Traditional Chinese medicine nursing crowd health management system based on big data

Abstract

The invention discloses a traditional Chinese medicine nursing crowd health management system based on big data, and belongs to the technical field of intelligent traditional Chinese medicine, and the system comprises a data preparation module, a data preprocessing module, a knowledge graph construction module, a patient constitution recognition module and a health management suggestion generation module. According to the invention, a knowledge graph is combined with patient physique identification to generate an overall technical process of health management suggestion generation, and deep fusion of a traditional Chinese medicine theory and a modern data technology is realized; according to the method, an entity extraction method of an improved transformer model is adopted, then knowledge graph construction is carried out, traditional Chinese medicine text semantic fuzziness is better processed, and knowledge graph construction quality is improved; the two-way long-short-term memory model combined with the switching feed-forward network is adopted for patient physique recognition, semantic information is deeply understood, good time sequence modeling performance is achieved, the adaptability of the model to traditional Chinese medicine complex semantics and features is enhanced, the robustness and accuracy of physique recognition are improved, and the accuracy of physique recognition is improved. And more scientific support is provided for personalized services of traditional Chinese medicine nursing.

Description

Technical Field

[0001] The present invention belongs to the field of smart Chinese medicine technology, and specifically refers to a health management system for Chinese medicine nursing population based on big data. Background Art

[0002] Big data-based health management of the TCM nursing population uses big data and artificial intelligence technology to analyze the health data of the TCM nursing population and data related to TCM nursing knowledge, and provide patients with precise health management suggestions. It aims to combine TCM theory with modern information technology, break the limitations of traditional TCM nursing experience, and promote the intelligent development of TCM nursing.

[0003] However, among the existing methods of population health management in TCM nursing, traditional TCM nursing mainly focuses on manual population health management, while manual methods are highly subjective and lack standardization. At the same time, existing intelligent technology solutions have technical problems such as insufficient integration and utilization of TCM nursing knowledge data; in the existing knowledge graph construction process in the field of TCM nursing, there are technical problems such as the unstructured and semantic ambiguity of TCM knowledge, the lack of high-quality standardized data, and insufficient accuracy of entity extraction; in the existing patient constitution identification process in the field of TCM nursing, the semantic complexity of TCM constitution description presents a high dimensionality and nonlinearity of data characteristics, and the traditional intelligent methods have insufficient technical problems in capturing the characteristics of patient health data and TCM questionnaire data. Summary of the invention

[0004] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides a TCM nursing population health management system based on big data. In view of the fact that in the existing TCM nursing population health management methods, traditional TCM nursing mainly focuses on manual population health management, and the manual method is highly subjective and lacks standardization. At the same time, the existing intelligent technical solutions have the technical problem of insufficient integration and utilization of TCM nursing knowledge data. This solution creatively adopts the knowledge graph combined with patient constitution identification to generate health management suggestions. The overall technical process, by integrating patient health data and TCM questionnaire data and combining machine learning models to identify patient constitutions, realizes the deep integration of traditional TCM theory and modern data technology, and improves the scientificity and accuracy of constitution identification. The entire process, from data preparation to health management suggestion generation, constitutes a complete closed-loop system, which fully reflects the innovation and practicality of data-driven TCM nursing management methods, and provides new ideas and new models for TCM health management. In view of the fact that in the existing TCM nursing field knowledge graph construction process, there are unstructured and semantic models of TCM knowledge. In order to solve the technical problems of fuzziness, lack of high-quality standardized data and insufficient entity extraction accuracy, this solution creatively adopts the entity extraction method of the improved transformer model to construct a knowledge graph, so as to better deal with the semantic ambiguity of TCM text and improve the quality of knowledge graph construction, laying a solid foundation for the subsequent intelligent application of patient constitution identification; in the existing TCM nursing field, the semantic complexity of TCM constitution description presents a high dimensionality and nonlinearity of data characteristics, and the traditional intelligent methods are insufficient to capture the characteristics of patient health data and TCM questionnaire data. This solution creatively adopts a bidirectional long short-term memory model combined with a switching feedforward network for patient constitution identification, which deeply understands semantic information and has good time series modeling performance, enhances the ability to extract constitution features, and at the same time, the switching feedforward network layer enhances the adaptability of the model to the complex semantics and features of TCM through dynamic branching mechanism and feature comprehensive processing, realizes personalized and accurate classification, thereby improving the robustness and accuracy of constitution identification, and providing more scientific support for personalized TCM nursing services.

[0005] The TCM nursing population health management system based on big data provided by the present invention includes: a data preparation module, a data preprocessing module, a knowledge graph construction module, a patient constitution identification module and a health management suggestion generation module;

[0006] The data preparation module is used to prepare the original data required for the health management of the TCM nursing population, specifically to collect patient health data and TCM nursing knowledge data, and send the patient health data and the TCM nursing knowledge data to the data preprocessing module;

[0007] The data preprocessing module performs data preprocessing on the patient's health data and the TCM nursing knowledge data to obtain the patient's health standard data and the TCM nursing knowledge standard data, and sends the patient's health standard data to the patient's constitution identification module, and sends the TCM nursing knowledge standard data to the knowledge graph construction module;

[0008] The knowledge graph construction module is used to construct a TCM nursing knowledge graph, specifically to construct a knowledge graph based on TCM nursing knowledge standard data to obtain a TCM nursing knowledge graph, and send the TCM nursing knowledge graph to the health management advice generation module;

[0009] The patient constitution identification module is used to identify the patient's constitution, specifically, based on the patient's health standard data, using a bidirectional long short-term memory model combined with a switching feedforward network to identify the patient's constitution, obtain the patient's constitution type, and send the patient's constitution type to the health management suggestion generation module;

[0010] The health management suggestion generation module is used for health management of the TCM nursing population, specifically generating health management suggestions based on the patient's physical constitution type and the TCM nursing knowledge graph.

