A traditional Chinese medicine prescription recommendation method based on the principle of monarch, minister, auxiliary and guide combination

Abstract

The application belongs to the technical field of traditional Chinese medicine prescription recommendation, and particularly relates to a traditional Chinese medicine prescription recommendation method based on the compatibility principle of monarch, minister, assistant and messenger. Unlike the existing single classifier for predicting a complete traditional Chinese medicine prescription, the application uses four classifiers (monarch, minister, assistant and messenger) to gradually generate a traditional Chinese medicine prescription in accordance with the compatibility principle of monarch, minister, assistant and messenger. The four classifiers constructed by the method can generate traditional Chinese medicines in accordance with respective roles, complete preliminary prescription generation, propose a prescription quality evaluation method, modify the preliminary prescription, and obtain a final prescription. The application can solve the defects of weak correlation of traditional Chinese medicines and lack of clinical significance in the existing automatic prescription generation model. The traditional Chinese medicine prescription generated by the model contains the knowledge of the compatibility of monarch, minister, assistant and messenger, and the generated prescription better reflects the thinking of clinical prescription writing, thereby providing auxiliary decision-making for the clinic.

Description

Technical Field

[0001] The present invention belongs to the technical field of traditional Chinese medicine prescription recommendation, and in particular relates to a traditional Chinese medicine prescription recommendation method based on the principle of monarch, minister, assistant and envoy compatibility. Background Art

[0002] The principle of monarch, minister, assistant, and envoy as a formula composition principle first appeared in the Inner Canon of Medicine and has a very long history of application in Traditional Chinese Medicine. The monarch drug, as the first of the monarch, minister, assistant, and envoy, acts on the primary symptom or disease, determining the primary focus of a prescription. The minister drug, as a secondary Chinese medicine, serves two functions: first, to assist the monarch drug and strengthen the treatment of the primary symptom or disease; second, to treat concurrent symptoms or diseases. The functions of adjuvant drugs include assisting, controlling, and counter-assisting. Assisting enhances the efficacy of the monarch and minister drugs or directly treats secondary concurrent symptoms; controlling mitigates the toxicity or potency of the monarch and minister drugs; and counter-assisting drugs are drugs with properties opposite to the monarch drug but capable of complementing it. The number of adjuvant drugs is greater than that of minister drugs. The function of the envoy drug is to guide the meridians and harmonize the other drugs, directing them to the site of the disease so that they can work together to expel pathogenic factors. Generally, only one or two envoy drugs are needed.

[0003] The paper “A Knowledge Graph Enhanced Topic Modeling Approach for Herb Recommendation” proposed the HC-KGETM model combining the seven emotions compatibility of traditional Chinese medicine, which simplifies the seven emotions compatibility into two states: one is the combination of traditional Chinese medicine with positive effects, called the positive chain combination L P , one is a combination of Chinese medicine with negative effects, called negative chain combination L N Each L P and L N Contains at least two Chinese medicines. P The probability of extracting Chinese medicine from L N The probability of the same subject is extracted to generate prescriptions that meet the compatibility rules.

[0004] Most existing prescription generation models use the LDA topic model, inferring syndromes from medical record data and then deriving a sequence of traditional Chinese medicines (TCMs) as a prescription. Some work also uses Naive Bayesian prediction to predict the probability of each drug in the monarch, minister, assistant, or envoy position. Existing methods ignore the interactions between TCMs and treat TCM prescription generation as a simple sequence prediction problem. While HC-KGETM considers TCM interactions, it simplifies them into two categories: positive and negative interactions, failing to capture the complex relationships between individual herbs in a TCM prescription. Furthermore, the model only considers the interactions between pairs of herbs, ignoring the potential connections between multiple herbs. Summary of the Invention

[0005] To solve the above problems, the application gradually pushes out monarch drugs, minister drugs, auxiliary drugs and assistant drugs according to the interaction among the four roles of monarch, minister, auxiliary and assistant, generates the final prescription, solves the problem of weak correlation in the generated prescription, and provides auxiliary decision for the TCM doctors.

