An outpatient information query and reservation management method based on big data
By constructing a big data outpatient environment model and a dynamic feature cloud, the problem of existing systems being unable to uniformly recognize multi-dimensional dynamic factors has been solved, enabling efficient utilization of outpatient resources and personalized appointment management.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SHANDONG PROVINCIAL HOSPITAL AFFILIATED TO SHANDONG FIRST MEDICAL UNIVERSITY (SHANDONG PROVINCIAL HOSPITAL)
- Filing Date
- 2026-03-18
- Publication Date
- 2026-05-19
AI Technical Summary
The existing outpatient information query and appointment management system cannot form a continuous and unified cognitive model of multi-dimensional dynamic factors, resulting in resource misallocation and local congestion, making it difficult to achieve global efficiency optimization and personalized and precise guidance.
By constructing a big data-based outpatient environment model, performing timestamp alignment and semantic association, generating a dynamic feature cloud, identifying high-frequency feature migration trajectories, planning personalized appointment strategies, and combining logical rule networks for path matching and routing.
It enables real-time, quantitative, and visual representation of the overall outpatient load and individual patient progress, automatically identifies efficient flow patterns, generates dynamic appointment strategies, and improves resource utilization efficiency and patient waiting urgency management.
Smart Images
Figure CN121860098B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of big data-driven smart outpatient management technology, specifically a method for outpatient information query and appointment management based on big data. Background Technology
[0002] Currently, outpatient information inquiry and appointment management mainly rely on the basic functional modules of hospital information systems. Existing technologies typically process discrete data streams from registration, triage, examination, and payment processes directly, and allocate resources and schedule time according to pre-set, fixed process rules. These methods are essentially based on static rule bases summarized from historical experience, performing simple retrieval, statistics, and queuing calculations on real-time inflowing multi-source, heterogeneous data.
[0003] These conventional solutions have inherent flaws. Their operational logic is built upon discrete, fragmented data items and rigid rules, failing to form a continuous, unified, and quantifiable dynamic cognitive model of the overall outpatient operation. The system cannot effectively represent the complex coupling relationships between multi-dimensional dynamic factors such as "service efficiency" and "urgency of demand," nor can it perceive the evolution of these relationships over time. Therefore, the generation of management strategies lacks a deep understanding of the complex situation formed by the intertwining of real-time outpatient load and individual patient behavior patterns, easily leading to resource misallocation and localized congestion.
[0004] Lacking the ability to macroscopically and dynamically analyze real-world patient flow processes, existing technologies struggle to automatically identify the most efficient and seamless service flow patterns proven in practice from massive amounts of historical and real-time interactive data. The system cannot plan an optimal path for new appointment requests that incorporates historical successes under dynamic constraints; it can only perform static matching of departments and time points, making it difficult to achieve global efficiency optimization and personalized, precise guidance. Summary of the Invention
[0005] The purpose of this invention is to provide a method for outpatient information query and appointment management based on big data, so as to solve the problems mentioned in the background art.
[0006] To achieve the above objectives, the present invention provides a method for outpatient information query and appointment management based on big data, the method comprising:
[0007] Original outpatient activity records are collected from distributed storage nodes, and an environment model describing the outpatient operation logic is established using the original outpatient activity records.
[0008] The system receives the operational logic constraints output by the environment model, and performs timestamp alignment and semantic association on the real-time inflow registration data, treatment node status data, and resource consumption data based on the operational logic constraints to form a set of outpatient related data with a unified spatiotemporal framework.
[0009] The outpatient related data set is analyzed at multiple granularities to extract pattern fragments that represent individual medical behavior and overall outpatient load. These pattern fragments are then projected into a dynamic feature space consisting of service efficiency and demand urgency dimensions to generate an outpatient dynamic feature cloud.
[0010] In the dynamic feature space, based on the density and evolution direction of feature points in the outpatient dynamic feature cloud, high-frequency feature migration trajectories are identified, and the feature migration trajectories are aggregated into several core service paths.
[0011] The real-time received query and appointment requests are matched and routed according to the core service path to calculate the appointment guidance strategy that serves the current request.
[0012] Preferably, the operation of establishing an environment model describing the outpatient operation logic using the original outpatient activity records specifically involves:
[0013] Analyze the event types and participant roles in the original outpatient activity records;
[0014] Construct an outpatient event sequence chain based on the chronological order of events and the interaction rules of the participants' roles;
[0015] Based on the transition probabilities of events and the changes in role states in the outpatient event sequence chain, a logical rule network for outpatient operation is derived, which serves as the environment model.
[0016] Preferably, the operation of aligning and semantically associating the real-time incoming registration data, treatment node status data, and resource consumption data according to the aforementioned operational logic constraints specifically involves:
[0017] The standard time window and role state dependency of each event defined in the logical rule network of the environment model are extracted as the running logic constraints;
[0018] Using the aforementioned operational logic constraints, logical stage labels are assigned to the registration data, the treatment node status data, and the resource consumption data;
[0019] Align the three types of data based on the logical stage labels and physical timestamps;
[0020] Based on the semantic relationships of the aligned data in the logical rule network, the association relationships between data entities are established, thereby forming the outpatient related data set.