[0011] Further, in the data preparation module, the patient health data includes TCM questionnaire data and medical record data, the TCM questionnaire data is obtained by the patient filling in the TCM body mass table, the medical record data is obtained from the medical system, and the TCM nursing knowledge data is obtained from the TCM database;

[0012] The medical record data includes past medical history and medication use records;

[0013] The TCM nursing knowledge data includes TCM theory data, TCM literature data, TCM nursing practice data and TCM data;

[0014] The TCM nursing practice data includes TCM nursing plans, nursing processes and precautions.

[0015] Furthermore, in the data preprocessing module, the data preprocessing includes data cleaning and data standardization processing, and the data cleaning includes invalid value elimination, outlier filtering and missing value filling.

[0016] Furthermore, in the knowledge graph construction module, there are an entity extraction unit, a relationship extraction unit, an entity alignment unit and a TCM nursing knowledge graph construction unit, which specifically include the following contents:

[0017] An entity extraction unit is used to extract key entities in the TCM nursing knowledge standard data, specifically using an improved transformer model to extract entities and obtain TCM nursing knowledge entities;

[0018] The improved transformer model includes a BERT embedding layer, a context encoding improvement layer, a feature decoding layer and an entity output layer;

[0019] The entity extraction unit includes the following contents:

[0020] Constructing a BERT embedding layer, specifically, in the BERT embedding layer, using a pre-trained BERT model to perform word embedding processing on the TCM nursing knowledge standard data to obtain TCM nursing knowledge word embedding features;

[0021] Construct a context coding improvement layer, specifically, use the encoder as the basic architecture of the context coding improvement layer, introduce relative position coding and multi-head self-attention mechanism into the context coding improvement layer, process the embedded features of TCM nursing knowledge through the context coding improvement layer, and obtain the context features of TCM nursing knowledge;

[0022] Constructing a feature decoding layer, specifically, by constructing a conditional random field scoring function, constructing a feature decoding layer objective function and calculating an optimal label sequence to obtain a feature decoding layer, and processing the context features of traditional Chinese medicine nursing knowledge through the feature decoding layer to obtain an optimal label sequence;

[0023] The calculation formula for constructing the conditional random field scoring function is:

[0024] ;

[0025] Where S(·) is the conditional random field scoring function, h is the TCM nursing knowledge context feature, y is the TCM nursing knowledge feature label, a is the TCM nursing knowledge context feature index, n is the number of TCM nursing knowledge context features, is the transfer weight, specifically the weight of the a-1th TCM nursing knowledge feature label transferred to the ath TCM nursing knowledge feature label, y a-1 is the a-1th TCM nursing knowledge feature label, specifically the label corresponding to the a-1th TCM nursing knowledge context feature, y a is the ath TCM nursing knowledge feature label, is the matching score, specifically the matching score between the a-th TCM nursing knowledge context feature and the a-th TCM nursing knowledge feature label, h a It is the ath TCM nursing knowledge context feature;

[0026] The calculation formula for constructing the feature decoding layer objective function is:

[0027] ;

[0028] In the formula, is the optimal label, used to represent the optimal TCM nursing knowledge feature label, argmax is to find the maximum value, used to find the TCM nursing knowledge feature label that maximizes the conditional probability, P(·) is the conditional probability function, exp(·) is the exponential function, is a candidate TCM nursing knowledge feature label, Y is a TCM nursing knowledge feature label sequence;

[0029] The calculating of the optimal label sequence specifically comprises calculating the optimal label sequence using the Viterbi algorithm according to the feature decoding layer objective function;

[0030] Constructing an entity output layer, specifically, performing entity mapping on the optimal label sequence through the entity output layer to obtain the TCM nursing knowledge entity;

[0031] The relationship extraction unit is used to extract the semantic relationship between entities, specifically, based on the TCM nursing knowledge entity and TCM nursing knowledge standard data, a bidirectional long-term and short-term relationship extraction model is used to extract the relationship, and the TCM nursing knowledge entity relationship is obtained;

[0032] The bidirectional long-short-term relationship extraction model includes a BERT embedding layer, a bidirectional long-short-term memory layer, an attention pooling layer and an output layer; the attention pooling layer includes an attention layer and a weighted pooling layer;

[0033] The entity alignment unit is used to merge similar entities. Specifically, based on the semantic similarity of entity pairs, the TCM nursing knowledge entities are aligned to obtain TCM nursing knowledge aligned entities, including the following contents:

[0034] Calculate the semantic similarity of entity pairs, the calculation formula is:

[0035] ;

[0036] In the formula, Sim uz is the semantic similarity of entity pairs, specifically refers to the semantic similarity between the u-th TCM nursing knowledge entity and the z-th TCM nursing knowledge entity, u is the first index of the TCM nursing knowledge entity, z is the second index of the TCM nursing knowledge entity, the second index of the TCM nursing knowledge entity is not equal to the first index of the TCM nursing knowledge entity, sim(·) is the similarity calculation function used to calculate the semantic similarity of entity pairs, ent u is the vector representation of the u-th TCM nursing knowledge entity, ent z is the vector representation of the zth TCM nursing knowledge entity, is the norm symbol;

[0037] Set an alignment threshold. When the semantic similarity of an entity pair exceeds the alignment threshold, align the corresponding two TCM nursing knowledge entities.

