Telemedicine doctor-patient allocation method and system based on text mining and stochastic programming

Abstract

The application discloses a kind of remote medical treatment doctor-patient allocation method and system based on text mining and random programming, belong to medical resource optimization scheduling field.The method includes obtaining the application information of patient;Text mining is carried out to the unstructured text data of application information to extract text features, and after the feature fusion of text features and the structured data of application information, input to classification model, obtain the classification result of the service time category to which patient belongs;Based on the classification result, estimate the probability distribution parameter of service time for each patient;The probability distribution parameter is input into two-stage random programming model, and a mathematical model is obtained;Using sample mean approximation method, the mathematical model is converted into a deterministic mixed integer linear programming model and solved, and the optimal doctor-patient allocation strategy is generated.The application can effectively utilize text information, accurately predict the service time category to which patient belongs and optimize the doctor-patient allocation scheme.

Description

Technical Field

[0001] This invention belongs to the field of medical resource optimization and scheduling technology, specifically involving a remote medical patient allocation method and system based on text mining and stochastic programming. Background Technology

[0002] Telemedicine effectively improves the accessibility and utilization efficiency of medical resources by connecting primary hospitals with central hospitals. In the operation and management of telemedicine, the core issues for reducing costs and improving service rates are: rationally allocating patients requesting consultations to remote doctors and setting precise service durations.

[0003] In existing technologies, telemedicine centers typically use fixed service times (such as 10 minutes or 20 minutes) to schedule all patients. However, the actual service time is highly uncertain due to various factors, including the complexity of the patient's condition. Using fixed durations leads to a large number of patients' actual service times far exceeding their appointment times, resulting in frequent cancellations or delays. Data shows that cancellation rates can exceed 30%, severely impacting resource utilization efficiency and patient satisfaction.

[0004] To address this issue, some existing studies have introduced stochastic programming methods, utilizing structured historical data such as patient age and gender to characterize the uncertainty of service time. However, these methods overlook the rich information contained in unstructured data (such as preliminary diagnostic texts written by primary care physicians). This textual data contains crucial information such as the patient's suspected disease, underlying conditions, and disease severity, which is of significant value for predicting service time, but it has not been effectively utilized in current decision-making models.

[0005] Therefore, how to effectively mine unstructured text information and combine it with optimization models to improve the accuracy of patient service time prediction, and thus develop a more cost-effective and efficient doctor-patient allocation scheme, is a technical problem that urgently needs to be solved. Summary of the Invention

[0006] The purpose of this invention is to provide a method and system for remote medical patient allocation based on text mining and stochastic programming, which can effectively utilize text information, accurately predict the service time category to which the patient belongs, and optimize the patient allocation scheme.

[0007] In a first aspect, the present invention provides a method for remote medical patient allocation based on text mining and stochastic programming, comprising the following steps: S1: Obtain the patient's request information, which includes structured data and unstructured text data; S2: Perform text mining on unstructured text data to extract text features, and then fuse the text features with structured data before inputting them into a classification model to obtain the classification results of the service time category to which the patient belongs; S3: Based on the classification results, estimate the probability distribution parameters of service time for each patient, where patients belonging to different service time categories are set to follow truncated probability distributions within different intervals; S4: Input the probability distribution parameters into the two-stage stochastic programming model to construct a mathematical model containing decision variables, objective function and constraints. The first stage of the mathematical model is used to make immediate allocation decisions before the actual service time is unknown. The second stage is used to make retrospective adjustment decisions under multiple discretized stochastic scenarios where the actual service time is known. S5: Using the sample mean approximation method, the mathematical model constructed in step S4 is transformed into a deterministic mixed integer linear programming model and solved to generate the optimal doctor-patient allocation strategy.

[0008] As an alternative implementation, step S2 specifically includes: Perform text normalization, stemming, and stop word removal on unstructured text data; The bag-of-words model is used to convert the processed text data into numerical feature vectors; Principal component analysis is used to reduce the dimensionality of numerical feature vectors and extract a predetermined number of principal components as text features. The extracted text features are concatenated with the structured data to form the final fused feature vector. The final feature vector is input into the gradient booster classification model, which outputs the classification result of whether the patient belongs to the normal service time category or the extended service time category.

