Dynamic user portrait generation and personalized recommendation method oriented to diagnosis accompanying service

Abstract

The invention discloses a diagnosis accompanying service-oriented dynamic user portrait generation and personalized recommendation method, particularly relates to the technical field of medical portrait personalized recommendation, and is used for solving the problem of insufficient user demand dynamic adaptability caused by recommendation homogenization and portrait update stiffness due to excessive dependence on historical behavior data in an existing recommendation system. A dynamically evolved user portrait is constructed by fusing time sequence characteristics of a user behavior sequence, time-space fluctuation characteristics of physiological indexes and spatial semantics of a doctor seeing environment, and the limitation that a traditional recommendation system is disjointed from medical logic in a medical scene is broken through through closed-loop optimization of a disease-service mapping rule and real-time feedback, so that the medical scene recommendation system is more accurate. Co-evolution of demand perception and resource scheduling is realized; the recommendation strategy based on dynamic weight fusion and diversity threshold screening gives consideration to accuracy of service matching and exploratory performance of potential requirements, and recommendation reliability and service response efficiency in a complex medical scene are effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of personalized recommendation for medical images, and more specifically, to a method for generating dynamic user portraits and personalized recommendations for companion diagnosis services. Background Art

[0002] The popularization of intelligent medical services has significantly increased the dynamics and diversity of user needs in the companion diagnosis scenario. In the existing technology, a recommendation system based on generative artificial intelligence constructs a dynamic user portrait by integrating multi-dimensional features such as user behavior data and health monitoring information to provide personalized companion diagnosis services, and usually relies on historical interaction data to continuously optimize the recommendation strategy to achieve an efficient match between medical resources and patient needs.

[0003] However, in the long-term service process, the existing generative recommendation methods are prone to cause the recommendation results to tend to be homogenized due to excessive reliance on historical behavior feedback, resulting in the update of the user portrait falling into local optimization. This phenomenon makes it difficult for the system to actively explore potential needs. Especially when the patient's condition or the companion diagnosis environment changes, the recommendation strategy lacks adaptability, thus affecting the service accuracy and the sustainable improvement of the user experience. Summary of the Invention

[0004] In order to overcome the above-mentioned defects of the existing technology, an embodiment of the present invention provides a method for generating dynamic user portraits and personalized recommendations for companion diagnosis services to solve the problems proposed in the above background art.

[0005] To achieve the above object, the present invention provides the following technical solutions: A method for generating dynamic user portraits and personalized recommendations for companion diagnosis services, comprising the following steps: S1. Real-time obtain the user behavior data, health monitoring data and medical environment data of the user in the companion diagnosis service scenario; S2. Perform discrete event encoding on the interaction operation type and timestamp in the user behavior data to generate a discrete event sequence; S3. Perform spatio-temporal association on the physiological index fluctuation value in the health monitoring data and the spatial position information in the medical environment data to generate a spatio-temporal association matrix; S4. Based on a preset rule set, extract features from the discrete event sequence and the spatio-temporal association matrix to generate an initial portrait containing user demand priority labels; S5. Replace and correct the conflicting labels in the initial portrait according to the preset disease-companion diagnosis service mapping rules to generate a dynamic user portrait; S6. Retrieve candidate companion diagnosis solutions based on the user demand priority labels in the dynamic user portrait, extract the service type identifiers of the candidate companion diagnosis solutions and calculate the frequencies; S7. Sort the candidate escort plans according to the urgency of the frequency calculation results and the user demand priority tags, and generate a recommended queue, and then screen the target escort plan.

[0006] In a preferred embodiment, the user behavior data, health monitoring data, and medical environment data in the escort service scenario are obtained in real time, including: Obtain user behavior data, including click operations, swipe operations, and voice command operations in the escort service interface, and record each interaction operation type and the corresponding timestamp; Obtain health monitoring data, including at least one physiological index such as heart rate, blood pressure, and blood oxygen saturation and its fluctuation value collected in real time by a medical sensor; Obtain medical environment data, including the spatial location information of the hospital department where the user is located collected by a positioning module and the medical treatment type identifier of the current medical department obtained through the hospital information system interface.

[0007] In a preferred embodiment, discrete event encoding is performed on the interaction operation type and timestamp in the user behavior data to generate a discrete event sequence, including: Map the interaction operation type in the user behavior data to a preset discrete operation code, and the discrete operation code includes a click operation code, a swipe operation code, and a voice operation code; Divide the timestamp in the user behavior data into time slot numbers according to a preset time window, and the time slot numbers are generated at fixed time intervals based on the system clock; Combine the discrete operation code and the corresponding time slot number in the order of the timestamp to generate a discrete event sequence.

[0008] In a preferred embodiment, the physiological index fluctuation value in the health monitoring data is spatio-temporally associated with the spatial location information in the medical environment data to generate a spatio-temporal association matrix, including: Align the physiological index fluctuation value in the health monitoring data with the spatial location information in the medical environment data in time through a preset time window; Based on a preset spatio-temporal association rule, perform association calculation on the physiological index fluctuation value and the corresponding spatial location information in the same time slot to generate a spatio-temporal association vector; Arrange the spatio-temporal association vectors corresponding to each time slot in a matrix structure in time order to generate a spatio-temporal association matrix. The row dimension of the matrix represents the time slot number, and the column dimension represents the combined characteristics of the physiological index type and the spatial location coordinates.

[0009] In a preferred embodiment, feature extraction is performed on the discrete event sequence and the spatio-temporal association matrix based on a preset rule set to generate an initial portrait including user demand priority tags, including: Mine frequent operation sequences from discrete event sequences based on preset behavior pattern recognition rules to generate behavior feature vectors; Perform threshold judgment on the physiological index fluctuation values in the spatio-temporal correlation matrix based on preset abnormal fluctuation detection rules to generate physiological environment feature vectors; Concatenate the behavior feature vectors and the physiological environment feature vectors according to the feature dimensions to generate an initial portrait, and the data structure of the initial portrait is a multi-dimensional feature array; Perform label mapping on the multi-dimensional feature array in the initial portrait based on a preset medical priority table to generate an initial portrait containing user demand priority labels.

[0010] In a preferred embodiment, the behavior pattern recognition rules include a preset minimum support threshold and a time window sliding step size, the abnormal fluctuation detection rules include a preset fluctuation threshold range and an adjacent time slot difference threshold, and the medical priority table stores the corresponding relationship between feature types and priority labels.

