A method, apparatus, electronic device, and storage medium for matching entity information.
By constructing an entity name dataset and a semantic feature matching method, the problem of difficulty in identifying matching relationships caused by differences in entity names was solved, and the accuracy and efficiency of entity information matching were improved. In particular, the accuracy of matching medical examination items with medical fee items was improved in the consistency review of medical order fees and tests.
Patent Information
- Application Number
- CN202211212739.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-09-30
- Publication Date
- 2026-11-10
- Estimated Expiration
- 2042-09-30
AI Technical Summary
Existing entity information matching methods struggle to accurately identify entity matching relationships when faced with differences in entity names, hierarchical structures, and affiliations, leading to erroneous judgments. This is particularly true in the consistency review of medical order fees and test results, where it is difficult to identify the matching relationship between medical examination items and medical fee items.
An entity name dataset is constructed to record the canonical names of each entity, as well as their hierarchical and attribution relationships. The canonical entity names of target entities are determined through semantic feature extraction and matching. An entity name tree and semantic feature matching method are used to improve the matching accuracy.
It improves the accuracy and efficiency of entity matching, enabling accurate identification of the matching relationship between medical examination items and medical fee items in the consistency review of medical order fees and tests, thereby reducing overcharging and illegal charges.
Smart Images

Figure CN115659963B_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of artificial intelligence technology, and in particular to an entity information matching method, apparatus, electronic device and storage medium. Background Technology
[0002] Entity matching is a common method for verifying entity matching relationships. For example, when verifying whether medical examination items and medical billing items in a medical record correspond to each other, this is usually achieved by checking whether the medical examination items in the medical record can be matched one-to-one with the medical billing items in the medical record.
[0003] In conventional entity information matching schemes, entity names are typically matched character by character to determine if a match is found. However, in many real-world scenarios, recorded entity names may differ, including abbreviations, misspellings, and codes. Furthermore, entities with hierarchical or subordinate relationships may not have identical names. In these situations, relying solely on character-based matching of entity names to determine a match often leads to incorrect results. Summary of the Invention
[0004] Based on the above-mentioned technological status, this application proposes an entity information matching method, apparatus, electronic device, and storage medium, which can improve the efficiency and accuracy of entity matching.
[0005] The first aspect of this application proposes an entity information matching method, including: Get the name of the target entity; The name of the target entity is matched with the entity names in a preset entity name dataset, and a first entity name that matches the name of the target entity is determined from the entity name dataset; the entity name dataset includes the canonical names of each entity, as well as the hierarchical and belonging relationship information between each entity; The first entity name is determined as the canonical entity name of the target entity.
[0006] A second aspect of this application provides an entity information matching device, comprising: The information acquisition unit is used to acquire the name of the target entity; An entity matching unit is used to match the name of the target entity with the entity names in a preset entity name dataset, and to determine the first entity name that matches the name of the target entity from the entity name dataset; the entity name dataset includes the canonical names of each entity, as well as the hierarchical and attribution information between each entity; The information determination unit is used to determine the first entity name as the canonical entity name of the target entity.
[0007] A third aspect of this application discloses an electronic device, comprising: Memory and processor; The memory is connected to the processor and is used to store programs; The processor is used to implement the above-described entity information matching method by running the program in the memory.
[0008] The fourth aspect of this application proposes a storage medium storing a computer program, which, when executed by a processor, implements the aforementioned entity information matching method.
[0009] The technical solution proposed in this application constructs an entity name dataset, which summarizes and records the canonical names of each entity, as well as the hierarchy and attribution relationships between entities. Based on this, for the obtained name of a target entity, the name of the target entity can be matched with the entity names in the aforementioned entity name dataset to determine the first entity name that matches the name of the target entity, which is then used as the canonical entity name of the target entity.
[0010] The above scheme determines the canonical entity name of the target entity by matching the obtained target entity name with the canonical entity names in the entity name dataset. This process converts the name of any obtained entity into a canonical entity name, facilitating the matching and comparison of different entity names to determine whether different entities are matching or identical.
[0011] Applying the above-mentioned technical solution of this application to the entity matching process can improve the accuracy and efficiency of entity matching. Attached Figure Description
[0012] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of this application. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.
[0013] Figure 1 This is a flowchart illustrating an entity information matching method provided in an embodiment of this application.
