Medical service process iterative decision method and system based on multi-source feedback information

Abstract

The application discloses a medical service process iterative decision method and system based on multi-source feedback information, and the method steps comprise the following steps: in step S01, multi-source feedback information of each link of a medical service process is acquired, and service satisfaction scores and service quality scores are obtained; in step S02, the medical service process is modeled as a dynamic Bayesian network model, an emotion value after state transition is calculated by using an emotion transmission function, and a quantitative transmission effect is obtained by using an emotion transmission coefficient, and a time sequence cascade model is trained; in step S03, according to the deviation between expert evaluation benchmark data and the satisfaction scores, the feedback information weight is adjusted; in step S04, a multi-objective optimization function is iteratively solved, and in the solving process, the time sequence cascade model is used to predict the satisfaction change under different medical service processes, and an optimal medical service process scheme is obtained. The application can optimize the service efficiency and resource utilization rate of medical services, and improve the medical experience of patients.

Description

Technical Field

[0001] This invention relates to the field of intelligent medical decision-making technology, and in particular to an iterative decision-making method and system for medical service processes based on multi-source feedback information. Background Technology

[0002] The medical service process involves multiple stages, such as triage, registration, waiting, consultation, examination, diagnosis, treatment, and follow-up. Different service stages result in varying service efficiency, resource utilization, and patient experience. Optimizing the medical service process can improve service efficiency, resource utilization, and patient experience. Current technologies for optimizing the medical service process typically follow this general procedure: First, patient feedback is collected through relatively simple channels, such as paper questionnaires (patients fill out after their visit to express their feelings about the medical service) or simple online rating methods (e.g., setting up a rating portal on the hospital's website). Then, the collected feedback is manually organized and analyzed. Staff read the questionnaires or online ratings, summarize the main problems raised by patients, and finally, based on the manually summarized problems and experience, develop improvement plans.

[0003] However, healthcare services follow a strict time-series process, from triage, registration, waiting, consultation, examination, diagnosis to treatment and follow-up. These stages not only have a temporal sequence but also exhibit strong emotional transmission and quality cascading effects. For example, a poor experience during registration can lead to a general decline in patient satisfaction in subsequent stages; this "halo effect" is particularly pronounced in healthcare scenarios. Traditional manual-based healthcare process optimization methods analyze each stage independently, ignoring the temporal dependencies and cascading effects between them. This fails to capture the temporal transmission characteristics of each stage, resulting in poor actual optimization outcomes.

[0004] Furthermore, traditional human-based methods for optimizing healthcare service processes rely directly on collected patient feedback for analysis. However, the healthcare field suffers from severe information asymmetry, where patients often cannot accurately assess the core elements of healthcare quality (such as diagnostic accuracy and the rationality of treatment plans) and instead focus excessively on superficial services (such as doctor's attitude and environmental comfort). This leads to a systematic bias in patient feedback, namely, a lack or distortion of feedback on core healthcare quality while overemphasizing feedback on ancillary services. Such feedback bias caused by information asymmetry further affects the actual effectiveness of healthcare service process optimization. Summary of the Invention

[0005] The technical problem to be solved by this invention is: In view of the technical problems existing in the prior art, this invention provides a medical service process iterative decision-making method and system based on multi-source feedback information. By modeling the temporal dependencies and cascading effects between various links in the medical service process, it can accurately improve the optimization effect of the medical service process, so as to optimize the service efficiency and resource utilization of medical services as much as possible, thereby improving the patient's medical experience.

[0006] To solve the above-mentioned technical problems, the technical solution proposed by this invention is as follows:

[0007] An iterative decision-making method for healthcare service processes based on multi-source feedback information, comprising the following steps:

[0008] Step S01: Obtain multi-source feedback information: Obtain service satisfaction feedback information and service quality indicator feedback information for each link of the current medical service process, and convert the service satisfaction feedback information into a service satisfaction score and calculate the service quality score based on the service quality indicator feedback information to form a set of feedback information. The service quality indicators include waiting time and service time.

[0009] Step S02, Temporal Cascade Effect Modeling: The medical service process is modeled as a dynamic Bayesian network model. In the model, each stage of the medical service process is used as a state node, and the satisfaction score is used as an observation node. The emotion transfer function is used to calculate the emotion value after the state transition between each state node, and the transmission effect is quantified by the emotion transfer coefficient. The emotion transfer function is a function that calculates the emotion value at the current moment based on the service quality score of each stage and the emotion value at the previous moment. The emotion value is used to represent satisfaction. The dynamic Bayesian network model is trained using a set of feedback information. During the training process, the service satisfaction score is used as the initial emotion value of each stage. After training, a temporal cascade model is obtained to predict the satisfaction of subsequent stages given a preceding stage.

[0010] Step S03, Feedback Deviation Correction: Calculate the deviation between the expert evaluation benchmark data and the service satisfaction scores, adjust the corresponding weights in each service satisfaction score according to the calculated deviation, and obtain the corrected satisfaction scores.

[0011] Step S04, Multi-objective optimization decision: Construct a multi-objective optimization function to maximize the increase in satisfaction, the increase in service efficiency, and the increase in resource cost. Iterate and solve the constructed multi-objective optimization function. During the solution process, based on the corrected satisfaction score and the service quality score, use the time series cascade model to predict the change in satisfaction at each stage under different medical service processes to obtain the increase in satisfaction. After multiple rounds of iterative solution, the optimal medical service process solution is output.

[0012] Furthermore, in step S01, before converting the service satisfaction feedback information into a satisfaction score, the process also includes mapping the service satisfaction feedback information into standard medical terminology. This step includes:

[0013] Step S101. Construct a medical terminology mapping dictionary;

[0014] Step S102. Knowledge Graph Construction: Construct a medical knowledge graph. , This represents entity nodes that include diseases, symptoms, examinations, and medications. Indicates the relationship between entities;

[0015] Step S103. Terminology Disambiguation: Disambiguation processing is performed on the multi-source feedback information set to obtain a preliminary standardized medical terminology set. During the disambiguation process, the constructed medical knowledge graph is used... Find all candidate mappings for the term to be disambiguated in the middle, and through Calculate each candidate mapping found Context similarity score ,in For words With candidate mapping The mutual information between points is used to measure the strength of the association. i Indicates the index of the candidate mapping. C For the context consisting of multiple words before and after the term to be disambiguated, For context C Words within, Indicator The inverse document frequency is used to select a candidate mapping as the standard term for the term to be disambiguated based on the context similarity score.

[0016] Step S104. Calculate the similarity between each medical term in the preliminary standardized medical term set, and cluster each medical term according to the calculated similarity to obtain the final standardized medical term set.

[0017] Step S105. Convert service satisfaction feedback information into standardized medical terms based on the final set of standardized medical terms.

