A method, system, device and medium for producing a special disease database

Abstract

The present invention provides a method, system, device and medium for producing a special disease database, which belongs to the field of medical big data technology. The method steps are as follows: first, construct a field production mode dictionary table to store recognition models, parsing and insertion scripts; by identifying the data source table, use the dictionary table to match the production mode; if there is no match, use a similar or basic model to identify the entity relationship, and when the recognition accuracy is lower than the threshold, obtain the recognition model through training or optimization; then, select or develop an adapted parsing and insertion script to produce the target field; finally, verify the database production accuracy, and update the production mode dictionary table after reaching the standard. The present invention realizes the rapid matching of existing resources in the production of special disease databases by constructing a production mode dictionary table, avoids repeated design, and improves efficiency; reduces waste by reusing adaptation resources; introduces similar model optimization to improve recognition accuracy; reduces manual dependence, and reduces technical requirements and costs.

Description

Background Art

[0002] With the development of medical information technology, building a disease-specific database is of great significance for clinical research, patient management, and optimization of disease diagnosis and treatment pathways. Traditional disease-specific database generation mainly involves code disassembly or identification of entities and relationships in power cases through a single model, and then parsing and organizing the results. First, building a disease-specific database by code disassembly requires a lot of manpower, which is not only inefficient and error-prone, but also prolongs the database construction cycle and has certain limitations on personnel capabilities. Secondly, due to different understandings of the same disease among different departments of the hospital, different terms, classifications, and diagnostic criteria are used. Therefore, the same entity (such as tumor) may be expressed differently in different department cases, representing different meanings, making it difficult for a single model to accurately identify and process the same entity in different department cases. The new generation of disease-specific database production method is to use different models for different fields for identification, then design parsing scripts to identify entities and relationships, obtain the content required for the disease-specific database field, and design insertion scripts to insert the required content into the specified position in the disease-specific database. This method ignores the similarity between the two fields. Even if the two fields are extremely similar, the model, parsing script, and insertion script need to be redesigned, resulting in a certain waste of resources.

[0003] This is the deficiency of the prior art. Therefore, it is very necessary to provide a method, system, equipment and medium for producing a special disease database to address the above-mentioned defects in the prior art. Summary of the invention

[0004] It is difficult to accurately identify and process the same entities in cases of different departments using a single model when building a special disease database. Disassembling cases through code takes a lot of time and is inefficient. It is also limited by the medical and database processing capabilities of personnel. The new generation method needs to use different models for identification for different fields, and then design parsing scripts and insertion scripts, ignoring the similarity of two fields, resulting in the defect of wasting resources. The present invention provides a special disease database production method, system, equipment and medium to solve the above technical problems.

[0005] In a first aspect, the present invention provides a method for producing a disease-specific database, comprising the following steps:

[0006] S1. Pre-build a dictionary table of production methods for fields in the disease-specific database to store the recognition model, parsing script, and insertion script of the fields;

[0007] S2. Obtain the data source table produced by the disease-specific database, and determine whether the target field can be queried in the production method dictionary table to find the required production method;

[0008] If yes, go to step S3;

[0009] If not, proceed to step S4;

[0010] S3. According to the target field, the corresponding recognition model, parsing script and insertion script are queried from the production mode dictionary table to produce the corresponding target field, and then the production ends;

[0011] S4. Use a similarity model or a basic model to identify entity relationships in the data source table, verify the recognition accuracy, and when the recognition accuracy is lower than a first threshold, obtain a recognition model by optimization or training, and identify entity relationships in the data source table by using the recognition model;

[0012] S5. Select an adapted parsing script and inserting script from the production mode dictionary table to produce the target field. If the adaptation is not possible, develop a parsing script and inserting script to produce the target field.

[0013] S6. Verify the accuracy of the production of the disease-specific database, and when the accuracy is higher than the second threshold, record the recognition model, parsing script and insertion script produced by the target field into the production method dictionary table.