[0014] Figure 2 This is a schematic diagram of an entity name tree provided in an embodiment of this application.
[0015] Figure 3 This is a schematic diagram of the entity name tree for medical testing entities provided in the embodiments of this application.
[0016] Figure 4 This is a schematic diagram of a tree-structured long short-term memory network provided in an embodiment of this application.
[0017] Figure 5 This is a schematic diagram of the structure of a classification model provided in an embodiment of this application.
[0018] Figure 6 This is a schematic diagram illustrating the processing steps of an entity information matching method provided in an embodiment of this application.
[0019] Figure 7 This is a schematic diagram of the structure of an entity information matching device provided in an embodiment of this application.
[0020] Figure 8 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation
[0021] The technical solutions of this application are applicable to entity information matching scenarios. Using the technical solutions of this application can improve the accuracy and efficiency of entity matching.
[0022] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0023] Application Overview Entity matching is a common method for verifying entity matching relationships. For example, when verifying whether medical examination items and medical billing items in a medical record correspond to each other, this is usually achieved by checking whether the medical examination items in the medical record can be matched one-to-one with the medical billing items in the medical record.
[0024] In conventional entity information matching schemes, entity names are typically matched character by character to determine if a match is found. However, in many real-world scenarios, recorded entity names may differ, including abbreviations, misspellings, and codes. Furthermore, entities with hierarchical or subordinate relationships may not have identical names. In these situations, relying solely on character-based matching of entity names to determine a match often leads to incorrect results.
[0025] For example, in medical insurance review processes, "consistency review of medical order charges and laboratory tests" is a key aspect of medical insurance violation penalties. A medical order refers to a doctor's instructions to a patient regarding laboratory tests based on the patient's condition and treatment needs, which includes a record of the medical examination items. However, during medical insurance settlement, the charges for these tests are determined based on a billing statement, which records the medical examination items and their amounts. Normally, the medical examination items in the medical order should match those in the billing statement. The consistency review of medical order charges and laboratory tests primarily verifies whether the medical examination items in the billing statement match those in the medical order; otherwise, issues such as overcharging or illegal charges may occur.
[0026] Medical orders are doctors' instructions, which tend to be more conversational and vary between hospitals. Fee lists also differ by region, making it extremely difficult to verify the consistency of medical order fees and testing. The main challenges are as follows: ① Inconsistent wording of the same test name. The billing statement and the doctor's order describe the same test in different ways, such as "thyroid function test" and "thyroid function test". These two express the same meaning, but the wording is different. If the consistency of the test name in the doctor's order is checked by matching the entity name, the matching relationship between "thyroid function test" and "thyroid function test" cannot be accurately identified.
[0027] ② The writing styles for included tests vary greatly. The billing list may charge for a specific test item (sub-item), while the medical order may have a major category name. For example, the doctor may order "thyroid function" in the medical order, but the billing list may have a charge for "thyroid-stimulating hormone test". In this case, the billing should be consistent, but due to the large differences in writing styles, the matching relationship between the two is usually not accurately identified in actual review.
[0028] In summary, existing entity name-based entity matching methods cannot achieve scientifically accurate matching results. Especially in the aforementioned business scenarios such as consistency verification of medical order billing and testing, simply using the names of medical examination item entities and medical billing item entities cannot accurately determine the matching relationship between the two entities.
[0029] Based on the above-mentioned technical problems, this application proposes an entity information matching method. This method can match entity names to standardized entity names, thereby solving the problem of inaccurate identification of entity matching relationships caused by differences in entity names and improving the accuracy of entity information matching.
[0030] Exemplary methods This application proposes an entity information matching method. This method can be applied, exemplarily, to hardware devices such as processors, electronic devices, and servers (including cloud servers), or packaged as a software program and executed. When the hardware device executes the processing procedure of the technical solution of this application, or when the aforementioned software program is run, the entity information matching method proposed in this application can be implemented. This application only provides exemplary descriptions of the specific processing procedure of the technical solution of this application and does not limit the specific implementation form of the technical solution of this application. Any technical implementation form that can execute the processing procedure of the technical solution of this application can be adopted by this application.
[0031] See Figure 1 As shown, the method includes: S101. Obtain the name of the target entity.
