A health and wellness service intelligent matching method and system based on multi-source big data

Abstract

The application discloses a kind of based on multi-source big data's health and wellness service intelligent matching method and system, it is related to health and wellness service matching technical field, including following steps: S1, obtain multi-source data and carry out pre-processing;S2, carry out sliding window segmentation, determine the associated strength between state fragment, and screen target state fragment;S3, according to associated strength, establish cross-domain connection edge, and the connection edge is configured to semantic quantization, constructs supply-demand fuzzy heterogeneous graph;S4, according to cross-domain connection edge type, carries out cross-domain aggregation to neighborhood feature, and carries out multi-round fusion update;S5, to cross-domain connection edge is matched flow distribution, and carries out discretization screening;S6, obtain feedback response data, and the connection edge right of supply-demand fuzzy heterogeneous graph is updated incrementally.The application can realize the accurate docking of health and wellness user demand and available health and wellness service resources, improve health and wellness service matching efficiency.

Description

Technical Field

[0001] This invention relates to the field of health and wellness service matching technology, and in particular to a method and system for intelligent matching of health and wellness services based on multi-source big data. Background Technology

[0002] With the increasing demand for health and wellness services, resources such as elderly care, rehabilitation nursing, health management, and institutional services are gradually concentrating on online platforms. Existing platforms typically help users find relevant health and wellness service resources by having them fill in their needs, select service types, set service locations, and appointment times. This approach is effective in information display and basic appointment booking, reducing the costs associated with manual consultations and offline comparisons.

[0003] In practical applications, the demand for health and wellness services cannot be fully expressed by a single label. A user's recent health changes, lifestyle habits, service records, feedback, and the availability of service resources all influence matching results. Most existing platforms still rely primarily on service categories, keywords, distance, or manual rules for matching, failing to adequately utilize the continuous changes between multi-source data and making it difficult to promptly determine which type of service is most suitable for a user.

[0004] Meanwhile, health and wellness service resources themselves have certain limitations. Different resources vary in service capacity, availability, and capacity. Existing solutions typically use simple sorting or conditional filtering for recommendations, lacking detailed processing of the matching relationship between user needs and service resources. When multiple users are matched with multiple service resources simultaneously, problems such as duplicate recommendation results, uneven resource allocation, and neglect of service capacity can easily occur.

[0005] Furthermore, feedback information after service completion is typically only displayed as an evaluation in existing systems, rarely participating in updates to subsequent matching rules. If the actual response of a certain type of service is poor, the system struggles to promptly correct its subsequent recommendation direction; conversely, if a resource consistently provides positive feedback, it is difficult to establish a stable positive impact during the matching process, resulting in a lack of continuous improvement capabilities for the matching results.

[0006] Therefore, how to provide a method and system for intelligent matching of health and wellness services based on multi-source big data is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0007] One objective of this invention is to propose an intelligent matching method and system for health and wellness services based on multi-source big data. This invention fully utilizes multi-source big data processing, supply-demand fuzzy heterogeneous graph modeling, gated graph convolutional networks, and Bayesian confidence update technology. It details the intelligent implementation process of health and wellness state sequence construction, target state segment screening, supply-demand relationship modeling, service matching and allocation, and feedback incremental correction. It has the advantages of high matching accuracy, reasonable resource allocation, and strong service adaptability.

[0008] According to an embodiment of the present invention, a method for intelligent matching of health and wellness services based on multi-source big data includes the following steps: S1. Acquire multi-source data in the health and wellness service scenario and preprocess it to form a health and wellness status sequence; S2. The health and wellness state sequence is segmented by a sliding window, and the temporal mutual information metric is used to determine the correlation strength between state segments. Combined with the dynamic time warping algorithm, target state segments with stable demand representation are screened out. S3. Map the target state fragments to demand nodes and the health and wellness service resources to resource nodes. Establish cross-domain connection edges based on the correlation strength and use intuitive fuzzy numbers to perform semantic quantization configuration of the connection edges to construct a supply and demand fuzzy heterogeneous graph. S4. In the supply and demand fuzzy heterogeneous graph, according to the cross-domain connection edge type, a gated graph convolutional network is used to perform cross-domain aggregation of the neighborhood features of demand nodes and resource nodes respectively, and the aggregated features are fused and updated in multiple rounds through state recursion to obtain the bilateral utility score. S5. Based on the bilateral utility score and the preset service capacity constraint, a multi-objective trade-off optimization method is used to match the flow allocation of cross-domain connection edges, and the matching flow is discretized and filtered to generate service matching results. S6. Obtain the feedback response data corresponding to the service matching result, and use the Bayesian confidence update method to incrementally update the connection edge weights of the supply and demand fuzzy heterogeneous graph.

[0009] Optionally, the preprocessing specifically includes: Source labeling is performed on multi-source data in the context of health and wellness services, including user-side data, service-side data, and feedback-side data. A unified index is established based on user identifier, service identifier, and time identifier. Data fields from different sources are mapped to a unified field set, and the field values ​​are encoded and converted. Based on a unified index, duplicate records in multi-source data are merged, missing fields are filled in, and data records that exceed the preset value range are marked as abnormal records and then removed. The data that has completed field mapping is time-aligned according to the preset time granularity, and the data records under the same index are spliced ​​together to obtain the health and wellness status vector. Arrange the health and wellness status vectors in chronological order to form a health and wellness status sequence.

