Information recommendation method based on bipartite graph, information recommendation device and terminal equipment

Abstract

The application is suitable for the field of artificial intelligence technology, and provides an information recommendation method based on bipartite graph, an information recommendation device based on bipartite graph, a terminal device and a storage medium. The generation method comprises the following steps: determining a hidden danger probability of each preset health hidden danger existing for a user based on preprocessed health information, and the health hidden danger is at least one; constructing a first bipartite graph based on the health hidden danger and at least two preset information elements; constructing a second bipartite graph based on the information elements and information element combinations, wherein the information element combination comprises any one information element or a combination formed by the information elements; and determining a target information element combination based on the hidden danger probability, the first bipartite graph and the second bipartite graph, wherein the target information element combination is used for generating health suggestion information recommended to the user. The method can provide personalized health suggestions for the user based on the health information of the user, and improve the accuracy of health suggestion recommendation. In addition, the application also relates to the blockchain technology.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, and in particular to a bipartite graph-based information recommendation method, a bipartite graph-based information recommendation device, a terminal device, and a computer readable storage medium. BACKGROUND

[0002] Nowadays, more and more people pay attention to their physical conditions and are willing to spend time and money on health examinations in hospitals. There are various types of health examination projects, and a complete health examination will generate a large number of electronic medical records, which contain rich health information. The mining of such health information can assist doctors in quickly providing health recommendation information.

[0003] The current health recommendation information recommendation method is that a doctor first determines auxiliary information according to the electronic medical records of a patient, then queries health recommendation information matching the auxiliary information from a database based on the auxiliary information, and finally recommends the health recommendation information to the patient. In this process, the health recommendation information is obtained by using a template, but the actual conditions of different patients may be different, so this method cannot provide personalized health recommendation information for patients, resulting in insufficient accuracy of health recommendation information recommendation. SUMMARY

[0004] Therefore, the embodiments of the present application provide a bipartite graph-based information recommendation method, a bipartite graph-based information recommendation device, a terminal device, and a computer readable storage medium, which can provide personalized health recommendations for users based on the health information of the users and improve the accuracy of health recommendation.

[0005] A first aspect of the embodiments of the present application provides a bipartite graph-based information recommendation method, comprising:

[0006] determining, based on the preprocessed health information, a probability of a user having each preset health risk, wherein the health risks include at least one;

[0007] constructing a first bipartite graph based on the health risks and at least two preset information elements;

[0008] constructing a second bipartite graph based on the information elements and information element combinations, wherein the information element combinations include any one of the information elements or combinations formed by the information elements;

[0009] determining a target information element combination based on the probability, the first bipartite graph, and the second bipartite graph, wherein the target information element combination is used to generate health recommendation information recommended to the user.

[0010] A second aspect of the embodiments of the present application provides a bipartite graph-based information recommendation device, comprising:

[0011] The first determining module is configured to determine a probability of each preset health risk existing in the user based on the preprocessed health information, wherein the health risk is at least one of the health risks.

[0012] The first constructing module is configured to construct a first bipartite graph based on the health risk and the at least two preset information elements.

[0013] The second constructing module is configured to construct a second bipartite graph based on the information elements and information element combinations, wherein the information element combinations include any one of the information elements or a combination of the information elements.

[0014] The second determining module is configured to determine a target information element combination based on the probability of the health risk, the first bipartite graph and the second bipartite graph, wherein the target information element combination is used to generate health suggestion information recommended to the user.

[0015] The third aspect of the embodiment of the present application provides a terminal device, which comprises a memory, a processor and a computer program stored in the memory and executable on the terminal device, and the processor implements each step of the information recommendation method based on the bipartite graph provided in the first aspect when executing the computer program.

[0016] The fourth aspect of the embodiment of the present application provides a computer readable storage medium, which stores a computer program, and each step of the information recommendation method based on the bipartite graph provided in the first aspect is implemented when the processor executes the computer program.

[0017] The information recommendation method based on the bipartite graph, the information recommendation device based on the bipartite graph, the terminal device and the computer readable storage medium provided in the embodiment of the present application have the following beneficial effects:

[0018] Based on the preprocessed health information of the user, the health risk possibly existing in the user is determined, that is, the probability of each preset health risk existing in the user is determined. Then, a first bipartite graph is generated with the health risk and the at least two preset information elements as vertices, and a second bipartite graph is generated based on the information elements and information element combinations as vertices, wherein the information element combinations include at least one of the information elements or a combination of the information elements. Finally, a target information element combination is determined based on the probability of the health risk and the two bipartite graphs, and the target information element combination can be used to generate health suggestion information recommended to the user. The health suggestion information recommended to the user by the method is generated based on the target information element combination determined based on the health information of the user, and is not directly selected from the pre-stored health suggestion information by using a template, so compared with the prior art, the method can provide personalized health suggestion for the user and improve the accuracy of health suggestion recommendation. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a flowchart illustrating the implementation of an information recommendation method based on a bipartite graph, as provided in an embodiment of this application.

