Nursing plan recommendation method based on customer profile data analysis

Abstract

This invention relates to the fields of nursing services and artificial intelligence technology, and particularly to a nursing plan recommendation method based on customer profile data analysis. The method includes the following steps: S1, collecting basic and historical nursing information of users; S2, constructing a multi-layered tag graph and generating user profiles; S3, generating a candidate nursing path set based on semantic paths; S4, extracting nursing plans adopted by users with similar profiles; S5, constructing a reward function that integrates ratings and structural information; S6, obtaining user feedback and dynamically adjusting the path structure weights; S7, outputting the recommended plan with the highest matching degree based on the optimization results. This invention integrates structured and unstructured information to construct a semantic graph, combines similar user plans and multi-dimensional feedback to construct a reward function, dynamically optimizes the path structure, and improves the matching and adaptability of the recommended plan; through rating ranking and environmental interference elimination, it improves the intelligence level and resource matching efficiency of nursing recommendations, and increases the breadth of application.

Description

Technical Field

[0001] This invention relates to the fields of nursing services and artificial intelligence technology, and in particular to a method for recommending nursing plans based on customer profile data analysis. Background Technology

[0002] With the accelerating aging of the population and the diversification of healthcare service demands, the importance of nursing services is becoming increasingly prominent. Traditional nursing plans rely heavily on the experience of caregivers and fixed rules, lacking systematic and intelligent support, making it difficult to meet the personalized and changing nursing needs of different users. In recent years, artificial intelligence technology, especially recommendation systems based on user profiles, has received widespread attention in the healthcare field. Deeply mining user basic information, behavioral data, and historical nursing records to achieve personalized nursing plan recommendations has become a research hotspot.

[0003] Existing nursing recommendation methods typically employ rule-based matching or simple similarity calculations, which suffer from limitations such as a single user profile dimension, a lack of dynamic adjustment of recommendation results, and insufficient resource matching. For example, some published patents and related literature propose recommendations based on basic user information and nursing records, but they fail to fully consider dynamic optimization based on user feedback and the constraints of nursing resources, resulting in insufficient accuracy and practicality of the recommendations. Furthermore, the recommendation process is often a black box operation, lacking necessary explainability, which affects users' trust and acceptance of the recommendation results.

[0004] Furthermore, nursing scenarios are complex and ever-changing, with dynamic shifts in user needs and resource supply. Traditional static recommendation models struggle to cope effectively, necessitating the introduction of advanced technologies such as multi-layered label graph construction and reinforcement learning to achieve multi-dimensional information fusion, dynamic path optimization, and intelligent resource allocation. By constructing a structured semantic label graph and combining it with a reward mechanism driven by real-time user feedback, the personalization, accuracy, and feasibility of nursing plan recommendations can be significantly improved, meeting the diverse needs of actual nursing services. Therefore, developing an intelligent nursing plan recommendation method based on customer profile data analysis and incorporating reinforcement learning feedback has significant technical and application value.

[0005] Chinese Patent Publication No. CN115835799A discloses an oral care device recommendation system for recommending types of oral care accessories to be used with oral care devices. The recommendations consider information about the user's oral geometry and, preferably, also user behavior information about how a particular user performs their oral care using the oral care device. The recommendations are based on modeling the mechanical interaction between one or more oral care accessories from a set of accessories and the user's oral geometry when the user performs an oral care routine. A cleanliness metric is determined from the modeling, representing the effectiveness of the oral care routine when using the one or more oral care accessories.

[0006] However, existing technologies primarily focus on matching and recommending oral care accessories. They generate cleaning metrics and recommend accessory types by modeling users' oral geometry and brushing behavior. In contrast, their recommendation dimensions are singular, limited to physical structure matching and static behavior analysis. They lack multi-dimensional label graph modeling and user feedback-driven dynamic optimization mechanisms, and do not address the availability of nursing resources or the elimination of environmental interference. Therefore, they struggle to achieve cross-scenario, personalized, and interpretable intelligent nursing care solutions. Summary of the Invention

[0007] To address this, the present invention provides a nursing plan recommendation method based on customer profile data analysis, which overcomes the problems in the prior art where the nursing plan recommendation execution method is unstable and the reproducible path is singular in complex nursing environments, making it difficult for the recommended plan to adapt to diverse nursing needs.

[0008] To achieve the above objectives, the present invention provides a method for recommending nursing care plans based on customer profile data analysis, comprising:

[0009] Obtain the target user information input by the client and convert the target user information into corresponding feature vector nodes;

[0010] The feature vector nodes are integrated based on the multi-layer tag graph construction method to construct a target user profile;

[0011] The multi-layer label graph construction method involves assigning each feature vector node to its corresponding label type through structured semantic mapping, and establishing associated edges based on the label hierarchy structure.

