Method and system for metadata matching applied to different business systems

Abstract

The application relates to a method and system for metadata matching applied to different business systems, wherein the method comprises the following steps: constructing a training set and a test set according to a prior knowledge base, training a hybrid algorithm model according to the training set and the test set, obtaining metadata to be matched, and querying whether there is a matching result in the prior knowledge base; if yes, the matching result is directly returned; if not, a candidate set of the metadata to be matched is obtained from the prior knowledge base through a coarse classification model, and a matching result of the metadata to be matched is obtained from the candidate set through the hybrid algorithm model. Through the application, the problems of low efficiency and poor precision of metadata matching between different business systems are solved, the complexity of the algorithm is reduced from the data level based on the prior knowledge base and the coarse classification model, and the accuracy of the matching is further improved through the hybrid algorithm model.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of term matching, and in particular to a method and system for metadata matching applied to different business systems. BACKGROUND

[0002] In a business system, metadata is descriptive information of data and information resources. As an important part of a business system, metadata plays a crucial role in the normal operation of the system. In a hospital, for example, a large number of business systems run in the production environment of the hospital. Because the metadata in different business systems is variously expressed, and the naming of metadata is based on specific business scenarios, the same metadata has different names, which greatly affects the association and application of hospital medical data.

[0003] In the past, the connection of data in different business systems was based on manual matching, which required a lot of time and manual input, greatly prolonging the online time of the business system. How to quickly, simply and efficiently match metadata has become an urgent problem to be solved.

[0004] Similarity algorithm is a general method in the technical field of term matching. However, the similarity algorithm generally refers to algorithms such as character surface and character order. The results of the above algorithms are not satisfactory, because metadata contains rich semantic classification and abstract semantic information.

[0005] At present, there is no effective solution to the problem of low efficiency and poor precision of metadata matching between different business systems in the related art. SUMMARY

[0006] The embodiments of the present application provide a method and system for metadata matching applied to different business systems, to at least solve the problem of low efficiency and poor precision of metadata matching between different business systems in the related art.

[0007] In a first aspect, the embodiments of the present application provide a method for metadata matching applied to different business systems, the method comprising:

[0008] constructing a training set and a test set according to a prior knowledge base, and training a hybrid algorithm model according to the training set and the test set;

[0009] obtaining metadata to be matched, and querying whether there is a matching result in the prior knowledge base;

[0010] if yes, directly returning the matching result;

[0011] if no, obtaining a candidate set of the metadata to be matched from the prior knowledge base through a coarse ranking model, and obtaining a matching result of the metadata to be matched from the candidate set through the hybrid algorithm model.

[0012] In some embodiments, the hybrid algorithm model comprises a feature extractor and a classifier, wherein the feature extractor comprises an edit distance algorithm model, a component radical similarity algorithm model and a semantic similarity algorithm model, and the classifier is trained based on an AdaBoost algorithm.

[0013] In some embodiments, obtaining the candidate set of the to-be-matched metadata from the prior knowledge base by a coarse ranking model comprises:

[0014] segmenting the vocabulary data in the prior knowledge base by using a forward-backward maximum matching algorithm, and according to the segmentation result, screening out the candidate set of the to-be-matched metadata by a coarse ranking model based on a BM25 algorithm.

[0015] In some embodiments, obtaining the matching result of the to-be-matched metadata from the candidate set by the hybrid algorithm model comprises:

[0016] obtaining the literal similarity feature of the to-be-matched metadata and the candidate set by an edit distance algorithm model;

[0017] splitting the vocabulary of the to-be-matched metadata and the candidate set according to the component radical, to obtain the character form similarity feature, by a component radical similarity algorithm model;

[0018] mapping the vocabulary of the to-be-matched metadata and the candidate set in a high-dimensional semantic space, to obtain the semantic similarity feature based on vector calculation, by a semantic similarity algorithm model;

[0019] obtaining the matching result of the to-be-matched metadata from the candidate set by a classifier according to the literal similarity feature, the character form similarity feature and the semantic similarity feature.