[0021] Figure 2 This is a flowchart illustrating an information recommendation method based on a bipartite graph in an application scenario provided by an embodiment of this application.

[0022] Figure 3 This is a schematic diagram of the structure of a first bipartite graph provided in an embodiment of this application;

[0023] Figure 4 This is a structural block diagram of an information recommendation device based on a bipartite graph provided in an embodiment of this application;

[0024] Figure 5 This is a structural block diagram of a terminal device provided in an embodiment of this application. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0026] The bipartite graph-based information recommendation method described in this application can be executed by a terminal device, such as a laptop, ultra-mobile personal computer (UMPC), netbook, or personal digital assistant (PDA).

[0027] The embodiments of this application can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence (AI) is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0028] The artificial intelligence basic technology generally includes technologies such as sensors, special artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation / interaction system, mechatronics, etc. The artificial intelligence software technology mainly includes computer vision technology, robot technology, biometric identification technology, speech processing technology, natural language processing technology, and machine learning / deep learning, etc.

[0029] In addition, the information recommendation method based on the bipartite graph related in the embodiments of the present application can be applied to the smart medical scene, thereby promoting the construction of smart cities.

[0030] Please refer to Figure 1 , Figure 1 An implementation flowchart of an information recommendation method based on a bipartite graph provided by the embodiments of the present application is shown. The generation method comprises:

[0031] Step 110, determining the risk probability of each preset health risk of the user based on the preprocessed health information.

[0032] In order to improve the convenience of data processing, after obtaining the data, the data can be preprocessed. There is no standard process for data preprocessing method, and the processing operation used is different for different tasks and different data attributes. Common data processing methods include but are not limited to missing data filling, removing unique attribute information, encoding operation and standardization operation, etc., wherein the standardization operation is a normalization operation on the data.

[0033] Therefore, in order to improve the efficiency of data processing, when determining the risk probability of each health risk of the user, the preprocessed health information can be used for determination. Specifically, the health risk refers to a predisposition that may cause the user to suffer from a certain disease, and at least one common health risk can be set in advance according to the probability of the health risk appearing in daily life. For each health risk, the health information can be used to determine the risk probability of the user having the health risk. When there are multiple health risk probabilities, multiple health risk probabilities corresponding to the health risk can be calculated according to the health information. The risk probability is the probability of the user having a certain health risk.

[0034] Optionally, the health information can be obtained through relevant information uploaded by the user, or can be obtained through a relevant application program after obtaining the user's permission. The relevant application program includes but is not limited to the hospital system or application program of the user's visit. The latter acquisition method does not require the user to download or upload, which can further improve the user's experience.

[0035] Optionally, the carrier of the health information can be multi-modal data, such as medical records, questionnaire files, and diversified data with genes as external manifestations. The application of multi-modal data is closer to real scenarios and can improve the convenience and comprehensiveness of health information acquisition.

[0036] Step 120: constructing a first bipartite graph based on the health risks and the preset at least two information elements.

[0037] To reduce the probability of a certain risk, the corresponding key information, i.e., information element, can be determined according to the attributes and characteristics of the health risk corresponding to the probability of the risk. Further, to accurately determine the effective information element, a bipartite graph can be constructed based on the health risks and the information elements, where the health risks and the information elements are the vertices of the bipartite graph. To distinguish from another bipartite graph mentioned later, the bipartite graph constructed based on the health risks and the information elements is referred to as a first bipartite graph, and the bipartite graph appearing later is referred to as a second bipartite graph.

[0038] Step 130: constructing a second bipartite graph based on the information elements and the information element combinations.

[0039] To obtain complete health recommendations, only the information elements are not enough, and effective information element combinations, i.e., the target information element combinations mentioned later, also need to be further determined based on the health elements. The information element combinations can include any one information element, or can include two or more information elements. When there are multiple information elements, the multiple information elements should be able to be combined together, i.e., the multiple information elements are compatible, and there is no mutual exclusion or mutual weakening or even cancellation.

[0040] Step 140: determining the target information element combinations based on the risk probabilities, the first bipartite graph, and the second bipartite graph.

[0041] After the risk probabilities are determined and the two bipartite graphs are constructed, the effective element combinations, i.e., the target information element combinations, can be determined through the risk probabilities and the two bipartite graphs. The number of the target information element combinations can be one or more. After the target information element combinations are determined, the user-specific personalized health recommendation information can be generated based on the target information element combinations, and the health recommendation information can be recommended to the user. The user can adjust the work and body according to the health recommendation information to reduce the risk probabilities of the health risks.