[0021] Preferably, the operation of performing multi-granularity parsing on the outpatient related data set specifically involves:
[0022] At the individual patient visit level, the complete data chain of a single patient in the outpatient related data set is traced, and the complete behavioral sequence from registration to departure is extracted as a fine-grained pattern segment.
[0023] At the departmental business granularity, all patient data from the same department at the same time period in the outpatient related data set are aggregated, and their resource consumption and process time distribution are statistically analyzed to form a medium-granularity pattern fragment.
[0024] The fine-grained pattern fragment and the medium-grained pattern fragment are used together as the pattern fragment.
[0025] Preferably, the operation of projecting the pattern fragments into the dynamic feature space to generate the outpatient dynamic feature cloud specifically involves:
[0026] The service efficiency dimension is defined, and its value is calculated based on the service completion volume per unit time and the resource idle rate in the mode segment.
[0027] The urgency dimension of the demand is defined, and its value is calculated based on the cumulative waiting time of patients in the pattern segment and the keyword weights of the symptom descriptions.
[0028] Calculate the coordinate values of each of the pattern fragments in the service efficiency dimension and the demand urgency dimension;
[0029] The distribution of the coordinates of all pattern fragments in the dynamic feature space constitutes the outpatient dynamic feature cloud.
[0030] The operation of calculating the coordinate values of each of the mode fragments in the service efficiency dimension and the demand urgency dimension specifically includes:
[0031] The number of patient services completed per unit time and the average idle rate of various resources are extracted from the pattern fragments as service efficiency sub-indicators.
[0032] The cumulative waiting time of patients and the keywords and their weights extracted from their symptom descriptions based on natural language processing technology are used as sub-indicators of demand urgency.
[0033] The data of each item in the service efficiency sub-indicator and the demand urgency sub-indicator were normalized to eliminate the influence of units.
[0034] The data of each item in the normalized service efficiency sub-indicator are weighted and summed according to the preset efficiency weights, and the result is used as the coordinate value of the mode segment on the service efficiency dimension.
[0035] The data of each item in the normalized demand urgency sub-index are weighted and summed according to the preset urgency weights, and the result is used as the coordinate value of the pattern segment on the demand urgency dimension.
[0036] Preferably, the operation of identifying feature migration trajectories based on the density and evolution direction of feature points within the outpatient dynamic feature cloud specifically involves:
[0037] Density clustering is performed on the outpatient dynamic feature cloud in the dynamic feature space to identify high-density regions of feature points;
[0038] Analyze the flow trend of feature points between the high-density regions; the flow trend constitutes the evolution direction.
[0039] The sequence of feature points that continuously traverses multiple high-density regions according to the evolution direction is identified as a feature migration trajectory.
[0040] Preferably, the operation of aggregating the feature migration trajectories into core service paths specifically involves:
[0041] A similarity comparison is performed on all identified feature migration trajectories;
[0042] Merge migration trajectories with similar starting points, ending points, and high-density regions along their routes;
[0043] A central path is calculated for each type of merged trajectory, and the central path serves as a core service path.
[0044] Preferably, the operation of matching real-time received queries and appointment requests according to the core service path specifically involves:
[0045] Parse the patient's initial status information contained in the query and appointment request;
[0046] The patient's initial state information is mapped onto the dynamic feature space to obtain an initial feature point;
[0047] Calculate the feature distance from the initial feature point to the starting point of each core service path;
[0048] The core service path with the smallest feature distance is selected as the path that matches the current request.
[0049] Preferably, the operation of routing for the matched request to calculate the appointment guidance strategy specifically includes:
[0050] On the matched core service path, the feasible positions on the path are determined based on the real-time node status fed back by the current outpatient associated data set;
[0051] Based on the logical rule network in the aforementioned environment model, the conditions and specific transaction sequences required to reach a feasible position from the initial feature point are deduced.
[0052] The conditions and transaction sequence are encapsulated to generate specific appointment time points, department nodes, and precautions, forming the final appointment guidance strategy.
[0053] Preferably, the operation of combining the logical rule network in the environment model to deduce the conditions and possible transaction sequences required to reach a feasible position from the initial feature point specifically includes:
[0054] Based on the patient's initial state mapped by the initial feature points, the corresponding initial role state node is located in the logical rule network.
[0055] Based on the target department or service node corresponding to the feasible location, locate the corresponding target role status node in the logical rule network;
[0056] Starting from the initial role state node and ending at the target role state node, search for all state transition paths that satisfy the event transition probability constraints in the logical rule network.
[0057] For each searched state transition path, based on the standard time window of each event on the path and the role state dependency, the specific sequence of tasks that the patient needs to complete, as well as the preconditions and estimated time for each task, are derived.
[0058] From all derived transaction sequences, the sequence with the shortest total estimated time and all preconditions are satisfied is selected as the optimal deduction result.