[0020] In some embodiments, training the hybrid algorithm model according to the training set and the test set comprises:

[0021] training a semantic similarity algorithm model according to the training set and the test set, wherein the semantic similarity algorithm model is a SimCSE algorithm model, and the SimCSE algorithm model is composed of a pre-trained Bert model and a contrastive learning method;

[0022] training a classifier by using an edit distance algorithm model, a component radical similarity algorithm model and a semantic similarity algorithm model as a feature extractor according to the training set and the test set, wherein the classifier is an AdaBoost algorithm model.

[0023] In some embodiments, the method further comprises:

[0024] The method and dependent environment for metadata matching applied to different systems are integrated into a docker container through container technology.

[0025] In some embodiments, after obtaining the metadata to be matched, the method comprises:

[0026] normalizing the punctuation of the metadata to be matched, unifying the character representation of the metadata to be matched, and deleting repeated and invalid semantic characters in the metadata to be matched to obtain processed metadata to be matched.

[0027] In some embodiments, before constructing the training set and the test set according to the prior knowledge base, the method comprises:

[0028] Obtaining metadata in books and the Internet, obtaining data synonyms and core semantic words of the metadata through data cleaning and proofreading, and then constructing a prior knowledge base.

[0029] In some embodiments, querying the prior knowledge base for a matching result comprises:

[0030] Querying the prior knowledge base for a matching result through whole word matching.

[0031] In a second aspect, the embodiments of the present application provide a system for metadata matching applied to different business systems, the system comprising a model training module and a metadata matching module.

[0032] The model training module is configured to construct a training set and a test set according to a prior knowledge base, and to train a hybrid algorithm model according to the training set and the test set.

[0033] The metadata matching module is configured to obtain metadata to be matched, and to query the prior knowledge base for a matching result.

[0034] If yes, the matching result is directly returned.

[0035] If no, a candidate set of the metadata to be matched is obtained from the prior knowledge base through a coarse ranking model, and a matching result of the metadata to be matched is obtained from the candidate set through the hybrid algorithm model.

[0036] Compared with the related art, the method and system for metadata matching applied to different business systems provided by the embodiments of the present application, a training set and a test set are constructed according to a prior knowledge base, and then a hybrid algorithm model is trained according to the training set and the test set; the metadata to be matched is obtained, and it is inquired whether there is a matching result in the prior knowledge base; if yes, the matching result is directly returned; if not, a candidate set of the metadata to be matched is obtained from the prior knowledge base through a coarse sorting model, and then a matching result of the metadata to be matched is obtained from the candidate set through the hybrid algorithm model. The problems of low efficiency and poor precision of metadata matching between different business systems are solved, the complexity of the algorithm is reduced from the data level based on the prior knowledge base and the coarse sorting model, and the accuracy of the matching is further improved by the hybrid algorithm model. BRIEF DESCRIPTION OF DRAWINGS

[0037] The drawings described herein are used to provide further understanding of the present application, and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application, and do not constitute an improper limitation on the present application. In the drawings:

[0038] Figure 1 is a step flowchart of a method for metadata matching applied to different business systems according to an embodiment of the present application;

[0039] Figure 2 is a flowchart of training of a semantic similarity algorithm model according to an embodiment of the present application;

[0040] Figure 3 is a flowchart of training of a classifier according to an embodiment of the present application;

[0041] Figure 4 is a flowchart of model matching after training according to a specific embodiment of the present application;

[0042] Figure 5 is a structural block diagram of a system for metadata matching applied to different business systems according to an embodiment of the present application;

[0043] Figure 6 is a schematic diagram of the internal structure of an electronic device according to an embodiment of the present application.

[0044] BRIEF DESCRIPTION OF DRAWINGS: 51, model training module; 52, metadata matching module. DETAILED DESCRIPTION

[0045] In order to make the objects, technical solutions and advantages of the present application clearer, the present application will be described and illustrated below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and should not be used to limit the present application. Based on the embodiments provided by the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort belong to the scope of the present application.

[0046] Obviously, the accompanying drawings in the following description are only some examples or embodiments of the present application, and for those of ordinary skill in the art, the present application can also be applied to other similar scenarios without creative effort based on the accompanying drawings. In addition, it can be understood that although the efforts made in the development process can be complex and lengthy, for those of ordinary skill in the art related to the content disclosed in the present application, some design, manufacture or production changes based on the technical content disclosed in the present application are only routine technical means and should not be understood as insufficient disclosure of the content disclosed in the present application.