[0014] Furthermore, the specific steps of step S1 are as follows:

[0015] S11. Construct a production mode dictionary table for each disease-specific database;

[0016] S12. Obtaining a recognition model for identifying entities and relationships of fields determined in the disease-specific database, a parsing script for parsing entities and relationships to obtain fields required by the disease-specific database, and an insertion script for inserting the fields required by the disease-specific database into corresponding positions in the disease-specific database;

[0017] S13. Using the concatenation result of the disease-specific library name and the field name as a keyword, and using the recognition model, parsing script, and insertion script of the production target field as content, a record of the production mode dictionary table is constructed;

[0018] S14. Build a model library based on the identification model in the production method dictionary table, build a parsing script library based on the parsing script in the production method dictionary table, and build an insertion script library based on the insertion script in the production method dictionary table.

[0019] Furthermore, the specific steps of step S2 are as follows:

[0020] S21. Obtain the name of the data source table representing the acquisition channel of the disease-specific database, and determine the target fields representing the content of the disease-specific database;

[0021] S22. concatenate the data source table name with the target field, and check whether there is a match between the concatenation result and the keyword in the production method dictionary table;

[0022] If yes, go to step S3;

[0023] If not, go to step S4.

[0024] Furthermore, the specific steps of step S4 are as follows:

[0025] S41. Randomly extract a set proportion of data from the data source table as test data;

[0026] S42. Obtain the target field of the data source table, and calculate the similarity between the target field name and the entity type identified by each recognition model in the model library;

[0027] If the similarity is greater than the threshold, proceed to step S43;

[0028] If the similarity is less than the threshold, proceed to step S44;

[0029] S43. Use the recognition model in the model library with a similarity greater than the threshold to identify entities and relationships in the target field in the data source table, and proceed to step S45;

[0030] S44. Use the pre-built basic model as the recognition model to perform entity and relationship recognition on the target field in the data source table;

[0031] S45. Calculate the recognition accuracy of the recognition model used using the test data;

[0032] If the recognition accuracy is greater than the first threshold, proceed to step S5;

[0033] If the recognition accuracy is less than or equal to the first threshold, proceed to step S46;

[0034] S46. Obtain the target field of the data source table, and calculate the similarity between the target field name and the entity type identified by each recognition model in the model library;

[0035] If there is no recognition model with a similarity greater than the threshold, proceed to step S47;

[0036] If there is a recognition model with a similarity greater than the threshold, proceed to step S48;

[0037] S47. Train the pre-built basic model to obtain a recognition model, and use the recognition model to identify entities and relationships in the target field in the data source table, and proceed to step S5;

[0038] S48. Optimize the recognition model with similarity greater than the threshold, and use the optimized recognition model to perform entity and relationship recognition on the target field in the data source table.

[0039] Furthermore, the specific steps of step S47 are as follows:

[0040] S471. Randomly extract a set proportion of data from the data source table as training data;

[0041] S472. Label the training data according to the source and target fields of the disease-specific database;

[0042] S473. Set the number of training times and hyperparameters for the basic model based on the target field;

[0043] S474. Input the training data into the basic model for iterative training until the model converges after a preset number of iterations, thereby obtaining a trained recognition model;

[0044] In step S48, the recognition model is optimized by adjusting hyperparameters, increasing training data, and adjusting the proportion of each field data in the training data.

[0045] Furthermore, the specific steps of step S5 are as follows:

[0046] S51. Select an adapted parsing script from the parsing script library of the production mode dictionary table;

[0047] If the parsing script cannot be adapted, go to step S52;

[0048] If the parsing script is adapted, the recognized entities and relationships are parsed using the adapted parsing script to obtain the content required by the disease-specific database field, and then the process goes to step S53;

[0049] S52. Develop a parsing script based on the identified entities and relationships, use the developed parsing script to parse the identified entities and relationships, obtain the content required by the disease library field, and add the developed parsing script to the parsing script library;

[0050] S53. Select an appropriate insert script from the insert script library of the production mode dictionary table;

[0051] If the insertion script cannot be adapted, go to step S54;

[0052] If the insertion script is adapted, the adapted insertion script is used to insert the required content of the parsed field of the special disease library into the designated position in the special disease library, and the process goes to step S55;

[0053] S54. Develop an insertion script based on the parsing results, use the developed insertion script to insert the required content of the parsed special disease library field into the specified position in the special disease library, and add the developed insertion script to the insertion script library.