[0011] In a preferred embodiment, replace and correct the conflicting labels in the initial portrait according to the preset disease-escort service mapping rules to generate a dynamic user portrait, including: Traverse the user demand priority labels in the initial portrait based on the preset disease-escort service mapping rule tree to detect the set of labels conflicting with the disease type nodes; Match the similarity of each label in the set of conflicting labels with the leaf nodes in the disease-escort service mapping rule tree, and screen out the labels with a confidence level lower than the preset conflict threshold as the labels to be corrected; Replace the labels to be corrected one by one according to the escort service type labels corresponding to the leaf nodes in the disease-escort service mapping rule tree to generate a set of corrected priority labels; Based on the label confidence weights in the set of corrected priority labels and the department diagnosis and treatment type identifiers in the real-time medical environment data, dynamically adjust the label weights to generate a dynamic user portrait, and the data structure of the dynamic user portrait is a set of labels with weight values.

[0012] In a preferred embodiment, retrieve candidate escort plans based on the user demand priority labels in the dynamic user portrait, extract the service type identifiers of the candidate escort plans and perform frequency calculation, including: Based on the user demand priority labels with weight values in the dynamic user portrait, retrieve candidate escort plans in the escort service resource library in descending order of label weights to generate an initial candidate set; Filter the initial candidate set by geographical fence and screen by department matching according to the spatial location coordinates and department diagnosis and treatment type identifiers in the real-time medical environment data to generate an optimized candidate set; Extract the service type identifiers in the companion diagnosis plan description text from the optimized candidate set. The service type identifiers are subjected to keyword matching and standardized mapping through a predefined companion diagnosis service classification dictionary; Perform sliding window frequency statistics on the service type identifiers based on a preset time window, calculate the sum of the occurrences of each service type in the recent several time windows, and generate a frequency calculation result.

[0013] In a preferred embodiment, sort the candidate companion diagnosis plans according to the frequency calculation result and the urgency of the user demand priority label, and generate a recommended queue and then screen the target companion diagnosis plan, including: Based on the frequency calculation result in the service type-frequency mapping table and the urgency value of the user demand priority label, calculate the dynamic sorting weight of the candidate companion diagnosis plan; Arrange the candidate companion diagnosis plans in descending order according to the dynamic sorting weight to generate a recommended queue; Perform double screening on the recommended queue based on a preset emergency threshold and diversity threshold. The screening conditions include that the dynamic sorting weight is greater than or equal to the emergency threshold and the service type identifier does not appear repeatedly within the preset time window; Mark the candidate companion diagnosis plans that simultaneously meet the emergency threshold and the diversity threshold as the target companion diagnosis plans to generate a target companion diagnosis plan list.

[0014] In a preferred embodiment, the dynamic sorting weight is the weighted sum of the frequency and the urgency. Each plan record in the recommended queue includes the dynamic sorting weight, the service type identifier, and the real-time position coordinates. The target companion diagnosis plan list is grouped and stored according to the department diagnosis and treatment type identifier.

[0015] Compared with the prior art, the present invention has the following beneficial effects: 1. Through the dynamic fusion of multi-source heterogeneous data and the collaborative optimization of medical knowledge constraints, the scene perception ability and personalized adaptation accuracy of companion diagnosis service recommendations are significantly improved; based on the discrete event encoding of user behavior data and the spatio-temporal correlation matrix of physiological environment information, a dynamic behavior sequence with temporal consistency and cross-dimensional feature expression is constructed, effectively solving the problem of insufficient scene perception caused by data fragmentation in traditional portrait generation; through the preset medical rules, conflict detection and correction are performed on the user demand labels to ensure the medical rationality and dynamic adaptability of the portrait labels. Especially when the condition suddenly changes or the companion diagnosis environment changes, it can quickly capture the potential needs of users and update the recommendation strategy in real time, avoiding recommendation solidification caused by historical data dependence; 2. The multi-dimensional candidate solution sorting and dual-threshold screening mechanism based on dynamic weights achieves a balance between precise recommendation and diversity exploration; through the dynamic integration of frequency statistics and urgency weights, combined with the grouped optimization of department diagnosis and treatment types, the recommendation results not only meet the timeliness requirements of the medical scenario but also actively explore the potential service needs of users; the closed-loop iterative design of the spatio-temporal correlation matrix and the disease-service mapping rules further enhances the system's adaptability to complex medical scenarios, ensures the efficient matching of escort resources and the continuous optimization of the user experience, providing reliable technical support for the personalized recommendation of intelligent medical services. Brief Description of the Drawings

[0016] Figure 1 It is a flowchart of the method for generating a dynamic user portrait and personalized recommendation for escort services according to the present invention. Detailed Embodiment

[0017] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0018] Embodiment: Figure 1 The method for generating a dynamic user portrait and personalized recommendation for escort services according to the present invention is given, which includes the following steps: S1. Real-time obtain the user behavior data, health monitoring data, and medical environment data of the user in the escort service scenario; S2. Discretely encode the interaction operation types and timestamps in the user behavior data to generate a discrete event sequence; S3. Spatially and temporally correlate the physiological index fluctuation values in the health monitoring data with the spatial position information in the medical environment data to generate a spatio-temporal correlation matrix; S4. Extract features from the discrete event sequence and the spatio-temporal correlation matrix based on a preset rule set to generate an initial portrait containing user demand priority tags; S5. Replace and correct the conflicting tags in the initial portrait according to the preset disease-escort service mapping rules to generate a dynamic user portrait; S6. Retrieve candidate escort solutions based on the user demand priority tags in the dynamic user portrait, extract the service type identifiers of the candidate escort solutions, and perform frequency calculation; S7. Sort the candidate escort solutions according to the frequency calculation results and the urgency of the user demand priority tags, generate a recommendation queue, and then screen the target escort solutions. S1. Real-time obtain the user behavior data, health monitoring data and medical environment data of the user in the accompaniment medical service scenario, including: The acquisition of user behavior data is realized through the interaction operations in the accompaniment medical service interface; the user behavior data includes the click operation, swipe operation and voice command operation triggered by the user in the accompaniment medical service interface. Among them, the click operation is captured and recorded as a click event through the front-end buried point code, the swipe operation is captured and recorded as a swipe event through the touch screen swipe trajectory recognition algorithm, the voice command operation is converted into a text command through the voice recognition interface and recorded as a voice event, and each operation type and the corresponding timestamp are generated and associated with storage through the system clock.

