An internet hospital resource scheduling method based on a service bus

Abstract

This application provides a service bus-based resource scheduling method for internet hospitals, relating to the field of medical information processing technology. The method includes: collecting doctor session data and analyzing disease complexity and waiting intervals to identify high-risk abandonment sessions with non-linearly increasing waiting times; dynamically adjusting the upper limit of concurrent doctor sessions, pausing allocation for simple diseases and prioritizing complex disease sessions, and locking resources until a substantial response is completed; resuming allocation and continuously updating the upper limit when the waiting interval decreases. The beneficial effects of this application are: through non-linear growth trend identification and response slot locking mechanisms, it accurately intervenes in high-risk abandonment sessions, avoiding the contradiction between doctor overload and long patient waiting times in the traditional static allocation model, achieving a dynamic balance between doctor load and patient experience, and significantly improving the resource scheduling efficiency and service quality of online consultations.

Description

Technical Field

[0001] This invention relates to the field of medical information processing technology, and in particular to a resource scheduling method for Internet hospitals based on a service bus. Background Technology

[0002] Internet hospitals, as an important form of modern medical services, allow patients to receive help from doctors at any time through online consultations. Their core value lies in efficiently matching medical resources with patient needs, thereby alleviating the pressure on offline medical care and improving overall service accessibility.

[0003] In this field, physician resources remain a relatively scarce and crucial element. The efficient allocation of physicians' online time directly determines the platform's capacity to receive patients and the quality of service. Currently, many internet hospital platforms adopt a unified scheduling approach based on a service bus, allowing doctors to open multiple consultation sessions simultaneously in order to serve more patients within the same timeframe. While this approach superficially appears to increase physician workload, it reveals significant shortcomings in actual operation. Due to the significant differences in the complexity of consultations for different diseases, simple illnesses such as colds and rashes typically require only a few rounds of dialogue, while complex illnesses such as chronic disease management and analysis of difficult cases often require doctors to maintain sustained attention and engage in multiple in-depth conversations. When a large number of concurrent sessions are available, the prolonged engagement of complex illnesses leads to doctors frequently switching between multiple sessions. During these switching processes, doctors need to reread historical conversations and recover contextual information, resulting in significant delays in the response time for simpler illnesses.

[0004] A further problem is that this lag is not evenly distributed, but exhibits a significant cumulative effect. Patients with simple illnesses are typically sensitive to response speed; if they don't receive a substantial answer after their initial question, their waiting time increases rapidly, leading them to abandon the session. Although abandoned sessions disappear from the doctor's interface, the time and effort previously invested cannot be recovered, resulting in a de facto waste of resources. Simultaneously, the platform's promised service response time cannot be fulfilled, leading to a decline in patient experience and damage to overall trust. While sessions for complex illnesses can continue, doctors are tied up by an excessively long concurrency count, unable to promptly see new patients, resulting in another form of resource waste. Therefore, an irreconcilable contradiction exists between the number of concurrent sessions doctors can handle and the actual ability of patients to receive timely responses.

[0005] How to reasonably control and dynamically adjust the concurrency limit of each doctor based on the differences in the complexity of different diseases, the time spent switching between doctors' contexts, and the actual waiting experience of patients, so as to avoid excessive resource consumption by complex diseases and ensure that the response to simple diseases does not exceed the service commitment, has become a key issue that needs to be addressed in the resource scheduling of Internet hospitals. Summary of the Invention

[0006] This invention provides a resource scheduling method for internet hospitals based on a service bus, comprising the following steps: S1. Collect the disease information, patient waiting interval and number of consultation rounds of the doctor's current online consultation session. Determine the consultation complexity level of each session based on the disease information and classify them into high complexity sessions and low complexity sessions. Statistically calculate the total number of current concurrent sessions, the distribution of the number of sessions at each complexity level and the average waiting interval. S2. Extract the response time threshold corresponding to each complexity level from the service bus scheduling platform, calculate the ratio of the average waiting interval to the corresponding response time threshold, and when the ratio of any complexity level exceeds the preset intervention threshold, input the patient waiting interval and the number of consultation rounds of each session into the growth trend recognition model to identify sessions in which the waiting interval increases non-linearly with the number of consultation rounds. S3. For sessions exhibiting non-linear growth, determine the substantive response, mark sessions that do not receive a substantive response as high-risk abandonment sessions, and extract the number of high-risk abandonment sessions and their complexity level. S4. Based on the number of high-risk abandoned sessions and their complexity level, the total number of current concurrent sessions, the distribution of the number of sessions at each complexity level, and the doctor's context recovery time and switching frequency, a load balancing adjustment strategy is adopted to determine the upper limit of the target number of concurrent patient reception sessions. S5. Determine whether to reduce the patient load based on the target maximum number of concurrent patient sessions and the current total number of concurrent sessions. When reduction is confirmed, pause the allocation of new low-complexity sessions to doctors and include the unallocated sessions in the cross-doctor scheduling scope.

[0007] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects: This invention discloses a service bus-based resource scheduling method for internet hospitals. Addressing the issues of excessive concurrent doctor session load, long patient waiting times, and the identification and handling of high-risk abandonment sessions in online consultations, it proposes a dynamic adjustment mechanism. This invention collects doctor session data in real time, analyzes disease complexity, waiting intervals, and consultation rounds, identifies high-risk abandonment sessions with non-linearly increasing waiting times, and dynamically adjusts the upper limit of concurrent doctor sessions using a load balancing strategy. It suspends the allocation of simple disease sessions and prioritizes processing complex disease sessions, ensuring resources are locked until a substantial response is received. Simultaneously, this invention resumes allocation and continuously updates the upper limit when the waiting interval decreases, achieving efficient resource utilization. Its core lies in accurately intervening in high-risk sessions through non-linear growth trend identification and response slot locking, balancing doctor load and patient experience, and ultimately improving the quality of consultation services and resource scheduling efficiency. Attached Figure Description

[0008] Figure 1This is a flowchart of a service bus-based Internet hospital resource scheduling method according to the present invention.

[0009] Figure 2 This is a flowchart of a session risk identification and substantive response determination method according to the present invention.

[0010] Figure 3 This is a flowchart illustrating the load balancing adjustment and resource scheduling execution process of the present invention. Detailed Implementation

[0011] The embodiments of this disclosure will now be described in detail with reference to the accompanying drawings.

[0012] The following specific examples illustrate the implementation of this disclosure. Those skilled in the art can easily understand other advantages and effects of this disclosure from the content disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of this disclosure, and not all of them. This disclosure can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of this disclosure. It should be noted that, in the absence of conflict, the following embodiments and features in the embodiments can be combined with each other. Based on the embodiments in this disclosure, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this disclosure.

