Information matching method and system for health management decision-making

Abstract

The present application discloses an information matching method and system for health management decision-making, including: obtaining decision subject information; structuring the decision subject information to obtain structured subject information; encoding the structured subject information to obtain subject information encoding results; calculating the correlation between the subject information encoding results and the health and medical information encoding results of each health and medical information; extracting the health and medical information corresponding to the health and medical information encoding results whose correlation exceeds a preset first correlation threshold, and obtaining the target health and medical information. The present application selects the target health and medical information whose correlation meets the preset conditions, and provides it to medical staff for reference. By pre-encoding and matching health and medical information of different directions and types, the matching dimension is increased, which effectively improves the limitations of the single content type and low decision support efficiency of health management decision-making content support, and ensures that the provided reference information is reliable and accurate.

Description

Technical Field

[0001] The present invention relates to the field of smart medical care, and in particular to an information matching method and system for health management decision-making. Background Art

[0002] At present, the clinical diagnosis and treatment process mainly relies on the medical staff's own knowledge reserves and personal experience to provide treatment plans for patients. However, there are great differences in the systematic education and training received by medical staff in my country, resulting in large differences in clinical diagnosis and treatment standards in different regions and uneven medical service levels, making it difficult to meet the medical and health needs of the people.

[0003] However, existing technical solutions can usually only provide a few single-aspect decision support such as triage suggestions, diagnostic references, and medication suggestions based on limited data such as diagnostic information, symptoms and signs extracted from the decision-making subject information, and the breadth and depth of the decision-making information are extremely limited. For example, medication recommendations usually only include limited information such as drug names, drug usage and dosage, and lack information such as reasonable medication warnings, interpretation of medication evidence, and medication burden. It is impossible to achieve homogeneous amplification of different categories of information through a single technical path, and it is difficult to quickly and efficiently support the complex decision-making needs in the whole cycle management of the disease from multiple dimensions with fine-grained data, and it is also impossible to cover the diverse decision-making needs in other health management decision-making directions other than clinical diagnosis and treatment.

[0004] On the other hand, existing technologies also lack intelligent decision-making assistance supported by multi-dimensional data in other health management-related scenarios outside of clinical diagnosis and treatment, such as community medical and nursing decision-making assistance for chronic disease populations, nutritional and health management assistance for various groups of people, hospital performance control, and medical quality management assistance.

[0005] To this end, a more comprehensive and effective information matching method for health management decision-making is needed to provide medical staff with effective reference information. Summary of the invention

[0006] In view of this, the purpose of the present invention is to provide an information matching method and system for health management decision-making, which can provide effective reference information for medical staff. The specific scheme is as follows:

[0007] An information matching method for health management decision-making, comprising:

[0008] Obtain information about decision-making entities;

[0009] Structuring the decision subject information to obtain structured subject information;

[0010] Performing encoding processing on the structured subject information to obtain a subject information encoding result for reflecting the characteristics of the structured subject information;

[0011] Calculating the correlation between the subject information coding result and the health care information coding result of each health care information;

[0012] Extracting health and medical information corresponding to the health and medical information encoding result whose correlation with the subject information encoding result exceeds a preset first correlation threshold, to obtain target health and medical information;

[0013] The health and medical information encoding result of each health and medical information is obtained by encoding each health and medical information in advance.

[0014] Optionally, the process of encoding the structured subject information to obtain a subject information encoding result for reflecting the characteristics of the structured subject information includes:

[0015] Using a preset information coding model to encode the structured subject information, to obtain a subject information encoding result that reflects the characteristics of the structured subject information;

[0016] Wherein, the information encoding model is pre-trained using historical decision-making subject information.

[0017] Optionally, the process of encoding the structured subject information using a preset information encoding model to obtain a subject information encoding result that reflects the characteristics of the structured subject information includes:

[0018] Calculating the representation vector of each feature in the structured subject information using the information encoding model;

[0019] The information encoding model is used to integrate the representation vectors of various features to obtain the subject information encoding result for reflecting the structured subject information features.

[0020] Optionally, the process of calculating the correlation between the subject information encoding result and the health care information encoding result of each health care information includes:

[0021] Calculating the correlation between the subject information encoding result and the health care information encoding result of each health care information by using a preset correlation calculation model;

[0022] The correlation calculation model is pre-trained using historical subject information coding and historical health and medical information coding results.

[0023] Optionally, also include:

[0024] The health and medical information is pre-classified based on the preset information, clustered and stored in the health and medical information database.

[0025] Optionally, the process of extracting health and medical information corresponding to the health and medical information encoding result whose correlation with the subject information encoding result exceeds a preset first correlation threshold to obtain target health and medical information includes:

[0026] Extracting the associated health and medical information and the mapped health and medical information whose correlation with the information encoding result of the associated health and medical information exceeds a preset second correlation threshold from the health and medical information database, and obtaining target health and medical information including the associated health and medical information and the mapped health and medical information;

[0027] The associated health and medical information is health and medical information corresponding to a health and medical information encoding result whose degree of association with the subject information encoding result exceeds a preset first association threshold.