[0059] Compared with the prior art, the beneficial effects of the present invention are:
[0060] A model of the outpatient environment is built by extracting raw records from distributed storage nodes. Based on the logical constraints output by the model, the real-time inflow of registration, treatment node status, and resource consumption data is timestamped and semantically correlated to construct an outpatient related data set with a unified spatiotemporal framework. This ensures that subsequent analysis is based on an accurate and coherent sequence of events in the patient visit process. By performing multi-granularity analysis on this data set, pattern fragments representing individual behavior and overall load are extracted and projected into a dynamic feature space composed of service efficiency and demand urgency dimensions, generating a dynamic feature cloud for the outpatient department. This method of transforming complex outpatient states into a distribution of feature points in a continuous mathematical space enables real-time, quantitative, and visual representation of the overall outpatient load and individual patient visit progress. Outpatient managers can intuitively perceive the shape, density, and movement trend of the "feature cloud," thereby identifying areas with low service efficiency or urgent patient waiting times. Traditional lagging monitoring methods relying on discrete indicators and reports are replaced by dynamic and continuous spatial state monitoring.
[0061] In the constructed dynamic feature space, based on the distribution density and evolution direction of feature points within the feature cloud, frequently occurring feature migration trajectories are identified, and these trajectories are aggregated into several core service paths. This process essentially involves automatically mining from massive amounts of real-world patient data efficient or frequently used workflow patterns repeatedly validated under different efficiency-urgency states. When receiving real-time query and appointment requests, the system matches and routes the requests to these core service paths based on the mapping of the current outpatient status in the feature space, calculating and generating appointment guidance strategies. This transforms the system from simply assigning patients an isolated registration time slot or clinic node, to planning a complete patient journey that moves within the feature space and integrates into historically efficient patterns. Scheduling decisions shift from static assignment based on fixed rules and average durations to intelligent planning based on dynamic feature space and data-driven path discovery. Attached Figure Description
[0062] Figure 1 This is a schematic diagram illustrating the working principle of the outpatient information query and appointment management method based on big data as described in this invention.
[0063] Figure 2 A flowchart for establishing an environmental model;
[0064] Figure 3 The flowchart for multi-granularity analysis;
[0065] Figure 4 A correlation diagram of density and evolution direction intensity of high-density regions in the dynamic feature space of outpatient clinics;
[0066] Figure 5 This is a scatter plot of the dynamic features of outpatient services. Detailed Implementation
[0067] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0068] Please see Figure 1 This invention provides a big data-based outpatient information query and appointment management method. The method includes: aggregating historical raw outpatient activity records from multiple distributed storage nodes, covering the entire process of registration, triage, examination, treatment, and payment; utilizing these raw outpatient activity records, establishing an environment model describing the outpatient operation logic by analyzing event sequences and role interactions. The environment model continuously outputs operational logic constraints. The system receives these constraints and processes real-time incoming registration data, treatment node status data, and resource consumption data. The processing includes timestamp alignment and semantic association based on logical constraints, thereby forming an outpatient-related data set with a unified spatiotemporal framework. Subsequently, the outpatient-related data set undergoes multi-granularity analysis, extracting pattern fragments from different levels that can characterize both individual patient behavior and reflect the overall outpatient load. These pattern fragments are projected into a dynamic feature space composed of service efficiency and demand urgency dimensions, forming a scattered distribution within this space, i.e., an outpatient dynamic feature cloud. In the dynamic feature space, based on the distribution density and flow evolution direction of feature points within the outpatient dynamic feature cloud, the system identifies frequently occurring feature point movement sequences, i.e., feature migration trajectories, and aggregates multiple similar trajectories into several representative core service paths. When the system receives query and appointment requests from terminals in real time, it matches and searches these requests based on the core service paths, ultimately calculating and generating a personalized appointment guidance strategy to serve the current request.
[0069] Example 1: See Figure 2This study analyzes the event types and participant roles in the original outpatient activity records. Event types include registration, waiting, treatment, and payment, while participant roles include patients, doctors, nurses, and departments. Based on the sequence of events and the interaction rules between participant roles, outpatient event sequence chains are constructed, each chain describing a complete patient visit. Based on the transition probabilities between events and the changes in role states within these event sequence chains, a logical rule network for outpatient operations is derived. This logical rule network serves as an environment model describing the outpatient operation logic, defining the conditions, sequence, and state dependencies of events. Standard time windows and role state dependencies for each event are extracted from the logical rule network in the environment model; this information serves as operational logic constraints. Using these constraints, corresponding logical stage labels, such as "awaiting treatment," "under examination," and "completed," are assigned to the real-time incoming registration data, treatment node status data, and resource consumption data. Based on the logical stage labels and the inherent physical timestamps of the data, the three types of data are time-aligned, ensuring that data belonging to the same logical stage or consecutive stages are aligned on the timeline. Based on the semantic relationships of the aligned data in the logical rule network, the association between data entities is established. For example, a patient's registration record, the real-time status of the department, and the resource consumption data of the department are associated to form an outpatient association data set.
[0070] In practical implementation, original outpatient activity records are collected from multiple distributed storage nodes, including distributed databases, hospital information systems, and IoT terminals. These records contain fields such as event timestamps, event types, patient identifiers, department identifiers, doctor identifiers, and operation content. The implementation then parses the event types and participant roles from these records. Event types include registration, triage, doctor consultation, examination / testing, prescription issuance, payment, and medication dispensing. Participant roles include patient, registration clerk, triage nurse, doctor, examination department, and pharmacy roles. Based on the order of events and the interaction rules between participant roles, an outpatient event sequence chain is constructed. In some embodiments, a logical rule network for outpatient operations is derived based on the transition probabilities and role state changes between events in a large number of outpatient event sequence chains. In this logical rule network, nodes represent role states, and edges represent events and transition probabilities. For example, a patient transitions from the state "Patient has registered, awaiting triage" to the state "Patient has been triaged, awaiting consultation" via the "Triage completed" event, with the transition probability being a conditional probability based on historical data statistics.