[0054] Furthermore, the specific steps of step S6 are as follows:

[0055] S61. Calculate the overall accuracy of the recognition model, insertion script, and parsing script used in the target field production using the test data;

[0056] If the overall accuracy is greater than the second threshold, the process goes to step S62;

[0057] If the overall accuracy is less than or equal to the second threshold, return to step S42;

[0058] S62. Add the recognition model, insertion script, and parsing script used in target field production to the production dictionary table.

[0059] In a second aspect, the present invention provides a disease-specific database production system, comprising:

[0060] The dictionary table construction module is used to pre-construct the production mode dictionary table of the fields in the special disease database, and store the recognition model, parsing script and insertion script of the fields;

[0061] The dictionary table query module is used to obtain the data source table produced by the special disease database and determine whether the target field can query the required production method in the production method dictionary table;

[0062] The module for obtaining the table lookup of the identification model of the disease-specific library is used to query the corresponding identification model, parse the script and insert the script from the production method dictionary table according to the target field to produce the corresponding target field when the required production method is found.

[0063] The module for building a model for identifying a disease-specific database is used to identify the entity relationships in the data source table using a similarity model or a basic model when the required production method is not found, verify the recognition accuracy, and obtain a recognition model through optimization or training when the recognition accuracy is lower than a first threshold, and identify the entity relationships in the data source table through the recognition model;

[0064] The module for parsing and inserting the identification results of the disease-specific database selects the appropriate parsing script and insertion script from the production method dictionary table to produce the target field. If the parsing script and insertion script cannot be adapted, the module develops the parsing script and insertion script to produce the target field.

[0065] The dictionary table adding module is used to verify the accuracy of the production of the special disease database, and when the accuracy is higher than the second threshold, the recognition model, parsing script and insertion script produced by the target field are recorded in the production method dictionary table.

[0066] In a third aspect, the present invention provides an electronic device, including a processor and a memory;

[0067] The memory is used to store a computer program, and the processor is used to call and run the computer program from the memory, so that the electronic device executes the method described in the first aspect above.

[0068] In a fourth aspect, the present invention provides a storage medium,

[0069] The storage medium stores instructions, which, when executed on a computer, enable the computer to execute the method described in the first aspect.

[0070] The beneficial effects of the present invention are:

[0071] The method, system, device and medium for producing a special disease database provided by the present invention can quickly match existing recognition models, parsing scripts and insertion scripts when producing a special disease database by constructing a production mode dictionary table, thereby avoiding the tedious process of repeated design of each field and improving the production efficiency of the special disease database; by reusing and adapting existing resources, unnecessary waste of resources is reduced and resource utilization is improved; and the selection and optimization of similar models and basic models are introduced. When a suitable recognition model cannot be adapted, a model with high similarity can be automatically selected or a new recognition model can be obtained by training the basic model, thereby improving the recognition accuracy of the same entity in department medical records from different sources; reducing dependence on manual labor, lowering the requirements for personnel's technical level, and reducing labor costs.

[0072] It can be seen that compared with the prior art, the present invention has outstanding substantive features and significant progress, and the beneficial effects of its implementation are also obvious. BRIEF DESCRIPTION OF THE DRAWINGS

[0073] 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, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0074] Figure 1 It is a schematic flow chart of the method for producing a disease-specific database of the present invention.

[0075] Figure 2 It is a schematic diagram of the disease-specific database production system of the present invention. DETAILED DESCRIPTION

[0076] The method for producing a disease-specific database can be applied to one or more electronic devices, wherein the electronic device is a device that can automatically perform numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes but is not limited to microprocessors, application specific integrated circuits (ASICs), programmable gate arrays (FPGAs), digital processors (DSPs), embedded devices, etc.