[0047] In the present application, the phrase "embodiments" means that the specific features, structures or characteristics described in conjunction with the embodiments can be included in at least one embodiment of the present application. The appearance of this phrase at various places in the specification does not necessarily mean the same embodiment, nor is it an independent or alternative embodiment to other embodiments. It is explicitly and implicitly understood by those of ordinary skill in the art that the embodiments described in the present application can be combined with other embodiments without conflict.

[0048] Unless otherwise defined, technical terms and scientific terms used in the present application shall have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. The terms "a", "an", "one", "this", and similar referents in the context of describing the application are to be construed to be inclusive, not exclusive. The terms "comprise", "comprising", "comprises", "include", "including", "includes", "contain", "containing", "contains", and any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, product, or apparatus that comprises a list of steps or units (elements) is not necessarily limited to the listed steps or units, but can include additional steps or units not expressly listed or can also include additional steps or units inherent to such process, method, product, or apparatus. The terms "connected", "coupled", and "coupling" are not limited to direct or physical connections, but can include indirect connections, such as electrical connections. The term "plurality" means two or more. The term "and / or" describes associated objects in association relationships, which means that there can be three relationships, such as "A and / or B", which means that A exists alone, A and B exist together, and B exists alone. The character " / " generally represents an "or" relationship between the associated objects. The terms "first", "second", "third", and the like are merely used to distinguish similar objects, and do not represent a specific order of the objects.

[0049] Currently, the connection of data of different business systems is based on manual matching method, which requires a lot of time and manual input, greatly prolonging the online time of the business system. How to quickly, simply and efficiently match metadata has become an urgent problem to be solved.

[0050] Similarity algorithm is a general method in the field of term matching technology. However, the similarity algorithm generally refers to the algorithm of text surface and text order. The matching result of the above algorithm is not satisfactory, because the metadata contains rich semantic classification and abstract semantic information.

[0051] Therefore, the present application proposes a metadata matching method and system applied to different business systems, which contains algorithm model and container technology encapsulation. The algorithm model is the combination of knowledge information and machine learning algorithm, and the container technology is to realize the rapid deployment and use in the hospital.

[0052] The algorithm model specifically includes a hybrid algorithm model, and an AdaBoost algorithm model as a classifier of the hybrid algorithm model. The core idea is to train different weak classifiers for the same training set, and then combine the different weak classifiers to form a strong classifier. Each algorithm (edit distance algorithm, radical similarity algorithm, and semantic similarity algorithm) in the hybrid algorithm model serves as a feature extractor to evaluate the similarity from multiple perspectives such as literal, sequence, and semantics. The AdaBoost algorithm uses numerous features to train an efficient and accurate strong classifier. The hybrid model of multiple similarities evaluates the matching degree of the metadata from multiple perspectives, improves the generalization performance of a single algorithm, and effectively broadens the algorithm usage scenarios.

[0053] In addition, the data of the hospital has strong privacy. In order to prevent data leakage, a set of metadata matching tools that can be quickly deployed and implemented in the hospital is needed. Container technology is to package applications and dependent environments into a lightweight and portable virtual environment, and then publish them to any popular Linux machine. The docker container is a specific embodiment of the container technology, which can embed a stable and secure system environment, and pre-integrate the files required by the service and program in the container, to quickly and efficiently deploy services or programs. Therefore, the container technology (specifically docker) is used to deploy the metadata matching tool to the hospital in a lightweight manner.

[0054] The embodiment of the present application provides a metadata matching method applied to different business systems, Figure 1 The step flow chart of the metadata matching method applied to different business systems according to the embodiment of the present application is shown in Figure 1 The method comprises the following steps:

[0055] In step S102, a training set and a test set are constructed according to a priori knowledge base, and then a hybrid algorithm model is trained according to the training set and the test set.

[0056] Specifically, the semantic similarity algorithm model is trained according to the training set and the test set, wherein the semantic similarity algorithm model is a SimCSE algorithm model, and the SimCSE algorithm model is composed of a pre-trained Bert model and a contrast learning method. The pre-trained Bert model adopts a Bert-wwm-ext model, which has the advantages of large training data set and large number of training steps, and can better interpret the semantic representation of Chinese words; the contrast learning method helps the pre-trained model to further explore the latent semantics and solve the collapse problem of the Bert model, so as to meet the reasonable representation of semantics in the vector space.