[0012] In the multi-layered label graph, a set of candidate paths is formed by constructing several semantic paths that connect the path starting point and the nursing plan node based on semantic association, with a specific category label node as the path starting point.

[0013] Based on historical sample data, nursing plans adopted by users similar to the target user profile are extracted, and the nursing plan with the highest adoption frequency is selected as the initial recommended nursing plan, and the corresponding semantic path is used as the initial recommendation path.

[0014] Constructing the reward function in a reinforcement learning model;

[0015] Obtain user feedback information under the initial recommended care plan and the corresponding feedback category after the reward function is applied, so as to perform corresponding optimization and adjustment on the key nodes and edge weights in the candidate path;

[0016] The comprehensive score of each path after optimization and adjustment is obtained. Based on user needs and after excluding environmental interference factors, the nursing plan associated with the highest-scoring semantic path in the feedback category is selected as the recommended nursing plan. Recommendation information is generated and sent to the client.

[0017] Furthermore, assigning each feature node to its corresponding label type through structured semantic mapping includes:

[0018] Basic information includes direct information and supplementary information;

[0019] The supplementary information is segmented based on semantic structure and punctuation marks to obtain several supplementary information word blocks;

[0020] Map basic information word blocks to the same semantic vector space;

[0021] Traverse each basic information word block, calculate its semantic matching degree with each tag type, and when the first matching degree result is obtained, map the feature vector node corresponding to the basic information word block to the corresponding tag type;

[0022] The label types include basic information label types and nursing label types.

[0023] Furthermore, establishing association edges based on the tag hierarchy and semantic relationships includes:

[0024] Determine whether the historical nursing information in the target user's information is valid;

[0025] If the historical nursing information is valid, the semantic path corresponding to the most recent historical nursing plan will be used as the initial recommended path.

[0026] If the historical nursing information is invalid, the label hierarchy is determined based on the nursing information knowledge graph, and each basic information label node is connected to the nursing label node to establish an association edge.

[0027] Furthermore, determining the validity of a target user's historical care information includes:

[0028] Retrieve the target user's historical nursing records within a preset time window;

[0029] Historical nursing records include the number of historical nursing protocols and the pathways of historical nursing protocols;

[0030] Determine whether there is at least one historical nursing care plan and whether the corresponding historical nursing care plan path is complete, so as to obtain a result on whether the historical nursing care information is effective.

[0031] Furthermore, extracting the care plans adopted by users similar to the target user profile includes:

[0032] The similarity between the target user's basic information tag nodes and the standard user profile's basic information tag nodes is determined to identify the corresponding similarity profile type.

[0033] When the judgment result is the second similarity profile type, the priority of the basic information type corresponding to each basic information tag node is determined based on the nursing information knowledge graph, and sorting is performed.

[0034] Obtain the basic information tag types that are in the first half of the priority interval after sorting, and update them to the basic information tag types to be analyzed;

[0035] The similarity thresholds of the basic information tag types to be analyzed are compared with the similarity thresholds of the core basic information tags to obtain the comparison results, and corresponding measures are implemented.

[0036] Furthermore, obtaining the comparison results and implementing corresponding measures includes:

[0037] If all the basic information tag types to be analyzed are greater than the core basic information tag similarity threshold, the first comparison result is obtained, and the tag type content of the second half priority interval is adjusted.

[0038] If the similarity of any basic information tag type to be analyzed is less than or equal to the core basic information tag similarity threshold, a second comparison result is obtained, and the similarity judgment step between the target user's basic information tag type and the standard user profile's basic information tag type is re-executed.

[0039] Furthermore, the reward function in the reinforcement learning model includes:

[0040] Obtain feedback from target users and define the categories and rating dimensions of the feedback;

[0041] The feedback information from target users is converted into rating values, and the rating values ​​are weighted and summed using a reward function before the function value is output.

[0042] The function value is compared with the standard function threshold, and the weights of the associated edges involved in the current semantic path are adjusted accordingly when the second function value comparison result is obtained.

[0043] Furthermore, adjusting the weights of the associated edges involved in the current semantic path includes:

[0044] Obtain negative feedback information from target users to identify the association edges between basic information tag nodes and nursing plan nodes that significantly affect the accuracy of recommendation results in the current initial recommendation path;

[0045] Based on the negative score output in the reward function, the weights of the identified associated edges are updated according to a preset decay factor.

[0046] Furthermore, excluding environmental interference factors includes:

[0047] Identify the nursing preference conditions set in the feedback information of target users;

[0048] If the nursing preference conditions cannot be met under the current nursing resources, the recommended direction is determined based on the preset strategy, and the nursing plan with the highest quality or the lowest price is selected from the available paths.

[0049] Send environmental limitation information to the client regarding the recommendation results.