[0013] Example 1 like Figures 1 to 3 As shown, this embodiment provides a best implementation of a service bus-based internet hospital resource scheduling method, which runs on a service bus scheduling platform. The service bus scheduling platform is the core scheduling hub of the internet hospital, responsible for managing doctors' online status, session allocation queues, response time slot resources, and cross-doctor scheduling strategies. This platform interacts with the online consultation front-end system, doctor's workbench client, and patient mobile application through a standardized message bus interface, acquiring basic data such as session status change events, message sending and receiving timestamps, and doctor operation behavior logs in real time. The method in this embodiment specifically includes the following steps: S1. Collect the disease information, patient waiting interval and number of consultation rounds of the doctor's current online consultation session. Determine the consultation complexity level of each session based on the disease information and classify them into high complexity sessions and low complexity sessions. Statistically calculate the total number of current concurrent sessions, the distribution of the number of sessions at each complexity level and the average waiting interval.

[0014] In this step, the session monitoring unit of the service bus scheduling platform collects real-time data on all currently active online consultation sessions of the doctor. Specifically, the session monitoring unit obtains the message sequence records and doctor operation logs for each active session from the online consultation system through the message bus interface. The message sequence records contain structured fields such as the sender identifier, sending timestamp, message content, and message type for each message in the session. The doctor operation logs record the complete sequence of operations performed by the doctor on the workbench client, including session switching, message reading events, and reply editing events.

[0015] Regarding the collection of disease information, the session monitoring unit extracts the patient's chief complaint text, entered diagnostic codes, or disease keywords from the session metadata of each active session. In a preferred embodiment, the service bus scheduling platform maintains a disease knowledge base, which predefines the mapping relationship between common diseases and their corresponding consultation complexity levels. The consultation complexity level is used to assess the difficulty of diagnosing the disease involved in the current session. In this embodiment, it is divided into two levels: high complexity and low complexity. Low complexity corresponds to common diseases with clear treatment paths that can usually be completed in a few rounds of dialogue, such as upper respiratory tract infections and mild rashes; high complexity corresponds to diseases involving multiple complications or requiring long-term continuous monitoring, such as comprehensive management of chronic diseases and differential diagnosis of difficult and complicated diseases. When a doctor enters disease keywords in the session or the system identifies disease characteristics from the patient's chief complaint text through the natural language processing module, it can query the disease knowledge base to obtain the corresponding complexity level. In another alternative implementation, the complexity level can be dynamically determined using a machine learning classification model. This model takes features such as the length of the patient's complaint text, the number of symptom keywords, and whether multiple examinations or tests are involved as input, and outputs a complexity score for the session, which is then mapped to a high or low complexity level. This complexity determination mechanism allows for an accurate understanding of the doctor's current session load structure, providing foundational data support for subsequent differentiated scheduling strategies.

[0016] To collect patient waiting interval data, the session monitoring unit extracts the sending timestamp of the doctor's latest reply message in each active session from the message sequence record. It then calculates the difference between the current system time and this timestamp to obtain the current patient waiting interval for that session. The physical meaning of the patient waiting interval is the length of time the patient has waited since the doctor's last reply; this indicator directly reflects the patient's current waiting experience. If the patient sends a new message during this period, the calculation cutoff time for the waiting interval is updated to the sending time of the patient's latest message to more accurately reflect the patient's actual waiting experience. The session monitoring unit dynamically tracks the patient waiting interval for each session by continuously comparing the message sequence within the session with the real-time system clock.

[0017] To collect the number of consultation rounds, the session monitoring unit performs structured analysis on the message sequence records of each active session. In this embodiment, a doctor's question or reply and the patient's corresponding answer or follow-up question are defined as a complete consultation round. The session monitoring unit identifies the alternating speaking patterns of doctors and patients by parsing the sender's role identifier and the sending time order of each message in the message sequence, and accumulates the count to obtain the number of consultation rounds for that session. The number of consultation rounds reflects the depth of the session and is a key sequence index for subsequently judging the growth trend of the waiting interval.

[0018] After completing the session-by-session data collection, the session monitoring unit performs statistical aggregation. First, it iterates through all currently active sessions to obtain the total number of concurrent sessions. Second, it categorizes and counts sessions based on their determined consultation complexity levels to obtain the distribution of sessions at each complexity level, i.e., the number of high-complexity sessions and the number of low-complexity sessions. Further, it calculates the average waiting interval for each session based on its determined consultation complexity level. It selects all sessions with a low consultation complexity level, aggregates the patient waiting intervals of these sessions, and calculates the arithmetic mean to obtain the average waiting interval for low-complexity sessions; similarly, it selects all sessions with a high consultation complexity level, aggregates the patient waiting intervals of these sessions, and calculates the arithmetic mean to obtain the average waiting interval for high-complexity sessions. The average waiting interval for each complexity level reflects the overall response efficiency of doctors at different levels of diagnostic difficulty. The average waiting interval for low-complexity sessions measures the response timeliness for routine cases, while the average waiting interval for high-complexity sessions measures the response timeliness for complex cases. Together, they constitute the key criteria for triggering subsequent scheduling intervention processes.

[0019] In addition, the session monitoring unit also simultaneously collects the doctor's context recovery time and switching frequency. Context recovery time is an important indicator for measuring the doctor's workload and efficiency in multi-session concurrent scenarios. Specifically, when a doctor switches between multiple concurrent consultation sessions, the time difference from activating a paused session window to starting actual input or operation in that session is the context recovery time for that switch. At the technical implementation level, the session monitoring unit measures this time by monitoring focus events on the doctor's front-end workbench interface: when the doctor clicks a session tab to make it active, the system records an activation timestamp; when the system detects that the doctor has entered the first character or performed a send operation in that session, it records an operation timestamp; the difference between the two timestamps is the context recovery time for that switch. The system records the recovery time of the most recent few sessions for each session and calculates its moving average as the doctor's current context recovery time indicator. Switching frequency is defined as the number of times the doctor performs an activation operation between different session windows within a preset time window. The session monitoring unit listens for focus switching events of the workbench interface components. Each detected change in the active session tab is counted as a switch, and the switching frequency is accumulated within a preset time window. By collecting context recovery time and switching frequency, the cognitive load level of doctors in a multi-session concurrent state can be quantified, providing key input parameters for subsequent load balancing adjustment strategies.

[0020] S2. Extract the response time threshold corresponding to each complexity level from the service bus scheduling platform, calculate the ratio of the average waiting interval to the corresponding response time threshold, and when the ratio of any complexity level exceeds the preset intervention threshold, input the patient waiting interval and the number of consultation rounds for each session into the growth trend recognition model to identify sessions where the waiting interval increases non-linearly with the number of consultation rounds.

[0021] In this step, the service bus scheduling platform first extracts the response time thresholds specified for each complexity level from its internally stored service commitment configuration document. The service commitment configuration document is a pre-defined standard document created by the internet hospital operator and stored in the service bus scheduling platform's database. It clearly specifies the response time time doctors should take to respond to patient messages at different consultation complexity levels. The response time thresholds are set based on statistical analysis of historical consultation data, medical industry service standards, and platform user experience standards. The response time thresholds corresponding to each complexity level are read from the service commitment configuration document via a query interface and cached in the scheduling engine's local memory to support subsequent real-time comparison calculations.