[0010] Optionally, S2 specifically includes: S21. Traverse the health and wellness state sequence according to the fixed window length and sliding step size, cut the continuous health and wellness state vector into multiple state segments, and record the time position corresponding to each state segment. S22. Extract the change amount of the health care state vector within the state segment to obtain the segment change sequence, and calculate the temporal mutual information value between different state segments based on the segment change sequence, and use the temporal mutual information value as the correlation strength between state segments. S23. State segments whose association strength reaches a preset association threshold are taken as candidate state segments. The dynamic time warping algorithm is used to align the segment change sequences between candidate state segments to obtain the warped distance between candidate state segments. S24. Candidate state segments are filtered based on correlation strength and regularity distance. Candidate state segments with correlation strength reaching a preset correlation threshold and regularity distance below a preset distance threshold are identified as target state segments with stable demand representation, wherein: The preset association threshold is determined based on the overall distribution of the association strength between state segments, and the preset distance threshold is determined based on the overall distribution of the regular distance between candidate state segments.

[0011] Optionally, S23 specifically includes: S231. Mark the state segments whose association strength reaches the preset association threshold as candidate state segments, extract the segment change sequences corresponding to the candidate state segments, and arrange the segment change sequences according to the time position. S232. Traverse the segment change sequence of the two candidate state segments, calculate the change difference value between different time positions, write the change difference value into the corresponding matrix position, and form a regularized cost matrix. S233. Starting from the beginning of the regularization cost matrix, the cumulative cost is calculated in three progressive directions: horizontal, vertical, and diagonal. The change difference value of the current matrix position is added to the minimum cumulative cost of the previous matrix position to form the cumulative regularization cost matrix. S234. Backtrack from the end position of the cumulative normalization cost matrix to the beginning position to determine the normalization path between the two candidate state segments, and determine the normalization distance between the candidate state segments based on the cumulative result of the change difference corresponding to the normalization path.

[0012] Optionally, S3 specifically includes: S31. Perform node-based processing on the target state fragments, converting each target state fragment into a requirement node, and maintaining a one-to-one correspondence between the requirement nodes and the target state fragments. S32. Perform node-based processing on health and wellness service resources, converting each health and wellness service resource into a resource node, and maintaining a one-to-one correspondence between resource nodes and health and wellness service resources; S33. Traverse the demand nodes and resource nodes one by one, and establish cross-domain connection edges for demand nodes and resource nodes that meet the cross-domain connection conditions based on the correlation strength and health care service resources corresponding to the target state fragment. S34. For each cross-domain connection edge, convert the association strength corresponding to the cross-domain connection edge into a first value representing the degree to which the connection is established. Based on the unmatched portion of the demand node and resource node in the cross-domain connection conditions, it is converted into a second value indicating the degree to which the connection is not established; The difference between the sum of the first and second values ​​and the upper limit of the intuitive fuzzy value is used as the third value to represent the degree of uncertainty of the connection. The first, second, and third values ​​are sequentially assigned to the corresponding cross-domain connection edges to form the intuitive fuzzy number of the cross-domain connection edge. A supply-demand fuzzy heterogeneous graph is constructed using demand nodes, resource nodes, and cross-domain connection edges configured with intuitive fuzzy numbers.

[0013] Optionally, S4 specifically includes: S41. In the supply and demand fuzzy heterogeneous graph, the data representations corresponding to the demand node and the resource node are used as node features, and the neighboring nodes of the current node are determined according to the intuitive fuzzy number corresponding to the cross-domain connection edge. S42. A gated graph convolutional network is used to perform cross-domain aggregation of neighborhood node features. The gate coefficient is calculated based on the intuitive fuzzy number corresponding to the cross-domain connection edge. The proportion of neighborhood node features participating in the aggregation is controlled according to the gate coefficient. The aggregated neighborhood node features are then passed to the corresponding nodes. S43. Using the aggregated node features as the current round state, perform state recursion on demand nodes and resource nodes, and use the current round state recursion result as the input for the next round of aggregation. Repeat cross-domain aggregation and state recursion to complete multiple rounds of fusion update. S44. Perform bidirectional association calculations on the demand nodes and resource nodes after multiple rounds of fusion updates. Based on the association results of the demand nodes and resource nodes with the demand nodes, calculate the bilateral utility scores between the corresponding nodes.

[0014] Optionally, S42 specifically includes: S421. Taking the current demand node or the current resource node as the center, determine the adjacent resource nodes or adjacent demand nodes along the cross-domain connection edge, and take the node characteristics of the adjacent nodes as the features to be aggregated. S422. Perform gating conversion on the intuitive fuzzy numbers corresponding to the cross-domain connection edges, take the values ​​corresponding to the degree of connection establishment as positive adjustment terms, and take the values ​​corresponding to the degree of connection failure and the degree of connection uncertainty as suppression adjustment terms, and calculate the gating coefficient of the cross-domain connection edges. S423. Apply the gating coefficient to the corresponding feature to be aggregated, increase the participation ratio for features to be aggregated with a higher gating coefficient, and decrease the participation ratio for features to be aggregated with a lower gating coefficient, to obtain the neighborhood features after gating. S424. In the gated graph convolutional network, the neighborhood features after gated processing are linearly transformed, and neighborhood weighted summation is performed according to the connection direction of the cross-domain connection edge. The summation result is combined with the current node features and then nonlinearly mapped to obtain the cross-domain aggregated features of the current node.

