if both the behavior compliance verification and the data consistency verification pass and are successfully returned, performing task closed loop; otherwise, re-checking is triggered. According to the scheme, the credibility and the execution degree of follow-up visit of mental disorders are improved through fusion of process chain optimization and data credibility modeling.

Description

Technical Field

[0001] The present invention belongs to the field of information processing technology, and in particular relates to a real-time evidence storage system and method for follow-up of health care based on robotic process automation. Background Art

[0002] Currently, community follow-up tasks for patients with mental disorders in my country are generally plagued by high execution intensity, fragmented workflows, cumbersome data entry, and a lack of consistent verification. In accordance with national public health requirements, physicians are required to regularly complete patient risk level assessments, disease progression records, medication compliance assessments, and management recommendations, and regularly enter these data into provincial information platforms. However, in current practice, task generation relies on manual judgment by physicians, which can easily miss high-risk cases. Furthermore, follow-up data is complex, often containing multimodal information such as audio, images, and forms. This data often needs to be manually entered into the platform item by item at a later stage, which is extremely inefficient.

[0003] In addition, since most provincial platforms do not provide standardized interfaces, automated integration is difficult to achieve. Doctors can only upload data manually through copying and pasting, which is prone to errors or uneven data quality. In the data collection process, the answers of patients with mental disorders often contain logical conflicts, emotional deviations or positioning distortions. Doctors need to rely on subjective experience to judge the credibility of information, and lack a systematic interactive process verification mechanism. Core issues such as whether the follow-up scene actually occurred, whether the images were taken on the spot, and whether the voice is from the patient himself cannot be confirmed through the existing data structure, resulting in a lack of regulatory measures and difficulty in timely detection of forged or formalized follow-up issues. In addition, there is a lack of effective behavioral verification and data traceability mechanisms after the task is executed, and the follow-up chain cannot be closed. In addition, most systems do not support offline collection and offline synchronization, which limits the feasibility of the program promotion in grassroots areas.

[0004] To this end, we propose a real-time evidence storage system and method for Jingkang follow-up based on robotic process automation to solve the above problems. Summary of the Invention

[0005] The purpose of the present invention is to solve the problems of low efficiency and low credibility in the existing technology, and to propose a real-time evidence storage system and method for Jingkang follow-up based on robotic process automation.

[0006] In order to achieve the above object, the present invention adopts the following technical solutions:

[0007] The real-time evidence storage method for Jingkang follow-up based on robotic process automation includes:

[0008] S1: Input first original data, which includes patient risk level, historical core symptom score sequence, task fulfillment rate, and doctor task schedulability score;

[0009] Constructing a scheduling scoring function, wherein the scheduling scoring function obtains a scheduling score by performing a weighted calculation on the first original data; sorting the patients from high to low based on the scheduling score, taking the top K positions to form a priority sequence; and assigning the patients in the priority sequence in order according to the doctor task schedulability score;

[0010] Outputting a task structure based on the priority sequence;

[0011] S2: During the follow-up, multimodal data collection is performed on the patient to obtain second original data, where the second original data includes location information, audio information, image information, time information, and structured information corresponding to the task form;

[0012] Constructing a trusted weight modified hash function, inputting the scheduling score and the second original data; obtaining a first encrypted fingerprint;

[0013] The trusted weight modified hash function introduces a collection trusted weight term;

[0014] Outputting a corresponding collection credibility level based on the collection credibility weight item;

[0015] S3: Perform an integrity check on the second original data, and call the trusted weight modified hash function to calculate the second encrypted fingerprint, and verify the consistency of the first encrypted fingerprint and the second encrypted fingerprint; if they are consistent, proceed to the subsequent process; otherwise, mark it as data corruption;

[0016] Verifying the structured information corresponding to the task form in the second original data according to the task form configuration in the task structure, and calculating a semantic consistency score, where the semantic consistency score is the ratio of the number of non-violated rules to the total number of rules;

[0017] Constructing a behavior rationality score, wherein the behavior rationality score is calculated by calculating the total task duration and the overlap of image and audio acquisition;

[0018] S4: Calculate the reporting priority based on the semantic consistency score and the behavior rationality score, and arrange all tasks to be reported in descending order according to the reporting priority to obtain a sequence to be processed;

[0019] Conduct task simulation execution and data entry;

[0020] Construct a feedback verification function, which determines whether the platform response text is successful by taking a screenshot and identifying it through OCR; and outputs the judgment result;

[0021] S5: Performing behavior compliance verification and data consistency judgment. The behavior compliance verification is performed by comparing the actual execution time of the follow-up task with the positioning coordinates to determine whether the task is completed within the specified time and geographical range; the data consistency verification is performed by comparing the first encrypted fingerprint and the second encrypted fingerprint;

[0022] If both the behavior compliance check and the data consistency check pass and the platform successfully receives a receipt, the task is closed; otherwise, a review is triggered.

