Diagnosis accompanying service reservation method and system

Abstract

The invention provides a diagnosis accompanying service reservation method and system, relates to the technical field of medical services, and aims to provide whole-process medical accompanying for patients. According to the invention, the whole process from demand analysis to service execution is automatic; according to the method, multi-modal feature modeling is carried out according to the requirements of the patient and the ability of the accompanying diagnosis personnel, and the matching algorithm for scoring based on multi-dimensional weighting is combined to intelligently match the appropriate accompanying diagnosis personnel, so that the accuracy of diagnosis is improved, and the accuracy of diagnosis is improved. Meanwhile, the knowledge graph is adopted to complement the requirements and pre-judge the risk, so that the service efficiency and the individuation level are remarkably improved; through service appointment, scheduling and management, and service process recording, feedback and evaluation functions, the efficiency and quality of accompanying diagnosis service are improved, and individual needs of different patients for accompanying in the process of seeing a doctor are met.

Description

Technical Field

[0001] The present invention relates to the technical field of medical services, and in particular to a method and system for booking accompanying medical services. Background Art

[0002] With the aging of the population and the increasing demand for medical service quality, the demand for accompanying services is growing. Currently, there are many problems with the existing accompanying service reservation methods on the market. For example, the matching is inaccurate, and the matching is simply based on time and location, without considering factors such as the professional skills and service evaluation of the accompanying person, which may result in patients not receiving appropriate accompanying services; the service process is not standardized, and there is a lack of effective tracking and management of the service process, making it difficult for patients to solve problems encountered during the medical treatment process in a timely manner; personalized services are insufficient, and the personalized needs of different patients regarding the gender, age, professional skills, etc. of the accompanying person cannot be met. Therefore, there is an urgent need for an accompanying service reservation method and system that can achieve intelligent, personalized, and full-process management. Summary of the Invention

[0003] The purpose of the present invention is to overcome the shortcomings of the existing technology and provide a method and system for booking accompanying medical services. Through dynamic matching, scheduling algorithms and knowledge graph drive, a method and system for integrating service booking, scheduling and management is realized, thereby improving the efficiency and quality of accompanying medical services, meeting the personalized needs of different patients for accompanying care during medical treatment, and solving the above problems.

[0004] The purpose of the present invention is mainly achieved through the following technical solutions:

[0005] A method for booking an accompanying medical service comprises the following steps:

[0006] Step S1. The patient enters the accompanying service reservation function through the medical platform and completes the patient information. The patient information includes: filled in personal information, selected medical information, accompanying needs, and notes on special requirements for accompanying personnel; among them, medical information includes the medical department and expected date of medical treatment; accompanying needs include whether one or more of registration services, payment services, medication services, and translation services are required.

[0007] Step S2: Analyze and process the patient information, perform multimodal feature modeling based on the medical information and the accompanying medical needs, establish the accompanying medical needs as a demand vector, and find the multidimensional features of the accompanying medical needs.

[0008] Step S3: Obtain the accompanying personnel information of the medical platform, including professional skills, service evaluation, available time, and geographic location; perform multimodal feature modeling based on the accompanying personnel information, establish the accompanying personnel information as a capability vector, and find a multidimensional portrait of the accompanying personnel.

[0009] Step S4: According to the demand vector and the ability vector, adopt a matching algorithm that scores based on multi-dimensional weighting to match the escort demand and the escort personnel information, calculate the comprehensive matching score, sort the escort personnel according to the comprehensive matching score, and generate a candidate list.

[0010] Step S5: Send the candidate list to the patient. The patient can choose an escort personnel or have the system automatically select the escort personnel with the highest matching degree. When the escort personnel is selected, send the candidate list to the patient. The patient can view the information of the escort personnel and independently select a satisfactory escort personnel. If the patient does not make an independent selection, conduct scheduling or select the one with the highest matching degree as the escort personnel.

[0011] Step S6: Generate an appointment arrangement and notify the patient and the escort personnel. After generating and notifying the appointment arrangement, after determining the escort personnel, generate an appointment arrangement and send it to the patient and the escort personnel simultaneously via text message and APP. After receiving the notice, the escort personnel confirm to take the order or raise an objection. If an objection is raised, re-match and return to Step S5.

[0012] Step S7: Conduct information processing and analysis. After receiving the patient information, first analyze the department of visit, obtain the common diseases and corresponding escort requirements of this department through the medical platform, establish a medical knowledge graph, complete the requirements of the medical knowledge graph through the visit process of the department, and use GNN to predict the service risk probability.

[0013] To further optimize the above technical solution, in Step S2, the multi-modal feature modeling refers to modeling the escort demand through multi-modal features, deconstructing the patient information into multi-dimensional features of medical-service-preference-time and space as the demand vector, generating a multi-dimensional demand vector through the TF-IDF text processing algorithm, and finding the multi-dimensional features of the escort demand.

[0014] Step S21: Process the input text of the patient information by segmenting it with a word segmentation tool; filter out the words without actual semantics and retain the keywords; normalize the keywords and map the synonyms to a unified label.

[0015] Step S22: Obtain medical features based on the medical visit information, set the feature dimensions of the medical features. The medical features include common departments, disease labels, and examination items. The medical features are the medical professional requirements directly related to the medical visit department, disease type, and examination items. Obtain service features based on the escort needs, set the feature dimensions of the service features. The service keywords of the service features include basic service labels, special service labels, service duration, and time window. The service features are the specific execution requirements and procedural operations of the escort service. Obtain preference features based on the information of the special requirements for the escort personnel in the remarks, set the feature dimensions of the preference features. The preference keywords of the preference features include personnel attributes, skill labels, and service styles. The preference features are the personalized attribute requirements of the patient for the escort personnel. Obtain spatio-temporal features based on the medical visit information, set the feature dimensions of the spatio-temporal features. The spatio-temporal keywords of the spatio-temporal features include the geographical location of the medical visit department and the time window. The spatio-temporal features are the geographical and time constraints related to the medical visit time and the geographical location of the medical visit department. Step S23: Calculate the keyword scores for all keywords using the TF-IDF text processing algorithm.

[0016] By calculating the word frequency of the keywords, the relative importance of the keywords in the current text of the patient information is measured. The calculation formula is:

[0017]

[0018] Among them, is the target keyword; is the text of the escort needs input by the patient; keyword in the text the number of occurrences.

[0019] By calculating the inverse document frequency of the keywords, the scarcity of the keywords is measured. The calculation formula is:

[0020]

[0021] Among them, is the full set of escort needs texts and all patient needs in the historical data of the medical platform; is the total number of texts; represents the number of texts containing the keyword .

[0022] The keyword score can be calculated by the obtained word frequency and inverse document frequency. The calculation formula is:

[0023]

[0024] Among them, , the higher the value, the stronger the representativeness of the keyword for the current text.

[0025] Step S24: Perform multi-modal feature modeling to obtain a multi-dimensional feature vector as the demand vector. Multiply the keyword scores of each keyword in the medical features, service features, preference features, and spatio-temporal features by the set weights to obtain feature values, and then concatenate the feature values of each keyword in turn to obtain the final demand vector.

[0026] To further optimize the above technical solution, in step S3, performing multi-modal feature modeling based on the escort personnel information means processing the escort personnel information to construct a multi-dimensional portrait of profession-service-spatio-temporal-skill as the ability vector, and using the Word2Vec semantic mapping algorithm to generate a multi-dimensional ability vector.

[0027] To further optimize the above technical solution, in step S3, using the Word2Vec semantic mapping algorithm to generate a multi-dimensional ability vector is to pre-train a medical domain Word2Vec model through the Word2Vec semantic mapping algorithm, obtain vectors from the escort personnel information, and generate word vectors. The establishment of the medical domain Word2Vec model specifically includes:

[0028] The medical domain Word2Vec model predicts the context word from the central word, and the calculation formula is:

[0029]

[0030] where, The input vector of the central word contains its semantic information.

[0031] The central word Predicts the probability distribution of the context word The larger the value indicates The more likely it is to be The context of.

[0032] The weight matrix from the input layer to the hidden layer stores the semantic association strength from the central word to all words.

[0033] The weight matrix from the hidden layer to the output layer maps the hidden layer features to the vocabulary space.

[0034] The bias term, The hidden layer bias term avoids the neuron output being constantly 0. The output layer bias term adjusts the prediction probabilities of each word.

[0035] where, specifically includes:

[0036] The product of the vector and the weight matrix, plus the bias, gives the original input of the hidden layer, which undergoes a linear transformation. The calculation formula is:

[0037]

[0038] The values are compressed to [-1, 1] through the tanh function to highlight key semantic features and perform non-linear activation. The calculation formula is:

[0039]

[0040] Calculate the product of the hidden layer features and the output weights to obtain the original scores of each vocabulary, and map the output layer. The calculation formula is:

[0041]

[0042] Finally, perform probability normalization. Convert the original scores to probabilities through Softmax so that the sum of the probabilities of all vocabularies is 1.

[0043]

[0044] After the training is completed, each row of the weight matrix of the input layer corresponds to the word vector of a vocabulary .

