Doctor scheduling optimization method for internet hospital

Abstract

The invention discloses a doctor scheduling optimization method for an internet hospital, and the method comprises the steps: firstly, obtaining personal information inputted by a patient when the patient initiates an online inquiry process, and chief complaint information about illness state description; then, the waiting duration of the patients and the illness state levels of the patients after the patients initiate the inquiry process are monitored, and the real-time waiting priority of each patient is calculated according to the waiting duration and the illness state levels; calculating the priority level of the patient per minute to obtain an updating result; and finally, according to the feature matrix of the patient side, searching for a similar patient group, then obtaining a candidate doctor list, and adopting a weighted Topsis algorithm to reorder the doctors on the Internet hospital. And according to the received real-time waiting priorities of the patients, carrying out final sorting on the doctors, and according to a final sorting result, carrying out scheduling optimization on the online doctors of the Internet hospital. According to the invention, the online inquiry order reception can be completed more quickly, and the doctor-seeing experience of the patient is improved.

Description

Technical Field

[0001] The present invention belongs to the field of intelligent medical technology, and in particular relates to a doctor scheduling optimization method for an Internet hospital. Background Art

[0002] Internet hospitals are a new medical treatment model that is becoming increasingly mature under the current integration of information technology and medical care. With the help of Internet hospitals, patients can go through a series of processes such as online registration, consultation, medical treatment, and drug delivery, and can complete an online consultation and follow-up visit without leaving home.

[0003] At present, online consultations in Internet hospitals are all conducted in fragmented time. The business process is as follows: patients log in to the Internet hospital system, select online departments, select online doctors, register, and doctors see patients during their free time, and start online consultations. If a doctor does not see a patient after a certain time, the order will be automatically canceled; because many doctors are busy with their daily work and slow to see patients, many patients have registered with designated doctors, but have not been seen for a long time, resulting in timeout cancellation, and the patients are told to register with other doctors, which in turn affects the patient's online medical experience. Summary of the invention

[0004] To solve the above problems, the purpose of the present invention is to provide a doctor scheduling optimization method for an Internet hospital, which dynamically adjusts the list of candidate doctors according to the patient's waiting priority and recommends them to the patient as early as possible so that the patient can switch in time and receive the consultation as soon as possible.

[0005] In order to achieve the above-mentioned invention object, the present invention adopts the following technical scheme:

[0006] A method for optimizing doctor scheduling in an Internet hospital comprises the following steps:

[0007] S1. Obtaining personal information and chief complaint information about the description of the condition entered by the patient when initiating the online consultation process; the chief complaint information includes text information and the number of pictures;

[0008] S2. Monitor the patient's waiting time T_wait and the patient's condition level C_severity after the patient initiates the consultation process, calculate the real-time waiting priority P_priority of each patient according to the patient's waiting time T_wait and the patient's condition level C_severity; and calculate the patient's priority level every minute to obtain an updated result;

[0009] S3. Based on the feature matrix on the patient side, find similar patient groups, obtain a list of candidate doctors, and use the weighted Topsis algorithm to reorder the doctors on the Internet hospital. According to the real-time waiting priority P_priority of the received patients, finally sort the doctors, and according to the final sorting results, optimize the scheduling of online doctors in the Internet hospital.

[0010] Furthermore, in the above step S2, the patient's condition level C_severity is comprehensively judged according to the character length of the condition description and whether there is a picture attachment, and the calculation formula is:

[0011]

[0012] Among them, T_length represents the character length of the description of the condition entered by the patient; image_count represents the number of image attachments uploaded by the patient.

[0013] Furthermore, in the above step S2, the patient waiting time T_wait, in minutes, is adjusted through a dynamic weighting mechanism.

[0014] T_wait = (current time - registration time) * W uergency

[0015] Among them, W uergency Indicates dynamic allocation based on the patient's condition level C_severity;

[0016]

[0017] Furthermore, in the above step S2, the real-time waiting priority P_priority of the patient is obtained by comprehensively calculating the patient's waiting time T_wait and the patient's condition level C_severity; the operation is:

[0018] S2.1. Normalize the patient's condition level C_severity and the patient's waiting time T_wait;

[0019] Normalized waiting time T_wait':

[0020]

[0021] Among them, Tmax represents the maximum waiting time;

[0022] Normalized patient condition level C_serverity':

[0023]

[0024] S2.2. Combine the normalized waiting time T_wait' and the normalized patient condition level C_serverity' to calculate the final patient's real-time waiting priority P_priority. The calculation formula is:

[0025] P_priority=w1·T_wait′+w2·C_serverity′

[0026] Among them, w1 is the weight of waiting time; w2 is the weight of disease level;

[0027] The above factors are monitored once every minute, and are comprehensively calculated through the above algorithm to obtain the patient's real-time waiting priority P_priority.