[0022] After obtaining the response time threshold, the scheduling engine divides the average waiting interval calculated in the previous step for each complexity level by its corresponding response time threshold to obtain the waiting interval ratio. This ratio physically represents the deviation of the average patient waiting time at that complexity level from the promised response time. A ratio of 1 indicates that the service commitment boundary has been reached, while a ratio greater than 1 indicates that the average waiting time has exceeded the promised response time. The scheduling engine then obtains a preset intervention threshold, a configurable parameter determined based on platform operational experience and the statistical distribution of patient abandonment behavior in historical data. When the waiting interval ratio for any complexity level exceeds the intervention threshold, it indicates that the patient's waiting experience at that level has approached or exceeded the tolerance boundary of the service commitment, requiring a deeper level of session state analysis. At this point, the scheduling engine uses the patient waiting interval and the number of consultation rounds for all active sessions as input datasets and feeds them into a growth trend recognition model to identify non-linear growth characteristics.

[0023] By setting an intervention threshold as a trigger condition instead of performing trend analysis on each session directly, the computationally intensive trend identification process can be effectively avoided when the system is running well overall, thereby reducing the computational overhead of the service bus scheduling platform while ensuring scheduling sensitivity.

[0024] Furthermore, the specific process by which the growth trend recognition model identifies the nonlinear growth features of the waiting interval sequence for each session is as follows: The patient waiting interval for each session is used as a time series observation, and the number of consultation rounds is used as the sequence index to construct a waiting interval time series. The waiting interval time series is preprocessed by calculating the incremental value of the waiting interval between adjacent rounds to obtain an incremental sequence. The incremental sequence is then fitted with a linear or nonlinear trend to determine whether the incremental sequence exhibits an increasing characteristic. When the incremental sequence exhibits an increasing characteristic, it is determined that the patient waiting interval for the corresponding session exhibits a nonlinear increase with the number of consultation rounds.

[0025] Specifically, the growth trend identification model first constructs a waiting interval time series for each active session. Let N be the total number of consultation rounds in a session. Then, using the sequence number n (n=1,2,...,N) of each round as the sequence index and the patient waiting interval W(n) corresponding to that round as the sequence observation, a waiting interval time series {W(1), W(2), ..., W(N)} is constructed. Here, W(n) represents the time the patient waits from sending a message to receiving a doctor's reply in the nth round of consultation.

[0026] After constructing the waiting interval time series, the growth trend identification model performs preprocessing operations on it. The core step of preprocessing is to calculate the incremental value of the waiting interval between adjacent rounds, that is, for the nth round and the (n-1)th round, calculate the incremental value ΔW(n) = W(n) - W(n-1), where n=2,3,...,N. This yields the incremental sequence {ΔW(2), ΔW(3), ..., ΔW(N)}. The physical meaning of the incremental sequence is the change in the patient waiting interval between each two adjacent rounds of consultation. If the value of the incremental sequence is positive and shows an increasing trend, it indicates that the patient waiting interval is not only increasing, but the rate of increase is also accelerating, which is a typical characteristic of nonlinear growth (such as quadratic growth or exponential growth). By converting the original waiting interval sequence into an incremental sequence for analysis, the interference of the difference in the baseline value of the waiting interval on the trend judgment can be eliminated, and the changing pattern of the growth rate can be captured more directly.

[0027] After obtaining the incremental sequence, the growth trend identification model performs trend fitting analysis on it. In a preferred embodiment, the model uses the least squares method to perform linear regression fitting on the incremental sequence, with the fitting model being ΔW(n) = α·n + β, where α is the fitting slope and β is the intercept. The physical meaning of the fitting slope α is the trend coefficient of the incremental sequence changing with the round index. When the fitting slope α is positive and its statistical significance meets the pre-set reliability requirements, it is determined that the incremental sequence exhibits an increasing characteristic, and further, it is determined that the patient waiting interval of the corresponding session exhibits a non-linear growth with the number of consultation rounds. If the fitting slope α is close to zero or is negative, it indicates that the incremental sequence does not have an increasing characteristic, and the change in the waiting interval of the corresponding session shows a linear or decelerating trend, and does not belong to a non-linear growth session.

[0028] In another alternative implementation, for sessions with a large number of rounds, the growth trend identification model can also use multinomial regression or nonparametric local regression methods to fit the incremental sequence to a nonlinear trend, in order to more accurately capture the nonlinear increasing patterns that may exist in the incremental sequence. When the fitting results show that the incremental sequence has a significant upward convex increasing trend, the corresponding session is also determined to exhibit nonlinear growth.

[0029] Through the aforementioned trend identification process, a subset of sessions where the waiting interval increases non-linearly with each round of consultation can be accurately screened from all active sessions. Patients in these sessions are experiencing a rapidly deteriorating waiting experience and are the high-risk group most likely to give up due to excessive waiting time. Timely identification of these sessions and initiation of subsequent intervention measures are crucial for preventing resource hoarding and maintaining service commitments.

[0030] S3. For sessions exhibiting non-linear growth, determine the substantive response, mark sessions that do not receive a substantive response as high-risk abandonment sessions, and extract the number of high-risk abandonment sessions and their complexity level.

[0031] In this step, the service bus scheduling platform performs a substantive response determination on each session identified in the previous step that exhibits non-linear growth characteristics. This distinguishes between sessions where the waiting interval, although non-linearly increasing, has already received effective medical guidance from a doctor, and sessions that are indeed in a prolonged state of ineffective waiting. The non-linear growth of the waiting interval only reflects a deteriorating trend over time; whether the patient has received substantive medical feedback is another key factor determining their willingness to abandon the session. If the patient has received a clear diagnosis, treatment plan, or examination recommendation from a doctor, the probability of abandoning the session will significantly decrease even if the subsequent waiting interval increases. Conversely, if the patient only receives greetings, follow-up questions, or procedural responses after a long wait, the probability of abandoning the session will increase significantly. Therefore, performing a logical AND operation between non-linear growth identification and substantive response determination is a necessary technical means to accurately locate sessions with high risk of abandonment.

[0032] Furthermore, the specific process for determining a substantive response is as follows: Extract the latest doctor's response content from the non-linearly growing conversation; perform semantic analysis on the latest response content to identify whether it contains at least one of a diagnostic conclusion, treatment plan, or examination suggestion; when the latest response content contains at least one of the above, the conversation is determined to have received a substantive response; when the latest response content does not contain any of the diagnostic conclusion, treatment plan, or examination suggestion, the conversation is determined not to have received a substantive response; thus, a substantive response determination result is obtained.