[0015] Optionally, S5 specifically includes: S51. Using the cross-domain connection edges in the supply-demand fuzzy heterogeneous graph as the allocation object, the bilateral utility scores corresponding to the cross-domain connection edges are used as the allocation basis, and the total amount of matching flows that resource nodes can handle is constrained according to the preset service capacity, wherein: The preset service capacity is the upper limit of the number of services that a resource node can undertake in the current matching round, and the matching flow is the service undertaking share allocated between the demand node and the resource node by the cross-domain connection edge. S52. Assign initial flow values ​​to cross-domain connection edges in descending order of bilateral utility scores, and accumulate the matched flows already received by each resource node during the assignment process; When the number of matching flows already handled by a resource node reaches the preset service capacity, stop adding more matching flows to the cross-domain connection edge connected to that resource node; S53. Compare the allocated matching flows of resource nodes with the preset service capacity. For matching flows that exceed the preset service capacity, reduce them from low to high according to the bilateral utility score, and allocate the reduced matching flows to cross-domain connection edges with remaining capacity. S54. Discretize and filter the adjusted matching flow, retain the cross-domain connection edges that meet the preset filtering conditions, and remove the cross-domain connection edges that are repeatedly accepted under the same demand node. Based on the retained cross-domain connection edges, form the service matching result.

[0016] Optionally, S6 specifically includes: S61. Obtain the feedback response data corresponding to the service matching result, and allocate the feedback response data to the corresponding cross-domain connection edge according to the cross-domain connection edge index in the service matching result. S62. Extract the positive and negative response values ​​from the feedback response data. Increase the feedback support value of the corresponding cross-domain connection edge based on the positive response value, and increase the feedback suppression value of the corresponding cross-domain connection edge based on the negative response value. S63. Using the current edge weight of the corresponding cross-domain connection edge as the prior value, and the feedback support value and feedback suppression value as the observation value, the Bayesian confidence update method is used to calculate the posterior confidence value. S64. Based on the posterior confidence value, the edge weights of the corresponding cross-domain connection edges are recursively corrected, and the intuitive fuzzy number of the cross-domain connection edges is recalculated according to the corrected edge weights to complete the incremental update of the supply and demand fuzzy heterogeneous graph.

[0017] According to an embodiment of the present invention, a smart matching system for health and wellness services based on multi-source big data includes: The data processing module is used to acquire multi-source data in the context of health and wellness services and preprocess it to form a health and wellness status sequence. The segment screening module is used to perform sliding window segmentation on the health and wellness state sequence, use temporal mutual information metric to determine the correlation strength between state segments, and combine dynamic time warping algorithm to screen target state segments with stable demand representation. The graph construction module is used to map target state fragments to demand nodes, and health and wellness service resources to resource nodes. It establishes cross-domain connection edges based on the correlation strength and uses intuitive fuzzy numbers to perform semantic quantization configuration of the connection edges to construct a supply and demand fuzzy heterogeneous graph. The aggregation scoring module is used in the supply and demand fuzzy heterogeneous graph to perform cross-domain aggregation of the neighborhood features of demand nodes and resource nodes according to the cross-domain connection edge type, and to obtain a two-sided utility score by performing multiple rounds of fusion and updating of the aggregated features through state recursion. The matching and allocation module is used to allocate matching flows to cross-domain connection edges based on bilateral utility scores and preset service capacity constraints, and to discretize and filter the matching flows to generate service matching results. The feedback update module is used to obtain the feedback response data corresponding to the service matching results and to incrementally update the connection edge weights of the supply and demand fuzzy heterogeneous graph using the Bayesian confidence update method.

[0018] The beneficial effects of this invention are: First, this invention forms a health and wellness status sequence through multi-source data processing, and combines sliding window, temporal mutual information and dynamic time warping to screen target status segments, so that user needs can be identified from continuously changing data, reducing matching deviations caused by relying solely on manual input or static labels.

[0019] Secondly, this invention converts target state fragments and health and wellness service resources into demand nodes and resource nodes, respectively. It constructs a supply-demand fuzzy heterogeneous graph through cross-domain connection edges and intuitive fuzzy numbers, and uses a gated graph convolutional network for cross-domain aggregation. This allows the connection between demand and resources to be quantified, improving the rationality of service matching and resource allocation.

[0020] Finally, based on the feedback response data corresponding to the service matching results, the present invention uses a Bayesian confidence update method to incrementally correct the connection edge weights, so that the matching relationship can be continuously adjusted with service feedback, reducing resource mismatch and duplicate recommendations, and improving the matching efficiency, resource utilization and service suitability of health and wellness services. Attached Figure Description

[0021] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is an overall flowchart of a smart matching method for health and wellness services based on multi-source big data proposed in this invention. Figure 2 This is a flowchart illustrating the cross-domain aggregation scoring process of an intelligent matching method for health and wellness services based on multi-source big data proposed in this invention. Figure 3 This is a module structure diagram of an intelligent matching system for health and wellness services based on multi-source big data proposed in this invention. Detailed Implementation