[0023] Preferably, the task structure includes patient number, assigned doctor number, scheduled time, recommended follow-up method, system automatic reminder setting, and task form configuration.

[0024] Preferably, the structured information corresponding to the task form in the second original data is a structured information set generated by a doctor filling in a standard structured form corresponding to the task form configuration in the task structure.

[0025] Preferably, the trusted weight modified hash function first uses SHA-256 and then uses AES-256 to encrypt the second original data.

[0026] Preferably, the collection trust weight items include the patient's voice and intonation abnormality index, the structured question and answer logic consistency score, and the position abnormality penalty; the patient's voice and intonation abnormality index is generated by the emotional state classifier after extracting the MFCC features using the terminal's built-in voice processing module; the structured question and answer logic consistency score is obtained by checking whether the answer is continuous and whether the previous and subsequent logic are self-consistent; the position abnormality penalty is determined by whether the collection location is in a preset legal area, and is 1 when the deviation exceeds the preset range, otherwise it is 0.

[0027] Preferably, the acquisition overlap indicates the degree of overlap between the picture and audio acquisition time, and is 1 if the picture is taken during the audio recording, and 0 otherwise.

[0028] Preferably, the task simulation execution includes the following steps:

[0029] Open the provincial platform webpage and locate the input box and task entrance;

[0030] Enter the patient information in the second raw data and navigate to the corresponding task page;

[0031] Fill form fields: automatically paste patient information fields into the form through the locally maintained field mapping table;

[0032] Upload attachments: Audio files and images are uploaded by RPA according to the file path;

[0033] Address positioning: Parse the coordinates into address text and fill it into the follow-up location field;

[0034] After submission, capture the feedback information and take screenshots for archiving. If it fails, record it in the exception pool.

[0035] The real-time evidence storage system for Jingkang follow-up based on robotic process automation is characterized by:

[0036] A task generation module is configured to input first raw data, the first raw data including a patient risk level, a historical core symptom score sequence, a task fulfillment rate, and a doctor's task schedulability score; construct a scheduling score function, the scheduling score function obtaining a scheduling score by performing a weighted calculation on the first raw data; sort the patients from high to low based on the scheduling score, and take the top K patients to form a priority sequence; assign the patients in the priority sequence in sequence according to the doctor's task schedulability score; and output a task structure based on the priority sequence;

[0037] A data acquisition module configured to acquire multimodal data from a patient during a follow-up visit to obtain second original data, the second original data including location information, audio information, image information, time information, and structured information corresponding to a task form; construct a trusted weight-corrected hash function, input the scheduling score and the second original data; obtain a first encrypted fingerprint; introduce a collection trusted weight term into the trusted weight-corrected hash function; and output a corresponding collection trusted level based on the collection trusted weight term.

[0038] a data verification module configured to perform an integrity check on the second original data, and to call the trusted weight modified hash function to calculate a second encrypted fingerprint, and verify the consistency of the first encrypted fingerprint and the second encrypted fingerprint; if they are consistent, the subsequent process is entered; otherwise, it is marked as data corruption; the structured information corresponding to the task form in the second original data is verified according to the task form configuration in the task structure, and a semantic consistency score is calculated, wherein the semantic consistency score is the ratio of the number of non-violated rules to the total number of rules; and a behavior rationality score is constructed, wherein the behavior rationality score is calculated by calculating the total task duration and the overlap of the image and audio acquisition;

[0039] A task execution module is configured to calculate a reporting priority based on the semantic consistency score and the behavior rationality score, arrange all tasks to be reported in descending order according to the reporting priority, and obtain a sequence to be processed; perform task simulation execution and data entry; construct a feedback verification function, which determines whether the platform response text is successful by taking a screenshot and identifying it through OCR; and output a judgment result;

[0040] The data verification module is configured to perform behavioral compliance verification and data consistency judgment. The behavioral compliance verification determines whether the task is completed within the specified time and geographical range by comparing the actual execution time of the follow-up task with the positioning coordinates; the data consistency verification is completed by comparing the first encrypted fingerprint and the second encrypted fingerprint; if both the behavioral compliance verification and the data consistency verification pass and the platform successfully receives a receipt, the task is closed; otherwise, a review is triggered.

[0041] In summary, the technical effects and advantages of the present invention are as follows: This solution calculates a follow-up urgency score based on structured indicators such as historical risk level, symptom fluctuation, and task fulfillment rate, achieving automatic task prioritization and intelligent matching of doctors. After the task is issued, the doctor uses an offline terminal to perform standardized data collection according to the task template, including multi-modal content such as positioning, recording, taking photos, and filling out forms. Simultaneously, the system extracts voice features and interactive logic indicators in real time to generate a trusted weighted hash fingerprint, ensuring the authenticity and verifiability of the data collection process. Subsequently, an embedded semantic rule engine and behavioral consistency function perform multi-dimensional analysis of the follow-up records, automatically identifying logical conflicts and abnormal collection behaviors, and calculating data trustworthiness scores based on this data to screen high-quality samples. For tasks that pass the screening, the system uses process automation technology to simulate manual operations to complete data reporting to the provincial platform, ensuring full-process standard storage without relying on interfaces. Finally, the system constructs a behavioral verification model to reversely verify the task completion through positioning, timestamps, and image fingerprints. Combined with feedback information, it outputs execution results and review flags, forming a trusted closed loop for follow-up tasks. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 Schematic diagram of the method flow in the present invention;

[0043] Figure 2 Schematic diagram of the system structure of the present invention. DETAILED DESCRIPTION

[0044] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0045] Jingkang: Health management for patients with severe mental disorders.