[0045] To further optimize the above technical solution, in step S3, the construction process of the multi-dimensional portrait specifically includes:

[0046] Step S321: The professional portrait obtains the professional skill text according to the professional skills. The nursing qualification level, nursing years, medical knowledge fields of expertise, and unique skills of the professional skills are respectively established into three word vectors, and the three word vectors of the professional skills are obtained through the pre-trained Word2Vec model in the medical field. At this time, the formula for the high-dimensional professional portrait is expressed as:

[0047]

[0048] Aggregate the multi-word vectors and generate professional portrait features through mean aggregation:

[0049]

[0050] Step S322: Perform the service portrait;

[0051] Obtain the service evaluation text according to the service evaluation. Perform sentiment analysis on the service evaluation for keywords, perform semantic mapping, extract positive evaluations, on-time rates, and service completion rates, and obtain the word vectors of the service evaluation through the pre-trained Word2Vec model in the medical field ,

[0052] Then, combine the word vectors with the sentiment weights. At this time, the service portrait features are:

[0053]

[0054] where is the set sentiment weight matrix.

[0055] Step S323: The spatio-temporal portrait obtains word vectors of available time, geographical location, and moving speed through a pre-trained Word2Vec model in the medical field according to the available time and geographical location , and splices them to obtain spatio-temporal portrait features:

[0056]

[0057] Step S324: Splice the professional portrait features, service portrait features, and spatio-temporal portrait features in sequence to obtain an ability vector as a multi-dimensional portrait.

[0058] To further optimize the above technical solution, in step S4, a matching algorithm based on multi-dimensional weighting for scoring is used to match the escort needs with the escort personnel information, calculate the comprehensive matching score, sort the escort personnel according to the comprehensive matching score, and generate a candidate list.

[0059] The matching algorithm dynamically adjusts the weight matrix based on the escort needs and escort personnel information, and calculates the comprehensive matching score using a matching algorithm based on multi-dimensional weighting for scoring.

[0060] The matching algorithm is based on the patient demand vector and the ability vector , and through the formula

[0061]

[0062] calculate the cosine similarity of each dimension between the patient demand vector and the ability vector , and combine the set weight matrix to generate the comprehensive matching score .

[0063] To further optimize the above technical solution, in step S7, a medical knowledge graph is established. The relationship between departments - processes - skills is obtained through the medical platform to establish the graph; the association between departments and diseases is found in the electronic medical records, the association between departments and processes is found through the process specifications in the medical guidelines, and the association between skills and processes is found in the medical literature, thereby establishing a medical knowledge graph.

[0064] Based on the department-disease relationship, when matching, give priority to selecting an escort with experience in that department; based on the disease-skill relationship, automatically complete the hidden skill requirements of the patient; based on the process-skill, generate an execution list for the escort service.

[0065] To further optimize the above technical solution, in step S7, a GNN is used to predict the service risk probability. By using a GNN graph neural network to predict the needs not explicitly expressed by the patient, such as the "depression patient" being associated with the "psychological counseling" skill, the calculation formula for the risk probability is:

[0066] 。

[0067] The present invention also includes an escort service reservation system. The system is applied to an escort service reservation method, including a user module, a data processing module, a matching module, and a service management module that are electrically connected in sequence; wherein,

[0068] The user module includes a patient terminal, an escort terminal, and a storage unit; it is used for the registration and information management of patients and escorts, records patient information, including: filled personal information, selected medical appointment information, escort needs, and information on special requirements for escorts, and records escort information, including professional skills, service evaluations, available time, and geographical location; the storage unit stores patient information, escort information, and the established medical knowledge graph.

[0069] The data processing module preprocesses the patient reservation information and the escort registration information, extracts key features; performs multi-modal feature modeling on the patient information, establishes the escort needs as a demand vector, and finds the multi-dimensional features of the escort needs; obtains the escort information on the medical platform for multi-modal feature modeling, establishes the escort information as an ability vector, and finds the multi-dimensional portrait of the escort.

[0070] The matching module uses a matching algorithm that scores based on multi-dimensional weighting to match the escort needs with the escort information, calculates the comprehensive matching score, sorts the escorts according to the comprehensive matching score, and generates a candidate list.

[0071] The service management module sends the candidate list to the patient. The patient can select an escort or have the system automatically select the escort with the highest matching degree; when an escort is selected, the candidate list is sent to the patient, and the patient can view the escort information and independently select a satisfactory escort; if the patient does not make an independent selection, the system schedules the escort or selects the one with the highest matching degree as the escort; generates an appointment arrangement and notifies the patient and the escort.

[0072] A computer program product includes a computer program. When the computer program is executed by a processor, it implements the steps of any one of the methods in an escort service reservation method.

[0073] In summary, compared with the prior art, the present invention has the following beneficial effects:

[0074] Through the technical solutions of dynamic matching, scheduling algorithms, and knowledge graph driving, the present invention meets the requirements of deep coupling between automated recommended escort personnel and medical scenarios.

[0075] Deep customization of medical scenarios, realizing intelligent complementation of requirements through the knowledge graph of department processes. Compared with traditional keyword matching, the accuracy of requirement parsing is higher. Because of medical coupling, the knowledge graph directly associates department processes with escort skills, reflecting the professionalism of medical scenarios.

[0076] By establishing mathematical modeling, clear algorithm formulas and objective functions, the present invention is made more portable.

[0077] Real-time monitoring of the dynamic environment, the scheduling makes the average response faster and the utilization rate of cross-regional resources is improved.

[0078] Preventing risks in advance, the GNN risk warning reduces the service interruption rate and significantly improves service reliability.

[0079] Through the deep combination of multi-algorithm fusion and medical domain knowledge, the present invention constructs an automated escort service method and system, filling the gaps in dynamic matching, intelligent scheduling, and risk prediction in the prior art, and having significant innovation and practical value. BRIEF DESCRIPTION OF THE DRAWINGS

[0080] The drawings described herein are used to provide a further understanding of the embodiments of the present invention, form a part of this application, and do not limit the embodiments of the present invention. In the drawings:

[0081] Figure 1 It is a step diagram of a method for booking an escort service;

[0082] Figure 2 It is a structural diagram of a system for booking an escort service. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0083] To make the objectives, technical solutions, and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below in combination with embodiments and drawings. The illustrative embodiments of the present invention and their descriptions are only used to explain the present invention and do not limit the present invention. The embodiments of the present application will be described below in combination with the drawings.

[0084] Traditional escort services have three major problems. The feature modeling is simple, only based on shallow features such as time and location for matching, and cannot capture deep needs such as "cancer patients require escort staff with chemotherapy knowledge". The scheduling strategy is static, using fixed rules to allocate orders (such as the nearest principle), lacking dynamic perception of the real-time flow of people in the hospital and the proficiency of personnel skills. Risk prevention and control are lacking, and potential conflicts such as "foreign patients and escort staff having language barriers" cannot be predicted. The present invention provides the following embodiments to solve the above technical problems.

[0085] This embodiment provides a method and system for booking escort services, aiming at specifically targeting the medical scenario. Based on the existing demand for escort services, improvements are made in the method for service booking, scheduling, and management. Through technical solutions driven by dynamic matching, scheduling algorithms, and knowledge graphs, the efficiency and quality of escort services are improved to meet the personalized needs of different patients for escort during the medical treatment process and to meet the deep coupling requirements of automated recommended escort personnel and medical scenarios.

[0086] In Embodiment 1:

[0087] Referring to Figure 1 - Figure 2 , which is the first embodiment of the present invention, this embodiment provides a method and system for booking escort services.

[0088] As Figure 1 shown, this embodiment provides a method for booking escort services, mainly including the following steps:

[0089] Step S1: The patient enters the escort service booking function through the medical platform, fills in the patient information, which includes the filled personal information, selected medical appointment information, escort needs, and information on special requirements for the escort staff; among them, the medical appointment information includes the department to visit and the expected medical appointment date; the escort needs include one or more of whether assistance is required for registration services, payment services, pharmacy services, and translation services.

[0090] Step S2: Analyze and process the patient information, perform multi-modal feature modeling based on the medical appointment information and escort needs, establish the escort needs as a demand vector, and find the multi-dimensional features of the escort needs.

[0091] Step S3: Obtain the escort staff information on the medical platform, including professional skills, service evaluations, available time, and geographical location; perform multi-modal feature modeling based on the escort staff information, establish the escort staff information as an ability vector, and find the multi-dimensional portrait of the escort staff.

[0092] Step S4: According to the demand vector and the ability vector, use a matching algorithm based on multi-dimensional weighting for scoring to match the escort needs with the escort staff information, calculate the comprehensive matching score, sort the escort staff according to the comprehensive matching score, and generate a candidate list.

[0093] Step S5: Send the candidate list to the patient, who can choose an accompanying person or automatically select the accompanying person with the highest matching degree. When selecting an accompanying person, the candidate list is sent to the patient, who can view the accompanying person information and independently select a satisfactory accompanying person. If the patient does not make an independent selection, scheduling will be carried out or the accompanying person with the highest matching degree will be selected.