[0071] In some embodiments, the standard time windows and role state dependencies of each event defined in the logical rule network of the environment model are extracted as runtime logic constraints. Taking the doctor's consultation event as an example, the runtime logic constraint defines its standard time window as the morning and afternoon periods. The role state dependency requires that before executing the doctor's consultation event, the patient role must be in the "triaged" state and the doctor role must be in the "on duty" state. Using runtime logic constraints, logical stage labels are assigned to real-time incoming registration data, treatment node status data, and resource consumption data. For example, a real-time incoming registration data is labeled with the "registration completed" logical stage label, a treatment node status data describing the doctor's current state is labeled with the "doctor on duty, idle" logical stage label, and a resource consumption data describing the number of people currently queuing in the CT room is labeled with the "examination resource, occupied" logical stage label. In specific implementations, the three types of data are aligned based on the logical stage labels and physical timestamps. Optionally, the association relationship between data entities is established based on the semantic relationship of the aligned data in the logical rule network. For example, by linking patient records in registration data, the target doctor status assigned to that patient in treatment node status data, and the resource usage of the target doctor's clinic in resource consumption data through patient identifiers, a single record in the outpatient associated data set can be formed. When calculating the event transition probability, a feasible calculation method is expressed by the following formula:
[0072]
[0073] Where: characters In a logical rule network, this means starting from the state... Through events Transition to state The probability of the character. This indicates that in the historical outpatient event sequence chain, observations were made from the state... Through the incident Transition to state The count of characters. This indicates that in the historical outpatient event sequence chain, observations were made from the state... Through the incident The total count of transitions to any possible successor state.
[0074] Example 2: See Figure 3At the individual patient visit granularity, the complete data chain of a single patient in the outpatient related data set is traced, from registration to leaving the hospital, extracting the patient's complete visit behavior sequence as a fine-grained pattern fragment. At the departmental business granularity, all patient data from the same department in the same time period in the outpatient related data set are aggregated, and the resource usage and process time distribution of the department during that time period are statistically analyzed to form a medium-grained pattern fragment. The fine-grained and medium-grained pattern fragments together constitute the pattern fragments used for subsequent analysis. Two dimensions are defined in the dynamic feature space: service efficiency and demand urgency. The value of the service efficiency dimension is calculated based on the service completion volume per unit time and resource idle rate in the pattern fragment, while the value of the demand urgency dimension is calculated based on the cumulative patient waiting time and keyword weights in the symptom description in the pattern fragment. For each pattern segment, coordinate values are calculated in the service efficiency and demand urgency dimensions. Specifically, the calculation process involves extracting the number of patient services completed per unit time and the average idle rate of various resources as service efficiency sub-indicators from the pattern segment; and extracting the cumulative waiting time of patients and keywords and their weights extracted from their symptom descriptions using natural language processing technology as demand urgency sub-indicators. The data for each component of both the service efficiency and demand urgency sub-indicators are normalized to eliminate the influence of dimensions. The normalized service efficiency sub-indicators are then weighted and summed according to preset efficiency weights, and the result is used as the coordinate value of the pattern segment in the service efficiency dimension. Similarly, the normalized demand urgency sub-indicators are weighted and summed according to preset urgency weights, and the result is used as the coordinate value of the pattern segment in the demand urgency dimension. The distribution of coordinate points of all pattern segments in the dynamic feature space constitutes the outpatient dynamic feature cloud.
[0075] In specific implementation, the input is an outpatient association data set formed according to the method in Example 1. The outpatient association data set includes aligned timestamps, logical stage labels, and semantically related patient data, department status data, and resource data. In specific implementation, when operating at the individual patient visit level, the system tracks the complete data chain of a single patient in the outpatient association data set. Based on the unique patient identifier, it retrieves all records of the patient from the registration event to the departure event from the outpatient association data set. These records are sorted according to the physical timestamps, and a complete ordered sequence of behaviors from registration to departure is extracted as a fine-grained pattern fragment. The content of the fine-grained pattern fragment includes the patient identifier, event sequence, time points of each event, and the departments and resources involved. In some embodiments, when operating at the departmental business granularity, the system aggregates all patient data from the same department within the same time period in the outpatient related data set. For example, selecting all records with a time window of "2023-10-27 09:00:00 to 2023-10-27 10:00:00" and a department identification of "Cardiology", the system statistically analyzes the number of patient services completed in Cardiology during this time period, the on-duty and off-duty hours of all physician resources, the utilization rate of examination equipment, and the distribution of average waiting time and average treatment time for patients, forming a medium-granularity pattern segment. The medium-granularity pattern segment records the department identification, time window, resource usage statistics, and a histogram of process time distribution. The fine-granularity pattern segment and the medium-granularity pattern segment together serve as pattern segments for subsequent projective analysis.