[0009] As an alternative implementation, in step S3, the truncation probability distribution is a truncation exponential distribution; the step of estimating the probability distribution parameters of the service time specifically includes: Obtain the preset duration threshold c, minimum service time a, and maximum service time b; For patients classified as having normal service time, their service time is set to follow a first-cut-off exponential distribution within the interval [a, c]. For patients classified as having extended service time, their service time is set to follow a second cutoff exponential distribution within the interval [c, b]. The rate parameters of the first and second truncated exponential distributions are estimated by using service time samples within the corresponding intervals of historical data.

[0010] As an alternative implementation, in step S4, the objective function of the mathematical model is to minimize the expected sum of the first-stage cost and the second-stage cost; The decision variables in the first phase include: variables instructing doctors whether to provide services, variables instructing patients whether to be assigned to a specific doctor, and variables instructing patients whether to postpone to the next cycle; the costs in the first phase include the doctor's fixed operating costs and the penalty costs for patient postponement, minus the benefits of serving the patient. The decision variables in the second phase include: variables indicating whether a patient is canceled under various random scenarios, and continuous variables indicating the length of time a doctor works overtime; the costs in the second phase include the cost of doctors working overtime and the cost of penalties for canceling patients.

[0011] As an alternative implementation, step S5, which involves using the sample mean approximation method to transform the mathematical model into a deterministic mixed-integer linear programming model, includes the following specific steps: Based on the cutoff probability distribution set for each patient in step S3, a predetermined number of N random scenarios are generated using Monte Carlo sampling. Each scenario contains the service time realized value of all patients in that scenario. Substitute the N random scenarios into the mathematical model, and use the sample mean of the second-stage cost under the N random scenarios to approximate the expected term of the second-stage cost, thereby obtaining a deterministic mixed-integer linear programming model. The deterministic mixed-integer linear programming model is solved by calling the mixed-integer linear programming solver.

[0012] The optimal doctor-patient allocation strategy includes at least: the set of doctors selected to provide services in the current planning cycle, the list of patients pre-assigned to each doctor, and the list of patients postponed to the next planning cycle. During the execution phase, if the total actual service time of any doctor's pre-assigned patients exceeds the sum of the doctor's planned working hours and the preset maximum overtime hours, then the appointments of some of the doctor's patients will be cancelled according to the predetermined priority rules.

[0013] Secondly, the present invention provides a remote medical patient allocation system based on text mining and stochastic programming, comprising: The data acquisition module is configured to acquire patient application information, which includes structured data and unstructured text data. The text mining and classification module is configured to: perform text mining on unstructured text data to extract text features, and then fuse the text features with structured data before inputting them into the classification model to obtain the classification results of the service time category to which the patient belongs; The distribution parameter determination module is configured to estimate the probability distribution parameters of service time for each patient based on the classification results, wherein patients belonging to different service time categories are set to follow truncated probability distributions within different intervals; The optimization model building module is configured to: input probability distribution parameters into a two-stage stochastic programming model to build a mathematical model containing decision variables, objective function and constraints. The first stage of the mathematical model is used to make immediate allocation decisions before the actual service time is unknown, and the second stage is used to make retrospective adjustment decisions under multiple discretized stochastic scenarios where the actual service time is known. The solution module is configured to use the sample mean approximation method to transform the mathematical model constructed by the optimization model construction module into a deterministic mixed integer linear programming model and solve it to generate the optimal doctor-patient allocation strategy.

[0014] Thirdly, the present invention provides an electronic device including a memory and a processor, and computer instructions stored in the memory and running on the processor, wherein the computer instructions, when executed by the processor, perform the method described in the first aspect.

[0015] Fourthly, the present invention provides a computer-readable storage medium for storing computer instructions, which, when executed by a processor, perform the method described in the first aspect.

[0016] Fifthly, the present invention provides a computer program product, including a computer program that, when executed by a processor, implements the method described in the first aspect.