[0042] Based on the pre-processed health information of the user, a health risk that the user may have is determined, i.e., a risk probability of the user having each preset health risk is determined. Then, a first bipartite graph is generated with the health risks and at least two preset information elements as vertices, and a second bipartite graph is generated with information elements and information element combinations as vertices. The information element combinations include at least one information element or a combination that can be formed by the information elements. Finally, a target information element combination is determined based on the risk probability and the two bipartite graphs. The target information element combination can be used to generate health recommendation information recommended to the user. The health recommendation information recommended to the user by the method is generated based on the target information element combination determined based on the health information of the user, rather than directly selecting pre-stored health recommendation information by using a template. Therefore, compared with the prior art, the method can provide personalized health recommendations to the user and improve the accuracy of health recommendation.

[0043] In some embodiments, in order to improve the convenience of data processing, before step 110, the method further includes:

[0044] A1, obtaining the medical record of the user.

[0045] A2, extracting health information from the medical record, the health information including structured health information and unstructured health information.

[0046] A3, performing a standardization operation on the structured health information and performing an encoding operation on the unstructured health information to obtain pre-processed health information.

[0047] In step 110, it is mentioned that the carrier of the health information includes but is not limited to medical records and health questionnaires filled out by the user, etc. The medical record covers various images obtained based on imaging, health examination reports, and doctor's diagnosis information, etc. Obviously, among various carriers of health information, the medical record is a relatively professional and more extensive type of carrier compared to the health questionnaire. Therefore, preferably, the medical record of the user can be obtained to efficiently obtain more comprehensive health information. The medical record includes electronic medical records and scanned copies of paper medical records, etc.

[0048] The extracted health information includes two types of data, one type of data is structured, and the other type of data is unstructured. Structured data, also known as row data, is data logically expressed and implemented by a two-dimensional table structure, strictly following data format and length specifications. Unstructured data is data whose structure is irregular or incomplete, has no pre-defined data model, and is not convenient to represent with a two-dimensional logical table of a database. Compared with structured data, unstructured data is a type of data that is easily encountered in data processing. Common unstructured data includes office documents, texts, pictures, various reports, and audio / video information, etc.

[0049] To facilitate subsequent processing of health information, different data preprocessing methods can be adopted for health information of two different data types. Specifically, for structured health information, standardization processing can be performed. Common standardization processing includes, but is not limited to, range standardization, standard deviation standardization, and linear proportion standardization. When selecting a standardization processing method, a corresponding standardization method can be selected for processing according to the requirements of the processed data. For example, one-hot encoding is performed on discrete features such as gender and allergy history to facilitate computer recognition and subsequent processing. Standardization processing is performed on continuous features such as age to eliminate the influence of scale differences between features and speed up model convergence. For unstructured health information, a word table such as <word, ID> can be constructed to map the words appearing in the unstructured health information to integer IDs. For example, "I have a headache for two days" can be converted to "2, 5, 6, 32, 70" according to the pre-constructed word table; "I feel a headache" can be converted to "2, 73, 75, 5, 6". The data preprocessing process is usually the first step in natural language processing tasks, because in natural language processing, a corresponding word vector is established for each word in the text as the input of the model. The establishment of the word table can correspond to the word vector table such as <ID, word vector>, and the corresponding word vector can be directly taken according to the ID corresponding to the word to input the model for calculation, thereby improving the calculation efficiency of the model.

[0050] In some embodiments, the step 110 specifically includes:

[0051] Step 111, feature extraction is performed on the preprocessed health information to obtain a feature vector.

[0052] Step 112, determining the probability of each health risk based on the feature vector.

[0053] After obtaining the preprocessed health information, more representative feature information can be extracted from the health information to generate a feature vector, and then the probability of each health risk of the user can be directly determined based on the feature vector.

[0054] Optionally, in the process of feature extraction, in order to extract representative features, different neural networks can be used to extract features from the two kinds of preprocessed health information respectively. For example, the preprocessed structured health information can be input into a fully connected neural network, i.e. a narrow sense deep neural network (DNN), to obtain the corresponding feature vector; the preprocessed unstructured health information can be input into a convolutional neural network (CNN) or a recurrent neural network (RNN) to obtain the corresponding feature vector. Although the health information of the two data types is extracted to obtain feature vectors, in this application, in order to distinguish between the two feature vectors, the feature vector based on structured health information can be denoted as the first feature vector e1=NN1(input1), and the feature vector based on unstructured health information can be denoted as the second feature vector e2=f1(input1), f∈{CNN,RNN}. The two feature vectors together are the feature vector corresponding to the health information, which is simply referred to as the feature vector e=concatenate(e1,e2). That is, the feature vector refers to the first feature vector and the second feature vector.