[0076] The dynamic feature space for projecting pattern segments is defined by a service efficiency dimension and a demand urgency dimension. The service efficiency dimension is calculated based on the number of services completed per unit time and the resource idle rate within the pattern segment. The demand urgency dimension is calculated based on the cumulative patient waiting time and the keyword weights in the symptom description within the pattern segment. The coordinate values of each pattern segment in the service efficiency and demand urgency dimensions are calculated as follows: The number of patient services completed per unit time and the average idle rate of various resources are extracted from the pattern segment as service efficiency sub-indicators. For example, for a 30-minute pattern segment, the number of patient services completed is 5, the doctor resource idle rate is 0.2, and the examination equipment idle rate is 0.5. The cumulative patient waiting time and the keywords and their weights extracted from the symptom description using natural language processing (NLP) are extracted from the pattern segment as demand urgency sub-indicators. For example, the cumulative waiting time is 45 minutes, and the symptom description, analyzed using NLP, yields a keyword set {"chest pain", "persistent", "dyspnea"} and corresponding weights {0.7, 0.5, 0.6}. The data for each component of the service efficiency sub-indicator and the demand urgency sub-indicator are normalized to eliminate the influence of dimensions. The normalization method uses minimum-maximum scaling. In practice, the normalized data for each component of the service efficiency sub-indicator are weighted and summed according to preset efficiency weights, and the result is used as the coordinate value of the pattern segment in the service efficiency dimension. Similarly, the normalized data for each component of the demand urgency sub-indicator are weighted and summed according to preset urgency weights, and the result is used as the coordinate value of the pattern segment in the demand urgency dimension. The distribution of the coordinate points of all pattern segments in the dynamic feature space constitutes the outpatient dynamic feature cloud, which is a two-dimensional point set. Optionally, the coordinate values of the service efficiency dimension... The calculation can be expressed by the following formula:
[0077]
[0078] Where: characters This represents the final coordinate value of the pattern fragment in the service efficiency dimension. (Character) This indicates the preset efficiency weights for each service completion item. (Character) This represents the normalized value of the number of patient services completed. (Character) This indicates the preset efficiency weight for the resource idle rate item. (Character) This represents the normalized average resource idle rate index value.
[0079] Example 3: Density clustering is performed on the outpatient dynamic feature cloud in the dynamic feature space to identify high-density areas with dense feature point distribution. The flow trend of feature points between high-density areas is analyzed, which constitutes the evolution direction of feature points. Sequences of feature points that continuously traverse multiple high-density areas according to the evolution direction are identified as feature migration trajectories, reflecting a common transfer path of the patient visit pattern in the feature space. All identified feature migration trajectories are compared for similarity, based on the starting and ending points of the trajectory and the high-density areas traversed. Feature migration trajectories with similar starting points, ending points, and high-density areas are merged to form several trajectory clusters. A central path is calculated for each merged trajectory cluster. This central path is obtained by calculating the geometric center or average path of all trajectory points within the cluster. This central path serves as a core service path, representing a typical outpatient service process pattern.
[0080] In specific implementations, the input is the outpatient dynamic feature cloud generated according to the method in Example 2. The outpatient dynamic feature cloud is a collection of feature points in a two-dimensional plane, where each feature point corresponds to the coordinates of a pattern segment in the service efficiency dimension and the demand urgency dimension. In specific implementations, density clustering is performed on the outpatient dynamic feature cloud in the dynamic feature space using a density-based spatial clustering algorithm. This algorithm can discover clusters of arbitrary shapes and identify noise points. After the algorithm is executed, high-density areas with dense feature point distribution are identified, and each high-density area is assigned a unique region identifier, representing a service state pattern that frequently occurs in the feature space. In some embodiments, the flow trend of feature points between high-density areas is analyzed. The flow trend constitutes the evolution direction. This is specifically achieved by analyzing the time-series attributes of feature points in the outpatient dynamic feature cloud. Multiple feature points belonging to the same patient's visit process or continuous time periods in the same department are connected in chronological order, and their movement vectors are calculated. The number and intensity of movement vectors from one high-density area to another are statistically analyzed to determine the main evolution direction. It is understandable that a sequence of feature points that are sequentially traversed through multiple high-density regions in accordance with the direction of evolution, are continuous in time and are spatially adjacent, can be identified as a feature migration trajectory.
[0081] All identified feature migration trajectories are compared for similarity, based on the trajectory's start and end positions, and the sequence of high-density regions it traverses. The start position of a feature migration trajectory is defined as the identifier of the high-density region to which the first feature point in the trajectory belongs, the end position is defined as the identifier of the high-density region to which the last feature point in the trajectory belongs, and the sequence of high-density regions traversed is defined as a list of all high-density region identifiers that the trajectory passes through in sequence. In practice, a similarity score is calculated between two feature migration trajectories. One feasible calculation method is represented by the following formula:
[0082]
[0083] Where: characters Represents feature migration trajectory With feature transfer trajectory Similarity score between characters. This represents the starting point similarity weight coefficient. (Character) This is an indicator function; the function value is 1 when the condition inside the parentheses is true, and 0 otherwise. (Character) Represents feature migration trajectory The starting region identifier. Character Represents feature migration trajectory The starting region identifier. Character This represents the endpoint similarity weight coefficient. (Character) Represents feature migration trajectory The endpoint region identifier. Character Represents feature migration trajectory The endpoint region identifier. Character This represents the weighting coefficients for path sequence similarity. (Character) Represents feature migration trajectory The sequence of regions through which the route passes With feature transfer trajectory The sequence of regions through which the route passes The length of the longest common subsequence. Optionally, a similarity threshold can be set to classify feature migration trajectories with similarity scores higher than the threshold as similar trajectories.