[0017] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention proposes a method and system for remote medical patient allocation based on text mining and stochastic programming. By deeply integrating unstructured text mining and stochastic programming models, it systematically solves the problems of resource misallocation and high cancellation rates caused by the high uncertainty of remote medical patient service time. Firstly, in terms of technical approach, this invention breaks through the limitations of traditional methods that rely solely on structured data such as age and gender. It innovatively extracts key semantic features such as disease complexity and suspected diseases from the initial diagnosis texts of primary care physicians, and combines this with a gradient boosting machine classification model to accurately classify patients into normal or extended service time categories. Experiments show that the classification model achieves an AUC of 90.41% and a sensitivity of 86.23% on the test set, significantly improving the ability to identify patients with complex conditions. Secondly, based on the classification results, this invention assigns differentiated cutoff distribution parameters to different patient categories, making the characterization of service time uncertainty more consistent with actual business logic. Compared with existing fixed-duration allocation methods, this significantly reduces resource waste caused by time estimation errors. Building upon this foundation, this invention constructs a two-stage stochastic programming optimization model, organically unifying the service assignment and postponement decisions in the first stage with the overtime and cancellation retrospective decisions in the second stage. The objective function considers doctor workload, patient postponement penalties, service benefits, and overtime and cancellation penalties, achieving the minimization of total operating costs or the maximization of total benefits. Actual data validation results show that after adopting the strategy proposed in this invention, the patient cancellation rate decreased from 32.4% to 18.0%-22.0%, and the average benefit increased by 41%-47% compared to the traditional fixed-duration strategy. Furthermore, it exhibits excellent robustness even when the service time distribution shifts. Therefore, this invention achieves significant technological advancements in improving the efficiency of telemedicine resource utilization, reducing operating costs, and enhancing the patient experience. Attached Figure Description

[0018] Figure 1 This is an overall flowchart of the telemedicine doctor-patient allocation method based on text mining and random programming disclosed in the embodiments of the present invention; Figure 2 This is a schematic diagram of the specific process for predicting patient service time using text information, as disclosed in an embodiment of the present invention. Figure 3 This is a trend chart of the individual variance and cumulative variance contribution rate of the principal components in the principal component analysis disclosed in the embodiments of the present invention; Figure 4 This is a schematic diagram of the confusion matrix of the gradient boosting machine classifier disclosed in the embodiments of the present invention on the training set and validation set; Figure 5 This is the ROC curve of the gradient booster classifier disclosed in the embodiments of the present invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0020] The technical solutions disclosed in the various embodiments of this application are described in detail below with reference to the accompanying drawings.

[0021] Example 1 like Figure 1 As shown, this embodiment provides a remote medical patient allocation method based on text mining and stochastic programming, including the following steps: S1: Obtain the patient's request information, which includes structured data and unstructured text data; S2: Perform text mining on unstructured text data to extract text features, and then fuse the text features with structured data before inputting them into a classification model to obtain the classification results of the service time category to which the patient belongs; S3: Based on the classification results, estimate the probability distribution parameters of service time for each patient, where patients belonging to different service time categories are set to follow truncated probability distributions within different intervals; S4: Input the probability distribution parameters into the two-stage stochastic programming model to construct a mathematical model containing decision variables, objective function and constraints. The first stage of the mathematical model is used to make immediate allocation decisions before the actual service time is unknown. The second stage is used to make retrospective adjustment decisions under multiple discretized stochastic scenarios where the actual service time is known. S5: Using the sample mean approximation method, the mathematical model constructed in step S4 is transformed into a deterministic mixed integer linear programming model and solved to generate the optimal doctor-patient allocation strategy.

[0022] The specific steps of this plan are as follows: Step S1: Obtain the patient's application information.

[0023] In this step, when a primary care physician applies for telemedicine services for a patient, the system acquires the application information submitted by the physician. Such information includes structured data and unstructured text data. Examples of the structured data include "gender" (e.g., male / female) and "age" (e.g., an integer ranging from 5 to 97 years old). The unstructured text data includes at least "preliminary diagnosis", which is descriptive text about the patient's condition, suspected diseases, basic medical history and the like written by the primary care physician based on the consultation, for example, "1. Severe pulmonary infection with fungal infection (pulmonary aspergillus infection); 2. IgA nephropathy; 3. Pressure ulcer; 4. Mental disorder due to skull defect after massive cerebral infarction surgery; 5. Malnutrition, anemia and hypoproteinemia".

[0024] Step S2: performing text mining on the unstructured text data to extract text features, performing feature fusion on the text features and the structured data, and inputting the fused features into a classification model to obtain a classification result of the service time category to which the patient belongs.

[0025] The objective of this step is to classify patients into two categories: normal service time (for example, service time ≤ 30 minutes) and extended service time (for example, service time > 30 minutes). As Figure 2 shows, it specifically includes the following sub-steps: Step S21: text normalization.