[0028] Optionally, also include:

[0029] Add new information categories and add health and medical information under any information category.

[0030] Optionally, before encoding the structured subject information, the method further includes:

[0031] Performing query intent analysis on the structured subject information using a preset intent analysis model to obtain a query intent corresponding to the structured subject information;

[0032] Using the preset feature-intent mapping relationship, determine whether the key features corresponding to the query intent are missing;

[0033] If missing, obtain the missing key features in the decision-making subject information from the corresponding database or receive the missing key features in the decision-making subject information supplemented by the user to obtain the complete decision-making subject information;

[0034] Reusing the complete decision subject information as decision subject information for structured processing to obtain the structured subject information;

[0035] The intent analysis model is pre-trained using historical structured subject information, and the feature-intent mapping relationship records the correspondence between the query intent and the corresponding key features.

[0036] Optional, including:

[0037] Randomly remove features from historical decision-making subject information;

[0038] Counting the error probability between each query intent and the missing feature that causes the correlation calculation model to calculate errors;

[0039] According to the error probability, the corresponding relationship between each query intent and key feature is determined, and the mapping relationship between the feature and intent is generated.

[0040] The present invention also discloses an information matching system for health management decision-making, comprising:

[0041] A subject information acquisition module is used to obtain decision-making subject information;

[0042] A structured processing module, used for performing structured processing on the decision subject information to obtain structured subject information;

[0043] An encoding processing module, used for encoding the structured subject information to obtain a subject information encoding result for reflecting the characteristics of the structured subject information;

[0044] A correlation calculation module, used to calculate the correlation between the subject information coding result and the health care information coding result of each health care information;

[0045] A medical information extraction module, used to extract health and medical information corresponding to the health and medical information encoding result whose correlation with the subject information encoding result exceeds a preset first correlation threshold, to obtain target health and medical information;

[0046] The health and medical information encoding result of each health and medical information is obtained by encoding each health and medical information in advance.

[0047] In the present invention, the information matching method for health management decision-making includes: obtaining decision-making subject information; structuring the decision-making subject information to obtain structured subject information; encoding the structured subject information to obtain a subject information encoding result for reflecting the characteristics of the structured subject information; calculating the correlation between the subject information encoding result and the health care information encoding results of each health care information; extracting the health care information corresponding to the health care information encoding result whose correlation with the subject information encoding result exceeds a preset first correlation threshold value to obtain the target health care information; wherein the health care information encoding result of each health care information is obtained by pre-encoding each health care information.

[0048] The present invention structures the decision-making subject information into a unified format, then performs encoding processing to obtain the subject information encoding result, calculates the correlation with a large amount of health and medical information encoding results that have been pre-encoded, screens out the health and medical information corresponding to the health and medical information encoding result whose correlation exceeds a preset first association threshold as the target health and medical information, and provides it to medical staff for reference. By pre-encoding health and medical information of different directions and types and using it for matching, the matching dimension is increased, so the limitations of health management decision support content, such as single content type, small amount of information, narrow application scenarios, and low decision support efficiency, can be effectively improved, thereby ensuring that the provided reference information is reliable and accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] 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 embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0050] Figure 1 A schematic flow chart of an information matching method for health management decision-making disclosed in an embodiment of the present invention;

[0051] Figure 2 A schematic diagram of a decision subject information label disclosed in an embodiment of the present invention;

[0052] Figure 3 A schematic flow chart of another information matching method for health management decision-making disclosed in an embodiment of the present invention;

[0053] Figure 4 A schematic diagram of the overall framework of an information matching method for health management decision-making disclosed in an embodiment of the present invention;

[0054] Figure 5 A schematic diagram of a health and medical data label and medical insurance data corresponding to the health and medical information B006361 disclosed in an embodiment of the present invention;

[0055] Figure 6 This is another data diagram of mapping the health and medical data label corresponding to the health and medical information X006259 and the data of the National Health Commission's "Guidelines for the Clinical Application of New Anti-tumor Drugs (2020 Edition)" disclosed in an embodiment of the present invention;

[0056] Figure 7 A schematic flow chart of another information matching method for health management decision-making disclosed in an embodiment of the present invention;

[0057] Figure 8A schematic diagram of the structure of an information matching system for health management decision-making disclosed in an embodiment of the present invention. DETAILED DESCRIPTION

[0058] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only 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.

[0059] The embodiment of the present invention discloses an information matching method for health management decision making, see Figure 1 As shown, the method includes:

[0060] S11: Obtain decision-making entity information.

