Doctor recommendation method based on capability comprehensive evaluation under multi-source heterogeneous correlation analysis

Abstract

The invention provides a doctor recommendation method based on capability comprehensive evaluation under multi-source heterogeneous correlation analysis, and relates to the field of medical services. The method comprises the following steps: firstly, taking the number of patients corresponding to diseases diagnosed and treated by doctors as ability indexes, generating a doctor ability matrix and performing normalization operation; secondly, for zero values in the normalized doctor capability matrix, sequentially obtaining doctor capability evaluation values based on doctor similarity and disease similarity based on multi-source heterogeneous data, and obtaining a similarity-driven comprehensive evaluation capability value after weighted fusion; and finally, based on the filled doctor capability matrix, combining quantitative information of severity and diagnosis and treatment difficulty of different diseases to obtain a recommendation index of a target doctor. According to the method, based on multi-source heterogeneous correlation analysis and fusion of the doctor similarity and the disease similarity, the ability value of the target doctor for the target disease which is not diagnosed and treated is comprehensively evaluated, the recommendation index of the target doctor is perfected, and experiments show that the method is significantly helpful for reasonably configuring limited medical resources.

Description

Technical Field

[0001] The present invention relates to the field of medical services, and in particular to a doctor recommendation method, system, storage medium and electronic device based on comprehensive ability evaluation under multi-source heterogeneous correlation analysis. Background Art

[0002] The shortage of medical resources has become a global problem and has received widespread attention from governments, public media and academia.

[0003] In related technologies, many studies focus on the allocation and optimization of medical resources. Considering that even general practitioners in primary hospitals receive the same standard of medical education and are trained in professional ethics, medical theoretical knowledge, and practical skills. Therefore, in fact, for the vast majority of common diseases that account for a large proportion of all diseases, the diagnosis and treatment capabilities of ordinary doctors are no less than those of top doctors. Patients prefer top doctors in high-level hospitals, often because they underestimate the capabilities of ordinary doctors.

[0004] In view of this, it is necessary to provide a technical solution for comprehensively evaluating doctors' abilities so as to recommend capable but neglected doctors to patients in a timely manner. Summary of the invention

[0005] 1. Technical issues to be solved

[0006] In view of the deficiencies in the prior art, the present invention provides a doctor recommendation method, system, storage medium and electronic device based on comprehensive ability evaluation under multi-source heterogeneous correlation analysis, which solves the technical problem of comprehensively evaluating doctor's ability.

[0007] (II) Technical solution

[0008] To achieve the above objectives, the present invention is implemented through the following technical solutions:

[0009] A doctor recommendation method based on comprehensive ability evaluation under multi-source heterogeneous correlation analysis, including:

[0010] Obtain a set of doctors and a set of disease types, use the number of patients corresponding to the diseases treated by the doctor as an indicator of his / her ability, generate a doctor ability matrix and perform normalization operations;

[0011] For the zero values ​​in the normalized doctor capability matrix, execute:

[0012] Based on the first multi-source heterogeneous data, obtaining a doctor ability evaluation value based on doctor similarity;

[0013] Based on the second multi-source heterogeneous data, obtaining a physician capability evaluation value based on disease similarity;

[0014] The doctor ability evaluation value based on doctor similarity and the doctor ability evaluation value based on disease similarity are weightedly integrated to obtain a similarity-driven comprehensive evaluation ability value, and fill in and replace the corresponding zero values ​​of the normalized doctor ability matrix;

[0015] Based on the filled physician competency matrix and combined with quantitative information on the severity and difficulty of diagnosis and treatment of different diseases, the recommendation index of the target physician is obtained.

[0016] Preferably, the first multi-source heterogeneous data includes pre-assigned doctor attribute information and non-zero ability values ​​of the target doctor and other doctors in the normalized doctor ability matrix, and the obtaining of the doctor ability evaluation value based on doctor similarity includes:

[0017] Based on the doctor attribute information, obtain doctors with similar internal basic attributes to the target doctor, and based on the non-zero capability values ​​of the internal similar doctors for the target disease that has not been diagnosed and treated by the target doctor, obtain the first capability value of the target doctor for the target disease;

[0018] Based on the non-zero capability values ​​of the target doctor and other doctors, obtaining doctors with similar external medical experience to the target doctor, and based on the non-zero capability values ​​of the external similar doctors for the target disease, obtaining a second capability value of the target doctor for the target disease;

[0019] The first capability value and the second capability value are weightedly fused to obtain a doctor capability evaluation value based on doctor similarity.

[0020] Preferably, the doctor attribute information includes the hospital where the doctor works, the level of experience, and the number of patients treated. The process of obtaining the first capability value includes:

[0021] The cosine distance is used to calculate the similarity of the internal basic attributes between the target doctor and other doctors; it is expressed as:

[0022]

[0023] Among them, DIS p,q Represents the target doctor doc represented by cosine distance p With doctor doc q The similarity of the internal basic attributes between them, subscript p≠q; ||·|| represents the vector length; IC i ={wh i ,dl i ,pn i} indicates doctor doc i The internal basic properties of wh i ,dl i ,pn i Respectively represent doctor doc istandard values ​​for the hospital where the candidate worked, level of experience, and number of patients treated;

[0024] Select DIS p,q The doctor corresponding to the positive number is doc q As a neighbor, build the target doctor doc p The first neighbor set of:

[0025]

[0026] Where DOC = {doc i |i∈{1,…,m}} represents the set of doctors, m is the number of doctors;

[0027] Use in Internal similarity in doctordoc q Dis j The non-zero capability value of the target doctor doc is obtained. p Dis j The first ability value of; expressed as:

[0028]

[0029] in, Indicates the target doctor doc p Dis j The first capability value of Doc-IS indicates the internal similarity of doctors; a q,j Indicates internal similarity doctor doc q Dis j A non-zero capability value.

[0030] Preferably, the process of acquiring the second capability value of the target doctor for the target disease includes:

[0031] The Pearson correlation coefficient is used to calculate the similarity of external medical experience between the target doctor and other doctors; it is expressed as:

[0032]

[0033] Among them, DES p,q Represents the target doctor doc represented by the Pearson correlation coefficient p With doctor doc q Similarity of external medical experience between p,q Indicates the target doctor doc p With doctor doc q Types of diseases treated; Respectively represent the target doctor doc p With doctor doc qThe average non-zero capability value of

[0034] Select DES p,q A positive number for doctor doc q As a neighbor, build the target doctor doc p The second neighbor set of:

[0035]

[0036] Where DIS = {dis j |j∈{1,…,n}} represents the set of disease types, and n is the number of disease types;

[0037] Use in Internal similarity in doctordoc q Dis j The non-zero capability value of the target doctor doc is obtained. p Dis j The second ability value of; expressed as:

[0038]

[0039] in, Indicates the target doctor doc p Dis j The second capability value of , the superscript Doc-IS represents the doctor external similarity.