[0026] processing the text content of the "preliminary diagnosis" field. All punctuation marks (such as commas, periods, and semicolons), numeric characters, irrelevant prefixes and suffixes are deleted, all English characters are uniformly converted to lowercase, and all spaces are normalized.

[0027] Step S22: stem extraction and stop word removal.

[0028] using the Natural Language Toolkit (NLTK) in Python to perform stem extraction on the text processed in the previous step, and removing stop words with no practical meaning such as "de" (of) and "le" (past tense marker in Chinese).

[0029] Step S23: meaningless word removal.

[0030] deleting words that appear less than ten times in the dataset. Most of these words contain errors, and it is difficult to understand their intended meanings. By performing this step, obstacles that affect the model from generating accurate predictions are effectively reduced. The fewer meaningless words contained in the dataset, the less likely the prediction model is to be misled.

[0031] Step S24: text digitization.

[0032] The Bag-of-Words (BOW) model was used to convert the text into numerical feature vectors. The BOW model ignores word grammar and word order, focusing only on the frequency of each word in the text. Since the preliminary diagnostic texts in this dataset are mostly incomplete sentences and lists of medical terms with weak semantic relationships between words, the BOW model is suitable and efficient. After conversion, each word becomes a feature dimension, with its value being the number of times it appears in that diagnostic record.

[0033] Step S25: Feature dimensionality reduction.

[0034] After BOW transformation, the feature dimensionality may be very high (e.g., 756 columns). To avoid the curse of dimensionality and overfitting, principal component analysis (PCA) is used for dimensionality reduction. Figure 3 As shown, the cumulative variance contribution rate of principal components is analyzed using the elbow method to select the number of principal components at the inflection point. The specific steps are as follows: (1) Calculate the individual variance for each principal component. The individual variance of each principal component represents the degree of variation in the data.

[0035] (2) Calculate the variation in variance between continuous principal components. This refers to the covariance between principal components, which is used to represent the correlation between them.

[0036] (3) The elbow method was used to determine the inflection point in the variance variation plot. Individual variances and variance variation were analyzed, i.e., a Screen test was performed. The plot shows the variance of each principal component and the percentage of the cumulative variance of these principal components. When an elbow appears on the plot, the number of principal components corresponding to this inflection point is considered to be the optimal dimension of the data, indicating that the contribution of additional principal components to explaining the total variance gradually decreases.

[0037] In this embodiment, the first 50 principal components are selected, which can explain 64% of the total variance, significantly reducing data dimensionality while retaining key information. For example... Figure 3 As shown, this was determined by analyzing the trend charts of individual variance and cumulative variance.

[0038] Step S26: Model classification.

[0039] Feature extraction is performed on the original structured data (such as age and gender) to obtain structured data features. The 50 principal component features obtained after dimensionality reduction are concatenated with the structured data features to form the final feature vector. This feature vector is then input into a pre-trained Gradient Boosting Machine (GBM) classifier to output the patient's category (0 indicates extended service time, 1 indicates normal service time).

[0040] Model training and validation: This embodiment uses GBM as the classification model due to its excellent performance in medical text classification tasks. The core idea of ​​GBM is to iteratively train the model, correcting errors in each iteration and progressively optimizing the model. Before training, the historical dataset (containing 7237 valid records of completed remote consultations) was divided into an 80% training set and a 20% test set. Grid search cross-validation was used to adjust the model's hyperparameters, resulting in a final model with 400 estimators, each with a maximum depth of 10. To address the class imbalance problem in the dataset, Synthetic Minority Oversampling Technique (SMOTE) was used to augment the training set, improving the model's ability to identify the minority class (extended service time).

[0041] The model performance was evaluated using 10-fold cross-validation. The model's performance on the test set is as follows: Figure 4 As shown in Table 1: the accuracy reached 81.85% (the accuracy of the classification model without text features was 49.2%), the sensitivity (the ability to correctly identify patients with extended service hours) was 86.23%, and the specificity (the ability to correctly identify patients with normal service hours) was 79.57%. Figure 5 As shown, the area under the ROC curve (AUC) of the model reaches 90.41%, indicating that the classifier has good performance.

[0042] Table 1. Model performance under different metrics;

[0043] Step S3: Based on the classification results, estimate the probability distribution parameters of service time for each patient. The probability distribution parameters are obtained by fitting the parameters using service time samples within the corresponding intervals in the historical data.