[0022] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0023] refer to Figures 1-2 A method for intelligent matching of health and wellness services based on multi-source big data includes the following steps: S1. Acquire multi-source data in the health and wellness service scenario and preprocess it to form a health and wellness status sequence; S2. The health and wellness state sequence is segmented by a sliding window, and the temporal mutual information metric is used to determine the correlation strength between state segments. Combined with the dynamic time warping algorithm, target state segments with stable demand representation are screened out. S3. Map the target state fragments to demand nodes and the health and wellness service resources to resource nodes. Establish cross-domain connection edges based on the correlation strength and use intuitive fuzzy numbers to perform semantic quantization configuration of the connection edges to construct a supply and demand fuzzy heterogeneous graph. S4. In the supply and demand fuzzy heterogeneous graph, according to the cross-domain connection edge type, a gated graph convolutional network is used to perform cross-domain aggregation of the neighborhood features of demand nodes and resource nodes respectively, and the aggregated features are fused and updated in multiple rounds through state recursion to obtain the bilateral utility score. S5. Based on the bilateral utility score and the preset service capacity constraint, a multi-objective trade-off optimization method is used to match the flow allocation of cross-domain connection edges, and the matching flow is discretized and filtered to generate service matching results. S6. Obtain the feedback response data corresponding to the service matching result, and use the Bayesian confidence update method to incrementally update the connection edge weights of the supply and demand fuzzy heterogeneous graph.

[0024] In this embodiment, the preprocessing specifically includes: Source labeling is performed on multi-source data in the health and wellness service scenario, where the multi-source data includes user-side data, service-side data, and feedback-side data. A unified index is established based on user identifier, service identifier, and time identifier. Data fields from different sources are mapped to a unified field set, and the field values ​​are encoded and converted. Based on a unified index, duplicate records in multi-source data are merged, missing fields are filled in, and data records that exceed the preset value range are marked as abnormal records and then removed. The data that has completed field mapping is time-aligned according to the preset time granularity, and the data records under the same index are spliced ​​together to obtain the health and wellness status vector. Arrange the health and wellness status vectors in chronological order to form a health and wellness status sequence; In this embodiment, the preset time granularity is set to 30 minutes, and data records falling into the same index and the same time granularity are spliced ​​together. Data records with a field missing rate of less than 20% are filled in using data values ​​taken from adjacent time granularities; data records with a field missing rate of more than 20% are not included in the status splicing. After the field values ​​are encoded and converted, they fall into the range of 0 to 1. The dimension of each health and wellness status vector is set to 64. For the health and wellness status vectors formed consecutively by the same user, they are arranged in chronological order to form a health and wellness status sequence. When more than three consecutive time granularities are missing, they are broken up in the health and wellness status sequence to avoid abnormal missing data affecting the screening of subsequent status segments.

[0025] In this embodiment, S2 specifically includes: S21. Traverse the health and wellness state sequence according to the fixed window length and sliding step size, cut the continuous health and wellness state vector into multiple state segments, and record the time position corresponding to each state segment. S22. Extract the change amount of the health care state vector within the state segment to obtain the segment change sequence, and calculate the temporal mutual information value between different state segments based on the segment change sequence, and use the temporal mutual information value as the correlation strength between state segments. S23. State segments whose association strength reaches a preset association threshold are taken as candidate state segments. The dynamic time warping algorithm is used to align the segment change sequences between candidate state segments to obtain the warped distance between candidate state segments. S24. Candidate state segments are filtered based on correlation strength and regularity distance. Candidate state segments with correlation strength reaching a preset correlation threshold and regularity distance below a preset distance threshold are identified as target state segments with stable demand representation, wherein: The preset association threshold is determined based on the overall distribution of the association strength between state segments, and the preset distance threshold is determined based on the overall distribution of the regular distance between candidate state segments.

[0026] In this embodiment, S23 specifically includes: S231. Mark the state segments whose association strength reaches the preset association threshold as candidate state segments, extract the segment change sequences corresponding to the candidate state segments, and arrange the segment change sequences according to the time position. S232. Traverse the segment change sequence of the two candidate state segments, calculate the change difference value between different time positions, write the change difference value into the corresponding matrix position, and form a regularized cost matrix. S233. Starting from the beginning of the regularization cost matrix, the cumulative cost is calculated in three progressive directions: horizontal, vertical, and diagonal. The change difference value of the current matrix position is added to the minimum cumulative cost of the previous matrix position to form the cumulative regularization cost matrix. S234. Backtrack from the end position of the cumulative regularization cost matrix to the beginning position to determine the regularization path between the two candidate state segments, and determine the regularization distance between the candidate state segments based on the cumulative change difference results corresponding to the regularization path. The lengths of the segment variation sequences are 6 and 8 respectively, and a 6×8 normalized cost matrix is ​​constructed accordingly. The difference in change is calculated by the absolute difference between the changes in state at two time points. Each change in state is calculated dimension by dimension in the order of the 64-dimensional health and wellness state vector and then summed. The cumulative cost calculation starts from the top left corner of the regularized cost matrix and proceeds recursively along the horizontal, vertical, and diagonal directions. During backtracking, start from the bottom right corner of the matrix and select the preceding positions with smaller cumulative costs in sequence until returning to the top left corner of the matrix to obtain a regular path; The differences in the changes at each position along the regular path are summed and then divided by the length of the regular path. When the obtained normalization distance is less than 0.35, the corresponding candidate state segment is retained for the selection of target state segment.