[0057] Preferably, Figure 2 The flow chart of training the semantic similarity algorithm model according to the embodiment of the present application is shown in Figure 2As shown, according to the prior knowledge base, the training set and the test set are constructed, and the data content ratio is 8:2. Among them, the synonym mapping is used as the positive sample of the training set and the test set, and the negative sample is to traverse each metadata in the prior knowledge base, and randomly select 16 metadata from the non-synonym set, such as the metadata "upper respiratory tract infection" and "upper respiratory infection", which are positive samples, and "skin infection", "respiratory infection", "trauma infection" and the like are used as negative samples.

[0058] The target of the simCSE algorithm model is to train an encoder f, so that:

[0059] score(f(x), f(x + )) > score(f(x), f(x - ))

[0060] Wherein, x + is a positive sample similar to x, x - is a negative sample similar to x, and score is a metric function to measure the similarity between samples;

[0061] The loss function of the simCSE algorithm model is the InfoNCE loss, and the formula is as follows:

[0062]

[0063] Among them, the corresponding sample x has 1 positive sample and N-1 negative samples, and the target is that the features of x and the positive sample are more similar, and the features of the N-1 negative samples are more dissimilar.

[0064] Since the pre-trained Bert model has learned a large amount of Chinese prediction information, and the Bert-wwm-ext model is used, it has the advantages of large training data set and large number of training steps, and can better interpret the semantic representation of Chinese words. It is reasonable to output the semantic representation of the metadata through pre-training. On this basis, the infoNCE loss with the negative example is calculated, and the model parameters are corrected through back propagation to ensure that different metadata have different characteristics in the high-dimensional vector space.

[0065] Specifically, according to the training set and the test set, the edit distance algorithm model, the radical similarity algorithm model and the semantic similarity algorithm model are used as feature extractors to obtain the multi-feature between the metadata, so as to train the classifier, wherein the classifier is an AdaBoost algorithm model, and a decision tree (Desision Tree) is used as a weak classifier of AdaBoost. The above multi-features are iteratively trained to obtain a strong classifier.

[0066] Preferably, Figure 3 is a flowchart of the classifier training according to the embodiments of the present application, such as Figure 3As shown, based on three algorithms of edit distance, component and semantic representation, features are extracted and strong classifier is trained. The steps are as follows:

[0067] The similarity of positive and negative examples is calculated by using edit distance algorithm, component similarity algorithm and semantic similarity algorithm respectively, which are x, y and z respectively. The statistics of all positive and negative examples form (x n ,y n ,z n )→{0,1}.

[0068] Initialize the weight distribution of training data, that is:

[0069]

[0070] Where w 1i is the weight and is uniformly distributed, and N is the number of samples.

[0071] Learn weak classifier using data set with weight distribution D m , that is:

[0072] G(x):X→{0,1}

[0073] Where G(x) is the combination of weak classifiers.

[0074] Calculate the error rate of weak classifier, that is:

[0075]

[0076] Where x i and y i are the i-th sample and label respectively, and w 1i is the weight of weak classifier.

[0077] Calculate G m (x) coefficient, that is e m is the error rate of weak classifier under the current weight distribution.

[0078] Update the weight distribution of training data set, increase the weight of error sample and decrease the weight of correct sample, that is:

[0079] D m+1 =(w m+1，1 ,..., w m+1，N )

[0080]

[0081] Z m is the normalization factor, that is:

[0082]

[0083] Constructing a linear combination of weak classifiers, namely:

[0084]

[0085] Thus, the strong classifier is obtained:

[0086] G(x) = sin(f(x))

[0087] Step S104, obtaining the metadata to be matched, querying whether there is a matching result in the priori knowledge base;

[0088] Preferably, after obtaining the metadata to be matched, the punctuation of the metadata to be matched is unified, such as "bacterial dysentery, other" is converted into "bacterial dysentery, other"; the character representation of the metadata to be matched is unified, such as "a group of streptococcal sepsis" is converted into "a group of streptococcal sepsis"; the repeated and invalid semantic characters in the metadata to be matched are deleted, and the processed metadata to be matched is obtained, such as "intestinal adhesion [with] obstruction" is converted into "intestinal adhesion with obstruction";

[0089] According to the processed metadata to be matched, it is queried whether there is a matching result in the priori knowledge base, wherein the query of the priori knowledge base is quickly realized according to the full word matching mode, and the returned result is quick and accurate; the priori knowledge base not only provides training and test data for the training model, but also reduces the metadata matching workload and effectively shortens the query time.