[0044] Based on the classification results of step S2, a different service time probability distribution is set for each patient.

[0045] First, set a duration threshold c (e.g., 30 minutes) to distinguish between normal and extended service times. Simultaneously, based on historical data or expert experience, set a minimum service time a (e.g., 2.99 minutes) and a maximum service time b (e.g., 99.99 minutes).

[0046] If a patient’s service time is classified into the normal service time category (Category 1), then their service time is modeled as following a truncated distribution in the interval [a, c], where the truncated probability distribution is a truncated exponential distribution.

[0047] If a patient’s service time is classified as an extended service time category (category 0), then their service time is modeled as following a truncated distribution in the interval [c, b], wherein the truncated probability distribution is a truncated exponential distribution.

[0048] The probability density function (PDF) of this truncated distribution is: Specifically, it is expressed as: (1); in, Let be a random variable, representing the situation. The service hours for patient i t Indicates time.

[0049] Step S4: Input the probability distribution parameters into the two-stage stochastic programming model to construct a mathematical model that includes decision variables, objective function and constraints. The first stage of the mathematical model is used to make immediate allocation decisions before the actual service time is unknown, and the second stage is used to make retrospective adjustment decisions under multiple discretized stochastic scenarios where the actual service time is known.

[0050] This step only completes the mathematical construction of the model and does not involve solving it; the constructed mathematical model serves as the input for step S5, which is then solved by step S5.

[0051] This step uses the service time probability distribution obtained in step S3 as input to construct a two-stage stochastic programming mathematical model to generate the optimal doctor-patient allocation scheme. The relevant symbols and meanings of the model are shown in Table 2.

[0052] Table 2. Symbols and meanings related to the model;

[0053] The specific model is as follows: First-stage decision-making (immediate decision-making): Decision variables include: y j : A 0-1 variable, indicating whether doctor j provides services during this period. y j A value of 1 indicates that remote doctor j is assigned to provide telemedicine services, while 0 indicates that he / she does not. This invention assumes that doctors are homogeneous, meaning that the quality of their services is the same.

[0054] x ij The variable is 0-1, indicating whether patient i has been pre-assigned to doctor j. It is 1 when patient i is assigned to remote doctor j, and 0 otherwise, indicating that patient i will not be served by remote doctor j.

[0055] u i : A 0-1 variable representing whether patient i is postponed to the next cycle. 1 indicates postponement, and 0 indicates no postponement.

[0056] h j : is the work cost of doctor j within time block T.

[0057] f i : This refers to the unit revenue that assigns patient i to the doctor.

[0058] r i If patient i postpones to the next cycle, there will be a penalty cost.

[0059] The objective function for the first stage is to minimize the deterministic cost: doctor's work cost + patient delay penalty cost - patient service revenue. The formula is expressed below.

[0060] (2); (3); (4); (5); (6); (7); (8); The constraints include: each patient is either assigned or postponed (3); the number of doctors providing services does not exceed a maximum of K (4); and the logical relationships between decision variables (5)(6), where constraint (5) is a symmetry-breaking constraint because the cost parameters are homogeneous. Constraint (6) will affect the variables. x and y To ensure that patient i is only assigned to the doctor j providing the service, and that the total appointment time for each doctor's patients does not exceed the planned time period T (7), constraint (8) defines the variable type, with all variables being 0-1 variables. Among them, service time The classification result of step S3 determines the duration, as shown in formula (9): if it is a normal category, it is set to short duration. (e.g., 10 minutes); if it is an extended category, set it to long duration. (e.g., 30 minutes).

[0061] (9); Second-stage decision-making (retrospective decision-making): First phase decision x ij , y j , u i Once determined, for each possible random scenario ω (Representing one possible value of service time), make a retrospective decision: : 0-1 variables, representing the scene ω Below, should doctor j cancel patient i's appointment? A value of 1 indicates that the patient has been cancelled, and doctor j will no longer provide telemedicine services to patient i in the current period; otherwise, remote doctor j will provide services to patient i as planned.

[0062] Continuous variables, representing variables in the scene ω Below is the overtime hours of Doctor J.

[0063] The objective function of the second stage is to minimize the expected value of overtime costs and patient cancellation penalty costs, as shown in formulas (10) and (11).