[0006] The technical scheme of the application is:

[0007] A TCM prescription recommendation method based on the compatibility principle of monarch, minister, auxiliary and assistant, wherein X represents a syndrome, Y represents a traditional Chinese medicine; wherein the syndrome X includes main symptoms X z and auxiliary symptoms X j , the traditional Chinese medicine Y includes monarch medicine Y j , minister medicine Y c , auxiliary medicine Y z and assistant medicine Y s , the traditional Chinese medicine is encoded by a feature vector according to three characteristics of drug nature, taste and meridian; the recommendation method comprises:

[0008] S1, constructing a monarch drug classifier to select a monarch drug:

[0009]

[0010] wherein, is to find a monarch drug Y j that maximizes the product, j P(Y j ) is the class prior of the monarch drug Y j , and is the conditional probability of each attribute in the main symptom:

[0011]

[0012]

[0013] wherein D represents a medical record sample set, each sample includes symptom and prescription data, represents a medical record sample set with the monarch drug Y j , represents a sample set consisting of the medical record with the monarch drug Y j , the i-th attribute of the main symptom is N J represents the number of types of monarch drugs in the sample set, and N i represents the possible values of the i-th attribute of the main symptom X z ;

[0014] S2, constructing a minister drug classifier to select a minister drug:

[0015]

[0016] wherein, P(X z ,Xj ,Y j |Y c ) indicates that in the minister drug Y c Under the condition of X z ,X j ,Y j The conditional joint probability of the drug Y c The class prior P(Y c ) and the conditional probability of each attribute combination Expressed as:

[0017]

[0018]

[0019] in, Indicates that the ministerial drug is Y c A collection of medical case samples, Indicates that the ministerial drug is Y c The attributes of the main symptom i, the concurrent symptom k, and the main medicine l in the medical record are Y j l A set of samples. N C Indicates the number of types of drugs in the sample set, N i 、N k 、N l Represents the main symptom X z The i-th attribute, and the proof of X j The kth attribute, monarch drug Y j The number of possible values ​​of the lth attribute. For the drug classifier, select the probability Chinese medicine is used as the prediction result of the classifier. The average number of auxiliary medicinal ingredients in the existing prescription is rounded up;

[0020] S3. Construct an adjuvant classifier to select adjuvants:

[0021]

[0022] Where P(X j ,Y j ,Y c |Yx) represents adjuvant Y z Under the condition of X j ,Y j ,Y c Because more than one adjuvant drug is selected in S2, when the adjuvant drug classifier is calculated, Y c The weighted average vector of all the agent feature vectors is used.

[0023] By analogy, adjuvant Y zThe class prior P(Y z ) and the conditional probability of each attribute combination It can be expressed as:

[0024]

[0025]

[0026] in, Indicates that the adjuvant is Y z A collection of medical case samples, Indicates that the adjuvant is Y z The attributes of the i-th concurrent symptom, the k-th main drug, and the l-th auxiliary drug in the medical case are Y j k 、 A set of samples, N Z Indicates the number of adjuvants in the sample set, N i ′、N k ′、N l ′ respectively represent concurrent proof X j The i-th attribute, monarch medicine Y j The kth attribute, minister medicine Y c The number of possible values ​​of the lth attribute; for the adjuvant classifier, select the probability Chinese medicine is used as the prediction result of the classifier. The average number of adjuvants in the existing prescriptions is rounded up;

[0027] S4. Construct a drug classifier to select drugs:

[0028]

[0029] Where P(X z ,X j ,Y j ,Y c ,Y z |Y s ) indicates that drug Y s Under the condition of X z ,X j ,Y j ,Y c ,Y z Because more than one adjuvant and auxiliary medicine are selected in S2 and S3, when the drug classifier is calculated, Y c and Y z The weighted average vector of all the adjuvant and auxiliary drug feature vectors was used respectively.

[0030] By analogy, drug Y S The class prior P(Y S) and the conditional probability of each attribute combination It can be expressed as:

[0031]

[0032]

[0033] in, Indicates that the drug is Y S A collection of medical case samples, Indicates that the drug is Y S In the medical record, the attributes of the main symptom i, the concurrent symptom k, the main drug l, the auxiliary drug ii, and the adjuvant drug kk are respectively A set of samples. N S Indicates the number of types of drugs used in the sample set, N i 、N k 、N l 、N ii 、N kk Represents the main symptom X z The i-th attribute, and the proof of X j The kth attribute, monarch drug Y j The first attribute, minister medicine Y c The second attribute, adjuvant Y z The number of possible values ​​of the kkth attribute. For the drug classifier, select the probability Chinese medicine is used as the prediction result of the classifier. Rounding the average number of medicinal ingredients in the existing prescription;

[0034] S5, based on the obtained monarch drug, minister drug, adjuvant drug and guiding drug, generate a preliminary prescription C = {Y j ,Y c ,Y z ,Y s};