[0040] Preferably, the second multi-source heterogeneous data includes predefined disease term annotation information and the number of historical diagnosis and treatment of the target disease and other diseases, and the obtaining of the physician capability evaluation value based on disease similarity includes:

[0041] Based on the disease term annotation information, a disease with an internal mechanism similar to the target disease is obtained, and based on the non-zero capability value of the target doctor for the internally similar disease, a third capability value of the target doctor for the target disease that the target doctor has not diagnosed and treated is obtained;

[0042] Based on the historical diagnosis and treatment times of the target disease and other diseases, diseases with similar external medical interactions to the target disease are obtained, and based on the non-zero capability value of the target doctor for the external similar disease, a third capability value of the target doctor for the target disease is obtained;

[0043] The third capability value and the fourth capability value are weightedly fused to obtain a doctor capability evaluation value based on disease similarity.

[0044] Preferably, the disease term annotation information at least includes genes annotated with terms in the disease ontology, and the process of obtaining the third capability value of the target doctor for the target disease includes:

[0045] Semantic measurement is used to estimate the intrinsic mechanism similarity between the target disease and other diseases; it is expressed as:

[0046]

[0047] Among them, MS u,v Represents the target disease represented by semantic measurement j Dis k The intrinsic mechanism similarity between them, subscript j≠k; target disease dis j Includes genes annotated with O terms, disease dis k Includes genes annotated with Q terms; Rs max (t jo ,t j ) indicates the target disease dis j The genes and diseases annotated by the oth term are k The maximum Rs value among all the genes annotated by the term; Rs max (t kq ,t k ) indicates disease k The qth term annotated gene is related to the target disease dis j The maximum Rs value among all the genes annotated by the term;

[0048] N(a∩b) represents the total number of genes annotated with both terms a and b in the disease ontology, and N(a∪b) represents the total number of genes annotated with terms a and / or b in the disease ontology; C(a) and C(b) represent the estimates of terms a and b as -log 2 The information content of p, P is the ratio of genes annotated with term a or b to the total number of genes annotated with any term;

[0049] Select MS j,k The disease corresponding to a positive number is dis k As neighbors, construct the target disease dis j The first neighbor set of:

[0050]

[0051] Where DIS = {dis j |j∈{1,...,n}} represents the set of disease types, n is the number of disease types;

[0052] Utilize the target doctor doc p right Intrinsically similar diseases k The non-zero capability value of the target doctor doc is obtained. p Disj The third ability value of ; expressed as:

[0053]

[0054] in, Indicates the target doctor doc p Dis j The third ability value of the superscript Dis-IS indicates the intrinsic similarity of the disease; a p,k Indicates the target doctor doc p Dis k A non-zero capability value.

[0055] Preferably, the process of acquiring the fourth capability value of the target doctor for the target disease includes:

[0056] The Pearson correlation coefficient is used to calculate the external medical interaction similarity between the target disease and other diseases; it is expressed as:

[0057]

[0058] Among them, DES j,k The target disease dis represented by Pearson correlation coefficient j Dis k Similarity of external medical experience between j,k Indicates that the target disease has been diagnosed and treated j Dis k A collection of doctors; Respectively represent the target disease dis j Dis k The average number of historical visits;

[0059] Select DES j,k Positive disease dis k As neighbors, construct the target disease dis j The second neighbor set of:

[0060]

[0061] Where DOC = {doc i |i∈{1,...,m}} represents the set of doctors, m is the number of doctors;

[0062] Utilize the target doctor doc p Yes Intrinsic similarity in disease k The non-zero capability value of the target doctor doc is obtained. p Dis jThe fourth ability value; expressed as:

[0063]

[0064] in, Indicates the target doctor doc p Dis j The fourth ability value of , the superscript Dis-ES indicates disease external similarity; Indicates the target doctor doc p The average non-zero capability value of .

[0065] A doctor recommendation system based on comprehensive ability evaluation under multi-source heterogeneous correlation analysis, including:

[0066] A generation module is used to obtain a set of doctors and a set of disease types, and use the number of patients corresponding to the diseases treated by the doctors as an indicator of their ability to generate a doctor ability matrix and perform normalization operations;

[0067] The filling module is used to perform the following operations on the zero values ​​in the normalized doctor ability matrix:

[0068] Based on the first multi-source heterogeneous data, obtaining a doctor ability evaluation value based on doctor similarity;

[0069] Based on the second multi-source heterogeneous data, obtaining a physician capability evaluation value based on disease similarity;

[0070] The doctor ability evaluation value based on doctor similarity and the doctor ability evaluation value based on disease similarity are weightedly integrated to obtain a similarity-driven comprehensive evaluation ability value, and fill in and replace the corresponding zero values ​​of the normalized doctor ability matrix;

[0071] The recommendation module is used to obtain the recommendation index of the target doctor based on the filled doctor capability matrix and the quantitative information of the severity and difficulty of diagnosis and treatment of different diseases.

[0072] A storage medium stores a computer program for recommending doctors based on comprehensive ability evaluation under multi-source heterogeneous correlation analysis, wherein the computer program enables a computer to execute the doctor recommendation method as described above.

[0073] An electronic device, comprising:

[0074] One or more processors; a memory; and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, the programs including a method for executing the doctor recommendation method as described above.

[0075] (III) Beneficial effects

[0076] The present invention provides a doctor recommendation method, system, storage medium and electronic device based on comprehensive ability evaluation under multi-source heterogeneous correlation analysis. Compared with the prior art, it has the following beneficial effects:

[0077] First, the present invention uses the number of patients corresponding to the diseases treated by the doctor as an indicator of his ability, generates a doctor ability matrix and performs normalization operation; secondly, for the zero value in the normalized doctor ability matrix, based on multi-source heterogeneous data, sequentially obtains the doctor ability evaluation value based on doctor similarity and disease similarity, and obtains the similarity-driven comprehensive evaluation ability value after weighted fusion; finally, based on the filled doctor ability matrix, combined with the quantitative information of the severity and treatment difficulty of different diseases, obtains the target doctor's recommendation index. Based on multi-source heterogeneous correlation analysis and fusion of doctor similarity and disease similarity, the present invention comprehensively evaluates the target doctor's ability value for the target disease that has not been treated, improves the target doctor's recommendation index, and significantly helps to rationally allocate limited medical resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0078] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0079] Figure 1 A flowchart of a doctor recommendation method based on comprehensive ability evaluation under multi-source heterogeneous correlation analysis provided by an embodiment of the present invention;

[0080] Figure 2 An example diagram of a doctor using HaoDF provided by an embodiment of the present invention;

[0081] Figures 3(a) to 3(d) These are comparison charts of MAE performance estimation using different methods tested in anorectal, urology, gastroenterology, and respiratory departments;

[0082] Figure 3(e) to 3(h) The following are comparison charts of RMSE performance estimation using different methods tested in anorectal, urological, gastroenterological, and respiratory departments. DETAILED DESCRIPTION

[0083] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention are clearly and completely described. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0084] The embodiments of the present application solve the technical problem of comprehensively evaluating doctors' abilities by providing a doctor recommendation method, system, storage medium and electronic device based on comprehensive ability evaluation under multi-source heterogeneous correlation analysis, which is conducive to timely recommending capable but neglected doctors to patients.