[0032] The target entity mentioned above can be any entity in any field or of any type, such as an organization, quantity, person's name, company, geographical location, product, activity, etc.
[0033] The name of the target entity mentioned above can be an entity word that represents the name of the target entity, such as organization name, quantity value, person name, company name, geographical location name, product name, event name, etc.
[0034] The number of names of the target entities mentioned above can be the name of one target entity, the names of multiple target entities, or multiple different names of one target entity.
[0035] The technical solution of this application embodiment is applicable to the business scenario of medical order fee test consistency review. Therefore, the name of the target entity mentioned above can be one or more of the names of medical examination items in the medical order of the medical report and the names of medical charge items in the charge list of the medical report.
[0036] S102. Match the name of the target entity with the entity names in the preset entity name dataset, and determine the first entity name that matches the name of the target entity from the entity name dataset.
[0037] Specifically, in this embodiment, the standardized names of entities in a specific domain or business scenario, as well as the hierarchical and attribution information between entities, are pre-collected to construct an entity name dataset. This entity name dataset not only records the standardized names of each entity in the specific domain or business scenario, but also records the hierarchical and attribution information between the entities. By querying this entity name dataset, the standardized names of entities can be determined, as well as the hierarchical and attribution relationships between entities.
[0038] The specific field or specific business scenario mentioned above can be determined according to the application field or application scenario of the embodiments of this application. Alternatively, the specific field or specific business scenario mentioned above can be all fields or all business scenarios. When the technical solution of the embodiments of this application is applied to a specific field or business scenario, the entity name dataset corresponding to the specific field or business scenario is selected from the entity name dataset corresponding to all fields or all business scenarios according to the specific field or business scenario applied.
[0039] For example, this application embodiment summarizes and aggregates the standardized names of entities in the medical field and the hierarchical and attribution relationships between entities, ultimately constructing a medical entity name dataset. This medical entity name dataset contains the standardized names of various medical entities, as well as the hierarchical and attribution relationships between these entities.
[0040] For example, in the aforementioned medical entity name dataset, the standardized name of the medical examination item "Thyroid function test" is recorded, as is the standardized name of the medical fee item "Thyroid-stimulating hormone test". Since "Thyroid-stimulating hormone test" is a subclass of "Thyroid function test", the aforementioned medical entity name dataset records both the standardized entity names "Thyroid function test" and "Thyroid-stimulating hormone test", and also records their hierarchical and subordinate relationship, that is, "Thyroid-stimulating hormone test" is a subclass of "Thyroid function test".
[0041] In practice, the specific data format of the aforementioned entity name dataset can be any format, and the hierarchical and attribution information between the various entities can also be represented in any form.
[0042] Based on the construction of the above entity name dataset, after obtaining the name of the target entity, this embodiment of the application matches the obtained name of the target entity with the entity names in the above entity name dataset, and determines the entity name that matches the name of the target entity from the entity name dataset as the first entity name.
[0043] Specifically, any feasible entity name matching method can be used to match the name of the target entity with the entity names in the entity name dataset. For example, character matching and / or semantic matching can be used to match the name of the target entity with the entity names in the entity name dataset.
[0044] By matching the name of the target entity with each entity name in the entity name dataset, one or more entity names that match the name of the target entity can be identified from the entity name dataset. All of these matched entity names can be used as the first entity name.
[0045] S103. Determine the first entity name as the canonical entity name of the target entity.
[0046] Specifically, since the entity names stored in the aforementioned entity name dataset are all canonical entity names of the corresponding entities, when it is determined that the name of the target entity matches the first entity name in the entity name dataset, the first entity name can be identified as the canonical entity name corresponding to the target entity. Therefore, the first entity name can be determined as the canonical entity name of the target entity.
[0047] For example, suppose that the medical order in a certain medical report contains the name of the target entity "thyroid function". After processing by the above scheme of this application, the entity name "thyroid function test" that matches "thyroid function" is found from the medical entity name dataset. Then, the entity name "thyroid function test" is the first entity name corresponding to the above "thyroid function", which is the standardized entity name corresponding to "thyroid function".
[0048] As described above, this application embodiment constructs an entity name dataset to summarize and record the canonical names of each entity, as well as the hierarchy and affiliation relationships between entities. Based on this, for the obtained name of a target entity, the name of the target entity can be matched with the entity names in the aforementioned entity name dataset to determine the first entity name that matches the target entity name, which is then used as the canonical entity name of the target entity.