[0018] Furthermore, in step S104, based on the semantic similarity of the knowledge graph... Category similarity and context vector cosine similarity The similarity between two medical terms is obtained by weighted calculation, where the semantic similarity from the knowledge graph is included. according to Calculations show that For two medical terms to be computed in the constructed medical knowledge graph Path distance and category similarity in To assign different values ​​based on whether two medical terms belong to the same category, the context vector cosine similarity is used. It is calculated based on the cosine distance between the context word vectors of the two medical terms to be calculated.

[0019] Furthermore, in step S02, the calculation expression for the emotion transfer function used is:

[0020]

[0021] in, For the link i At the present moment t The sentiment score is used to reflect service satisfaction. For the link i In the previous moment The emotional value is used to reflect the memory effect of emotions. For the link i Service quality rating For the link i The set of preceding stages, To start from the link j To the stage i The emotion transmission coefficient; This is a memory decay factor used to control the degree of influence of historical emotions. Weighting of service quality for the current stage. Weighting of the impact of emotions in the preceding stages;

[0022] The initial emotion values ​​at each stage are obtained from service satisfaction feedback information, and the service quality scores at each stage are obtained from the service quality score feedback information. After training the dynamic Bayesian network model using historical feedback information, the emotion transmission coefficients between each stage are learned. The final time-series cascade model is obtained: ,in, Indicates about , as well as The function, Represents the set of preceding stages The emotional value of all stages in the process. Indicates model parameters.

[0023] Further, step S03 includes:

[0024] Step S301: Preliminary screening: Calculate the information entropy of each satisfaction rating in the service satisfaction rating set, and perform preliminary screening of the service satisfaction rating set based on the information entropy of each rating, removing rating information with information entropy lower than a preset threshold, to obtain a preliminarily adjusted service satisfaction rating set.

[0025] Step S302, Weight Correction: According to the formula Calculate the corrected weighting coefficients for each rating in the service satisfaction rating set. , Indicates the basic weight. This indicates the initial adjusted satisfaction rating. This represents the expert evaluation benchmark value. This indicates the adjustment parameter used to control the rate of weight decay;

[0026] Step S303, Correction Output: Use the corrected weighting coefficients The satisfaction score is obtained after correction to account for the bias.

[0027] Furthermore, in step S04, the calculation expression of the constructed multi-objective optimization function is as follows:

[0028]

[0029] in, , , , as well as Indicates the weighting coefficient. Indicates the increase in satisfaction. Indicates the increase in cost. Indicates the amount of efficiency improvement. Indicates the specified evaluation indicators. This indicates the penalty for violating constraints, which includes hard constraints and soft constraints. The hard constraints include the minimum treatment time constraint. , Indicates the actual treatment time. Indicates the shortest treatment time. The safety factor is represented by the soft constraint, which includes workload constraints, equipment utilization constraints, and cost control constraints. The weight coefficient of the hard constraint is greater than that of the soft constraint.

[0030] Furthermore, in step S04, an improved genetic algorithm is used to solve the multi-objective optimization function, wherein the mutation probability in the improved genetic algorithm is... Based on individual fitness values Average fitness value of the population The adaptive determination yields the following calculation expression: .

[0031] Furthermore, step S04 is followed by a hierarchical iterative verification step, including:

[0032] Based on the first time period, adjust one or more links where the satisfaction level is lower than the preset threshold, and iterate again to determine the optimal medical service process solution.

[0033] The implementation effect of the optimal medical service process plan is evaluated according to the second time period, and the parameters in the multi-objective optimization function and the parameters of the dynamic Bayesian network model are adaptively adjusted based on the evaluation results of the implementation effect.

[0034] The time-series cascaded model was retrained according to the third time period, and the medical terminology mapping dictionary was updated in order to re-update and evaluate the model.

[0035] The period lengths of the first time period, the second time period, and the third time period increase sequentially.

[0036] Furthermore, based on the evaluation results of the implementation effect, the parameters in the multi-objective optimization function and the parameters of the dynamic Bayesian network model are adaptively adjusted using the Bayesian optimization method. The parameters of the dynamic Bayesian network model include the emotion transfer coefficient in the emotion transfer function and the dynamic Bayesian network structure. The parameters in the multi-objective optimization function include the weight coefficients of each optimization objective and the penalty coefficients in the constraint penalty terms. The implementation effect includes the satisfaction improvement rate, the cascade efficiency improvement degree, and the constraint satisfaction degree. When the satisfaction improvement rate is less than a preset threshold, the emotion transfer coefficient in the emotion transfer function is increased. When the cascade efficiency improvement degree is less than a preset threshold, the dynamic Bayesian network structure is adjusted. When the constraint satisfaction degree is lower than a preset threshold, the penalty coefficient in the constraint penalty terms is increased.

[0037] A computer system includes a processor and a memory, the memory being used to store a computer program, and the processor being used to execute the computer program to perform the iterative decision-making method for a medical service process as described above.

[0038] Compared with the prior art, the advantages of the present invention are as follows:

[0039] 1. This invention models the medical service process as a dynamic Bayesian network model, making full use of the dynamic Bayesian network to characterize the temporal characteristics of the medical service. The model uses an emotion transfer function to calculate the emotion value after the state transition between each state node and quantifies the transfer effect through the emotion transfer coefficient. By quantifying the temporal dependence and emotion transfer effect between each link of the medical service process through the emotion transfer function, it can effectively quantify the impact of the preceding link on the satisfaction of the subsequent link, accurately capture the "halo effect" in the medical service process, and based on the temporal cascade model obtained after training, it can clearly capture the mutual influence pattern between each link in the medical process, accurately analyze how changes in patient satisfaction are transmitted to subsequent links, and thus accurately predict the chain reaction of service links.

[0040] 2. This invention, by combining multi-objective optimization function solutions, can form a complete closed-loop optimization system. It fully utilizes multi-source feedback information and trained time-series cascade models to iteratively solve the medical service process, ultimately determining the optimal medical service process solution that maximizes satisfaction, service efficiency, and minimizes resource costs. This can precisely improve the optimization effect of the medical service process, optimize service efficiency and resource utilization as much as possible, thereby improving the patient's medical experience and achieving precise optimization and continuous improvement of medical services.

[0041] 3. This invention uses expert evaluation benchmark data to adjust and correct the weight of satisfaction feedback information. Compared with traditional feedback analysis methods, it can significantly reduce evaluation bias caused by differences in information understanding, making the data analysis results more consistent with the actual situation of medical services. Attached Figure Description

[0042] Figure 1 This is a schematic diagram illustrating the implementation process of the iterative decision-making method for medical service processes based on multi-source feedback information in this embodiment.