[0027] In this embodiment, S3 specifically includes: S31. Perform node-based processing on the target state fragments, converting each target state fragment into a requirement node, and maintaining a one-to-one correspondence between the requirement nodes and the target state fragments. S32. Perform node-based processing on health and wellness service resources, converting each health and wellness service resource into a resource node, and maintaining a one-to-one correspondence between resource nodes and health and wellness service resources; S33. Iterate through the demand nodes and resource nodes one by one. Based on the correlation strength of the target state segment and the health care service resources, establish cross-domain connection edges for demand nodes and resource nodes that meet the cross-domain connection conditions. The cross-domain connection conditions are: the correlation strength of the target state segment reaches the preset connection threshold, and the health care service resources and the target state segment meet the preset connection relationship. S34. For each cross-domain connection edge, convert the association strength corresponding to the cross-domain connection edge into a first value representing the degree to which the connection is established. Based on the unmatched portion of the demand node and resource node in the cross-domain connection conditions, it is converted into a second value indicating the degree to which the connection is not established; The difference between the sum of the first and second values ​​and the upper limit of the intuitive fuzzy value is used as the third value to represent the degree of uncertainty of the connection. The first, second, and third values ​​are sequentially assigned to the corresponding cross-domain connection edges to form the intuitive fuzzy number of the cross-domain connection edge. A supply-demand fuzzy heterogeneous graph is constructed using demand nodes, resource nodes, and cross-domain connection edges configured with intuitive fuzzy numbers.

[0028] In this embodiment, S4 specifically includes: S41. In the supply and demand fuzzy heterogeneous graph, the data representations corresponding to the demand node and the resource node are used as node features, and the neighboring nodes of the current node are determined according to the intuitive fuzzy number corresponding to the cross-domain connection edge. S42. A gated graph convolutional network is used to perform cross-domain aggregation of neighborhood node features. The gate coefficient is calculated based on the intuitive fuzzy number corresponding to the cross-domain connection edge. The proportion of neighborhood node features participating in the aggregation is controlled according to the gate coefficient. The aggregated neighborhood node features are then passed to the corresponding nodes. S43. Using the aggregated node features as the current round state, perform state recursion on demand nodes and resource nodes, and use the current round state recursion result as the input for the next round of aggregation. Repeat cross-domain aggregation and state recursion to complete multiple rounds of fusion update. S44. Perform bidirectional association calculations on the demand nodes and resource nodes after multiple rounds of fusion updates. Based on the association results of the demand nodes and resource nodes with the demand nodes, calculate the bilateral utility scores between the corresponding nodes.

[0029] In this embodiment, S42 specifically includes: S421. Taking the current demand node or the current resource node as the center, determine the adjacent resource nodes or adjacent demand nodes along the cross-domain connection edge, and take the node characteristics of the adjacent nodes as the features to be aggregated. S422. Perform gating conversion on the intuitive fuzzy numbers corresponding to the cross-domain connection edges, take the values ​​corresponding to the degree of connection establishment as positive adjustment terms, and take the values ​​corresponding to the degree of connection failure and the degree of connection uncertainty as suppression adjustment terms, and calculate the gating coefficient of the cross-domain connection edges. S423. Apply the gating coefficient to the corresponding feature to be aggregated, increase the participation ratio for features to be aggregated with a higher gating coefficient, and decrease the participation ratio for features to be aggregated with a lower gating coefficient, to obtain the neighborhood features after gating. S424. In the gated graph convolutional network, the neighborhood features after gated processing are linearly transformed, and neighborhood weighted summation is performed according to the connection direction of the cross-domain connection edge. The summation result is combined with the current node features and then nonlinearly mapped to obtain the cross-domain aggregated features of the current node.

[0030] In this embodiment, S5 specifically includes: S51. Using the cross-domain connection edges in the supply-demand fuzzy heterogeneous graph as the allocation object, the bilateral utility scores corresponding to the cross-domain connection edges are used as the allocation basis, and the total amount of matching flows that resource nodes can handle is constrained according to the preset service capacity, wherein: The preset service capacity is the maximum number of services that a resource node can handle in the current matching round. The matching flow is the service share allocated between demand nodes and resource nodes by the cross-domain connection edge; S52. Assign initial flow values ​​to cross-domain connection edges in descending order of bilateral utility scores, and accumulate the matched flows already received by each resource node during the assignment process; When the number of matching flows already handled by a resource node reaches the preset service capacity, stop adding more matching flows to the cross-domain connection edge connected to that resource node; S53. Compare the allocated matching flows of resource nodes with the preset service capacity. For matching flows that exceed the preset service capacity, reduce them from low to high according to the bilateral utility score, and allocate the reduced matching flows to cross-domain connection edges with remaining capacity. S54. Discretize and filter the adjusted matching flow, retain the cross-domain connection edges that meet the preset filtering conditions, and remove the cross-domain connection edges that are repeatedly accepted under the same demand node. Based on the retained cross-domain connection edges, form the service matching result.

[0031] In this embodiment, S6 specifically includes: S61. Obtain the feedback response data corresponding to the service matching result, and allocate the feedback response data to the corresponding cross-domain connection edge according to the cross-domain connection edge index in the service matching result, wherein the feedback response data is the response record formed after the service matching result is executed; S62. Extract the positive and negative response values ​​from the feedback response data. Increase the feedback support value of the corresponding cross-domain connection edge based on the positive response value, and increase the feedback suppression value of the corresponding cross-domain connection edge based on the negative response value. S63. Using the current edge weight of the corresponding cross-domain connection edge as the prior value, and the feedback support value and feedback suppression value as the observation value, the Bayesian confidence update method is used to calculate the posterior confidence value. S64. Based on the posterior confidence value, the edge weights of the corresponding cross-domain connection edges are recursively corrected, and the intuitive fuzzy number of the cross-domain connection edges is recalculated according to the corrected edge weights to complete the incremental update of the supply and demand fuzzy heterogeneous graph.