[0050] Furthermore, sending recommendation information to the client includes:

[0051] The recommended care plans are transformed into structured data presentations and sent to target users in a graphic and text format;

[0052] The data displayed includes the reasons for recommendation, matching criteria, and key tags.

[0053] Compared with existing technologies, the advantages of this invention lie in its integration of customer profiling modeling, semantic path analysis, and reinforcement learning mechanisms to achieve intelligent nursing solution recommendations for complex nursing environments. By constructing a multi-layered tag graph, user basic information and historical nursing records are structurally categorized into a tag system, forming a semantic path set, thus improving the feature coverage breadth and path interpretability of the recommendations. Initial recommendation schemes are extracted based on similar user behaviors, enhancing the individual matching of the schemes. Furthermore, a reinforcement learning reward function is constructed, dynamically adjusting path weights based on user positive and negative feedback information, strengthening the adaptability and closed-loop optimization capability of the recommendation strategy. Simultaneously, a nursing resource and environmental factor exclusion mechanism is introduced before recommendation execution to ensure the feasibility and execution stability of the recommendation results. This method effectively solves the problems of single path, slow response, and unreproducible results in traditional nursing recommendations, possessing high adaptability, interpretability, and feasibility, significantly improving the efficiency of intelligent nursing services and user trust.

[0054] Furthermore, by setting a historical data window and structurally acquiring the target user's historical nursing records, the effectiveness of nursing plans can be judged by combining the number of times they are adopted with the completeness of the associated semantic paths. By verifying the continuity of basic information tag nodes, nursing tag nodes, and nursing plan nodes, the accuracy of historical nursing information in the initial recommendation path selection and similarity profile generation process can be improved, effectively avoiding interference caused by incomplete paths or distorted information on the recommendation results.

[0055] Furthermore, by quantifying the comprehensive performance of user feedback through weighted summation, we ensure that different feedback dimensions contribute reasonably to the overall evaluation. By combining the weights of each associated edge in the path and the dynamic mapping adjustment function based on feedback, we achieve real-time optimization of the path structure, reduce the weight of path parts with poor feedback, and improve the adaptability and accuracy of the recommended path. Attached Figure Description

[0056] Figure 1 This is a schematic diagram of the nursing plan recommendation method based on customer profile data analysis according to an embodiment of the present invention;

[0057] Figure 2 This is a logical decision graph for establishing associated edges based on the tag hierarchy structure and semantic association in an embodiment of the present invention;

[0058] Figure 3 This is a logic diagram for determining the validity of a target user's historical nursing information in an embodiment of the present invention.

[0059] Figure 4 This is a schematic diagram illustrating the structure of adjusting the weights of associated edges involved in the current semantic path according to an embodiment of the present invention. Detailed Implementation

[0060] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0061] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0062] It should be noted that in the description of this invention, the terms "upper", "lower", "left", "right", "inner", "outer", etc., which indicate directions or positional relationships, are based on the directions or positional relationships shown in the accompanying drawings. This is only for the convenience of description and is not intended to indicate or imply that the device or element must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, it should not be construed as a limitation of this invention.

[0063] Furthermore, it should be noted that, in the description of this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0064] Please see Figure 1 As shown, it is a structural schematic diagram of a nursing plan recommendation method based on customer profile data analysis according to an embodiment of the present invention. The present invention provides a nursing plan recommendation method based on customer profile data analysis, including:

[0065] Step S1: Obtain the target user information input by the client and convert the target user information into corresponding feature vector nodes;

[0066] Step S2: Integrate the feature vector nodes based on the multi-layer label graph construction method to construct the target user profile;

[0067] The multi-layer label graph construction method involves assigning each feature vector node to its corresponding label type through structured semantic mapping, and establishing associated edges based on the label hierarchy structure.

[0068] Step S3: In the multi-layered label graph, using a specific category label node as the starting point of the path, construct several semantic paths that connect the starting point of the path with the nursing plan node based on semantic association, forming a candidate path set.

[0069] Step S4: Extract nursing plans adopted by users similar to the target user profile based on historical sample data, select the nursing plan with the highest adoption frequency as the initial recommended nursing plan, and use the corresponding semantic path as the initial recommended path.

[0070] Step S5: Construct the reward function in the reinforcement learning model;

[0071] Step S6: Obtain the user's feedback information under the initial recommended care plan and the corresponding feedback category after the reward function is calculated, so as to perform corresponding optimization and adjustment on the key nodes and edge weights in the candidate path;

[0072] Step S7: Obtain the comprehensive score of each path after optimization and adjustment. Based on user needs and after excluding environmental interference factors, select the nursing plan associated with the semantic path with the highest score in its respective feedback category as the recommended nursing plan, form recommendation information and send it to the client.