[0061] Specifically, when health management is needed for a patient, the decision-making subject information of the patient is needed. The decision-making subject information includes the information needed to describe the patient, wherein the decision-making subject information may include at least one of user basic information, outpatient medical record information, inpatient medical record information, physical examination information, medical order information, examination information, test information, surgery information, and traditional Chinese medicine treatment information.

[0062] S12: Structuring the decision subject information to obtain structured subject information.

[0063] Specifically, in order to facilitate storage and subsequent invocation of decision-making subject information, the decision-making subject information is structured to obtain structured subject information in a unified format. By structuring the decision-making subject information, a large amount of structured subject information can be effectively stored for a long time. At the same time, due to the unified format, it is also convenient to analyze, process and call the structured subject information in the future.

[0064] Specifically, in the process of structuring the decision-making subject information, the decision-making subject information can be feature extracted and simplified. For example, the decision-making subject information label is set, and the corresponding feature label is extracted from the decision-making subject information according to the preset label, so as to obtain the simplified structured subject information.

[0065] For example, the decision-making subject information includes: the patient, female, 63 years old, was admitted to the hospital due to "cough and left chest pain for more than one month". The patient had cough without obvious cause more than a month ago, mainly irritating dry cough, a small amount of sputum, accompanied by left chest pain, dull pain, obvious when inhaling, not radiating to other parts, no obvious chest tightness and shortness of breath, no chills, fever, night sweats, and did not attract attention at first, and the symptoms did not improve. She was treated in our hospital 2 weeks ago, and the chest CT showed: multiple lesions in both lungs, and the lesions in the left upper lung did not exclude tumors. PET / CT showed: metastasis to both lungs, metastasis to the left cervical lymph nodes, bilateral clavicular lymph nodes, mediastinal lymph nodes, bilateral hilar lymph nodes, brain metastasis, and right adrenal gland metastasis. Electronic bronchoscopy and lung biopsy of the left main bronchus and left upper lobe were performed: a neoplasm was found in the left upper lobe bronchus, which was brittle and easily bleeding. A biopsy was taken and the pathological results showed: cancer cells were found, suggesting NSCLC adenocarcinoma; immunohistochemistry results showed: ALK-ventana (-), BRAF (-), EGFR (19) (0), EGFR (21) (0), Ki-67 (+30%), PD-L1 (TPS 10%), ROS-1 (-). Diagnosis: left upper lobe bronchial adenocarcinoma, non-small cell lung cancer, cT4N3M1c (IVB stage), ALK (-), BRAF (-), EGFR (-), Ki-67 (+30%), ROS-1 (-), PD-L1 (TPS 10%). After using the label extraction model to extract the decision subject information labels in the decision subject information, we can obtain labels including gender, age, symptom and sign labels, disease name labels, case classification labels, disease stage labels and molecular classification labels. Figure 2 The decision-making entity information label shown.

[0066] Among them, the decision-making subject information label may include at least one of the following: subject basic information label, symptom and sign label, disease classification label, medication name label, drug usage and dosage label, surgical classification label, radiotherapy plan label, other treatment label, cost label, and scientific research label; among them, the subject basic information label may further include at least one of a gender label, an age label, a native place label, an occupation label, an ethnicity label, a marriage label, a region label, a medical institution type label, a department label, a community label, a consumables label, and an equipment label; the disease classification label may include at least one of an ICD code label, a disease name label, a pathological classification label, a molecular classification label, a prognosis classification label, a disease stage label, a treatment stage label, and an optional treatment plan classification label.

[0067] S13: Encoding the structured subject information to obtain a subject information encoding result that reflects the characteristics of the structured subject information.

[0068] Specifically, the features recorded in text form are difficult to retrieve and the retrieval efficiency is low. In order to match the health and medical information associated with the structured subject information, the structured subject information is encoded. The encoding process is used to further extract and highlight the features of the structured subject information, and a subject information encoding result that can reflect the features of the structured subject information is obtained. Feature matching through the encoding result will be faster and more accurate.

[0069] S14: Calculate the correlation between the subject information coding result and the health care information coding result of each health care information.

[0070] Specifically, after obtaining the subject information coding result, it can be matched with the health care information coding result of each health care information. During the matching process, the correlation between the subject information coding result and the health care information coding result of each health care information is calculated based on various coefficients such as the similarity and weight of the coding results.

[0071] The health and medical information encoding result of each health and medical information is obtained by encoding each health and medical information in advance.

[0072] Specifically, a large amount of health and medical information is collected in advance. In order to effectively match the health and medical information with the subject information coding results of the decision-making subject information, it is also necessary to perform structured processing on the health and medical information to obtain structured medical information, and then perform coding processing to obtain each health and medical information coding result corresponding to each structured medical information, so that the correlation degree can be calculated with the subject information coding results of the structured subject information later. Among them, the health and medical information can include various medical-related data, such as clinical research data, pharmaceutical data, and medical data, etc., to cover all the content that may be involved in the decision-making subject information as comprehensively as possible, to ensure that any decision-making subject information can find the corresponding health and medical information, and to ensure that corresponding information support can be provided to users.