[0094] Step S6: Generate an appointment and notify the patient and accompanying person. After the appointment is generated and notified, the accompanying person is determined, and the appointment is generated and sent to the patient and accompanying person simultaneously via SMS and APP notification. After receiving the notification, the accompanying person confirms the order or raises an objection. If an objection is raised, re-matching is performed and the process returns to step S5.

[0095] Step S7: Perform information processing and analysis. After receiving the patient information, first analyze the department where the patient is being treated. Obtain the common diseases and corresponding accompanying requirements of the department through the medical platform, establish a medical knowledge graph, complete the medical knowledge graph through the department's treatment process, and use GNN to predict the service risk probability.

[0096] Accompanying services rely primarily on manual scheduling and empirical judgment to complete matching. Traditional appointment systems typically perform simple screening based solely on appointment time and department location, lacking a multi-dimensional analysis of patient needs. For example, when elderly patients require assistance in obtaining medication and mobility support, the system is unable to identify the nursing skills and physical requirements of the accompanying personnel. Existing technologies have not established a medical knowledge association mechanism, resulting in hidden needs not being covered and process omissions being prone to occur during the service process. In addition, the service confirmation process lacks a two-way feedback mechanism. When the accompanying personnel are unable to accept the order, the matching process must be restarted, causing service delays.

[0097] To address these issues, we first analyzed and discovered that insufficient matching accuracy stems from inadequate demand expression and incomplete capability modeling. Traditional systems simplify requirements into combinations of labels, failing to mine deep-level feature associations. By introducing multimodal modeling techniques, we transformed textual information into a computable vector space representation. Furthermore, considering the dynamic nature of medical scenarios, we designed a matching algorithm with adjustable weights to accommodate requirements of varying priorities. To address the reliability issues of service confirmation, we established a feedback mechanism for objections, creating a closed-loop management system. Finally, we combined knowledge graph technology to complete potential requirements and build a risk warning system.

[0098] Therefore, this embodiment proposes a method for booking escort services, including the following steps: The patient fills in personal information, medical appointment information, and escort service requirements through a medical platform; multi-modal feature modeling is performed on the patient information to generate a requirement vector; the information of the escort personnel is obtained and modeled to generate a capacity vector; a multi-dimensional weighted algorithm is used to calculate the matching score to generate a candidate list; the patient selects or the system automatically matches the escort personnel; a reservation notice is generated and objection feedback is processed; a medical knowledge graph is established to predict service risks.

[0099] Among them, multi-modal feature modeling refers to the process of converting unstructured text into structured vectors. The TF-IDF algorithm is used to extract keyword weights, and a requirement vector is constructed through four dimensions: medical-service-preference-time and space. The construction of the capacity vector involves converting the professional qualifications and service records of the escort personnel into numerical features, and using the Word2Vec model to generate semantic vectors. The dynamic weight matrix is used to adjust the matching priorities of each dimension in different scenarios, and weight allocation is achieved through preset rules or machine learning models. The construction of the medical knowledge graph establishes the association relationship between departments, diseases, and processes by parsing electronic medical records and medical guidelines, and specifically uses a graph database to store the relationship data of nodes and edges.

[0100] After the patient submits a reservation request, the system performs word segmentation and feature extraction on the escort service requirement text. For example, "fracture follow-up consultation requires wheelchair assistance" is parsed into the orthopedics label in medical features and the action assistance label in service features. The nursing experience data of the escort personnel is converted into a skill vector through a semantic model, including dimensions such as orthopedic nursing duration and wheelchair proficiency. The matching algorithm calculates the similarity between the requirement vector and the capacity vector in each dimension, and automatically increases the weight of the time and space features in the emergency scenario. After the candidate list is generated, the patient can view the service evaluation details of the escort personnel for manual selection. A dual confirmation mechanism is set in the confirmation link, and the re-matching process is triggered when the escort personnel rejects the order. The knowledge graph automatically associates patients with depression with psychological counseling skills to supplement the requirements that the patient has not explicitly stated.

[0101] Traditional methods can only use keyword matching and cannot handle semantic associations. For example, they cannot recognize the association between "postoperative care" and "wound treatment skills". This solution calculates the semantic similarity through a vector space model and can discover potential matching relationships across labels. Existing systems lack an objection handling mechanism in the service confirmation link. This solution sets a re-matching trigger condition, and when the escort personnel rejects the order, the matching of secondary candidate personnel is automatically started, reducing the number of times of manual intervention. Traditional risk prediction relies on manual experience judgment. This solution analyzes the topological relationship between diseases, processes, and skills through a graph neural network and automatically generates risk warning indicators.

[0102] This embodiment realizes the precise matching of patient needs and the capabilities of escort personnel, solving the problem that traditional systems ignore the relevance of professional skills. The established two-way confirmation mechanism improves the success rate of service order acceptance and reduces process interruptions caused by personnel rejecting orders. The introduction of a medical knowledge graph can automatically complete the escort needs not clearly expressed by patients. For example, it automatically adds a medicine pickup assistance item for patients with limited mobility. The dynamic weight adjustment mechanism adapts to the priority changes in different medical scenarios to ensure that emergency patients are preferentially matched with nearby available personnel.

[0103] As Figure 2 shown, this embodiment also provides an escort service reservation system, including: a user module, a data processing module, a matching module, and a service management module that are electrically connected in sequence; among them,

[0104] The user module includes a patient terminal, an escort personnel terminal, and a storage unit; it is used for the registration and information management of patients and escort personnel, records patient information, including: filled personal information, selected medical appointment information, escort needs, and information on special requirements for escort personnel, and records escort personnel information, including professional skills, service evaluations, available time, and geographical location; the storage unit stores patient information, escort personnel information, and the established medical knowledge graph.

[0105] The data processing module preprocesses the patient reservation information and the escort personnel registration information, extracts key features; performs multi-modal feature modeling on patient information, establishes the escort needs as a demand vector, and finds the multi-dimensional features of the escort needs; obtains the escort personnel information on the medical platform for multi-modal feature modeling, establishes the escort personnel information as a capability vector, and finds the multi-dimensional portrait of the escort personnel.

[0106] The matching module uses a matching algorithm based on multi-dimensional weighting for scoring to match the escort needs with the escort personnel information, calculates the comprehensive matching score, sorts the escort personnel according to the comprehensive matching score, and generates a candidate list.

[0107] The service management module sends the candidate list to the patient. The patient can select an escort personnel or have the system automatically select the escort personnel with the highest matching degree; when an escort personnel is selected, the candidate list is sent to the patient, and the patient can view the escort personnel information and independently select a satisfactory escort personnel; if the patient does not make an independent selection, scheduling is performed or the one with the highest matching degree is selected as the escort personnel; a reservation arrangement is generated and the patient and the escort personnel are notified.

[0108] This embodiment also proposes an accompanying diagnosis service reservation system, which includes a user module, a data processing module, a matching module, and a service management module that are sequentially electrically connected. The user module includes a patient terminal, an accompanying diagnosis personnel terminal, and a storage unit, which is used to manage registration information and store a medical knowledge graph; the data processing module performs multi-modal feature modeling on patient needs and accompanying diagnosis personnel information to generate a demand vector and a capability vector; the matching module uses a multi-dimensional weighted algorithm to calculate a comprehensive matching score and generate a candidate list; the service management module realizes functions such as candidate list push, appointment arrangement generation, and notification.

[0109] Among them, the storage unit of the user module refers to a database system for persistently storing structured data and a knowledge graph, which is implemented by a distributed database cluster and ensures high availability through a data partitioning and replication mechanism. The multi-modal feature modeling of the data processing module refers to a technology for converting unstructured text information into a structured vector representation, which is implemented by natural language processing algorithms, such as a feature extraction method combining TF-IDF and Word2Vec. The multi-dimensional weighted algorithm of the matching module refers to a similarity calculation method for dynamically adjusting the weights of each feature dimension, which is implemented by a mathematical model of multiplying the cosine similarity with a dynamic weight matrix and adapts to different scenario requirements through parameter tuning. The candidate list push mechanism of the service management module refers to a recommendation system based on priority sorting, which is implemented by a message queue and a push service to ensure the real-time and reliable transmission of information.

[0110] The user module collects structured data such as the department to visit and accompanying diagnosis needs filled in by the patient, and at the same time records qualification information such as the qualification certificates and service evaluations of the accompanying diagnosis personnel. The storage unit integrates historical service data and medical guide documents to construct a knowledge graph. The data processing module performs word segmentation on the text "need English translation service" input by the patient, extracts "English translation" as a service feature keyword, and generates a demand vector after calculating its weight through TF-IDF. At the same time, this module analyzes the description of "having an international nursing certification" in the resume of the accompanying diagnosis personnel and maps it to professional portrait features through a pre-trained word vector model. The matching module calculates the cosine similarity between the demand vector and the capability vector, automatically increases the weight coefficient of the spatio-temporal features when the patient selects the emergency department, and generates a candidate list sorted in descending order of matching degree. The service management module pushes the detailed information of the top three accompanying diagnosis personnel to the patient through the APP message channel. If the patient does not make a selection within two hours, the one with the highest matching degree will be automatically assigned, and the appointment time and reception location will be synchronously notified to both parties by text message.