[0038] The TCM nursing knowledge graph construction unit constructs the TCM nursing knowledge graph based on the TCM nursing knowledge alignment entities and TCM nursing knowledge entity relationships.

[0039] Further, in the patient constitution recognition module, the bidirectional long short-term memory model combined with the switched feedforward network includes a BERT embedding layer, a self-attention layer, a bidirectional long short-term memory layer, a switched feedforward network layer and a classification layer;

[0040] The patient constitution recognition module is provided with a BERT embedding layer construction unit, a self-attention layer construction unit, a bidirectional long short-term memory layer construction unit, a switching feedforward network layer construction unit, a classification layer construction unit, a model training unit and a patient constitution type calculation unit, and specifically includes the following contents:

[0041] A BERT embedding layer construction unit, specifically, in the BERT embedding layer, a pre-trained BERT model is used to perform word embedding processing on the patient health standard data to obtain the patient health word embedding features;

[0042] The self-attention layer construction unit is used to capture the global dependency between words. Specifically, the self-attention mechanism is used to process the patient health word embedding features to obtain the patient health global features;

[0043] The bidirectional long short-term memory layer construction unit is used to capture the deep semantics of the context. Specifically, the forward long short-term memory unit and the backward long short-term memory unit are set in the bidirectional long short-term memory layer. The patient's global health features are processed through the bidirectional long short-term memory layer to obtain the patient's health semantic features. The calculation formula is:

[0044] ;

[0045] In the formula, H t is the patient health semantic feature at time step t, t is the time step index, concat(·) is the feature concatenation operation, LSTM fw (·) is the forward long short-term memory unit, is the global health characteristic of the patient at time step t, is the forward hidden state at time step t-1, is the forward memory state of the t-1th time step, LSTM bw (·) is the backward long short-term memory unit, is the backward hidden state at the t-1th time step, is the backward memory state at the t-1th time step;

[0046] The switch feed-forward network layer building unit is used to dynamically adapt to the patient's health semantic features, including the following:

[0047] Constructing a multi-branch feedforward neural network layer, specifically setting M parallel feedforward neural network layers, each feedforward neural network layer independently learns the nonlinear features in the patient's health semantic features;

[0048] The feedforward neural network layer includes two fully connected layers, and the calculation formula is:

[0049] ;

[0050] In the formula, FFN i (·) is the i-th feedforward neural network layer, i is the feedforward neural network layer index, H is the patient health semantic feature, GELU(·) is the GELU activation function, ReLU(·) is the ReLU activation function, W 2i is the weight of the second fully connected layer in the i-th feedforward neural network layer, W 1i is the weight of the first fully connected layer in the i-th feedforward neural network layer, b 1i is the first fully connected layer bias term in the i-th feedforward neural network layer, b 2i is the bias term of the second fully connected layer in the i-th feedforward neural network layer;

[0051] Setting a router mechanism, specifically using a router to dynamically allocate the weight of each feedforward neural network layer according to the patient's health semantic features to obtain branch weights;

[0052] The comprehensive multi-branch output is used to integrate the output results of the multi-branch feedforward neural network. Specifically, according to the branch weight, the output features of each feedforward neural network are weighted and summed to obtain the comprehensive health features of the patient. The calculation formula is:

[0053] ;

[0054] In the formula, O FFN is the comprehensive health characteristics of the patient, M is the number of feedforward neural network layers, and p i is the branch weight, specifically the branch weight of the i-th feedforward neural network layer;

[0055] Output feature normalization, specifically, connecting the patient's comprehensive health features and the patient's health semantic features through residuals, and performing normalization processing to obtain the patient's comprehensive health normative features;

[0056] The classification layer construction unit specifically processes the comprehensive normative features of patient health through the softmax activation function and the fully connected layer to obtain the model classification results;

[0057] The model training unit is specifically configured to construct a bidirectional long short-term memory model combined with a switching feedforward network through the BERT embedding layer construction unit, the self-attention layer construction unit, the bidirectional long short-term memory layer construction unit, the switching feedforward network layer construction unit and the classification layer construction unit, and perform model training to obtain a patient constitution recognition model;

[0058] The patient's physical constitution type calculation unit specifically identifies the patient's physical constitution through a patient's physical constitution identification model to obtain the patient's physical constitution type.

[0059] Further, in the health management suggestion generation module, the health management suggestion is generated by searching the health management suggestion corresponding to the patient's physical constitution type from the TCM nursing knowledge graph according to the patient's physical constitution type to obtain the health management suggestion;

[0060] The patient's constitution types include mild constitution, qi deficiency constitution, yang deficiency constitution, yin deficiency constitution, phlegm-damp constitution, damp-heat constitution, blood stasis constitution, qi stagnation constitution and special constitution;

[0061] The health management recommendations include dietary recommendations, exercise recommendations, medicated diet recommendations and massage treatments.

[0062] The beneficial effects achieved by the present invention using the above scheme are as follows:

[0063] (1) In the existing TCM nursing population health management methods, traditional TCM nursing mainly focuses on manual population health management, which is highly subjective and lacks standardization. At the same time, the existing intelligent technology solutions have technical problems such as insufficient integration and utilization of TCM nursing knowledge data. This solution creatively adopts the overall technical process of combining knowledge graph with patient constitution identification to generate health management suggestions. By integrating patient health data and TCM questionnaire data and combining machine learning models to identify patient constitutions, it realizes the deep integration of traditional TCM theory and modern data technology, and improves the scientificity and accuracy of constitution identification. The entire process, from data preparation to health management suggestion generation, constitutes a complete closed-loop system, which fully reflects the innovation and practicality of data-driven TCM nursing management methods and provides new ideas and new models for TCM health management.