[0028] Furthermore, the above step S3 includes the following sub-steps:

[0029] S3.1. Collect online information of Internet hospital patients

[0030] When a patient makes an online appointment in an Internet hospital and the doctor fails to see the patient for a preset period of time, the feature vector V of the patient’s online information is collected. Online_Data ,

[0031] V Online_Data = { gender ,ν age ,ν Symptoms ,ν Location ,ν habits_vec}

[0032] Among them, ν gender Indicates the patient's gender; age represents the actual age characteristics of the patients after standardized processing; ν Symptoms Indicates the patient's main complaint; Location represents the patient's geographical location; habits_vec Indicates the patient's health habits;

[0033] S3.2. Collect offline information of patients and construct fusion feature matrix

[0034] Based on the patient's medical treatment code, the feature vector V of offline information is obtained from the offline information system of the medical institution Offline_Data ,

[0035] V Offline_Data = { past_history ,ν present_illness ,ν surgical_history ,ν inspection ,ν medications}

[0036] Among them, νpast_history Indicates the patient's medical history; present_illness Indicates the patient's current medical history; surgical_history represents the patient's surgical and treatment history; inspection Indicates the patient's examination and test information; medications represents the patient's medication use record;

[0037] According to the feature vector V of the online information Online_Data and the feature vector V of offline information Offline_Data , construct the merged data matrix V input ,

[0038]

[0039] The feature vector V of the online information is Online_Data and the feature vector V of offline information Offuline_Data Perform SVD compression;

[0040] S3.3. Find similar patients based on the combined feature matrix

[0041] From the Internet hospital’s previous patient database, find similar patient groups and calculate the data matrix V for each previous patient input , find the TOP N most similar patient groups through matrix weighted cosine similarity calculation; the operation is:

[0042] The weighted cosine similarity calculation formula is:

[0043]

[0044] Among them, n represents the total number of patients in the previous patient database;

[0045] v new ,i represents the value of the new patient on the i-th feature;

[0046] v historcal ,i represents the value of the historical patient in the database on the i-th feature;

[0047] w i represents the weight of the i-th feature, that is, the weight of online and offline features;

[0048] Based on the patient’s real-time waiting priority P_priority, w i Perform dynamic adjustments to dynamically adjust the weights of online features;

[0049] S3.4. Obtain a list of candidate doctors based on similar patient groups

[0050] Obtain the list of TOP N doctors based on the medical records of the same department of TOP N patients; at the same time, determine whether the doctor is online at the Internet hospital, delete the offline doctors from the list of candidate doctors, and sort the candidate doctors for the first time;

[0051] According to the ranking proportion of different patients, each doctor is assigned a weight, and finally a re-ranked list of candidate doctors is obtained;

[0052] The higher the ranking of the patient, the higher the weight of the associated doctor. The basic weight of the doctor W is calculated using the following formula: d :

[0053]

[0054] Where N represents the total number of patients; rank i Indicates the ranking of the patient in TOP N; C id Indicates whether the i-th patient is related to doctor d. If so, C id is 1, if there is no association, C id is 0;

[0055] S3.5. Obtain doctor-related features and perform secondary sorting using the weighted Topsis method

[0056] S3.51. Based on the list of candidate doctors, obtain four characteristics of each doctor: admission rate, professional compliance, average consultation time, and praise rate;

[0057] S3.52. Data Standardization

[0058] The characteristic value of each doctor is normalized, and the normalization formula is as follows:

[0059]

[0060] Among them, X is the original eigenvalue; X max and X min Represent the maximum and minimum values ​​of the eigenvalues ​​respectively;

[0061] S3.53. Constructing a weighted decision matrix

[0062] The decision matrix R is in the form of:

[0063]

[0064] Among them, w′1 represents the initialization weight of the average consultation time; w′2 represents the initialization weight of professional compliance; w′3 represents the initialization weight of the favorable comment rate; w′4 represents the initialization weight of the consultation rate; X ij Represents the value of the i-th sample on the j-th index;

[0065] S3.54. Calculation of ideal solutions and negative ideal solutions

[0066] According to the properties of each column indicator, calculate the ideal solution A + and negative ideal solution A - :

[0067]

[0068] Among them, w j represents the weight of the jth indicator;

[0069] S3.55, distance calculation

[0070] Compute the Euclidean distance of each doctor from the ideal solution and the negative ideal solution:

[0071]

[0072] in, represents the Euclidean distance between each doctor and the ideal solution; m represents the total number of decision indicators; represents the value of the jth evaluation indicator in the ideal solution; represents the Euclidean distance between each doctor and the negative ideal solution; represents the value of the jth evaluation indicator in the negative ideal solution;

[0073] S3.56, Calculate relative closeness

[0074] Based on the distance between the ideal solution and the negative ideal solution, calculate the relative closeness C i

[0075]

[0076] The greater the relative closeness, the closer the doctor is to the ideal solution and the higher the recommendation priority;

[0077] S3.6. According to the received real-time waiting priority P_priority of the patient, the doctor ranking is dynamically adjusted again.

[0078] Furthermore, the above step S3.1 includes:

[0079] (1) ν gender Use binary coding: 1 for male and 0 for female;

[0080]

[0081] (2) ν age Expressed as numerical values, through standardization,

[0082]

[0083] Where age represents the actual age of the patient; μ age and σ age They represent the mean and standard deviation of age, respectively, which are obtained by calculating the information of this patient and the information of patients who have successfully visited the hospital in the past;

[0084] (3) ν Symptoms Stored in natural language, converted into vectors using WordEmbedding technology,

[0085] ν Symptoms =Word2Vec(symptoms)

[0086] (4) ν Location Taking provinces as units, we use unique hot encoding to represent the provinces in China. Let P be the set of all possible provinces. For any province p i ∈P, the index in the set is i;

[0087] For a specific province p k , whose one-hot encoding vector ν Location It is expressed as:

[0088]

[0089] ν Location is a vector of length |P|, where the i-th position of the vector is 1 if and only if i=k, and all other positions are 0;

[0090] (5) ν habits_vec Including but not limited to the patient's eating habits, smoking and drinking conditions, and exercise frequency; this information is natural language text entered by the patient and represented by a TF-IDF vector.