[0085] The technical solution in the embodiment of the present application is to solve the above technical problems, and the overall idea is as follows:

[0086] The embodiment of the present invention is used for, but not limited to, medical resource allocation. By comprehensively evaluating the real ability of doctors, it provides support for obtaining accurate doctor recommendation indexes and rankings. The main technical contributions of the solution are as follows:

[0087] (1) Based on doctor similarity and disease similarity, a new method for evaluating doctors' ability to diagnose and treat diseases for which they have no experience is proposed.

[0088] (2) Doctor similarity is further divided into internal similarity (essential characteristics) and external similarity (medical experience).

[0089] (3) Regarding disease similarity, it is proposed to estimate the internal similarity of diseases based on the mechanisms of diseases, and to measure the external similarity of diseases through the interaction between diseases and doctors.

[0090] (4) Based on the doctor similarity and disease similarity, the doctor-driven ability and disease-driven ability of each disease are estimated respectively, and they are integrated into the doctor's comprehensive evaluation ability value, forming the basis for doctor recommendation and ranking.

[0091] In order to better understand the above technical solution, the above technical solution will be described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0092] Embodiment 1:

[0093] like Figure 1 As shown, an embodiment of the present invention provides a doctor recommendation method based on comprehensive ability evaluation under multi-source heterogeneous correlation analysis, including:

[0094] S1. Obtain a set of doctors and a set of disease types, use the number of patients corresponding to the diseases treated by the doctors as the indicator of their ability, generate a doctor ability matrix and perform normalization operation;

[0095] S2. For the zero values ​​in the normalized doctor ability matrix, execute:

[0096] S21. Based on the first multi-source heterogeneous data, obtaining a doctor ability evaluation value based on doctor similarity;

[0097] S22, based on the second multi-source heterogeneous data, obtaining a doctor ability evaluation value based on disease similarity;

[0098] S23, weighted fusion of the doctor ability evaluation value based on doctor similarity and the doctor ability evaluation value based on disease similarity, obtain a similarity-driven comprehensive evaluation ability value, and fill in and replace the corresponding zero values ​​of the normalized doctor ability matrix;

[0099] S3. Based on the filled physician competency matrix, combined with the quantitative information on the severity of different diseases and the difficulty of diagnosis and treatment, obtain the recommendation index of the target physician.

[0100] The embodiment of the present invention is based on multi-source heterogeneous correlation analysis and integrates doctor similarity and disease similarity, comprehensively evaluates the ability value of the target doctor for the target disease that has not been diagnosed and treated, improves the recommendation index of the target doctor, and is significantly helpful for the rational allocation of limited medical resources.

[0101] Next, each step of the above scheme will be described in detail:

[0102] In step S1, a set of doctors and a set of disease types are obtained, and the number of patients corresponding to the diseases treated by the doctors is used as an indicator of their ability to generate a doctor ability matrix and perform a normalization operation.

[0103] In this step, we first define:

[0104] DOC = {doc i |i∈{1,...,m}} represents the set of doctors, and m is the number of doctors.

[0105] DIS = {dis j |j∈{1,...,n}} represents the set of disease types, and n is the number of disease types.

[0106] Considering that the doctor's ability can be improved by accumulating experience through diagnosis and treatment of patients, the number of patients corresponding to the diseases (types) that the doctor has diagnosed and treated is then used as an indicator of his ability, resulting in the doctor's ability matrix shown in Table 1.

[0107] Table 1 Doctors' competence matrix

[0108]

[0109] In Table 1, A = {a i,j|i∈{1,...,m},j∈{1,...,n}} represents the physician competence matrix associated with the set DOC and DIS, a i,j Indicates doctor doc i Dis j The ability value is doctor doc i Diagnosed diseases j The corresponding number of patients.

[0110] Furthermore, in order to facilitate subsequent estimation, a i,j Normalized to [0,1], as follows:

[0111]

[0112] in, and Respectively indicate the doctor's j The maximum and minimum capabilities of the .

[0113] In order to estimate the comprehensive ability of doctors, the embodiment of the present invention estimates the ability of the "missing" doctors in these matrices according to the similarities between doctors and diseases, as shown in step S2:

[0114] In step S2, for the zero values ​​in the normalized doctor ability matrix, the following operations are performed:

[0115] S21. Based on the first multi-source heterogeneous data, obtain a doctor ability evaluation value based on doctor similarity.

[0116] In an optional embodiment, the first multi-source heterogeneous data includes pre-assigned doctor attribute information and non-zero ability values ​​of the target doctor and other doctors in a normalized doctor ability matrix.

[0117] Accordingly, the step of obtaining the doctor ability evaluation value based on doctor similarity includes the following steps:

[0118] S101. Based on the doctor attribute information, obtain doctors with similar internal basic attributes to the target doctor, and based on the non-zero capability values ​​of the internal similar doctors for the target disease that the target doctor has not diagnosed and treated, obtain the first capability value of the target doctor for the target disease.

[0119] S102. Based on the non-zero capability values ​​of the target doctor and other doctors, obtain doctors with similar external medical experience to the target doctor, and based on the non-zero capability values ​​of the external similar doctors for the target disease, obtain a second capability value of the target doctor for the target disease.

[0120] S103: Weightedly fuse the first capability value and the second capability value to obtain a doctor capability evaluation value based on doctor similarity.

[0121] In step S101, the first capability value of the target doctor for the target disease is obtained, which specifically includes:

[0122] Considering doctors with similar basic attributes, such as the hospital where they work, the level of their seniority, and the number of patients they have, their medical abilities are usually similar, as these attributes can reflect their level of experience in the medical field. For example, a chief physician with extensive medical experience in a high-level hospital should be very competent and skilled.

[0123] On this basis, the embodiment of the present invention sets the doctor's attribute information to include the working hospital, qualification level and the number of patients treated.