[0064] (2) In view of the technical problems in the construction of the existing knowledge graph in the field of traditional Chinese medicine nursing, such as the unstructured and semantic ambiguity of traditional Chinese medicine knowledge, the lack of high-quality standardized data, and the insufficient accuracy of entity extraction, this solution creatively adopts the entity extraction method of the improved transformer model to construct the knowledge graph, so as to better handle the semantic ambiguity of traditional Chinese medicine text and improve the quality of knowledge graph construction, laying a solid foundation for the subsequent intelligent application of patient constitution identification.

[0065] (3) In the existing process of patient constitution identification in the field of traditional Chinese medicine nursing, there is a problem that the semantic complexity of TCM constitution description presents a high dimensionality and nonlinearity of data characteristics, and the traditional intelligent methods are insufficient to capture the characteristics of patient health data and TCM questionnaire data. This solution creatively adopts a bidirectional long short-term memory model combined with a switching feedforward network to identify patient constitutions. It deeply understands semantic information and has good time series modeling performance, which enhances the ability to extract constitution features. At the same time, the switching feedforward network layer enhances the model's adaptability to complex TCM semantics and features through a dynamic branching mechanism and feature comprehensive processing, achieving personalized and accurate classification, thereby improving the robustness and accuracy of constitution identification and providing more scientific support for personalized TCM nursing services. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] Figure 1 A schematic diagram of the structure of the TCM nursing population health management system based on big data provided by the present invention;

[0067] Figure 2 Schematic diagram of building modules for knowledge graph;

[0068] Figure 3 It is a flow chart of the entity extraction unit;

[0069] Figure 4 This is a flow chart of the patient constitution identification module.

[0070] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0071] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0072] In the description of the present invention, it should be understood that terms such as “upper”, “lower”, “front”, “back”, “left”, “right”, “top”, “bottom”, “inside” and “outside” indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore should not be understood as limiting the present invention.

[0073] Example 1, see Figure 1The TCM nursing population health management system based on big data provided by the present invention includes: a data preparation module, a data preprocessing module, a knowledge graph construction module, a patient constitution identification module and a health management suggestion generation module;

[0074] The data preparation module is used to prepare the original data required for the health management of the TCM nursing population, specifically to collect patient health data and TCM nursing knowledge data, and send the patient health data and the TCM nursing knowledge data to the data preprocessing module;

[0075] The data preprocessing module performs data preprocessing on the patient's health data and the TCM nursing knowledge data to obtain the patient's health standard data and the TCM nursing knowledge standard data, and sends the patient's health standard data to the patient's constitution identification module, and sends the TCM nursing knowledge standard data to the knowledge graph construction module;

[0076] The knowledge graph construction module is used to construct a TCM nursing knowledge graph, specifically to construct a knowledge graph based on TCM nursing knowledge standard data to obtain a TCM nursing knowledge graph, and send the TCM nursing knowledge graph to the health management advice generation module;

[0077] The patient constitution identification module is used to identify the patient's constitution, specifically, based on the patient's health standard data, using a bidirectional long short-term memory model combined with a switching feedforward network to identify the patient's constitution, obtain the patient's constitution type, and send the patient's constitution type to the health management suggestion generation module;

[0078] The health management suggestion generation module is used for health management of TCM nursing population, specifically generating health management suggestions based on the patient's physical type and TCM nursing knowledge graph;

[0079] By performing the above operations, it is found that in the existing TCM nursing population health management methods, traditional TCM nursing mainly focuses on manual population health management, which is highly subjective and lacks standardization. At the same time, the existing intelligent technology solutions have technical problems such as insufficient integration and utilization of TCM nursing knowledge data. This solution creatively adopts the knowledge graph combined with patient constitution identification to generate health management suggestions. The overall technical process, which integrates patient health data and TCM questionnaire data and combines machine learning models to identify patient constitutions, realizes the deep integration of traditional TCM theory and modern data technology, and improves the scientificity and accuracy of constitution identification. The entire process, from data preparation to health management suggestion generation, constitutes a complete closed-loop system, which fully reflects the innovation and practicality of data-driven TCM nursing management methods, and provides new ideas and new models for TCM health management.

[0080] Example 2, see Figure 1This embodiment is based on the above embodiment. In the data preparation module, the patient health data includes TCM questionnaire data and medical record data. The TCM questionnaire data is obtained by the patient filling in the TCM body mass table, the medical record data is obtained from the medical system, and the TCM nursing knowledge data is obtained from the TCM database;

[0081] The medical record data includes past medical history and medication use records;

[0082] The TCM nursing knowledge data includes TCM theory data, TCM literature data, TCM nursing practice data and TCM data;

[0083] The TCM nursing practice data includes TCM nursing plans, nursing processes and precautions.

[0084] Example 3, see Figure 1 This embodiment is based on the above embodiment. In the data preprocessing module, the data preprocessing includes data cleaning and data standardization processing. The data cleaning includes invalid value elimination, outlier filtering and missing value filling.