[0064]

[0065] st(3)-(6),(8)(10); in, (11); (12); (13); (14); (15); Among them, parameters b j This represents the unit overtime cost for doctor j, and there are relationships between the parameters. h j / T ≤ b j This is because the cost of overtime work is higher than the cost of normal working hours. g i This indicates the cancellation of the penalty cost for patient i.

[0066] The constraints include: the total actual service time of physician j (serving patients who were not cancelled) does not exceed the sum of normal working hours and overtime (12); only pre-assigned patients can be cancelled (13); and overtime does not exceed the maximum allowed overtime duration B (14). Constraint (15) defines the type of decision variable. The second-stage problem is always feasible; in the worst case, decision-makers can cancel all patients.

[0067] S5: Using the sample mean approximation method, the mathematical model constructed in step S4 is transformed into a deterministic mixed integer linear programming model and solved to generate the optimal doctor-patient allocation strategy.

[0068] This embodiment employs the Sample Average Approximation (SAA) method, generating a sufficient number of discrete scenarios Ω (in Python, generating multiple service times conforming to the target distribution) to approximate the true uncertain distribution. The Gurobi solver is then used in Python to solve the aforementioned two-stage stochastic programming mathematical model. The resulting solution... x ij , y j , u i The optimal "immediate" allocation strategy is to specifically assign which doctors, which patients to each doctor, and which patients should be postponed.

[0069] In another embodiment, the feature dimensionality reduction in step S24 can also be replaced by methods such as linear discriminant analysis (LDA) or t-distributed random neighborhood embedding (t-SNE). In addition to GBM, the classification model in step S25 can also be random forest, support vector machine (SVM) or deep learning model (such as TextCNN), as long as it can achieve the purpose of text classification. In the optimization model in step S4, the objective function can also be directly set to maximize the total benefit, and the doctor cost and penalty cost are treated as negative benefits; the deterministic model in formulas (2)-(8) and the stochastic programming model in formulas (10)-(15) can be selected according to the actual operational needs. For example, in scenarios with stable demand and low uncertainty, the deterministic model with less computation can be used.

[0070] Performance verification: To verify the technical effectiveness of this invention, the inventors conducted numerical experiments on a real dataset from a telemedicine center. The experiments compared the two novel strategies proposed in this invention (Strategy II: the machine learning-based deterministic model for the first stage of this scheme; Strategy IV: the two-stage stochastic programming model of this scheme) with existing strategies (Strategy I: fixed duration; Strategy III: a stochastic model based on historical data).k The number of minutes added to the average service time.

[0071] The experimental results are shown in Tables 3 to 6.

[0072] Table 3. Out-of-sample performance comparison of strategy I and strategy II;

[0073] Table 4. Out-of-sample performance comparison of strategy III and strategy IV;

[0074] Table 5. Comparison of strategy FI (the theoretical "idealized optimal strategy", i.e., knowing the actual service time of each patient in advance) and other strategies under complete information;

[0075] Table 6. Delay and cancellation rates for various strategies under different κ values;

[0076] The results show that, compared with the existing fixed-duration strategy (Strategy I), Strategy II of the present invention improves average revenue by 41%-47% and reduces the cancellation rate from 32.4% to 18.0-22.0%. Strategy IV of the present invention performs better in most cases, especially when the service time distribution changes (e.g., becomes a normal distribution or a worst-case distribution), its robustness is significantly better than Strategy III. For example, under a normal distribution, the average revenue of Strategy IV is 42% higher than that of Strategy III. These experimental data fully demonstrate the significant beneficial effects of the proposed method in reducing costs, improving revenue, and increasing resource utilization efficiency.

[0077] Example 2 This embodiment provides a remote medical patient allocation system based on text mining and stochastic programming, including: The data acquisition module is configured to acquire patient application information, which includes structured data and unstructured text data. The text mining and classification module is configured to: perform text mining on unstructured text data to extract text features, and then fuse the text features with structured data before inputting them into the classification model to obtain the classification results of the service time category to which the patient belongs; The distribution parameter determination module is configured to estimate the probability distribution parameters of service time for each patient based on the classification results, wherein patients belonging to different service time categories are set to follow truncated probability distributions within different intervals; The optimization model building module is configured to: input probability distribution parameters into a two-stage stochastic programming model to build a mathematical model containing decision variables, objective function and constraints. The first stage of the mathematical model is used to make immediate allocation decisions before the actual service time is unknown, and the second stage is used to make retrospective adjustment decisions under multiple discretized stochastic scenarios where the actual service time is known. The solution module is configured to use the sample mean approximation method to transform the mathematical model constructed by the optimization model construction module into a deterministic mixed integer linear programming model and solve it to generate the optimal doctor-patient allocation strategy.