[0019] The acquisition of health monitoring data is realized by the real-time collection of physiological indicators through medical sensors; the health monitoring data includes at least one physiological indicator such as heart rate, blood pressure and blood oxygen saturation collected by wearable medical sensors. Among them, the heart rate is calculated by measuring the pulse waveform through a photoelectric sensor, the blood pressure is calculated by measuring the arterial pressure fluctuation value through a pressure sensor, and the blood oxygen saturation is calculated by the difference in the absorption rates of red light and infrared light. The fluctuation value of each physiological indicator is generated by calculating the standard deviation or moving average of the measured values within adjacent time windows.

[0020] The acquisition of medical environment data is realized through the positioning module and the hospital information system interface; the medical environment data includes the spatial location information of the hospital department where the user is located and the diagnosis and treatment type identifier of the current medical department. Among them, the spatial location information is obtained through the GPS signal of the positioning module or the Wi-Fi fingerprint positioning technology and converted into the hospital department coordinates, and the diagnosis and treatment type identifier is generated by calling the predefined department coding table through the hospital information system interface to match the current medical department name. The types of the hospital information system interface include the HL7 protocol interface or the FHIR standard interface, and the identity authentication and data permission verification are carried out through the OAuth 2.0 protocol when the interface is called.

[0021] S2. Discretely encode the interaction operation type and timestamp in the user behavior data to generate a discrete event sequence, including: The interaction operation types in the user behavior data include click operation, swipe operation and voice command operation; the click operation is mapped to a preset discrete operation code, and the click operation code is a numerical code. For example, the click operation code is 1, the swipe operation is mapped to a preset discrete operation code, and the swipe operation code is a character code. For example, the swipe operation code is S01, the voice command operation is mapped to a preset discrete operation code, and the voice operation code is a mixed code. For example, the voice operation code is V-001. The mapping rule of the discrete operation code is realized through a preset coding look-up table. The coding look-up table stores the corresponding relationship between the operation type and the code. The structure of the coding look-up table is in the form of key-value pairs, where the key is the operation type name and the value is the corresponding code.

[0022] The discretization of the timestamp is achieved based on the division of a preset time window; the preset time window is a time interval with a fixed time length. For example, the preset time window is 5 minutes. The time slot number is generated by dividing the timestamp by the preset time window length and taking the integer operation. The generation of the time slot number is based on the timer function of the system clock, and the timing accuracy of the system clock is at the millisecond level. The correspondence between the timestamp and the time slot number is stored in a hash table. The key of the hash table is the integer value of the timestamp, and the value is the corresponding time slot number. Hash conflicts are resolved by linear probing.

[0023] The generation of the discrete event sequence is achieved through the combination of discrete operation encoding and the time slot number; the discrete operation encoding and the corresponding time slot number are arranged and combined in the time order of the timestamp to generate a discrete event sequence. Each event item in the discrete event sequence includes a discrete operation encoding field and a time slot number field. The discrete operation encoding field stores the encoded value after mapping the operation type, and the time slot number field stores the time slot number corresponding to the timestamp. The data format of the event item is a JSON string, including key-value pairs of the discrete operation encoding and the time slot number.

[0024] S3. Spatially and temporally correlate the physiological index fluctuation value in the health monitoring data with the spatial location information in the medical treatment environment data to generate a spatio-temporal correlation matrix, including: The physiological index fluctuation value in the health monitoring data and the spatial location information in the medical treatment environment data are time-aligned through a preset time window; the length of the time window is the same as the preset time window in S2. The time slot number is divided based on a fixed time interval generated by the system clock. The physiological index fluctuation value and the spatial location information are matched and aligned according to the time slot number. The physiological index data and the spatial location coordinates within the same time slot are correlated by matching the timestamp range.

[0025] The physiological index fluctuation value and the spatial location information within the same time slot are correlated and calculated through preset spatio-temporal correlation rules; the spatio-temporal correlation rules include combining the numerical value of the physiological index fluctuation value with the latitude and longitude information of the spatial location coordinates to generate a spatio-temporal correlation vector. The dimensions of the spatio-temporal correlation vector include the physiological index type, the latitude value, and the longitude value of the spatial location coordinates. For example, when the heart rate is the physiological index type, the latitude is 39.9042, and the longitude is 116.4074, the spatio-temporal correlation vector is [heart rate, 39.9042, 116.4074].

[0026] The spatio-temporal correlation vectors corresponding to each time slot are arranged in a matrix structure in chronological order; the row dimension of the spatio-temporal correlation matrix represents the time slot number, and the column dimension represents the combined characteristics of the physiological index type and the spatial position coordinates. The combined characteristics are composed of the concatenation of the physiological index type name and the longitude and latitude of the spatial position coordinates. For example, the combined characteristics of heart rate and 39.9042N, 116.4074E are "heart rate_39.9042_116.4074". Each element of the spatio-temporal correlation matrix stores the correlation vector value corresponding to the time slot and the combined characteristics.

[0027] S4. Feature extraction is performed on the discrete event sequence and the spatio-temporal correlation matrix based on a preset rule set to generate an initial portrait containing user demand priority tags, including: The behavior pattern recognition rule is used to extract frequent operation sequences from the discrete event sequence and generate behavior feature vectors; the minimum support threshold in the behavior pattern recognition rule represents the minimum number of times a certain operation sequence appears within a time window. For example, the minimum support threshold is set to 5 times, and the time window sliding step represents the interval between adjacent time windows. For example, the time window sliding step is set to 1 minute. The mining of frequent operation sequences is achieved by traversing the operation types within each time window of the discrete event sequence and counting the operation sequence combinations that meet the minimum support threshold. For example, when the click operation appears 5 times and the swipe operation appears 3 times within the time window, this operation sequence is recorded as a frequent sequence. Each dimension of the behavior feature vector corresponds to the occurrence frequency of a frequent operation sequence.