[0035] S6. Conduct quality evaluation and Chinese medicine substitution on the preliminary prescription to obtain the final prescription:

[0036] Using the traditional Chinese medicines in existing prescriptions to establish a traditional Chinese medicine-traditional Chinese medicine network, using traditional Chinese medicines as nodes of the network, and defining the weight of the edge between two nodes as e ij , e ij is the number of prescriptions with Chinese medicine corresponding to node i and node j; the distance between two Chinese medicines is defined as d ij , and if TCM i and TCM j are directly connected, then:

[0037]

[0038] otherwise

[0039]

[0040] where p ij is the number of hops of the shortest path between Chinese medicine i and Chinese medicine j;

[0041] Evaluation of the preliminary prescription:

[0042]

[0043] Where m is the total number of Chinese medicines in the prescription, S p The smaller the value, the more commonly used drug pairs or drug pairs with similar distances are included in the generated prescription;

[0044] Using S p Formula to calculate the average prescription score of existing prescriptions The evaluation S obtained from the initial prescription p and For comparison, if or The preliminary prescription is not modified and is output as the final prescription; otherwise, the preliminary prescription is modified:

[0045] By the distance formula d ij Select the drug pair with the longest distance in the preliminary prescription C<a,b> , use the following formula to calculate the closeness centrality of Chinese medicine a:

[0046]

[0047] where d aj is the distance between Chinese medicine a and any Chinese medicine j, and C(d b ). If C(d a )>C(d b ), then replace the Chinese medicine b by selecting a Chinese medicine c from the existing prescription that is closest to the Chinese medicine b in terms of medicinal properties, taste and meridian code, and the Chinese medicine c is not in the preliminary prescription C; if C(d a ) <C(d b ), similarly replace the Chinese medicine a; if C(d a )=C(d b ) then any one of a and b can be replaced. Get the revised prescription and recalculate S for the revised prescription p , and reconnect with Compare until the recalculated prescription S p Rating less than Finally, the final prescription is obtained and output.

[0048] The beneficial effects of the present invention are:

[0049] This invention addresses the shortcomings of existing automatic prescription generation models, which suffer from weak relevance and lack of clinical significance. The TCM prescriptions generated by this model incorporate the knowledge of the compatibility of monarch, minister, adjuvant, and guiding herbs. The generated prescriptions better reflect clinical prescribing thinking and provide support for clinical decision-making. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 It is the overall flow chart of the present invention.

[0051] Figure 2 It is a schematic diagram of the Chinese medicine-Chinese medicine network structure. DETAILED DESCRIPTION

[0052] The present invention is described in detail below with reference to the accompanying drawings.

[0053] The key point of this invention is that, unlike existing methods where a single classifier predicts a complete TCM prescription, this method uses four classifiers (Monarch, Minister, Assistant, and Envoy) to gradually generate TCM prescriptions that adhere to the Monarch, Minister, Assistant, and Envoy compatibility rules. These four classifiers, constructed using this method, can generate TCM herbs that match their respective roles, completing the initial prescription generation. A method for evaluating prescription quality is then proposed to refine the initial prescription and obtain the final one.

[0054] In the present invention, X is defined as syndrome, and Y is defined as Chinese medicine. z and combined certificate X j According to the role of Chinese medicine in prescriptions, it is divided into monarch medicine Y j 、Chemical Y c , adjuvant Y z and drug Y s .

[0055] The vector encoding of the main and concurrent symptoms refers to the national standard classification and coding of TCM diseases and syndromes. The construction of the TCM feature vector uses three characteristics of the TCM: nature, flavor, and meridian. TCM has four properties and five flavors. The four properties refer to the four different medicinal properties of the medicine: cold, hot, warm, and cool; the five flavors refer to the five different medicinal flavors of the medicine: sour, bitter, sweet, pungent, and salty. The meridian refers to the property of TCM that it has selective therapeutic effects on the twelve meridians of the human body. Each TCM is modeled as a 21-dimensional feature vector (four properties, five flavors, and twelve meridians). For each attribute, the presence is encoded as 1, and the absence is encoded as 0.