[0043] Figure 2 This is a schematic diagram illustrating the complete implementation process of iterative decision-making in the medical service process in this embodiment. Detailed Implementation

[0044] The present invention will be further described below with reference to the accompanying drawings and specific preferred embodiments, but this does not limit the scope of protection of the present invention.

[0045] like Figure 1 As shown, the steps of the iterative decision-making method for medical service processes based on multi-source feedback information in this embodiment include:

[0046] Step S01: Obtaining multi-source feedback information: Obtain service satisfaction feedback information and service quality indicator feedback information for each link of the current medical service process (including registration, triage, waiting, consultation, examination, diagnosis, treatment, etc.), and convert the service satisfaction feedback information into a service satisfaction score and calculate the service quality score based on the service quality indicator feedback information to form a set of feedback information. Service quality indicators include waiting time and service time, etc.

[0047] Step S02, Temporal Cascade Effect Modeling: The medical service process is modeled as a dynamic Bayesian network model. In the model, each link of the medical service process is used as a state node and the satisfaction score is used as an observation node. The emotion transfer function is used to calculate the emotion value after the state transition between each state node and the transmission effect is quantified by the emotion transfer coefficient. The emotion transfer function is a function that calculates the emotion value at the current moment based on the service quality score of each link and the emotion value at the previous moment. The emotion value is used to represent the satisfaction. The dynamic Bayesian network model is trained using the feedback information set. During the training process, the service satisfaction score of each link is used as the initial emotion value of each link. After training, a temporal cascade model is obtained to predict the satisfaction of subsequent links given the preceding links.

[0048] Step S03, Feedback Deviation Correction: Calculate the deviation between the expert evaluation benchmark data and the service satisfaction scores, adjust the corresponding weights in each service satisfaction score according to the calculated deviation, and obtain the corrected satisfaction scores.

[0049] Step S04, Multi-objective optimization decision: Construct a multi-objective optimization function to maximize the increase in satisfaction, the increase in service efficiency, and the increase in resource cost. Iterate and solve the constructed multi-objective optimization function. During the solution process, use a time series cascade model based on the corrected satisfaction score and service quality score to predict the change in satisfaction at each stage under different medical service processes, and obtain the increase in satisfaction. After multiple rounds of iterative solution, the optimal medical service process solution is output.

[0050] This embodiment models the medical service process as a dynamic Bayesian network, fully utilizing the dynamic Bayesian network to characterize the temporal characteristics of the medical service. The model uses an emotion transfer function to calculate the emotion value after state transitions between each state node and quantifies the transfer effect through an emotion transfer coefficient. By quantifying the temporal dependencies and emotion transfer effects between each stage of the medical service process through the emotion transfer function, it can effectively quantify the impact of preceding stages on the satisfaction of subsequent stages, accurately capturing the "halo effect" in the medical service process. Therefore, based on the temporal cascade model obtained after training, it can clearly capture the mutual influence patterns between each stage of the medical process and accurately analyze how changes in patient satisfaction are transmitted to subsequent stages. It can accurately predict the chain reaction of service links, and given the preceding links, it can accurately predict the satisfaction of subsequent links. Combined with the solution of multi-objective optimization functions, it can form a complete closed-loop optimization system. It makes full use of multi-source feedback information, including satisfaction feedback and service quality scores, as well as trained temporal cascade models to iteratively solve the medical service process. Finally, it decides on the optimal medical service process solution that maximizes the improvement of satisfaction, service efficiency, and minimizes resource costs. It can accurately improve the optimization effect of the medical service process, optimize the service efficiency and resource utilization of medical services as much as possible, thereby improving the patient's medical experience and achieving precise optimization and continuous improvement of medical services.

[0051] At the same time, by using expert evaluation benchmark data to adjust and correct the weight of satisfaction feedback information, compared with traditional feedback analysis methods, the evaluation bias caused by differences in information understanding can be greatly reduced, making the data analysis results more consistent with the actual situation of medical services.

[0052] In a specific application embodiment, step S01 can collect textual descriptions of patients or relevant personnel regarding satisfaction with each stage of the medical service process through various means, as satisfaction feedback information; at the same time, service quality scores for each stage of the medical service process are collected from various feedback platforms, and corresponding timestamp information is obtained synchronously during the collection of the above feedback information.

[0053] Medical texts often suffer from a mixture of technical terms and semantic ambiguity. Satisfaction feedback frequently mixes medical terminology with colloquial expressions, and often exhibits synonyms with different meanings. For example, a patient might say, "I waited two hours for the ultrasound," "The wait for the X-ray was too long," or "The blood draw area was chaotic." These statements relate to different medical examinations, and the same examination may have multiple colloquial names (e.g., "B-ultrasound," "color Doppler ultrasound," and "ultrasound" all refer to ultrasound examinations). General text analysis methods cannot accurately identify these unique medical-specific synonyms, leading to the same problem being categorized differently and affecting the accuracy of problem clustering. To address these issues, step S01 of this embodiment, before converting service satisfaction feedback information into a satisfaction score, includes mapping the service satisfaction feedback text information into standard medical terminology. This step includes:

[0054] Step S101. Construct a medical terminology mapping dictionary;

[0055] Step S102. Knowledge Graph Construction: Construct a medical knowledge graph. , This represents entity nodes that include diseases, symptoms, examinations, and medications. It indicates the relationship between entities, such as "examination items - applicable diseases", "diseases - common symptoms", "medicines - treatment of diseases", etc.

[0056] Step S103. Terminology Disambiguation: Disambiguation processing is performed on the multi-source feedback information set to obtain a preliminary standardized medical terminology set. During the disambiguation process, all candidate mappings of the term to be disambiguated are searched in the constructed medical knowledge graph G, and the results of each candidate mapping are calculated. Context similarity score :

[0057] (1)

[0058] in, For words With candidate mapping Point-to-point mutual information between words, used to measure word... With candidate mapping The strength of the correlation between them i Indicates the index of the candidate mapping. C For the context consisting of multiple words before and after the term to be disambiguated, For context C Words within, Indicator The inverse document frequency is used to select a candidate mapping as the standard term for the term to be disambiguated based on the context similarity score.

[0059] Step S104. Calculate the similarity between each medical term in the preliminary standardized medical term set, and cluster each medical term according to the calculated similarity to obtain the final standardized medical term set.

[0060] Step S105. Convert service satisfaction feedback information into standardized medical terms based on the final set of standardized medical terms.