[0049] The above scheme determines the canonical entity name of the target entity by matching the obtained target entity name with the canonical entity names in the entity name dataset. This process converts the name of any obtained entity into a canonical entity name, facilitating the matching and comparison of different entity names to determine whether different entities are matching or identical.
[0050] Applying the above-mentioned technical solution of this application to the entity matching process can improve the accuracy and efficiency of entity matching.
[0051] As a preferred implementation, the entity name dataset described above is constructed in the form of an entity name tree in this application embodiment.
[0052] For the specific form of this entity name tree, please refer to [link / reference]. Figure 2 As shown. In Figure 2 In the entity name tree shown, each circle represents a tree node, and each tree node is an entity name. Starting from the root node, each level of non-leaf node has a parent node and child nodes. The lines connecting the tree nodes represent the relationships between them, that is, the relationships between entity names. The direction of the connecting lines is from the parent node to the child node. By drawing connecting lines between the tree nodes in the entity name tree according to the hierarchy and affiliation of different entities, the purpose of marking the hierarchy and affiliation of different entities can be achieved.
[0053] visible, Figure 2 The entity name tree shown not only records the names of each entity as tree nodes, but also records the hierarchy and affiliation relationships between different entity names through the connecting lines between the tree nodes.
[0054] For example, when the technical solution of this application is applied to the business scenario of medical order billing and test consistency review, clinical laboratory expertise and clinical laboratory knowledge system are used to summarize medical laboratory entities and sort out the hierarchy and affiliation relationship between different medical laboratory entities to obtain an entity name tree containing medical laboratory entities and the hierarchy and affiliation relationship between medical laboratory entities.
[0055] like Figure 3 As shown, the entity name tree is constructed in a top-down manner as described below: First, first-level labels are determined, such as clinical blood and body fluid tests, clinical chemistry tests, etc.; then, each first-level label is expanded hierarchically. For example, clinical blood and body fluid tests include general blood examinations, bone marrow cytology tests, anemia tests, etc., so these test names are second-level labels, and their parent label is clinical blood and body fluid tests; next, all second-level labels are expanded hierarchically. For example, general blood examinations include blood cell analysis, blood cell morphology examination, erythrocyte sedimentation rate measurement, and these labels are also third-level labels, and their parent label is general blood examinations; in this way, each layer of labels is iteratively expanded downwards, and finally the entity name tree is constructed.
[0056] It should be noted that the entity name dataset, such as the entity name tree, proposed in the embodiments of this application can be dynamically expanded in real time. For example, in practical application scenarios, low-frequency entities may not be fully included in the entity name dataset or entity name tree during construction. In such cases, low-frequency entities can be added to the entity name dataset or entity name tree when they are obtained in the actual application scenario.
[0057] In addition, in an entity name dataset or entity name tree, an entity can correspond to multiple different entity names. For example, an entity can correspond to a standard academic name, as well as one or more common names.
[0058] As a preferred implementation, the embodiments of this application achieve matching of the name of the target entity with the entity names in the entity name set through the semantic features of the entity.
[0059] Specifically, by performing the following steps A1-A2, the name of the target entity is matched with the entity names in the entity name dataset, and the first entity name that matches the name of the target entity is determined from the entity name dataset: A1. Extract the semantic features of the target entity's name, and extract the semantic features of the canonical names of each entity in the entity name dataset based on the hierarchy and attribution relationships between entities.
[0060] For example, semantic features of the target entity's name can be obtained by extracting semantic features from the target entity's name. For instance, the semantic features of the target entity's name can be extracted using a Bidirectional Long Short-Term Memory (BiLSTM) network model.
[0061] Similarly, by extracting semantic features from the canonical names of each entity in the entity name dataset, we can obtain the semantic features of the canonical names of each entity. The semantic feature extraction process described above can employ any feature extraction scheme.
[0062] As a preferred implementation, since the entity names in the entity name dataset have hierarchical and attribution relationships, that is, there are associations between entity names at different levels, these associations can also be mapped to the semantic features of entity names at different levels.