[0056] like Figure 1 As shown, the present invention includes the following steps:

[0057] Step 1: Construct a monarch drug classifier and select the monarch drug according to the main symptoms:

[0058] Since the monarch drug is only related to the main symptom, when predicting the monarch drug, only the main symptom X z , the output is the monarch drug Y jThe first Naive Bayes classifier (the monarch drug classifier) ​​selects the drug with the highest probability as the monarch drug (only one is needed). Naive Bayes assumes that each attribute independently affects the classification result, which leads to the following assumptions:

[0059]

[0060] P(Y j |X z ) indicates that in the main evidence X z Under the condition that the monarch drug is Y j The conditional probability of Indicates the main symptom X z The value of the i-th attribute, Indicates that the monarch drug is Y j Main symptom X z The value of the i-th attribute is The conditional probability of . ∏ represents the product, and d is the dimension of the symptom. P(Y j ) and P(X z ) are respectively the monarch drug Y j and main symptom X z Class priors.

[0061] P(X z ) is the evidence factor used for normalization. For a certain sample X z , evidence factor P(X z ) and class label Y j It is irrelevant, that is, it is the same for all the main drug labels, so the estimated P(Y j |X z ) is transformed into how to estimate the prior P(Y j ) and likelihood P(X z ) is not calculated.

[0062] P(Y j ) is calculated as follows:

[0063]

[0064] Where D represents the entire medical record sample set, and each sample includes symptom and prescription data. Indicates that the monarch drug is Y j The sample collection.

[0065] Since the syndrome coding is discrete, the conditional probability of each attribute in the main syndrome It can be expressed as:

[0066]

[0067] in, express The value of the i-th attribute is A collection of samples.

[0068] In order to avoid the small number of samples is 0, and the Laplace correction is used to convert the class prior P(y j ) and the conditional probability of each attribute Modify to:

[0069]

[0070]

[0071] where N J Indicates the number of types of monarch drugs in the sample set. N i Indicates the main symptom X z The possible values ​​of the i-th attribute.

[0072] The monarch drug classifier is expressed as follows:

[0073]

[0074] in, is to find the The monarch drug Y with the largest product j , Y is a set of Chinese medicine,

[0075] For the monarch drug classifier, only one Chinese medicine, that is, the drug with the highest probability, needs to be selected as the monarch drug.

[0076] Step 2: Construct a ministerial drug classifier and select ministerial drugs based on the main symptoms, concomitant symptoms, and monarch drugs:

[0077] The ministerial drug is related to the main symptom, the secondary symptom and the monarch drug. The ministerial drug classifier inputs the main symptom X z , and certificate X j The result Y of the Hejun medicine classifier j , the output is the minister medicine Y c The ministerial drug classifier is expressed as:

[0078]

[0079] Where P(X z ,X j ,Y j |Y c ) indicates that in the minister drug Y c Under the condition of X z ,X j ,Y j For the drug classifier, select the conditional joint probability of Chinese medicine is used as the prediction result of the classifier. The average number of auxiliary medicinal ingredients in all prescriptions in existing medical records is rounded up.

[0080] Step 3: Construct an adjuvant classifier and select adjuvants based on concurrent symptoms, main drugs, and auxiliary drugs:

[0081] The adjuvant drug is related to the monarch drug, the minister drug and the concurrent symptoms. The adjuvant drug classifier inputs the monarch drug Y j 、Chemical Y c , and combined certificate X j , the output is adjuvant Y z The adjuvant classifier is expressed as:

[0082]

[0083] Where P(X j ,X j ,Y c |Y z ) indicates that adjuvant Y z Under the condition of X j ,Y j ,Y c For the adjuvant classifier, select the conditional joint probability of Chinese medicine is used as the prediction result of the classifier. The average number of adjuvant ingredients in all prescriptions in existing medical records is rounded up.

[0084] Step 4: Construct a guide drug classifier and select guide drugs based on the main symptoms, secondary symptoms, monarch drugs, minister drugs, and adjuvant drugs:

[0085] Since the drug needs to be introduced into the meridian, it requires knowledge of each drug and information of the main symptom and the concurrent symptom. z , and certificate X j ,junyao Y j 、Chemical Y c and adjuvant Y z , the output is drug Y s The drug classifier is expressed as:

[0086]

[0087] Where P(X z ,X j ,Y j ,Y c ,Y z |Y s ) indicates that drug Y s Under the condition of X z ,X j ,Y j ,Y c ,Y z The conditional joint probability of . For the drug classifier, select the probability Chinese medicine is used as the prediction result of the classifier. The average number of medicinal ingredients in all prescriptions in existing medical records is rounded up.

[0088] Step 5: Generate preliminary prescription:

[0089] The combined monarch, minister, adjuvant and guiding drugs generate a preliminary prescription C = {Y j ,Y c ,Y z ,Y s}.