[0028] The abnormal fluctuation detection rule is used to identify abnormal fluctuations of physiological indicators from the spatio-temporal correlation matrix and generate physiological environment feature vectors; the fluctuation threshold range in the abnormal fluctuation detection rule represents the allowable fluctuation interval of physiological indicators in the normal state. For example, the fluctuation threshold range of heart rate is set to 60 to 100 beats per minute, and the fluctuation threshold range of blood pressure is set to 90 to 140 mmHg for systolic pressure. The adjacent time slot difference threshold represents the maximum allowable difference in the fluctuation of physiological indicators between adjacent time slots. For example, the adjacent time slot difference threshold of heart rate is set to 30 beats per minute. Abnormal fluctuation detection is performed by comparing the physiological indicator fluctuation values of each time slot in the spatio-temporal correlation matrix with the fluctuation threshold range. If it exceeds the threshold range or the adjacent time slot difference exceeds the difference threshold, it is marked as abnormal. Each dimension of the physiological environment feature vector corresponds to the number of abnormal fluctuations of a physiological indicator within the time window.

[0029] The concatenation of the behavioral feature vector and the physiological environment feature vector is achieved through the sequential combination of feature dimensions; the dimensions of the behavioral feature vector include the types of frequent operation sequences and their occurrence frequencies, such as the frequency of click operation sequences and the frequency of slide operation sequences, and the dimensions of the physiological environment feature vector include the types of physiological indicators and the number of abnormal fluctuations, such as the number of abnormal heart rates and the number of abnormal blood pressures. When concatenating, each dimension of the behavioral feature vector is arranged in order, followed by each dimension of the physiological environment feature vector, generating a multi-dimensional feature array of the initial portrait. The data structure of the multi-dimensional feature array is a numeric array, and each element stores the feature value of the corresponding dimension.

[0030] The medical priority table is used to map the multi-dimensional feature array to user demand priority labels; the medical priority table is a two-dimensional table structure, with rows representing feature types and columns representing priority labels. The table content stores the correspondence between feature types and priority labels. For example, when the feature type is "number of abnormal heart rates ≥ 3 times", the corresponding priority label is "urgent escort demand", and when the feature type is "click operation frequency ≥ 10 times", the corresponding priority label is "high-frequency operation demand". The label mapping process generates the final set of user demand priority labels by traversing each feature value in the multi-dimensional feature array, matching the eligible feature types according to the medical priority table, and accumulating the matching priority labels according to the preset weights. For example, when the number of abnormal heart rates is 4 times, the "urgent escort demand" label is matched, and when the click operation frequency is 12 times, the "high-frequency operation demand" label is matched. The final label set includes "urgent escort demand" and "high-frequency operation demand".

[0031] The data storage format of the initial portrait is a structured data object; the data structure of the initial portrait includes a multi-dimensional feature array field and a user demand priority label set field. The multi-dimensional feature array field stores the concatenated feature values, and the user demand priority label set field stores the generated label list. The metadata of the data object includes the time window number, user identifier, and generation timestamp. The construction of the medical priority table is based on the medical knowledge base and historical service data; the medical knowledge base stores the association rules between disease types and escort service types, such as that coronary heart disease patients need to be preferentially matched with cardiology escort staff, and the historical service data stores the matching records of user historical behaviors and escort service results, such as the number of actual successful appointments after the user clicks on the cardiology escort staff's profile. The medical priority table generates the priority mapping relationship by statistically analyzing the co-occurrence frequency of feature types and successful matching labels in the historical service data and combining the rule weights of the medical knowledge base. For example, the co-occurrence frequency of "number of abnormal heart rates ≥ 3 times" and "urgent escort demand" is 90%, and the weight of this rule in the medical knowledge base is 0.9, then the priority mapping weight is set to 0.9.

[0032] The generation process of user demand priority tags is achieved through weighted summation; each matched priority tag is weighted and accumulated according to the weights in the medical priority table. The weight value represents the influence degree of the tag on the final recommendation result. For example, the weight of "urgent escort demand" is 0.9, and the weight of "high-frequency operation demand" is 0.7. When the total score after weighted accumulation exceeds the preset priority threshold, the tag is added to the final tag set. For example, if the priority threshold is set to 0.8, the score of "urgent escort demand" is 0.9, which is higher than the threshold and is retained, while the score of "high-frequency operation demand" is 0.7, which is lower than the threshold and is filtered out.

[0033] The update mechanism of the initial portrait is dynamically adjusted based on real-time feedback data; the actual selection behavior of the user for the recommended escort plan is used as feedback data. For example, if the user selects an escort corresponding to the "urgent escort demand" tag, the weights of the corresponding feature types in the medical priority table are updated according to the feedback data. For example, the weights of "heart rate abnormality times ≥ 3 times" and "urgent escort demand" are adjusted from 0.9 to 0.95. The multi-dimensional feature array of the initial portrait recalculates the tag mapping result according to the new weight values to generate an updated set of user demand priority tags.

[0034] S5. Replace and correct the conflicting tags in the initial portrait according to the preset disease-escort service mapping rules to generate a dynamic user portrait, including: The disease-escort service mapping rule is a disease-escort service mapping rule tree, which is a hierarchical tree structure. The root node is the disease type, the sub-node is the escort service type, and the leaf node is the specific escort service tag; the disease type nodes include disease classifications such as coronary heart disease, diabetes, and postoperative care, etc. The escort service type nodes include professional nurse escort, psychological counseling escort, emergency transfer escort, etc. The leaf node tags store the mapping relationship between the escort service type and the disease type. For example, coronary heart disease is mapped to professional nurse escort, and postoperative care is mapped to emergency transfer escort. The construction of the rule tree is based on the disease diagnosis and treatment guidelines and historical escort service records in the medical knowledge base.

[0035] The set of user demand priority tags in the initial portrait is detected for conflicts by traversing the disease type nodes of the disease-escort service mapping rule tree; the conflict detection process includes matching the user demand priority tag with the name of the disease type node in the rule tree. If the disease type corresponding to the tag is inconsistent with the diagnosis and treatment type identifier of the current visiting department, it is determined as a conflicting tag. For example, if the current visiting department is the Department of Cardiology and the user demand priority tag contains "diabetes care demand", it is marked as a conflicting tag set. Each tag in the conflicting tag set stores the original tag name and the confidence weight.

