CN-121075534-B - Nursing scheme recommendation method based on customer portrait data analysis

CN121075534BCN 121075534 BCN121075534 BCN 121075534BCN-121075534-B

Abstract

The invention relates to the technical field of nursing service and artificial intelligence, in particular to a nursing scheme recommending method based on customer portrait data analysis. The method comprises the steps of S1, collecting user basic and historical nursing information, S2, constructing a multi-layer label map and generating a user image, S3, generating a candidate nursing path set based on semantic paths, S4, extracting nursing schemes adopted by users with similar images, S5, constructing a reward function fusing scores and structure information, S6, acquiring user feedback and dynamically adjusting path structure weights, and S7, outputting a recommended scheme with highest matching degree based on an optimization result. The method combines structured and unstructured information to construct semantic graphs, combines similar user schemes and multidimensional feedback to construct rewarding functions, dynamically optimizes path structures, improves matching and self-adaption of recommended schemes, improves intelligent level and resource adaptation efficiency of nursing recommendation through grading sorting and environment interference elimination, and improves application breadth.

Inventors

ZHANG LEI
LIU WEI
WU BO

Assignees

北京邦伴数字科技有限公司

Dates

Publication Date: 20260505
Application Date: 20250729

Claims (7)

1. A care plan recommendation method based on customer representation data analysis, comprising: acquiring target user information input by a client, and converting the target user information into corresponding feature vector nodes; integrating the feature vector nodes based on a multi-layer label graph construction mode to construct a target user portrait; The multi-layer label map construction mode is that each feature vector node is attributed to a corresponding label type through structural semantic mapping, and an associated edge is established based on a label hierarchical structure; in the multi-layer label atlas, a specific category label node is taken as a path starting point, a plurality of semantic paths with the path starting point connected with the care scheme node are constructed based on semantic association, and a candidate path set is formed; Extracting a nursing scheme adopted by a user similar to the target user portrait based on historical sample data, selecting the nursing scheme with highest adoption frequency as an initial recommended nursing scheme, and taking the corresponding semantic path as an initial recommended path; constructing a reward function in the reinforcement learning model; acquiring feedback information of a user under an initial recommended nursing scheme and a feedback category corresponding to the feedback information after the feedback information is operated by the reward function, so as to execute corresponding optimization adjustment on key nodes and edge weights in candidate paths; Acquiring the comprehensive score of each path after optimization and adjustment, selecting a nursing scheme associated with a semantic path with the highest score under the category of the feedback type as a recommended nursing scheme based on the user requirement and after the environmental interference factor is eliminated, forming recommended information and sending the recommended information to the client; Constructing the reward function in the reinforcement learning model includes: acquiring feedback information of a target user, and defining the category and grading dimension of the feedback information; converting feedback information of the target user into a scoring value, weighting the scoring value by a reward function, and outputting a function value; comparing the function value with a standard function threshold value, and correspondingly adjusting the weight of the related edge involved in the current semantic path when a second function value comparison result is obtained; wherein the corresponding adjustment of the weights of the associated edges involved in the current semantic path comprises: Acquiring negative feedback information of a target user to identify an associated edge between a basic information label node and a nursing scheme node which have obvious influence on the accuracy of a recommendation result in a current initial recommendation path; According to the negative scoring value output in the reward function, carrying out numerical value updating processing on the weight of the correlation edge which completes recognition according to a preset attenuation factor; Wherein, excluding the environmental interference factor includes: Identifying the nursing preference conditions set in the feedback information of the target user; if the nursing preference condition cannot be met under the current nursing resource condition, determining a preferentially selected recommended direction based on a preset strategy, and selecting a nursing scheme with highest quality or lowest price from the selectable paths; and sending the environment limit description information of the recommended result to the client.
2. The customer care plan recommendation method based on customer representation data analysis according to claim 1, wherein attributing feature nodes to corresponding label types through structured semantic mapping comprises: the basic information comprises direct information and supplementary information; Dividing the supplementary information based on the semantic structure and punctuation marks to obtain a plurality of supplementary information word blocks; mapping the basic information word blocks to the same semantic vector space; Traversing each basic information word block, calculating semantic matching degree between each basic information word block and each label type, and mapping and attributing the feature vector node corresponding to the basic information word block to the corresponding label type when a first matching degree result is obtained; The label types comprise basic information label types and nursing label types.
3. The customer care regimen recommendation method based on customer representation data analysis of claim 1, wherein establishing an association edge based on a label hierarchy comprises: judging whether the historical nursing information in the target user information is effective or not; If the history nursing information is effective, taking a semantic path corresponding to the latest history nursing scheme as an initial recommended path; if the historical nursing information is invalid, determining a label hierarchical structure based on the nursing information knowledge graph, and connecting each basic information label node with the nursing label node to establish an associated edge.
4. The method for recommending care plan based on analysis of customer image data as claimed in claim 3, wherein determining whether the historical care information of the target user is valid comprises: Acquiring a historical nursing record of a target user in a preset time window; The historical care record comprises historical care scheme times and historical care scheme paths; judging whether at least one historical care scheme exists or not, and judging whether the historical care information is effective or not according to the fact that the corresponding historical care scheme is complete in path.
5. The method of claim 1, wherein extracting the care plan taken by the user similar to the target user profile comprises: performing similarity judgment on the basic information label node of the target user and the basic information label node of the standard user image to determine the type of the similarity image; When the judgment result is the second similarity portrait type, determining the priority of the basic information type corresponding to each basic information label node based on the nursing information knowledge graph, and performing sorting processing; acquiring basic information label types in a first half priority interval after sequencing treatment, and updating the basic information label types into basic information label types to be analyzed; And comparing the types of the basic information labels to be analyzed with the similarity threshold value of the core basic information labels to obtain a comparison result, and executing corresponding measures.
6. The method of claim 5, wherein obtaining a comparison result and performing a corresponding measure comprises: If the types of the basic information labels to be analyzed are all larger than the core basic information label similarity threshold, a first comparison result is obtained, and the label type content of the second half-priority interval is adjusted; If the similarity of any basic information label type to be analyzed is smaller than or equal to the core basic information label similarity threshold value, a second comparison result is obtained, and the similarity judging step of the target user basic information label type and the standard user portrait basic information label type is re-executed.
7. The care plan recommendation method based on customer image data analysis according to claim 1, the method is characterized in that the step of sending the recommendation information to the client comprises the following steps: converting the recommended nursing scheme into structured display data, and sending the structured display data to a target user in a graphic form; the display data comprises recommendation reasons, matching bases and key tag items.