[0085] Example 4, see Figure 1 , Figure 2 and Figure 3 This embodiment is based on the above embodiment. In the knowledge graph construction module, an entity extraction unit, a relationship extraction unit, an entity alignment unit and a TCM nursing knowledge graph construction unit are provided, which specifically include the following contents:

[0086] An entity extraction unit is used to extract key entities in the TCM nursing knowledge standard data, specifically using an improved transformer model to extract entities and obtain TCM nursing knowledge entities;

[0087] The improved transformer model includes a BERT embedding layer, a context encoding improvement layer, a feature decoding layer and an entity output layer;

[0088] The entity extraction unit includes the following contents:

[0089] Constructing a BERT embedding layer, specifically, in the BERT embedding layer, using a pre-trained BERT model to perform word embedding processing on the TCM nursing knowledge standard data to obtain TCM nursing knowledge word embedding features;

[0090] Construct a context coding improvement layer, specifically, use the encoder as the basic architecture of the context coding improvement layer, introduce relative position coding and multi-head self-attention mechanism into the context coding improvement layer, process the embedded features of TCM nursing knowledge through the context coding improvement layer, and obtain the context features of TCM nursing knowledge;

[0091] The calculation formula of the multi-head self-attention mechanism is:

[0092] ;

[0093] Where MultiHead(·) is the multi-head attention mechanism, Q is the query matrix, K is the key matrix, V is the value matrix, concat(·) is the feature concatenation operation, head1 is the calculation result of the first self-attention head, head2 is the calculation result of the second self-attention head, and head A is the calculation result of the Ath self-attention head, A is the number of self-attention heads, W O is the multi-head self-attention output weight;

[0094] Constructing a feature decoding layer, specifically, by constructing a conditional random field scoring function, constructing a feature decoding layer objective function and calculating an optimal label sequence to obtain a feature decoding layer, and processing the context features of traditional Chinese medicine nursing knowledge through the feature decoding layer to obtain an optimal label sequence;

[0095] The calculation formula for constructing the conditional random field scoring function is:

[0096] ;

[0097] Where S(·) is the conditional random field scoring function, h is the TCM nursing knowledge context feature, y is the TCM nursing knowledge feature label, a is the TCM nursing knowledge context feature index, n is the number of TCM nursing knowledge context features, is the transfer weight, specifically the weight of the a-1th TCM nursing knowledge feature label transferred to the ath TCM nursing knowledge feature label, y a-1 is the a-1th TCM nursing knowledge feature label, specifically the label corresponding to the a-1th TCM nursing knowledge context feature, y a is the ath TCM nursing knowledge feature label, is the matching score, specifically the matching score between the a-th TCM nursing knowledge context feature and the a-th TCM nursing knowledge feature label, h a It is the ath TCM nursing knowledge context feature;

[0098] The calculation formula for constructing the feature decoding layer objective function is:

[0099] ;

[0100] In the formula, is the optimal label, used to represent the optimal TCM nursing knowledge feature label, argmax is to find the maximum value, used to find the TCM nursing knowledge feature label that maximizes the conditional probability, P(·) is the conditional probability function, exp(·) is the exponential function, is a candidate TCM nursing knowledge feature label, Y is a TCM nursing knowledge feature label sequence;

[0101] The calculating of the optimal label sequence specifically comprises calculating the optimal label sequence using the Viterbi algorithm according to the feature decoding layer objective function;

[0102] Constructing an entity output layer, specifically, performing entity mapping on the optimal label sequence through the entity output layer to obtain the TCM nursing knowledge entity;

[0103] The relationship extraction unit is used to extract the semantic relationship between entities, specifically, based on the TCM nursing knowledge entity and TCM nursing knowledge standard data, a bidirectional long-term and short-term relationship extraction model is used to extract the relationship, and the TCM nursing knowledge entity relationship is obtained;

[0104] The bidirectional long-short-term relationship extraction model includes a BERT embedding layer, a bidirectional long-short-term memory layer, an attention pooling layer and an output layer; the attention pooling layer includes an attention layer and a weighted pooling layer;

[0105] The entity alignment unit is used to merge similar entities. Specifically, based on the semantic similarity of entity pairs, the TCM nursing knowledge entities are aligned to obtain TCM nursing knowledge aligned entities, including the following contents:

[0106] Calculate the semantic similarity of entity pairs, the calculation formula is:

[0107] ;

[0108] In the formula, Sim uz is the semantic similarity of entity pairs, specifically refers to the semantic similarity between the u-th TCM nursing knowledge entity and the z-th TCM nursing knowledge entity, u is the first index of the TCM nursing knowledge entity, z is the second index of the TCM nursing knowledge entity, the second index of the TCM nursing knowledge entity is not equal to the first index of the TCM nursing knowledge entity, sim(·) is the similarity calculation function used to calculate the semantic similarity of entity pairs, ent u is the vector representation of the u-th TCM nursing knowledge entity, ent z is the vector representation of the zth TCM nursing knowledge entity, is the norm symbol;

[0109] An alignment threshold is set. When the semantic similarity of an entity pair exceeds the alignment threshold, the corresponding two TCM nursing knowledge entities are aligned. The alignment threshold is 0.9.

[0110] The TCM nursing knowledge graph construction unit constructs the TCM nursing knowledge graph based on the TCM nursing knowledge alignment entities and TCM nursing knowledge entity relationships;

[0111] By performing the above operations, this solution creatively adopts the entity extraction method of the improved transformer model to construct the knowledge graph, aiming at the technical problems of the unstructured and semantic ambiguity of TCM knowledge, the lack of high-quality standardized data, and the insufficient accuracy of entity extraction in the existing knowledge graph construction process in the field of TCM nursing. This can better handle the semantic ambiguity of TCM text and improve the quality of knowledge graph construction, laying a solid foundation for the subsequent intelligent application of patient constitution identification.