[0036] Each tag in the conflict tag set is matched for similarity with the escort service tag of the leaf node of the rule tree; the similarity matching is achieved by calculating the text similarity of the tag names. For example, the edit distance algorithm is used to calculate the similarity between the conflict tag "Diabetes care needs" and the leaf node tag "Diabetes professional escort". If the similarity is lower than the preset conflict threshold, for example, the conflict threshold is set to 0.7, it is marked as a tag to be corrected. The screening results of the tags to be corrected are stored as a list, including the original tag name, similarity score, and the corresponding leaf node tag.

[0037] The tags to be corrected are replaced one by one according to the escort service tags of the leaf nodes of the rule tree; the replacement rule is to replace the original name of the tag to be corrected with the name of the escort service tag of the corresponding leaf node. For example, the conflict tag "Diabetes care needs" is replaced by "Diabetes professional escort". The confidence weight of the original tag is inherited by the priority tag set after replacement. For example, the original tag confidence is 0.8, and the tag confidence remains 0.8 after replacement. The corrected priority tag set stores the replaced tag name and weight value.

[0038] The corrected priority tag set dynamically adjusts the weight according to the department diagnosis and treatment type identifier in the real-time medical environment data; the weight adjustment rule is to weight the tag confidence based on the importance coefficient of the department diagnosis and treatment type. For example, the importance coefficient of the cardiology department is 1.2, and the importance coefficient of the psychology department is 0.9. The dynamic adjustment formula is adjusted weight = original weight × importance coefficient. For example, the original weight of the tag "Professional nurse escort" is 0.8, and when the department is the cardiology department, the adjusted weight is 0.8 × 1.2 = 0.96. The data structure of the dynamic user portrait is a tag set with weight values. For example, {"labels": [{"name": "Professional nurse escort", "weight": 0.96}, {"name": "Emergency transfer escort", "weight": 0.75}]}.

[0039] The generation process of the dynamic user portrait is achieved through tag weight sorting and threshold screening; the tag weight sorting arranges the tags in descending order according to the adjusted weight value. For example, 0.96 is ranked before 0.75. When the preset priority threshold is 0.8, only the tags with weight ≥ 0.8 are retained. For example, the tag "Professional nurse escort" with a weight of 0.96 is retained, and the tag "Emergency transfer escort" with a weight of 0.75 is filtered. The final dynamic user portrait only contains high-weight tags that meet the threshold. For example, {"labels": [{"name": "Professional nurse escort", "weight": 0.96}]}.

[0040] The update mechanism of the disease-accompanying service mapping rule tree is based on user feedback data and the iteration of the medical knowledge base; the user feedback data includes the actual selection behavior of users for accompanying service tags. For example, if a user selects "professional nurse accompanying service" instead of the system-recommended "psychological counseling accompanying service", the mapping weight of the corresponding leaf node in the rule tree is adjusted according to the number of feedbacks. For example, the weight of "professional nurse accompanying service" is increased by 0.05. When the medical knowledge base is iterated, new disease types or accompanying service types are implemented through dynamic insertion of tree nodes. For example, a new node of "oncology rehabilitation accompanying service" is added and mounted under the root node of "postoperative care".

[0041] The storage and transmission format of the dynamic user profile is a standardized JSON object; the JSON object contains the label name, weight value, generation timestamp, and associated department information for medical treatment. For example, {"labels":[{"name":"professional nurse accompanying service","weight":0.96}],"metadata":{"department":"Cardiology","timestamp":"2023-10-01 14:30:00"}}. The naming and type definition of data fields are declared in advance in the system data dictionary to ensure the consistency of cross-module data parsing.

[0042] The conflict threshold for label similarity matching is jointly set through historical data statistics and medical expert experience; historical data statistics include analyzing the matching records of past conflicting labels and correct accompanying service labels. For example, statistics show that the mis-matching rate of labels with similarity lower than 0.7 exceeds 30%. Medical experts suggest setting the threshold to 0.7 based on clinical experience. The dynamic adjustment of the conflict threshold is achieved through periodic evaluation of the mis-matching rate. For example, the threshold is recalculated every quarter. If the mis-matching rate rises, the threshold is lowered; otherwise, it is raised.

[0043] The department importance coefficient in the weight adjustment rule is derived from the hospital department priority configuration table; the hospital department priority configuration table is formulated by the hospital management department according to the department's patient reception volume, proportion of critical and severe cases, and resource allocation situation. For example, because the Cardiology department has a large patient reception volume and many emergency cases, the importance coefficient is set to 1.2. Because the Psychology department mainly provides non-emergency services, the importance coefficient is set to 0.9. The configuration table is updated synchronously at regular intervals through the hospital information system interface to ensure that the coefficient value is consistent with the actual operation requirements.

[0044] S6. Retrieve candidate accompanying service plans based on the user demand priority labels in the dynamic user profile, extract the service type identifiers of the candidate accompanying service plans and calculate the frequencies, including: The user demand priority tags with weight values in the dynamic user profile are used to retrieve candidate escort plans in the escort service repository and generate an initial candidate set; the generation process of the initial candidate set is based on the descending order of tag weight values, and tags with higher weight values are retrieved and matched with escort plans first. For example, the tag "Professional Nurse Escort" with a weight of 0.96 retrieves an escort with a service type of "Professional Nursing" first, followed by the tag "Psychological Counseling" with a weight of 0.75. Each escort plan record in the escort service repository contains a service type description, geographical location coordinates, and the type of department it belongs to. The initial candidate set is generated by traversing the repository records and matching the tag names. For example, when matching the "Professional Nurse Escort" tag, all plans with service type descriptions containing keywords such as "Professional Nursing" and "Nurse Escort" are retrieved.

[0045] The spatial location coordinates and department diagnosis and treatment type identifiers in the real-time medical environment data are used to perform geographical fence filtering and department matching screening on the initial candidate set; geographical fence filtering is achieved by calculating the straight-line distance between the geographical location coordinates of the candidate escort plan and the user's current location coordinates, and only the plans with a distance less than the preset fence radius are retained. For example, the fence radius is set to 500 meters. Department matching screening is achieved by comparing the type of department the candidate escort plan belongs to with the current medical department diagnosis and treatment type identifier, and only the plans with exactly the same department type are retained. For example, when the current department is Cardiology, escorts belonging to the Cardiology department are screened out, and the optimized candidate set stores the filtered escort plan list.