[0073] In this embodiment, the target user profile is a personalized dynamic semantic graph, which exists in the form of a multi-layered label graph. It includes feature vector nodes transformed from basic information and historical care information, label nodes mapped from user feature vectors, and associated edges established based on label hierarchy and semantic relevance. The node weights and associated edges are dynamically optimized through reinforcement learning based on user feedback.

[0074] The target user information includes basic information and historical nursing information.

[0075] By integrating customer profiling modeling, semantic path analysis, and reinforcement learning mechanisms, this invention achieves intelligent nursing solution recommendation for complex nursing environments. By constructing a multi-layered tag graph, user basic information and historical nursing records are structurally categorized into a tag system, forming a semantic path set, thus improving the feature coverage breadth and path interpretability of the recommendations. Initial recommendation schemes are extracted based on similar user behaviors, enhancing the individual matching of the schemes. Furthermore, a reinforcement learning reward function is constructed, dynamically adjusting path weights based on user positive and negative feedback information, strengthening the adaptability and closed-loop optimization capability of the recommendation strategy. Simultaneously, a nursing resource and environmental factor exclusion mechanism is introduced before recommendation execution to ensure the feasibility and execution stability of the recommendation results. This method effectively solves the problems of single path, slow response, and unreproducible results in traditional nursing recommendations, possessing high adaptability, interpretability, and feasibility, significantly improving the efficiency of intelligent nursing services and user trust.

[0076] Specifically, assigning each feature node to its corresponding label type through structured semantic mapping includes:

[0077] Basic information includes direct information and supplementary information;

[0078] The supplementary information is segmented based on semantic structure and punctuation marks to obtain several supplementary information word blocks;

[0079] Map basic information word blocks to the same semantic vector space;

[0080] Traverse each basic information word block, calculate its semantic matching degree with each tag type, and when the first matching degree result is obtained, map the feature vector node corresponding to the basic information word block to the corresponding tag type;

[0081] The label types include basic information label types and nursing label types.

[0082] In this embodiment, the basic information input data in the target user information is divided into direct information and supplementary information;

[0083] The basic information includes structured fields such as age group, gender, classification of underlying diseases, skin integrity status, consciousness and communication ability, excretory function status, mobility status, and infection risk status.

[0084] The supplementary information refers to open-ended text content filled in by the target user;

[0085] The supplementary information is analyzed semantically using a natural language processing model, and the text paragraphs are semantically segmented based on the position of punctuation marks to form several semantically complete and clearly defined supplementary information word blocks.

[0086] All direct information chunks and supplementary information chunks are mapped to a unified semantic vector space;

[0087] Cosine similarity is used to calculate the degree of matching between the text content of the feature vector node and the label type in the semantic vector space, which is used as the semantic matching degree.

[0088] Among them, the text content of the feature vector node is the direct information word block and the supplementary information word block;

[0089] A semantic matching degree threshold is set, which is obtained based on historical semantic annotation data and has a value range of 0.72 to 0.85. Preferably, in this embodiment, the semantic matching degree threshold is set to 0.80.

[0090] Compare the cosine similarity with the semantic matching threshold;

[0091] If the cosine similarity is greater than the semantic matching threshold, the word block has a valid semantic relationship with the label type, and the feature vector node corresponding to the word block is mapped to the corresponding label type;

[0092] If the cosine similarity is less than or equal to the semantic matching threshold, continue to traverse the semantic vectors corresponding to other tag type nodes, recalculate the similarity and determine whether the threshold condition is met.

[0093] If the similarity between a certain word block and multiple tag types is greater than the semantic matching threshold, the best matching tag is selected and mapped according to the hierarchical priority of the tag type, so as to be assigned to the corresponding tag type.

[0094] The hierarchical priority is determined based on the types of tags in the nursing information knowledge graph. Different tag types have different importance in user profile construction, and their priority changes accordingly.

[0095] A semantic path refers to a sequence of paths that start from a specific label node and connect the nursing plan nodes step by step, relying on the semantic relevance between the label nodes in the graph.

[0096] Nursing information knowledge graph is a structured semantic graph constructed in the nursing service domain to describe nursing-related tags, attributes and their relationships. The graph integrates historical nursing records, tag definitions and nursing operation data, and uses tag nodes as basic units to construct a complete graph structure through causal or correlational relationships between tags, providing a logical basis for semantic path generation, similar user identification and solution recommendation.

[0097] Tag hierarchy refers to dividing tag types into multiple levels of categories and establishing hierarchical relationships between them.

[0098] Each tag type contains multiple nodes; after completing semantic mapping, the system transforms user basic information or supplementary information word blocks into feature vectors, and assigns them to a certain tag type according to semantic matching degree, forming a tag node set. The existence of multiple nodes allows the same tag type to express more detailed individual differences.