[0111] Traditional escort systems only match personnel based on geographical location and time window, without considering key factors such as professional skills and service evaluation. This system quantifies implicit indicators such as medical professional capabilities and service historical performance through multi-modal feature modeling, solving the problem of poor adaptability caused by single-dimensional matching. Existing technologies using fixed-weight algorithms cannot adapt to different medical scenarios, while this system realizes differential strategies such as priority matching for emergencies and professional matching for difficult and complicated diseases through dynamically adjusting the weight matrix. In addition, the traditional system lacks an artificial confirmation link for the candidate list. This system combines automated recommendation with manual selection, retaining the patient's final decision-making power while ensuring matching efficiency.

[0112] This embodiment achieves high-precision matching between escort personnel and patient needs. For example, pediatric patients visiting the doctor are automatically associated with escort personnel with child care experience. The system can automatically optimize the spatio-temporal weights according to the emergency scenario to ensure quick matching of available personnel nearby. By structurally storing the medical knowledge graph, the implicit needs of patients are automatically supplemented during the matching process. For example, additional requirements for the use of assistive mobile devices are added for orthopedic patients. The two-way confirmation mechanism for the candidate list reduces the dispute rate during the service process, and the objection feedback data is used to continuously optimize the matching model parameters.

[0113] The working principle of this embodiment is:

[0114] Step S1: Fill in patient information; the patient logs in to the medical platform, finds the escort service entry on the home page and clicks to enter. After entering, complete the filling of patient information, including the filled personal information, selected medical appointment information, escort needs, and information on special requirements for escort personnel; among them, the medical appointment information includes the department to visit and the expected medical appointment date; the escort needs include one or more of whether assistance is required for registration service, payment service, pharmacy service, translation service (for foreign patients or special situations); whether there are special requirements for the gender, age, professional skills (such as nursing experience, medical knowledge reserve) of the escort personnel, as well as the appointment time and duration.

[0115] The feature modeling module uses the TF-IDF text processing unit and the Word2Vec semantic mapping unit to generate the escort demand vector and the escort personnel ability vector.

[0116] Step S1 can be applied to the user module of the system. The user module includes a patient terminal, an escort personnel terminal, and a storage unit; it is used for the registration and information management of patients and escort personnel, records the patient's personal basic information, medical appointment needs, reservation information, evaluation information, as well as the professional skills, service evaluation, available time, geographical location, etc. of the escort personnel, and supports personalized settings;

[0117] Step S2: Analyze and process the patient information, perform multi-modal feature modeling based on the medical visit information and accompaniment needs, and establish the accompaniment needs as a demand vector.

[0118] The multi-modal feature modeling models the accompaniment needs through multi-modal features, deconstructs the patient information into medical-service-preference-space-time multi-dimensional features as the demand vector, generates a multi-dimensional demand vector through the TF-IDF text processing algorithm, and finds the multi-dimensional features of the accompaniment needs; processes the input text of the patient information by segmenting it with a word segmentation tool; filters out words without actual semantics, and retains the keywords; normalizes the keywords and maps synonyms to a unified label; obtains the medical features according to the medical visit information, sets the feature dimensions of the medical features, and the medical features include common departments, disease labels, and examination items; the medical features are the medical professional needs directly related to the medical visit department, disease type, and examination items; obtains the service features according to the accompaniment needs, sets the feature dimensions of the service features, and the service keywords of the service features include basic service labels, special service labels, service duration, and time window; the service features are the specific execution requirements and procedural operations of the accompaniment service; obtains the preference features according to the information on the special requirements for the accompaniment personnel in the remarks, sets the feature dimensions of the preference features, and the preference keywords of the preference features include personnel attributes, skill labels, and service styles; the preference features are the personalized attribute requirements of the patient for the accompaniment personnel; obtains the space-time features according to the medical visit information, sets the feature dimensions of the space-time features, and the space-time keywords of the space-time features include the geographical location of the medical visit department and the time window; the space-time features are the geographical and time constraints related to the medical visit time and the geographical location of the medical visit department; calculates the keyword scores for all keywords using the TF-IDF text processing algorithm; measures the relative importance of the keywords in the current text of the patient information by calculating the word frequency of the keywords; measures the scarcity of the keywords by calculating the inverse document frequency of the keywords; calculates the keyword scores from the obtained word frequency and inverse document frequency, and the higher the value, the stronger the representativeness of the keyword for the current text; performs multi-modal feature modeling to obtain a multi-dimensional feature vector as the demand vector; multiplies the keyword scores of each keyword of the medical features, service features, preference features, and space-time features by the set weights to obtain the feature values, and then concatenates the feature values of each keyword in turn to obtain the final demand vector. Specifically as follows:

[0119] Step S21: Process the input text of the patient information by segmenting it with a word segmentation tool; filter out words without actual semantics such as "need", "assist", "um", "want", etc., and retain the keywords; normalize the keywords and map synonyms to a unified label.

[0120] Step S22: Obtain medical features based on the medical visit information. The medical feature dimension refers to a professional requirement label system directly associated with the department of visit, disease type, and examination items, which is implemented using the standardized department names and disease codes in the medical knowledge base and is used to accurately identify the medical professional attributes of patients. Set the medical feature dimension. The medical features include common departments, disease labels, and examination items. Medical features are the medical professional requirements directly related to the department of visit, disease type, and examination items. For example, the department names in the patient's input text (such as "Thoracic Surgery", "Pediatrics"), the keyword descriptions of the disease (such as "Lung Cancer", "Diabetes", "Postoperative Rehabilitation"), and the examination / treatment items (such as "PET-CT", "Chemotherapy", "Dialysis").

[0121] Obtain service features based on the information about special requirements for the escort personnel in the remarks. The service feature dimension refers to the operation processes and time constraints involved in the escort service execution, which is implemented using a preset basic service label and special service label system and is used to decompose the key links in the service process. Set the service feature dimension. The service keywords of the service features include basic service labels, special service labels, service duration, and time window. Service features are the specific execution requirements and procedural operations of the escort service. Basic service labels ("Registration Assistance", "Payment Guidance", "Medication Collection Escort", "Report Interpretation"), special service labels ("Wheelchair Rental", "Dialect Translation", "Foreign Escort", "Psychological Counseling"), service duration (such as "4 hours", "All day"), time window (such as "Starting at 9 am").

[0122] Obtain preference features based on the information about special requirements for the escort personnel in the remarks. The preference feature dimension refers to the personalized requirements of the patient for the attributes and skills of the escort personnel, which is implemented using a personnel attribute classification tree and a skill label library and is used to analyze implicit needs. Set the preference feature dimension. The preference keywords of the preference features include personnel attributes, skill labels, and service styles. Preference features are the personalized attribute requirements of the patient for the escort personnel. Personnel attribute keywords ("Female", "Under 50 years old", "Local escort"), professional skill preferences ("Nursing experience ≥ 5 years", "With first aid certificate", "CET-6"), service style preferences ("Patient and meticulous", "Strong communication ability", "Familiar with medical insurance policies").

[0123] Based on the medical department and expected date of visit, spatiotemporal features are derived. The spatiotemporal feature dimension refers to the constraints of the visit time window and geographic location. This dimension is implemented using timestamp encoding and geographic coordinate conversion, and is used to quantify spatiotemporal matching elements. The spatiotemporal feature dimension is set. The spatiotemporal keywords of the spatiotemporal feature include the geographic location of the visit department and the time window. The spatiotemporal features are geographical and time constraints related to the visit time and the location of the visit department. The address of the visit department (determines the geographic location of the hospital where the department is located). The time window includes the appointment date (such as "2025-04-20"), the time period ("morning", "afternoon", "all day"), and the geographic location preferences of the accompanying personnel (such as "only accompany in the neurology department" and "same administrative district").

[0124] Step S23: Calculate keyword scores for all keywords using the TF-IDF text processing algorithm. The TF-IDF text processing algorithm is a keyword weight calculation method based on word frequency and inverse document frequency. It combines local word frequency statistics with global corpus analysis to quantify the representativeness and scarcity of keywords in the demand text.

[0125] The relative importance of keywords in the current text of patient information is measured by calculating the frequency of keywords. The calculation formula is:

[0126]

[0127] in, is the target keyword; It is the text of the patient's request for accompanying consultation; Keywords In the text The number of times it appears in .

[0128] For example, if "varicocele ligation" appears twice, The value changes inversely with the total number of words in the text.

[0129] The scarcity of keywords is measured by calculating the inverse document frequency of keywords. The calculation formula is:

[0130]

[0131] in, It is the complete set of accompanying medical consultation requirements and also all the patient requirements in the historical data of the medical platform; is the total number of texts; Represents keywords The number of texts.

[0132] If "laparoscopic surgery" only appears in 5% of the texts, then The value is higher.

[0133] The keyword scores can be calculated from the obtained word frequencies and inverse document frequencies. The calculation formula is as follows:

[0134]

[0135] where , and the higher the value, the stronger the representativeness of the keyword for the current text.