[0090] Step S106, if yes, the matching result is directly returned;

[0091] Step S108, if no, the candidate set of the metadata to be matched is obtained from the priori knowledge base through the coarse sorting model, and the matching result of the metadata to be matched is obtained from the candidate set through the hybrid algorithm model.

[0092] Specifically, the word segmentation is performed on the vocabulary data in the priori knowledge base by using the forward and backward maximum matching algorithm, and according to the word segmentation result, the candidate set of the metadata to be matched is screened out by the coarse sorting model based on the BM25 algorithm.

[0093] Preferably, the coarse screening model is based on the BM25 algorithm, the word segmenter splits the vocabulary in the metadata set, the BM25 calculates the relevance degree with the matched vocabulary, and the steps are as follows:

[0094] Load the core word dictionary in the priori knowledge base into the system cache, and the core word dictionary is the ontology vocabulary of the metadata;

[0095] Obtain the metadata set, call the forward and backward maximum matching algorithm to cut the word, and return the word segmentation result;

[0096] According to the word segmentation result, the relevance is calculated by BM25 to obtain a candidate set of words with high relevance.

[0097] Specifically, the literal similarity feature of the to-be-matched metadata and the candidate set is obtained by using an edit distance algorithm model; the word shape similarity feature is obtained by splitting the words of the to-be-matched metadata and the candidate set according to the components of the words by using a component similarity algorithm model; and the semantic similarity feature is obtained by mapping the words of the to-be-matched metadata and the candidate set in a high-dimensional semantic space based on a semantic similarity algorithm model.

[0098] Finally, the matching result of the to-be-matched metadata is obtained from the candidate set by a classifier according to the literal similarity feature, the word shape similarity feature, and the semantic similarity feature.

[0099] By steps S102 to S108 of the embodiment of the present application, the problem of low efficiency and poor accuracy of metadata matching between different business systems is solved, the complexity of the algorithm is reduced from the data level based on the prior knowledge base and the coarse sorting model, and the mixed algorithm model further improves the accuracy of the matching.

[0100] In some embodiments, before constructing the training set and the test set according to the prior knowledge base in step S102, the metadata in books and the Internet is obtained, the data synonyms and core semantic words of the metadata are obtained through data cleaning and checking, and then the prior knowledge base is constructed.

[0101] Specifically, according to the metadata categories and characteristics, the existing metadata knowledge from multiple sources is integrated, the metadata synonyms and core semantic words are obtained through data cleaning and checking (such as manual checking), the prior mapping relationship is stored in the form of key-value pair, such as being sorted into the corresponding form of “upper respiratory tract infection” and “upper respiratory infection”, and then the prior knowledge base is constructed.

[0102] In some embodiments, after the model is trained according to the above steps, the method model and the dependent environment of the metadata matching of different systems are integrated into a docker container through container technology.

[0103] Specifically, the container technology encapsulation is to encapsulate the foregoing algorithm model into a new system environment by using container technology. The image encapsulated by Dockek contains the complete application system environment and the dependencies of the algorithm model. The program in the container can run independently on any server and single machine, and the term matching function is stably implemented.

[0104] The embodiment of the present application provides a metadata matching method applied to different business systems, Figure 4 is a process schematic diagram of model matching trained according to the embodiment of the present application, as Figure 4 shown.

[0105] The metadata to be matched, "upper respiratory tract infection", needs to go through a data preprocessing layer first. The specific process is similar to the preprocessing process in step S104 of the foregoing embodiment.

[0106] After preprocessing, the query construction priori knowledge base, such as the return result, is the final result, or it enters the algorithm coarse ranking.

[0107] In the algorithm coarse ranking, the tokenizer is used for word segmentation, the BM25 algorithm is used to calculate the relevance, some similar words are preliminarily screened out as candidate metadata, and then enter the fine ranking module. For example, the query example "upper respiratory tract infection", the coarse ranking result is "respiratory tract infection", "respiratory infection", "skin infection", "head infection", etc.