[0032] refer to Figure 3 A smart matching system for health and wellness services based on multi-source big data includes: The data processing module is used to acquire multi-source data in the context of health and wellness services and preprocess it to form a health and wellness status sequence. The segment screening module is used to perform sliding window segmentation on the health and wellness state sequence, use temporal mutual information metric to determine the correlation strength between state segments, and combine dynamic time warping algorithm to screen target state segments with stable demand representation. The graph construction module is used to map target state fragments to demand nodes, and health and wellness service resources to resource nodes. It establishes cross-domain connection edges based on the correlation strength and uses intuitive fuzzy numbers to perform semantic quantization configuration of the connection edges to construct a supply and demand fuzzy heterogeneous graph. The aggregation scoring module is used in the supply and demand fuzzy heterogeneous graph to perform cross-domain aggregation of the neighborhood features of demand nodes and resource nodes according to the cross-domain connection edge type, and to obtain a two-sided utility score by performing multiple rounds of fusion and updating of the aggregated features through state recursion. The matching and allocation module is used to allocate matching flows to cross-domain connection edges based on bilateral utility scores and preset service capacity constraints, and to discretize and filter the matching flows to generate service matching results. The feedback update module is used to obtain the feedback response data corresponding to the service matching results and to incrementally update the connection edge weights of the supply and demand fuzzy heterogeneous graph using the Bayesian confidence update method.

[0033] Example 1: To verify the feasibility of this invention in practice, it was applied to a service matching scenario of a comprehensive elderly care service platform. In this scenario, the platform simultaneously integrates multiple service resources, including home care, rehabilitation nursing, health consultation, medical assistance, and daily care, and user needs are relatively dispersed. Some users only filled in simple service intentions, some users had continuous health records and historical appointment records, and some users left evaluations and cancellation reasons after the service.

[0034] Traditional methods rely mainly on service type, distance, and manual screening for recommendations, which can easily lead to inaccurate judgment of user needs, repeated recommendations of popular resources, and some resources being idle while others are overloaded.

[0035] In this scenario, a batch of continuously running data was selected for verification. The data included 3,860 health and wellness users, 742 allocable service resources, and 18,420 historical service records. The system first preprocessed the user-side data, service-side data, and feedback-side data to form a health and wellness status sequence.

[0036] The health and wellness status sequence is then segmented using a sliding window. The correlation strength between status segments is filtered through temporal mutual information, and segments with unstable trends are eliminated using a dynamic time warping algorithm, resulting in target status segments that reflect actual needs. For example, some users may not immediately initiate an appointment after experiencing health fluctuations, but subsequently have continuous needs for rehabilitation care, in-home assessments, and health consultations. This invention can identify these delayed needs through the temporal relationships between status segments, rather than simply recommending services based on the user's current information.

[0037] After identifying the target state fragments, the system converts the target state fragments into demand nodes and the health and wellness service resources into resource nodes, and establishes cross-domain connection edges based on the association strength. For cases where there is uncertainty in the connection relationship, intuitive fuzzy numbers are used to quantify the cross-domain connection edges, so that the relationships of "suitable to undertake", "unsuitable to undertake", and "temporarily uncertain" can be expressed in the graph.

[0038] Subsequently, the gated graph convolutional network performs cross-domain aggregation along cross-domain connection edges, enabling demand nodes to absorb resource-side states and resource nodes to obtain demand-side change information. After multiple rounds of state recursion, the system obtains a two-sided utility score between demand nodes and resource nodes, and performs matching flow allocation based on the preset service capacity of resource nodes. In cases where the same user is repeatedly served by multiple resources or where some resource capacities are full, the system reduces and transfers the matching flow, and then uses discretization filtering to form the final service matching result.

[0039] In the verification, the average matching time of the traditional matching method was 7.8 minutes, which was reduced to 3.1 minutes by this invention; the user confirmation rate was 68.4% under the traditional method, reaching 82.7% under this invention; and the resource capacity overrun rate was 11.6% under the traditional method, reduced to 3.8% under this invention. In the feedback processing after service completion, the system uses the confirmation result, completion result, and evaluation result as feedback response data, and corrects the connection edge weights in the supply-demand fuzzy heterogeneous graph through Bayesian confidence updates. After three rounds of feedback updates, the duplicate recommendation rate of low-evaluation services decreased from 14.2% to 5.1%, and the resource utilization rate of high-feedback services increased from 71.5% to 86.9%.

[0040] Table 1 Comparison of the operational effects of matching health and wellness services

[0041] As can be seen from the data in Table 1, the present invention can not only improve the single matching result, but also continuously correct the subsequent matching relationship based on service feedback, making it suitable for application scenarios where the supply and demand of health and wellness services are subject to significant dynamic changes.

[0042] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for intelligent matching of health and wellness services based on multi-source big data, characterized in that, Includes the following steps: S1. Acquire multi-source data in the health and wellness service scenario and preprocess it to form a health and wellness status sequence; S2. The health and wellness state sequence is segmented by a sliding window, and the temporal mutual information metric is used to determine the correlation strength between state segments. Combined with the dynamic time warping algorithm, target state segments with stable demand representation are screened out. S3. Map the target state fragments to demand nodes and the health and wellness service resources to resource nodes. Establish cross-domain connection edges based on the correlation strength and use intuitive fuzzy numbers to perform semantic quantization configuration of the connection edges to construct a supply and demand fuzzy heterogeneous graph. S4. In the supply and demand fuzzy heterogeneous graph, according to the cross-domain connection edge type, a gated graph convolutional network is used to perform cross-domain aggregation of the neighborhood features of demand nodes and resource nodes respectively, and the aggregated features are fused and updated in multiple rounds through state recursion to obtain the bilateral utility score. S5. Based on the bilateral utility score and the preset service capacity constraint, a multi-objective trade-off optimization method is used to match the flow allocation of cross-domain connection edges, and the matching flow is discretized and filtered to generate service matching results. S6. Obtain the feedback response data corresponding to the service matching result, and use the Bayesian confidence update method to incrementally update the connection edge weights of the supply and demand fuzzy heterogeneous graph.