[0077] An electronic device can be any electronic product that can interact with a user, such as a personal computer, a tablet computer, a smart phone, a personal digital assistant (PDA), an interactive network television (IPTV), etc. The network where the device is located includes but is not limited to the Internet, a wide area network, a metropolitan area network, a local area network, a virtual private network (VPN), etc.

[0078] 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 should fall within the scope of protection of the present invention.

[0079] See also Figure 1 The present invention shows a specific implementation of a method for producing a disease database, comprising the following steps:

[0080] S1. Pre-build a dictionary table of production methods for fields in the disease-specific database to store the recognition model, parsing script, and insertion script of the fields;

[0081] S2. Obtain the data source table produced by the disease-specific database, and determine whether the target field can be queried in the production method dictionary table to find the required production method;

[0082] If yes, go to step S3;

[0083] If not, proceed to step S4;

[0084] S3. According to the target field, the corresponding recognition model, parsing script and insertion script are queried from the production mode dictionary table to produce the corresponding target field, and then the production ends;

[0085] S4. Use a similarity model or a basic model to identify entity relationships in the data source table, verify the recognition accuracy, and when the recognition accuracy is lower than a first threshold, obtain a recognition model by optimization or training, and identify entity relationships in the data source table by using the recognition model;

[0086] S5. Select an adapted parsing script and inserting script from the production mode dictionary table to produce the target field. If the adaptation is not possible, develop a parsing script and inserting script to produce the target field.

[0087] S6. Verify the accuracy of the production of the disease-specific database, and when the accuracy is higher than the second threshold, record the recognition model, parsing script and insertion script produced by the target field into the production method dictionary table.

[0088] In this embodiment, the specific steps of step S1 are as follows:

[0089] S11. Construct a production mode dictionary table for each disease-specific database;

[0090] S12. Obtaining a recognition model for identifying entities and relationships of fields determined in the disease-specific database, a parsing script for parsing entities and relationships to obtain fields required by the disease-specific database, and an insertion script for inserting the fields required by the disease-specific database into corresponding positions in the disease-specific database;

[0091] The recognition model is mainly used to identify entities and relationships in electronic medical records. Entities refer to words or phrases with descriptive meanings, usually names of people, places, organizations, diseases, etc. Relationships refer to the relationship between two entities.

[0092] The recognition model identifies the contact entity relationship example as follows:

[0093] {"start":1,"end":3,"text":"Syphilis","labels":["Disease"]},{"start":22,"end":30,"text":"Vesicular-bullous lesions","labels":["Clinical findings"]},{"from_id":"Syphilis","to_id":"Vesicular-bullous lesions","type":"relation","direction":"right","labels":["Clinical manifestations"]}

[0094] Among them, start and end represent the position of the entity in the electronic medical record. For example, the position of syphilis is from the first character to the third character. Text represents the content of the entity, and label represents the label corresponding to the entity. From_id represents the starting entity in the relationship, and to_id represents the ending entity in the relationship. In the above example, it means that there is a clinical relationship between syphilis and vesicular-bullous skin lesions. Since each electronic medical record may have a large number of entities and relationships, it is necessary to use a python script to filter the recognition results of the model according to the label of the entity and the existing relationship, and finally obtain the content required by the field of the special disease library.

[0095] S13. Using the concatenation result of the disease-specific library name and the field name as a keyword, and using the recognition model, parsing script, and insertion script of the production target field as content, a record of the production mode dictionary table is constructed;

[0096] For example, the keywords of the production method dictionary are "Lung Cancer Special Disease Database_Lung Cancer Staging", and the records of the production method dictionary are specific production processes, such as "Model 1_Parsing Script 1_Insert Script 1", where the model refers to which model is used to identify the entities and relationships in the electronic medical record; the parsing script refers to which pre-constructed python script is used to parse the entities and relationships identified by the model, so as to obtain the required content of the special disease database field; the insert script refers to which pre-constructed sql script is used to insert the content into the corresponding position of the special disease database after obtaining the required content of the special disease database field. The complete dictionary table sample is as follows: {"Lung Cancer Special Disease Database_Lung Cancer Staging": "Model 1_Parsing Script 1_Insert Script 1", "Lung Cancer Special Disease Database_Onset Time": "Model 2_Parsing Script 2_Insert Script 2"};