[0055] Optionally, after obtaining the feature vector, the neural network can be used to determine the probability of the user having each health risk based on the feature vector. Specifically, the health risk probability is determined by the fully connected layer of the neural network. The number of neurons of the fully connected layer is the same as the number of preset health risks, so each output of the fully connected layer can be used as the health risk probability of each health risk, which can be represented as p h =NN2(e), p h ∈R |H| where H is the set of all health risks, and p i h indicates the probability of the user having the i-th health risk.

[0056] In some embodiments, in order to improve the accuracy of the target information element combination determination, the above step 140 specifically includes:

[0057] Step 141, determining the weight of each information element based on the health risk probability and the first bipartite graph.

[0058] Step 142, determining the determination probability of each information element combination being determined as the target information element combination based on the weight and the second bipartite graph.

[0059] Step 143, determining the information element combination with a determination probability greater than a preset probability threshold as the target information element combination.

[0060] After the risk probability is determined, the risk probability can be input into the first bipartite graph to determine the weight of each information element. After the weight is determined, the weight can be input into the second bipartite graph to determine the determination probability of each information element combination being determined as the target information element combination. After the determination probability corresponding to each information element combination is obtained, each determination probability can be compared with a preset probability threshold to determine whether the information element combination corresponding to the determination probability is the target information element combination. For each information element combination, when the determination probability corresponding thereto is greater than the probability threshold, the information element combination corresponding to the determination probability is the target information element combination; when the determination probability corresponding thereto is less than or equal to the probability threshold, the information element combination corresponding to the determination probability is not the target information element combination. The probability threshold can be set according to experience, for example, 0.7. After all target information element combinations are determined, the health suggestion information can be generated according to the target information element combination. Let y represent the health suggestion, then y = [y1, y2, …, y i , …], wherein:

[0061]

[0062] In some embodiments, the above step 141 specifically includes:

[0063] Step 1411, calculating a first adjacency matrix and a first matrix parameter of the first bipartite graph based on the risk probability.

[0064] Step 1412, optimizing the first matrix parameter based on the first adjacency matrix.

[0065] Step 1413, determining the weight of each information element based on the optimized first matrix parameter.

[0066] For each information element, the parameter matrix and the adjacency matrix between the health risk and the information element can be calculated first, the parameters in the parameter matrix are optimized based on the adjacency matrix, and the weight of each first information element is determined based on the parameters. In view of the same concept appearing later, in order to distinguish, here, the parameter matrix is denoted as the first parameter matrix, denoted by w1∈R |H|×|E| ; the adjacency matrix is denoted as the first adjacency matrix, denoted by A1∈R |H|×|E| ; and E is the set of information elements.

[0067] Specifically, the calculation formula of the weight is as follows:

[0068]

[0069] wherein, is the multiplication of elements, where the element is not the information element, but each element in the matrix; b1 is the bias term, and f1 is the activation function. In fact, the calculation process can be regarded as performing a full connection operation on the health risks and information elements, and the parameter matrix used to perform the full connection operation is replaced by the adjacency matrix of the second bipartite graph.

[0070] In some embodiments, the step 142 specifically includes:

[0071] Step 1421, calculating a second adjacency matrix of the second bipartite graph and a second matrix parameter based on the weight.

[0072] Step 1422, optimizing the second matrix parameter based on the second adjacency matrix.

[0073] Step 1423, determining the probability that each information element combination is determined as the target information element combination based on the optimized second matrix parameter.

[0074] For each information element combination, the parameter matrix and the adjacency matrix between the information element and the information element combination can be calculated first, denoted as the second parameter matrix and the second adjacency matrix respectively, the second parameter matrix is denoted as w2∈R |H|×|E| , and the second adjacency matrix is denoted as A2∈R |H|×|E| . Through the second adjacency matrix, each parameter in the second parameter matrix is optimized to determine the determination probability that each information element combination is determined as the target information element combination, and the calculation formula of the determination probability is as follows:

[0075]

[0076] Where b2 is the bias term, and f2 is the activation function.

[0077] In some embodiments, in order to further improve the personalization and accuracy of the health recommendation information, after the step 140, the method further includes:

[0078] Step B1, determining the candidate health recommendation information based on the target information element combination.

[0079] Step B2, correcting the candidate health recommendation information based on the contraindication information of the user.

[0080] Step B3, determining the corrected candidate health recommendation information as the health recommendation information.

[0081] In order to avoid that the pushed health advice information is not feasible due to the user's contraindication information, or that other health risks are caused after the health advice information is adopted, before the health advice information is finally formed, after the target information element combination is determined, the candidate health advice information can be determined based on the target information element combination, and then the candidate health advice information is revised according to the user's contraindication information. The process can be revised by the doctor according to the contraindication information, or revised according to the preset rules to form the final health advice information. The health advice information is closer to the actual needs of the user, and the reliability and feasibility of the health advice information can be improved.