[0108] The algorithm fine ranking is to calculate the similarity from the aspects of literal, radical, semantic, etc. for each candidate metadata word, and then input the similarity feature into the AdaBoost as a model to obtain the final similarity. The value range of the similarity is between 0 and 1. The closer to 0 indicates that the similarity between the metadata is poor, and the closer to 1 indicates that the similarity between the metadata is higher.

[0109] It should be noted that the steps shown in the above process or the flowchart of the accompanying drawings can be executed in a computer system such as a group of computer executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from here.

[0110] The embodiment of the present application provides a metadata matching system applied to different business systems, Figure 5 The structure block diagram of the metadata matching system applied to different business systems according to the embodiment of the present application is shown as Figure 5 The system includes a model training module 51 and a metadata matching module 52.

[0111] The model training module 51 is used to construct a training set and a test set according to a priori knowledge base, and then train a hybrid algorithm model according to the training set and the test set.

[0112] The metadata matching module 52 is used to obtain metadata to be matched, and query whether there is a matching result in the priori knowledge base.

[0113] If yes, the matching result is directly returned.

[0114] If no, the candidate set of the metadata to be matched is obtained from the priori knowledge base through a coarse ranking model, and then the matching result of the metadata to be matched is obtained from the candidate set through a hybrid algorithm model.

[0115] Through the model training module 51 and the metadata matching module 52 in the embodiment of the present application, the problem of low efficiency and poor accuracy of metadata matching between different business systems is solved, the complexity of the algorithm is reduced from the data level based on the prior knowledge base and the rough sorting model, and the mixed algorithm model further improves the accuracy of the matching.

[0116] It should be noted that each of the above modules can be a functional module or a program module, and can be implemented by software or hardware. For the modules implemented by hardware, each of the above modules can be located in the same processor; or each of the above modules can also be located in different processors in any combination.

[0117] The embodiment also provides an electronic device including a memory and a processor, the memory storing a computer program, and the processor being configured to execute the computer program to perform the steps in any of the method embodiments.

[0118] Optionally, the electronic device can further include a transmission device and an input / output device, wherein the transmission device is connected with the processor, and the input / output device is connected with the processor.

[0119] It should be noted that the specific examples in the embodiment can refer to the examples described in the above embodiments and optional implementation manners, and the embodiment will not be described here.

[0120] In addition, in combination with the method for metadata matching applied to different business systems in the above embodiments, the embodiment of the present application can provide a storage medium for implementation. The storage medium stores a computer program; the computer program is executed by a processor to implement any of the methods for metadata matching applied to different business systems in the above embodiments.

[0121] In one embodiment, a computer device is provided, which can be a terminal. The computer device comprises a processor, a memory, a network interface, a display screen and an input device connected through a system bus. The processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device comprises a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for running the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is configured to communicate with an external terminal through a network connection. The computer program, when executed by the processor, implements a method for metadata matching applied to different business systems. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer overlaid on the display screen, or a key, trackball or touchpad arranged on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.

[0122] In one embodiment, Figure 6 is a schematic diagram of the internal structure of an electronic device according to an embodiment of the present application, as Figure 6 shown, an electronic device is provided, which can be a server, and a schematic diagram of the internal structure of the electronic device can be as Figure 6 shown. The electronic device comprises a processor, a network interface, an internal memory and a non-volatile memory connected through an internal bus, wherein the non-volatile memory stores an operating system, a computer program and a database. The processor is configured to provide computing and control capabilities, the network interface is configured to communicate with an external terminal through a network connection, the internal memory is configured to provide an environment for running the operating system and the computer program, the computer program, when executed by the processor, implements a method for metadata matching applied to different business systems, and the database is configured to store data.

[0123] Those skilled in the art can understand that Figure 6 the structure shown in the above embodiment is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the electronic device to which the scheme of the present application is applied. The specific electronic device can comprise more or fewer components than those shown in the diagram, or combine certain components, or have a different arrangement of components.