[0073] Among them, when the health and medical information is structured, the health and medical information can also be labeled, and health and medical information labels can be set. According to the preset labels, corresponding feature labels are extracted from the health and medical information to obtain simplified structured medical information.

[0074] S15: Extract health and medical information corresponding to the health and medical information encoding result whose correlation with the subject information encoding result exceeds a preset first correlation threshold, and obtain target health and medical information.

[0075] Specifically, after comparing the correlation between the subject information coding result and each health and medical information coding result, the health and medical information coding result whose correlation exceeds the preset first correlation threshold is selected and identified as the health and medical information coding result suitable for the subject, and the remaining health and medical information coding results with low correlation can be considered to be information irrelevant to the decision-making subject information or information that has been covered by health and medical information with high correlation. It can be understood that after determining the health and medical information coding result that exceeds the preset first correlation threshold, the health and medical information corresponding to each health and medical information coding result can be used as the target health and medical information and provided to users, such as medical staff, as reference information to provide information reference for patients.

[0076] It can be seen that the embodiment of the present invention structures the decision-making subject information into a unified format, then performs encoding processing to obtain the subject information encoding result, calculates the correlation with a large amount of health and medical information encoding results that have been pre-encoded, and screens out the health and medical information corresponding to the health and medical information encoding result whose correlation exceeds the preset first association threshold as the target health and medical information, which is provided to medical staff for reference. By pre-encoding health and medical information of different directions and types and using it for matching, the matching dimension is increased, so it can effectively improve the limitations of health management decision-making content support, such as single content type, small amount of information, narrow application scenarios, and low decision support efficiency, and ensure the reliability and accuracy of the reference information provided.

[0077] The embodiment of the present invention discloses a specific information matching method for health management decision-making. Compared with the previous embodiment, this embodiment further explains and optimizes the technical solution. Figure 3 and Figure 4 As shown, specifically:

[0078] S21: Classify the health care information based on preset information in advance, and cluster and store them in a health care information database.

[0079] Specifically, in order to effectively store health and medical information, a health and medical information database can be established. In order to further improve retrieval efficiency, health and medical information can be clustered and stored based on preset information classifications. In this way, when the query direction of the decision-making subject information can be clearly defined, the health and medical information of at least one corresponding information classification can be retrieved for subsequent correlation calculations.

[0080] S22: Obtain decision-making subject information;

[0081] S23: Structuring the decision subject information to obtain structured subject information;

[0082] S24: Encoding the structured subject information using a preset information encoding model to obtain a subject information encoding result that reflects the characteristics of the structured subject information.

[0083] Specifically, by pre-training with historical decision subject information to obtain an information encoding model, the model can be used to encode structured subject information more quickly and accurately.

[0084] Further, S24 uses a preset information coding model to encode the structured subject information to obtain a subject information encoding result for reflecting the characteristics of the structured subject information, which may specifically include S241 and S242; wherein,

[0085] S241: Calculate the representation vector of each feature in the structured subject information using the information encoding model;

[0086] S242: Using the information coding model to integrate the representation vectors of each feature, obtain a subject information coding result for reflecting the structured subject information features.

[0087] Specifically, the information encoding model can first use word embedding technology (including but not limited to word2vec and glove) or pre-trained models (including but not limited to BERT (Bidirectional Encoder Representations from Transformers), XLNet, GPT2, CNN (Convolutional Neural Networks), RNN (Recurrent Neural Network), LSTM (Long Short-Term Memory) and GRU (Gate Recurrent Unit)) to obtain the representation vector of each feature of the decision-making subject information, and then integrate the feature representation vectors involved in the decision-making subject information (for example, splicing or averaging, etc.), and the integration result is the information encoding result. It can be understood that the encoding result of health and medical information is also obtained in the same way.

[0088] S25: Calculate the correlation between the subject information coding result and the health care information coding result of each health care information by using a preset correlation calculation model;

[0089] Specifically, similarly, by using natural language technology to establish a correlation calculation model, the calculation speed and accuracy can be effectively improved, wherein the correlation calculation model is pre-trained using historical subject information encoding and historical health and medical information encoding results.

[0090] S26: Extracting the associated health and medical information and the mapped health and medical information whose correlation with the information encoding result of the associated health and medical information exceeds a preset second correlation threshold from the health and medical information database to obtain target health and medical information including the associated health and medical information and the mapped health and medical information.

[0091] Specifically, since health and medical information is clustered and stored in the health and medical information database, and an efficient mapping relationship is pre-constructed between associated health and medical information and mapped health and medical information, while extracting associated health and medical information from the health and medical information database, mapped health and medical information whose correlation with the information encoding result of the associated health and medical information exceeds a preset second correlation threshold can also be extracted as auxiliary information based on information classification and mapping relationship, and provided together as target health and medical information to medical staff for reference, thereby increasing the richness of the reference content.