[0090] Step 6: Conduct quality evaluation and Chinese medicine substitution on the preliminary prescription to obtain the final prescription:

[0091] (1) Construction of Chinese medicine co-occurrence network

[0092] According to the classic prescriptions in Treatise on Febrile Diseases, a Chinese medicine-Chinese medicine network is established, in which the nodes are all the Chinese medicines that have appeared. i Indicates that the edge is the co-occurrence information of Chinese medicine and classic prescription. For example, Chinese medicine d f and Chinese medicine e In the 20 classic prescriptions, d f and d e The edge weight e fe Set to 20. Figure 2 is an example graph of a network.

[0093] Define the TCM distance d ij If TCM i and TCM j are directly connected, their direct distance is calculated as follows:

[0094]

[0095] where e ij Represents the edge weight of Chinese medicine i and Chinese medicine j. If Chinese medicine i and Chinese medicine j are not directly connected, assuming there are n-1 nodes between them, then their indirect distance d indirect Defined as:

[0096]

[0097] where p ij is the number of hops of the shortest path between Chinese medicine i and Chinese medicine j. Figure 2 In, p af =1,p ab = 3. The denominator is the sum of the edge weights on the links of TCM i and TCM j.

[0098] It is important to note that instead of directly dividing the total path sum of the entire prescription by the weighted sum, the distance values ​​for each drug pair are calculated first and then summed. This prevents distant drug pairs from weakening the effect of nearby drug pairs. This also increases the impact of core drug pairs on the prescription score. For example, if a prescription includes an extremely distant drug pair, the numerator will become very large, and the prescription score will deteriorate significantly. However, the effect of such a distant drug pair should be smaller than that of a nearby drug pair. The effect of a distant drug pair with a long path can be reduced by dividing it by the sum of the weights along the path. However, the effect of drug pairs with short paths and extremely high weights will still be highlighted.

[0099] (2) Prescription evaluation and modification to generate the final prescription

[0100] For the initial prescription C={Y j ,Y c ,Y z ,Y s}, use the following formula to evaluate the score of the prescription:

[0101]

[0102] Where m is the total number of Chinese medicines in the prescription, d ij is the direct or indirect distance between Chinese medicine i and Chinese medicine j. p The smaller the value, the more commonly used drug pairs or drug pairs with similar distances are contained in the generated prescription.

[0103] According to S p Formula for calculating the average prescription score of classic prescriptions in Treatise on Febrile Diseases The score S of the prescription will be automatically generated p Comparison with the average prescription score of classic prescriptions If the score is lower than the average prescription score of the classic prescription, no correction is required and the preliminary prescription is directly used as the final prescription; if the score is higher than the average prescription score of the classic prescription, the uncommon drug pairs need to be replaced. The specific method is as follows:

[0104] For the drug pair with the longest distance in the automatically generated prescription C<a,b> Replace the commonly used drug pairs in classic prescriptions, and the replacement strategy is to choose<a,b> The Chinese medicine with lower closeness centrality is replaced. Closeness centrality can measure the distance between a node and all nodes. The larger the closeness centrality, the closer the point is to any other point. a The closeness centrality of is calculated as follows:

[0105]

[0106] where d aj is the direct or indirect distance between Chinese medicine a and any Chinese medicine j.<a,b> When making a replacement, such as C(d a)>C(d b ), then replace Chinese medicine b. Select a Chinese medicine from the commonly used medicine set of Chinese medicine a based on the classic prescription that is most similar to Chinese medicine b in terms of medicinal properties, taste, and meridian code, and replace Chinese medicine b with a Chinese medicine that is not in the automatically generated prescription Y. After the replacement, recalculate the S of the new prescription. p Rating, if the rating is still greater than Continue to replace the farthest drug pair until the generated prescription S p Rating less than Get the final prescription.