[0046] The escort plan description text in the optimized candidate set is used to extract service type identifiers and perform standardized mapping through a predefined escort service classification dictionary; the escort service classification dictionary is in a key-value pair structure, where the key is the standardized service type name and the value is a set of description text keywords. For example, the keywords corresponding to the key "Professional Nurse Escort" include "Professional Nursing", "Nurse", "Escort", etc. The extraction of service type identifiers is achieved by traversing the description text and matching keywords. For example, the description text "Provide professional nursing services and be accompanied by a senior nurse during the visit" matches the "Professional Nurse Escort" identifier. Standardized mapping converts the extracted identifier into a unified classification name. For example, "Nurse Escort" is mapped to "Professional Nurse Escort".

[0047] The frequency statistics of service type identifiers are calculated using a sliding window based on a preset time window; the length of the time window is the same as the preset time window in S2. For example, the time window is 5 minutes and the sliding window step size is set to 1 minute. Frequency statistics are obtained by recording the number of occurrences of each service type identifier in the recent several time windows and accumulating them. For example, the total number of occurrences of the "Professional Nurse Escort" identifier in the recent 3 time windows is counted. The generated service type-frequency mapping table stores each service type identifier and its cumulative frequency. For example, {"Professional Nurse Escort": 15, "Psychological Counseling": 8}.

[0048] The generation rules of the service type-frequency mapping table include the sliding update of the time window and the frequency clearing mechanism; the sliding update of the time window is triggered according to the system clock. After each sliding step time interval, the frequency record of the earliest time window is removed and the frequency of the current window is added. The frequency clearing mechanism is automatically reset when the maximum number of statistical windows is exceeded. For example, the maximum number of statistical windows is set to 10. After exceeding the limit, all frequencies are reset to zero and recounted to ensure the timeliness of statistical data.

[0049] The construction and updating of the classification dictionary of accompanying medical services are based on the medical terminology database and historical service description texts; the medical terminology database includes standardized service type names and corresponding medical terms. For example, the corresponding terms for "professional nurse accompanying medical consultation" include "nursing service" and "medical accompaniment". The historical service description texts use natural language processing technology to extract high-frequency keywords. For example, the words "accompany", "nursing" and "transfer" are extracted from the descriptions of historical accompanying medical consultation plans. The update cycle of the classification dictionary is set to once a week. New keywords are added to the dictionary after manual review. For example, the keyword "postoperative accompanying medical consultation" is added and mapped to the "postoperative nursing accompanying medical consultation" logo.

[0050] The correlation between frequency statistics and service type identification is monitored and verified through a real-time data visualization interface; the data visualization interface displays the frequency change curve and time window distribution of each service type identification. For example, a line chart shows the frequency fluctuation of "professional nurse accompanying the patient" in the last hour. The verification rules include threshold alarms for sudden increases or decreases in frequency. For example, an alarm is triggered when the frequency of a single window exceeds 200% of the historical average, ensuring the rationality of statistical data and anomaly detection capabilities.

[0051] The fence radius and department matching rules in geographic fence filtering are dynamically set through the hospital management configuration table; the hospital management configuration table stores the fence radius and matching rules corresponding to different department types. For example, the fence radius of the cardiology department is 500 meters and strict department matching is required, and the fence radius of the emergency department is 1000 meters and cross-department matching is allowed. The configuration table is updated regularly and synchronously through the hospital information system interface to ensure that the filtering rules are consistent with the actual operation strategy.

[0052] The data storage format of the optimized candidate set is a structured list; each entry in the structured list contains the accompanying treatment plan identifier, service type identifier, geographic location coordinates, department type and frequency statistics. The naming and type definition of the data field are pre-declared in the system metadata table to ensure the consistency of cross-module data interaction.

[0053] S7. Sort candidate accompanying treatment plans according to the frequency calculation results and the urgency of the user's demand priority label, generate a recommendation queue, and then select the target accompanying treatment plan, including: The frequency calculation result in the service type - frequency mapping table and the urgency value of the user requirement priority label are used to calculate the dynamic sorting weight of the candidate escort plans; the calculation rule of the dynamic sorting weight is the weighted sum of the frequency calculation result and the urgency value. For example, the frequency weight coefficient is set to 0.6, and the urgency weight coefficient is set to 0.4. The dynamic sorting weight = frequency × 0.6 + urgency × 0.4. The weight coefficient is dynamically adjusted according to the influence ratio of frequency and urgency on the recommendation success rate in historical service data. For example, when the frequency influence accounts for 60%, the coefficient is set to 0.6. The calculation result of the dynamic sorting weight is reserved to two decimal places. For example, when the frequency is 15 times and the urgency is 0.9, the weight = 15 × 0.6 + 0.9 × 0.4 = 9.36.

[0054] The candidate escort plans are sorted in descending order according to the dynamic sorting weight to generate a recommendation queue; the sorting rule of the recommendation queue is to arrange the dynamic sorting weights from high to low. For example, the plan with a weight of 9.36 is ranked before the plan with a weight of 8.75. Each plan record contains the dynamic sorting weight, service type identifier, and real - time location coordinates. For example, [{"scheme_id":"S001","weight":9.36,"service_type":"Professional nurse escort","location":"39.9042,116.4074"}]. The data storage format of the recommendation queue is an ordered list, stored in descending order of the weight value.

[0055] The preset emergency threshold and diversity threshold are used for double - screening the recommendation queue; the emergency threshold represents the minimum qualified value of the plan weight. For example, the emergency threshold is set to 8.0, and only the plans with a weight ≥ 8.0 are retained. The diversity threshold represents the maximum number of repetitions of the same service type identifier within a preset time window. For example, the time window is 1 hour and the repetition threshold is set to 1 time. When screening, it is checked whether there is already a plan with the same service type identifier recommended within the current time window. If not, it is allowed to be selected. Each plan in the recommendation queue needs to meet both the weight ≥ emergency threshold and the service type not being repeated. For example, the "Professional nurse escort" plan with a weight of 9.36 is retained when it appears for the first time within the time window, and is filtered if there is already a similar plan.

[0056] Candidate escort plans that meet both the emergency threshold and the diversity threshold are marked as target escort plans and a list of target escort plans is generated; the list of target escort plans is grouped and stored according to the department diagnosis and treatment type identifier. For example, the target plans for cardiology are stored in the "cardiology" group, and the target plans for the emergency department are stored in the "emergency" group. The plans within each group are sorted in descending order of weight. The data structure of the list is a nested dictionary, with the outer key being the department diagnosis and treatment type identifier and the inner value being an array of plan records. For example, {"cardiology": [{"scheme_id": "S001", "weight": 9.36}], "emergency": []}. When storing the data, a timestamp and a version number are attached. For example, the version number is used to distinguish lists generated at different times.