[0136] Step S24: Perform multi-modal feature modeling to obtain a multi-dimensional feature vector as the demand vector; multiply the keyword scores of each keyword of the medical features, service features, preference features, and spatio-temporal features by the set weights to obtain feature values, and then concatenate the keyword feature values of each keyword in turn to obtain the final demand vector.

[0137] Step S2 can be applied to the patient side of the user module of the system to support patient registration, login, filling and management of personal information, medical needs, appointment information, evaluation information, etc. Patients can perform personalized settings, such as setting commonly visited hospitals, preferred characteristics of accompanying personnel, etc. Patient information is stored in the storage unit of the user module of the system. The accompanying demand text input by the patient is first semantically segmented by a word segmentation tool, and stop words without actual semantics are filtered out, and the core keywords are retained. For example, "Department of Cardiology" may be segmented into "cardiovascular" and "internal medicine", and unified into a standardized department name through synonym mapping. Then, medical features are extracted from the medical information respectively. For example, "angina pectoris reexamination" is mapped to a disease label and an examination item; service features are extracted from the accompanying demand. For example, "assist in registering and picking up medicine" is decomposed into basic service labels; preference features are extracted from special requirements. For example, "possess first aid skills" is converted into a skill label; spatio-temporal features are extracted from the medical treatment time and location. The weight score of each keyword is calculated by the TF-IDF algorithm. For example, the frequently occurring "registration" obtains a higher word frequency score, but if the word is prevalent in the historical data, the inverse document frequency score decreases, while the low-frequency professional term "coronary angiography" obtains a higher comprehensive score due to its scarcity. Finally, the feature values of the four dimensions are weighted and concatenated according to the preset weights to form a structured demand vector.

[0138] Traditional accompanying demand processing only relies on simple keyword matching and does not establish a multi-dimensional feature classification system, resulting in the omission of key elements such as professional attributes and spatio-temporal constraints. Existing methods use fixed word frequency statistics and cannot distinguish the importance differences between high-frequency general words and low-frequency professional words. For example, "examination" and "magnetic resonance imaging" may be treated equally. This solution constructs a multi-dimensional feature framework of medical-service-preference-spatio-temporal, combined with the dynamic weight calculation of TF-IDF, so that professional terms and service process elements can be accurately quantified and characterized, solving the problem of incomplete feature extraction caused by unstructured text.

[0139] This embodiment realizes the multi-dimensional and precise analysis of the accompanying diagnosis needs of patients, converts the free text description into a structured feature vector, and ensures the complete extraction of key elements such as medical professional attributes, service process requirements, personalized preferences, and spatio-temporal constraints. This solution eliminates the expression differences through semantic normalization processing, and uses a statistical weighting algorithm to distinguish the importance of features, providing accurate and quantifiable input data for subsequent intelligent matching, effectively avoiding the matching deviation caused by information fragmentation.

[0140] Step S3: Obtain the information of the accompanying diagnosis personnel on the medical platform, including professional skills, service evaluation, available time, and geographical location.

[0141] Analyze and process the information filled in by the accompanying diagnosis personnel, perform multi-modal feature modeling based on professional skills, service evaluation, available time, and geographical location, establish the information of the accompanying diagnosis personnel as an ability vector, and find the multi-dimensional portrait of the accompanying diagnosis personnel.

[0142] The acquisition of the information of the accompanying diagnosis personnel is through the need for the accompanying diagnosis personnel to fill in detailed personal information when registering on the medical platform, including:

[0143] Professional skills, nursing qualification certificate level, nursing years, medical knowledge fields of expertise, unique skills; service evaluation, scores and feedback from past patients, the scores are given by patients, and the feedback content can include service attitude and professional ability; available time, specific to morning, afternoon, or a certain time period, geographical location, the community where the accompanying diagnosis personnel are located, and the distance from the hospital for medical treatment.

[0144] The multi-modal feature modeling of the ability of the accompanying diagnosis personnel processes the information of the accompanying diagnosis personnel to construct a multi-dimensional portrait of profession-service-spatio-temporal-skills, and uses the Word2Vec semantic mapping algorithm to generate a multi-dimensional ability vector.

[0145] Generating a multi-dimensional ability vector using the Word2Vec semantic mapping algorithm is to pre-train a medical domain Word2Vec model through the Word2Vec semantic mapping algorithm, obtain vectors from the information of the accompanying diagnosis personnel, and generate word vectors.

[0146] Word2Vec converts natural language into low-dimensional dense vectors (word embeddings) through a neural network language model, so that semantically related words have similar distances in the vector space.

[0147] Step S31: The medical domain Word2Vec model is established to predict context words from the central word, and the calculation formula is:

[0148]

[0149] Among them, Central word The input vector, such as the input vector of the central word (e.g., the field of expertise is "Emergency Department"), contains its semantic information.

[0150] Central word Predict context words The probability distribution, The larger the value indicates The more likely it is to be The context of.

[0151] The weight matrix from the input layer to the hidden layer stores the semantic association strength from the central word to all words.

[0152] The weight matrix from the hidden layer to the output layer maps the hidden layer features to the vocabulary space.

[0153] Bias term, The hidden layer bias term avoids the neuron output being constantly 0, The output layer bias term adjusts the prediction probabilities of each word.

[0154] Among them, specifically including:

[0155] Taking the Emergency Department as an example, we calculate the product of the "Emergency Department" vector and the weight matrix, add the bias, and obtain the original input of the hidden layer, performing a linear transformation. The calculation formula is:

[0156]

[0157] Compress the value to [-1, 1] through the tanh function, highlight the key semantic features, and perform non-linear activation. The calculation formula is:

[0158]

[0159] The activation value of features related to "first aid" is relatively high.

[0160] Calculate the product of the hidden layer features and the output weights to obtain the original scores of each word, and map to the output layer. The calculation formula is:

[0161]

[0162] The score of "cardiopulmonary resuscitation" is 5.8, and the score of "urology" is -2.1.

[0163] Finally, perform probability normalization, convert the original scores to probabilities through Softmax, and make the sum of all word probabilities equal to 1,

[0164]

[0165] The probability of "cardiopulmonary resuscitation" is 0.9, and the probability of "urology department" is 0.05.

[0166] After the training is completed, each row of the input layer weight matrix corresponds to the word vector of a vocabulary; in the accompanying diagnosis scenario, the word vector of "N1-level nurse" is the row vector in the matrix with the index of "N1-level nurse" .

[0167] Step S321: The professional portrait obtains the professional skill text according to the professional skills, and respectively establishes four word vectors for the nursing qualification level, nursing years, medical knowledge fields of expertise, and unique skills of the professional skills. The four word vectors of the professional skills are obtained through the pre-trained medical field Word2Vec model. At this time, the high-dimensional professional portrait formula is expressed as:

[0168]

[0169] Aggregate the multi-word vectors and generate professional portrait features through mean aggregation:

[0170]

[0171] Step S322: Conduct a service portrait.

[0172] Obtain the service evaluation text according to the service evaluation, conduct sentiment analysis on the service evaluation for keywords such as arriving on time, being patient in communication, and being proficient in getting medicine, perform semantic mapping, extract positive evaluations, on-time rate, and service completion rate, and obtain the word vectors of the service evaluation through the pre-trained medical field Word2Vec model ,

[0173] Then combine the word vectors with the sentiment weights. At this time, the service portrait features:

[0174] [[ID=�6]]

[0175] where is the set sentiment weight matrix.

[0176] Step S323: The spatio-temporal portrait obtains the word vectors of available time, geographical location, and moving speed through the pre-trained medical field Word2Vec model according to the available time and geographical location , and splices them to obtain the spatio-temporal portrait features .

[0177] Step S324: Splice the professional portrait features, service portrait features, and spatio-temporal portrait features in sequence to obtain the ability vector as the multi-dimensional portrait.

[0178] Step S3 can be applied to the end of the escort staff in the user module of the system. When registering, the escort staff needs to go through identity verification and qualification review, fill in information such as professional skills, service evaluations, available time, geographical location, etc., and can update and manage their own information, view appointment orders, service records, and evaluation information. The information of the escort staff is stored in the storage unit of the user module of the system. The method first extracts keywords such as the level of nursing qualification certificate and the number of years of nursing from the professional skills text of the escort staff, and generates initial word vectors through a pre-trained Word2Vec model in the medical field. In the linear transformation stage from the input layer to the hidden layer, the word vectors are multiplied by the weight matrix and the bias term is added. For example, "5 years of nursing experience" is converted into a numerical feature containing a time dimension. Subsequently, a non-linear transformation is performed through the tanh function to screen out high-value features such as "good at postoperative care" and suppress interference from irrelevant information. The weight matrix of the output layer maps the features of the hidden layer to the service ability dimension. For example, "fast moving speed" is mapped into a spatio-temporal feature vector. Finally, a probability distribution is generated through Softmax, so that "emergency department escort experience" has a higher matching weight when matching the needs of emergency department visits. After training is completed, the word vectors in the weight matrix of the input layer constitute the basis of the ability vector of the escort staff.