[0099] Associative edges are connecting lines that connect different label nodes, representing the direct relationship between these label nodes;

[0100] Associative edges describe the connection relationships between labels and also serve the purpose of path scoring and weight updates;

[0101] During reinforcement learning, the system dynamically intervenes in path generation by adjusting the weights of associated edges, thereby making recommendations more consistent with user feedback and environmental conditions.

[0102] The basic information of target users is divided into direct information and supplementary information. The supplementary information is subjected to semantic structure analysis and segmentation to extract word blocks with clear semantic boundaries. All word blocks are mapped to a unified semantic vector space. Cosine similarity is used to calculate the matching degree between text and tag type. Combined with matching degree threshold and hierarchical priority filtering, the accuracy of word block attribution is improved. This structured semantic mapping realizes the structured expression of unstructured information and improves the information fusion accuracy and recommendation path stability of the multi-layer tag graph construction of user profile.

[0103] Please see Figure 2 As shown, it is the logical decision diagram for establishing association edges based on the tag hierarchy structure and semantic association in an embodiment of the present invention;

[0104] Specifically, establishing association edges based on label hierarchy and semantic association includes:

[0105] Determine whether the historical nursing information in the target user's information is valid;

[0106] If the historical nursing information is valid, the semantic path corresponding to the most recent historical nursing plan will be used as the initial recommended path.

[0107] If the historical nursing information is invalid, the label hierarchy is determined based on the nursing information knowledge graph, and each basic information label node is connected to the nursing label node to establish an association edge.

[0108] Please see Figure 3 As shown, it is a logic determination diagram for judging whether the historical nursing information of the target user is valid according to an embodiment of the present invention;

[0109] Specifically, determining the validity of a target user's historical care information includes:

[0110] Retrieve the target user's historical nursing records within a preset time window;

[0111] Historical nursing records include the number of historical nursing protocols and the pathways of historical nursing protocols;

[0112] Determine whether there is at least one historical nursing care plan and whether the corresponding historical nursing care plan path is complete, so as to obtain a result on whether the historical nursing care information is effective.

[0113] In this embodiment, the system sets the historical data window to 30 days and obtains the target user's historical nursing records on the nursing platform within this time period.

[0114] The historical nursing records include the nursing protocol number adopted for each nursing service, the adoption time, and the semantic path associated with the nursing protocol in the multi-layer tag map;

[0115] The number of times a nursing plan is adopted is counted. If no nursing plan is adopted within a preset time window, the target user's historical nursing information is deemed invalid.

[0116] If there are records of nursing plan adoption, then it is further determined whether the semantic path corresponding to each nursing record is complete;

[0117] The semantic path integrity refers to whether the number of basic information tag nodes, the number of nursing tag nodes, and the continuity and semantic direction consistency of nursing plan nodes involved in the path meet the set standards.

[0118] If at least one semantic path structure is complete, the system determines that the target user's historical nursing information is valid and selects the semantic path of the most recent nursing record as the source of the initial recommendation path or similarity profile path.

[0119] If a complete semantic path does not exist, the historical nursing information is deemed invalid.

[0120] By setting a historical data window and obtaining the target user's historical nursing records in a structured manner, the effectiveness of nursing plans can be judged by combining the number of times they are adopted with the completeness of the associated semantic paths. By verifying the continuity of basic information tag nodes, nursing tag nodes, and nursing plan nodes, the accuracy of the adaptation of historical nursing information in the initial recommendation path selection and similarity profile generation process can be improved, effectively avoiding the interference of incomplete paths or information distortion on the recommendation results.

[0121] Specifically, extracting the care plans adopted by users similar to the target user profile includes:

[0122] The similarity between the target user's basic information tag nodes and the standard user profile's basic information tag nodes is determined to identify the corresponding similarity profile type.

[0123] When the judgment result is the second similarity profile type, the priority of the basic information type corresponding to each basic information tag node is determined based on the nursing information knowledge graph, and sorted in order of priority from high to low.

[0124] Obtain the basic information tag types that are in the first half of the priority interval after sorting, and update them to the basic information tag types to be analyzed;

[0125] The similarity thresholds of the basic information tag types to be analyzed are compared with the similarity thresholds of the core basic information tags to obtain the comparison results, and corresponding measures are implemented.

[0126] In this embodiment, the standard user profile is not unique, but is a set of typical user profiles constructed based on a large amount of user data through cluster analysis and historical user profile samples.

[0127] This set covers the basic information features of different user groups, serving as a reference for judging the similarity of target user profiles;

[0128] Standard user profiles contain multiple categories, and the category is determined based on similarity during matching;

[0129] The similarity between the target user's basic information tag nodes and the standard user profile's basic information tag nodes is determined. If the target user's basic information tag nodes completely overlap with the standard user profile's basic information tag nodes, the first similarity profile type is obtained. The nursing plan with the highest adoption frequency under the standard user profile is selected as the initial recommended nursing plan, and its corresponding semantic path is used as the initial recommended path.