[0124] Specifically, all hospitals are divided into three levels, namely "A", "B" and "C". Each level has three sub-levels, namely "AAA", "AA" and "A", among which "AAA" is the highest level. And, in order to measure the similarity of doctors based on their internal basic attributes, Table 2 quantifies the hospital where they work and the doctor's qualification level using numbers from 0 to 1. For example, for the hospital where they work, a high level is assigned a large number, and the doctor's qualification level is also assigned a large number. In addition, in order to further refine the similarity measurement, the number of patients of the doctor is standardized to the same range as the doctor's hospital and qualification level.

[0125] Assume IC i ={wh i ,dl i ,pn i} represents the attribute vector of doctor doci, wh i ,dl i ,pn i They represent the doctor's working hospital, qualification level, and number of patients he has treated, as shown in Table 2:

[0126] Table 2 Basic attributes of doctors

[0127]

[0128]

[0129] For the target physician, the goal here is to estimate their competence for the disease without medical experience based on the experience of physicians with similar internal basic attributes to the target physician.

[0130] First, the cosine distance is used to calculate the similarity of the internal basic attributes between the target doctor and other doctors; it is expressed as:

[0131]

[0132] Among them, DIS p,q Represents the target doctor doc represented by cosine distance p With doctor doc q The similarity of the internal basic attributes between them, subscript p≠q; ||·|| represents the vector length; IC i ={wh i ,dl i ,pn i} indicates doctor doc i The internal basic properties of wh i ,dl i ,pn i Respectively represent doctor doc i The standard values ​​for the hospital where the candidate worked, the level of experience, and the number of patients treated.

[0133] It can be understood that the DIS obtained above p,q In the interval [0,1], DIS p,q The larger the value, the higher the similarity of the internal basic attributes of the two compared doctors.

[0134] Second, select DIS p,q The doctor corresponding to the positive number is doc q As a neighbor, build the target doctor doc p The first neighbor set of:

[0135]

[0136] Where DOC = {doc i |i∈{1,...,m}} represents the set of doctors, and m is the number of doctors.

[0137] Finally, using Internal similarity in doctordoc q Dis j The non-zero capability value of the target doctor doc is obtained. p Dis j The first ability value of; expressed as:

[0138]

[0139] in, Indicates the target doctor doc p Dis j The first capability value of Doc-IS indicates the internal similarity of doctors; a q,j Indicates internal similarity doctor doc q Dis j A non-zero capability value.

[0140] In step S102, the second capability value of the target doctor for the target disease is obtained, which specifically includes:

[0141] In addition to the internal basic attributes of doctors, the existing doctor capabilities (i.e., non-zero values ​​in the doctor capability matrix) can reflect the doctor relationship from an external perspective. Specifically, if two doctors have similar capabilities for certain diseases, then their medical skills will also be similar, and thus they will be able to diagnose and treat other diseases that require similar skill levels. Since this logic is the same as the collaborative filtering technology used in the recommendation system, the embodiment of the present invention introduces this technology to estimate the doctor's capabilities based on the external similarity of the doctors.

[0142] For the target physician, the goal here is to estimate their competence for the disease without medical experience based on the experience of physicians with similar external medical experience to the target physician.

[0143] First, since the Pearson correlation coefficient (PCC) is widely used to estimate the degree of linear association between two entities, the Pearson correlation coefficient is used to calculate the similarity of external medical experience between the target doctor and other doctors; it is expressed as:

[0144]

[0145] Among them, DES p,q Represents the target doctor doc represented by the Pearson correlation coefficient p With doctor doc q Similarity of external medical experience between p,q Indicates the target doctor doc p With doctor doc q Types of diseases treated; Respectively represent the target doctor doc p With doctor doc q The average non-zero capability value of .

[0146] It should be noted that by the set DOC = {doc i Each doctor in |i∈{1,...,m}} performs the above calculations, and the embodiment of the present invention can generate a doctor's external medical experience similarity matrix, as shown in Table 3:

[0147] Table 3 Similarity matrix of doctors' external medical experience

[0148]

[0149]

[0150] Understandably, due to DES p,q and DES q,pThe matrix shown in Table 3 is a symmetric matrix whose diagonal elements are all equal to 1.

[0151] Second, select DES p,q A positive number for doctor doc q As a neighbor, build the target doctor doc p The second neighbor set of:

[0152]

[0153] Where DIS = {dis j |j∈{1,...,n}} represents the set of disease types, and n is the number of disease types.

[0154] Finally, using Internal similarity in doctordoc q Dis j The non-zero capability value of the target doctor doc is obtained. p Dis j The second ability value of; expressed as:

[0155]

[0156] in, Indicates the target doctor doc p Dis j The second capability value of , the superscript Doc-IS represents the doctor external similarity.

[0157] In step S103, obtaining the doctor ability evaluation value based on the doctor similarity includes:

[0158] It should be noted that a Doc-IS and a Doc-ES The similarity of the doctor's ability between two different doctors is estimated from the perspective of internal basic attributes and external medical experience. These are combined together as shown in the following formula:

[0159]

[0160] in, Represents the target doctor doc obtained based on doctor similarity p Dis j The physician competence assessment value is Indicates a Doc-IS and a Doc-ES The combination of , whose value falls in the interval [0,1]; α represents the weight parameter.

[0161] S22. Based on the second multi-source heterogeneous data, obtain a doctor ability evaluation value based on disease similarity.

[0162] In an optional embodiment, the second multi-source heterogeneous data includes pre-defined disease term annotation information and the historical diagnosis and treatment times of the target disease and other diseases.

[0163] Accordingly, the step of obtaining the doctor's ability evaluation value based on disease similarity includes the following steps:

[0164] S201. Based on the disease term annotation information, diseases with similar internal mechanisms to the target disease are obtained, and based on the target doctor's non-zero ability value for the internally similar disease, a third ability value of the target doctor for the target disease that the target doctor has not diagnosed and treated is obtained.

[0165] S202. Based on the historical number of times the target disease and other diseases have been diagnosed and treated, diseases with similar external medical interactions to the target disease are obtained, and based on the non-zero capability value of the target doctor for the external similar diseases, a third capability value of the target doctor for the target disease is obtained.

[0166] S203: Weightedly fuse the third capability value and the fourth capability value to obtain a doctor capability evaluation value based on disease similarity.

[0167] In step S201, the third capability value of the target doctor for the target disease is obtained, which specifically includes:

[0168] Given that some diseases have similar mechanisms, such as pathogenesis and genes, their diagnosis and treatment share similarities. That is, if a doctor is good at treating a disease, they should be good at treating diseases similar to that disease. Therefore, the doctor's existing ability in similar diseases can be used to estimate the doctor's ability in diseases for which the doctor has no medical experience. Disease mechanism (or internal) similarity is often estimated using ontological metrics in semantic measurements between gene ontology processes, which are determined based on the genes known to be involved and are related to different diseases.

[0169] On this basis, the embodiment of the present invention sets the disease term annotation information to at least include genes annotated with terms in the disease ontology.