[0082] In some embodiments, the above generation method further comprises:

[0083] The above health information, risk probability, first bipartite graph, second bipartite graph, target information element combination and / or health advice information are deployed into a blockchain.

[0084] In order to ensure the security of the data and the fairness and transparency to the user, the health information, risk probability, first bipartite graph, second bipartite graph, target information element combination and / or health advice information can be deployed into a blockchain for storage. The user can then download the health information, risk probability, first bipartite graph, second bipartite graph, target information element combination and / or health advice information from the blockchain through their own device to check whether these data have been tampered with. The blockchain referred to in this embodiment is a new application mode of distributed data storage, point-to-point transmission, consensus mechanism and encryption algorithm and other computer technologies. The blockchain is essentially a decentralized database, which is a series of data blocks associated using cryptographic methods, each data block contains a batch of network transaction information, which is used to verify the validity (anti-fake) of the information and generate the next block. The blockchain can include a blockchain underlying platform, a platform product service layer, and an application service layer.

[0085] For the sake of understanding, the size of the serial number of each step in the above embodiments does not mean the order of execution, the execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0086] For the sake of understanding, the following is an actual application scenario to illustrate the bipartite graph-based information recommendation method proposed in the present application. Specifically, the information is prescription information, and correspondingly, the drug is the diagnosis result, the information element is the drug component, and the information element combination is the drug. Referring to Figure 2 The prescription bipartite graph-based information recommendation method can be implemented by the following steps:

[0087] First, the electronic medical record of the patient is obtained to obtain health information, the health information including structured health information and unstructured health information, and the two kinds of health information are preprocessed respectively to obtain preprocessed health information.

[0088] Then, the health information after preprocessing can be used to represent the physical condition of the patient, that is, the corresponding features are extracted from the health information to generate the corresponding feature vector. Specifically, in the process of feature extraction of health information, different neural networks can be used to process the two kinds of health information. The first feature vector e1 = NN1(input1) is obtained based on the structured health information, and the second feature vector e2 = f1(input1) is obtained based on the unstructured health information, f e {CNN, RNN}, and the two feature vectors are combined to obtain the feature vector corresponding to the health information, which is simply referred to as the feature vector e = concatenate(e1, e2).

[0089] Secondly, the feature vector can be used to generate a diagnosis result through a fully connected layer, that is, the probability of a certain diagnosis for the patient is obtained. The number of neurons of the fully connected layer is equal to the number of all diagnoses in the electronic medical record, so the output of the fully connected layer can be regarded as the probability p d = NN2(e), p d e R |D| , where D is the set of all diagnoses, and p i d indicates the probability of the patient suffering from the i-th diagnosis.

[0090] Thirdly, a candidate prescription is generated based on the diagnosis result. Before generating the candidate prescription, in order to improve the accuracy of the candidate prescription generation, two bipartite graphs can be constructed according to medical knowledge, which are "diagnosis-drug component" bipartite graph and "drug component-drug" bipartite graph, and the corresponding adjacency matrices are A1 e R |D|×|C| ,A2 e R |C|×|Drug| , where D, C, and Drug are the sets of all diagnoses, all drug components, and all drugs. After the two bipartite graphs are constructed, the weight of the drug component can be calculated through the "diagnosis-drug component" bipartite graph where w1 e R |D|×|C| is the parameter matrix when the bipartite graph is fully connected, is the element-wise multiplication, and b1 is the bias term, and f1 is the activation function.

[0091] Referring to Figure 3 , Figure 3 Fig. 1 shows a structural diagram of the "diagnosis-drug component" bipartite graph, and the connection relationship between the two vertices on the two sides of the bipartite graph is determined according to the adjacency matrix A1 and the parameter matrix w1. Specifically, the adjacency matrix A1 and the parameter matrix w1 of the diagnosis-drug component bipartite graph can be seen from the following table.

[0092] ​ Drug Ingredient 1 Drug Ingredient 2 Drug Ingredient 3 Drug Ingredient 4 Diagnosis 1 1 0 1 0 Diagnosis 2 0 1 0 0 Diagnosis 3 0 1 0 1

[0093] Table 1

[0094]

[0095] Table 2

[0096] Combining two tables and Figure 3 As can be seen, in the process of determining the weight of the drug component, such as the above two tables, there are 3 diagnoses and 4 drug components in total, so the left side of the bipartite graph is 3 nodes and the right side is 4 nodes, and the adjacency matrix size is 3*4, the value is non-zero or 1, 1 The two elements in the adjacency matrix have an edge, indicating the structure of the “diagnosis-drug component” bipartite graph. The parameter matrix size is also 3*4, and the value is the weight of the left and right sides. In the specific calculation, the purpose is to optimize the weight in the parameter matrix, and the adjacency matrix is unchanged. The left vertex weight is set to 0 by using the adjacency matrix*parameter matrix between the left vertex and the right vertex without an edge. The output of the right vertex calculated in this way is based on the left node with a connected edge.