[0092] For example, see Figure 5 and Figure 6 As shown in the figure, after labeling and comparing the information coding results, the correlation between "Other Information - Medical Insurance Data B006361" and "Other Information - National Health Commission "Guidelines for the Clinical Application of New Anti-tumor Drugs (2020 Edition)" X006259" is 97%, which exceeds the preset threshold of 95%. Therefore, information X006259 can be used as the mapping health information of information B006361 under the category of "Other Information - National Health Commission "Guidelines for the Clinical Application of New Anti-tumor Drugs (2020 Edition)". Structured subject information is as follows Figure 2 As shown, after the structured subject information is labeled and encoded, it is compared with the information encoding result of information B006361, and the correlation is 93%, which exceeds the preset threshold of 90%. Therefore, "other information-medical insurance data B006361" is used as the associated data in the target data of this decision, and "other information-National Health Commission's "Guidelines for the Clinical Application of New Anti-tumor Drugs (2020 Edition)" X006259" is used as the mapping data in the target data of this decision, and they are output together. It can be seen that health and medical information is clustered and stored in the health and medical information database, and an efficient mapping relationship is constructed between the associated health and medical information and the mapped health and medical information, which can quickly obtain various types of associated health and medical information and their mapped health information in the required application scenarios, and provide multi-dimensional information reference. Among them, the associated health and medical information is the health and medical information corresponding to the health and medical information encoding result whose correlation with the subject information encoding result exceeds the preset first correlation threshold.

[0093] The health and medical information encoding result of each health and medical information is obtained by encoding each health and medical information in advance.

[0094] S27: Add new information categories and add health and medical information under any information category.

[0095] Specifically, the health and medical information database can continuously expand information categories and the data stored in a certain information category to meet the new direction of medical and health decision-making query needs.

[0096] The embodiment of the present invention also discloses a specific information matching method for health management decision making, see Figure 7 As shown, specifically:

[0097] S31: Obtain decision-making subject information;

[0098] S32: Structuring the decision subject information to obtain structured subject information;

[0099] S33: Analyze the query intent of the structured subject information using a preset intent analysis model to obtain the query intent corresponding to the structured subject information.

[0100] Specifically, in order to improve the analysis speed and avoid querying all the health and medical information in the health and medical information database, and also to avoid the situation where the matching effect is poor due to the lack of key information in the decision-making subject information, after obtaining the structured subject information, the preset intention analysis model is used to perform query intention analysis on the structured subject information, so as to obtain the query direction of the decision-making subject information, that is, the query intention corresponding to the structured subject information.

[0101] Among them, the intent analysis model is pre-trained using historical structured subject information, and the feature-intent mapping relationship records the correspondence between the query intent and the corresponding key features.

[0102] It is understandable that the analysis of query intent may be achieved using natural language processing technology, or may be directly input and specified by the user when inputting decision subject information.

[0103] S34: Using the preset feature-intent mapping relationship, determine whether the key features corresponding to the query intent are missing.

[0104] Specifically, a mapping relationship between query intent and key features, i.e., a preset feature and intent mapping relationship, is established in advance, so that the key features required to achieve the query effect of the query intent can be determined according to the query intent. In this way, it can be determined whether the corresponding key features are missing in the structured subject information.

[0105] Specifically, the structured subject information may include feature labels. According to the labels of the key features corresponding to each intent recorded in the feature-intent mapping relationship, it is determined whether the structured subject information includes the labels of the key features corresponding to the current query intent. If not, it can be determined that the key features are missing.

[0106] S35: If missing, obtain the missing key features in the decision subject information from the corresponding database or receive the missing key features in the decision subject information supplemented by the user to obtain complete decision subject information;

[0107] S36: The complete decision subject information is used again as the decision subject information so as to perform structured processing and obtain structured subject information.

[0108] Specifically, if the key features corresponding to the query intent are missing, the missing key features in the decision-making subject information can be automatically obtained from the corresponding database. For example, the missing key features can be obtained from a database including the personal information of the subject corresponding to the decision-making subject information. After supplementing the missing key features in the decision-making subject information, the supplemented complete decision-making subject information is re-obtained as the new decision-making subject information, and S32 is re-executed for structured processing to obtain structured subject information, so as to execute the subsequent steps S33 to S39 to re-match the information for health management decisions, effectively avoiding the situation where the reference information finally calculated is wrong and affects the user's judgment. Of course, the user can also actively supplement it. After the key features corresponding to the query intent are missing, only a reminder message is generated to remind the decision-making subject information that the key features are missing, allowing the user to supplement them by themselves. After receiving the missing key features in the decision-making subject information supplemented by the user, the complete decision-making subject information can also be obtained.

[0109] Furthermore, in the case of automatically supplementing the key features missing in the decision subject information, prompt information on whether to allow automatic supplementation can be generated. After the user inputs an instruction to allow automatic supplementation, the key features missing in the decision subject information are automatically supplemented.