[0033] Specifically, the semantic analysis module of the service bus scheduling platform first extracts the latest reply from the message sequence records for each session that exhibits a non-linear growth characteristic. The latest reply refers to the complete text of the last message sent by the doctor in that session. The semantic analysis module performs semantic parsing on the latest reply to determine whether it contains at least one piece of information with substantial diagnostic and treatment value, such as a diagnostic conclusion, treatment plan, or examination suggestion.

[0034] Preferably, the specific implementation of semantic parsing of the latest response content is as follows: The latest response content is segmented and keywords are extracted; the extracted keywords are matched with a preset medical substantive response terminology database, which includes a diagnostic terminology set, a treatment plan terminology set, and an examination item set; the keyword hit rate is calculated based on the matching results; when the keyword hit rate reaches a preset matching threshold, an intent recognition algorithm is used to analyze the sentence structure of the latest response content to determine whether the latest response content contains guiding statements; based on the keyword hit rate and the determination result of whether the latest response content contains guiding statements, the substantive response judgment result is comprehensively determined.

[0035] Specifically, the semantic analysis module first performs word segmentation on the doctor's latest reply, dividing the continuous natural language text into independent word units. After word segmentation, it extracts keywords using part-of-speech tagging and named entity recognition technologies, focusing on extracting key information such as medical terms, disease names, drug names, examination names, and descriptions of diagnostic and treatment actions.

[0036] Subsequently, the semantic analysis module matches the extracted keywords with a pre-defined medical substantive response thesaurus. This thesaurus is a structured terminology knowledge base pre-built and maintained by the service bus scheduling platform, containing three subsets: a diagnostic terminology set, which includes terms such as diagnostic descriptions of common diseases, syndrome names, and differential diagnostic conclusions; a treatment plan terminology set, which includes terms such as drug treatment instructions, physical therapy plans, surgical recommendations, and lifestyle intervention recommendations; and an examination item set, which includes the names of common examination and testing items such as laboratory tests, imaging tests, and functional tests. The semantic analysis module performs a matching operation on each extracted keyword with the terms in the above three subsets, and calculates the ratio of the number of matched keywords to the total number of extracted keywords to obtain the keyword hit rate. The physical meaning of the keyword hit rate is the proportion of information related to substantive diagnostic and treatment terminology in the doctor's latest response content. The higher this ratio, the more likely the response content contains substantive diagnostic and treatment guidance.

[0037] When the keyword hit rate reaches the preset matching threshold, the semantic analysis module further employs an intent recognition algorithm to analyze the sentence structure of the doctor's latest response. The preset matching threshold is set based on the results of cross-validation on historical annotated corpora, selecting a threshold level that achieves a reasonable balance between accuracy and recall. The purpose of the intent recognition algorithm is to determine whether the response contains guiding statements, i.e., statements with clear diagnostic or actionable implications. Characteristics of guiding statements include, but are not limited to: predicate structures such as "suggest," "should," "need," "can be taken," and "go for examination"; conclusive statements such as "diagnosed as," "considered as," and "preliminary judgment"; or operational details such as specific drug dosage, usage, and examination sites. When the response simultaneously meets the criteria of a keyword hit rate reaching the matching threshold and containing guiding statements, the conversation is considered to have received a substantive response; otherwise, it is considered not to have received a substantive response.

[0038] By employing a dual-judgment mechanism that combines keyword hit rate and guidance statement identification, it is possible to effectively distinguish between doctor responses containing substantial diagnostic and treatment value and those that are merely procedural responses or follow-up questions. This avoids misjudging non-substantive responses such as "Please wait" or "Are there any other symptoms?" as completed treatment guidance. This two-layer filtering mechanism improves the accuracy of substantive response identification, thereby making the marking of high-risk abandonment sessions more precise and reliable.

[0039] After determining the substantive response, the service bus scheduling platform performs a logical AND operation on each session exhibiting non-linear growth characteristics: when a session simultaneously meets both conditions—a non-linearly increasing waiting interval and no substantive response—it is marked as a high-risk abandonment session. After marking, the platform extracts the total number of all marked high-risk abandonment sessions and the diagnostic complexity level of each session, grouping and counting them according to complexity level to obtain the distribution of high-risk abandonment sessions across high and low complexity levels. This data will serve as a key input for subsequent load balancing adjustment strategies.

[0040] S4. Based on the number of high-risk sessions and their complexity level, the total number of current concurrent sessions, the distribution of sessions at each complexity level, and the doctor's context recovery time and switching frequency, a load balancing adjustment strategy is adopted to determine the upper limit of the target number of concurrent patient reception sessions.

[0041] In this step, the load balancing adjustment module of the service bus scheduling platform integrates the multi-dimensional data collected and calculated in the preceding steps, and through systematic evaluation and adjustment calculations, determines the upper limit of the target number of concurrent patient sessions that doctors should maintain. This upper limit must be determined to maximize the doctor's effective patient reception capacity while ensuring that the patient's waiting experience does not exceed the service commitment boundary, and to avoid resource idleness caused by excessive reduction in concurrency.

[0042] Furthermore, the specific process for determining the target upper limit of concurrent patient reception sessions using the load balancing adjustment strategy is as follows: Calculate the ratio of high-risk abandonment sessions to the total number of sessions at each complexity level to obtain the risk proportion for each level; comprehensively evaluate the risk proportion for each level with the current total number of concurrent sessions to determine the current load state; assess the impact of the context recovery time on the doctor's effective response capability based on the current load state and the context recovery time; combine the impact level with the switching frequency to perform a weighted adjustment calculation on the current total number of concurrent sessions to obtain a suggested value for the number of concurrent sessions; compare and verify the suggested value for the number of concurrent sessions with the preset doctor reception capability constraints, which include the maximum number of concurrent sessions set by the doctor or the upper limit of reception configured by the platform, to determine the target upper limit of concurrent patient reception sessions.

[0043] Specifically, the load balancing adjustment module first calculates the proportion of sessions with high abandonment risk within each complexity level. Let the total number of high-complexity sessions be N_H, and the number of sessions with high abandonment risk be R_H. Then, the risk proportion of the high-complexity level is P_H = R_H / N_H. Let the total number of low-complexity sessions be N_L, and the number of sessions with high abandonment risk be R_L. Then, the risk proportion of the low-complexity level is P_L = R_L / N_L. It is important to note that when the total number of sessions in a certain level is zero, that level is not included in the risk proportion calculation; its risk proportion is directly recorded as zero to avoid division by zero anomalies. The risk proportion of each level reflects the degree of session quality deterioration at that complexity level; a higher risk proportion indicates greater scheduling pressure at that level.

[0044] Subsequently, the load balancing adjustment module comprehensively evaluates the risk proportion of each level along with the total number of concurrent sessions to determine the current load status. The current load status is a comprehensive quantitative description of the overall workload of doctors, considering not only the absolute number of concurrent sessions but also the distribution of high-risk sessions within each level. In actual evaluation, if the risk proportion of high-complexity levels is significantly higher than that of low-complexity levels, it indicates that the backlog of complex sessions is the main problem in the current scheduling; conversely, if the risk proportion of low-complexity levels is higher, it indicates that the response lag of simple sessions is more prominent. The load balancing adjustment module determines the direction and intensity of adjustment based on these differentiated risk distribution characteristics.