[0097] Referring to the above two tables and Figure 3 It can be seen that for drug component 1 and diagnosis 2, the value of the two vertices in the adjacency matrix is 0, so there is no edge between the two vertices. When 0 is multiplied by 0.28 in the parameter matrix, the weight of diagnosis 2 is zero. For drug component 1 and diagnosis 1, the value of the two vertices in the adjacency matrix is 1, so there is no edge between the two vertices. Correspondingly, after multiplying the two matrices, 1 is multiplied by 0.15 in the parameter matrix, and the weight of 0.15 is directly involved in the calculation.

[0098] Similarly, the probability of all drugs being prescribed is calculated by the “drug component-drug” bipartite graph Where w2∈R |C|×|Drug| is the parameter matrix when the bipartite graph is fully connected, b2 is the bias term, and f2 is the activation function. If p i > probability threshold, indicating that the i-th drug will be prescribed; if p i ≤ probability threshold, indicating that the i-th drug will not be prescribed. The probability threshold is set according to the empirical value, and is generally selected as 0.5. Therefore, the candidate prescription is: y = [y1, y2, …, y i , …], where:

[0099]

[0100] Finally, based on the contraindications in the drug instructions, drugs contraindicated due to age and allergy history in candidate prescriptions y can be removed, resulting in a revised recommended prescription. This revised prescription report is then returned for doctors to review and reference. The prescription report includes: the revised recommended prescription and the probability of the patient corresponding to each diagnosis and / or the probability of the drug component corresponding to each diagnosis and / or the probability of the drug corresponding to each drug component. This makes the prescription information more interpretable: the prescription process is displayed by constructing a "diagnosis-drug component" and "drug component-drug" bipartite graph, and the final prescription report returns the patient's corresponding diagnosis, drug component, and drug probability, which is beneficial for doctors' reference and for the learning of younger doctors.

[0101] It should be understood that the data in the table above is consistent with... Figure 2 The data in the two tables do not have a corresponding relationship; the two tables are only related to each other. Figure 3 The calculation process will be further explained.

[0102] Furthermore, this application also provides an information recommendation device based on a bipartite graph.

[0103] Please see Figure 4 , Figure 4 This is a structural block diagram of an information recommendation device based on a bipartite graph, provided in an embodiment of this application. In this embodiment, the terminal device includes units used for execution... Figure 1 The steps in the corresponding embodiments. Please refer to the details. Figure 1 And the relevant descriptions in the embodiments corresponding to the figures. For ease of explanation, only the parts relevant to this embodiment are shown. See also Figure 4 The bipartite graph-based information recommendation device 40 includes:

[0104] The first determining module 41 is used to determine the probability of a user having each preset health risk based on the preprocessed health information, wherein there is at least one health risk.

[0105] The first construction module 42 is used to construct a first bipartite graph based on health risks and at least two preset information elements;

[0106] The second construction module 43 is used to construct a second bipartite graph based on information elements and combinations of information elements. The combination of information elements includes any type of information element or any combination that information elements can form.

[0107] The second determining module 44 is used to determine the combination of target information elements based on the probability of hidden danger, the first bipartite graph, and the second bipartite graph. The combination of target information elements is used to generate health advice information recommended to the user.

[0108] As an embodiment of this application, the information recommendation device 40 may further include:

[0109] The acquisition module is configured to acquire a medical record of the user before determining the risk probability of each preset health risk of the user based on the preprocessed health information.

[0110] The extraction module is configured to extract the health information from the medical record, the health information including structured health information and unstructured health information.

[0111] The preprocessing module is configured to perform a standardization operation on the structured health information and perform an encoding operation on the unstructured health information to obtain the preprocessed health information.

[0112] As an embodiment of the present application, the first determination module 41 can include:

[0113] The extraction unit is configured to perform feature extraction on the preprocessed health information to obtain a feature vector.

[0114] The first determination unit is configured to determine the risk probability of each health risk of the user based on the feature vector.

[0115] As an embodiment of the present application, the second determination module 44 can include:

[0116] The second determination unit is configured to determine the weight of each information element based on the risk probability and the first bipartite graph.

[0117] The third determination unit is configured to determine the determination probability of each information element combination being determined as the target information element combination based on the weight and the second bipartite graph.

[0118] The fourth determination unit is configured to determine the information element combination with a determination probability greater than a preset probability threshold as the target information element combination.

[0119] As an embodiment of the present application, the second determination unit can include:

[0120] The calculation sub-unit is configured to calculate a first adjacency matrix and a first matrix parameter of the first bipartite graph based on the risk probability.