[0063] Therefore, in this embodiment of the application, when extracting the semantic features of the canonical names of each entity in the entity name dataset, the semantic features are extracted based on the hierarchy and attribution relationship between the entities. Specifically, the semantic features of entity names are extracted sequentially from highest to lowest level. When extracting the semantic features of higher-level entity names, the semantic features of lower-level entity names belonging to that higher-level entity name are referenced. This ensures that the extracted semantic features of the higher-level entity names include the semantic features of the lower-level entity names.
[0064] When the entity name dataset is an entity name tree, the semantic features of the entity names of each tree node in the entity name tree can be extracted through the following steps A11-A12: A11. Extract the semantic features of the canonical names of the first-level tree nodes.
[0065] A12. Based on the semantic features of the canonical names of the first-level tree nodes, extract the semantic features of the canonical names of the second-level tree nodes.
[0066] In this context, the first-level tree node mentioned above is a child node of the second-level tree node mentioned above.
[0067] That is, for any two adjacent tree nodes with a hierarchical relationship in the entity name tree described above, first extract the semantic features of the canonical name of the first-level tree node, and then extract the semantic features of the canonical name of the second-level tree node based on the semantic features of the canonical name of the first-level tree node. By repeatedly executing the above processes A11 and A12, the semantic features of each entity name in the entity name tree can be extracted. When extracting the semantic features of each level of tree node, the Long Short-Term Memory (LSTM) network can be used to encode the entity name of the tree node to obtain the semantic features of the tree node.
[0068] In a preferred embodiment, when extracting the semantic features of the entity names of each tree node in the entity name tree, the entity name tree can be input into a pre-trained tree-structured long short-term memory network (Tree-LSTM). The Tree-LSTM then extracts the semantic features of the canonical names of the first-level tree nodes and, based on the semantic features of the canonical names of the first-level tree nodes, extracts the semantic features of the canonical names of the second-level tree nodes, thereby obtaining the semantic features of the canonical names of each node in the entity name tree.
[0069] See Figure 4 As shown, similar to a standard LSTM, each Tree-LSTM cell ( Figure 4 The box shown in the diagram will have an input vector. Each Tree-LSTM unit includes similar input gates, output gates, cell state, and hidden layer output. The difference is that the updates to the gate vectors and cell state in a Tree-LSTM unit depend on the states of all its associated child nodes. Furthermore, compared to the single forget gate of an LSTM, a Tree-LSTM has multiple forget gates, each corresponding to a child node k in the current unit. This allows the Tree-LSTM to selectively retrieve information from its child nodes.
[0070] This application uses Tree-LSTM to obtain the semantic representation of the medical entity name tree, and it can take into account the hierarchical relationship between nodes. Specifically, the vector representation of each tree node of the medical entity name tree is used as input. The output of each node corresponds to the hidden state (i.e., semantic features) of that node. , i=1,2,…,N, where N is the number of tree nodes in the entity name tree.
[0071] By extracting the semantic features of each entity name in the entity name tree through the above embodiments of this application, the hierarchical relationship of semantic features can be comprehensively considered, thereby extracting more accurate and more informative semantic features of entity names.
[0072] A2. Based on the semantic features of the target entity's name and the semantic features of the canonical names of each entity in the entity name dataset, determine the first entity name that matches the target entity's name from the entity name dataset.
[0073] After obtaining the semantic features of the target entity's name and the semantic features of the canonical names of each entity in the entity name dataset, the entity name with the highest similarity to the target entity's name or with a similarity greater than a set similarity threshold can be determined from the entity name dataset by calculating the similarity between the semantic features of the target entity's name and the semantic features of the canonical names of each entity in the entity name dataset. This entity name is then used as the first entity name that matches the target entity's name.
[0074] As a more preferred implementation, this application treats the matching process between the name of the target entity and the canonical names of each entity in the entity name dataset as a multi-classification process. That is, the canonical names of each entity in the entity name dataset are used as classification labels, and the name of the target entity is classified within each classification label. The classification result is the canonical entity name that matches the name of the target entity.
[0075] Specifically, a classification model can be pre-trained to classify entity names, with the classification labels set to the canonical names of each entity in the entity name dataset. During model training, the semantic features of the entity names are used as input. The model classifies the input entity names by calculating the similarity between the semantic features of the input entity names and the semantic features of the canonical names of each entity in the entity name dataset, thus obtaining the canonical name classification result corresponding to the input entity names.