[0097] S14. Construct a model library based on the recognition model in the production mode dictionary table, construct a parsing script library based on the parsing script in the production mode dictionary table, and construct an insertion script library based on the insertion script in the production mode dictionary table;

[0098] The specific steps of step S2 are as follows:

[0099] S21. Obtain the name of the data source table representing the acquisition channel of the disease-specific database, and determine the target fields representing the content of the disease-specific database;

[0100] For example, the disease discovery time in the lung cancer database comes from the hospital admission record, and the type of lung cancer comes from the test and examination. Both the hospital admission record and the test and examination can be used as the data source table of the lung cancer database.

[0101] S22. concatenate the data source table name with the target field, and check whether there is a match between the concatenation result and the keyword in the production method dictionary table;

[0102] If yes, go to step S3;

[0103] If not, proceed to step S4;

[0104] The specific steps of step S4 are as follows:

[0105] S41. Randomly extract a set proportion of data from the data source table as test data; for example, extract 10% of the data as test data;

[0106] S42. Obtain the target field of the data source table, and calculate the similarity between the target field name and the entity type identified by each recognition model in the model library;

[0107] If the similarity is greater than the threshold, proceed to step S43;

[0108] If the similarity is less than the threshold, proceed to step S44;

[0109] S43. Use the recognition model in the model library with a similarity greater than the threshold to identify entities and relationships in the target field in the data source table, and proceed to step S45;

[0110] S44. Use the pre-built basic model as the recognition model to perform entity and relationship recognition on the target field in the data source table;

[0111] For example, if the field of the disease-specific library to be produced is the onset time, a model for identifying time-type entities is selected from the model library; if there is no model similar to the field type of the disease-specific library, a pre-built basic model is used for identification;

[0112] S45. Calculate the recognition accuracy of the recognition model used using the test data;

[0113] If the recognition accuracy is greater than the first threshold, proceed to step S5;

[0114] If the recognition accuracy is less than or equal to the first threshold, proceed to step S46;

[0115] For example, the first threshold is set to 90%;

[0116] S46. Obtain the target field of the data source table, and calculate the similarity between the target field name and the entity type identified by each recognition model in the model library;

[0117] If there is no recognition model with a similarity greater than the threshold, proceed to step S47;

[0118] If there is a recognition model with a similarity greater than the threshold, proceed to step S48;

[0119] S47. Train the pre-built basic model to obtain a recognition model, and use the recognition model to identify entities and relationships of the target fields in the data source table, and proceed to step S5; the specific steps of step S47 are as follows:

[0120] S471. Randomly extract a set proportion of data from the data source table as training data;

[0121] S472. Label the training data according to the source and target fields of the disease-specific database;

[0122] S473. Set the number of training times and hyperparameters for the basic model based on the target field; the basic model can be a deep model global pointer, a machine learning model hidden Markov model, and a piece-wise CNN model;

[0123] S474. Input the training data into the basic model for iterative training until the model converges after a preset number of iterations, thereby obtaining a trained recognition model;

[0124] Taking the global pointer model as an example, the training process is as follows: First, extract a part of the data from the data source table for annotation; build a global pointer model and set hyperparameters such as the number of training times and batch_size. Input the data in the training set into the initial deep model for iterative training. In each iterative training, convert the input data into vectors, calculate the logit score of each vector, and generate a prediction matrix based on the calculated logit score of each vector. Then, use the cross entropy loss function to calculate the loss between the prediction matrix and the true label matrix corresponding to the data in the training set. Then, calculate the model gradient through the back propagation algorithm based on the calculated loss, and use the Adam optimizer to update the model parameters until the model converges or reaches the pre-set number of iterations. The training ends and a trained deep model is obtained.