[0091] Furthermore, the above step S3.51 includes:

[0092] (1) Consultation rate: This is the ratio of the number of successful consultations by a doctor to the total number of registrations within a certain period of time, reflecting the availability of the doctor;

[0093]

[0094] (2) Professional Compatibility: This measures the degree of match between the doctor’s specialist qualifications and the current registration department, and is determined by analyzing information including but not limited to the doctor’s scope of practice, research direction, and published papers; the operation is as follows:

[0095] First, obtain the doctor information, registration department information, and mapping table;

[0096] Secondly, the doctor information and the registration department information are converted into feature vectors for similarity calculation; the scope of practice and department classification are represented by one-hot encoding, the research direction and diagnosis and treatment keywords are represented by word2vec, and the practice experience is converted into a standardized score;

[0097] Finally, the doctor feature vector is matched with the registration department vector, and the semantic similarity is calculated using the Doc2Vec model. The overall professional conformity is calculated by combining the similarities of multiple features:

[0098] Professional conformity w2 = α· similarity of practice scope + β· similarity of scientific research direction + γ· practice experience score, where α, β, and γ are weight coefficients, which are adjusted through historical data;

[0099] (3) Average consultation time: The doctor’s historical consultation records are collected to calculate the average duration of each consultation, which is used to evaluate the doctor’s efficiency;

[0100]

[0101] (4) Rate of favorable comments: Based on the patient evaluation system, the doctor’s performance in terms of patient satisfaction is obtained, calculated as the ratio of favorable comments to the total number of comments;

[0102]

[0103] Furthermore, the above step S3.6 includes the following sub-steps:

[0104] S3.61. Dynamically adjust the doctor's admission weight

[0105] According to the patient's real-time waiting priority P_priority, adjust the weight of the doctor's "acceptance rate" feature;

[0106] S3.62. Update of the physician priority list

[0107] The adjusted weights will be used to recalculate the comprehensive score of each doctor and generate an updated doctor priority list; the updated priority list will be pushed to the Internet hospital's scheduling system in real time;

[0108] S3.63. Intelligent scheduling is performed based on the updated doctor priority list.

[0109] Furthermore, the above step S3.62 performs the following operations:

[0110] (1) Push doctors: According to the updated doctor list, the recommended doctors are dynamically adjusted for waiting patients, and patients can modify their appointments with one click;

[0111] (2) Automatic polling: The list of patients who have not been seen will continue to be dynamically adjusted to ensure that the waiting time for each patient is as short as possible;

[0112] (3) Feedback mechanism: After each consultation, the patient’s waiting time, doctor’s response time, and final consultation duration are recorded to optimize subsequent scheduling strategies.

[0113] A doctor scheduling optimization system for an Internet hospital, comprising:

[0114] The data collection module is used to obtain personal information and chief complaint information about the description of the condition entered by the patient when initiating the online consultation process; the chief complaint information includes text information and the number of pictures;

[0115] The real-time monitoring module is used to monitor the patient's condition level C_severity and the patient's waiting time T_wait after the patient initiates the consultation process, and calculate the real-time waiting priority P_priority of each patient according to the patient's waiting time T_wait and the patient's condition level C_severity; and calculate the patient's priority level every minute to obtain an updated result;

[0116] The dynamic scheduling module is used to find similar patient groups based on the feature matrix of the patient side, then obtain a list of candidate doctors, and use a weighted Topsis algorithm to reorder the doctors on the Internet hospital; finally sort the doctors according to the real-time waiting priority P_priority of the patients received, and according to the final sorting results, optimize the scheduling of online doctors in the Internet hospital.

[0117] Due to the adoption of the above-mentioned technical solution, the present invention has the following advantages:

[0118] The doctor scheduling optimization method and system of the Internet hospital of the present invention utilizes a real-time scheduling module to calculate the patient priority at the minute level, transmits the priority information to a dynamic scheduling module, can dynamically update the doctor's weight, find out the doctor who is most likely to see the patient, and thus improve the seeing rate; the weighted cosine similarity calculation method adopted not only considers the directional similarity of each feature vector, but also reflects the importance of different medical features in diagnosis and treatment through weights, is suitable for processing feature vectors after SVD dimension reduction, and can effectively capture the key medical feature similarity between patients; the patient's priority is dynamically updated by the real-time scheduling module, and the seeing rate weight is appropriately increased according to the priority, so that the seeing of online consultation orders can be completed faster, and the medical experience of patients can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0119] Figure 1 It is a flow chart of the doctor scheduling optimization method of the Internet hospital of the present invention;

[0120] Figure 2 It is a structural block diagram of the doctor scheduling optimization system of the Internet hospital of the present invention. DETAILED DESCRIPTION

[0121] The technical solution of the present invention is further described in detail below through the drawings and embodiments.