[0170] For the target physician, the goal here is to estimate their competence in the absence of medical experience with the disease based on their experience with diseases that have similar underlying mechanisms to the target disease.

[0171] First, a semantic measure is used to estimate the intrinsic mechanism similarity between the target disease and other diseases; it is expressed as:

[0172]

[0173] Among them, MS u,vRepresents the target disease represented by semantic measurement j Dis k The intrinsic mechanism similarity between them, subscript j≠k; target disease dis j Includes genes annotated with O terms, disease dis k Includes genes annotated with Q terms; Rs max (t jo ,t j ) indicates the target disease dis j The genes and diseases annotated by the oth term are k The maximum Rs value among all the genes annotated by the term; Rs max (t kq ,t k ) indicates disease k The qth term annotated gene is related to the target disease dis j The maximum Rs value among all the genes annotated by the term;

[0174] N(a∩b) represents the total number of genes annotated with both terms a and b in the disease ontology, and N(a∪b) represents the total number of genes annotated with terms a and / or b in the disease ontology; C(a) and C(b) represent the estimates of terms a and b as -log 2 The information content of p is the ratio of genes annotated with term a or b to the total number of genes annotated with any term.

[0175] It should be noted that by the set DIS = {dis j The above calculation is performed for each disease in |j∈{1,...,n}}. The embodiment of the present invention can generate a disease internal mechanism similarity matrix, the content of which can refer to the doctor's external medical experience similarity matrix given in Table 3, which will not be repeated here.

[0176] Second, choose MS j,k The disease corresponding to a positive number is dis k As neighbors, construct the target disease dis j The first neighbor set of:

[0177]

[0178] Where DIS = {dis j |j∈{1,...,n}} represents the set of disease types, and n is the number of disease types.

[0179] Finally, using the target doctor doc p right Intrinsically similar diseases k The non-zero capability value of the target doctor doc is obtained. pDis j The third ability value of ; expressed as:

[0180]

[0181] in, Indicates the target doctor doc p Dis j The third ability value of the superscript Dis-IS indicates the intrinsic similarity of the disease; a p,k Indicates the target doctor doc p Dis k A non-zero capability value.

[0182] In step S202, the fourth capability value of the target doctor for the target disease is obtained, which specifically includes:

[0183] In addition to the similarity of the intrinsic mechanisms of diseases, the existing information in the doctor's ability matrix can reflect the disease similarity relationship from another external perspective. For example, if most doctors have similar abilities in two diseases, then the diagnosis and treatment of the two diseases have strong external similarities. Therefore, if a doctor has experience in diagnosing and treating one of the diseases but not the other, then the doctor's existing ability can be used to estimate his ability in the other disease. Similar to estimating the external similarity of doctors, the embodiment of the present invention introduces collaborative filtering to estimate the doctor's ability based on the intrinsic similarity relationship of diseases.

[0184] For the target physician, the goal here is to estimate their competence for the disease without medical experience based on the experience of physicians who have similar external medical interactions to the target physician.

[0185] First, the Pearson correlation coefficient is used to calculate the external medical interaction similarity between the target disease and other diseases; it is expressed as:

[0186]

[0187] Among them, DES j,k The target disease dis represented by Pearson correlation coefficient j Dis k Similarity of external medical experience between j,k Indicates that the target disease has been diagnosed and treated j Dis k A collection of doctors; Respectively represent the target disease dis j Dis k The average number of historical visits.

[0188] Second, select DES j,k Positive disease dis k As neighbors, construct the target disease dis j The second neighbor set of:

[0189]

[0190] Where DOC = {doc i |i∈{1,...,m}} represents the set of doctors, and m is the number of doctors.

[0191] Finally, using the target doctor doc p Yes Intrinsic similarity in disease k The non-zero capability value of the target doctor doc is obtained. p Dis j The fourth ability value; expressed as:

[0192]

[0193] in, Indicates the target doctor doc p Dis j The fourth ability value of , the superscript Dis-ES indicates disease external similarity; Indicates the target doctor doc p The average non-zero capability value of .

[0194] In step S203, obtaining the doctor's ability evaluation value based on disease similarity specifically includes:

[0195] It should be noted that and The similarity of the physician capabilities of two different doctors is estimated from the perspectives of internal mechanisms and external medical interactions. These are combined together as shown in the following formula:

[0196]

[0197] in, Represents the target doctor doc obtained based on disease similarity p Dis j The physician competence assessment value is express and The combination of , whose value falls in the interval [0,1]; β represents the weight parameter.

[0198] S23. Weighted fusion of the doctor ability evaluation value based on doctor similarity and the doctor ability evaluation value based on disease similarity to obtain a similarity-driven comprehensive evaluation ability value, and fill in and replace the corresponding zero values ​​of the normalized doctor ability matrix.

[0199] The doctor ability evaluation value based on doctor similarity and the doctor ability evaluation value based on disease similarity are weighted and fused to obtain a similarity-driven comprehensive evaluation ability value, which is expressed as:

[0200]

[0201] in, Indicates the target doctor doc p For its undiagnosed target disease j A comprehensive estimate of the physician's ability value; γ represents the weight parameter, which has similar meaning and function to α and β.

[0202] On this basis, the obtained comprehensive evaluation ability values ​​can be used to fill in and replace the corresponding zero values ​​of the normalized doctor ability matrix.

[0203] In step S3, based on the filled physician capability matrix and combined with the quantitative information of severity and diagnosis and treatment difficulty of different diseases, the recommendation index of the target physician is obtained.

[0204] For the target doctor, based on the filled doctor capability matrix, the recommendation index can be calculated as follows:

[0205]

[0206] Among them, DA p represents the target doctor's recommendation index, sd j Indicates the target disease dis j Quantitative information on the severity of the disease and the difficulty of treatment.

[0207] On this basis, the embodiment of the present invention can sort all doctors in non-ascending order based on the DA value of each doctor, and recommend them to the corresponding patients.

[0208] So far, the embodiment of the present invention has completed the entire process of the doctor recommendation method based on comprehensive ability evaluation under multi-source heterogeneous correlation analysis.

[0209] In order to help understand the superiority of the doctor recommendation method provided by the embodiment of the present invention, the following experimental description is provided:

[0210] 1) Data source

[0211] In order to test the performance of the proposed physician competence evaluation and physician recommendation, data were collected from multiple sources, including physician basic information, physician medical experience, and disease ontology information. The first two parts of the data come from the HaoDF online platform. As of 2019, HaoDF has collected information on 610,000 doctors from 9,917 hospitals in China. Among them, 240,000 doctors have registered with their real names on the platform and provide direct online medical services to patients. The basic information and medical experience of doctors can be obtained from the platform.