[0057] The frequency weight coefficient and the emergency weight coefficient in the dynamic sorting weight are determined through historical matching record statistics; historical matching record statistics include analyzing the influence ratio of frequency and emergency on users' selection of escort plans in the past year. For example, if the statistics show that 60% of users' selections are affected by frequency and 40% are affected by emergency, the weight coefficients are set to 0.6 and 0.4. The update period of the weight coefficients is once a month, and the coefficient values are adjusted according to the latest statistical results. For example, if the next month's statistics show that the frequency influence rises to 65%, the coefficients are adjusted to 0.65 and 0.35.

[0058] The setting of the emergency threshold and the diversity threshold is based on the priority strategy of the medical scenario and user experience feedback; the emergency threshold is dynamically adjusted according to the critical level of the medical department. For example, the threshold for the emergency department is set to 7.0, for cardiology it is set to 8.0, and for psychiatry it is set to 6.0. The diversity threshold is determined by analyzing the negative feedback rate of users for repeated recommended plans. For example, when repeated recommendations lead to 10% of users canceling the service, the threshold is set to 1 time / hour. The data is stored in the hospital management configuration table and is updated synchronously through a scheduled task.

[0059] The department grouping storage rule of the target escort plan list is based on the resource allocation strategy of the hospital departments; the resource allocation strategy stipulates that the escort plans of different departments need to be managed independently to ensure rapid response. For example, the cardiology plans are stored independently to facilitate quick matching of specialist escort staff. When grouping and storing, the storage partition is allocated according to the hash value of the department diagnosis and treatment type identifier. For example, the hash function converts the department name into an integer partition number to ensure uniform data distribution and efficient query.

[0060] The version number and timestamp of the target companion diagnosis plan list are used for data traceability and consistency verification; the version number is an incrementing integer value, and the version number is incremented by 1 each time the list is updated. The timestamp records the specific time when the list is generated. For example, the version number 202410011200 represents the first version of the list generated at 12:00 on October 1, 2024. When performing data verification, data integrity is ensured by comparing the version number and timestamp. If the version number is discontinuous or the timestamp is abnormal, the alarm mechanism is triggered.

[0061] The duplicate detection of service type identifiers is implemented through cached records within a sliding time window; the cached records store the service type identifiers recommended in the recent period and their recommended times. For example, the cached records retain data within 1 hour. Each time a new plan is screened, it is checked whether there are records with the same service type identifier and a time difference of less than 1 hour in the cache. If so, the recommendation is rejected; otherwise, the current plan identifier is added to the cache, and the cached data is automatically cleared after the time window expires.

[0062] The results of dynamic sorting weight calculation and threshold screening are monitored in real time through a data visualization interface; the visualization interface displays the weight distribution, threshold lines, and screening results of each plan in the recommendation queue. For example, a bar chart shows the number of plans with a weight ≥ 8.0, and the screening results are distinguished by different colors for approved and filtered plans. The monitoring data is refreshed every 5 seconds to ensure that the operation personnel can grasp the status of the recommendation system in real time.

[0063] The above formulas are all dimensionless and take their numerical values for calculation. The formulas are obtained by collecting a large amount of data for software simulation to obtain a formula closest to the actual situation. The preset parameters and threshold selection in the formulas are set by technicians in this field according to the actual situation.

[0064] It should be noted that the present invention can be deployed on the device itself to achieve embedded applications, or it can also run on a PC or other terminals with a user interface, so as to meet various hardware environments and usage requirements.

[0065] The above embodiments can be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the processes or functions described in the embodiments of the present application are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center by wire (such as infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that contains one or more collections of available media. The available media can be magnetic media (such as floppy disks, hard disks, magnetic tapes), optical media (such as DVDs), or semiconductor media. The semiconductor media can be a solid-state drive.

[0066] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the systems, devices, and modules described above can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated herein.

[0067] In several embodiments provided in the present application, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the modules is only a logical function division, and there can be other division methods in actual implementation. For example, multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the couplings or direct couplings or communication connections shown or discussed with each other can be through some interfaces, and the indirect couplings or communication connections of the devices or modules can be in electrical, mechanical, or other forms.

[0068] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical modules. They can be located in one place or distributed to multiple network modules. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0069] In addition, in each embodiment of the present application, each functional module can be integrated into a processing module, or each module can exist physically alone, or two or more modules can be integrated into one module.

[0070] If the above-mentioned function is implemented in the form of a software functional module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art, or a part of this 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 for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes such as USB flash drives, mobile hard disks, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical discs.

[0071] The above is only the specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed in the present application, and all should be covered by the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

[0072] Finally: The above is only the preferred embodiment of the present invention and is not used to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for generating a dynamic user profile and personalized recommendation for companion diagnosis services, characterized in that It includes the following steps: S1. Real-time obtain the user behavior data, health monitoring data, and medical environment data of the user in the accompaniment service scenario; S2. Discretely event code the interaction operation types and timestamps in the user behavior data to generate a discrete event sequence; S3. Spatially and temporally correlate the physiological index fluctuation values in the health monitoring data with the spatial location information in the medical environment data to generate a spatio-temporal correlation matrix; S4. Based on a preset rule set, extract features from the discrete event sequence and the spatio-temporal correlation matrix to generate an initial portrait containing user demand priority labels; S5. According to the preset disease-accompaniment service mapping rules, replace and correct the conflicting labels in the initial portrait to generate a dynamic user portrait; S6. Based on the user demand priority labels in the dynamic user portrait, retrieve candidate accompaniment plans, extract the service type identifiers of the candidate accompaniment plans, and calculate the frequencies; S7. Sort the candidate accompaniment plans according to the frequency calculation results and the urgency of the user demand priority labels, generate a recommended queue, and then screen the target accompaniment plan.