[0122] like Figure 1 As shown, a doctor scheduling optimization method for an Internet hospital includes the following steps:

[0123] S1. Obtain the personal information and the main complaint information about the description of the condition entered by the patient when initiating the online consultation process; the consultation information includes text information and the number of pictures;

[0124] S2. Monitor the patient's waiting time T_wait and the patient's condition level C_severity after the patient initiates the consultation process, calculate the real-time waiting priority P_priority of each patient according to the patient's waiting time T_wait and the patient's condition level C_severity; and calculate the patient's priority level every minute to obtain an updated result;

[0125] The patient's condition level C_severity is comprehensively judged based on the length of the condition description and whether there are picture attachments (the longer the character length and the more picture attachments, the higher the level), and is divided into 1 to 3 levels, with 3 being the highest. The calculation formula is:

[0126]

[0127] T_length indicates the length of the patient's condition description in characters; image_count indicates the number of image attachments uploaded by the patient, including but not limited to X-rays and CT scans;

[0128] The patient waiting time T_wait, in minutes, is adjusted through a dynamic weighting mechanism;

[0129] T_wait = (current time - registration time) * W uergency

[0130] Among them, W uergency Indicates dynamic allocation based on the patient's condition level C_severity;

[0131]

[0132] For example, the registration time of a mild patient (C_serverity=1) is 10:00, and the current time is 11:00. The waiting time T_wait is:

[0133] T_wait = (11:00-10:00) × 0.8 = 60 minutes × 0.8 = 48 minutes;

[0134] The real-time waiting priority P_priority of the patient is obtained by comprehensively calculating the patient's waiting time T_wait and the patient's condition level C_severity; the operation is:

[0135] S2.1 Since the patient's condition level C_severity and the patient's waiting time T_wait have different value ranges, in order to balance the impact of the two, the patient's condition level C_severity and the patient's waiting time T_wait are normalized;

[0136] Normalized waiting time T_wait':

[0137]

[0138] Wherein, Tmax represents the maximum waiting time, preferably, Tmax is 24 hours, i.e. 1440 minutes;

[0139] Normalized patient condition level C_serverity':

[0140]

[0141] Since the severity level C_severity can only take integer values ​​1, 2, and 3, it is directly converted into a continuous value, and severity level 3 is considered the most severe;

[0142] S2.2. Combining the normalized waiting time T_wait' and the normalized patient condition level C_serverity', the final patient real-time waiting priority P_priority calculation formula is:

[0143] P_priority=w1·T_wait′+w2·C_serverity′

[0144] Among them, w1 is the weight of the waiting time; w2 is the weight of the disease level; the value of w1 is 0~1, the value of w2 is 0~1, and w1+w2=1;

[0145] The above factors are monitored once every minute, and the above algorithm is used to perform comprehensive calculations to obtain the patient's real-time waiting priority P_priority;

[0146] For example, suppose a patient with severe illness C_severity=3 has a waiting time of 720 minutes, w1 is set to 0.8, and w2 is set to 0.2. The calculation process is:

[0147] Normalized waiting time T_wait'=720÷1440=0.5;

[0148] Normalized disease severity level

[0149] Get the real-time waiting priority P_priority=0.8*0.5+0.2*1.0=0.4+0.2=0.6;

[0150] S3. Find similar patient groups based on the feature matrix of the patient side, obtain a list of candidate doctors, and use the weighted Topsis algorithm to reorder the doctors on the Internet hospital. Perform a final sorting of the doctors based on the real-time waiting priority P_priority of the received patients. According to the final sorting result, schedule and optimize the online doctors of the Internet hospital to complete the online consultation or consultation as soon as possible. The operation is as follows:

[0151] S3.1. Collecting patients’ online information

[0152] When a patient makes an online appointment in an Internet hospital and the doctor does not see the patient for a preset period of time (e.g., 12 hours), the feature vector V of the patient's online information is collected. Online_Data ,

[0153] V Online_Data = { gender ,ν age ,ν Symptoms ,ν Location ,ν habits_vec}

[0154] Among them, ν gender Indicates the patient's gender; age represents the actual age characteristics of the patients after standardized processing; ν Symptoms Indicates the patient's main complaint; Location represents the patient's geographical location, that is, the patient's place of residence; habits_vec Indicates the patient's health habits;

[0155] (1) ν gender Use binary coding: 1 for male and 0 for female;

[0156]

[0157] (2) ν age Expressed as numerical values, normalized to eliminate differences in dimension and range,

[0158]

[0159] Among them, age represents the actual age of the patient, μ age and σ age They represent the mean and standard deviation of age, respectively, which are obtained by calculating the information of this patient and the information of patients who have successfully visited the hospital in the past;

[0160] Eliminate dimension effects: In data, the numerical ranges of different variables may vary greatly. For example, age may be between 0 and 100, while other variables may be binary variables, that is, 0 or 1. This difference will cause variables with a larger range to dominate during model training or calculation, affecting model performance. Standardization can ensure that the contribution of the age variable to the model will not be overly amplified due to its large numerical range, thereby balancing the weights of other variables.

[0161] (3) ν Symptoms Stored in natural language, converted into vectors using WordEmbedding technology,

[0162] ν Symptoms =Word2Vec(symptoms)

[0163] (4) ν Location There may be correlation between disease type and epidemiology, specifically in terms of provinces, and one-hot encoding is used to represent provinces in the country; one-hot encoding creates a new feature column for each province in the dataset, and for each row, only the column corresponding to the province in that row is marked as 1, and the columns of the remaining provinces are marked as 0;

[0164] Let P be the set of all possible provinces. For any province p i ∈P, whose index in the set is i;

[0165] For a specific province p k , whose one-hot encoding vector ν Location It is expressed as:

[0166]

[0167] Here, ν Location is a vector with a length equal to |P| (the total number of provinces), where the i-th position of the vector is 1 if and only if i=k, and all other positions are 0;

[0168] (5) ν habits_vec Including but not limited to the patient's eating habits, smoking and drinking conditions, and exercise frequency; this information is input by the patient and is in the form of natural language text, represented by TF-IDF vectors;

[0169] S3.2. Collect offline information of patients and construct fusion feature matrix

[0170] Based on the patient's medical treatment code, the feature vector V of offline information is obtained from the offline information system of the medical institution Offline_Data ,