[0084] Feature migration trajectories with similar start and end points and passing through high-density areas are merged. This merging operation is performed by assigning similar feature migration trajectories to the same trajectory cluster. A central path is calculated for each merged trajectory cluster. This central path is calculated by extracting a sequence of common high-density areas traversed by all feature migration trajectories within the cluster, and then taking the geometric center of the feature point coordinates of all trajectories within that region for each region in the sequence. This central path serves as a core service path. It can be understood that a core service path is an abstract path in the dynamic feature space representing a typical outpatient workflow pattern. For example, a core service path might be expressed as "Path P: From region A (coordinate center) -> region B (coordinate center) -> region C (coordinate center)".
[0085] See Figure 4 This is a high-density region density and evolution direction intensity correlation map of the outpatient dynamic feature space. It is a core analysis tool in the feature migration trajectory recognition stage and is used to show the correspondence between the feature point distribution density and the flow trend intensity between different regions.
[0086] Example 4: Parsing the initial patient status information contained in real-time received query and appointment requests. This information may include basic patient information, chief symptoms, expected appointment time, etc. The initial patient status information is mapped onto a dynamic feature space, and its coordinates in the dimensions of service efficiency and urgency are calculated to obtain an initial feature point. The feature distance from this initial feature point to the starting point of each core service path is calculated, using metrics such as Euclidean distance. The core service path with the smallest feature distance is selected as the path matching the current request. On the matched core service path, based on the real-time node status feedback from the current outpatient related data set, the specific location available for the patient to proceed on the path is determined. Combining the logical rule network in the environment model, the conditions and possible transaction sequences required to reach the proceedable location from the state represented by the initial feature point are deduced. The deduced conditions and transaction sequences are encapsulated to generate specific appointment times, suggested department nodes, and related precautions, forming the final appointment guidance strategy serving the request.
[0087] In specific implementation, the input includes a set of core service paths aggregated according to the method in Example 3, and query and appointment requests received in real time from the user terminal. Query and appointment requests are usually transmitted in the form of structured data messages. The initial patient status information contained in the query and appointment requests is parsed. The initial patient status information may include fields such as patient age, gender, chief complaint text, expected appointment time, and past medical history identifiers. The initial patient status information is mapped to a dynamic feature space to obtain an initial feature point. The mapping process follows the rules defined in Example 2, namely, extracting keyword weights based on the chief complaint text using natural language processing technology, estimating the initial waiting time based on the expected appointment time, and calculating the initial values of the two dimensions of service efficiency and urgency of demand as the coordinates of the initial feature point. In some embodiments, the feature distance from the initial feature point to the starting point of each core service path is calculated. The starting point of a core service path is represented by the coordinate center of the first region in its defined central path sequence, and the feature distance is measured using Euclidean distance. The core service path with the smallest feature distance is selected as the path matching the current request. The matching logic is to find a typical service flow pattern in the feature space that is closest to the patient's initial status. As can be understood, Table 1 shows a simplified example of matching calculation, which includes three preset core service paths and their distance calculations to a specific initial feature point.
[0088] Table 1: Calculation Table of Feature Distance Between Initial Feature Point and Core Service Path Starting Point
[0089] Core Service Path ID Path starting point coordinates (service efficiency, urgency of demand) Initial feature point coordinates (service efficiency, urgency of demand) Feature distance Path_001 (0.15,0.80) (0.18,0.76) 0.05 Path_002 (0.40,0.50) (0.18,0.76) 0.47 Path_003 (0.10,0.90) (0.18,0.76) 0.14
[0090] In practice, according to the calculation results in Table 1, the path with the smallest feature distance is Path_001, so Path_001 is selected as the path that matches the current request.
[0091] On the core service path of the matching, based on the real-time node status feedback from the current outpatient related data set, the traversable positions on the path are determined. The real-time node status includes the current number of patients waiting in each department, the real-time on-duty status of doctors, and the estimated idle time of examination equipment. Combining the logical rule network in the environment model, the conditions and possible transaction sequences required to reach the traversable position from the initial feature point are deduced. The deduction process includes querying the logical rule network to confirm that from the patient's initial state "request submitted, awaiting triage" to the target state "examination scheduled," the "triage event" and the "examination appointment event" need to be triggered sequentially, and the preconditions such as the "triage event" requiring complete basic patient information and the "examination appointment event" requiring the patient to have been triaged and the target examination equipment to be about to be idle are met. The deduced conditions and transaction sequences are encapsulated to generate specific appointment time points, department nodes, and precautions, forming the final appointment guidance strategy. Optionally, the appointment guidance strategy could be described as: "Recommended appointment: Echocardiogram, Appointment time: 10:30 AM today, Department: Ultrasound Room 3, Note: Please complete triage at the triage desk on the first floor first, and arrive at the Ultrasound Department for registration before 10:20 AM." (Characteristic distance) The calculation is expressed by the following formula:
[0092]
[0093] Where: characters This represents the feature distance from the initial feature point to the starting point of a core service path. (Character) This represents the coordinates of the initial feature points along the service efficiency dimension. (Character) This represents the coordinates of the starting point of this core service path on the service efficiency dimension. (Character) This represents the coordinates of the initial feature point along the dimension of urgency. (Character) This indicates the coordinate value of the starting point of this core service path in terms of the urgency of demand.