[0179] When the traditional method uses a general semantic model to process medical texts, it cannot accurately distinguish the professional differences between "postoperative care" and "daily care", and low-frequency skill features are easily ignored. Through dedicated training data in the medical field and hidden layer bias design, this solution enables precise vector expressions for professional service capabilities such as "blood dialysis escort". The linear feature extraction method in the prior art is difficult to capture the non-linear association between "psychological counseling" and "depression patients", while the tanh activation function of this solution can effectively model such complex relationships.

[0180] This embodiment realizes the precise quantitative expression of the professional skills of the escort staff. For example, converting "third-level nursing qualification" into a computable feature value supports dynamic matching with the needs of patients. The semantic association strength of medical terms is dynamically adjusted through the weight matrix. For example, establishing a feature association between "orthopedic rehabilitation" and "operation of physical therapy instruments". Low-frequency but important skill features are retained through the bias term. For example, the ability of "sign language translation" can still be effectively recognized when matching deaf patients. Non-linear activation processing makes the feature space have better separability. For example, distinguishing the service difficulty differences between "ordinary medicine pick-up service" and "pick-up and delivery of special cold-chain drugs".

[0181] Steps S2 and 3 are also processed based on the data processing module to preprocess the patient appointment information and the escort staff registration information and extract key features.

[0182] The data processing module cleans, classifies, and extracts features from the appointment information of patients and the registration information of accompanying personnel, converting unstructured data into structured data for algorithm processing. For example, converting the accompanying needs of patients into specific labels such as "registration assistance" and "translation service", and converting the professional skills of accompanying personnel into labels such as "nursing experience" and "medical knowledge", and then mapping the labels into vectors through the TF-IDF text processing algorithm and the Word2Vec semantic mapping algorithm.

[0183] Step S4: Use a matching algorithm based on multi-dimensional weighting for scoring to match the accompanying needs with the information of accompanying personnel, calculate the comprehensive matching score, sort the accompanying personnel according to the comprehensive matching score, and generate a candidate list. The matching algorithm is based on the patient demand vector and the ability vector, calculates the cosine similarity of each dimension between the patient demand vector and the ability vector through a formula, and generates a comprehensive matching score in combination with the set weight matrix.

[0184] The matching algorithm dynamically adjusts the weight matrix based on the accompanying needs and the information of accompanying personnel, and the calculation of the comprehensive matching score uses a matching algorithm based on multi-dimensional weighting for scoring.

[0185] The matching algorithm is based on the patient demand vector and the ability vector , through the formula

[0186]

[0187] calculate the cosine similarity of each dimension between the patient demand vector and the ability vector , and generate a comprehensive matching score in combination with the set weight matrix .

[0188] And when calculating the matching weights, integrate the dynamic weighted matching algorithm and the context awareness module, and support the weights of the emergency-routine scenarios to be adaptively adjusted according to real-time data and verified by the recommended accompanying personnel and the accompanying personnel finally actually selected by the patient.

[0189] Step S4 can be applied to the matching module. The matching module uses a multi-dimensional weighted scoring algorithm for matching to generate a matching result; uses the above multi-dimensional weighted scoring algorithm for matching, and dynamically generates a matching result according to the real-time updated information of patients and accompanying personnel. The algorithm can continuously learn and optimize, adjust the weights of each dimension by analyzing historical matching data and evaluation results, and improve the accuracy of matching.

[0190] ​In step S4, first, the medical appointment information, accompanying needs, and special requirements filled in by the patient are transformed into a demand vector through multi-modal feature modeling. At the same time, the professional skills, service evaluations, and spatio-temporal information of the accompanying personnel are transformed into an ability vector. Subsequently, the cosine similarity between the demand vector and the ability vector is calculated for each of the four dimensions of medical care, service, preference, and spatio-temporal. Among them, the medical dimension evaluates the matching degree between the professional ability of the accompanying personnel and the patient's condition, the service dimension evaluates the fit between the service item execution ability and the patient's needs, the preference dimension evaluates the compliance between the personnel attributes and the patient's personalized requirements, and the spatio-temporal dimension evaluates the coincidence degree between the geographical location and the time window. Further, a dynamic weight is assigned to each dimension through a preset weight matrix. For example, a higher weight is assigned to the spatio-temporal dimension in the emergency scenario, and a higher weight is assigned to the medical dimension in the chronic disease follow-up scenario. Finally, the similarity of each dimension is multiplied by the corresponding weight and then summed to generate a comprehensive matching score and sort the candidates according to the score to generate a candidate list.

[0191] Traditional accompanying care matching methods only perform single-dimensional screening based on the appointment time or department location, without considering the in-depth relationship between service capabilities and patient needs, resulting in the matching results being unable to meet the requirements of complex medical scenarios. This solution realizes a comprehensive evaluation of medical professional capabilities, service execution levels, personalized preferences, and spatio-temporal constraints through multi-dimensional feature modeling and dynamic weight adjustment, effectively solving the problem of insufficient accuracy caused by single-dimensional matching.

[0192] This embodiment realizes multi-dimensional precise matching between accompanying care needs and personnel capabilities, adapts to the matching strategy requirements of different medical scenarios through dynamic weight configuration, uses a comprehensive scoring mechanism to quantitatively evaluate the service suitability of accompanying care personnel, and the finally generated candidate list not only ensures the objectivity of the matching results but also reserves an autonomous selection space for patients.

[0193] Step S5: Send the candidate list to the patient, and the patient can select the accompanying care personnel or have the most highly matched accompanying care personnel automatically selected.

[0194] When selecting the accompanying care personnel, the candidate list is sent to the patient, and the patient can view the detailed information of the accompanying care personnel, including photos, service evaluations, professional skill introductions, and past service cases, and autonomously select a satisfactory accompanying care personnel; if the patient does not make an autonomous selection, the one with the highest matching degree is selected as the accompanying care personnel.

[0195] Step S6: Generate an appointment arrangement and notify the patient and the accompanying care personnel.

[0196] Reservation Arrangement Generation and Notification: After determining the accompanying medical staff, generate a reservation arrangement. The content of the reservation arrangement includes the appointment time, location, information of the accompanying medical staff, and service process, and simultaneously send it to the patient and the accompanying medical staff via text message, APP notification, and email; after receiving the notification, the accompanying medical staff confirms taking the order or raises an objection. If an objection is raised, re-match and return to step S5.

[0197] Steps S5 and S6 can be applied in the service management module to generate reservation arrangements. In the actual implementation process, the service management module also includes scheduling and monitoring of the accompanying medical service, recording the service process, handling patient feedback, and having a service warning function; among them,

[0198] The generation of the reservation arrangement is based on the matching result and the information of the patient and the accompanying medical staff, generates a detailed reservation arrangement, and ensures the accuracy of information such as time and location.

[0199] Service scheduling and monitoring is to monitor the progress of the accompanying medical service in real time, such as whether the accompanying medical staff arrives on time, whether the service is carried out as planned, etc. When unexpected situations such as the accompanying medical staff being late or suddenly falling ill occur, the system triggers the service warning function, automatically finds a backup accompanying medical staff for replacement, and promptly notifies the patient.

[0200] Step S7: Conduct information processing and analysis. After receiving the patient information, first analyze the department of visit, obtain the common diseases and corresponding accompanying medical needs of this department through the medical platform, establish a medical knowledge graph through the relationship between department - process - skill obtained from the medical platform, and construct a knowledge framework through the association between department and disease in the electronic medical record, the association between department and process in the medical guide, and the association between skill and process in the medical literature; based on the department - disease relationship, preferentially match accompanying medical staff with experience in the corresponding department, automatically complete the skill requirements not clearly proposed by the patient based on the disease - skill relationship, and generate a standardized execution list for the accompanying medical service based on the process - skill relationship. For example, for some departments, the accompanying medical staff may need to have certain nursing knowledge to better assist the patient.

[0201] Establish medical knowledge graph reasoning, automatically complete the requirements of the medical knowledge graph through the department visit process (such as pediatrics is associated with "getting medicine + medical order interpretation"), and use GNN to predict the service risk probability.

[0202] Establish a graph by obtaining the relationship between department - process - skill through the medical platform; find the association between department and disease in the electronic medical record, find the association between department and process through the process specifications in the medical guide, and find the association between skill and process in the medical literature, so as to establish a medical knowledge graph.

[0203] The medical knowledge graph module constructs a department - process - skill association graph to achieve enhanced requirement analysis and risk warning.

[0204] In the medical knowledge graph, through the department-disease relationship, when matching, the experience escort in this department is preferentially selected; through the disease-skill relationship, the implicit skill requirements of patients are automatically completed; through the process-skill, an execution list of escort services is generated.

[0205] The GNN is used to predict the service risk probability. By using the GNN graph neural network to predict the unexpressed needs of patients, such as the "depression patient" being associated with the "psychological counseling" skill, the calculation formula of the risk probability: the patient needs "Russian communication" and the escort does not have this skill → the risk probability is 100%.