[0121] The optimization sub-unit is configured to optimize the first matrix parameter based on the first adjacency matrix.

[0122] The determination sub-unit is configured to determine the weight of each information element based on the optimized first matrix parameter.

[0123] As an embodiment of the present application, the bipartite graph-based information recommendation device 40 can further include:

[0124] The third determination module is configured to determine candidate health recommendation information based on the target information element combination after determining the target information element combination based on the risk probability, the first bipartite graph, and the second bipartite graph.

[0125] a correction module, configured to correct the candidate health advice information based on the contraindication information of the user;

[0126] a fourth determination module, configured to determine the corrected candidate health advice information as the health advice information.

[0127] As an embodiment of the present application, the information recommendation apparatus 40 based on bipartite graph can further include:

[0128] a data deployment module, configured to deploy the health information, the risk probability, the first bipartite graph, the second bipartite graph, the target information element combination and / or the health advice information into a blockchain.

[0129] It should be understood that, Figure 4 the structure block diagram of the information recommendation apparatus based on bipartite graph is shown, each unit is used to execute Figure 1 each step in the corresponding embodiment, and for Figure 1 each step in the corresponding embodiment has been explained in detail in the above embodiment, please refer to Figure 1 and Figure 1 the related description in the corresponding embodiment, which will not be repeated here.

[0130] Figure 5 is a structure block diagram of a terminal device provided by another embodiment of the present application. As Figure 5 shown, the terminal device 50 of this embodiment includes a processor 51, a memory 52, and a computer program 53 stored in the memory 52 and executable on the processor 51, such as a program of the information recommendation method based on bipartite graph. The processor 51 executes the above computer program 53 to realize the steps in each embodiment of the above various information recommendation methods based on bipartite graph, such as Figure 1 shown as 110 to 140. Alternatively, the processor 51 executes the computer program 53 to realize the functions of each module in the corresponding embodiment, such as Figure 4 the functions of the modules 41 to 44 shown in Figure 4 , please refer to Figure 4 the related description in the corresponding embodiment, which will not be repeated here.

[0131] For example, the computer program 53 can be divided into one or more units, which are stored in the memory 52 and executed by the processor 51 to complete the present application. The one or more units can be a series of computer program instruction segments capable of completing a specific function, which are used to describe the execution process of the computer program 53 in the terminal 50. For example, the computer program 53 can be divided into the first determining module 41, the first constructing module 42, the second constructing module 43, and the second determining module 44, and the specific functions of the modules are described above.

[0132] The terminal device can include, but is not limited to, the processor 51 and the memory 52. Those skilled in the art can understand that Figure 5 The terminal device 50 is only an example and does not constitute a limitation on the terminal device 50, and can include more or fewer components than the illustration, or combine certain components, or different components, for example, the terminal device can also include an input / output device, a network access device, a bus, etc.

[0133] The processor 51 can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor.

[0134] The memory 52 can be an internal storage unit of the terminal device 50, such as a hard disk or a memory of the terminal device 50. The memory 52 can also be an external storage device of the terminal device 50, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. Further, the memory 52 can include both the internal storage unit and the external storage device of the terminal device 50. The memory 52 is used to store the computer program and other programs and data required by the terminal device. The memory 52 can also be used to temporarily store data that has been output or will be output.

[0135] The above-described embodiments are only used to illustrate the technical solutions of the present application, but not limit them; although the present application is described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement to part of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be included in the protection scope of the present application.

Claims

1. A bipartite graph-based information recommendation method, characterized by, The information recommendation method comprises: determining, based on the preprocessed health information, a risk probability of the user having each preset health risk, the health risk being at least one; constructing a first bipartite graph based on the health risk and at least two preset information elements, the information element being a drug component; constructing a second bipartite graph based on the information element and an information element combination, the information element combination comprising any one information element or a combination formed by the information element, and the information element combination being a drug; determining, based on the risk probability, the first bipartite graph and the second bipartite graph, a target information element combination, the target information element combination being used to generate health suggestion information recommended to the user; the determining, based on the risk probability, the first bipartite graph and the second bipartite graph, a target information element combination, comprises: determining a weight of each information element based on the risk probability and the first bipartite graph; determining a determination probability of each information element combination being determined as the target information element combination based on the weight and the second bipartite graph; determining, as the target information element combination, an information element combination having a determination probability greater than a preset probability threshold; the determining a weight of each information element based on the risk probability and the first bipartite graph, comprises: calculating a first adjacency matrix and a first matrix parameter of the first bipartite graph based on the risk probability; optimizing the first matrix parameter based on the first adjacency matrix; determining a weight of each information element based on the optimized first matrix parameter; a calculation formula of the weight is as follows: P c = f 1（（ w 1 A 1） p d + b1） The w 1 ∈ R |D|×|C| denotes a first parameter matrix; the A 1 ∈ R |D|×|C| denotes a first adjacency matrix; the D denotes a set of all diagnoses, the C denotes a set of all drug components; the denotes a multiplication of matrix elements; the b1 denotes a first bias term, the f 1 denotes a first activation function; the determining a determination probability of each information element combination being determined as the target information element combination based on the weight and the second bipartite graph, comprises: calculating a second adjacency matrix and a second matrix parameter of the second bipartite graph based on the weight; optimizing the second matrix parameter based on the second adjacency matrix; determining a probability of each information element combination being determined as the target information element combination based on the optimized second matrix parameter; a calculation formula of the probability is as follows: P = f 2(( w 2 A 2) P c + b2) The w 2∈R |C|×|Drug| denotes a second parameter matrix; the A 2∈R |C|×|Drug| denotes a second adjacency matrix; the Drug denotes a set of all drugs; the b2denotes a second bias term; the f 2denotes a second activation function.