[0125] S48. Optimize the recognition model whose similarity is greater than the threshold, and use the optimized recognition model to perform entity and relationship recognition on the target field in the data source table; optimize the recognition model by adjusting hyperparameters, increasing training data, and adjusting the proportion of data in each field in the training data;

[0126] The specific steps of step S5 are as follows:

[0127] S51. Select an adapted parsing script from the parsing script library of the production mode dictionary table;

[0128] If the parsing script cannot be adapted, go to step S52;

[0129] If the parsing script is adapted, the recognized entities and relationships are parsed using the adapted parsing script to obtain the content required by the disease-specific database field, and then the process goes to step S53;

[0130] S52. Develop a parsing script based on the identified entities and relationships, use the developed parsing script to parse the identified entities and relationships, obtain the content required by the disease library field, and add the developed parsing script to the parsing script library;

[0131] S53. Select an appropriate insert script from the insert script library of the production mode dictionary table;

[0132] If the insertion script cannot be adapted, go to step S54;

[0133] If the insertion script is adapted, the adapted insertion script is used to insert the required content of the parsed field of the special disease library into the designated position in the special disease library, and the process goes to step S55;

[0134] S54. Develop an insertion script based on the parsing results, use the developed insertion script to insert the required content of the parsed disease-specific library field into a specified position in the disease-specific library, and add the developed insertion script to the insertion script library;

[0135] The specific steps of step S6 are as follows:

[0136] S61. Calculate the overall accuracy of the recognition model, insertion script, and parsing script used in the target field production using the test data;

[0137] If the overall accuracy is greater than the second threshold, the process goes to step S62;

[0138] If the overall accuracy is less than or equal to the second threshold, return to step S42;

[0139] For example, the second threshold may be set to 95%;

[0140] S62. Add the recognition model, insertion script, and parsing script used in target field production to the production dictionary table.

[0141] It should be understood that the order of execution of the steps in the above embodiment does not mean that 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 embodiment of the present invention.

[0142] The following is an embodiment of a disease-specific database production system provided in an embodiment of the present disclosure. This system and the disease-specific database production methods of the above-mentioned embodiments belong to the same inventive concept. For details not described in detail in the embodiment of the disease-specific database production system, please refer to the above-mentioned disease-specific database production system embodiment.

[0143] like Figure 2 As shown, the system includes:

[0144] The dictionary table construction module is used to pre-construct the production mode dictionary table of the fields in the special disease database, and store the recognition model, parsing script and insertion script of the fields;

[0145] The dictionary table query module is used to obtain the data source table produced by the special disease database and determine whether the target field can query the required production method in the production method dictionary table;

[0146] The module for obtaining the table lookup of the identification model of the disease-specific library is used to query the corresponding identification model, parse the script and insert the script from the production method dictionary table according to the target field to produce the corresponding target field when the required production method is found.

[0147] The module for building a model for identifying a disease-specific database is used to identify the entity relationships in the data source table using a similarity model or a basic model when the required production method is not found, verify the recognition accuracy, and obtain a recognition model through optimization or training when the recognition accuracy is lower than a first threshold, and identify the entity relationships in the data source table through the recognition model;

[0148] The module for parsing and inserting the identification results of the disease-specific database selects the appropriate parsing script and insertion script from the production method dictionary table to produce the target field. If the parsing script and insertion script cannot be adapted, the module develops the parsing script and insertion script to produce the target field.

[0149] The dictionary table adding module is used to verify the accuracy of the production of the special disease database, and when the accuracy is higher than the second threshold, the recognition model, parsing script and insertion script produced by the target field are recorded in the production method dictionary table.

[0150] The method for producing a disease-specific database provided by the present invention is a combination of the modules and algorithm steps of each example described in the embodiments disclosed herein, and can be implemented by 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 terms of function in the above description. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can 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.