[0206] A medical knowledge graph refers to a semantic network that associates medical entities through structured data. It is realized by using data extraction tools to extract the co-occurrence relationship between departments and diseases from electronic medical records, parsing the mapping rules between process steps and departments from medical guidelines, and mining the collaborative mode between skills and processes from medical literature. Through knowledge fusion technology, entity ambiguity is eliminated and a unified knowledge representation is established. The department-disease relationship refers to establishing an association rule between the visiting department and common diseases in the knowledge graph, which is realized by using the association rule mining algorithm to analyze the frequent item sets of departments and diseases in historical medical records, enabling the system to infer potential disease types based on the patient's visiting department. The process-skill relationship refers to establishing a mapping relationship between standardized medical process nodes and the required escort skills, which is realized by using natural language processing technology to parse the co-occurrence relationship between process descriptions and skill keywords in medical literature, and is used to automatically load necessary skill items when generating the service execution list.

[0207] In step S7, when the patient selects the Department of Neurology as the visiting department, the system automatically identifies the possible psychological counseling needs of the patient based on the high-frequency association between the Department of Neurology and depression in the knowledge graph, and screens out the personnel with psychological intervention skills when matching escort personnel. At the same time, the system extracts the standard visiting process of the Department of Neurology from medical guidelines, such as the precautions before electroencephalogram examination and the operation specifications of drug allergy tests, associates these process nodes with the operation guides in the escort skill library, and generates a service list including steps such as emotion soothing, examination accompaniment, and medication reminder. When the escort personnel execute the service, the system pushes the corresponding skill guides according to the process nodes in the list to ensure that the service steps comply with medical norms.

[0208] Traditional methods only perform keyword matching based on the explicit needs of patients, unable to identify the implicit needs of depression patients for psychological counseling, and the escort service process depends on manual experience and is prone to missing key links. This solution realizes the collaborative optimization of demand prediction and process specification through the three-layer association mechanism of the knowledge graph, mapping department selection to potential diseases, diseases to necessary skills, and processes to standardized operations.

[0209] This embodiment can accurately identify the special skill requirements not clearly expressed by patients, such as automatically completing the psychological counseling skill requirements for neurology patients; generate a standardized service list covering all process nodes to reduce service omissions caused by unfamiliarity with the process; and avoid medical risks caused by non-standardized operations through the binding relationship between processes and skills.

[0210] In step S7, a medical knowledge graph is established and stored in the storage unit of the user module of the system.

[0211] In a specific embodiment, it further includes step S8 and step S9.

[0212] In step S8, during the accompaniment service process, the accompaniment personnel record the service situation in real time, and the patient can give feedback on problems and needs.

[0213] The service record and feedback module is used to store all service records during the accompaniment service process, including service time, service content, and patient feedback, for subsequent query and analysis, and classify the patient's feedback to solve the patient's problems and needs in a timely manner.

[0214] In step S9, after the service ends, the system reminds the patient to conduct a service evaluation, and the evaluation result is used as the basis for subsequent matching of the accompaniment personnel.

[0215] The service evaluation and complaint module is used to collect the evaluation information of patients, evaluate and rank the services of the accompaniment personnel. For patient complaints, detailed service records are retrieved for investigation, and corresponding measures are taken for the accompaniment personnel according to the investigation results, such as warnings and suspension of services. The specific operation process for a patient to make an appointment for accompaniment service is as follows:

[0216] The patient enters on the medical platform: "On [date] [time] in the morning, visit the Thoracic Surgery Department of XX Tumor Hospital, need 4 hours of accompaniment, require a female accompaniment personnel with experience in caring for lung cancer patients and able to assist in interpreting inspection reports";

[0217] Extract keywords "Thoracic Surgery Department of Tumor Hospital", "Lung Cancer Nursing", "Female" through TF-IDF to generate a demand vector , where the medical sub-vector includes features such as Thoracic Surgery Department (set weight 0.8), Lung Cancer Nursing (set weight 0.7), etc.

[0218] Search the accompaniment personnel database, screen out female personnel with at least 2 years of lung cancer nursing experience, generate their ability vectors , and calculate the dynamic matching degree S.

[0219] The information of the top 3 accompaniment personnel in terms of matching degree is pushed to the patient. After the patient makes a selection, the nearest accompaniment personnel is assigned through the DDPG algorithm, and it is obtained that the accompaniment personnel [name] is 2.8 kilometers away from the hospital and is expected to arrive in 12 minutes;

[0220] During the service process, through the knowledge graph, it is found that a CT examination needs to be done first for patients visiting this department, and the accompanying staff is automatically reminded to guide the patient to make a priority appointment for the examination; after the service ends, the patient's evaluation triggers the update of the ability vector. If the feedback is "the report interpretation is clear", the weight value of the "professional skills" dimension of this accompanying staff will be increased.

[0221] Through the above steps of innovative design and rigorous implementation, the accompanying service reservation method provided by the present invention, through an intelligent matching algorithm and a full-process service management, and combined with the system's collaborative work of multiple modules, ensures the accurate collection of data, the accuracy of intelligent recommendation, the efficient docking of patients, accompanying staff and the hospital, and the update of real-time data. This architecture not only optimizes the process of the accompanying service, but also effectively improves the efficiency and quality of the accompanying service through the intelligent matching algorithm and the full-process service management, meets the personalized needs of patients, and has good application prospects and innovation.

[0222] In Embodiment 2:

[0223] The second embodiment of the present invention is different from the previous embodiment in that:

[0224] When the function is implemented in the form of a software functional unit and can be used as an independent product for sale or use, it can be stored in a storage medium readable by a computer at this time. Based on this, the technical solution of the present invention, in essence, either the part that contributes to the prior art or part of the content of this solution, can be presented in the form of a computer software product.

[0225] This software product is stored in a storage medium and contains a series of instructions that can enable computer devices such as personal computers, servers or network devices to execute all or part of the steps of the methods described in various embodiments of the present invention. The above storage media cover various media such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, and optical discs that can store program codes.

[0226] The logic and steps presented in the flowchart or described in other forms can be regarded as an ordered set of executable instructions that perform logical functions. They can be specifically implemented in any computer-readable medium for use by an instruction execution system, apparatus, or device (such as a computer-based system, a system including a processor, or other systems that can obtain and execute instructions from the instruction execution system, apparatus, or device), or in conjunction with these instruction execution systems, apparatuses, or devices. Within the scope of this specification, a "computer-readable medium" refers to any device that can contain, store, communicate, propagate, or transmit a program for use by an instruction execution system, apparatus, or device, or in conjunction with these instruction execution systems, apparatuses, or devices.

[0227] More specific examples of computer-readable media are given below, but this is not an exhaustive list: electrical connection parts with one or more wirings (belonging to electronic devices), portable computer disk cartridges (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber devices, and portable compact disc read-only memory (CDROM). In addition, a computer-readable medium can even be paper or other suitable media on which a program can be printed. This is because, by optically scanning the paper or other media, followed by editing, interpretation, or other appropriate processing when necessary, the program can be obtained in electronic form and then stored in a computer memory.

[0228] It should be understood that various parts of the present invention can be implemented by hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be achieved by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, any one or a combination of the following techniques well-known in the art can be used: discrete logic circuits with logic gate circuits that implement the logical functions of data signals, application-specific integrated circuits with appropriate combinational logic gate circuits, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0229] The above-described specific embodiments have further elaborated on the purpose, technical solutions, and beneficial effects of the present invention. It should be understood that the above is only the specific embodiments of the present invention and is not used to limit the protection scope of the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for booking escort services, characterized in that, It includes the following steps: Step S1: The patient enters the accompanying diagnosis service reservation function through the medical platform and completes the filling of patient information, which includes the filled personal information, selected medical appointment information, accompanying diagnosis needs, and information on special requirements for the accompanying diagnosis personnel; among them, the medical appointment information includes the department to be visited and the expected medical appointment date; the accompanying diagnosis needs include one or more of whether assistance in registration service, payment service, pharmacy pick-up service, and translation service is required; Step S2: Analyze and process the patient information, perform multi-modal feature modeling based on the medical appointment information and accompanying diagnosis needs, establish the accompanying diagnosis needs as a demand vector, and find the multi-dimensional features of the accompanying diagnosis needs; Step S3: Obtain the information of the accompanying diagnosis personnel on the medical platform, including professional skills, service evaluations, available time, and geographical location; perform multi-modal feature modeling based on the information of the accompanying diagnosis personnel, establish the information of the accompanying diagnosis personnel as an ability vector, and find the multi-dimensional portrait of the accompanying diagnosis personnel; Step S4: According to the demand vector and the ability vector, use a matching algorithm based on multi-dimensional weighting for scoring to match the accompanying diagnosis needs with the information of the accompanying diagnosis personnel, calculate the comprehensive matching score, sort the accompanying diagnosis personnel according to the comprehensive matching score, and generate a candidate list; Step S5: Send the candidate list to the patient. The patient can select the accompanying diagnosis personnel or have the accompanying diagnosis personnel with the highest matching degree automatically selected; when the accompanying diagnosis personnel are selected, send the candidate list to the patient. The patient can view the information of the accompanying diagnosis personnel and independently select a satisfactory accompanying diagnosis personnel; if the patient does not make an independent selection, perform scheduling or select the one with the highest matching degree as the accompanying diagnosis personnel; Step S6: Generate a reservation arrangement and notify the patient and the accompanying diagnosis personnel; for the generation and notification of the reservation arrangement, after determining the accompanying diagnosis personnel, generate a reservation arrangement and send it to the patient and the accompanying diagnosis personnel simultaneously via text message and APP; after receiving the notification, the accompanying diagnosis personnel confirm to accept the order or raise an objection. If an objection is raised, re-match and return to Step S5; Step S7: Perform information processing and analysis. After receiving the patient information, first analyze the department to be visited, obtain the common diseases and corresponding accompanying diagnosis needs of this department through the medical platform, establish a medical knowledge graph, complete the requirements of the medical knowledge graph through the medical appointment process of the department, and use GNN to predict the service risk probability.