[0110] Specifically, of course, if there is no missing key feature corresponding to the query intent, then continue to execute S37 to perform the next matching process.

[0111] S37: Encoding the structured subject information to obtain a subject information encoding result that reflects the characteristics of the structured subject information.

[0112] S38: Calculating the correlation between the subject information coding result and the health care information coding result of each health care information;

[0113] S39: Extract the health and medical information corresponding to the health and medical information encoding result whose correlation with the subject information encoding result exceeds a preset first correlation threshold, and obtain the target health and medical information.

[0114] The health and medical information encoding result of each health and medical information is obtained by encoding each health and medical information in advance.

[0115] Furthermore, the process of generating the feature-intention mapping relationship in the above S34 may specifically include S341 to S343; wherein,

[0116] S341: Randomly remove features from historical decision-making subject information.

[0117] Specifically, in order to obtain the impact of each feature on different query intentions, some features are randomly removed from the historical decision subject information. In this way, when the correlation is calculated for the historical decision subject information with missing features, the missing features will cause errors in the correlation calculation. Therefore, the current missing features are the key features of the correct health and medical information that cannot be associated due to the missing features.

[0118] For example, n features in the historical decision subject information are randomly removed, where n∈[1, the maximum number of features in the current decision subject information).

[0119] S342: Count the error probability of the correlation calculation model causing calculation errors between each query intent and the missing features.

[0120] Specifically, each historical decision subject information can support multiple query intentions, such as rehabilitation query intention, treatment query intention or drug information query intention, etc. Therefore, by continuously removing new random features using a historical decision subject information, the error probability of the correlation calculation model causing errors in calculations between each query intention that the historical decision subject information can cover and the missing features can be counted. Of course, the query intentions that a historical decision subject information can cover are limited, and the samples are limited. Therefore, new historical decision subject information will be continuously used to remove random features, and the error probability of the correlation calculation model causing errors in calculations between each query intention and the missing features can be counted. Finally, the error probability of the correlation calculation model causing errors in calculations between each query intention and the missing features with a wide enough coverage can be counted.

[0121] Specifically, due to the missing features, the association calculation model can no longer calculate the health care information corresponding to the missing features, resulting in calculation errors. Therefore, by counting the error probability of the association calculation model causing calculation errors between each query intent and the missing features, the key features of the query intent can be inferred.

[0122] For example, due to the missing feature A, the error probability of the correlation calculation model calculating the historical decision-making subject information and the health care information corresponding to the query intention B increased from 10% to 80%.

[0123] S343: Determine the correspondence between each query intent and key features based on the error probability, and generate a mapping relationship between features and intent.

[0124] Specifically, based on the error probability of the correlation calculation model causing calculation errors between each query intent and the missing features, the closeness of the relationship between each feature and each query intent can be inferred. When a feature is missing, the error probability of one or more query intent correlation calculation models calculating errors increases significantly, exceeding the preset error probability. It is highly likely that the feature is a key feature of this query intent. For example, due to the missing feature A, the error probability of the correlation calculation model calculating the historical decision-making subject information and the health and medical information corresponding to query intent B increases from 10% to 80%, exceeding the preset error probability of 50%. This means that feature A seriously affects the calculation accuracy of the correlation calculation model for the health and medical information in the direction of query intent B. Therefore, feature A is a key feature for query intent B. By repeatedly counting the correspondence between each query intent and the key feature, a feature and intent mapping relationship covering each query intent and the key feature can be generated.

[0125] Specifically, the judgment between key features and query intent can be based on the association formula;

[0126] The associated formula is:

[0127] In the formula, feature M is composed of n features in S341, Indicates the probability that feature M is the key feature of query intent K.

[0128] Correspondingly, the embodiment of the present invention also discloses an information matching system for health management decision making, see Figure 8 As shown, the system includes:

[0129] The subject information acquisition module 11 is used to acquire decision subject information;

[0130] The structured processing module 12 is used to perform structured processing on the decision subject information to obtain structured subject information;

[0131] The encoding processing module 13 is used to encode the structured subject information to obtain a subject information encoding result that reflects the characteristics of the structured subject information;

[0132] The correlation calculation module 14 is used to calculate the correlation between the subject information coding result and the health care information coding result of each health care information;

[0133] The medical information extraction module 15 is used to extract the health and medical information corresponding to the health and medical information encoding result whose correlation with the subject information encoding result exceeds a preset first correlation threshold, and obtain the target health and medical information;

[0134] The health and medical information encoding result of each health and medical information is obtained by encoding each health and medical information in advance.