[0045] After determining the current load status, the load balancing adjustment module introduces a context recovery time metric to assess its impact on doctors' effective response capabilities. A longer context recovery time indicates that doctors need to spend more time re-understanding the patient's condition and the dialogue context each time they switch sessions. This time is considered non-effective treatment time, directly weakening the doctor's actual response capability. The assessment of the impact is based on the product effect of context recovery time and the total number of concurrent sessions: when both the number of concurrent sessions and the recovery time are high, the proportion of effective treatment time for doctors will decrease sharply, and the impact will exhibit a non-linear amplification trend.

[0046] Preferably, the weighted adjustment operation is implemented as follows: the number of high-risk abandonment sessions, the distribution of session numbers at each complexity level, the total number of current concurrent sessions, the context recovery time, and the switching frequency are used as adjustment factors; each adjustment factor is normalized to eliminate dimensional differences; weight coefficients are assigned to each normalized adjustment factor, and the weight coefficients are determined based on the correlation strength between each adjustment factor and the patient waiting interval; a weighted summation operation is performed on each normalized adjustment factor and its corresponding weight coefficient to obtain a comprehensive load score; the recommended value for the number of concurrent sessions is determined by querying a preset mapping table between the comprehensive load score range and the upper limit of the number of concurrent sessions based on the comprehensive load score.

[0047] Specifically, the load balancing adjustment module uses the following five dimensions as adjustment factors: the first adjustment factor is the number of sessions with high abandonment risk, reflecting the scale of sessions currently at risk of abandonment; the second adjustment factor is the distribution of the number of sessions at each complexity level, quantified by the ratio of the number of high-complexity sessions to the total number of concurrent sessions, reflecting the structural characteristics of the session load; the third adjustment factor is the total number of current concurrent sessions, reflecting the absolute load level of doctors; the fourth adjustment factor is the context recovery time, reflecting the cognitive cost of doctors when switching sessions; and the fifth adjustment factor is the switching frequency, reflecting the frequency of doctors' attention being distracted.

[0048] Because the five adjustment factors mentioned above have different units of measurement—for example, the number of high-risk sessions is measured in terms of quantity, context recovery time in terms of time, and handover frequency in terms of frequency—they cannot be directly weighted. Therefore, the load balancing adjustment module first normalizes each adjustment factor, mapping them to a unified dimensionless numerical range. The normalization method can be min-max normalization or Z-score standardization, the specific choice depending on the distribution characteristics of each factor. After normalization, the values ​​of each factor are all within a comparable numerical range, eliminating the impact of unitary differences on the weighting calculation.

[0049] After normalization, the load balancing adjustment module assigns weight coefficients to each normalized adjustment factor. The weight coefficients are determined based on the correlation strength analysis of each adjustment factor with patient waiting intervals. In a preferred embodiment, the weight coefficients are determined through multiple regression analysis of historical operational data: using the average patient waiting interval as the dependent variable and the historical values ​​of each adjustment factor as independent variables, a regression fit is performed. The standardized regression coefficients of each independent variable reflect the explanatory power of that factor for changes in waiting intervals. Factors with higher correlation strength are assigned larger weight coefficients. The sum of the weight coefficients is normalized to 1 to ensure the consistency of the overall score.

[0050] After determining the weighting coefficients, the load balancing adjustment module performs a weighted summation operation on each normalized adjustment factor and its corresponding weighting coefficient. Let the five normalized adjustment factors be x1, x2, x3, x4, and x5, and the corresponding weighting coefficients be w1, w2, w3, w4, and w5. Then, the formula for calculating the comprehensive load score S is: S = w1·x1 + w2·x2 + w3·x3 + w4·x4 + w5·x5, where S is the comprehensive load score, x1 to x5 are the normalized number of high-risk sessions, the proportion of high-complexity sessions, the total number of current concurrent sessions, context recovery time, and switching frequency, respectively, and w1 to w5 are the corresponding weighting coefficients, satisfying w1+w2+w3+w4+w5=1. A larger comprehensive load score S indicates a greater overall load pressure on the doctor, requiring a more significant reduction in the number of concurrent sessions. Since all factors are normalized and the sum of their weight coefficients is 1, the comprehensive load score S ranges from 0 to 1, where 0 indicates the lightest load and 1 indicates the heaviest load.

[0051] Finally, the load balancing adjustment module queries a pre-defined mapping table between load score intervals and concurrent access limits based on the comprehensive load score to determine the recommended concurrent access limit. This mapping table uses pre-configured segmented mapping rules from the service bus scheduling platform, dividing the continuous value space of the comprehensive load score into several discrete score intervals, each corresponding to a recommended concurrent access limit. Higher score intervals correspond to lower concurrent access limits, reflecting the scheduling principle of "heavier load, lower concurrency." After obtaining the recommended concurrent access limit, the load balancing adjustment module compares and verifies it with pre-defined doctor capacity constraints. These constraints include the maximum number of concurrent sessions set by the individual doctor in the system and the access limit configured by the platform operator; the smaller of these two values ​​is taken as the upper bound of the constraint. The final target concurrent access session limit is the smaller of the recommended concurrent access limit and the upper bound of the constraint, ensuring that the target limit meets the scheduling needs of load balancing without exceeding the actual access limit of doctors.

[0052] Through the above-mentioned multi-dimensional weighted evaluation and segmented mapping load balancing adjustment strategy, the upper limit of the target concurrent consultation session can be determined in a refined manner according to the doctor's current actual load status, risk distribution characteristics and cognitive load level, avoiding the problems of excessive reduction or insufficient adjustment that may be caused by simple and crude fixed threshold adjustment.

[0053] S5. Determine whether to reduce the patient load based on the target maximum number of concurrent patient sessions and the current total number of concurrent sessions. When reduction is confirmed, pause the allocation of new low-complexity sessions to doctors and include the unallocated sessions in the cross-doctor scheduling scope.

[0054] Specifically, in this embodiment of the application, step S5 further includes: including low-complexity sessions in high-risk abandonment sessions in the cross-doctor scheduling scope, allocating high-complexity sessions to continuous response time slots and locking doctor resources until the session receives a substantial response or the session ends; and restoring session allocation when the average waiting interval falls below the response time threshold.

[0055] In this step, the service bus scheduling platform executes specific load reduction operations and session scheduling strategies based on the comparison between the target maximum number of concurrent patient sessions determined in the previous step and the current total number of concurrent sessions. This step is the core of the entire resource scheduling method, involving multiple collaborative steps such as load reduction determination, pausing new session allocation, transferring cross-doctor sessions, locking high-complexity session resources, and resuming allocation.