[0094] Example 5: Based on the patient's initial state mapped by the initial feature points, locate the corresponding initial role state node in the logical rule network. Based on the target department or service node corresponding to the traversable position, locate the corresponding target role state node in the logical rule network. Starting from the initial role state node and ending at the target role state node, search for all state transition paths that satisfy the event transition probability constraints in the logical rule network. For each searched state transition path, based on the standard time window of each event on the path and the role state dependency, deduce the specific sequence of transactions the patient needs to complete, as well as the preconditions and estimated time for each transaction. From all derived transaction sequences, select the sequence with the shortest total estimated time and all preconditions satisfied as the optimal deduction result. This optimal deduction result is used to generate the specific step arrangement in the appointment guidance strategy.
[0095] In practical implementation, the inputs to the simulation operation are the matched core service path, the patient's initial state mapped by the initial feature points, the target state determined by the drivable location, and the logical rule network in the environment model. Based on the patient's initial state mapped by the initial feature points, the corresponding initial role state node is located in the logical rule network. For example, if the patient's initial state information includes "Complaint: Chest pain, appointment request submitted via mobile application but no on-site triage," it is mapped to the "Patient Role: Request Submitted, No Triage" state node in the logical rule network. Based on the target department or service node corresponding to the drivable location, the corresponding target role state node is located in the logical rule network. For example, if the drivable location corresponds to "CT scan room, expected to be available in 10 minutes," it is mapped to the "Examination Resource: CT Room, Status: Soon to be Available, Associated Patient Status: Examination Appointed" target role state node in the logical rule network.
[0096] Starting from the initial role state node and ending at the target role state node, all state transition paths satisfying event transition probability constraints are searched within the logical rule network. In some embodiments, the logical rule network is modeled as a directed graph, where nodes represent role states and edges represent events and their corresponding transition probabilities. The search process traverses all possible paths from the starting node to the ending node, filtering out paths where the transition probability of any event is below a preset threshold, such as paths with a transition probability below 0.1, ultimately yielding N candidate state transition paths. For each searched state transition path, based on the standard time window of each event on the path and its dependency on the role state, the specific sequence of tasks the patient needs to complete, as well as the preconditions and estimated time for each task, are derived.
[0097] From all derived transaction sequences, the sequence with the shortest total estimated time and all preconditions are satisfied is selected as the optimal derivation result. It can be understood that the calculation of the total estimated time includes the standard time of each transaction and the waiting time for transfers between transactions. A feasible total estimated time... The calculation formula is as follows:
[0098]
[0099] Where: characters This represents the total estimated time for the transaction sequence corresponding to the k-th candidate state transition path. (Character) This represents the total number of transactions contained in the transaction sequence derived from the k-th path. (Character) This represents the estimated time taken for the i-th transaction in the transaction sequence of the k-th path. (Character) This represents the estimated transfer or waiting time required to satisfy the preconditions of transaction (j+1) after completing transaction j. During filtering, the system verifies whether the preconditions of all transactions in each path's transaction sequence can be satisfied under the current real-time outpatient status, such as checking if the "doctor on duty" condition is met, and selects the appropriate path from all paths where the conditions are satisfied. The path with the smallest value is taken as the optimal projection result. Optionally, the optimal projection result will include a specific list of transactions, the preconditions for each transaction, and the expected time point. This result will be directly used to generate the step descriptions and schedules in the final appointment guidance strategy.
[0100] See Figure 5 This is a scatter plot of the outpatient dynamic feature cloud, a core visualization tool in the dynamic feature cloud generation stage. It displays the distribution of outpatient pattern segments in the two-dimensional feature space of "service efficiency" and "demand urgency." Feature points cover most of the feature space, but are more densely distributed in the "medium service efficiency + high demand urgency" interval, indicating that this pattern is common in outpatient settings. A few extreme points exist, representing special low-urgency scenarios. There is no obvious strong linear correlation, indicating that the relationship between service efficiency and demand urgency is complex and requires further identification of typical areas using density clustering. This type of chart provides a global overview of outpatient service status, helping to quickly locate dense areas in the feature space and providing basic visualization support for subsequent density clustering and feature migration trajectory recognition.
[0101] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.