[0046] like Figure 1 As shown in the figure, the real-time evidence storage method for Jingkang follow-up based on robotic process automation includes:

[0047] S1: Input first original data, which includes patient risk level, historical core symptom score sequence, task fulfillment rate, and doctor task schedulability score;

[0048] Constructing a scheduling scoring function, wherein the scheduling scoring function obtains a scheduling score by performing a weighted calculation on the first original data; sorting the patients from high to low based on the scheduling score, taking the top K positions to form a priority sequence; and assigning the patients in the priority sequence in order according to the doctor task schedulability score;

[0049] Outputting a task structure based on the priority sequence;

[0050] S2: During the follow-up, multimodal data collection is performed on the patient to obtain second original data, where the second original data includes location information, audio information, image information, time information, and structured information corresponding to the task form;

[0051] Constructing a trusted weight modified hash function, inputting the scheduling score and the second original data; obtaining a first encrypted fingerprint;

[0052] The trusted weight modified hash function introduces a collection trusted weight term;

[0053] Outputting a corresponding collection credibility level based on the collection credibility weight item;

[0054] S3: Perform an integrity check on the second original data, and call the trusted weight modified hash function to calculate the second encrypted fingerprint, and verify the consistency of the first encrypted fingerprint and the second encrypted fingerprint; if they are consistent, proceed to the subsequent process; otherwise, mark it as data corruption;

[0055] Verifying the structured information corresponding to the task form in the second original data according to the task form configuration in the task structure, and calculating a semantic consistency score, where the semantic consistency score is the ratio of the number of non-violated rules to the total number of rules;

[0056] Constructing a behavior rationality score, wherein the behavior rationality score is calculated by calculating the total task duration and the overlap of image and audio acquisition;

[0057] S4: Calculate the reporting priority based on the semantic consistency score and the behavior rationality score, and arrange all tasks to be reported in descending order according to the reporting priority to obtain a sequence to be processed;

[0058] Conduct task simulation execution and data entry;

[0059] Construct a feedback verification function, which determines whether the platform response text is successful by taking a screenshot and identifying it through OCR; and outputs the judgment result;

[0060] S5: Performing behavior compliance verification and data consistency judgment. The behavior compliance verification is performed by comparing the actual execution time of the follow-up task with the positioning coordinates to determine whether the task is completed within the specified time and geographical range; the data consistency verification is performed by comparing the first encrypted fingerprint and the second encrypted fingerprint;

[0061] If both the behavior compliance check and the data consistency check pass and the platform successfully receives a receipt, the task is closed; otherwise, a review is triggered.

[0062] The specific steps are as follows:

[0063] Step 1: Intelligent task generation and scheduling

[0064] This step automatically generates personalized follow-up tasks for patients with severe mental disorders. These tasks must include: time window, follow-up method, form template, and physician assignment. Tasks are assigned based on patient condition fluctuations and behavioral manifestations. A comprehensive score is used to prioritize urgent cases, ensuring that limited resources are used for those most in need. This mechanism, based on structured and behavioral data, enables precise and automated task scheduling, a prerequisite for subsequent data collection and archiving.

[0065] The input for this step is the following four items extracted from the system database:

[0066] R i : The patient's current risk level (1–5), filled in by the last follow-up physician in accordance with national standards on the form;

[0067] S i : The past three core symptom score series (such as hallucinations, self-harm, etc., with a unit of 0–10) are derived from the historical follow-up structured form;

[0068] C i : Task fulfillment rate, the number of times the system has completed tasks in the past three times / 3, directly counted from the task record table;

[0069] A j : The doctor's task schedulability score is defined as the number of available time slots per week / total number of time slots for the doctor, and is automatically calculated by the system.

[0070] The system constructs a scheduling scoring function U for each patient i , defined as follows:

[0071] U i =α·R i +β·std(S i )+γ·(1-C i )

[0072] in:

[0073] U i: Scheduling urgency score, used to sort task priorities;

[0074] R i : Current risk level, derived from the doctor's last follow-up visit;

[0075] std(S i ):symptom fluctuation value, S i The standard deviation of the series indicates the volatility of the disease;

[0076] C i : Patient task fulfillment rate, derived from system logs;

[0077] The coefficients α, β, and γ are set to 0.4, 0.4, and 0.2 by default and can be adjusted by the system administrator.