[0151] Those skilled in the art will appreciate that various aspects of the method for producing a disease-specific database provided by the present invention can be implemented as a system, method or program product. Therefore, various aspects of the present disclosure can be specifically implemented in the following forms, namely: a complete hardware implementation, a complete software implementation (including firmware, microcode, etc.), or a combination of hardware and software, which can be collectively referred to as "circuit", "module" or "system".

[0152] In the non-transitory computer-readable storage medium storing the method for producing a disease database, it can be, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination thereof. More specific examples (a non-exhaustive list) of readable storage media include: an electrical connection with one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof.

[0153] Although the present invention has been described in detail with reference to the accompanying drawings and in combination with the preferred embodiments, the present invention is not limited thereto. Without departing from the spirit and essence of the present invention, a person of ordinary skill in the art may make various equivalent modifications or substitutions to the embodiments of the present invention, and these modifications or substitutions shall be within the scope of the present invention. Any person of ordinary skill in the art may easily think of changes or substitutions within the technical scope disclosed by the present invention, and these shall be within the scope of protection of the present invention.

Claims

1. A method for producing a disease-specific database, characterized in that: The steps include: S1. Pre-build a dictionary table of production methods for fields in the disease database to store the recognition model, parsing script and insertion script of the field; the specific steps of step S1 are as follows: S11. Construct a production mode dictionary table for each disease-specific database; S12. Obtaining a recognition model for identifying entities and relationships of fields determined in the disease-specific database, a parsing script for parsing entities and relationships to obtain fields required by the disease-specific database, and an insertion script for inserting the fields required by the disease-specific database into corresponding positions in the disease-specific database; S13. Using the concatenation result of the disease-specific library name and the field name as a keyword, and using the recognition model, parsing script, and insertion script of the production target field as content, a record of the production mode dictionary table is constructed; S14. Construct a model library based on the recognition model in the production mode dictionary table, construct a parsing script library based on the parsing script in the production mode dictionary table, and construct an insertion script library based on the insertion script in the production mode dictionary table; S2. Obtain the data source table produced by the disease-specific database, and determine whether the target field can be queried in the production method dictionary table to find the required production method; If yes, go to step S3; If not, proceed to step S41; S3. According to the target field, the corresponding recognition model, parsing script and insertion script are queried from the production mode dictionary table to produce the corresponding target field, and then the production ends; S41. Randomly extract a set proportion of data from the data source table as test data; S42. Obtain the target field of the data source table, and calculate the similarity between the target field name and the entity type identified by each recognition model in the model library; If the similarity is greater than the threshold, proceed to step S43; If the similarity is less than the threshold, proceed to step S44; S43. Use the recognition model in the model library with a similarity greater than the threshold to identify entities and relationships in the target field in the data source table, and proceed to step S45; S44. Use the pre-built basic model as the recognition model to perform entity and relationship recognition on the target field in the data source table; S45. Calculate the recognition accuracy of the recognition model used using the test data; If the recognition accuracy is greater than the first threshold, proceed to step S5; If the recognition accuracy is less than or equal to the first threshold, proceed to step S46; S46. Obtain the target field of the data source table, and calculate the similarity between the target field name and the entity type identified by each recognition model in the model library; If there is no recognition model with a similarity greater than the threshold, proceed to step S47; If there is a recognition model with a similarity greater than the threshold, proceed to step S48; S47. Train the pre-built basic model to obtain a recognition model, and use the recognition model to identify entities and relationships in the target field in the data source table, and proceed to step S5; S48. Optimizing the recognition model whose similarity is greater than the threshold, and using the optimized recognition model to perform entity and relationship recognition on the target field in the data source table; S5. Select an adapted parsing script and inserting script from the production mode dictionary table to produce the target field. If the adaptation is not possible, develop a parsing script and inserting script to produce the target field. S6. Verify the accuracy of the production of the disease-specific database, and when the accuracy is higher than the second threshold, record the recognition model, parsing script and insertion script produced by the target field into the production method dictionary table.