[0076] For example, the embodiments of this application achieve the classification processing of the names of target entities through the following steps B1-B2: B1. Perform feature fusion processing on the semantic features of the name of the target entity and the semantic features of the canonical names of each entity in the entity name dataset to obtain fused features.
[0077] Specifically, the semantic features of the target entity's name are fused with the semantic features of the standard names of each entity in the entity name dataset. For example, the semantic features of the target entity's name are concatenated or superimposed with the semantic features of the standard names of each entity in the entity name dataset. This integrates the semantic features of the standard names of each entity into the semantic features of the target entity's name, which will help to classify the target entity's name more accurately within the standard names of each entity.
[0078] As an optional implementation, this embodiment of the application achieves the fusion of the semantic features of the target entity's name with the semantic features of the canonical names of each entity in the entity name dataset through the following steps B11-B13: B11. Determine the semantic features of the target entity's name and their relevance to the semantic features of the canonical names of each entity in the entity name dataset, and obtain a relevance sequence.
[0079] The dimension of the relevance sequence is the same as the number of canonical names in the entity name dataset.
[0080] Specifically, assuming the semantic features of the target entity's name extracted according to the above processing are A, A is compared with the semantic features Y of the canonical names of each entity in the entity name dataset: { (i=1,2,…,N, where N is the number of each entity in the entity name dataset) Attention is calculated to obtain A in each semantic feature Attention distribution on (i=1,2,…,N, where N is the number of each entity in the entity name dataset), attention distribution This means that for A, its relation to The level of attention, that is, with The relevance. Therefore, attention distribution. It is a relevance sequence consisting of the relevance between the semantic features of the target entity's name and the semantic features of the canonical names of each entity in the entity name dataset.
[0081] B12. Based on the relevance sequence, the semantic features of the canonical names of each entity in the entity name dataset are weighted and fused to obtain reference features.
[0082] Specifically, the relevance of each entity's specification name in the above relevance sequence is used as the weight of the corresponding entity specification name, and the semantic features of the specification names of each entity in the entity name dataset are weighted and summed to obtain the reference feature att.
[0083] B13. The reference features are fused with the semantic features of the name of the target entity to obtain the fused features.
[0084] Specifically, the semantic feature A of the target entity's name is concatenated with the aforementioned reference feature att to obtain the fused feature.
[0085] B2. Based on the fusion features, classify the name of the target entity as the first entity name in the entity name dataset.
[0086] Specifically, by performing softmax classification on the aforementioned fusion features for each entity name in the corresponding entity name dataset, the probability of classifying the target entity's name into any of the entity names in the dataset can be obtained. The entity name with the highest classification probability is then selected as the first entity name that matches the target entity.
[0087] As one implementation, when representing the entity name dataset using an entity name tree, see [link to relevant documentation]. Figure 5 As shown, this embodiment of the application employs a classification model based on a BiLSTM+TreeLSTM+Attention architecture to match the name of the target entity with the entity names in the entity name tree. Figure 5 As shown, the name of the target entity is encoded using BiLSTM+SelfAttention to obtain its representation; TreeLSTM is used to encode the entity names in the entity name tree, where each node in the tree represents an entity. This encoding method can implicitly imply the relationship between parent and child classes at different levels; finally, the Attention layer is used to fuse information between the target entity and the entities in the entity name tree to complete the target entity classification prediction.
[0088] Specifically, the input to the classification model is the name of the target entity and an entity name tree, where the name of the target entity is e:{ , ... Each node in the entity name tree is t. { , , ... }, It refers to the words in the name of the target entity. This refers to the entity name of each node in the entity name tree. First, pre-trained word vectors are used to convert the target entity name and the entity names in the entity name tree into vector representations. Then, these vector representations are fed into the following modules to extract semantic representations: ① Extract the semantic representation of the name of the target entity See Figure 5As shown, a bidirectional long short-term memory (BiLSTM) network is used as the semantic representation extraction layer for the target entity's name. The word feature matrix is input into the BiLSTM to calculate the hidden states at multiple time steps. The hidden states of the forward and backward LSTMs at each time step are taken and concatenated to obtain the feature matrix extracted by the bidirectional LSTM network. The output serves as the input to the next layer of the network. Next, Self-Attention is used to extract higher-level semantic representations. Each hidden state performs attention calculations with all states within the entity to learn the dependencies within the entity, capture its internal structure, and finally obtain the attention vector A for the name of the target entity.