[0135] It can be seen that the embodiment of the present invention structures the decision-making subject information into a unified format, then performs encoding processing to obtain the subject information encoding result, calculates the correlation with a large amount of health and medical information encoding results that have been pre-encoded, and screens out the health and medical information corresponding to the health and medical information encoding result whose correlation exceeds the preset first association threshold as the target health and medical information, which is provided to medical staff for reference. By pre-encoding health and medical information of different directions and types and using it for matching, the matching dimension is increased, so it can effectively improve the limitations of health management decision-making content support, such as single content type, small amount of information, narrow application scenarios, and low decision support efficiency, and ensure the reliability and accuracy of the reference information provided.

[0136] Specifically, the encoding processing module 13 is specifically used to encode the structured subject information using a preset information encoding model to obtain a subject information encoding result that reflects the characteristics of the structured subject information;

[0137] Among them, the information encoding model is pre-trained using the historical decision-making subject information.

[0138] Specifically, the encoding processing module 13 includes a feature vector calculation unit and a vector integration unit; wherein,

[0139] A feature vector calculation unit, used to calculate the representation vector of each feature in the structured subject information by using the information encoding model;

[0140] The vector integration unit is used to integrate the representation vectors of various features using the information coding model to obtain the main information coding result for reflecting the structured main information characteristics.

[0141] Specifically, the above-mentioned correlation calculation module 14 is specifically used to calculate the correlation between the subject information coding result and the health care information coding result of each health care information by using a preset correlation calculation model;

[0142] Among them, the correlation calculation model is pre-trained using the historical subject information coding and historical health and medical information coding results.

[0143] Specifically, it may also include a cluster storage module; wherein,

[0144] The cluster storage module is used to pre-classify the health and medical information based on the preset information, and cluster and store it in the health and medical information database.

[0145] Specifically, the medical information extraction module 15 is specifically used to extract the associated health and medical information and the mapped health and medical information whose correlation with the information encoding result of the associated health and medical information exceeds a preset second correlation threshold from the health and medical information database, and obtain the target health and medical information including the associated health and medical information and the mapped health and medical information;

[0146] The associated health and medical information is health and medical information corresponding to the health and medical information encoding result whose correlation with the subject information encoding result exceeds a preset first correlation threshold.

[0147] Specifically, it also includes a new module for medical information; among them,

[0148] The new medical information module is used to add new information categories and add health and medical information under any information category.

[0149] Specifically, it may also include an intention query module, a feature missing judgment module and a feature missing reminder module; wherein,

[0150] An intention query module is used to perform query intention analysis on structured subject information using a preset intention analysis model to obtain a query intention corresponding to the structured subject information;

[0151] A feature missing judgment module is used to use the preset feature-intent mapping relationship to judge whether the key features corresponding to the query intent are missing;

[0152] A feature supplement module is used to obtain the missing key features in the decision subject information from the corresponding database to obtain complete decision subject information if the feature missing judgment module determines that the key features are missing;

[0153] A matching calling module, used for calling the subject information obtaining module to obtain the complete decision subject information as new decision subject information;

[0154] Among them, the intent analysis model is pre-trained using historical structured subject information, and the feature-intent mapping relationship records the correspondence between the query intent and the corresponding key features.

[0155] Specifically, the feature missing judgment module includes:

[0156] A feature removal unit, used to randomly remove features from historical decision subject information;

[0157] Count the error probability between each query intent and the missing features that causes the correlation calculation model to calculate errors;

[0158] According to the error probability, the correspondence between each query intent and key features is determined, and the mapping relationship between features and intent is generated.

[0159] Finally, 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 presence of other identical elements in the process, method, article or device including the elements.

[0160] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0161] The technical content provided by the present invention is introduced in detail above. Specific examples are used in this article to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea. At the same time, for general technical personnel in the field, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on the present invention.

Claims

1. An information matching method for health management decision making, It is characterized in that include: Obtain information about decision-making entities; The decision subject information is structured to obtain structured subject information; wherein, a decision subject information label is set, and corresponding feature labels are extracted from the decision subject information according to preset labels to obtain simplified structured subject information; the decision subject information label includes a subject basic information label, a symptom and sign label, a disease classification label, a medication name label, a medication usage and dosage label, a surgery classification label, a radiotherapy plan label, other treatment labels, a cost label, and a scientific research label; Performing encoding processing on the structured subject information to obtain a subject information encoding result for reflecting the characteristics of the structured subject information; Calculating the correlation between the subject information coding result and the health care information coding result of each health care information; Extracting health and medical information corresponding to the health and medical information encoding result whose correlation with the subject information encoding result exceeds a preset first correlation threshold, to obtain target health and medical information; The health and medical information encoding result of each health and medical information is obtained by encoding each health and medical information in advance; The process of encoding the structured subject information to obtain a subject information encoding result for reflecting the characteristics of the structured subject information includes: Using a preset information coding model to encode the structured subject information, to obtain a subject information coding result for reflecting the characteristics of the structured subject information; Wherein, the information encoding model is pre-trained using historical decision-making subject information; Before encoding the structured subject information, the method further includes: Performing query intent analysis on the structured subject information using a preset intent analysis model to obtain a query intent corresponding to the structured subject information; Using the preset feature-intent mapping relationship, determine whether the key features corresponding to the query intent are missing; If missing, obtain the missing key features in the decision-making subject information from the corresponding database or receive the missing key features in the decision-making subject information supplemented by the user to obtain the complete decision-making subject information; Reusing the complete decision subject information as decision subject information for structured processing to obtain the structured subject information; The intention analysis model is pre-trained using historical structured subject information, and the feature-intention mapping relationship records the corresponding relationship between the query intention and the corresponding key features; The process of generating the mapping relationship between features and intentions includes: Randomly remove features from historical decision-making subject information; Count the error probability between each query intent and the missing features that causes the correlation calculation model to calculate errors; According to the error probability, determining the corresponding relationship between each query intent and the key feature, and generating the mapping relationship between the feature and the intent; Among them, the correlation calculation model is: ; The missing feature M is obtained by removing the historical decision subject information Item features constitute, Indicates missing features The query intent Probability of error.