[0112] Example 5, see Figure 1 and Figure 4 , this embodiment is based on the above embodiment, in the patient constitution recognition module, the bidirectional long short-term memory model combined with the switching feedforward network includes a BERT embedding layer, a self-attention layer, a bidirectional long short-term memory layer, a switching feedforward network layer and a classification layer;

[0113] The patient constitution recognition module is provided with a BERT embedding layer construction unit, a self-attention layer construction unit, a bidirectional long short-term memory layer construction unit, a switching feedforward network layer construction unit, a classification layer construction unit, a model training unit and a patient constitution type calculation unit, and specifically includes the following contents:

[0114] A BERT embedding layer construction unit, specifically, in the BERT embedding layer, a pre-trained BERT model is used to perform word embedding processing on the patient health standard data to obtain the patient health word embedding features;

[0115] The self-attention layer construction unit is used to capture the global dependency between words. Specifically, the self-attention mechanism is used to process the patient health word embedding features to obtain the patient health global features;

[0116] The self-attention mechanism is used to process the patient health word embedding features to obtain the calculation formula of the patient health global feature:

[0117] ;

[0118] In the formula, H att is the global feature of patient health, Attention(·) is the self-attention mechanism, Q is the query matrix, K is the key matrix, V is the value matrix, Softmax(·) is the Softmax activation function, T is the transpose operation, and d k is the key vector dimension;

[0119] The bidirectional long short-term memory layer construction unit is used to capture the deep semantics of the context. Specifically, the forward long short-term memory unit and the backward long short-term memory unit are set in the bidirectional long short-term memory layer. The patient's global health features are processed through the bidirectional long short-term memory layer to obtain the patient's health semantic features. The calculation formula is:

[0120] ;

[0121] In the formula, H t is the patient health semantic feature at time step t, t is the time step index, concat(·) is the feature concatenation operation, LSTM fw (·) is the forward long short-term memory unit, is the global health characteristic of the patient at time step t, is the forward hidden state at time step t-1, is the forward memory state of the t-1th time step, LSTM bw (·) is the backward long short-term memory unit, is the backward hidden state at the t-1th time step, is the backward memory state at the t-1th time step;

[0122] The switch feed-forward network layer building unit is used to dynamically adapt to the patient's health semantic features, including the following:

[0123] Constructing a multi-branch feedforward neural network layer, specifically setting M parallel feedforward neural network layers, each feedforward neural network layer independently learns the nonlinear features in the patient's health semantic features;

[0124] The feedforward neural network layer includes two fully connected layers, and the calculation formula is:

[0125] ;

[0126] In the formula, FFN i (·) is the i-th feedforward neural network layer, i is the feedforward neural network layer index, H is the patient health semantic feature, GELU(·) is the GELU activation function, ReLU(·) is the ReLU activation function, W 2i is the weight of the second fully connected layer in the i-th feedforward neural network layer, W 1i is the weight of the first fully connected layer in the i-th feedforward neural network layer, b 1i is the first fully connected layer bias term in the i-th feedforward neural network layer, b 2i is the bias term of the second fully connected layer in the i-th feedforward neural network layer;

[0127] Setting a router mechanism, specifically using a router to dynamically allocate the weight of each feedforward neural network layer according to the patient's health semantic features to obtain branch weights;

[0128] The comprehensive multi-branch output is used to integrate the output results of the multi-branch feedforward neural network. Specifically, according to the branch weight, the output features of each feedforward neural network are weighted and summed to obtain the comprehensive health features of the patient. The calculation formula is:

[0129] ;

[0130] In the formula, O FFN is the comprehensive health characteristics of the patient, M is the number of feedforward neural network layers, and p i is the branch weight, specifically the branch weight of the i-th feedforward neural network layer;

[0131] Output feature normalization, specifically, connecting the patient's comprehensive health features and the patient's health semantic features through residuals, and performing normalization processing to obtain the patient's comprehensive health normative features;

[0132] The classification layer construction unit specifically processes the comprehensive normative features of patient health through the softmax activation function and the fully connected layer to obtain the model classification results;

[0133] The model training unit is specifically configured to construct a bidirectional long short-term memory model combined with a switching feedforward network through the BERT embedding layer construction unit, the self-attention layer construction unit, the bidirectional long short-term memory layer construction unit, the switching feedforward network layer construction unit and the classification layer construction unit, and perform model training to obtain a patient constitution recognition model;

[0134] A patient constitution type calculation unit, specifically, identifies the patient constitution through a patient constitution identification model to obtain the patient constitution type;

[0135] By performing the above operations, in order to address the technical problems in the existing TCM nursing field where the semantic complexity of TCM constitution description presents a high-dimensionality and nonlinearity of data characteristics, and the traditional intelligent methods are insufficient in capturing the characteristics of patient health data and TCM questionnaire data, this solution creatively adopts a bidirectional long short-term memory model combined with a switching feedforward network to identify patient constitutions. It has a deep understanding of semantic information and good temporal modeling performance, and enhances the ability to extract constitution features. At the same time, the switching feedforward network layer enhances the adaptability of the model to the complex semantics and features of TCM through a dynamic branching mechanism and comprehensive feature processing, and realizes personalized and accurate classification, thereby improving the robustness and accuracy of constitution identification and providing more scientific support for personalized TCM nursing services.

[0136] Example 6, see Figure 1 This embodiment is based on the above embodiment. In the health management suggestion generation module, the health management suggestion is generated by searching the TCM nursing knowledge graph for health management suggestions corresponding to the patient's physical constitution type according to the patient's physical constitution type to obtain the health management suggestions;

[0137] The patient's constitution types include mild constitution, qi deficiency constitution, yang deficiency constitution, yin deficiency constitution, phlegm-damp constitution, damp-heat constitution, blood stasis constitution, qi stagnation constitution and special constitution;

[0138] The health management recommendations include dietary recommendations, exercise recommendations, medicated diet recommendations and massage treatments.