Claims
1. A method for outpatient information query and appointment management based on big data, characterized in that, The method includes the following operations: The system aggregates original outpatient activity records from distributed storage nodes and uses these records to build an environment model describing the outpatient operation logic. Specifically, this includes: parsing the event types and participant roles in the original outpatient activity records; constructing an outpatient event sequence chain based on the order of events and the interaction rules of participant roles; and deriving a logical rule network for outpatient operation based on the transition probabilities of events and the changes in role states in the outpatient event sequence chain. This logical rule network serves as the environment model. The system receives the operational logic constraints output by the environment model. Based on these constraints, it performs timestamp alignment and semantic association on the real-time incoming registration data, treatment node status data, and resource consumption data to form an outpatient related data set with a unified spatiotemporal framework. Specifically, this includes: extracting the standard time windows and role state dependencies of each event defined by the logical rule network in the environment model as the operational logic constraints; using these constraints to assign logical stage labels to the registration data, treatment node status data, and resource consumption data; aligning the three types of data based on the logical stage labels and physical timestamps; and establishing associations between data entities based on the semantic connections of the aligned data in the logical rule network, thereby forming the outpatient related data set. The outpatient-related data set is analyzed at multiple granularities to extract pattern fragments representing individual patient behavior and overall outpatient load. These pattern fragments are then projected into a dynamic feature space composed of service efficiency and demand urgency dimensions to generate an outpatient dynamic feature cloud. Specifically, this includes: at the individual patient granularity, tracing the complete data chain of a single patient in the outpatient-related data set and extracting the complete behavioral sequence from registration to departure as a fine-grained pattern fragment; at the departmental business granularity, aggregating all patient data from the same department within the same time period in the outpatient-related data set, statistically analyzing their resource usage and process time distribution to form a medium-grained pattern fragment; combining the fine-grained and medium-grained pattern fragments as the overall pattern fragment; defining the service efficiency dimension, whose value is calculated based on the service completion volume and resource idle rate per unit time in the pattern fragment; defining the demand urgency dimension, whose value is calculated based on the cumulative patient waiting time and keyword weights in the symptom description in the pattern fragment; and calculating the relationship between service efficiency and demand urgency for each pattern fragment. The coordinate values of all pattern segments in the dynamic feature space constitute the outpatient dynamic feature cloud. The operation of calculating the coordinate values of each pattern segment in the service efficiency and demand urgency dimensions specifically includes: extracting the number of patient services completed per unit time and the average idle rate of various resources from the pattern segment as service efficiency sub-indicators; extracting the cumulative waiting time of patients and keywords and their weights extracted from their symptom descriptions based on natural language processing technology from the pattern segment as demand urgency sub-indicators; normalizing the data of each item in the service efficiency and demand urgency sub-indicators to eliminate the influence of dimensions; weighting and summing the data of each item in the normalized service efficiency sub-indicator according to preset efficiency weights, and using the result as the coordinate value of the pattern segment in the service efficiency dimension; and weighting and summing the data of each item in the normalized demand urgency sub-indicator according to preset urgency weights, and using the result as the coordinate value of the pattern segment in the demand urgency dimension. In the dynamic feature space, based on the density and evolution direction of feature points within the outpatient dynamic feature cloud, frequently occurring feature migration trajectories are identified, and these trajectories are aggregated into several core service paths. Specifically, this includes: performing density clustering on the outpatient dynamic feature cloud in the dynamic feature space to identify high-density regions of feature points; analyzing the flow trend of feature points between these high-density regions, where the flow trend constitutes the evolution direction; identifying a sequence of feature points that continuously traverses multiple high-density regions according to the evolution direction as a feature migration trajectory; comparing the similarity of all identified feature migration trajectories; merging feature migration trajectories with similar starting points, ending points, and traversing high-density regions; and calculating a central path for each merged trajectory, where the central path serves as a core service path. The real-time received query and appointment requests are matched and routed according to the core service path to calculate the appointment guidance strategy that serves the current request.
2. The outpatient information query and appointment management method based on big data as described in claim 1, characterized in that, The specific operation of matching real-time received query and appointment requests according to the core service path is as follows: Parse the patient's initial status information contained in the query and appointment request; The patient's initial state information is mapped onto the dynamic feature space to obtain an initial feature point; Calculate the feature distance from the initial feature point to the starting point of each core service path; The core service path with the smallest feature distance is selected as the path that matches the current request.
3. The outpatient information query and appointment management method based on big data as described in claim 2, characterized in that, The specific operation of routing the matched requests to calculate the appointment guidance strategy is as follows: On the matched core service path, the feasible positions on the path are determined based on the real-time node status fed back by the current outpatient associated data set; Based on the logical rule network in the aforementioned environment model, the conditions and specific transaction sequences required to reach a feasible position from the initial feature point are deduced. The conditions and transaction sequence are encapsulated to generate specific appointment time points, department nodes, and precautions, forming the final appointment guidance strategy.
4. The method according to claim 3, characterized in that, The operation of combining the logical rule network in the environment model to deduce the conditions and specific transaction sequence required to reach a feasible position from the initial feature point specifically includes: Based on the patient's initial state mapped by the initial feature points, the corresponding initial role state node is located in the logical rule network. Based on the target department or service node corresponding to the feasible location, locate the corresponding target role status node in the logical rule network; Starting from the initial role state node and ending at the target role state node, search for all state transition paths that satisfy the event transition probability constraints in the logical rule network. For each searched state transition path, based on the standard time window of each event on the path and the role state dependency, the specific sequence of tasks that the patient needs to complete, as well as the preconditions and estimated time for each task, are derived. From all derived transaction sequences, the sequence with the shortest total estimated time and all preconditions are satisfied is selected as the optimal deduction result.