[0130] If the basic information tag nodes of the target user do not completely overlap with the basic information tag nodes of the standard user profile, a second similarity profile type is obtained. Based on the nursing information knowledge graph, the priority of the basic information type corresponding to each basic information tag node is determined, and they are sorted from high to low.

[0131] The nursing information knowledge graph is a multi-layered tag semantic graph structure, which constructs multiple basic information tag nodes including age group, gender, basic disease classification, skin integrity status, consciousness and communication ability, excretory function status, mobility status and infection risk status, and establishes association with nursing tag nodes through semantic edges;

[0132] Priority determination comprehensively considers the following three dimensions: First, the co-occurrence frequency of basic information tag types and nursing tag nodes in historical nursing pathways; second, the semantic association strength between tag nodes, i.e., the degree of semantic coupling and logical relevance between tags; and third, the breadth of path coverage of the tag type in the knowledge graph, i.e., its participation range in different nursing pathways. Therefore, a first-half priority interval and a second-half priority interval are divided. The first-half priority interval represents tag types that have a greater impact on nursing pathways. Taking the stroke nursing scenario as an example, the "consciousness and communication ability" tag has a significant co-occurrence relationship with nursing solutions such as swallowing management, aspiration prevention, and communication reconstruction, and in... The "mobility status" is the initial node in most pathways and is therefore given the highest priority. Secondly, it is frequently involved in transfer care, limb function rehabilitation, and fall risk prevention programs, and is also among the top priorities. Thirdly, the "excretory function status" is closely related to pressure ulcer prevention and indwelling catheter management, and is of medium priority. The "infection risk status" is highly relevant to specific care programs, but its coverage is relatively limited. Although the "skin integrity status" is related to some pathways, its overall participation is low, and its priority is slightly lower. While labels such as "underlying disease classification," "age group," and "gender" are common, their direct impact on care pathways is relatively weak, and therefore they are ranked lower.

[0133] The first half of the priority interval, in descending order of priority, is: "awareness and communication ability", "mobility ability status", "excretory function status" and "infection risk status".

[0134] The latter half of the priority interval includes "skin integrity status", "underlying disease classification", "age group" and "gender";

[0135] Based on this, the tag types located in the first half of the priority interval are extracted as the basic information tag types to be analyzed.

[0136] By comparing standard user profiles with target user profiles, and identifying the second similarity profile type, the priority of tag types is determined based on the nursing information knowledge graph, and key tags are extracted for comparison. This can improve the accurate coverage of core nursing elements in the recommendation path and enhance the stability and relevance of the recommendation results even when the tags do not match completely.

[0137] Specifically, obtaining the comparison results and implementing corresponding measures includes:

[0138] If all the basic information tag types to be analyzed are greater than the core basic information tag similarity threshold, the first comparison result is obtained, and the tag type content of the second half priority interval is adjusted.

[0139] If the similarity of any basic information tag type to be analyzed is less than or equal to the core basic information tag similarity threshold, a second comparison result is obtained, and the similarity judgment step between the target user's basic information tag type and the standard user profile's basic information tag type is re-executed.

[0140] In this embodiment, the similarity of the basic information tag types to be analyzed is the similarity between the basic information tag node to be analyzed and the tag whose semantics are closest to it;

[0141] The similarity threshold for core basic information tags is set according to the actual application scenario and data characteristics, and is usually between 0.7 and 0.9. In this embodiment, 0.8 is preferred.

[0142] The basic information tag nodes to be analyzed are compared with the preset similarity threshold of core basic information tags.

[0143] If the similarity of the basic information tag types to be analyzed is greater than the threshold, the first comparison result is obtained. At this time, the content of the basic information tag types in the second half of the priority interval is appropriately adjusted. That is, under the current profile structure, the tag content is adjusted in order from low to high priority to adapt to the standard user profile.

[0144] If the similarity of any basic information tag type to be analyzed is less than or equal to the threshold, a second comparison result is obtained. At this time, the comparison result is incorrect, and the basic information tag node to be analyzed needs to be compared with the core basic information tag similarity threshold again.

[0145] By introducing a core basic information tag similarity threshold mechanism and implementing different adjustment strategies based on the comparison results, the matching degree of the target user profile in key tag dimensions can be effectively identified. When the similarity meets expectations, the content of low-priority tags is dynamically adjusted to adapt to the standard profile, improving the overall profile structure's capacity to support the recommendation path and semantic consistency. When the similarity does not meet expectations, the tag comparison process is re-triggered, enhancing the system's ability to respond to profile matching anomalies.