[0171] V Offline_Data = { past_history ,ν present_illness ,ν surgical_history ,ν inspection ,ν medications}

[0172] Among them, ν past_history Represents the patient's medical history, that is, the patient's previous medical condition, including the diseases diagnosed and the treatments received, which is free text data; present_illness The patient's current medical history refers to the course of the patient's condition before this visit, including the onset of symptoms, the course of change, and previous treatment responses; surgical_history The patient's surgical and treatment history, which is a detailed record of the type, time and results of previous surgeries and other treatments the patient received; inspection Indicates the patient's examination and test information, that is, the type, time and results of the examination and test the patient received; medications It indicates the patient's medication use record, i.e. the name, dosage, and frequency of medication currently and in the past used by the patient;

[0173] According to the feature vector V of the online information Online_Data and the feature vector V of offline information Offline_Data , construct the merged data matrix V input ,

[0174]

[0175] Since the feature vector V of the online information Online_Data and the feature vector V of offline information Offline_Data are all long vectors composed of five sub-information, respectively representing the feature vector V of the online information Online_Data and the feature vector V of offline information Offline_Data SVD compression can keep the lengths of the two long vectors consistent and increase the computing speed.

[0176] S3.3. Find similar patients based on the combined feature matrix

[0177] From the Internet hospital’s previous patient database, find similar patient groups and calculate the data matrix V for each previous patient input ; Through matrix weighted cosine similarity calculation, find the top N most similar patient groups; the operation is:

[0178] The weighted cosine similarity calculation formula is:

[0179]

[0180] Among them, n represents the total number of patients in the previous patient database;

[0181] v new ,i represents the value of the new patient on the i-th feature;

[0182] v hisorical ,i represents the value of the historical patient in the database on the i-th feature;

[0183] w i represents the weight of the i-th feature, that is, the weight of online and offline features;

[0184] In the present invention, both online features and offline features contain 5 feature values, and the initial weights are all 0.1. Based on the patient's real-time waiting priority P_priority, w i Perform dynamic adjustments to dynamically adjust the weights of online features;

[0185] P_priority value Online feature weights before adjustment Adjusted online feature weights P_priority>0.75 (high priority) Adjusted to 0.15 Adjust to 0.05 0.5 < P_priority ≤ 0.75 (Normal processing) Adjust to 0.1 Adjust to 0.1 P_priority≤0.5 (low priority) Adjust to 0.05 Adjusted to 0.15

[0186] S3.4. Obtain a list of candidate doctors based on similar patient groups

[0187] Obtain the list of TOP N doctors based on the same department medical records of TOP N patients (the same department means the same department as the one registered online this time); at the same time, determine whether the doctor is online in the Internet hospital. If not, it means that the doctor is not efficient at seeing patients at this time, and delete the doctor from the list of candidate doctors, thereby improving scheduling optimization and ranking the candidate doctors for the first time;

[0188] According to the ranking proportion of different patients, each doctor is assigned a weight, and finally a re-ranked list of candidate doctors is obtained;

[0189] The higher the ranking of the patient, the higher the weight of the associated doctor. The basic weight of the doctor W is calculated using the following formula: d :

[0190]

[0191] Where N represents the total number of patients; rank i Indicates the ranking of the patient in TOP N; C id Indicates whether the i-th patient is related to doctor d. If so, C id is 1, if there is no association, C id is 0;

[0192] S3.5. Obtain doctor-related features and perform secondary sorting using the weighted Topsis method

[0193] S3.51. Based on the list of candidate doctors, obtain four characteristics of each doctor: admission rate, professional compliance, average consultation time, and praise rate;

[0194] (1) Consultation rate: This is the ratio of the number of successful consultations by a doctor to the total number of registrations within a certain period of time, reflecting the availability of the doctor;

[0195]

[0196] (2) Professional Compatibility: This measures the degree of match between the doctor’s specialist qualifications and the current registration department, and is determined by analyzing information including but not limited to the doctor’s scope of practice, research direction, and published papers; the operation is as follows:

[0197] First, obtain the doctor's information (including practice scope, scientific research direction, and practice experience), the registration department information (including the department to which the current registration belongs and the corresponding diagnosis and treatment scope), and the mapping table (the mapping rules between professional departments and doctor's practice direction and scientific research field, which are used to evaluate the matching degree between the two);

[0198] Secondly, the doctor information and the registration department information are converted into feature vectors for similarity calculation; the scope of practice and department classification are represented by one-hot encoding, the research direction and diagnosis and treatment keywords are represented by word2vec, and the practice experience is converted into a standardized score;

[0199] Finally, the doctor feature vector is matched with the registration department vector, and the semantic similarity is calculated using the Doc2Vec model. The overall professional conformity is calculated by combining the similarities of multiple features:

[0200] Professional conformity w2 = α· similarity of practice scope + β· similarity of scientific research direction + γ· practice experience score, where α, β, and γ are weight coefficients, which are adjusted through historical data;

[0201] (3) Average consultation time: The doctor’s historical consultation records are collected to calculate the average duration of each consultation, which is used to evaluate the doctor’s efficiency;

[0202]

[0203] (4) Rate of favorable comments: Based on the patient evaluation system, the doctor’s performance in terms of patient satisfaction is obtained, calculated as the ratio of favorable comments to the total number of comments;

[0204]

[0205] S3.52. Data Standardization

[0206] In order to eliminate the dimensional differences between features, the above eigenvalues ​​are normalized. The normalization formula is as follows:

[0207]

[0208] Among them, X is the original eigenvalue; X max and X min Represent the maximum and minimum values ​​of the eigenvalues ​​respectively;

[0209] S3.53. Constructing a weighted decision matrix

[0210] Combined with actual clinical needs and feature importance, an initial weight is assigned to each feature;

[0211] The decision matrix R is in the form of:

[0212]

[0213] Among them, w′1 represents the initialization weight of the average consultation time; w′2 represents the initialization weight of professional compliance; w′3 represents the initialization weight of the favorable comment rate; w′4 represents the initialization weight of the consultation rate; X ij Represents the value of the i-th sample on the j-th index;

[0214] S3.54. Calculation of ideal solutions and negative ideal solutions

[0215] According to the attributes of each column indicator (positive or negative), calculate the ideal solution A + (Ideal Solution) and negative ideal solution A - (NegativeIdeal Solution):

[0216]

[0217] Among them, w j represents the weight of the jth indicator;

[0218] S3.55, distance calculation

[0219] Compute the Euclidean distance of each doctor from the ideal solution and the negative ideal solution:

[0220]

[0221] in, represents the Euclidean distance between each doctor and the ideal solution; m represents the total number of decision indicators, that is, the number of evaluation criteria (for example, consultation time, praise rate, etc.); represents the value of the jth evaluation indicator in the ideal solution; represents the Euclidean distance between each doctor and the negative ideal solution; represents the value of the jth evaluation index in the negative ideal solution, that is, the worst value of each evaluation criterion in the negative ideal solution;

[0222] The ideal solution mentioned above refers to the best possible value of each evaluation indicator, which usually represents a theoretically optimal state. The smaller the gap between the scores of all samples and the ideal solution, the closer the sample is to the ideal state.

[0223] S3.56, Calculate relative closeness

[0224] Based on the distance between the ideal solution and the negative ideal solution, calculate the relative closeness C i

[0225]

[0226] The greater the relative closeness, the closer the doctor is to the ideal solution and the higher the recommendation priority;

[0227] S3.6. Dynamically adjust the doctor ranking again based on the real-time waiting priority P_priority of the received patient

[0228] S3.61. Dynamically adjust the doctor's admission weight

[0229] According to the patient's real-time waiting priority P_priority, the weight of the doctor's "admission rate" feature is adjusted to achieve a doctor ranking that better meets the patient's needs. Preferably, the specific adjustment strategy is:

[0230] P_priority value Adjusted reception rate feature weights P_priority>0.75 (high priority) Adjust the reception rate feature weight to 0.4 0.5 < P_priority ≤ 0.75 (Normal processing) Adjust the reception rate feature weight to 0.3 P_priority≤0.5 (low priority) Adjust the feature weight of the reception rate to 0.2

[0231] Through the above adjustments, the doctor ranking can be dynamically changed at different waiting levels, so that patients with long waiting times can be seen more quickly;

[0232] S3.62. Update of the physician priority list

[0233] The adjusted weights will be used to recalculate the comprehensive score of each doctor and generate an updated doctor priority list. The updated priority list will be pushed to the Internet hospital's scheduling system in real time to assign appropriate doctors to patients.

[0234] S3.63, Intelligent Scheduling Execution

[0235] Based on the updated physician priority list, do the following:

[0236] (1) Push doctors: According to the updated doctor list, the recommended doctors are dynamically adjusted for waiting patients, and patients can modify their appointments with one click;

[0237] (2) Automatic polling: The list of patients who have not been seen will continue to be dynamically adjusted to ensure that the waiting time for each patient is as short as possible;

[0238] (3) Feedback mechanism: After each consultation, the patient’s waiting time, doctor’s response time, and final consultation duration are recorded to optimize subsequent scheduling strategies.

[0239] like Figure 2 As shown, a doctor scheduling optimization system for an Internet hospital includes:

[0240] The data collection module is used to obtain personal information and chief complaint information about the description of the condition input by the patient when initiating the consultation process; the chief complaint information includes text information and the number of pictures;

[0241] The real-time monitoring module performs the operation in step S2 of the above-mentioned doctor scheduling optimization method of the Internet hospital, and is used to monitor the patient's waiting time T_wait and the patient's condition level C_severity after the patient initiates the consultation process, and calculates the real-time waiting priority level P_priority of each patient according to the patient's waiting time T_wait and the patient's condition level C_severity; and calculates the patient's priority level every minute to obtain an updated result;

[0242] The dynamic scheduling module executes the operation in step S3 of the above-mentioned doctor scheduling optimization method of the Internet hospital, and is used to find similar patient groups according to the feature matrix of the patient side, and then obtain a list of candidate doctors, and use the weighted Topsis algorithm to re-sort the doctors on the Internet hospital; according to the real-time waiting priority P_priority of the received patients, the doctors are finally sorted, and according to the final sorting result, the online doctors of the Internet hospital are scheduled and optimized to complete the online consultation or diagnosis as soon as possible.

[0243] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for optimizing doctor scheduling in an Internet hospital, characterized by: It includes the following steps: S1. Obtaining personal information and chief complaint information about the description of the condition entered by the patient when initiating the online consultation process; the chief complaint information includes text information and the number of pictures; S2. Monitor the patient's waiting time T_wait and the patient's condition level C_severity after the patient initiates the consultation process, calculate the real-time waiting priority P_priority of each patient according to the patient's waiting time T_wait and the patient's condition level C_severity; and calculate the patient's priority level every minute to obtain an updated result; S3. Based on the feature matrix on the patient side, find similar patient groups, obtain a list of candidate doctors, and use the weighted Topsis algorithm to reorder the doctors on the Internet hospital. According to the real-time waiting priority P_priority of the received patients, finally sort the doctors, and according to the final sorting results, optimize the scheduling of online doctors in the Internet hospital.