[0078] For example:

[0079] Patient A: R i =4, S i =[6,8,4], C i =0.33, then:

[0080]

[0081] System with U i Sort all patients and take the first k patients to form a high priority queue. Each patient task will be matched to the doctor with the highest idleness (i.e., A j The allocated time period is the earliest one among the remaining available time periods in the current task cycle window.

[0082] Then generate the task structure P i , the fields contain:

[0083] Patient number;

[0084] Assigning a doctor number;

[0085] Recommended follow-up methods (such as R i >4 means home visit, the rest means phone / video call);

[0086] Form template type (such as R i >3Use the enhanced form);

[0087] Booking timestamp;

[0088] System automatic reminder settings (such as T-24 hour push).

[0089] Output:

[0090] P i : Task structure, including scheduling results, executors, time and task form configuration;

[0091] U i : Scheduling score, used for abnormal task management and subsequent analysis.

[0092] This step is achieved by constructing a scheduling score U i , integrating the patient's risk level, symptom volatility and historical compliance, while introducing the doctor's scheduling schedulability as a resource constraint, thus dynamically constructing the optimal follow-up task plan. Output P i A standard task package is provided for subsequent offline data collection. The entire solution features streamlined variables, fully utilized inputs, operational outputs, and clear scoring logic, fully meeting engineering reproducibility requirements and patent inventiveness requirements.

[0093] Step 2: Offline follow-up terminal data collection and evidence storage

[0094] This step is the core link in the entire system closed loop to achieve "on-site data collection" and "real-time trusted evidence storage", which inherits the task structure P generated in step 1. i , multimodal data is collected from patients through offline follow-up terminals carried by doctors, and encrypted data is stored by constructing a weighted fingerprint generation function for mental disorder follow-up scenarios. The innovation of this step lies in not only ensuring the multimodal integrity of the collected content (structured + voice + image + location information), but also designing a credibility adjustment mechanism for "mental illness characteristic data" - such as adding emotional interference terms and interactive stability regularization terms, to achieve dynamic modeling and weighted evidence storage of data credibility weights, effectively avoiding traditional problems such as "distortion of follow-up interaction" and "metaphysical audio and video evidence."

[0095] The doctor enters the task interface through the follow-up terminal (a tablet device with a dedicated App installed), and the system automatically calls out the task structure P i Once the task is started, the terminal will execute the following modular collection process according to the established template:

[0096] Positioning module: automatically collect GPS location information i (WGS-84 format), verify the follow-up location;

[0097] Audio module: Recording at the beginning of the task, audio file A i Save in .m4a format, and the acquisition duration automatically matches the follow-up period;

[0098] Image module: Doctors are required to take at least 2 pictures of the scene environment i (such as medication photos, patient status), and take a photo of the patient's ID card as a basis for two-way identity binding;

[0099] Form module: Display template number T iThe corresponding standard structured form, the doctor fills in all fields in real time to generate a structured information set F i ;

[0100] Timing module: records the above acquisition timestamp sequence T F ,T A ,T V ,T L , used for collection consistency verification.

[0101] Finally, the system aggregates the above information into a collection data packet:

[0102] D i ={F i ,A i ,V i ,L i ,T F ,T A ,T V ,T L ,P i}

[0103] Innovation 1: Constructing a weighted trusted hash generation model for mental disorder follow-up scenarios

[0104] Traditional hash evidence storage methods usually use irreversible encryption functions, while this invention introduces a "dynamic correction mechanism for collection quality" to construct the following trusted weight-corrected hash function:

[0105] Z i =Hash AES256 (D i ||U i Q i )

[0106] in:

[0107] Z i : The encrypted fingerprint generated in the end serves as the unique index for the entire process evidence storage;

[0108] Hash AES256 (·): Encrypt the result using SHA-256 and then AES-256.

[0109] D i : Original collection data packet;

[0110] U i : The scheduling urgency score generated in step 1;

[0111] Q i : Collect trust weight items, taking into account factors such as emotional fluctuations, response consistency, and location anomaly penalties during the interaction process. The specific definitions are as follows:

[0112] Q i =λ1·E i +λ2·S i +λ3·(1-δ(L i ))

[0113] in:

[0114] E i : The patient's voice intonation abnormality index is generated by extracting MFCC features using the terminal's built-in voice processing module and then passing it through the emotional state classifier (pre-set model), ranging from 0 to 1;

[0115] S i : Score for logical consistency of structured questions and answers, which systematically checks whether the answers are continuous and logically consistent (e.g., “I took my medication on time” but “my symptoms worsened significantly”);

[0116] δ(L i ): Whether the collection location is within the preset legal area. If it deviates from the patient's registered home address by more than 1 km, it is 1, and normally it is 0;

[0117] λ1,λ2,λ3: are weighted coefficients of three influencing factors. The default values ​​are 0.3, 0.4, and 0.3, respectively. Background policy adjustment is supported.