Claims

1. A method for recommending TCM prescriptions based on the principle of monarch, minister, assistant and envoy compatibility, defining Indicates symptoms, Indicates Chinese medicine; symptoms Including main symptoms and combined certificate ,traditional Chinese medicine Including monarch drug 、Minister medicine , adjuvant and the drug , Chinese medicine uses three features, namely, medicinal properties, taste and meridians, to encode feature vectors; the characteristics are, The recommended methods include: S1. Construct a monarch drug classifier to select the monarch drug: ， in, is to find the The monarch drug with the largest product , King of medicines The class prior, is the conditional probability of each attribute in the main evidence: ， ， in Represents a medical case sample set, each sample includes symptom and prescription data, Indicates that the monarch drug is A collection of medical case samples, Indicates that the monarch drug is The main symptom in the medical record The attribute is A collection of samples, Indicates the number of types of monarch drugs in the sample set, Indicates the main symptom No. The number of possible values ​​for an attribute; S2. Construct a drug classifier to select drugs: ， in, Indicates that the medicine Under the conditions, The conditional joint probability of Class prior and the conditional probability of each attribute combination Expressed as: ， ， in, Indicates that the ministerial drug is A collection of medical case samples, Indicates that the ministerial drug is The main symptom in the medical record , and the first The first The attributes are A collection of samples, Indicates the number of types of drugs in the sample set, 、 、 Respectively represent the main symptoms No. Attributes, certification No. Attributes, monarch medicine No. The number of possible values ​​of an attribute; for the herbal medicine classifier, select the probability Chinese medicine is used as the prediction result of the classifier. The average number of auxiliary medicinal ingredients in the existing prescription is rounded up; S3. Construct an adjuvant classifier to select adjuvants: ， in Indicates adjuvant Under the conditions, The conditional joint probability of , because there is more than one ministerial drug selected in S2, when the adjuvant classifier is calculated Using the weighted average vector of all the adjuvant drug feature vectors, the adjuvant drug Class prior and the conditional probability of each attribute combination It can be expressed as: ， ， in, Indicates adjuvant A collection of medical case samples, Indicates adjuvant The medical records of The first The first The attributes are A collection of samples, Indicates the number of adjuvant types in the sample set, 、 、 Respectively indicate concurrent certification No. Attributes, monarch medicine No. Attributes, ministers No. The number of possible values ​​of an attribute; for the adjuvant classifier, select the probability Chinese medicine is used as the prediction result of the classifier. The average number of adjuvants in the existing prescriptions is rounded up; S4. Construct a drug classifier to select drugs: ， in Indicates that the drug is being used Under the conditions, The conditional joint probability of S2 and S3 is that there are more than one auxiliary medicine and adjuvant medicine selected in the calculation of the drug classifier. as well as Use the weighted average vector of all the adjuvant and ministerial drug feature vectors to make the drug Class prior and the conditional probability of each attribute combination It can be expressed as: ， ， in, Indicates that the drug is A collection of medical case samples, Indicates that the drug is The main symptom in the medical record , and the first The first The first Adjuvant The attributes are A collection of samples, Indicates the number of types of drugs used in the sample set, 、 、 、 、 Respectively represent the main symptoms No. Attributes, certification No. Attributes, monarch medicine No. Attributes, ministers No. Attributes, adjuvants No. The number of possible values ​​of an attribute; for the drug classifier, select the probability Chinese medicine is used as the prediction result of the classifier. Rounding the average number of medicinal ingredients in the existing prescription; S5. Generate a preliminary prescription based on the obtained monarch drug, minister drug, adjuvant drug and guiding drug ; S6. Conduct quality evaluation and Chinese medicine substitution on the preliminary prescription to obtain the final prescription: Use the traditional Chinese medicines that appear in existing prescriptions to establish a traditional Chinese medicine-traditional Chinese medicine network, use traditional Chinese medicine as the node of the network, and define the edge weight between two nodes as , is the number of prescriptions with Chinese medicine corresponding to node i and node j; the distance between two Chinese medicines is defined as , and if Chinese medicine and Chinese medicine Directly connected, then: ， otherwise ， in For traditional Chinese medicine and Chinese medicine The number of hops of the shortest path between them; Evaluation of the preliminary prescription: ， in is the total number of Chinese medicines prescribed, The smaller the value, the more commonly used drug pairs or drug pairs with similar distances are included in the generated prescription; use Formula to calculate the average prescription score of existing prescriptions , the evaluation of the initial prescription and For comparison, if or , the preliminary prescription is not modified and is output as the final prescription; otherwise, the preliminary prescription is modified: By distance formula Select a preliminary prescription The longest distance between the two medicine pairs , use the following formula to calculate Chinese medicine The closeness centrality of: ， in For traditional Chinese medicine and any Chinese medicine The distance between ,if , then Chinese medicine Replace the medicine by selecting a Chinese medicine from the existing prescription that is closest to the Chinese medicine b in terms of medicinal properties, taste and meridian code. , and Chinese medicine Not in initial prescription In; if , the same applies to Chinese medicine Replace; if Then you can replace any one of a and b to get the revised prescription, and recalculate the revised prescription , and reconnect with Compare until the prescription is recalculated Rating less than Finally, the final prescription is obtained and output.

Images