2. The intelligent matching method for health and wellness services based on multi-source big data according to claim 1, characterized in that, The preprocessing specifically includes: Source labeling is performed on multi-source data in the context of health and wellness services, including user-side data, service-side data, and feedback-side data. A unified index is established based on user identifier, service identifier, and time identifier. Data fields from different sources are mapped to a unified field set, and the field values ​​are encoded and converted. Based on a unified index, duplicate records in multi-source data are merged, missing fields are filled in, and data records that exceed the preset value range are marked as abnormal records and then removed. The data that has completed field mapping is time-aligned according to the preset time granularity, and the data records under the same index are spliced ​​together to obtain the health and wellness status vector. Arrange the health and wellness status vectors in chronological order to form a health and wellness status sequence.

3. The intelligent matching method for health and wellness services based on multi-source big data according to claim 1, characterized in that, S2 specifically includes: S21. Traverse the health and wellness state sequence according to the fixed window length and sliding step size, cut the continuous health and wellness state vector into multiple state segments, and record the time position corresponding to each state segment. S22. Extract the change amount of the health care state vector within the state segment to obtain the segment change sequence, and calculate the temporal mutual information value between different state segments based on the segment change sequence, and use the temporal mutual information value as the correlation strength between state segments. S23. State segments whose association strength reaches a preset association threshold are taken as candidate state segments. The dynamic time warping algorithm is used to align the segment change sequences between candidate state segments to obtain the warped distance between candidate state segments. S24. Candidate state segments are filtered based on correlation strength and regularity distance. Candidate state segments with correlation strength reaching a preset correlation threshold and regularity distance below a preset distance threshold are identified as target state segments with stable demand representation, wherein: The preset association threshold is determined based on the overall distribution of the association strength between state segments, and the preset distance threshold is determined based on the overall distribution of the regular distance between candidate state segments.

4. The intelligent matching method for health and wellness services based on multi-source big data according to claim 3, characterized in that, S23 specifically includes: S231. Mark the state segments whose association strength reaches the preset association threshold as candidate state segments, extract the segment change sequences corresponding to the candidate state segments, and arrange the segment change sequences according to the time position. S232. Traverse the segment change sequence of the two candidate state segments, calculate the change difference value between different time positions, write the change difference value into the corresponding matrix position, and form a regularized cost matrix. S233. Starting from the beginning of the regularization cost matrix, the cumulative cost is calculated in three progressive directions: horizontal, vertical, and diagonal. The change difference value of the current matrix position is added to the minimum cumulative cost of the previous matrix position to form the cumulative regularization cost matrix. S234. Backtrack from the end position of the cumulative normalization cost matrix to the beginning position to determine the normalization path between the two candidate state segments, and determine the normalization distance between the candidate state segments based on the cumulative result of the change difference corresponding to the normalization path.

5. The intelligent matching method for health and wellness services based on multi-source big data according to claim 1, characterized in that, S3 specifically includes: S31. Perform node-based processing on the target state fragments, converting each target state fragment into a requirement node, and maintaining a one-to-one correspondence between the requirement nodes and the target state fragments. S32. Perform node-based processing on health and wellness service resources, converting each health and wellness service resource into a resource node, and maintaining a one-to-one correspondence between resource nodes and health and wellness service resources; S33. Traverse the demand nodes and resource nodes one by one, and establish cross-domain connection edges for demand nodes and resource nodes that meet the cross-domain connection conditions based on the correlation strength and health care service resources corresponding to the target state fragment. S34. For each cross-domain connection edge, convert the association strength corresponding to the cross-domain connection edge into a first value representing the degree to which the connection is established. Based on the unmatched portion of the demand node and resource node in the cross-domain connection conditions, it is converted into a second value indicating the degree to which the connection is not established; The difference between the sum of the first and second values ​​and the upper limit of the intuitive fuzzy value is used as the third value to represent the degree of uncertainty of the connection. The first, second, and third values ​​are sequentially assigned to the corresponding cross-domain connection edges to form the intuitive fuzzy number of the cross-domain connection edge. A supply-demand fuzzy heterogeneous graph is constructed using demand nodes, resource nodes, and cross-domain connection edges configured with intuitive fuzzy numbers.

6. The intelligent matching method for health and wellness services based on multi-source big data according to claim 1, characterized in that, S4 specifically includes: S41. In the supply and demand fuzzy heterogeneous graph, the data representations corresponding to the demand node and the resource node are used as node features, and the neighboring nodes of the current node are determined according to the intuitive fuzzy number corresponding to the cross-domain connection edge. S42. A gated graph convolutional network is used to perform cross-domain aggregation of neighborhood node features. The gate coefficient is calculated based on the intuitive fuzzy number corresponding to the cross-domain connection edge. The proportion of neighborhood node features participating in the aggregation is controlled according to the gate coefficient. The aggregated neighborhood node features are then passed to the corresponding nodes. S43. Using the aggregated node features as the current round state, perform state recursion on demand nodes and resource nodes, and use the current round state recursion result as the input for the next round of aggregation. Repeat cross-domain aggregation and state recursion to complete multiple rounds of fusion update. S44. Perform bidirectional association calculations on the demand nodes and resource nodes after multiple rounds of fusion updates. Based on the association results of the demand nodes and resource nodes with the demand nodes, calculate the bilateral utility scores between the corresponding nodes.