[0118] This model constructs quantitative penalties for common interference items (such as cognitive bias and mood swings) in the follow-up of patients with mental disorders, effectively improving the credibility level of the evidence content. i =2.4, E i =0.6, S i =0.8,δ(L i )=0, then:

[0119] Q i =0.3·0.6+0.4·0.8+0.3·1=0.18+0.32+0.3=0.8

[0120] final:

[0121] Z i =Hash AES256 (D i ||2.4·0.8)

[0122] At this time, the hash result is dynamically coupled with the task urgency and collection quality, and has complete anti-tampering capabilities.

[0123] Innovation 2: Coupling of evidence-level marking and subsequent task return mechanism

[0124] The system will Z i Simultaneously mark the trust level CRED when generating i =Ui Q i , for subsequent regulatory reflux, such as CRED i Tasks with a score <1.5 can be screened by the backend system and entered into the "manual review pool" for the next cycle, thus achieving a closed loop of task scheduling and data trust.

[0125] Output

[0126] Z i : The final evidence hash fingerprint serves as the primary key for subsequent platform storage and regulatory verification;

[0127] CRED i : Collect confidence levels for downstream evaluation and redistribution of abnormal tasks;

[0128] D i : Completely collect data packets and push them to the background management platform when connected to the Internet.

[0129] Step 3: Semantic consistency analysis and data verification

[0130] The core goal of this step is to collect and encrypt the multimodal follow-up data D in the previous stage. i Perform semantic consistency checks and anomaly screening before structured modeling to ensure data quality, semantic coherence, and behavioral credibility, providing a reliable data foundation for subsequent structured modeling and task archiving. This step is the "filtering threshold" of the entire follow-up data processing process, and its design logic should strictly focus on the three goals of "removing false data, identifying abnormal interactions, and enhancing data credibility."

[0131] Data consistency analysis is divided into two parts:

[0132] 1. Data packet integrity and fingerprint consistency verification (basic verification)

[0133] The system first i The internal fields are checked for integrity one by one to ensure that key fields are not missing, and the encryption function is called to recalculate the hash value of the data packet. Verify it with the original evidence fingerprint Z i Is it consistent?

[0134]

[0135] like Immediately mark the data as "data integrity violation" and output flag i =False, the data will not enter the subsequent process.

[0136] For example: If the user illegally tampers with F i Even if Z is not updated when the certificate is saved, i, here the hash comparison can accurately identify it.

[0137] 2. Semantic consistency evaluation based on template structure

[0138] For each follow-up form F i , the system is based on its task structure P i The template number T provided in i Call the corresponding rule engine Perform combined validation on fields. Each rule is a formal logic or conditional constraint, for example:

[0139] If medication compliance = poor, then the disease control should not be good;

[0140] If there is an attack tendency = yes, then the risk level ≤ 2 is considered a logical conflict.

[0141] Assume that n rules are triggered and the number of violations is c, then the semantic consistency score is:

[0142]

[0143] For example: if 2 out of 5 rules trigger a conflict, then This score will serve as an important indicator for subsequent credible data screening.

[0144] 3. Checking the rationality of behavior timing

[0145] To prevent tasks from being perfunctory (e.g. completing a form + recording + photo within 2 minutes), the system calculates the total task execution time ΔT based on the timestamp recorded by the terminal. i , and analyze the image V i With Audio A i The acquisition overlap ρ i , construct the following rationality index:

[0146]

[0147] in:

[0148] ΔT i : The duration from task start to task submission (in seconds), recorded by the terminal;

[0149] ρ i : Indicates the degree of overlap between the image and audio capture time. If the photo is taken during the audio recording, it is 1, otherwise it is 0;

[0150] τ: Smoothing constant, the default setting is 300 seconds, which means that it is recommended that the task be completed in no less than 5 minutes;

[0151] O i : Comprehensive operation rationality score, the closer to 1, the more standardized the task execution.

[0152] For example, if the total duration of the task is 180 seconds and the image and audio are not synchronized, then ρ i =0,O i =0, indicating that the task is very likely to be a false operation.

[0153] Output

[0154] S i : Structural field logical consistency score;

[0155] O i : Behavior rationality score;

[0156] flag i : A flag indicating whether the data has passed consistency and integrity checks, for use in downstream modeling steps.

[0157] Step 4: Automatic task execution and data reporting

[0158] enter:

[0159] S i : Structured field consistency score;

[0160] O i : Behavior rationality score;

[0161] flag i : Indicates whether the data has passed verification. True means it can be reported;

[0162] D i :Complete multimodal data package after verification (including F i ,A i ,V i ,L i ,P i );

[0163] Z i : The encrypted evidence fingerprint generated by the data packet is used for logging and backtracking;

[0164] This step consists of three parts:

[0165] 1. Task Execution Priority Sorting

[0166] In order to reasonably schedule system execution resources, it is necessary to first calculate the priority W of the task based on the score. i , so that high-quality data is processed first. The calculation formula is:

[0167] W i =α·S i +β·O i

[0168] in:

[0169] α=0.6,β=0.4, which is the best value verified by experiments;

[0170] W i ∈[0,1], the larger the value, the higher the priority;

[0171] Example: If S i =0.8,O i =0.7, then W i =0.6·0.8+0.4·0.7=0.76, higher priority.