[0061] In this embodiment, the medical terminology mapping dictionary constructed in step S101 may specifically include a multi-level standardization system, such as the following four-level standardization system:

[0062] First layer: Disease diagnosis terminology mapping;

[0063] Collect and organize common disease names and their correspondence with ICD-10 codes, for example:

[0064] {

[0065] "Common Cold": ["J00", "Acute Nasopharyngitis", "Upper Respiratory Tract Infection"],

[0066] "Diabetes": ["E11", "Type 2 Diabetes", "Diabetes Mellitus"],

[0067] "Hypertension": ["I10", "Primary Hypertension"]

[0068] Second layer: Check the project terminology mapping;

[0069] Establish a mapping between inspection items and their corresponding directory codes, for example:

[0070] {

[0071] "B-mode ultrasound": ["310100001", "Ultrasound examination", "Color Doppler ultrasound", "B-mode ultrasound"],

[0072] "X-ray": ["310300001", "X-ray examination", "X-ray photography", "radiological examination"],

[0073] "CT": ["310400001", "Computed Tomography", "CT Examination"]

[0074] }

[0075] The third layer: Establishing the correspondence between drug names and corresponding drug codes;

[0076] Fourth layer: Establish a standard terminology mapping between department names and medical procedures.

[0077] This embodiment solves the problem of mixed professional terms in medical texts by constructing a multi-level mapping dictionary that covers standard systems such as disease coding, examination item coding, and drug coding.

[0078] In step S102 of this embodiment, a medical knowledge graph is constructed. To achieve terminology disambiguation based on medical knowledge graphs, during the disambiguation process, for each term to be disambiguated... and its context (e.g., terminology) (10 words before and 10 words after the word) Search for the term to be disambiguated in the constructed medical knowledge graph. All possible mappings Then, the context similarity score is calculated according to equation (1). The mapping with the highest score is selected as the term to be disambiguated. The corresponding standard terminology can achieve accurate mapping. Based on the construction of a medical terminology mapping dictionary, this embodiment uses the above-mentioned knowledge graph-based disambiguation method to achieve accurate understanding of medical text and solve the problem of synonyms and variants such as "B-ultrasound / color Doppler ultrasound / ultrasound".

[0079] In step S103 of this embodiment, based on the semantic similarity of the knowledge graph... Category similarity and context vector cosine similarity The similarity between two medical terms can be obtained by weighted calculation, for example, using the following formula. Three-dimensional semantic similarity between two medical terms:

[0080] (2)

[0081] in, For semantic similarity weights, For category similarity weights, For context similarity weights, For semantic similarity in a knowledge graph, reflecting the path distance of terms within a medical knowledge graph, it can be specifically categorized as follows: Calculations show that For two medical terms to be computed in the constructed medical knowledge graph Path distance and category similarity in To determine whether two medical terms belong to the same medical category, different values ​​are assigned based on whether they are in the same category. For example, category similarity is used when the terms are in the same category. The cosine similarity of the context vectors is 1 at most, and not 0 at all. It is calculated based on the cosine distance between the context word vectors of the two medical terms to be calculated.

[0082] This embodiment calculates the semantic similarity between medical terms in three dimensions, integrating semantic, category, and contextual information, as described above. Based on the calculated similarity, the medical terms are clustered to obtain a final standardized medical terminology set. This significantly improves the accuracy of medical terminology recognition, enabling effective understanding of professional concepts expressed in different ways and ensuring high accuracy in information processing. The standardized medical terminology set can be used to standardize real-time service satisfaction feedback information, converting it into standardized medical terms. Patients' colloquial expressions (such as "B-ultrasound" and "filming") can be mapped to standardized medical terms (such as "ultrasound examination" and "X-ray photography") to accurately understand the specific issues raised. Simultaneously, text can be mapped to corresponding categories, grouping synonymous and heterogeneous expressions into the same category. For example, terms like "B-ultrasound," "color Doppler ultrasound," and "ultrasound-related terms" are unified under "ultrasound examination," preventing the same issues from being identified in different categories.

[0083] After converting service satisfaction feedback into standardized medical terminology, sentiment analysis can be performed on the converted standardized medical terminology to extract the sentiment tendency and intensity in the text, thereby converting the textual feedback into a numerical form of satisfaction score. Specifically, a sentiment dictionary or a pre-trained sentiment analysis model can be used to map the text to a satisfaction score of 0-10 based on the sentiment tendency (positive, negative, neutral) and intensity expressed in the service satisfaction text. For example, negative feedback expressing excessively long waiting times, such as "I waited two hours for my ultrasound," can be mapped to a lower satisfaction score (e.g., 2-4 points), while positive feedback such as "The doctor was very patient" can be mapped to a higher satisfaction score (e.g., 7-9 points).

[0084] In step S02 of this embodiment, based on each stage of the medical service process, the medical service process can be modeled as a 7-node dynamic Bayesian network (DBN) to quantitatively analyze the temporal dependence and emotion transfer effect of the medical service. The dependence between stages is characterized by the state transition probability, and the emotion transfer function is used. The system calculates the emotional values ​​after state transitions between each state node and quantifies the transmission effect through the emotional transmission coefficient. This allows for the quantification of the temporal dependence and emotional transmission effect between each stage of the medical service process, thereby capturing the "halo effect" in each stage of the medical service process.

[0085] Specifically, the healthcare service process can be modeled as a dynamic Bayesian network as follows:

[0086] If each step in the medical service process is treated as a state node, then the state node is defined as follows:

[0087]

[0088] Definition of observation node:

[0089] Waiting time and service duration reflect the efficiency of each step and can serve as objective indicators for service quality scoring. This embodiment uses waiting time and service duration as observation nodes to calculate the service quality score. In dynamic Bayesian networks, It is a key input to the emotion transfer function, and can be included by factors such as waiting time. Service duration The service quality indicators are calculated as follows: ,in Other service quality indicators. Specifically, during the training phase, service quality scores are calculated using observational data on waiting time and service duration. Then, the emotion transfer coefficient is learned through the emotion transfer function. During the prediction phase, given the waiting time and service duration, the corresponding service quality score can be predicted. By combining the emotional values ​​from the preceding stages, the satisfaction level of subsequent stages can be predicted.

[0090] In step S02 of this embodiment, the calculation expression of the emotion transfer function can be specifically expressed as follows:

[0091] (3)

[0092] in, For the link i At the present moment t The emotion value reflects service satisfaction, and the specific value range can be configured as [0, 10]. For the link i In the previous moment The emotional value is used to reflect the memory effect of emotions. For the link i The service quality score can be calculated in advance based on service quality indicators such as waiting time, service time, and equipment utilization. For the link i The set of preceding stages, To start from the link j To the stage i The emotion transmission coefficient can be learned from historical data; This is a memory decay factor used to control the degree of influence of historical emotions. Weighting of service quality for the current stage. The weight of the emotional influence in the preceding stages.