[0139] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0140] While the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that many changes, modifications, substitutions and variations can be made to the embodiments without departing from the principles and spirit of the invention.

[0141] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.

Claims

1. A health management system for TCM nursing population based on big data, characterized by: It includes data preparation module, data preprocessing module, knowledge graph construction module, patient constitution identification module and health management suggestion generation module; The data preparation module is used to prepare the original data required for the health management of the TCM nursing population, specifically to collect patient health data and TCM nursing knowledge data, and send the patient health data and the TCM nursing knowledge data to the data preprocessing module; The data preprocessing module performs data preprocessing on the patient's health data and the TCM nursing knowledge data to obtain the patient's health standard data and the TCM nursing knowledge standard data, and sends the patient's health standard data to the patient's constitution identification module, and sends the TCM nursing knowledge standard data to the knowledge graph construction module; The knowledge graph construction module is used to construct a TCM nursing knowledge graph, specifically to construct a knowledge graph based on TCM nursing knowledge standard data to obtain a TCM nursing knowledge graph, and send the TCM nursing knowledge graph to the health management advice generation module; The patient constitution identification module is used to identify the patient's constitution, specifically, based on the patient's health standard data, using a bidirectional long short-term memory model combined with a switching feedforward network to identify the patient's constitution, obtain the patient's constitution type, and send the patient's constitution type to the health management suggestion generation module; The health management suggestion generation module is used for health management of the TCM nursing population, specifically generating health management suggestions based on the patient's physical constitution type and the TCM nursing knowledge graph.

2. The TCM nursing population health management system based on big data according to claim 1 is characterized by: In the knowledge graph construction module, there are entity extraction unit, relationship extraction unit, entity alignment unit and TCM nursing knowledge graph construction unit, which specifically include the following contents: An entity extraction unit is used to extract key entities in the TCM nursing knowledge standard data, specifically using an improved transformer model to extract entities and obtain TCM nursing knowledge entities; The improved transformer model includes a BERT embedding layer, a context encoding improvement layer, a feature decoding layer and an entity output layer; The entity extraction unit includes the following contents: Constructing a BERT embedding layer, specifically, in the BERT embedding layer, using a pre-trained BERT model to perform word embedding processing on the TCM nursing knowledge standard data to obtain TCM nursing knowledge word embedding features; Construct a context coding improvement layer, specifically, use the encoder as the basic architecture of the context coding improvement layer, introduce relative position coding and multi-head self-attention mechanism into the context coding improvement layer, process the embedded features of TCM nursing knowledge through the context coding improvement layer, and obtain the context features of TCM nursing knowledge; Constructing a feature decoding layer, specifically, by constructing a conditional random field scoring function, constructing a feature decoding layer objective function and calculating an optimal label sequence to obtain a feature decoding layer, and processing the context features of traditional Chinese medicine nursing knowledge through the feature decoding layer to obtain an optimal label sequence; Constructing an entity output layer, specifically, performing entity mapping on the optimal label sequence through the entity output layer to obtain the TCM nursing knowledge entity; The relationship extraction unit is used to extract the semantic relationship between entities, specifically, based on the TCM nursing knowledge entity and TCM nursing knowledge standard data, a bidirectional long-term and short-term relationship extraction model is used to extract the relationship, and the TCM nursing knowledge entity relationship is obtained; The bidirectional long-short-term relationship extraction model includes a BERT embedding layer, a bidirectional long-short-term memory layer, an attention pooling layer and an output layer; the attention pooling layer includes an attention layer and a weighted pooling layer; The entity alignment unit is used to merge similar entities. Specifically, based on the semantic similarity of entity pairs, the TCM nursing knowledge entities are aligned to obtain TCM nursing knowledge aligned entities, including the following contents: Calculate the semantic similarity of entity pairs, the calculation formula is: ； In the formula, Sim uz is the semantic similarity of entity pairs, specifically refers to the semantic similarity between the u-th TCM nursing knowledge entity and the z-th TCM nursing knowledge entity, u is the first index of the TCM nursing knowledge entity, z is the second index of the TCM nursing knowledge entity, the second index of the TCM nursing knowledge entity is not equal to the first index of the TCM nursing knowledge entity, sim(·) is the similarity calculation function used to calculate the semantic similarity of entity pairs, ent u is the vector representation of the u-th TCM nursing knowledge entity, ent z is the vector representation of the zth TCM nursing knowledge entity, is the norm symbol; Set an alignment threshold. When the semantic similarity of an entity pair exceeds the alignment threshold, align the corresponding two TCM nursing knowledge entities. The TCM nursing knowledge graph construction unit constructs the TCM nursing knowledge graph based on the TCM nursing knowledge alignment entities and TCM nursing knowledge entity relationships.