2. The method for generating a dynamic user profile and personalized recommendation for companion diagnosis services according to claim 1, characterized in that Real-time obtain the user behavior data, health monitoring data, and medical environment data of the user in the accompaniment service scenario, including: Obtain user behavior data, including click operations, swipe operations, and voice command operations in the accompaniment service interface, and record each interaction operation type and the corresponding timestamp; Obtain health monitoring data, including at least one physiological index and its fluctuation value among heart rate, blood pressure, and blood oxygen saturation collected in real time by medical sensors; Obtain medical environment data, including the spatial location information of the hospital department where the user is located collected by the positioning module and the diagnosis and treatment type identifier of the current diagnosis and treatment department obtained through the hospital information system interface.

3. The method for generating a dynamic user profile and personalized recommendation for companion diagnosis and treatment services according to claim 1, wherein Discretely event code the interaction operation types and timestamps in the user behavior data to generate a discrete event sequence, including: Map the interaction operation types in the user behavior data to preset discrete operation codes, and the discrete operation codes include click operation codes, swipe operation codes, and voice operation codes; Divide the timestamps in the user behavior data into time slot numbers according to a preset time window, and the time slot numbers are generated at fixed time intervals based on the system clock; Combine the discrete operation codes and the corresponding time slot numbers in timestamp order to generate a discrete event sequence.

4. The method for generating a dynamic user profile and personalized recommendation for companion diagnosis services according to claim 1, characterized in that Spatially and temporally correlate the physiological index fluctuation values in the health monitoring data with the spatial location information in the medical environment data to generate a spatio-temporal correlation matrix, including: Align the physiological index fluctuation values in the health monitoring data with the spatial location information in the medical environment data in time through a preset time window; Based on preset spatio-temporal correlation rules, perform correlation calculations on the physiological index fluctuation values and the corresponding spatial location information in the same time slot to generate spatio-temporal correlation vectors; Arrange the spatio-temporal correlation vectors corresponding to each time slot in matrix structure in time order to generate a spatio-temporal correlation matrix. The row dimension of the matrix represents the time slot number, and the column dimension represents the combined features of the physiological index type and the spatial location coordinates.

5. The dynamic user portrait generation and personalized recommendation method for companion diagnosis service according to claim 1, characterized in that Based on a preset rule set, extract features from the discrete event sequence and the spatio-temporal correlation matrix to generate an initial portrait containing user demand priority labels, including: Mine frequent operation sequences from the discrete event sequence based on the preset behavior pattern recognition rules to generate a behavior feature vector; Perform threshold judgment on the physiological index fluctuation values in the spatio-temporal correlation matrix based on the preset abnormal fluctuation detection rules to generate a physiological environment feature vector; Concatenate the behavior feature vector and the physiological environment feature vector according to the feature dimension to generate an initial portrait, and the data structure of the initial portrait is a multi-dimensional feature array; Perform label mapping on the multi-dimensional feature array in the initial portrait based on the preset medical priority table to generate an initial portrait containing user demand priority labels.

6. The method for generating a dynamic user profile and personalized recommendation for companion diagnosis and treatment services according to claim 5, characterized in that, The behavior pattern recognition rules include the preset minimum support threshold and the time window sliding step size, the abnormal fluctuation detection rules include the preset fluctuation threshold range and the adjacent time slot difference threshold, and the medical priority table stores the corresponding relationship between the feature type and the priority label.

7. The method for generating a dynamic user profile and personalized recommendation for companion diagnosis and treatment services according to claim 1, wherein Replace and correct the conflicting labels in the initial portrait according to the preset disease-accompanying service mapping rules to generate a dynamic user portrait, including: Traverse the user demand priority labels in the initial portrait based on the preset disease-accompanying service mapping rule tree to detect the set of labels conflicting with the disease type nodes; Perform similarity matching on each label in the set of conflicting labels with the leaf nodes in the disease-accompanying service mapping rule tree, and filter out the labels with a confidence level lower than the preset conflict threshold as the labels to be corrected; Perform one-to-one replacement on the labels to be corrected according to the accompanying service type labels corresponding to the leaf nodes in the disease-accompanying service mapping rule tree to generate a set of corrected priority labels; Dynamically adjust the label weights based on the label confidence weights in the set of corrected priority labels and the department diagnosis and treatment type identifiers in the real-time medical environment data to generate a dynamic user portrait, and the data structure of the dynamic user portrait is a set of labels with weight values.

8. The method for generating a dynamic user profile and personalized recommendation for companion diagnosis services according to claim 1, wherein Retrieve candidate accompanying service plans based on the user demand priority labels in the dynamic user portrait, extract the service type identifiers of the candidate accompanying service plans and perform frequency calculation, including: Based on the user demand priority labels with weight values in the dynamic user portrait, retrieve candidate accompanying service plans in the accompanying service resource library in descending order of label weight to generate an initial candidate set; Perform geographical fence filtering and department matching screening on the initial candidate set according to the spatial location coordinates and department diagnosis and treatment type identifiers in the real-time medical environment data to generate an optimized candidate set; Extract the service type identifiers from the accompanying service plan description text in the optimized candidate set, and the service type identifiers are subject to keyword matching and standardized mapping through a predefined accompanying service classification dictionary; Perform sliding window frequency statistics on the service type identifiers based on the preset time window, calculate the total number of occurrences of each service type in the recent several time windows, and generate a frequency calculation result.

9. The dynamic user portrait generation and personalized recommendation method for companion diagnosis services according to claim 1, characterized in that Sort the candidate accompanying service plans according to the frequency calculation result and the urgency of the user demand priority labels, generate a recommended queue and then screen the target accompanying service plan, including: Calculate the dynamic sorting weight of the candidate accompanying service plan based on the frequency calculation result in the service type-frequency mapping table and the urgency value of the user demand priority label; Arrange the candidate escort plans in descending order according to the dynamic sorting weight to generate a recommendation queue; Perform double screening on the recommendation queue based on the preset emergency threshold and diversity threshold. The screening conditions include that the dynamic sorting weight is greater than or equal to the emergency threshold and the service type identifier has not appeared repeatedly within the preset time window; Mark the candidate escort plans that simultaneously meet the emergency threshold and the diversity threshold as the target escort plans to generate a list of target escort plans.

10. The method for generating a dynamic user profile and personalized recommendation for companion diagnosis services according to claim 9, wherein The dynamic sorting weight is the weighted sum of the frequency and the urgency. Each plan record in the recommendation queue includes the dynamic sorting weight, the service type identifier, and the real-time position coordinates. The list of target escort plans is grouped and stored according to the department diagnosis and treatment type identifier.

Images