[0093] Specifically, regarding the initial emotional value, taking the first step (such as registration) as an example, Collect patient satisfaction scores upon initial admission to the hospital. If no historical data is available, a neutral value (e.g., 5 points) can be used, or the scores can be obtained from historical data statistics. For subsequent procedures... i ( i≧ 2) The service satisfaction score at the beginning of this stage is taken as the basis for evaluation; that is, the satisfaction score for this stage is obtained from the service satisfaction feedback information. During the training phase, the collected service satisfaction scores are used as the initial emotional values ​​for each stage, and then the emotional values ​​for each stage at subsequent times are calculated according to the emotional transfer function of Equation (3). During the prediction phase, the initial emotional value of the first stage can be preset or obtained from historical data, and then the emotional values ​​for subsequent stages are calculated according to the emotional transfer function of Equation (3) and the emotional values ​​of the preceding stages.

[0094] In this embodiment, the transfer coefficients can be learned by training the dynamic Bayesian network with historical feedback data. This leads to a time-series prediction model, enabling cascaded satisfaction prediction. Specifically, a large amount of satisfaction data from the complete medical process is collected, including satisfaction scores, service quality, and timestamps at each stage. The maximum likelihood estimation method is then used to learn the transfer coefficients. The optimal parameters are solved by gradient descent optimization algorithm. The emotional value of each link at the initial time is obtained from the patient satisfaction feedback information. That is, the satisfaction score can be used as the emotional value at the initial time of the corresponding link. The service quality score of each link is obtained from the service quality score feedback information.

[0095] In this embodiment, after training the dynamic Bayesian network model, the emotion transmission coefficients between each stage are learned. The time-series cascaded model is finally obtained from the trained dynamic Bayesian network model: ,in, Indicates about , as well as The function, Table of Pre-processing Stages The emotional values ​​of all stages, i.e. , For the link i The set of preceding stages, Preliminary stage j The emotion value from the previous moment is used to calculate the current stage using the emotion transfer function. i The emotional value, This represents the model parameters. Based on this time-series cascade model, given the satisfaction level of preceding stages, the satisfaction level of subsequent stages can be predicted. Specifically, based on the sentiment values ​​of all preceding stages, the service quality score of the next stage, and the time frame of the next stage in the previous stage... The emotional value can be used to predict the emotional value of the next stage, i.e. the satisfaction level of the next stage, according to formula (3).

[0096] The specific steps of step S03 in this embodiment include:

[0097] Step S301, Preliminary Screening: Calculate the information entropy of each satisfaction rating in the service satisfaction rating set, for example, by using the following expression: ,in Indicates a single feedback message. The percentage of scores for each evaluation dimension The information entropy of feedback information reflects the degree of dispersion of feedback information. The larger the value, the more objective the feedback. The service satisfaction score set is initially screened based on the information entropy of each score information to remove score information with information entropy lower than the preset threshold, resulting in a preliminarily adjusted service satisfaction score set.

[0098] Step S302, Weight Correction: Calculate the corrected weight coefficients for each rating in the service satisfaction rating set according to the following formula. :

[0099] (4)

[0100] in, Indicates the basic weight. This indicates the initial adjusted satisfaction rating. This represents the expert evaluation benchmark value. This indicates the adjustment parameter used to control the rate of weight decay;

[0101] Step S303, Correction Output: Use the corrected weighting coefficients Obtain the corrected satisfaction score that takes into account the bias. :

[0102] (5)

[0103] in, Indicates the first i The service satisfaction rating corresponding to each piece of feedback information Represents the corrected weight coefficient of the i-th feedback message. .

[0104] The corrected feedback information set consists of the feedback information of the corrected satisfaction score and service quality score, as well as the corresponding time information.

[0105] In a specific application embodiment, the scores of medical experts on various issues can be obtained in advance and used as expert evaluation benchmark values ​​to construct an issue-expert evaluation benchmark value mapping table. After obtaining the preliminary adjusted satisfaction score feedback information, the corresponding expert evaluation benchmark value is found from the issue-expert evaluation benchmark value mapping table for each feedback information, and the corresponding corrected weight coefficient is calculated according to formula (4). Then, the deviation-corrected satisfaction data is obtained according to formula (5).

[0106] This embodiment uses the above-mentioned two-layer correction mechanism to correct the collected satisfaction feedback information. By calculating the feedback information entropy to identify cognitive biases for preliminary screening, low-quality feedback can be identified and filtered out. Then, expert evaluation benchmarks are introduced to adjust the weights and reduce cognitive biases, which can improve the objectivity and reliability of feedback data and solve the problem of information asymmetry.

[0107] In step S04 of this embodiment, a multi-objective optimization function is constructed with the objectives of maximizing the increase in satisfaction, minimizing the increase in cost, and maximizing the increase in efficiency. The calculation expression is as follows:

[0108] (6)

[0109] in, , , , as well as Indicates the weighting coefficient. This indicates the amount of improvement in satisfaction data after correction; Indicates the increase in cost; Indicates the amount of efficiency improvement. Indicates the specified evaluation indicators. Indicates the penalty for constraint violation, for example , Indicates the first i The degree of violation of a constraint, Indicates the first i The penalty coefficient for each constraint.

[0110] In a specific application embodiment, the increase in cost The calculation expression is: ,in The total cost of the optimized medical service process plan (including labor costs, equipment usage costs, management costs, etc.) The total cost of the baseline healthcare service process is given. During the iterative solution process, the cost of each candidate solution is calculated and then compared with the baseline solution to obtain the cost increase of the candidate solution. .

[0111] Efficiency improvement The calculation expression is: ,in This indicates the reduction in waiting time (waiting time of the baseline plan minus waiting time of the new plan). This indicates the reduction in service duration (baseline service duration minus new service duration). and The weighting coefficients are used. Since the observation nodes include waiting time and service time, the observation data can be used to calculate the efficiency improvement. During the iterative solution process, the time series cascade model is used to predict the changes in waiting time and service time under different process schemes, and the efficiency improvement can then be calculated.

[0112] This embodiment improves overall satisfaction. Increase in costs Efficiency improvement By constructing a multi-objective optimization function, we can comprehensively evaluate the satisfaction, cost, and efficiency of the medical service process, and then select the optimal medical service process solution through iterative solution.

[0113] Improvements in healthcare services are subject to strict resource constraints (such as fixed equipment capacity and number of doctors) and safety constraints (such as the inability to reduce necessary treatment time and simplify examination procedures). A lack of quantitative modeling of these constraints may lead to infeasible or dangerous improvement suggestions. To address these issues, this embodiment includes both hard and soft constraints. The hard constraints correspond to safety constraints, including minimum treatment time constraints and safe operation constraints. The minimum treatment time constraint is as follows: ,in, Indicates the actual treatment time. Indicates the shortest treatment time. This represents the safety factor, which can be adjusted within the range of [1.0, 1.2] based on the specific treatment types and treatment outcomes included in different candidate protocols. For example, regarding the treatment phase, if a candidate protocol includes high-risk procedures (such as interventional surgery), then... If it is a standard treatment type, then set .