[0212] Figure 2 An example of a doctor using HaoDF is shown. Specifically, "Zhang*" who works at XXXX Hospital in XX City is a chief anorectal doctor who has treated 56 patients, including 43 patients with hemorrhoids, 8 patients with perianal abscesses, and 5 patients with anal fistulas. In addition, the level of his hospital is "BB".

[0213] The public information of doctors from the HaoDF official website from January 2010 to September 2020 was obtained and data processed. The basic information consists of the doctor's "name", "doctor level", "working hospital (level)" and "clinical department". In addition, the number of patients with each disease was extracted as medical experience. In addition, in order to test the robustness and universality of the proposed ability assessment and doctor recommendation, four clinical departments, anorectal department, urology department, gastroenterology department and respiratory department, were selected for experiments. Table 4 gives the statistical information of the dataset:

[0214] Table 4 Dataset statistics (4 clinical departments in XX city)

[0215]

[0216] As shown in Table 4, the anorectal department contains 261 doctors and 22 diseases, forming a 261*22 doctor capability matrix with a density of 51.95%. In this matrix, there are 5742 doctor skills, and the goal is to estimate the remaining 2759 doctor skills. The statistics of the other three clinical departments are similar to those of the anorectal department, and the "missing" doctor capabilities account for about 50% of all doctor capabilities in each clinical department. For disease-related information, the topological structure of the disease in the disease ontology was used, and the UMLS (MeSH, SNOMED-CT, ICD9) provided by Mathur was used. Use the disease ontology (DO ver.3) vocabulary.

[0217] 2) Performance indicators

[0218] 2.1) Doctors’ ability assessment indicators

[0219] In the proposed model, doctor ability evaluation is a key step that directly affects doctor ranking and recommendation. Mean absolute error (MAE) is widely used to measure the error between the estimated value and the actual value. It is introduced to measure the estimation accuracy as follows:

[0220]

[0221] Among them, a i,j and They represent the sample doctor's ability (i.e., medical experience) and the estimated doctor's ability, respectively. f(X) represents a threshold function, which is equal to 0 only when x=0, and is equal to 1 in all other cases.

[0222] Compared to MAE, root mean square error (RMSE) is another commonly used error measurement method, which amplifies relatively large errors and reduces relatively small errors according to the following formula:

[0223]

[0224] In the experiment, the above two indicators are used to evaluate the ability of doctors. The values ​​of these two indicators fall in the interval [0,1], and the lower the value, the better the performance.

[0225] 2.2) Doctor recommended indicators

[0226] The purpose of physician ranking and recommendation is to allocate limited medical resources more rationally. In order to test the rationality, a physician ranking benchmark based on the preference of most people for choosing physicians should be proposed first. Based on this obvious preference for hospital level, a benchmark physician ranking can be generated and show how patients choose the top-ranked physician. Table 5 shows an example of generating a benchmark ranking using information from the dataset:

[0227] Table 5 Benchmark rankings of 5 anorectal doctors

[0228] doctor Working hospital (level) Doctor level Number of patients Ranking <![CDATA[doc 8 ]]> A Chief Physician 1232 3 <![CDATA[doc 24 ]]> AAA Attending Physician 428 2 <![CDATA[doc 104 ]]> A Associate Chief Physician 2198 4 <![CDATA[doc 131 ]]> AAA Chief Physician 1092 1 <![CDATA[doc 198 ]]> A Associate Chief Physician 1901 5

[0229] There are 5 doctors in the anorectal department, including doc 8 ,doc 24 ,doc 104 ,doc 131 ,doc 198 Doctor No. 131 is the chief physician of a Class A tertiary hospital and ranks first. 24 Working in a high-level hospital, although the doctor is an attending physician and has relatively less medical experience, their allocation ranking is second. The other 3 doctors in the "A" hospital, regardless of their medical experience less than doc 104 and doc 198 , patients are more inclined to choose chief physician doc8 . Selection of deputy chief physicians in hospitals of the same level 104 and doc 198 Only then do patients begin to consider their medical experience.

[0230] In addition, the commonly used ordinal correlation index, Kendall Rank (KRCC), is introduced in the difference measurement as follows:

[0231]

[0232] Among them, DOC i represents the set of doctors in clinical department i, |DOC i | represents the number of elements in the collection. and They represent the recommendation index and the generated ranking of two doctors who rank in the same position (i.e., PST) in the benchmark. g(x) is a binary threshold function, which is equal to 0 (x<0) or 1 (x>0), and KR is in the range of 0 to 1. The larger the KR value, the greater the difference between the two rankings, indicating that the allocation of medical resources is more reasonable.

[0233] 3) Experimental results

[0234] The proposed physician competence evaluation is compared with other commonly used state-of-the-art evaluation methods, including user-based collaborative filtering (UCF), item-based collaborative filtering (ICF), support vector machine (SVM), matrix factorization (MF), and neural network (NN). Since there are 3 weighting parameters in the proposed similarity-driven physician competence estimation (SAE), multiple models can be derived, as shown in Table 6:

[0235] Table 6 SAE model with different parameter settings

[0236]

[0237] In the comparative experiments, α, β, and γ are all set to 0.5, which means that all similarities have the same weight in the evaluation of doctor ability. When γ = 0, β can be set to any value because the doctor's disease-driven ability is not considered. SAE is simplified to Doc-SAE, which is based only on doctor similarity. Doc-SAE can be further simplified to Doc-I-SAE (i.e., α = 1) and Doc-E-SAE (i.e., α = 0), which consider internal similarity and external similarity, respectively. It is worth noting that Doc-E-SAE is exactly the same as UCF. γ = 1 means that only disease-driven doctor ability is considered, and other models with this γ setting are similar to the model with γ = 0.

[0238] Specifically, the proposed SAE, doctor-driven physician ability estimation (Doc-SAE), physician ability estimation based on internal similarity of physicians (Doc-I-SAE), disease-driven physician ability estimation (Dis-SAE), and disease external similarity-based physician ability estimation (Dis-I-SAE) are compared with UCF, ICF, SVM, MF, and NN on the physician ability matrices of anorectal, urology, gastroenterology, and respiratory departments. In these experiments, 10% to 90% (every 10%) of the sample physician abilities are randomly selected as training sets, and the rest are used as test sets.

[0239] Figure 3(a) to 3(h) The estimated performance is shown on four medical departments and is measured by MAE (corresponding to Figures 3(a) to 3(d) ) and RMSE (corresponding to Figure 3(e) to 3(h) ) were evaluated. In general, all the curves in these figures decrease as the data density increases. This verifies the validity and reliability of the dataset and proves the value of these models in the evaluation of physician competence. In fact, more data provides more information to make more accurate estimates.