[0146] Specifically, constructing the reward function in a reinforcement learning model includes:

[0147] Obtain feedback from target users and define the categories and rating dimensions of the feedback;

[0148] The feedback information from target users is converted into rating values, and the rating values ​​are weighted and summed using a reward function before the function value is output.

[0149] The function value is compared with the standard function threshold, and the weights of the associated edges involved in the current semantic path are adjusted accordingly when the second function value is obtained and the comparison result is obtained.

[0150] In this embodiment, the reward function is,

[0151]

[0152] Where R(u,P) is the comprehensive reward value of target user u under the current path P;

[0153] ω i For the feedback dimension weights, satisfying ∑ω i =1;

[0154] f i (u) represents the user feedback score for the i-th dimension;

[0155] λ j This is the adjustment factor for the associated edges;

[0156] W j (P) represents the weight of the j-th associated edge in the path;

[0157] M j (fb(u)) is a mapping adjustment function based on user feedback fb(u), which comes from the feedback-path adjustment mapping table and is used to dynamically adjust the weight of the corresponding edge. The value range is usually (0,1], the smaller the value, the more the weight needs to be reduced, and 1 indicates no adjustment.

[0158] n represents the number of feedback dimensions;

[0159] Number of edges associated with path m;

[0160] Based on the feedback information from the target users, a pre-set feedback and path weight adjustment mapping table is invoked, and the weight of each associated edge in the recommended path is dynamically adjusted through the mapping adjustment function;

[0161] This reward function quantifies the feedback information of the target user by weighting and summing the scores of multiple feedback dimensions with their corresponding weights.

[0162] The function contains the weights of each associated edge in the path. These weights represent the importance of the relationship between each label node in the path structure. Based on user feedback, the mapping adjustment function obtains adjustment coefficients from the preset feedback-path mapping table and dynamically adjusts the weights of each associated edge. The adjustment coefficient ranges from 0 to 1. The smaller the value, the more the weight of the corresponding associated edge needs to be reduced, reflecting that the part of the path is less adaptable to the current feedback.

[0163] By quantifying the overall performance of user feedback through weighted summation, we ensure that different feedback dimensions contribute reasonably to the overall evaluation. By combining the weights of each associated edge in the path and the dynamic mapping adjustment function based on feedback, we achieve real-time optimization of the path structure, reduce the weight of path parts with poor feedback, and improve the adaptability and accuracy of recommended paths.

[0164] Please see Figure 4 As shown, it is a schematic diagram of the structure for adjusting the weights of the associated edges involved in the current semantic path according to an embodiment of the present invention;

[0165] Specifically, adjusting the weights of the associated edges involved in the current semantic path includes:

[0166] Obtain negative feedback information from target users to identify the association edges between basic information tag nodes and nursing plan nodes that significantly affect the accuracy of recommendation results in the current initial recommendation path;

[0167] Based on the negative score output in the reward function, the weights of the identified associated edges are updated according to a preset decay factor.

[0168] In this embodiment, the calculation formula for updating the weight of the associated edge according to a preset decay factor is as follows:

[0169]

[0170] in, The weight of the j-th associated edge before the update;

[0171] The updated weights;

[0172] δ is a preset attenuation factor with a value range of (0,1]. The smaller the value, the stronger the weight attenuation. In this embodiment, the value is related to the corresponding associated edge and is not unique.

[0173] By using negative feedback to guide the process, the weights of key related edges that affect recommendation accuracy are attenuated, effectively weakening the influence of unfavorable path components. Dynamically adjusting the weights continuously optimizes the recommendation path structure, improving the accuracy and personalization of recommendation results, and enhancing the system's responsiveness and adaptability to user needs.

[0174] Specifically, excluding environmental interference factors includes:

[0175] Identify the nursing preference conditions set in the feedback information of target users;

[0176] If the nursing preference conditions cannot be met under the current nursing resources, the recommended direction is determined based on the preset strategy, and the nursing plan with the highest quality or the lowest price is selected from the available paths.

[0177] Send environmental limitation information to the client regarding the recommendation results.

[0178] Specifically, sending recommendation information to the client includes:

[0179] The recommended care plans are transformed into structured data presentations and sent to target users in a graphic and text format;

[0180] The data displayed includes the reasons for recommendation, matching criteria, and key tags.