[0114] Safety constraints include the requirement that all operations comply with safe operating procedures. Soft constraints correspond to resource efficiency constraints, including constraints on physician workload, equipment utilization, and cost control, for example: , This represents the actual workload of a single doctor. This indicates the maximum working capacity of a single doctor, and the equipment utilization rate. , This represents equipment utilization (calculated as the ratio of actual equipment usage time to total available time), and cost control constraints: , This represents the actual total cost of the optimized solution. This indicates the upper limit of the total cost budget. The penalty coefficients for hard constraints and soft constraints are configured with different values; the weighting coefficient of a hard constraint is greater than that of a soft constraint. For example, a hard constraint... soft constraints Combining the above hard and soft constraints, the constraint violation penalty term can be expressed as:

[0115] (7)

[0116] in, Represents a set of hard constraints. This represents the set of soft constraints.

[0117] This embodiment constructs the aforementioned hierarchical constraint optimization model, which clearly distinguishes between hard constraints (safety constraints) and soft constraints (efficiency objectives) in the multi-objective optimization model. The high penalty coefficient ensures compliance with safety constraints, ensuring that all improvement measures meet medical operating procedures and safety requirements, avoiding safety hazards caused by process adjustments. Thus, service optimization can be achieved while ensuring medical safety, maintaining a balance between medical efficiency and quality.

[0118] In step S04 of this embodiment, an improved genetic algorithm is used to solve the multi-objective optimization function. The improved genetic algorithm includes a mutation probability... Based on individual fitness values Average fitness value of the population The adaptive determination yields the following calculation expression:

[0119] (8)

[0120] This embodiment uses an improved genetic algorithm to solve the multi-objective optimization function, and the convergence efficiency can be effectively improved by using an adaptive mutation rate.

[0121] Specifically, the detailed steps for solving the problem using the improved genetic algorithm are as follows:

[0122] Step 1: Population Initialization

[0123] Generate 100 initial solutions that satisfy the hard constraints, and use domain knowledge as a guide, where 70% of the domain knowledge is randomly generated and 30% can be obtained based on historical excellent solutions.

[0124] Step 2: Fitness Function Design

[0125] The following fitness function can be used:

[0126] (9)

[0127] in, Let F represent the fitness value, and let F represent the value of the multi-objective optimization function (as calculated according to equation (7)). ), This indicates the penalty item (the penalty item for constraint violation), i.e. This is used to penalize solutions that do not meet the constraints.

[0128] Step 3: Implement crossover operations

[0129] A repair operator is used to ensure that offspring meet hard constraints, and crossover probabilities are set. Specifically, by combining hard constraint configuration repair strategies, after crossover operations, offspring individuals that do not meet the hard constraints are repaired. For example, for the shortest treatment time constraint, if... If the actual treatment time for that step is not met, the fitness value is adjusted to satisfy the constraints. After repair, the fitness value is recalculated to ensure that the offspring meet the hard constraints before proceeding to subsequent mutation and selection operations. This ensures that the genetic algorithm's search space remains within the feasible region, avoiding the generation of infeasible solutions that do not meet medical safety requirements.

[0130] Step 4: Design of Mutation Operations

[0131] According to the formula Adaptively adjust mutation probability.

[0132] In this embodiment, a hierarchical iterative verification step is included after step S04, including:

[0133] Based on the first time period, adjust one or more links where the satisfaction level is lower than the preset threshold, and iterate again to determine the optimal medical service process solution.

[0134] The implementation effect of the optimal medical service process plan is evaluated according to the second time period, and the parameters in the multi-objective optimization function and the parameters of the dynamic Bayesian network model are adaptively adjusted based on the evaluation results of the implementation effect.

[0135] The time-series cascaded model was retrained according to the third time period, and the medical terminology mapping dictionary was updated in order to re-update and evaluate the model.

[0136] The lengths of the first, second, and third time periods increase sequentially.

[0137] Specifically, the three cycles mentioned above correspond to short-term iteration (e.g., a 2-week cycle), medium-term iteration (e.g., a monthly cycle), and long-term iteration (e.g., a quarterly cycle), respectively. Short-term iterations can enable rapid responses to urgent issues, focusing on areas with high patient complaint rates and quickly verifying the effectiveness of improvement plans. Medium-term iterations can evaluate the effects of multiple optimal medical service processes to analyze the long-term impact of improvement plans and adjust model parameters based on the evaluation results. Long-term iterations can comprehensively and systematically evaluate and update the model by periodically retraining the temporal cascade model and updating the medical terminology mapping dictionary.

[0138] In a specific application embodiment, the implementation process of short-term iteration includes:

[0139] (1) Identification of key links: Based on the service satisfaction feedback information collected in step S01, calculate the complaint rate of each link according to a short period of time (e.g., 2 weeks): ,in For the link The number of complaints (satisfaction scores below a preset threshold, such as below 3 points). For the link The total number of feedback responses is used to identify key areas of focus where the complaint rate exceeds a preset threshold (e.g., complaint rate > 20%).

[0140] (2) Rapid optimization: Increase the weight coefficient of the identified key focus links in the multi-objective optimization function. (Increase in satisfaction) The weight of the link is increased, or the constraint penalty coefficient of the link is increased. The time series cascade model is used to predict the change in satisfaction of the link. The resource allocation and service process of the link are optimized. The multi-objective optimization decision of step S04 is re-executed, and the search space is focused on the identified key links and their predecessor links (because the emotion transmission effect will affect the subsequent links).

[0141] (3) Rapid verification: After implementing the optimization plan, collect feedback data in the next short period and calculate the improvement effect: ,in To optimize the process Average satisfaction The average satisfaction level before optimization is used; if the improvement is not significant ( If so, then adjust the optimization strategy or parameters.

[0142] In this embodiment, based on the evaluation results of the implementation effect, the parameters in the multi-objective optimization function and the parameters of the dynamic Bayesian network model are adaptively adjusted using the Bayesian optimization method. The parameters of the dynamic Bayesian network model include the emotion transfer coefficient in the emotion transfer function and the dynamic Bayesian network structure. The parameters in the multi-objective optimization function include the weight coefficients of each optimization objective and the penalty coefficients in the constraint penalty terms. The implementation effect includes the increase in satisfaction, the improvement in cascade efficiency, and the degree of constraint satisfaction / constraint violation. When the increase in satisfaction is less than a preset threshold, the emotion transfer coefficient in the emotion transfer function is increased. When the improvement in cascade efficiency is less than a preset threshold, the dynamic Bayesian network structure is adjusted. When the degree of constraint satisfaction is lower than a preset threshold, the penalty coefficient in the constraint penalty terms is increased.

[0143] In a specific application example, the satisfaction improvement rate It can be calculated using the following expression:

[0144] (10)

[0145] in, This indicates the level of satisfaction after optimization. This represents the satisfaction level before optimization.