[0240] As shown in Figure 3(a), all proposed models (i.e., curves with open shapes) outperform other compared models (i.e., curves with solid shapes). In particular, SAE performs significantly better than other proposed models. Among the four methods, Doc-SAE and Dis-SAE have similar performance and outperform Doc-I-SAE and Dis-I-SAE. It is worth noting that the doctor-driven models (i.e., Doc-SAE and Doc-I-SAE) slightly outperform the corresponding disease-driven models (Dis-SAE and Doc-I-SAE). These results indicate that the evaluation of doctor ability should be conducted from two aspects: doctor similarity and disease similarity. Doctor-driven ability includes the basic attributes of doctors and more accurately reflects the ability of doctors.

[0241] Among the compared models, the performance relationship of UCF (i.e., Doc-E-SAE) and ICF (i.e., Dis-E-SAE) is similar to that of Doc-SAE and Dis-SAE, while UCF and ICF outperform SVM, NN, and MF. In terms of the similarity between doctors and diseases, the intrinsic similarity based on the basic attributes of the doctors themselves is always more significant than the external similarity based on the doctors' existing abilities. However, external similarity cannot be abandoned because SAE based on both internal and external similarities performs best in the experiment. Although SVM, MF, and NN generally perform well in predicting missing information, they are unsatisfactory in estimating doctors' abilities. This is because these three models only predict missing data based on the data already available in the matrix, ignoring the context of the problem and the meaning of the data.

[0242] The experimental results evaluated by MAE in urology (Figure 3(b)), gastroenterology (Figure 3(c)), and respiratory medicine (Figure 3(d)) are similar to the conclusions of anorectal medicine (Figure 3(d)) (Figure 3(a)). This shows that the above analysis and conclusions are reliable. However, the fluctuation range of the RMSE curve is greater than that of the MAE curve. Nevertheless, the experimental results evaluated by RMSE are similar to those evaluated by MAE, which improves reliability.

[0243] Embodiment 2:

[0244] The embodiment of the present invention provides a doctor recommendation system based on comprehensive ability evaluation under multi-source heterogeneous correlation analysis, including:

[0245] A generation module is used to obtain a set of doctors and a set of disease types, and use the number of patients corresponding to the diseases treated by the doctors as an indicator of their ability to generate a doctor ability matrix and perform normalization operations;

[0246] The filling module is used to perform the following operations on the zero values ​​in the normalized doctor ability matrix:

[0247] Based on the first multi-source heterogeneous data, obtaining a doctor ability evaluation value based on doctor similarity;

[0248] Based on the second multi-source heterogeneous data, obtaining a physician capability evaluation value based on disease similarity;

[0249] The doctor ability evaluation value based on doctor similarity and the doctor ability evaluation value based on disease similarity are weightedly integrated to obtain a similarity-driven comprehensive evaluation ability value, and fill in and replace the corresponding zero values ​​of the normalized doctor ability matrix;

[0250] The recommendation module is used to obtain the recommendation index of the target doctor based on the filled doctor capability matrix and the quantitative information of the severity and difficulty of diagnosis and treatment of different diseases.

[0251] Embodiment 3:

[0252] An embodiment of the present invention provides a storage medium storing a computer program for recommending doctors based on comprehensive ability assessment under multi-source heterogeneous correlation analysis, wherein the computer program enables a computer to execute the doctor recommendation method as described in Example 1.

[0253] Embodiment 4:

[0254] An embodiment of the present invention provides an electronic device, including:

[0255] One or more processors; a memory; and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, the programs including a method for executing the doctor recommendation method as described in Example 1.

[0256] It can be understood that the doctor recommendation system, storage medium and electronic device based on comprehensive ability evaluation under multi-source heterogeneous correlation analysis provided in the embodiments of the present invention correspond to the doctor recommendation method based on comprehensive ability evaluation under multi-source heterogeneous correlation analysis provided in the embodiments of the present invention. The explanations, examples and beneficial effects of the relevant contents can refer to the corresponding parts in the doctor recommendation method and will not be repeated here.

[0257] In summary, compared with the prior art, the present invention has the following beneficial effects:

[0258] The embodiment of the present invention is based on multi-source heterogeneous correlation analysis and integrates doctor similarity and disease similarity to comprehensively evaluate the target doctor's ability value for the target disease that has not been diagnosed and treated, and improve the target doctor's recommendation index. Experiments have shown that it is significantly helpful for the rational allocation of limited medical resources.

[0259] 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. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0260] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A doctor recommendation method based on comprehensive ability evaluation under multi-source heterogeneous correlation analysis, characterized in that: include: Obtain a set of doctors and a set of disease types, use the number of patients corresponding to the diseases treated by the doctor as an indicator of his / her ability, generate a doctor ability matrix and perform normalization operations; For the zero values ​​in the normalized doctor capability matrix, execute: Based on the first multi-source heterogeneous data, obtaining a doctor ability evaluation value based on doctor similarity; Based on the second multi-source heterogeneous data, obtaining a physician capability evaluation value based on disease similarity; The doctor ability evaluation value based on doctor similarity and the doctor ability evaluation value based on disease similarity are weightedly integrated to obtain a similarity-driven comprehensive evaluation ability value, and fill in and replace the corresponding zero values ​​of the normalized doctor ability matrix; Based on the filled physician competency matrix and combined with quantitative information on the severity and difficulty of diagnosis and treatment of different diseases, the recommendation index of the target physician is obtained.

2. The doctor recommendation method according to claim 1, characterized in that: The first multi-source heterogeneous data includes pre-assigned doctor attribute information and non-zero ability values ​​of the target doctor and other doctors in the normalized doctor ability matrix, and the obtaining of the doctor ability evaluation value based on doctor similarity includes: Based on the doctor attribute information, obtain doctors with similar internal basic attributes to the target doctor, and based on the non-zero capability values ​​of the internal similar doctors for the target disease that has not been diagnosed and treated by the target doctor, obtain the first capability value of the target doctor for the target disease; Based on the non-zero capability values ​​of the target doctor and other doctors, obtaining doctors with similar external medical experience to the target doctor, and based on the non-zero capability values ​​of the external similar doctors for the target disease, obtaining a second capability value of the target doctor for the target disease; The first capability value and the second capability value are weightedly fused to obtain a doctor capability evaluation value based on doctor similarity.