[0078] It should be noted that the above modules correspond to the steps in Embodiment 1, and the examples and application scenarios implemented by the above modules and their corresponding steps are the same, but are not limited to the content disclosed in Embodiment 1. It should also be noted that the above modules can be executed in a computer system as part of the system.

[0079] In further embodiments, the following is also provided: An electronic device includes a memory and a processor, as well as computer instructions stored in the memory and running on the processor, which, when executed by the processor, perform the method described in Embodiment 1. For brevity, further details are omitted here.

[0080] It should be understood that in this embodiment, the processor can be a central processing unit (CPU), or it can be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor.

[0081] A computer-readable storage medium for storing computer instructions that, when executed by a processor, perform the method of Embodiment 1.

[0082] The method in Example 1 can be directly executed by a hardware processor, or it can be executed by a combination of hardware and software modules within the processor. The software modules can reside in readily available storage media in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, or registers. This storage medium is located in memory; the processor reads information from the memory and, in conjunction with its hardware, completes the steps of the above method. To avoid repetition, a detailed description is not provided here.

[0083] A computer program product includes a computer program that, when executed by a processor, implements the method in Embodiment 1.

[0084] The present invention also provides at least one computer program product tangibly stored on a non-transitory computer-readable storage medium. The computer program product includes computer-executable instructions, such as instructions included in program modules, which execute in a device on a target real or virtual processor to perform the processes / methods described above. Typically, program modules include routines, programs, libraries, objects, classes, components, data structures, etc., that perform specific tasks or implement specific abstract data types. In various embodiments, the functionality of program modules can be combined or divided among program modules as needed. The machine-executable instructions for the program modules can execute within a local or distributed device. In a distributed device, the program modules can reside in both local and remote storage media.

[0085] The computer program code used to implement the methods of the present invention may be written in one or more programming languages. This computer program code may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the computer or other programmable data processing device, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a computer, partially on a computer, as a stand-alone software package, partially on a computer and partially on a remote computer, or entirely on a remote computer or server.

[0086] In the context of this invention, computer program code or related data may be carried by any suitable carrier to enable a device, apparatus, or processor to perform the various processes and operations described above. Examples of carriers include signals, computer-readable media, and the like. Examples of signals may include electrical, optical, radio, sound, or other forms of propagation signals, such as carrier waves, infrared signals, etc.

[0087] Those skilled in the art will recognize that the units and algorithm steps described in conjunction with the embodiments herein can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0088] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.

Claims

1. A telemedicine doctor-patient allocation method based on text mining and stochastic programming, characterized in that, Includes the following steps: S1: Obtain the patient's application information, which includes structured data and unstructured text data; S2: Perform text mining on the unstructured text data to extract text features, and then fuse the text features with the structured data and input them into the classification model to obtain the classification result of the service time category to which the patient belongs; S3: Based on the classification results, estimate the probability distribution parameters of service time for each patient, wherein patients belonging to different service time categories are set to follow truncated probability distributions within different intervals; S4: Input the probability distribution parameters into a two-stage stochastic programming model to construct a mathematical model containing decision variables, objective function and constraints. The first stage of the mathematical model is used to make an immediate allocation decision before the actual value of the service time is unknown, and the second stage is used to make a retrospective adjustment decision under multiple discretized stochastic scenarios where the actual value of the service time is known. S5: Using the sample mean approximation method, the mathematical model constructed in step S4 is transformed into a deterministic mixed integer linear programming model and solved to generate the optimal doctor-patient allocation strategy.

2. The telemedicine doctor-patient allocation method based on text mining and stochastic programming as described in claim 1, characterized in that, Step S2 specifically includes: The unstructured text data is processed by text normalization, stemming, and stop word removal. The bag-of-words model is used to convert the processed text data into numerical feature vectors; Principal component analysis is used to reduce the dimensionality of the numerical feature vector and extract a predetermined number of principal components as text features. The extracted text features are concatenated with the structured data to form the final fused feature vector. The final feature vector is input into the gradient booster classification model, which outputs the classification result of whether the patient belongs to the normal service time category or the extended service time category.