2. The method for optimizing doctor scheduling in an Internet hospital according to claim 1 is characterized in that: In step S2, the patient's condition level C_severity is comprehensively judged based on the length of the character description of the condition and whether there is a picture attachment. The calculation formula is: Among them, T_length represents the character length of the description of the condition entered by the patient; image_count represents the number of image attachments uploaded by the patient.

3. The method for optimizing doctor scheduling in an Internet hospital according to claim 1 is characterized in that: In step S2, the patient waiting time T_wait, in minutes, is adjusted through a dynamic weighting mechanism. T_wait = (current time - registration time) * W uergency Among them, W uergency Indicates dynamic allocation based on the patient's condition level C_severity; 4. The method for optimizing doctor scheduling in an Internet hospital according to claim 1 is characterized in that: In step S2, the real-time waiting priority P_priority of the patient is obtained by comprehensively calculating the patient's waiting time T_wait and the patient's condition level C_severity; the operation is: S2.

1. Normalize the patient's condition level C_severity and the patient's waiting time T_wait; Normalized waiting time T_wait': Among them, Tmax represents the maximum waiting time; Normalized patient condition level C_serverity': S2.

2. Combine the normalized waiting time T_wait' and the normalized patient condition level C_serverity' to calculate the final patient's real-time waiting priority P_priority. The calculation formula is: P_priority=w1·T_wait'+w2·C_serverity' Among them, w1 is the weight of waiting time; w2 is the weight of disease level; The above factors are monitored once every minute, and are comprehensively calculated through the above algorithm to obtain the patient's real-time waiting priority P_priority.

5. The method for optimizing doctor scheduling in an Internet hospital according to claim 1 is characterized in that: In step S3, it includes the following sub-steps: S3.

1. Collect online information of Internet hospital patients When a patient makes an online appointment in an Internet hospital and the doctor fails to see the patient for a preset period of time, the feature vector V of the patient’s online information is collected. Online_Data , V Online_Data ={ν gender ,n age ,n Symptoms ,n Location ,n habits_vec } Among them, ν gender Indicates the patient's gender; age represents the actual age characteristics of the patients after standardized processing; ν Symptoms Indicates the patient's main complaint; Location represents the patient's geographical location; habits_vec Indicates the patient's health habits; S3.

2. Collect offline information of patients and construct fusion feature matrix Based on the patient's medical treatment code, the feature vector V of offline information is obtained from the offline information system of the medical institution Offline_Data , V Offline_Data ={ν past_history ,n present_illness ,n surgical_history ,n inspection ,n medications } Among them, ν past_history Indicates the patient's medical history; present_illness Indicates the patient's current medical history; surgical_history represents the patient's surgical and treatment history; inspection Indicates the patient's examination and test information; medications represents the patient's medication use record; According to the feature vector V of the online information Online_Data and the feature vector V of offline information Offline_Data , construct the merged data matrix V input , The feature vector V of the online information is Online_Data and the feature vector V of offline information Offline_Data Perform SVD compression; S3.

3. Find similar patients based on the combined feature matrix From the Internet hospital’s previous patient database, find similar patient groups and calculate the data matrix V for each previous patient input , find the TOP N most similar patient groups through matrix weighted cosine similarity calculation; the operation is: The weighted cosine similarity calculation formula is: Among them, n represents the total number of patients in the previous patient database; v new ,i represents the value of the new patient on the i-th feature; v historical ,i represents the value of the historical patient in the database on the i-th feature; w i represents the weight of the i-th feature, that is, the weight of online and offline features; Based on the patient’s real-time waiting priority P_priority, w i Perform dynamic adjustments to dynamically adjust the weights of online features; S3.

4. Obtain a list of candidate doctors based on similar patient groups Obtain the list of TOP N doctors based on the medical records of the same department of TOP N patients; at the same time, determine whether the doctor is online at the Internet hospital, delete the offline doctors from the list of candidate doctors, and sort the candidate doctors for the first time; According to the ranking proportion of different patients, each doctor is assigned a weight, and finally a re-ranked list of candidate doctors is obtained; The higher the ranking of the patient, the higher the weight of the associated doctor. The basic weight of the doctor W is calculated using the following formula: d : Where N represents the total number of patients; rank i Indicates the patient's ranking in TOPN; C id Indicates whether the i-th patient is related to doctor d. If so, C id is 1, if there is no association, C id is 0; S3.

5. Obtain doctor-related features and perform secondary sorting using the weighted Topsis method S3.

51. Based on the list of candidate doctors, obtain four characteristics of each doctor: admission rate, professional compliance, average consultation time, and praise rate; S3.

52. Data Standardization The characteristic value of each doctor is normalized, and the normalization formula is as follows: Among them, X is the original eigenvalue; X max and X min Represent the maximum and minimum values ​​of the eigenvalues ​​respectively; S3.

53. Constructing a weighted decision matrix The decision matrix R is in the form of: Among them, w′1 represents the initialization weight of the average consultation time; w′2 represents the initialization weight of professional compliance; w′3 represents the initialization weight of the favorable comment rate; w′4 represents the initialization weight of the consultation rate; X ij Represents the value of the i-th sample on the j-th index; S3.