[0089] ② Extract the semantic representation of entity names from the entity name tree. BiLSTM can process sequential information (a sequentially linked list), but it cannot process data with a tree structure. Therefore, BiLSTM cannot extract semantic representations from entity name trees. In order to encode each node of the entity name tree, this embodiment uses Tree-Structured LSTM to extract semantic features from each node of the entity name tree.
[0090] Similar to a standard LSTM, each Tree-LSTM unit has an input vector, and each cell includes a similar input gate, output gate, cell state, and hidden layer output. The difference is that the updates to the gate vectors and cell state in a Tree-LSTM unit depend on the states of all its associated child nodes. Furthermore, unlike the single forget gate of an LSTM, a Tree-LSTM has multiple forget gates, each corresponding to a child node k in the current unit. This allows the Tree-LSTM to selectively retrieve information from child nodes. This embodiment uses a Tree-LSTM to obtain the semantic representation of an entity name tree, taking into account the hierarchical relationships between nodes. The vector representation of each tree node is used as the input to the Tree-LSTM. The output of each node corresponds to the hidden state of that node. That is, the semantic representation of the entity name of the node.
[0091] ③ Information fusion and classification of the target entity's name and the entity names in the entity name tree To fully integrate the entity name tag dependency information of the entity name tree into the model, this embodiment of the application associates vector A with the hidden state Y of each entity name in the entity name tree:{ Perform attention calculations to obtain the attention distribution. Attention distribution This means that, with respect to vector A, for The degree of vector attention, in obtaining the attention distribution Subsequently, in this embodiment of the application, a soft attention selection mechanism is used to summarize the vectors, and finally the fused feature vector att is obtained.
[0092] in, The hyperparameter represents the relationship between vector A and vector B. The relevance weights are determined as model parameters during model training; N represents the total number of entity names in the entity name tree.
[0093] Finally, the Attention vector A and att of the target entity's name are concatenated. The concatenated vector [A;att] is then passed through a fully connected layer to achieve multi-class classification, obtaining the probability of classifying the Attention vector A of the target entity's name as each entity name in the entity name tree. The entity name corresponding to the highest probability is the entity name that matches the target entity's name, thus obtaining the first entity name that matches the target entity's name.
[0094] Figure 5 The classification model shown transforms the matching process between the target entity's name and the entity names in the entity name tree into a classification process for the target entity's name, treating each entity name in the entity name tree as a classification label. In the context of... Figure 5 The model shown can be trained using the cross-entropy loss function.
[0095] The technical solutions described in the above embodiments of this application all determine the first entity name that matches the name of the target entity from the entity name dataset based on semantic features, thereby determining the canonical entity name of the target entity. In another embodiment, character matching can also be used first to search for a canonical entity name that matches the name of the target entity from the entity name dataset. If no match is found, the semantic feature-based matching method described in the above embodiments can then be further adopted.
[0096] See Figure 6As shown, for the name of the target entity, the rule matching module first performs character matching between the name of the target entity and the entity names in the entity name dataset. If the rule matching module can match a standard entity name corresponding to the name of the target entity from the entity name dataset, the process ends. If the rule matching module fails to match a standard entity name corresponding to the name of the target entity from the entity name dataset, the semantic matching module performs semantic matching between the name of the target entity and the entity names in the entity name dataset by executing the matching scheme introduced in the above embodiments. If the semantic matching module can match a standard entity name corresponding to the name of the target entity from the entity name dataset, the process ends. If the semantic matching module fails to match a standard entity name corresponding to the name of the target entity from the entity name dataset, it indicates that the name of the target entity is a low-frequency entity name, and the standard name of this entity may not be inherently contained in the entity name dataset. At this time, the extension supplement module is utilized to add the name of the target entity to the entity name dataset.
[0097] The specific processing procedure of the above rule matching module performing character matching between the name of the target entity and the entity names in the entity name dataset can be referred to the following steps C1-C3: C1. Construct a regular matching expression corresponding to each entity name in the preset entity name dataset.