[0172] Press W to report all tasks i Arrange in descending order to form a queue to be processed

[0173] 2. Automatic Task Simulation Execution and Data Entry

[0174] The process of manually logging into the provincial platform is simulated based on RPA (such as UiPath or Python+PyAutoGUI), which includes the following steps:

[0175] Open the provincial platform webpage and locate the input box and task entrance;

[0176] Enter patient information (from F i ) and navigate to the corresponding task page;

[0177] Fill in form fields: Use the locally maintained "field mapping table" to fill in form fields. i The fields are automatically pasted into the form;

[0178] Upload attachment: A i Audio files, V i The image is uploaded by RPA according to the file path (the upload is located in the central evidence server);

[0179] Address positioning: L i The coordinates are parsed into address text to fill in the "Follow-up Location" field;

[0180] After submission, capture the feedback information and take screenshots for archiving. If it fails, record it in the exception pool.

[0181] This process does not call the platform API, but 100% simulates mouse / keyboard interaction, bypassing interface limitations and achieving universal deployability.

[0182] 3. Reporting feedback verification and backtracking log generation

[0183] In order to improve the maintainability of the system, a feedback verification function R is constructed i :

[0184]

[0185] This function determines whether the screenshot OCR recognizes the platform response text successfully. The judgment is based on the rule library (such as keywords such as "task completed" and "saved").

[0186] The system will Z i 、R i , current timestamp, execution log, and operation screenshot path are written into the local SQLite log table together to form a complete evidence chain for the management background to review or compare.

[0187] Step 5: Follow-up task execution and result confirmation

[0188] The core of this step is to verify system behavior and data consistency, outputting clear indicators of "follow-up task completion" and "whether a review has been triggered." It is a key component in achieving full-process traceability and regulatory compliance, ensuring that data is not only submitted but also truly reflects on-site execution, eliminating issues such as "formal reporting" and "pseudo-execution."

[0189] In order to complete the task execution confirmation, this step introduces two key modules:

[0190] 1. Behavioral Compliance Verification

[0191] By comparing the actual execution time T of the follow-up task i and positioning coordinates L i , to determine whether the task is completed within the specified time and geographical scope. The behavior scoring function is defined as follows:

[0192] B i =δ1·in_time(T i )+δ2·in_geo(L i )

[0193] in:

[0194] in_time(T i )∈{0,1}:If T i If the time window is within the specified time window (e.g. 08:00–20:00), it is 1, otherwise it is 0;

[0195] in_geo(L i )∈{0,1}:If L i If it falls within a 100-meter radius of the designated community, the value is 1, otherwise it is 0;

[0196] The default value is δ1 = δ2 = 0.5, which can be fine-tuned according to the actual management strategy;

[0197] If B i=1, indicating that the behavior is fully compliant.

[0198] Example: The doctor completes the follow-up at 14:30 within the community center coordinates (23.123456,113.123456), meeting the time and location conditions. i =1.

[0199] 2. Data consistency verification and final judgment

[0200] Hash fingerprint Z of the data before uploading i and log i The hash H generated by the screenshot file i Compare and determine whether the data is consistent:

[0201] H i =SHA256(screenshot_img i ), screenshots are from log i ;

[0202] If Z i =H i , indicating that the data source is reliable and has not been modified.

[0203] The final task confirmation function is defined as:

[0204]

[0205] If the task behavior is compliant, the data is consistent and the platform successfully receives a receipt, that is, C i =1, the task is closed; otherwise, a review is triggered.

[0206] Output

[0207] C i : Whether the task is executed in a closed loop, 1 means completed, 0 means failed;

[0208] review_flag i : Whether to enter the review process, defined as 1-C i .

[0209] The technical solutions in the above-mentioned embodiments of the present application have at least the following technical effects or advantages: This solution calculates the follow-up urgency score based on structured indicators such as historical risk level, symptom fluctuation, and task fulfillment rate, thereby achieving automatic task prioritization and intelligent matching of doctors. After the task is issued, the doctor uses an offline terminal to perform standardized data collection according to the task template, including multi-modal content such as positioning, recording, taking pictures, and filling out forms. At the same time, the system extracts voice features and interaction logic indicators in real time to generate a trusted weighted hash fingerprint, ensuring the authenticity and verifiability of the data collection process. Subsequently, the follow-up records are analyzed from multiple dimensions using an embedded semantic rule engine and behavioral consistency function to automatically identify logical conflicts and abnormal collection behaviors, and based on this, a data trust score is calculated to screen high-quality samples. For tasks that pass the screening, the system uses process automation technology to simulate manual operations to complete the data reporting to the provincial platform, ensuring the full process standard storage without relying on interfaces. Finally, the system constructs a behavior verification model to reversely compare the task completion through positioning, timestamps, and image fingerprints, and outputs the execution results and review marks based on the feedback information, forming a trusted closed loop for the follow-up task.