[0146] Cascade efficiency improvement The calculation expression is:

[0147] (11)

[0148] in, This represents the aggregate value of all emotion transfer coefficients in the optimized cascade model, for example, all emotion transfer coefficients. The L2 norm or mean is used to measure the overall strength of emotion transmission between optimized stages. This represents the aggregated value of all emotion transmission coefficients in the cascade model before optimization, used to measure the overall strength of emotion transmission between stages before optimization.

[0149] Constraint satisfaction The calculation expression is:

[0150] (12)

[0151] in, This represents the number of constraints that are satisfied, i.e., the number of constraint terms in the optimized solution that satisfy all hard and soft constraints. This indicates the total number of constraints, that is, the total number of all hard and soft constraints.

[0152] In a specific application embodiment, the steps for adaptively adjusting parameters using the Bayesian optimization method include:

[0153] Step 1: Evaluation Function Construction

[0154] Construct the evaluation function ,in, Indicates the combination of parameters to be evaluated. Indicating accuracy, Indicates efficiency. Indicates the degree of constraint violation. , as well as Indicates weight;

[0155] Step 2: Parameter Space Modeling

[0156] The parameter space is modeled using Gaussian processes (GP), and the next set of parameters is selected through the desired improvement (EI):

[0157] (13)

[0158] in, This represents the currently known best evaluation function value, that is, the optimal combination of parameters found so far during the parameter search process. Corresponding evaluation function value .

[0159] Step 3: Parameter Adjustment Rules

[0160] The parameters are adjusted based on the increase in satisfaction, the improvement in cascading efficiency, and the degree of constraint satisfaction. Specifically, if the satisfaction increase rate... Increase the emotion transmission coefficient Cascade efficiency improvement Adjust the DBN network structure; if the constraint violation rate Increase the penalty coefficient .

[0161] Furthermore, intelligent allocation algorithms can be used to improve the utilization rate of core resources such as specialist appointment slots and examination equipment, thereby further shortening patients' waiting time and improving the overall operational efficiency of medical services.

[0162] This application employs the above-mentioned method and a layered, progressive processing flow: standardization of medical terminology → modeling of temporal cascade effects → feedback bias correction → multi-objective optimization decision-making → iterative verification and adjustment. This forms a complete intelligent closed-loop process for optimizing medical service quality, from patient feedback collection to the implementation of optimization decisions. It can accurately determine the optimal medical service process scheme that maximizes satisfaction, service efficiency, and minimizes resource costs, thereby optimizing the service efficiency and resource utilization of medical services and improving the patient's medical experience.

[0163] To achieve the above method, such as Figure 2 As shown, this embodiment constructs a medical service process iterative decision-making system to implement medical service process iterative decision-making according to the above scheme, including:

[0164] The data input layer is used to receive patient satisfaction feedback, service quality score feedback, medical resource data, expert evaluation benchmark data, and historical feedback information (satisfaction data and service quality information), etc.

[0165] The preprocessing layer is used for medical terminology standardization, terminology disambiguation, and semantic similarity calculation to map patient satisfaction feedback information into a set of standard medical terms.

[0166] The analysis and modeling layer is used to construct a dynamic Bayesian network and design an emotion transfer function. Historical feedback information is used to train the model to obtain a time-series cascade model for cascade effect prediction.

[0167] The correction and optimization layer is used for feedback bias correction and multi-objective optimization by function construction and solution.

[0168] The decision output layer is used to output the optimal medical service process obtained through multiple rounds of iterative solutions. Optionally, it can also output the constraint satisfaction evaluation and optimization effect prediction results.

[0169] Optionally, a verification and adjustment layer is also included to verify the implementation effect of the output scheme and to perform adaptive parameter adjustment in a hierarchical iterative manner.

[0170] The medical service process iterative decision-making system in this embodiment corresponds to the method for constructing medical service process iterative decision-making, and will not be described in detail here.

[0171] This embodiment also provides a computer system, including a processor and a memory, wherein the memory is used to store a computer program and the processor is used to execute the computer program to perform the method described above.

[0172] It is understood that the method described in this embodiment can be executed by a single device, such as a computer or server, or it can be applied to a distributed scenario where multiple devices cooperate to complete the task. In a distributed scenario, one of the multiple devices may execute only one or more steps of the method described in this embodiment, and the multiple devices interact to complete the method. The processor can be implemented using a general-purpose CPU, microprocessor, application-specific integrated circuit, or one or more integrated circuits, and is used to execute relevant programs to implement the method described in this embodiment. The memory can be implemented using read-only memory (ROM), random access memory (RAM), static storage devices, and dynamic storage devices. The memory can store the operating system and other applications. When the method described in this embodiment is implemented through software or firmware, the relevant program code is stored in the memory and called and executed by the processor.

[0173] Those skilled in the art will understand that the above embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-readable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create a machine for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The functions specified in one or more boxes. These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable apparatus for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0174] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the invention. Therefore, any simple modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention should fall within the protection scope of the present invention.

Claims

1. A medical service process iterative decision-making method based on multi-source feedback information, characterized by the steps of Comprise: Step S01, multi-source feedback information acquisition: acquire service satisfaction feedback information and service quality index feedback information of each link of the current medical service process, and convert the service satisfaction feedback information to obtain service satisfaction scores and calculate the service quality index feedback information to obtain service quality scores to form a feedback information set, the service quality index includes waiting time and service time; Step S02, time sequence cascade effect modeling: model the medical service process as a dynamic Bayesian network model, model the links of the medical service process as state nodes, and the satisfaction scores as observation nodes, use an emotion transfer function to calculate the emotion values after state transition between the state nodes and quantify the transfer effect through an emotion transfer coefficient, the emotion transfer function is a function of calculating the emotion value at the current time according to the service quality scores of each link and the emotion value at the previous time, the emotion value is used to represent the satisfaction, the dynamic Bayesian network model is trained using the feedback information set, during the training process, the service satisfaction scores of each link are used as the initial emotion values of each link, and after the training is completed, a time sequence cascade model is obtained for predicting the satisfaction of the subsequent link under the given previous link; Step S03, feedback bias correction: calculate the bias between the expert evaluation benchmark data and each service satisfaction score, adjust the corresponding weight in each service satisfaction score according to the calculated bias to obtain the corrected satisfaction score; Step S04, multi-objective optimization decision: construct a multi-objective optimization function to maximize the satisfaction improvement, service efficiency increase and minimize the increase of resource cost, iteratively solve the constructed multi-objective optimization function, during the solving process, predict the satisfaction change of each link under different medical service processes based on the corrected satisfaction score and the service quality score using the time sequence cascade model, obtain the satisfaction improvement, and after multiple iterations, obtain the optimal medical service process scheme output; In step S04, the calculation expression of the constructed multi-objective optimization function is: wherein, , , , and denote weight coefficients, denote satisfaction improvement amounts, denote cost increase amounts, denote efficiency improvement amounts, denote specified evaluation indexes, denote constraint violation penalty terms, the constraint violation penalty terms including hard constraint terms and soft constraint terms, the hard constraint terms including a shortest diagnosis and treatment time constraint: , denote actual diagnosis and treatment times, denote shortest diagnosis and treatment times, denote safety coefficients, the soft constraint terms including a workload constraint, a device utilization rate constraint, and a cost control constraint, the weight coefficients of the hard constraint terms being greater than the weight coefficients of the soft constraint terms.