[0056] First, the scheduling engine compares the target maximum number of concurrent patient sessions with the current total number of concurrent sessions. When the current total number of concurrent sessions exceeds the target maximum number of concurrent patient sessions, it confirms the need to reduce the patient load. The first step in the reduction operation is to pause the allocation of new low-complexity sessions for that doctor. The scheduling engine marks the doctor's status as "low-complexity paused allocation" in the session allocation queue. Afterward, when new low-complexity session requests enter the allocation queue, the allocation engine will skip that doctor and no longer push new low-complexity sessions to them. Low-complexity sessions that fail to be allocated to this doctor are included in the cross-doctor scheduling scope.

[0057] Furthermore, the cross-doctor scheduling scope refers to the scheduling queue of other online doctors in the service bus scheduling platform that can receive new sessions; the continuous response time slot is a scheduling resource unit in which a doctor focuses on processing a single session within a specified time period; when a low-complexity session in a high-risk abandonment session is included in the cross-doctor scheduling scope, it is reallocated to other available doctor resources by the service bus scheduling platform.

[0058] Specifically, the cross-doctor scheduling scope is a logical subset of the global resource pool of the service bus scheduling platform, including the scheduling queues of other doctors currently online whose total number of concurrent sessions has not reached their target concurrent consultation session limit. When a doctor suspends accepting new low-complexity sessions due to excessive load, the low-complexity sessions that should have been allocated to that doctor, as well as the low-complexity sessions that the doctor currently holds with high risk of abandonment, are all included in the cross-doctor scheduling scope. Within the cross-doctor scheduling scope, the service bus scheduling platform reallocates these sessions based on factors such as the professional domain matching degree of other available doctors, current load level, and historical response efficiency, transferring these sessions to doctors with relatively ample load to ensure that patients can receive a response within a reasonable time. Through the cross-doctor scheduling mechanism, low-complexity sessions can be prevented from accumulating in the waiting queue for a long time due to the excessive load of a single doctor, effectively improving the platform's overall session processing capacity and patient response efficiency.

[0059] A continuous response time slot is a dedicated time resource unit reserved by the service bus scheduling platform for sessions requiring centralized processing. Within a continuous response time slot, the doctor's attention and operational resources are locked onto a single designated session, unaffected by message pushes from other sessions, ensuring that the doctor can provide continuous and in-depth diagnostic and treatment services for that session. The duration of a continuous response time slot is dynamically set based on the response time threshold corresponding to the complexity level of the session and the current waiting interval timeout level.

[0060] For high-complexity sessions within high-abandonment-risk sessions, the service bus scheduling platform allocates them to continuous response time slots and locks physician resources. Since high-complexity sessions require physicians to invest significant time in thinking and communication, and if such sessions are interrupted or abandoned by the patient, the physician's previously accumulated in-depth diagnostic and treatment efforts will be completely wasted. Therefore, a resource locking strategy is adopted for high-complexity, high-abandonment-risk sessions to ensure they receive priority and continuous physician attention.

[0061] Preferably, the specific process of allocating high-complexity sessions in high-abandonment-risk sessions to continuous response time slots and locking doctor resources is as follows: Based on the patient waiting interval and the response time threshold corresponding to the complexity level of each high-complexity session in the high-abandon-risk sessions, calculate the overflow amount of the patient waiting interval exceeding the response time threshold; calculate the waiting interval growth rate based on the change in the patient waiting interval between adjacent rounds; perform a comprehensive scoring calculation based on the overflow amount, the number of consultation rounds of the corresponding session, and the waiting interval growth rate to obtain the risk score of each high-complexity session; determine the allocation priority according to the risk score from high to low; allocate continuous response time slots to each high-complexity session in the service bus scheduling platform according to the allocation priority; lock doctor resources to the corresponding high-complexity session within the continuous response time slot, and block message pushes from other sessions; continuously monitor whether the locked high-complexity session receives the substantial response, and when the substantial response is received or the session status becomes closed, release the locking of the doctor resources.

[0062] Specifically, the scheduling engine first calculates a risk score for each high-risk, high-complexity session to determine processing priority. The risk score is calculated based on a comprehensive score across three dimensions. The first dimension is the overflow, which is the difference between the current patient waiting interval and the response time threshold corresponding to its complexity level. The physical meaning of the overflow is the absolute amount by which the patient's waiting time exceeds the service commitment boundary; a larger overflow indicates a more severe breach of the service commitment for that session. If the patient's waiting interval has not yet exceeded the response time threshold (i.e., the overflow is negative), the overflow is set to zero for subsequent calculations. The second dimension is the waiting interval growth rate, obtained by calculating the ratio of the change in patient waiting interval between the two most recent adjacent rounds to the waiting interval of the previous round. The waiting interval growth rate reflects the accelerating rate of waiting time deterioration; a higher growth rate indicates that the patient's waiting experience is deteriorating at a faster pace. When calculating the growth rate, if the waiting interval of the previous round is zero or a very small value, a very small correction constant is introduced into the denominator to ensure numerical stability and prevent the division result from tending towards infinity. The third dimension is the number of consultation rounds. The more rounds, the more diagnostic and treatment efforts the doctor has invested in the session. If the patient abandons the session at this point, it will lead to a greater waste of resources.

[0063] The specific method for calculating the comprehensive score is as follows: After normalizing the overflow amount, waiting interval growth rate, and number of consultation rounds, the scores are weighted and summed according to preset weighting coefficients to obtain the risk score for the session. The formula for calculating the risk score is: F = λ1·O' + λ2·G' + λ3·R', where F is the risk score, O' is the normalized overflow amount, G' is the normalized waiting interval growth rate, R' is the normalized number of consultation rounds, and λ1, λ2, and λ3 are the corresponding weighting coefficients, satisfying λ1+λ2+λ3=1. The principle for setting the weighting coefficients is: the overflow amount reflects the severity of the current default and is given a higher weight; the waiting interval growth rate reflects the urgency of the deterioration trend and is given a medium weight; the number of consultation rounds reflects the scale of sunk costs and is given a secondary weight. Through the above weighted scoring mechanism, the three dimensions of the current default severity, deterioration rate, and amount of resources invested can be organically combined to accurately quantify the risk urgency of each high-complexity session.

[0064] The scheduling engine sorts all high-risk, high-complexity sessions according to their risk scores from highest to lowest to determine allocation priorities. Sessions with the highest risk scores receive priority allocation of continuous response time slots. The scheduling engine then allocates continuous response time slots to each high-complexity session in the service bus scheduling platform according to allocation priority. Within a continuous response time slot, the scheduling engine locks physician resources to the corresponding high-complexity session. The locking operation is implemented as follows: the scheduling engine sends a resource locking command to the physician's workbench client via the service bus. Upon receiving the locking command, the physician's workbench automatically switches the interface focus to the target session and simultaneously blocks push notifications from other sessions, ensuring that the physician is not disturbed by other sessions during the locking period and can concentrate on providing in-depth diagnostic and treatment services to the target high-complexity session.