3. The TCM nursing population health management system based on big data according to claim 2 is characterized by: In the construction feature decoding layer, the calculation formula of the construction conditional random field scoring function is: ； Where S(·) is the conditional random field scoring function, h is the TCM nursing knowledge context feature, y is the TCM nursing knowledge feature label, a is the TCM nursing knowledge context feature index, n is the number of TCM nursing knowledge context features, is the transfer weight, specifically the weight of the a-1th TCM nursing knowledge feature label transferred to the ath TCM nursing knowledge feature label, y a-1 is the a-1th TCM nursing knowledge feature label, specifically the label corresponding to the a-1th TCM nursing knowledge context feature, y a is the ath TCM nursing knowledge feature label, is the matching score, specifically the matching score between the a-th TCM nursing knowledge context feature and the a-th TCM nursing knowledge feature label, h a It is the ath TCM nursing knowledge context feature; The calculation formula for constructing the feature decoding layer objective function is: ； In the formula, is the optimal label, used to represent the optimal TCM nursing knowledge feature label, argmax is to find the maximum value, used to find the TCM nursing knowledge feature label that maximizes the conditional probability, P(·) is the conditional probability function, exp(·) is the exponential function, is a candidate TCM nursing knowledge feature label, Y is a TCM nursing knowledge feature label sequence; The calculating of the optimal label sequence specifically involves using a Viterbi algorithm to calculate the optimal label sequence based on a feature decoding layer objective function.

4. The TCM nursing population health management system based on big data according to claim 3 is characterized by: In the patient constitution recognition module, the bidirectional long short-term memory model combined with the switched feedforward network includes a BERT embedding layer, a self-attention layer, a bidirectional long short-term memory layer, a switched feedforward network layer and a classification layer; The patient constitution recognition module is provided with a BERT embedding layer construction unit, a self-attention layer construction unit, a bidirectional long short-term memory layer construction unit, a switching feedforward network layer construction unit, a classification layer construction unit, a model training unit and a patient constitution type calculation unit, and specifically includes the following contents: A BERT embedding layer construction unit, specifically, in the BERT embedding layer, a pre-trained BERT model is used to perform word embedding processing on the patient health standard data to obtain the patient health word embedding features; The self-attention layer construction unit is used to capture the global dependency between words. Specifically, the self-attention mechanism is used to process the patient health word embedding features to obtain the patient health global features; The bidirectional long short-term memory layer construction unit is used to capture the deep semantics of the context. Specifically, the forward long short-term memory unit and the backward long short-term memory unit are set in the bidirectional long short-term memory layer. The patient's global health features are processed through the bidirectional long short-term memory layer to obtain the patient's health semantic features. The calculation formula is: ； In the formula, H t is the patient health semantic feature at time step t, t is the time step index, concat(·) is the feature concatenation operation, LSTM fw (·) is the forward long short-term memory unit, is the global health characteristic of the patient at time step t, is the forward hidden state at the t-1th time step, is the forward memory state of the t-1th time step, LSTM bw (·) is the backward long short-term memory unit, is the backward hidden state at the t-1th time step, is the backward memory state at the t-1th time step; The switch feed-forward network layer building unit is used to dynamically adapt to the patient's health semantic features, including the following: Constructing a multi-branch feedforward neural network layer, specifically setting M parallel feedforward neural network layers, each feedforward neural network layer independently learns the nonlinear features in the patient's health semantic features; The feedforward neural network layer includes two fully connected layers, and the calculation formula is: ； In the formula, FFN i (·) is the i-th feedforward neural network layer, i is the feedforward neural network layer index, H is the patient health semantic feature, GELU(·) is the GELU activation function, ReLU(·) is the ReLU activation function, W 2i is the weight of the second fully connected layer in the i-th feedforward neural network layer, W 1i is the weight of the first fully connected layer in the i-th feedforward neural network layer, b 1i is the first fully connected layer bias term in the i-th feedforward neural network layer, b 2i is the bias term of the second fully connected layer in the i-th feedforward neural network layer; Setting a router mechanism, specifically using a router to dynamically allocate the weight of each feedforward neural network layer according to the patient's health semantic features to obtain branch weights; The comprehensive multi-branch output is used to integrate the output results of the multi-branch feedforward neural network. Specifically, according to the branch weight, the output features of each feedforward neural network are weighted and summed to obtain the comprehensive health features of the patient. The calculation formula is: ； In the formula, O FFN is the comprehensive health characteristics of the patient, M is the number of feedforward neural network layers, and p i is the branch weight, specifically the branch weight of the i-th feedforward neural network layer; Output feature normalization, specifically, connecting the patient's comprehensive health features and the patient's health semantic features through residuals, and performing normalization processing to obtain the patient's comprehensive health normative features; The classification layer construction unit specifically processes the comprehensive normative features of patient health through the softmax activation function and the fully connected layer to obtain the model classification results; The model training unit is specifically configured to construct a bidirectional long short-term memory model combined with a switching feedforward network through the BERT embedding layer construction unit, the self-attention layer construction unit, the bidirectional long short-term memory layer construction unit, the switching feedforward network layer construction unit and the classification layer construction unit, and perform model training to obtain a patient constitution recognition model; The patient's physical constitution type calculation unit specifically identifies the patient's physical constitution through a patient's physical constitution identification model to obtain the patient's physical constitution type.

5. The TCM nursing population health management system based on big data according to claim 4 is characterized by: In the health management suggestion generation module, the health management suggestions are generated by specifically querying the health management suggestions corresponding to the patient's physical constitution type from the traditional Chinese medicine nursing knowledge graph based on the patient's physical constitution type to obtain the health management suggestions.

6. The TCM nursing population health management system based on big data according to claim 5 is characterized by: In the data preprocessing module, the data preprocessing includes data cleaning and data standardization processing, and the data cleaning includes invalid value elimination, outlier filtering and missing value filling.

7. The TCM nursing population health management system based on big data according to claim 6 is characterized by: In the data preparation module, the patient health data includes TCM questionnaire data and medical record data. The TCM questionnaire data is obtained by the patient filling out the TCM body mass scale, the medical record data is obtained from the medical system, and the TCM nursing knowledge data is obtained from the TCM database.

Images