[0210] The embodiment of the present application also provides a real-time evidence storage system for Jingkang follow-up based on robotic process automation, such as Figure 2 Shown, including:

[0211] A task generation module is configured to input first raw data, the first raw data including a patient risk level, a historical core symptom score sequence, a task fulfillment rate, and a doctor's task schedulability score; construct a scheduling score function, the scheduling score function obtaining a scheduling score by performing a weighted calculation on the first raw data; sort the patients from high to low based on the scheduling score, and take the top K patients to form a priority sequence; assign the patients in the priority sequence in sequence according to the doctor's task schedulability score; and output a task structure based on the priority sequence;

[0212] A data acquisition module configured to acquire multimodal data from a patient during a follow-up visit to obtain second original data, the second original data including location information, audio information, image information, time information, and structured information corresponding to a task form; construct a trusted weight-corrected hash function, input the scheduling score and the second original data; obtain a first encrypted fingerprint; introduce a collection trusted weight term into the trusted weight-corrected hash function; and output a corresponding collection trusted level based on the collection trusted weight term.

[0213] a data verification module configured to perform an integrity check on the second original data, and to call the trusted weight modified hash function to calculate a second encrypted fingerprint, and verify the consistency of the first encrypted fingerprint and the second encrypted fingerprint; if they are consistent, the subsequent process is entered; otherwise, it is marked as data corruption; the structured information corresponding to the task form in the second original data is verified according to the task form configuration in the task structure, and a semantic consistency score is calculated, wherein the semantic consistency score is the ratio of the number of non-violated rules to the total number of rules; and a behavior rationality score is constructed, wherein the behavior rationality score is calculated by calculating the total task duration and the overlap of the image and audio acquisition;

[0214] A task execution module is configured to calculate a reporting priority based on the semantic consistency score and the behavior rationality score, arrange all tasks to be reported in descending order according to the reporting priority, and obtain a sequence to be processed; perform task simulation execution and data entry; construct a feedback verification function, which determines whether the platform response text is successful by taking a screenshot and identifying it through OCR; and output a judgment result;

[0215] The data verification module is configured to perform behavioral compliance verification and data consistency judgment. The behavioral compliance verification determines whether the task is completed within the specified time and geographical range by comparing the actual execution time of the follow-up task with the positioning coordinates; the data consistency verification is completed by comparing the first encrypted fingerprint and the second encrypted fingerprint; if both the behavioral compliance verification and the data consistency verification pass and the platform successfully receives a receipt, the task is closed; otherwise, a review is triggered.

[0216] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. The real-time evidence storage method for Jingkang follow-up based on robotic process automation is characterized by: include: S1: Input first original data, which includes patient risk level, historical core symptom score sequence, task fulfillment rate, and doctor task schedulability score; Constructing a scheduling scoring function, wherein the scheduling scoring function obtains a scheduling score by performing a weighted calculation on the first original data; sorting the patients from high to low based on the scheduling score, taking the top K positions to form a priority sequence; and assigning the patients in the priority sequence in order according to the doctor task schedulability score; Outputting a task structure based on the priority sequence; S2: During the follow-up, multimodal data collection is performed on the patient to obtain second original data, where the second original data includes location information, audio information, image information, time information, and structured information corresponding to the task form; Constructing a trusted weight modified hash function, inputting the scheduling score and the second original data; obtaining a first encrypted fingerprint; The trusted weight modified hash function introduces a collection trusted weight term; Outputting a corresponding collection credibility level based on the collection credibility weight item; S3: Perform an integrity check on the second original data, and call the trusted weight modified hash function to calculate the second encrypted fingerprint, and verify the consistency of the first encrypted fingerprint and the second encrypted fingerprint; if they are consistent, proceed to the subsequent process; otherwise, mark it as data corruption; Verifying the structured information corresponding to the task form in the second original data according to the task form configuration in the task structure, and calculating a semantic consistency score, where the semantic consistency score is the ratio of the number of non-violated rules to the total number of rules; Constructing a behavior rationality score, wherein the behavior rationality score is calculated by calculating the total task duration and the overlap of image and audio acquisition; S4: Calculate the reporting priority based on the semantic consistency score and the behavior rationality score, and arrange all tasks to be reported in descending order according to the reporting priority to obtain a sequence to be processed; Conduct task simulation execution and data entry; Construct a feedback verification function, which determines whether the platform response text is successful by taking a screenshot and identifying it through OCR; and outputs the judgment result; S5: Performing behavior compliance verification and data consistency judgment. The behavior compliance verification is performed by comparing the actual execution time of the follow-up task with the positioning coordinates to determine whether the task is completed within the specified time and geographical range; the data consistency verification is performed by comparing the first encrypted fingerprint and the second encrypted fingerprint; If both the behavior compliance check and the data consistency check pass and the platform successfully receives a receipt, the task is closed; Otherwise, a review is triggered.