[0065] During resource locking, the service bus scheduling platform continuously monitors whether the locked high-complexity sessions have received a substantial response. The monitoring mechanism reuses the substantial response determination logic from the aforementioned steps. When it detects that the content of a doctor's response in the session passes the substantial response determination (i.e., includes at least one of a diagnostic conclusion, treatment plan, or examination suggestion), the scheduling engine immediately releases the lock on the doctor's resources, resumes message pushing for other sessions, and allows the doctor to re-enter the multi-session concurrent processing state. Similarly, if the session status becomes closed due to the patient's active closure or timeout mechanism, the scheduling engine also immediately releases the resource lock.

[0066] Furthermore, within the continuous response time slot, if the locked high-complexity session does not receive the substantial response within the preset time slot period and the session status does not change to closed, the continuous response time slot is extended to the next available time slot period and the doctor resource is kept locked; at the same time, the low-complexity sessions waiting to be allocated by the newly connected service bus scheduling platform during the locking period are redirected to other available doctor resources in the service bus scheduling platform whose current load is lower than the target concurrent consultation session limit.

[0067] Specifically, each continuous response time slot has a preset time slot period, which is dynamically determined based on the historical average processing time of high-complexity sessions and the current overflow level. If a locked high-complexity session neither receives a substantial response nor closes at the end of a time slot period, it indicates that the diagnosis and treatment difficulty of the session is higher than expected or the doctor needs more continuous processing time. In this case, the scheduling engine automatically extends the continuous response time slot to the next available time slot period while maintaining the locked state of the doctor's resources. During the time slot extension, because the doctor's resources are continuously locked and cannot process other sessions, newly accessed low-complexity sessions on the service bus scheduling platform cannot be assigned to that doctor. The scheduling engine redirects these newly accessed low-complexity sessions to other available doctor resources across the doctor scheduling scope, specifically selecting doctors whose current total number of concurrent sessions is lower than their target concurrent consultation session limit for allocation, ensuring that these low-complexity sessions do not get stuck in long waiting times due to the resource locking of a single doctor. By combining time slot extension with a low-complexity session redirection mechanism, it is possible to ensure that high-complexity, high-risk sessions receive sufficient attention while avoiding complete interruption of responses to low-complexity sessions, thus maintaining the platform's overall session processing capacity.

[0068] Further, when the average waiting interval of the low-complexity sessions falls below the corresponding response time threshold, session allocation is restored and the target concurrent patient access session limit is updated. Specifically, this includes: after completing the locking process for the high-complexity sessions in the high-abandonment-risk sessions, re-collecting the patient waiting interval of the low-complexity sessions and calculating the average waiting interval of the low-complexity sessions; comparing the average waiting interval of the low-complexity sessions with the corresponding response time threshold; when the average waiting interval of the low-complexity sessions is lower than the response time threshold, re-allocating new low-complexity sessions to doctors; and re-executing the collection and statistics operations and load balancing adjustment strategy based on the current total number of concurrent sessions after restoration and the distribution of the number of sessions at each complexity level, and updating the target concurrent patient access session limit.

[0069] Specifically, after completing the continuous response time slot allocation and resource locking processing for high-risk, high-complexity sessions, the service bus scheduling platform enters the recovery monitoring phase. During this phase, the session monitoring unit continuously re-collects patient waiting intervals for currently active low-complexity sessions and calculates the average waiting interval for these sessions in real time. Since the previous load reduction operation has paused the allocation of new low-complexity sessions, and high-risk, low-complexity sessions have been transferred to other physicians, the average waiting interval for low-complexity sessions is expected to gradually decrease over time as physicians focus on processing the remaining sessions.

[0070] The session monitoring unit continuously compares the recalculated average waiting interval for low-complexity sessions with the corresponding response time threshold. When the average waiting interval for low-complexity sessions drops below the response time threshold, it indicates that the doctor's current response efficiency has recovered to within the range allowed by the service commitment. At this point, the scheduling engine removes the "low-complexity paused allocation" status flag from the doctor and resumes allocating new low-complexity sessions for that doctor. After resuming allocation, the scheduling engine, based on the current total number of concurrent sessions after recovery and the distribution of session numbers at each complexity level, re-executes the data collection and statistical operations and load balancing adjustment strategies in the aforementioned steps, calculates the updated comprehensive load score, and queries the mapping table to determine the new target upper limit for the number of concurrent patient sessions.

[0071] The aforementioned recovery mechanism forms a complete dynamic adjustment closed loop: when doctors are overloaded, the system intervenes by reducing the number of concurrent sessions, suspending the allocation of low-complexity sessions, transferring low-complexity sessions, and locking high-complexity session resources; once the intervention is effective and the waiting interval returns to a reasonable level, the system automatically resumes session allocation and reassesses the load status, forming a continuous dynamic equilibrium adjustment. This closed-loop mechanism ensures that the number of concurrent consultation sessions for doctors is always maintained at a level that matches their current actual processing capacity, avoiding both the worsening of patient waiting and session abandonment caused by overload, and the idle doctor resources and wasted consultation capacity caused by excessive reduction, thereby achieving a dynamic balance between doctor resource utilization and service commitment fulfillment rate.

[0072] Through the coordinated execution of the above steps, the service bus-based internet hospital resource scheduling method in this embodiment can perceive the load status of doctors in multi-session concurrent scenarios in real time, accurately identify sessions facing high risk of abandonment due to non-linear increases in waiting intervals and lack of substantial responses, and dynamically determine the upper limit of the number of concurrent consultation sessions for doctors through a multi-dimensional weighted load balancing adjustment strategy. At the execution level, by pausing the allocation of low-complexity sessions, transferring across doctor sessions, locking high-complexity session resources, and implementing an allocation recovery mechanism based on waiting interval decline, it achieves refined scheduling of doctor resources, effectively preventing complex disease sessions from excessively crowding out the response slots of simple disease sessions, reducing the probability of patients actively abandoning sessions due to excessive waiting, ensuring the fulfillment rate of platform service commitments, and improving the overall quality and resource utilization efficiency of online consultation services in internet hospitals.