3. The telemedicine doctor-patient allocation method based on text mining and stochastic programming as described in claim 1, characterized in that, In step S3, the truncation probability distribution is a truncation exponential distribution; the step of estimating the probability distribution parameters of the service time specifically includes: Obtain the preset duration threshold c, minimum service time a, and maximum service time b; For patients classified as having normal service time, their service time is set to follow a first-cut-off exponential distribution within the interval [a, c]. For patients classified as having extended service time, their service time is set to follow a second cutoff exponential distribution within the interval [c, b]. The rate parameters of the first and second truncated exponential distributions are respectively obtained by estimating the parameters using service time samples within the corresponding intervals of historical data.

4. The telemedicine doctor-patient allocation method based on text mining and stochastic programming as described in claim 1, characterized in that, In step S4, the objective function of the mathematical model is to minimize the expected sum of the costs of the first stage and the costs of the second stage; The decision variables in the first stage include: variables instructing whether a doctor provides services, variables instructing whether a patient is assigned to a specific doctor, and variables instructing whether a patient postpones to the next cycle; the costs in the first stage include the doctor's fixed operating costs and the penalty costs for patient postponement, minus the benefits of serving the patient. The decision variables in the second stage include: variables indicating whether a patient is canceled under each random scenario, and continuous variables indicating the length of time a doctor works overtime; the costs in the second stage include the cost of doctors working overtime and the cost of penalties for canceling patients.

5. The telemedicine doctor-patient allocation method based on text mining and stochastic programming as described in claim 4, characterized in that, In step S5, the specific steps for transforming the mathematical model into a deterministic mixed-integer linear programming model using the sample mean approximation method include: Based on the cutoff probability distribution set for each patient in step S3, a predetermined number of N random scenarios are generated using Monte Carlo sampling. Each scenario contains the service time realized value of all patients in that scenario. Substituting the N random scenarios into the mathematical model, the expected term of the second-stage cost is approximated by the sample mean of the second-stage cost under the N random scenarios, thereby obtaining a deterministic mixed-integer linear programming model. The deterministic mixed-integer linear programming model is solved by calling the mixed-integer linear programming solver.

6. The telemedicine doctor-patient allocation method based on text mining and stochastic programming as described in claim 1, characterized in that, The method further includes step S6: outputting and executing the optimal doctor-patient allocation strategy; The optimal doctor-patient allocation strategy includes at least: the set of doctors selected to provide services in the current planning cycle, the list of patients pre-assigned to each doctor, and the list of patients postponed to the next planning cycle. During the execution phase, if the total actual service time of any doctor's pre-assigned patients exceeds the sum of the doctor's planned working hours and the preset maximum overtime hours, then the appointments of some of the doctor's patients will be cancelled according to the predetermined priority rules.

7. A remote medical patient allocation system based on text mining and stochastic programming, characterized in that, include: The data acquisition module is configured to acquire the patient's application information, which includes structured data and unstructured text data. The text mining and classification module is configured to: perform text mining on the unstructured text data to extract text features, and then fuse the text features with the structured data and input them into the classification model to obtain the classification result of the service time category to which the patient belongs; The distribution parameter determination module is configured to: estimate the probability distribution parameters of service time for each patient based on the classification results, wherein patients belonging to different service time categories are set to follow truncated probability distributions within different intervals; The optimization model building module is configured to: input the probability distribution parameters into a two-stage stochastic programming model to build a mathematical model containing decision variables, objective function and constraints. The first stage of the mathematical model is used to make an immediate allocation decision before the actual value of the service time is unknown, and the second stage is used to make a retrospective adjustment decision under multiple discretized stochastic scenarios where the actual value of the service time is known. The solution module is configured to use the sample mean approximation method to transform the mathematical model constructed by the optimization model construction module into a deterministic mixed integer linear programming model and solve it to generate the optimal doctor-patient allocation strategy.

8. An electronic device, characterized in that, It includes a memory and a processor, as well as computer instructions stored in the memory and running on the processor, which, when executed by the processor, perform the method according to any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, Used to store computer instructions, which, when executed by a processor, perform the method described in any one of claims 1-6.

10. A computer program product, characterized in that, Includes a computer program, which, when executed by a processor, implements the method described in any one of claims 1-6.

Images