2. The method for producing a disease-specific database according to claim 1, characterized in that: The specific steps of step S2 are as follows: S21. Obtain the name of the data source table representing the acquisition channel of the disease-specific database, and determine the target fields representing the content of the disease-specific database; S22. concatenate the data source table name with the target field, and check whether there is a match between the concatenation result and the keyword in the production method dictionary table; If yes, go to step S3; If not, go to step S41.

3. The method for producing a disease-specific database according to claim 2, characterized in that: The specific steps of step S47 are as follows: S471. Randomly extract a set proportion of data from the data source table as training data; S472. Label the training data according to the source and target fields of the disease-specific database; S473. Set the number of training times and hyperparameters for the basic model based on the target field; S474. Input the training data into the basic model for iterative training until the model converges after a preset number of iterations, thereby obtaining a trained recognition model; In step S48, the recognition model is optimized by adjusting hyperparameters, increasing training data, and adjusting the proportion of each field data in the training data.

4. The method for producing a disease-specific database according to claim 2, characterized in that: The specific steps of step S5 are as follows: S51. Select an adapted parsing script from the parsing script library of the production mode dictionary table; If the parsing script cannot be adapted, go to step S52; If the parsing script is adapted, the recognized entities and relationships are parsed using the adapted parsing script to obtain the content required by the disease-specific database field, and then the process goes to step S53; S52. Develop a parsing script based on the identified entities and relationships, use the developed parsing script to parse the identified entities and relationships, obtain the content required by the disease library field, and add the developed parsing script to the parsing script library; S53. Select an appropriate insert script from the insert script library of the production mode dictionary table; If the insertion script cannot be adapted, go to step S54; If the insertion script is adapted, the adapted insertion script is used to insert the required content of the parsed field of the special disease library into the designated position in the special disease library, and the process goes to step S55; S54. Develop an insertion script based on the parsing results, use the developed insertion script to insert the required content of the parsed special disease library field into the specified position in the special disease library, and add the developed insertion script to the insertion script library.

5. The method for producing a disease-specific database according to claim 4, characterized in that: The specific steps of step S6 are as follows: S61. Calculate the overall accuracy of the recognition model, insertion script, and parsing script used in the target field production using the test data; If the overall accuracy is greater than the second threshold, the process goes to step S62; If the overall accuracy is less than or equal to the second threshold, return to step S42; S62. Add the recognition model, insertion script, and parsing script used in target field production to the production dictionary table.

6. A system for producing a disease-specific database for implementing the method for producing a disease-specific database according to any one of claims 1 to 5, characterized in that: include: The dictionary table construction module is used to pre-construct the production mode dictionary table of the fields in the special disease database, and store the recognition model, parsing script and insertion script of the fields; The dictionary table query module is used to obtain the data source table produced by the special disease database and determine whether the target field can query the required production method in the production method dictionary table; The module for obtaining the table lookup of the identification model of the disease-specific library is used to query the corresponding identification model, parse the script and insert the script from the production method dictionary table according to the target field to produce the corresponding target field when the required production method is found. The module for building a model for identifying a disease-specific database is used to identify the entity relationships in the data source table using a similarity model or a basic model when the required production method is not found, verify the recognition accuracy, and obtain a recognition model through optimization or training when the recognition accuracy is lower than a first threshold, and identify the entity relationships in the data source table through the recognition model; The module for parsing and inserting the identification results of the disease-specific database selects the appropriate parsing script and insertion script from the production method dictionary table to produce the target field. If the parsing script and insertion script cannot be adapted, the module develops the parsing script and insertion script to produce the target field. The dictionary table adding module is used to verify the accuracy of the production of the special disease database, and when the accuracy is higher than the second threshold, the recognition model, parsing script and insertion script produced by the target field are recorded in the production method dictionary table.

7. An electronic device, characterized in that: including a processor and a memory; The memory is used to store a computer program, and the processor is used to call and run the computer program from the memory, so that the electronic device executes the method described in any one of claims 1 to 5.

8. A storage medium, characterized in that: The storage medium stores instructions, which, when executed on a computer, enable the computer to execute the method described in any one of claims 1 to 5.

Images