[0073] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

[0074] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0075] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0076] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

Claims

1. A service bus-based Internet hospital resource scheduling method, characterized in that, Includes the following steps: S1. Collect the disease information, patient waiting interval and number of consultation rounds of the doctor's current online consultation session. Determine the consultation complexity level of each session based on the disease information and classify them into high complexity sessions and low complexity sessions. Statistically calculate the total number of current concurrent sessions, the distribution of the number of sessions at each complexity level and the average waiting interval. S2. Extract the response time threshold corresponding to each complexity level from the service bus scheduling platform, calculate the ratio of the average waiting interval to the corresponding response time threshold, and when the ratio of any complexity level exceeds the preset intervention threshold, input the patient waiting interval and the number of consultation rounds of each session into the growth trend recognition model to identify sessions in which the waiting interval increases non-linearly with the number of consultation rounds. S3. For sessions exhibiting non-linear growth, determine the substantive response, mark sessions that do not receive a substantive response as high-risk abandonment sessions, and extract the number of high-risk abandonment sessions and their complexity level. S4. Based on the number of high-risk abandoned sessions and their complexity level, the total number of current concurrent sessions, the distribution of the number of sessions at each complexity level, and the doctor's context recovery time and switching frequency, a load balancing adjustment strategy is adopted to determine the upper limit of the target number of concurrent patient reception sessions. S5. Determine whether to reduce the patient load based on the target maximum number of concurrent patient sessions and the current total number of concurrent sessions. When reduction is confirmed, pause the allocation of new low-complexity sessions to doctors and include the unallocated sessions in the cross-doctor scheduling scope.

2. The method of claim 1, wherein, In step S2, the patient waiting interval and the number of consultation rounds for each session are input into the growth trend recognition model for identification, specifically including: The patient waiting interval for each session is used as a time series observation, and the number of consultation rounds is used as the sequence index to construct a waiting interval time series. The waiting interval time series is preprocessed by calculating the incremental value of the waiting interval between adjacent rounds to obtain an incremental sequence. The incremental sequence is then fitted with a linear or nonlinear trend to determine whether the incremental sequence exhibits an increasing characteristic. When the incremental sequence exhibits an increasing characteristic, it is determined that the patient waiting interval for the corresponding session exhibits a nonlinear increase with the number of consultation rounds.

3. The method according to claim 1, characterized in that, In step S3, a substantive response determination is made for the sessions exhibiting non-linear growth, specifically including: Extract the latest doctor's replies from the conversation, which exhibits non-linear growth. Semantic analysis is performed on the latest response content to identify whether the latest response content contains at least one of the following: diagnostic conclusion, treatment plan, or examination suggestion; If yes, it is determined that a substantive response has been received; if no, it is determined that the session has not received a substantive response.

4. The method according to claim 1, characterized in that, In step S4, a load balancing adjustment strategy is used to determine the upper limit of the target concurrent consultation session count, specifically including: Calculate the ratio of high-risk abandonment sessions to the total number of sessions within each complexity level to obtain the risk percentage for each level; The risk percentage of each level is comprehensively evaluated with the total number of current concurrent sessions to determine the current load status. Based on the current load status and the context recovery time, the impact of the context recovery time on the doctor's effective response capability is assessed. Combining the impact level and the switching frequency, a weighted adjustment calculation is performed on the total number of current concurrent sessions to obtain a suggested value for the number of concurrent sessions. The suggested value for the number of concurrent sessions is compared and verified with the preset doctor's consultation capacity constraint, which includes the maximum number of concurrent sessions set by the doctor or the consultation limit configured by the platform, to determine the upper limit of the target number of concurrent consultation sessions.

5. The method according to claim 1, characterized in that, Step S5 further includes: Based on the patient waiting interval and the response time threshold of the corresponding complexity level in each of the high-risk abandonment sessions, the overflow amount of the patient waiting interval exceeding the response time threshold is calculated; the waiting interval growth rate is calculated based on the change of the patient waiting interval in adjacent rounds; a comprehensive scoring calculation is performed based on the overflow amount, the number of consultation rounds in the corresponding session, and the waiting interval growth rate to obtain the risk score of each of the high-complexity sessions; the allocation priority is determined according to the risk score from high to low; continuous response time slots are allocated to each of the high-complexity sessions in the service bus scheduling platform according to the allocation priority; within the continuous response time slots, doctor resources are locked to the corresponding high-complexity session, and message pushes from other sessions are blocked; the locking of doctor resources is continuously monitored to see if the locked high-complexity session receives the substantial response, and when the substantial response is received or the session status becomes closed, the locking of doctor resources is released.

6. The method according to claim 1, characterized in that, In step S5, when the average waiting interval of the low-complexity session falls below the corresponding response time threshold, session allocation is restored and the target concurrent consultation session limit is updated. Specifically, this includes: After completing the locking process for the high-complexity sessions in the high-risk abandonment sessions, the patient waiting intervals of the low-complexity sessions are re-collected and the average waiting interval of the low-complexity sessions is calculated; the average waiting interval of the low-complexity sessions is compared with the corresponding response time threshold; when the average waiting interval of the low-complexity sessions is lower than the response time threshold, a new low-complexity session is reassigned to the doctor; based on the current total number of concurrent sessions after the reassignment and the distribution of the number of sessions at each complexity level, the collection and statistics operations in step S1 and the load balancing adjustment strategy in step S4 are re-executed to update the target concurrent patient reception session limit.

7. The method according to claim 5, characterized in that, In step S5: The cross-doctor scheduling scope refers to the scheduling queue of other online doctors in the service bus scheduling platform that can receive new sessions; the continuous response time slot is a scheduling resource unit in which a doctor focuses on processing a single session within a specified time period; when a low-complexity session in a high-risk abandonment session is included in the cross-doctor scheduling scope, it is reallocated to other available doctor resources by the service bus scheduling platform.

8. The method according to claim 4, characterized in that, The weighted adjustment calculation for the total number of current concurrent sessions specifically includes: The number of high-risk abandonment sessions, the distribution of session numbers at each complexity level, the total number of current concurrent sessions, the context recovery time, and the switching frequency are used as adjustment factors. Each adjustment factor is normalized to eliminate dimensional differences. Weight coefficients are assigned to each normalized adjustment factor, determined based on the correlation strength between each adjustment factor and the patient waiting interval. A weighted summation is performed on each normalized adjustment factor and its corresponding weight coefficient to obtain a comprehensive load score. The recommended concurrent session number is determined by querying a preset mapping table between load score intervals and concurrent session limits based on the comprehensive load score.

9. The method according to claim 3, characterized in that, Semantic parsing of the latest reply content, specifically including: The latest response content is segmented and keywords are extracted. The extracted keywords are matched with a preset medical substantive response terminology database, which includes diagnostic terminology, treatment terminology, and examination items. The keyword hit rate is calculated based on the matching results. When the keyword hit rate reaches a preset matching threshold, an intent recognition algorithm is used to analyze the sentence structure of the latest response content to determine whether it contains guiding statements. The substantive response containing guiding statements is determined based on the keyword hit rate and the latest response content.

10. The method according to claim 5, characterized in that, Locking doctor resources to the corresponding high-complexity session within the continuous response time slot also includes: If, within the continuous response time slot, the locked high-complexity session does not receive a substantial response within the preset time slot period and the session status does not change to closed, the continuous response time slot is extended to the next available time slot period and the locking of the doctor resource is maintained; at the same time, the low-complexity sessions waiting to be allocated by the newly accessed service bus scheduling platform during the locking period are redirected to other available doctor resources.

Images