2. The method for booking an accompaniment medical service according to claim 1, wherein In Step S2, the multi-modal feature modeling refers to modeling the accompanying diagnosis needs through multi-modal features, deconstructing the patient information into multi-dimensional features of medical-service-preference-time and space as the demand vector, generating a multi-dimensional demand vector through the TF-IDF text processing algorithm, and finding the multi-dimensional features of the accompanying diagnosis needs; Step S21: Process the input text of the patient information by using a word segmentation tool to perform segmentation; Filter out the words without actual semantics and retain the keywords; normalize the keywords and map the synonyms to a unified label; Step S22: Obtain medical features based on the medical visit information, set the feature dimensions of the medical features, and the service keywords of the medical features include common departments, disease labels, and examination items; obtain service features based on the accompaniment needs, set the feature dimensions of the service features, and the service keywords of the service features include basic service labels, special service labels, service duration, and time window; obtain preference features based on the information about the special requirements for the accompaniment personnel in the remarks, set the feature dimensions of the preference features, and the preference keywords of the preference features include personnel attributes, skill labels, and service styles; obtain spatio-temporal features based on the medical visit information, set the feature dimensions of the spatio-temporal features, and the spatio-temporal keywords of the spatio-temporal features include the geographical location of the medical department and the time window. Step S23: Calculate the keyword scores for all keywords using the TF-IDF text processing algorithm. By calculating the word frequency of the keywords, to measure the relative importance of the keywords in the current text of the patient information, the calculation formula is: Among them, is the target keyword; is the text of the accompanying consultation demand input by the patient; keyword appears in the text the number of times; By calculating the inverse document frequency of the keywords, to measure the scarcity of the keywords, the calculation formula is: Among them, is the complete set of escort service demand texts and all patient demands in the historical data of the medical platform; is the total number of texts; represents the number of texts containing the keyword ; The keyword scores can be calculated by multiplying the obtained word frequency and inverse document frequency, and the calculation formula is: Among them, , the higher the value, the stronger the representativeness of the keyword for the current text; Step S24: Perform multi-modal feature modeling to obtain a multi-dimensional feature vector as the demand vector; multiply the keyword scores of each keyword of the medical features, service features, preference features, and spatio-temporal features by the set weights to obtain feature values, and then splice the feature values of each keyword in turn to obtain the final demand vector.

3. The method for booking an escort service according to claim 1, wherein, In step S3, the multi-modal feature modeling according to the accompaniment personnel information refers to processing the accompaniment personnel information to construct a multi-dimensional portrait of profession-service-spatio-temporal-skill as the ability vector, and using the Word2Vec semantic mapping algorithm to generate a multi-dimensional ability vector.

4. A method for booking an escort service according to any one of claims 1 and 3, characterized in that, In step S3, the generation of the multi-dimensional ability vector using the Word2Vec semantic mapping algorithm is to pre-train a medical domain Word2Vec model through the Word2Vec semantic mapping algorithm, obtain vectors from the accompaniment personnel information, and generate word vectors. The establishment of the medical domain Word2Vec model specifically includes: The establishment of the medical domain Word2Vec model predicts the context words from the central word, and the calculation formula is: Among them, central word input vector, including its semantic information; Central word Predicted context word Probability distribution of, The larger the value indicates The more likely it is to be The context of; The weight matrix from the input layer to the hidden layer stores the semantic association strength from the central word to all words. The weight matrix from the hidden layer to the output layer maps the hidden layer features to the vocabulary space; Bias term; Hidden layer bias term to prevent the neuron output from being constantly 0; Output layer bias term to adjust the prediction probabilities of each word; Among them, it specifically includes: The product of the vector and the weight matrix, plus the bias, to obtain the original input of the hidden layer, and a linear transformation is performed, and the calculation formula is: Compress the value to [-1, 1] through the tanh function to highlight the key semantic features and perform non-linear activation, and the calculation formula is: Calculate the product of the hidden layer features and the output weights to obtain the original scores of each vocabulary, and map the output layer, and the calculation formula is: Finally, perform probability normalization, and convert the original scores to probabilities through Softmax so that the sum of all vocabulary probabilities is 1. After training is completed, each row of the weight matrix of the input layer corresponds to the word vector of a vocabulary .

5. A method for booking an escort service according to any one of claims 1 and 3, characterized in that, In step S3, the construction process of the multi-dimensional portrait specifically includes: Step S321: The professional portrait obtains the professional skill text according to professional skills, and respectively establishes three word vectors for the nursing qualification level, nursing years, medical knowledge fields of expertise, and unique skills of professional skills. The three word vectors of professional skills are respectively obtained through a pre-trained medical field Word2Vec model. At this time, the high-dimensional professional portrait formula is expressed as: Aggregate multiple word vectors and generate professional portrait features through mean aggregation: Step S322: Perform a service portrait; Obtain service evaluation text based on service evaluations, perform sentiment analysis on the service evaluations to obtain keywords, conduct semantic mapping, extract positive evaluations, on-time rate, and service completion rate, and obtain word vectors of the service evaluations through a pre-trained Word2Vec model in the medical field , Then combine the word vector with the sentiment weight. At this time, the service portrait features: Among them is the set emotional weight matrix; Step S323: Based on the available time and geographical location, the spatio-temporal image obtains the word vectors of the available time, geographical location, and moving speed through a pre-trained Word2Vec model in the medical field , and splices them to obtain spatio-temporal image features: Step S324: Sequentially splice the professional portrait features, service portrait features, and spatio-temporal portrait features to obtain an ability vector as a multi-dimensional portrait.

6. The method for booking an accompaniment medical service according to claim 1, wherein In step S4, a matching algorithm based on multi-dimensional weighting for scoring is used to match the escort demand and escort personnel information, calculate the comprehensive matching score, sort the escort personnel according to the comprehensive matching score, and generate a candidate list; For the matching algorithm, based on the escort demand and escort personnel information, the weight matrix is dynamically adjusted, and the calculation of the comprehensive matching score uses a matching algorithm based on multi-dimensional weighting for scoring; The matching algorithm is based on the patient demand vector and the ability vector , through the formula Calculate the patient demand vector and the ability vector to calculate the cosine similarity of each dimension, and combine with the set weight matrix to generate a comprehensive matching score .

7. A method for booking an escort service according to claim 1, wherein, In step S7, for the establishment of the medical knowledge graph, the relationship among departments - processes - skills is obtained through the medical platform to establish the graph; the association between departments and diseases is found in the electronic medical records, the association between departments and processes is found through the process specifications in the medical guidelines, and the association between skills and processes is found in the medical literature, so as to establish the medical knowledge graph; Through the department-disease relationship, give priority to selecting an escort with experience in this department during matching; For the disease-skill relationship, automatically complete the hidden skill requirements of the patient; through process-skill, generate an escort service execution list.

8. A method for booking an escort service according to any one of claims 1 and 7, characterized in that, In step S7, for the prediction of service risk probability using GNN, the calculation formula of the risk probability: 。 9. A companion diagnosis service reservation system, characterized in that, The system is applied to a method for booking an escort service according to any one of claims 1 to 8, including a user module, a data processing module, a matching module, and a service management module that are electrically connected in sequence; among them, The user module includes a patient terminal, an escort personnel terminal, and a storage unit; it is used for the registration and information management of patients and escort personnel, records patient information, including: filled personal information, selected medical appointment information, escort demand, and information on special requirements for escort personnel, and records escort personnel information, including professional skills, service evaluation, available time, and geographical location; the storage unit stores patient information, escort personnel information, and the established medical knowledge graph; The data processing module preprocesses the patient appointment information and escort personnel registration information, extracts key features; performs multi-modal feature modeling on patient information, establishes the escort demand as a demand vector, and finds the multi-dimensional features of the escort demand; obtains the escort personnel information on the medical platform for multi-modal feature modeling, establishes the escort personnel information as an ability vector, and finds the multi-dimensional portrait of the escort personnel; The matching module uses a matching algorithm based on multi-dimensional weighting for scoring to match the escort demand and escort personnel information, calculate the comprehensive matching score, sort the escort personnel according to the comprehensive matching score, and generate a candidate list; The service management module sends the candidate list to the patient. The patient can select an escort or the system automatically selects the escort with the highest matching degree. When selecting an escort, the candidate list is sent to the patient, and the patient can view the information of the escorts and independently select a satisfactory escort. If the patient does not make an independent selection, the system selects the one with the highest matching degree as the escort. Then, an appointment arrangement is generated and the patient and the escort are notified.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 8.

Images