2. The information matching method for health management decision-making according to claim 1, It is characterized in that The process of encoding the structured subject information using a preset information encoding model to obtain a subject information encoding result for reflecting the characteristics of the structured subject information includes: Calculating the representation vector of each feature in the structured subject information using the information encoding model; The information encoding model is used to integrate the representation vectors of various features to obtain the subject information encoding result for reflecting the structured subject information characteristics.

3. The information matching method for health management decision-making according to claim 1, It is characterized in that Also includes: The health and medical information is pre-classified based on the preset information, clustered and stored in the health and medical information database.

4. The information matching method for health management decision-making according to claim 3, It is characterized in that The process of extracting health and medical information corresponding to the health and medical information encoding result whose correlation with the subject information encoding result exceeds a preset first correlation threshold to obtain target health and medical information includes: Extracting the associated health and medical information and the mapped health and medical information whose correlation with the information encoding result of the associated health and medical information exceeds a preset second correlation threshold from the health and medical information database, and obtaining target health and medical information including the associated health and medical information and the mapped health and medical information; The associated health and medical information is health and medical information corresponding to a health and medical information encoding result whose degree of association with the subject information encoding result exceeds a preset first association threshold.

5. The information matching method for health management decision-making according to claim 3, It is characterized in that Also includes: Add new information categories and add health and medical information under any information category.

6. An information matching system for health management decision making, It is characterized in that include: A subject information acquisition module is used to obtain decision-making subject information; A structured processing module is used to perform structured processing on the decision subject information to obtain structured subject information; wherein, the decision subject information label is set, and the corresponding feature label is extracted from the decision subject information according to the preset label to obtain simplified structured subject information; the decision subject information label includes subject basic information label, symptom and sign label, disease classification label, medication name label, medication usage and dosage label, surgery classification label, radiotherapy plan label, other treatment label, cost label and scientific research label; An encoding processing module, used for encoding the structured subject information to obtain a subject information encoding result for reflecting the characteristics of the structured subject information; A correlation calculation module, used to calculate the correlation between the subject information coding result and the health care information coding result of each health care information; A medical information extraction module, used to extract health and medical information corresponding to the health and medical information encoding result whose correlation with the subject information encoding result exceeds a preset first correlation threshold, to obtain target health and medical information; The health and medical information encoding result of each health and medical information is obtained by encoding each health and medical information in advance; Wherein, the encoding processing module includes: A coding processing unit, used to perform coding processing on the structured subject information using a preset information coding model to obtain a subject information coding result for reflecting the characteristics of the structured subject information; Wherein, the information encoding model is pre-trained using historical decision-making subject information; An intention query module, used to perform a query intention analysis on the structured subject information using a preset intention analysis model to obtain a query intention corresponding to the structured subject information; A feature missing judgment module is used to judge whether the key feature corresponding to the query intent is missing by using a preset feature-intent mapping relationship; A feature supplementation module is used to obtain the missing key features in the decision-making subject information from the corresponding database if they are missing, or to receive the missing key features in the decision-making subject information supplemented by the user, so as to obtain the complete decision-making subject information; A matching calling module, used to use the complete decision subject information as decision subject information again, so as to perform structured processing and obtain the structured subject information; The intention analysis model is pre-trained using historical structured subject information, and the feature-intention mapping relationship records the corresponding relationship between the query intention and the corresponding key features; Among them, the feature missing judgment module includes: The feature removal unit is used to randomly remove features from the historical decision subject information; count the error probability between each query intent and the missing feature that causes the correlation calculation model to calculate errors; determine the corresponding relationship between each query intent and the key feature according to the error probability, and generate the feature-intent mapping relationship; wherein the correlation calculation model is: ； The missing feature M is obtained by removing the historical decision subject information Item features constitute, Indicates missing features The query intent Probability of error.

Images