[0124] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer readable storage medium, and when the computer program is executed, the processes of the above-mentioned embodiments of the methods can be included. Any reference to memory, storage, databases, or other media in the embodiments provided by the present application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration but not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0125] Those skilled in the art should understand that any combination of the technical features of the above-mentioned embodiments can be made. In order to make the description simple, all possible combinations of the technical features in the above-mentioned embodiments are not described, however, as long as the combination of the technical features does not exist contradictory, it should be considered as the scope of the present application.

[0126] The above-mentioned embodiments only express several implementation manners of the present application, and the description is more specific and detailed, but it should not be understood as a limitation on the scope of the patent. It should be pointed out that for ordinary skilled in the art, without departing from the concept of the present application, a number of modifications and improvements can be made, which are all within the scope of the present application. Therefore, the scope of the patent of the present application should be subject to the appended claims.

Claims

1. A method for metadata matching applied to different business systems, characterized in that, The method includes: A training set and a test set are constructed based on a prior knowledge base, and then a hybrid algorithm model is trained based on the training set and the test set. The hybrid algorithm model and its dependent environment are integrated into a Docker container using container technology for deployment within the hospital. The hybrid algorithm model is used for metadata matching between different business systems within the hospital. The hybrid algorithm model includes a feature extractor and a classifier. The feature extractor includes an edit distance algorithm model, a radical similarity algorithm model, and a semantic similarity algorithm model. The classifier is trained based on the AdaBoost algorithm. Obtain the metadata to be matched, and query whether there is a matching result in the prior knowledge base; If so, return the matching result directly; If not, then the candidate set of the metadata to be matched is obtained from the prior knowledge base through the coarse ranking model, and then the literal similarity features between the metadata to be matched and the candidate set are obtained through the edit distance algorithm model. Using the radical similarity algorithm model, the metadata to be matched and the vocabulary of the candidate set are split according to radicals to obtain character shape similarity features; The semantic similarity algorithm model maps the metadata to be matched and the vocabulary of the candidate set into a high-dimensional semantic space, and obtains semantic similarity features based on vector calculation. Based on the literal similarity features, the glyph similarity features, and the semantic similarity features, a classifier is used to obtain the matching result of the metadata to be matched from the candidate set.

2. The method according to claim 1, characterized in that, The candidate set of metadata to be matched obtained from the prior knowledge base through the coarse-ranking model includes: The prior knowledge base vocabulary data is segmented using the forward and backward maximum matching algorithm. Based on the segmentation results, a candidate set of the metadata to be matched is selected using a coarse-ranking model based on the BM25 algorithm.

3. The method according to claim 1, characterized in that, Based on the training set and the test set, the hybrid algorithm model trained includes: The semantic similarity algorithm model is trained based on the training set and the test set. The semantic similarity algorithm model is the SimCSE algorithm model, which consists of a pre-trained BERT model and a contrastive learning method. Based on the training set and the test set, the classifier is trained using the edit distance algorithm model, the radical similarity algorithm model and the semantic similarity algorithm model as feature extractors, wherein the classifier is the AdaBoost algorithm model.

4. The method according to claim 1, characterized in that, After obtaining the metadata to be matched, the method includes: The punctuation marks of the metadata to be matched are standardized, the character representation of the metadata to be matched is unified, and characters with duplicate and invalid semantics in the metadata to be matched are deleted to obtain the processed metadata to be matched.

5. The method according to claim 1, characterized in that, Before constructing the training and test sets based on the prior knowledge base, the method includes: Metadata from books and the Internet is acquired, and data synonyms and core semantic words of the metadata are obtained through data cleaning and verification, thereby constructing a priori knowledge base.

6. The method according to claim 1, characterized in that, Querying whether the prior knowledge base has a matching result includes: The prior knowledge base is queried to determine if a matching result exists by using whole-word matching.

7. A system for matching metadata applied to different business systems, characterized in that, The system is used to perform the method according to any one of claims 1 to 6, the system comprising a model training module and a metadata matching module; The model training module is used to construct a training set and a test set based on a prior knowledge base, and then train a hybrid algorithm model based on the training set and the test set. The metadata matching module is used to obtain the metadata to be matched and to query whether there is a matching result in the prior knowledge base. If so, return the matching result directly; If not, then a candidate set of the metadata to be matched is obtained from the prior knowledge base through a coarse-ranking model, and the matching result of the metadata to be matched is obtained from the candidate set through the hybrid algorithm model.

Images