2. The information recommendation method according to claim 1, characterized by, Before the determining, based on the preprocessed health information, a risk probability of the user having each preset health risk, the method comprises: obtaining a medical record of the user; extracting health information from the medical record, the health information comprising structured health information and unstructured health information; performing a standardization operation on the structured health information and an encoding operation on the unstructured health information to obtain the preprocessed health information.

3. The information recommendation method according to claim 1, characterized by, The determining, based on the preprocessed health information, a risk probability of the user having each preset health risk, comprises: performing feature extraction on the preprocessed health information to obtain a feature vector; determining, based on the feature vector, the risk probability of the user having each health risk.

4. The information recommendation method according to any one of claims 1 to 3, characterized by, After the determining, based on the risk probability, the first bipartite graph and the second bipartite graph, a target information element combination, the method further comprises: determining candidate health suggestion information based on the target information element combination; correcting the candidate health suggestion information based on the user's contraindication information; The revised candidate health recommendation information is determined as the health recommendation information.

5. The information recommendation method according to any one of claims 1 to 3, characterized by, The generation method further includes: The health information, the risk probability, the first bipartite graph, the second bipartite graph, the target information element combination and / or the health recommendation information are deployed into a blockchain.

6. An information recommendation device characterized by comprising: The information recommendation device includes: A first determination module is configured to determine, based on preprocessed health information, a risk probability of a user having each preset health risk, wherein the health risks include at least one; A first construction module is configured to construct a first bipartite graph based on the health risks and at least two preset information elements, wherein the information elements are drug components; A second construction module is configured to construct a second bipartite graph based on the information elements and information element combinations, wherein the information element combinations include any one of the information elements or combinations formed by the information elements; and the information element combinations are drugs; A second determination module is configured to determine, based on the risk probability, the first bipartite graph and the second bipartite graph, a target information element combination, wherein the target information element combination is used to generate health recommendation information recommended to the user; The second determination module includes: A second determination unit is configured to determine, based on the risk probability and the first bipartite graph, a weight of each information element; A third determination unit is configured to determine, based on the weight and the second bipartite graph, a determination probability of each information element combination being determined as the target information element combination; A fourth determination unit is configured to determine, as the target information element combination, an information element combination whose determination probability is greater than a preset probability threshold; The second determination unit is specifically configured to: Calculate a first adjacency matrix and a first matrix parameter of the first bipartite graph based on the risk probability; Optimize the first matrix parameter based on the first adjacency matrix; Determine the weight of each information element based on the optimized first matrix parameter; The calculation formula of the weight is as follows: P c = f 1（（ w 1 A 1） p d + b1） The w 1 ∈ R |D|×|C| denotes a first parameter matrix; the A 1 ∈ R |D|×|C| denotes a first adjacency matrix; the D denotes a set of all diagnoses, the C denotes a set of all drug components; the denotes a multiplication of matrix elements; the b1 denotes a first bias term, the f 1 denotes a first activation function; The third determination unit is specifically configured to: Calculate a second adjacency matrix and a second matrix parameter of the second bipartite graph based on the weight; Optimize the second matrix parameter based on the second adjacency matrix; Determine, based on the optimized second matrix parameter, a probability of each information element combination being determined as the target information element combination; The calculation formula of the probability is as follows: P = f 2(( w 2 A 2) P c + b2) The w 2∈R |C|×|Drug| denotes a second parameter matrix; the A 2∈R |C|×|Drug| denotes a second adjacency matrix; the Drug denotes a set of all drugs; the b2denotes a second bias term; the f 2denotes a second activation function.

7. A terminal device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, The processor executes the computer program to implement the steps of the method according to any one of claims 1 to 5.

8. A computer-readable storage medium storing a computer program, the computer-readable storage medium comprising: The computer program is executed by the processor to implement the steps of the method according to any one of claims 1 to 5.

Images