Description

Nursing scheme recommendation method based on customer portrait data analysis Technical Field The invention relates to the technical field of nursing service and artificial intelligence, in particular to a nursing scheme recommending method based on customer portrait data analysis. Background With the acceleration of population aging and the diversification of medical service demands, the importance of nursing services is increasingly highlighted. The traditional nursing scheme is formulated by relying on the experience and fixed rules of nursing staff, lacks systematic and intelligent support, and is difficult to meet the personalized and changeable nursing demands of different users. In recent years, artificial intelligence technology, particularly recommendation systems based on user portraits, have gained widespread attention in the medical health field. Through deep mining of user basic information, behavior data and historical nursing records, personalized nursing scheme recommendation becomes a research hotspot. The existing nursing recommendation method generally adopts rule-based matching or simple similarity calculation, and has the problems of single dimension of the user portrait, lack of dynamic adjustment of recommendation results, insufficient resource matching and the like. For example, some published patents and related documents propose recommendation of solutions based on user basic information and care records, but do not fully consider dynamic optimization of user feedback and constraint limits of care resources, resulting in insufficient recommendation accuracy and practicality. Meanwhile, most of recommendation processes are black box operation, and necessary interpretation is lacked, so that trust and acceptance of a user on recommendation results are affected. In addition, the nursing scene is complex and changeable, dynamic changes exist in user demands and resource supply, the traditional static recommendation model is difficult to effectively cope with, and advanced technologies such as multi-layer label map construction, reinforcement learning and the like are required to be introduced, so that multidimensional information fusion, dynamic path optimization and intelligent resource regulation are realized. By constructing the structured semantic tag map and combining a reward mechanism driven by real-time feedback of a user, individuation, accuracy and practicality of nursing scheme recommendation can be remarkably improved, and diversified requirements of actual nursing services are met. Therefore, developing an intelligent care scheme recommendation method based on customer portrait data analysis and fusion reinforcement learning feedback drive has important technical significance and application value. Chinese patent publication No. CN115835799a discloses an oral care device recommendation system for recommending the type of oral care attachments used with an oral care device. It is recommended to consider information about the oral geometry of the user and preferably also user behavior information about the way a particular user is taking his oral care with the oral care device. The recommendation is based on modeling of mechanical interactions between one or more of a set of oral care attachments and the oral geometry of the user when the user performs an oral care routine. A cleaning metric is determined from the modeling, the cleaning metric representing the effectiveness of an oral care routine when the one or more oral care attachments are used. At the same time, however, the prior art is primarily directed to matching recommendations for oral care attachments, generating cleaning metrics and recommending attachment types by modeling user oral geometry information and brushing behavior. In contrast, the recommendation dimension is single, limited to physical structure matching and static behavior analysis, lacks a dynamic optimization mechanism driven by multidimensional label map modeling and user feedback, and does not relate to nursing resource availability and environmental interference elimination, so that cross-scene, personalized and explanatory intelligent nursing scheme recommendation is difficult to realize. Disclosure of Invention Therefore, the invention provides a nursing scheme recommending method based on customer portrait data analysis, which is used for solving the problems that in the prior art, under a complex nursing environment, a recommending execution mode of a nursing mode is unstable and a reproducible path is single, so that a recommending scheme is difficult to adapt to diversified nursing requirements. In order to achieve the above object, the present invention provides a care plan recommendation method based on customer portrait data analysis, including: acquiring target user information input by a client, and converting the target user information into corresponding feature vector nodes; integrating the feature vector nodes based on a multi-