2. The real-time evidence storage method for follow-up of Jingkang based on robotic process automation according to claim 1 is characterized in that: The task structure includes patient number, assigned doctor number, scheduled time, recommended follow-up method, system automatic reminder setting, and task form configuration.

3. The real-time evidence storage method for follow-up of Jingkang based on robotic process automation according to claim 1 is characterized in that: The structured information corresponding to the task form in the second original data is a structured information set generated by a doctor filling in a standard structured form corresponding to the task form configuration in the task structure.

4. The real-time evidence storage method for follow-up of Jingkang based on robotic process automation according to claim 1 is characterized in that: The trusted weight modified hash function first uses SHA-256 and then uses AES-256 to encrypt the second original data.

5. The real-time evidence storage method for follow-up of Jingkang based on robotic process automation according to claim 1 is characterized in that: The collection trust weight items include the patient's voice and intonation abnormality index, the structured question and answer logic consistency score, and the position abnormality penalty; the patient's voice and intonation abnormality index is generated by extracting MFCC features using the terminal's built-in voice processing module and then using the emotional state classifier, ranging from 0 to 1; the structured question and answer logic consistency score is obtained by checking whether the answer is continuous and whether the previous and subsequent logic are self-consistent; The position anomaly penalty is determined by whether the acquisition location is within a preset legal area. If the deviation exceeds the preset range, the value is 1; otherwise, the value is 0.

6. The real-time evidence storage method for follow-up of Jingkang based on robotic process automation according to claim 1 is characterized in that: The acquisition overlap indicates the degree of overlap between the image and audio acquisition times. If the photo is taken during the audio recording, the value is 1; otherwise, the value is 0.

7. The real-time evidence storage method for follow-up of Jingkang based on robotic process automation according to claim 1 is characterized in that: The task simulation execution includes the following steps: Open the provincial platform webpage and locate the input box and task entrance; Enter the patient information in the second raw data and navigate to the corresponding task page; Fill form fields: automatically paste patient information fields into the form through the locally maintained field mapping table; Upload attachments: Audio files and images are uploaded by RPA according to the file path; Address positioning: Parse the coordinates into address text and fill it into the follow-up location field; After submission, capture the feedback information and take screenshots for archiving. If it fails, record it in the exception pool.

8. The real-time evidence storage system for Jingkang follow-up based on robotic process automation is characterized by: include: a task generation module, the task generation module being configured to input first raw data, the first raw data including patient risk level, historical core symptom score sequence, task fulfillment rate, and physician task schedulability score; Constructing a scheduling scoring function, wherein the scheduling scoring function obtains a scheduling score by performing a weighted calculation on the first raw data; sorting the patients from high to low based on the scheduling score, and taking the top K patients to form a priority sequence; assigning the patients in the priority sequence in sequence according to the doctor task schedulability score; and outputting a task structure based on the priority sequence; a data acquisition module configured to collect multimodal data from the patient during the follow-up process to obtain second original data, wherein the second original data includes position information, audio information, image information, time information, and structured information corresponding to the task form; Constructing a trusted weight modified hash function, inputting the scheduling score and the second original data; obtaining a first encrypted fingerprint; introducing a collection trusted weight term into the trusted weight modified hash function; Outputting a corresponding collection credibility level based on the collection credibility weight item; a data verification module configured to perform an integrity check on the second original data, and to call the trusted weight modified hash function to calculate a second encrypted fingerprint, and verify the consistency of the first encrypted fingerprint and the second encrypted fingerprint; if they are consistent, the subsequent process is entered; otherwise, it is marked as data corruption; the structured information corresponding to the task form in the second original data is verified according to the task form configuration in the task structure, and a semantic consistency score is calculated, wherein the semantic consistency score is the ratio of the number of non-violated rules to the total number of rules; and a behavior rationality score is constructed, wherein the behavior rationality score is calculated by calculating the total task duration and the overlap of the image and audio acquisition; A task execution module is configured to calculate a reporting priority based on the semantic consistency score and the behavior rationality score, arrange all tasks to be reported in descending order according to the reporting priority, and obtain a sequence to be processed; perform task simulation execution and data entry; construct a feedback verification function, which determines whether the platform response text is successful by taking a screenshot and identifying it through OCR; and output a judgment result; The data verification module is configured to perform behavioral compliance verification and data consistency judgment. The behavioral compliance verification determines whether the task is completed within the specified time and geographical range by comparing the actual execution time of the follow-up task with the positioning coordinates; the data consistency verification is completed by comparing the first encrypted fingerprint and the second encrypted fingerprint; if both the behavioral compliance verification and the data consistency verification pass and the platform successfully receives a receipt, the task is closed; otherwise, a review is triggered.

Images