3. The doctor recommendation method according to claim 2, characterized in that: The doctor attribute information includes the hospital where he works, the level of experience, and the number of patients he has treated. The process of obtaining the first capability value includes: The cosine distance is used to calculate the similarity of the internal basic attributes between the target doctor and other doctors; it is expressed as: Among them, DIS p,q Represents the target doctor doc represented by cosine distance p With doctor doc q The similarity of the internal basic attributes between them, subscript p≠q; ||·|| represents the vector length; IC i ={wh i ,dl i ,pn i } indicates doctor doc i The internal basic properties of wh i ,dl i ,pn i Respectively represent doctor doc i standard values ​​for the hospital where the candidate worked, level of experience, and number of patients treated; Select DIS p,q The doctor corresponding to the positive number is doc q As a neighbor, build the target doctor doc p The first neighbor set of: Where DOC = {doc i |i∈{1,…,m}} represents the set of doctors, m is the number of doctors; Use in Internal similarity in doctordoc q Dis j The non-zero capability value of the target doctor doc is obtained. p Dis j The first ability value of; expressed as: in, Indicates the target doctor doc p Dis j The first capability value of Doc-IS indicates the internal similarity of doctors; a q,j Indicates internal similarity doctor doc q Dis j A non-zero capability value.

4. The doctor recommendation method according to claim 3, characterized in that: The process of acquiring the second capability value of the target doctor for the target disease includes: The Pearson correlation coefficient is used to calculate the similarity of external medical experience between the target doctor and other doctors; it is expressed as: Among them, DES p,q Represents the target doctor doc represented by the Pearson correlation coefficient p With doctor doc q Similarity of external medical experience between p,q Indicates the target doctor doc p With doctor doc q Types of diseases treated; Respectively represent the target doctor doc p With doctor doc q The average non-zero capability value of Select DES p,q A positive number for doctor doc q As a neighbor, build the target doctor doc p The second neighbor set of: Where DIS = {dis j |j∈{1,…,n}} represents the set of disease types, and n is the number of disease types; Use in Internal similarity in doctordoc q Dis j The non-zero capability value of the target doctor doc is obtained. p Dis j The second ability value of; expressed as: in, Indicates the target doctor doc p Dis j The second capability value of , the superscript Doc-IS represents the doctor external similarity.

5. The doctor recommendation method according to claim 1, characterized in that: The second multi-source heterogeneous data includes pre-defined disease term annotation information and the number of historical diagnosis and treatment of the target disease and other diseases, and the obtaining of the physician capability evaluation value based on disease similarity includes: Based on the disease term annotation information, a disease with an internal mechanism similar to the target disease is obtained, and based on the non-zero capability value of the target doctor for the internally similar disease, a third capability value of the target doctor for the target disease that the target doctor has not diagnosed and treated is obtained; Based on the historical diagnosis and treatment times of the target disease and other diseases, diseases with similar external medical interactions to the target disease are obtained, and based on the non-zero capability value of the target doctor for the external similar disease, a third capability value of the target doctor for the target disease is obtained; The third capability value and the fourth capability value are weightedly fused to obtain a doctor capability evaluation value based on disease similarity.

6. The doctor recommendation method according to claim 5, characterized in that: The disease term annotation information at least includes genes annotated with terms in the disease ontology, and the target doctor's third capability value acquisition process for the target disease includes: Semantic measurement is used to estimate the intrinsic mechanism similarity between the target disease and other diseases; it is expressed as: Among them, MS u,v Represents the target disease represented by semantic measurement j Dis k The intrinsic mechanism similarity between them, subscript j≠k; target disease dis j Includes genes annotated with O terms, disease dis k Includes genes annotated with Q terms; Rs max (t jo ,t j ) indicates the target disease dis j The genes and diseases annotated by the oth term are k The maximum Rs value among all the genes annotated by the term; Rs max (t kq ,t k ) indicates disease k The qth term annotated gene is related to the target disease dis j The maximum Rs value among all the genes annotated by the term; N(a∩b) represents the total number of genes annotated with both terms a and b in the disease ontology, and N(a∪b) represents the total number of genes annotated with terms a and / or b in the disease ontology; C(a) and C(b) represent the information content of terms a and b estimated as -log2p, respectively, and P is the ratio of genes annotated with terms a or b to the total number of genes annotated with any term; Select MS j,k The disease corresponding to a positive number is dis k As neighbors, construct the target disease dis j The first neighbor set of: Where DIS = {dis j |j∈{1,...,n}} represents the set of disease types, n is the number of disease types; Utilize the target doctor doc p right Similar diseases in k The non-zero capability value of the target doctor doc is obtained. p Dis j The third ability value of ; expressed as: in, Indicates the target doctor doc p Dis j The third ability value of the superscript Dis-IS indicates the intrinsic similarity of the disease; a p,k Indicates the target doctor doc p Dis k A non-zero capability value.

7. The doctor recommendation method according to claim 6, characterized in that: The process of obtaining the fourth capability value of the target doctor for the target disease includes: The Pearson correlation coefficient is used to calculate the external medical interaction similarity between the target disease and other diseases; it is expressed as: Among them, DES j,k The target disease dis represented by Pearson correlation coefficient j Dis k Similarity of external medical experience between j,k Indicates that the target disease has been diagnosed and treated j Dis k A collection of doctors; Respectively represent the target disease dis j Dis k The average number of historical visits; Select DES j,k Positive disease dis k As neighbors, construct the target disease dis j The second neighbor set of: Where DOC = {doc i |i∈{1,…,m}} represents the set of doctors, m is the number of doctors; Utilize the target doctor doc p Yes Intrinsic similarity in disease k The non-zero capability value of the target doctor doc is obtained. p Dis j The fourth ability value of; expressed as: in, Indicates the target doctor doc p Dis j The fourth ability value of , the superscript Dis-ES indicates disease external similarity; Indicates the target doctor doc p The average non-zero capability value of .

8. A doctor recommendation system based on comprehensive ability evaluation under multi-source heterogeneous correlation analysis, characterized in that: include: A generation module is used to obtain a set of doctors and a set of disease types, and use the number of patients corresponding to the diseases treated by the doctors as an indicator of their ability to generate a doctor ability matrix and perform normalization operations; The filling module is used to perform the following operations on the zero values ​​in the normalized doctor ability matrix: Based on the first multi-source heterogeneous data, obtaining a doctor ability evaluation value based on doctor similarity; Based on the second multi-source heterogeneous data, obtaining a physician capability evaluation value based on disease similarity; The doctor ability evaluation value based on doctor similarity and the doctor ability evaluation value based on disease similarity are weightedly integrated to obtain a similarity-driven comprehensive evaluation ability value, and fill in and replace the corresponding zero values ​​of the normalized doctor ability matrix; The recommendation module is used to obtain the recommendation index of the target doctor based on the filled doctor capability matrix and the quantitative information of the severity and difficulty of diagnosis and treatment of different diseases.

9. A storage medium, characterized in that: It stores a computer program for recommending doctors based on comprehensive ability evaluation under multi-source heterogeneous correlation analysis, wherein the computer program enables the computer to execute the doctor recommendation method as described in any one of claims 1 to 7.

10. An electronic device, characterized in that: include: one or more processors; Memory; and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the programs including instructions for executing the doctor recommendation method according to any one of claims 1 to 7.

Images