7. The intelligent matching method for health and wellness services based on multi-source big data according to claim 6, characterized in that, S42 specifically includes: S421. Taking the current demand node or the current resource node as the center, determine the adjacent resource nodes or adjacent demand nodes along the cross-domain connection edge, and take the node characteristics of the adjacent nodes as the features to be aggregated. S422. Perform gating conversion on the intuitive fuzzy numbers corresponding to the cross-domain connection edges, take the values ​​corresponding to the degree of connection establishment as positive adjustment terms, and take the values ​​corresponding to the degree of connection failure and the degree of connection uncertainty as suppression adjustment terms, and calculate the gating coefficient of the cross-domain connection edges. S423. Apply the gating coefficient to the corresponding feature to be aggregated, increase the participation ratio for features to be aggregated with a higher gating coefficient, and decrease the participation ratio for features to be aggregated with a lower gating coefficient, to obtain the neighborhood features after gating. S424. In the gated graph convolutional network, the neighborhood features after gated processing are linearly transformed, and neighborhood weighted summation is performed according to the connection direction of the cross-domain connection edge. The summation result is combined with the current node features and then nonlinearly mapped to obtain the cross-domain aggregated features of the current node.

8. The intelligent matching method for health and wellness services based on multi-source big data according to claim 1, characterized in that, S5 specifically includes: S51. Using the cross-domain connection edges in the supply-demand fuzzy heterogeneous graph as the allocation object, the bilateral utility scores corresponding to the cross-domain connection edges are used as the allocation basis, and the total amount of matching flows that resource nodes can handle is constrained according to the preset service capacity, wherein: The preset service capacity is the upper limit of the number of services that a resource node can undertake in the current matching round, and the matching flow is the service undertaking share allocated between the demand node and the resource node by the cross-domain connection edge. S52. Assign initial flow values ​​to cross-domain connection edges in descending order of bilateral utility scores, and accumulate the matched flows already received by each resource node during the assignment process; When the number of matching flows already handled by a resource node reaches the preset service capacity, stop adding more matching flows to the cross-domain connection edge connected to that resource node; S53. Compare the allocated matching flows of resource nodes with the preset service capacity. For matching flows that exceed the preset service capacity, reduce them from low to high according to the bilateral utility score, and allocate the reduced matching flows to cross-domain connection edges with remaining capacity. S54. Discretize and filter the adjusted matching flow, retain the cross-domain connection edges that meet the preset filtering conditions, and remove the cross-domain connection edges that are repeatedly accepted under the same demand node. Based on the retained cross-domain connection edges, form the service matching result.

9. The intelligent matching method for health and wellness services based on multi-source big data according to claim 1, characterized in that, S6 specifically includes: S61. Obtain the feedback response data corresponding to the service matching result, and allocate the feedback response data to the corresponding cross-domain connection edge according to the cross-domain connection edge index in the service matching result. S62. Extract the positive and negative response values ​​from the feedback response data. Increase the feedback support value of the corresponding cross-domain connection edge based on the positive response value, and increase the feedback suppression value of the corresponding cross-domain connection edge based on the negative response value. S63. Using the current edge weight of the corresponding cross-domain connection edge as the prior value, and the feedback support value and feedback suppression value as the observation value, the Bayesian confidence update method is used to calculate the posterior confidence value. S64. Based on the posterior confidence value, the edge weights of the corresponding cross-domain connection edges are recursively corrected, and the intuitive fuzzy number of the cross-domain connection edges is recalculated according to the corrected edge weights to complete the incremental update of the supply and demand fuzzy heterogeneous graph.

10. A smart matching system for health and wellness services based on multi-source big data, comprising executing the smart matching method for health and wellness services based on multi-source big data as described in any one of claims 1 to 9, characterized in that, include: The data processing module is used to acquire multi-source data in the context of health and wellness services and preprocess it to form a health and wellness status sequence. The segment screening module is used to perform sliding window segmentation on the health and wellness state sequence, use temporal mutual information metric to determine the correlation strength between state segments, and combine dynamic time warping algorithm to screen target state segments with stable demand representation. The graph construction module is used to map target state fragments to demand nodes, and health and wellness service resources to resource nodes. It establishes cross-domain connection edges based on the correlation strength and uses intuitive fuzzy numbers to perform semantic quantization configuration of the connection edges to construct a supply and demand fuzzy heterogeneous graph. The aggregation scoring module is used in the supply and demand fuzzy heterogeneous graph to perform cross-domain aggregation of the neighborhood features of demand nodes and resource nodes according to the cross-domain connection edge type, and to obtain a two-sided utility score by performing multiple rounds of fusion and updating of the aggregated features through state recursion. The matching and allocation module is used to allocate matching flows to cross-domain connection edges based on bilateral utility scores and preset service capacity constraints, and to discretize and filter the matching flows to generate service matching results. The feedback update module is used to obtain the feedback response data corresponding to the service matching results and to incrementally update the connection edge weights of the supply and demand fuzzy heterogeneous graph using the Bayesian confidence update method.

Images