54. Calculation of ideal solutions and negative ideal solutions According to the properties of each column indicator, calculate the ideal solution A + and negative ideal solution A - : Among them, w j represents the weight of the jth indicator; S3.55, distance calculation Compute the Euclidean distance of each doctor from the ideal solution and the negative ideal solution: in, represents the Euclidean distance between each doctor and the ideal solution; m represents the total number of decision indicators; represents the value of the jth evaluation indicator in the ideal solution; represents the Euclidean distance between each doctor and the negative ideal solution; represents the value of the jth evaluation indicator in the negative ideal solution; S3.56, Calculate relative closeness Based on the distance between the ideal solution and the negative ideal solution, calculate the relative closeness C i The greater the relative closeness, the closer the doctor is to the ideal solution and the higher the recommendation priority; S3.

6. According to the received real-time waiting priority P_priority of the patient, the doctor ranking is dynamically adjusted again.

6. The method for optimizing doctor scheduling in an Internet hospital according to claim 5 is characterized in that: Step S3.1 includes: (1) ν gender Use binary coding: 1 for male and 0 for female; (2) ν age Expressed as numerical values, through standardization, Where age represents the actual age of the patient; μ age and σ age They represent the mean and standard deviation of age, respectively, which are obtained by calculating the information of this patient and the information of patients who have successfully visited the hospital in the past; (3) ν Symptoms Stored in natural language, converted into vectors using WordEmbedding technology, ν Symptoms =Word2Vec(symptoms) (4) ν Location Taking provinces as units, we use one-hot encoding to represent the provinces in China. Let P be the set of all possible provinces. For any province p i ∈P, the index in the set is i; For a specific province p k , whose one-hot encoding vector ν Location It is expressed as: ν Location is a vector of length |P|, where the i-th position of the vector is 1 if and only if i=k, and all other positions are 0; (5) ν habits_vec Including but not limited to the patient's eating habits, smoking and drinking conditions, and exercise frequency; this information is natural language text entered by the patient and represented by a TF-IDF vector.

7. The method for optimizing doctor scheduling in an Internet hospital according to claim 5 is characterized in that: Step S3.51 includes: (1) Consultation rate: This is the ratio of the number of successful consultations by a doctor to the total number of registrations within a certain period of time, reflecting the availability of the doctor; (2) Professional Compatibility: This measures the degree of match between the doctor’s specialist qualifications and the current registration department, and is determined by analyzing information including but not limited to the doctor’s scope of practice, research direction, and published papers; the operation is as follows: First, obtain the doctor information, registration department information, and mapping table; Secondly, the doctor information and the registration department information are converted into feature vectors for similarity calculation; the scope of practice and department classification are represented by one-hot encoding, the research direction and diagnosis and treatment keywords are represented by word2vec, and the practice experience is converted into a standardized score; Finally, the doctor feature vector is matched with the registration department vector, and the semantic similarity is calculated using the Doc2Vec model. The overall professional conformity is calculated by combining the similarities of multiple features: Professional conformity w2 = α· similarity of practice scope + β· similarity of scientific research direction + γ· practice experience score, where α, β, and γ are weight coefficients, which are adjusted through historical data; (3) Average consultation time: The doctor’s historical consultation records are collected to calculate the average duration of each consultation, which is used to evaluate the doctor’s efficiency; (4) Rate of favorable comments: Based on the patient evaluation system, the doctor’s performance in terms of patient satisfaction is obtained, calculated as the ratio of favorable comments to the total number of comments; 8. The method for optimizing doctor scheduling in an Internet hospital according to claim 5 is characterized in that: Step S3.6 includes the following sub-steps: S3.

61. Dynamically adjust the doctor's admission weight Adjust the weight of the doctor's "acceptance rate" feature according to the patient's real-time waiting priority P_priority; S3.

62. Update of the physician priority list The adjusted weights will be used to recalculate the comprehensive score of each doctor and generate an updated doctor priority list; the updated priority list will be pushed to the Internet hospital's scheduling system in real time; S3.

63. Intelligent scheduling is performed based on the updated doctor priority list.

9. The method for optimizing doctor scheduling in an Internet hospital according to claim 5 is characterized in that: Step S3.62, perform the following operations: (1) Push doctors: According to the updated doctor list, the recommended doctors are dynamically adjusted for waiting patients, and patients can modify their appointments with one click; (2) Automatic polling: The list of patients who have not been seen will continue to be dynamically adjusted to ensure that the waiting time for each patient is as short as possible; (3) Feedback mechanism: After each consultation, the patient’s waiting time, doctor’s response time, and final consultation duration are recorded to optimize subsequent scheduling strategies.

10. A doctor scheduling optimization system for an Internet hospital, characterized in that: include: The data collection module is used to obtain personal information and main complaint information about the description of the condition entered by the patient when initiating the online consultation process; The main complaint information includes text information and the number of pictures; The real-time monitoring module is used to monitor the patient's condition level C_severity and the patient's waiting time T_wait after the patient initiates the consultation process, and calculate the real-time waiting priority P_priority of each patient according to the patient's waiting time T_wait and the patient's condition level C_severity; and calculate the patient's priority level every minute to obtain an updated result; The dynamic scheduling module is used to find similar patient groups based on the feature matrix of the patient side, then obtain a list of candidate doctors, and use a weighted Topsis algorithm to reorder the doctors on the Internet hospital; finally sort the doctors according to the real-time waiting priority P_priority of the patients received, and according to the final sorting results, optimize the scheduling of online doctors in the Internet hospital.

Images