[0181] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

[0182] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for recommending nursing care plans based on customer profile data analysis, characterized in that, include: Obtain the target user information input by the client and convert the target user information into corresponding feature vector nodes; The feature vector nodes are integrated based on the multi-layer tag graph construction method to construct a target user profile; The multi-layer label graph construction method involves assigning each feature vector node to its corresponding label type through structured semantic mapping, and establishing associated edges based on the label hierarchy structure. In the multi-layered label graph, a set of candidate paths is formed by constructing several semantic paths that connect the path starting point and the nursing plan node based on semantic association, with a specific category label node as the path starting point. Based on historical sample data, nursing plans adopted by users similar to the target user profile are extracted, and the nursing plan with the highest adoption frequency is selected as the initial recommended nursing plan, and the corresponding semantic path is used as the initial recommendation path. Constructing the reward function in a reinforcement learning model; Obtain user feedback information under the initial recommended care plan and the corresponding feedback category after the reward function is applied, so as to perform corresponding optimization and adjustment on the key nodes and edge weights in the candidate path; The comprehensive score of each path after optimization and adjustment is obtained. Based on user needs and after excluding environmental interference factors, the nursing plan associated with the highest-scoring semantic path in the feedback category is selected as the recommended nursing plan. Recommendation information is generated and sent to the client. The reward function in constructing a reinforcement learning model includes: Obtain feedback from target users and define the categories and rating dimensions of the feedback; The feedback information from target users is converted into rating values, and the rating values ​​are weighted and summed using a reward function before the function value is output. The function value is compared with the standard function threshold, and the weights of the associated edges involved in the current semantic path are adjusted accordingly when the second function value comparison result is obtained. The corresponding adjustments to the weights of the associated edges involved in the current semantic path include: Obtain negative feedback information from target users to identify the association edges between basic information tag nodes and nursing plan nodes that significantly affect the accuracy of recommendation results in the current initial recommendation path; Based on the negative score value output in the reward function, the weights of the identified related edges are updated according to a preset decay factor. Among them, environmental interference factors that must be excluded include: Identify the nursing preference conditions set in the feedback information of target users; If the nursing preference conditions cannot be met under the current nursing resources, the recommended direction is determined based on the preset strategy, and the nursing plan with the highest quality or the lowest price is selected from the available paths. Send environmental limitation information to the client regarding the recommendation results.

2. The nursing plan recommendation method based on customer profile data analysis according to claim 1, characterized in that, Assigning each feature node to its corresponding label type through structured semantic mapping includes: Basic information includes direct information and supplementary information; The supplementary information is segmented based on semantic structure and punctuation marks to obtain several supplementary information word blocks; Map basic information word blocks to the same semantic vector space; Traverse each basic information word block, calculate its semantic matching degree with each tag type, and when the first matching degree result is obtained, map the feature vector node corresponding to the basic information word block to the corresponding tag type; The label types include basic information label types and nursing label types.

3. The nursing plan recommendation method based on customer profile data analysis according to claim 1, characterized in that, Establishing associated edges based on the label hierarchy includes: Determine whether the historical nursing information in the target user's information is valid; If the historical nursing information is valid, the semantic path corresponding to the most recent historical nursing plan will be used as the initial recommended path. If the historical nursing information is invalid, the label hierarchy is determined based on the nursing information knowledge graph, and each basic information label node is connected to the nursing label node to establish an association edge.

4. The nursing plan recommendation method based on customer profile data analysis according to claim 3, characterized in that, Determining the validity of a target user's historical care information includes: Retrieve the target user's historical nursing records within a preset time window; Historical nursing records include the number of historical nursing protocols and the pathways of historical nursing protocols; Determine whether there is at least one historical nursing care plan and whether the corresponding historical nursing care plan path is complete, so as to obtain a result on whether the historical nursing care information is effective.

5. The nursing plan recommendation method based on customer profile data analysis according to claim 1, characterized in that, Extracting care plans adopted by users similar to the target user profile includes: The similarity between the target user's basic information tag nodes and the standard user profile's basic information tag nodes is determined to identify the corresponding similarity profile type. When the judgment result is the second similarity profile type, the priority of the basic information type corresponding to each basic information tag node is determined based on the nursing information knowledge graph, and sorting is performed. Obtain the basic information tag types that are in the first half of the priority interval after sorting, and update them to the basic information tag types to be analyzed; The similarity thresholds of the basic information tag types to be analyzed are compared with the similarity thresholds of the core basic information tags to obtain the comparison results, and corresponding measures are implemented.

6. The nursing plan recommendation method based on customer profile data analysis according to claim 5, characterized in that, Obtaining the comparison results and implementing corresponding measures includes: If all the basic information tag types to be analyzed are greater than the core basic information tag similarity threshold, the first comparison result is obtained, and the tag type content of the second half priority interval is adjusted. If the similarity of any basic information tag type to be analyzed is less than or equal to the core basic information tag similarity threshold, a second comparison result is obtained, and the similarity judgment step between the target user's basic information tag type and the standard user profile's basic information tag type is re-executed.

7. The nursing plan recommendation method based on customer profile data analysis according to claim 1, characterized in that, Sending recommendation information to the client includes: The recommended care plans are transformed into structured data presentations and sent to target users in a graphic and text format; The data displayed includes the reasons for recommendation, matching criteria, and key tags.

Images