2. The medical service procedure iterative decision-making method based on multi-source feedback information according to claim 1, characterized in that, In step S01, before converting the service satisfaction feedback information into satisfaction scores, it also includes mapping the service satisfaction feedback information into standard medical terms, the steps include: Step S101. Construct a medical term mapping dictionary; Step S102. Knowledge graph construction: constructing a medical knowledge graph , representing entity nodes including diseases, symptoms, examinations, and drugs, representing relationships between entities; Step S103. Term disambiguation: a preliminary set of standardized medical terms is obtained by disambiguating the multi-source feedback information set, in the disambiguating process, by looking up all candidate mappings of the term to be disambiguated in the constructed medical knowledge graph and calculating the context similarity score of each candidate mapping found , wherein is the point mutual information between the word and the candidate mapping for measuring the strength of the association, represents the serial number index of the candidate mapping, i is the context composed of multiple words before and after the term to be disambiguated, C is the word in the context C , represents the inverse document frequency of the word , and one candidate mapping is selected as the standard term of the term to be disambiguated according to the context similarity score. Step S104. Calculate the similarity between each medical term in the preliminary standardized medical term set, cluster each medical term according to the calculated similarity, and obtain the final standardized medical term set; Step S105. Convert the service satisfaction feedback information into standardized medical terms according to the final standardized medical term set.

3. The medical service procedure iterative decision-making method based on multi-source feedback information according to claim 2, characterized in that, In step S104, the similarity between the two medical terms is calculated by weighting the knowledge graph semantic similarity , the category similarity , and the context vector cosine similarity , wherein the knowledge graph semantic similarity is calculated according to , is the path distance of the two medical terms to be calculated in the constructed medical knowledge graph , the category similarity is a value according to whether the two medical terms to be calculated are of the same category, and the context vector cosine similarity is calculated according to the cosine distance between the context word vectors of the two medical terms to be calculated.

4. The medical service procedure iterative decision-making method based on multi-source feedback information according to claim 1, characterized in that, In step S02, the calculation expression of the emotion transfer function used is: in, For the link i At the present moment t The sentiment score is used to reflect service satisfaction. For the link i In the previous moment The emotional value is used to reflect the memory effect of emotions. For the link i Service quality rating For the link i The set of preceding stages, To start from the link j To the stage i The emotion transmission coefficient; This is a memory decay factor used to control the degree of influence of historical emotions. Weighting of service quality for the current stage. Weighting of the impact of emotions in the preceding stages; In the process of training the dynamic Bayesian network model using the feedback information set, the service quality scores of each link are obtained from the service quality scores calculated according to the feedback information of the service quality indicators, and the emotion transmission coefficients between the links are learned Finally, a time series cascade model is obtained: Wherein, represents a function about , and , represents the emotion values of all links in the previous link set , represents the model parameters.

5. The medical service procedure iterative decision-making method based on multi-source feedback information according to claim 1, characterized in that, The steps of step S03 include: Step S301, preliminary screening: calculate the information entropy of each satisfaction score information in the service satisfaction score set, preliminarily screen the service satisfaction score set according to the information entropy of each score information, remove the score information with information entropy lower than a preset threshold, and obtain a preliminary adjusted service satisfaction score set; Step S302, weight correction: according to the formula Calculate the corrected weight coefficient of each score information in the service satisfaction score set , represents the basic weight, represents the preliminary adjusted satisfaction score, represents the expert evaluation reference value, represents the adjustment parameter for controlling the weight decay speed; Step S303, correction output: using the corrected weight coefficient A corrected satisfaction score considering the bias is obtained.

6. The medical service procedure iterative decision-making method based on multi-source feedback information according to any one of claims 1-5, characterized in that, In the step S04, the multi-objective optimization function is solved by using an improved genetic algorithm, in which the mutation probability is determined by the following formula: p = p0 + (p1-p0) * (1- (fmax-fmin) / (fmax-fmin) ), where p0 and p1 are the initial mutation probability and the final mutation probability, respectively, fmax and fmin are the maximum and minimum fitness values, respectively. According to the individual fitness value , the population average fitness value is determined adaptively, and the calculation expression is as follows: .

7. The medical service procedure iterative decision-making method based on multi-source feedback information according to any one of claims 1-5, characterized in that, The step S04 is followed by a layered iteration verification step, comprising: adjusting one or more links with a satisfaction degree lower than a preset threshold according to a first time period, and re-iterating to determine an optimal medical service process scheme; evaluating an implementation effect of the optimal medical service process scheme according to a second time period, and adaptively adjusting parameters in the multi-objective optimization function and parameters of the dynamic Bayesian network model according to an evaluation result of the implementation effect; re-training the time-series cascade model and updating the medical term mapping dictionary according to a third time period, to re-perform model updating and evaluation; period lengths of the first time period, the second time period and the third time period are sequentially increased.

8. The medical service procedure iterative decision-making method based on multi-source feedback information according to claim 7, characterized in that, The parameters in the multi-objective optimization function and the parameters of the dynamic Bayesian network model are adaptively adjusted according to the evaluation result of the implementation effect by using a Bayesian optimization method, the parameters of the dynamic Bayesian network model include emotion transmission coefficients in an emotion transmission function and a dynamic Bayesian network structure, and the parameters in the multi-objective optimization function include weight coefficients of each optimization target and a penalty coefficient in a constraint penalty term, wherein the implementation effect includes a satisfaction degree improvement rate, a cascade efficiency improvement degree and a constraint satisfaction degree, when the satisfaction degree improvement rate is less than a preset threshold, the emotion transmission coefficients in the emotion transmission function are increased, when the cascade efficiency improvement degree is less than a preset threshold, the dynamic Bayesian network structure is adjusted, and when the constraint satisfaction degree is lower than a preset threshold, the penalty coefficient in the constraint penalty term is increased.

9. A computer system comprising a processor and a memory for storing a computer program, characterised in that, The processor is configured to execute the computer program to perform the medical service process iteration decision method according to any one of claims 1-8. The processor is configured to execute the computer program to perform the medical service process iteration decision method according to any one of claims 1-8.

Images