[0098] Specifically, for each entity name in the entity name dataset, strings with no discriminatory significance are first removed, for example, characters such as "test", "examination", "measurement", "trial", "screening" and "detection" in medical examination item entities are removed. Then the remaining strings are used to construct a regular matching expression, for example, the regular matching expression for "cross-matching test" is "intersect. cross. match. blood".
[0099] When performing the character matching process for the first time, it is necessary to construct a regular matching expression corresponding to each entity name in the entity name dataset. After the regular matching expression corresponding to each entity name in the entity name dataset is determined, in the subsequent application process, step C1 can be skipped, and subsequent steps C2 and C3 can be directly executed to realize the character matching processing between the name of the target entity and the entity names in the entity name dataset.
[0100] C2. Match the regular matching expression corresponding to each entity name in the entity name dataset with the name of the target entity.
[0101] C3. If the regular expression corresponding to any entity name in the entity name dataset successfully matches the name of the target entity, the entity name corresponding to the regular expression that successfully matches the name of the target entity is determined as the first entity name that matches the name of the target entity.
[0102] Specifically, the name of the target entity is matched against the regular expression of each canonical entity name in the entity name dataset. If a match is found, the canonical entity name is determined as the first entity name that matches the name of the target entity.
[0103] Compared to semantic matching, character matching is faster. Therefore, we can first use character matching to match the name of the target entity with the entity names in the entity name dataset. If character matching fails to find an entity name that matches the name of the target entity in the entity name dataset, then semantic matching can be used. This can significantly improve the efficiency of finding the first entity name that matches the name of the target entity in the entity name dataset.
[0104] As another example, when the technical solution of this application is applied to the consistency review of medical order billing and testing, each item in the set of medical test items in the medical orders of the same medical record and each item in the set of medical charge items in the billing list can be used as the aforementioned target entity. The aforementioned entity name dataset is an entity name tree that summarizes medical test entities and outlines the hierarchy and affiliation relationships between different medical test entities based on clinical laboratory expertise and the clinical laboratory knowledge system.
[0105] By performing the processing described in the above embodiments of this application, the standardized entity name corresponding to each medical testing item and the standardized entity name corresponding to each medical billing item can be determined from the entity name dataset.
[0106] Based on this, the standardized entity names of each medical testing item can be matched with the standardized entity names of each medical billing item using the entity name dataset to detect whether there are any illegal medical billing items in the medical records.
[0107] The illegal medical charges are those for which there are no matching medical testing items.
[0108] Specifically, for each medical charge item in the billing list of the medical record to be reviewed, a matching medical test item is found in the medical test items included in the doctor's orders of the medical record. If no matching medical test item is found for a certain medical charge item, it means that the charge item is an illegal charge item, that is, an overcharge item.
[0109] In a preferred embodiment, after determining the standardized names of medical test items in the medical records to be reviewed and the standardized names of medical charge items, this application embodiment determines, based on the entity name dataset, the first associated standardized name and the second associated standardized name corresponding to the standardized name of each medical test item, and the standardized names of each medical test item and the first associated standardized name and the second associated standardized name corresponding to each medical test item form a set of medical test item names for each medical test item.
[0110] Wherein, the first associated specification name is the specification entity name of the medical testing item at the next higher level, and the second associated specification name is the specification entity name of the medical testing item at the next lower level.
[0111] Specifically, for each standardized name of a medical testing item, based on the hierarchical and attribution relationship information between entity names recorded in the entity name dataset, the parent entity name and child entity name of the standardized name of the medical testing item are determined from the entity name dataset and used as the first associated standardized name and the second associated standardized name of the medical testing item, respectively.
[0112] Then, the standardized name of each medical test item in the medical record to be reviewed, along with its corresponding first and second associated standardized names, are summarized to obtain the set of medical test item names for that medical test item. It can be seen that this set of medical test item names not only includes the standardized names of the medical test items in the medical record, but also the standardized names of the parent and child items of that medical test item.
[0113] Based on this, the standardized entity names of each medical charge item in the medical record to be reviewed are matched with the set of medical test item names for each medical test item in the medical record. For any medical charge item A, if the standardized entity name of medical charge item A matches any entity name in the set of medical test item names for any medical test item B, then medical charge item A is considered to match medical test item B.
[0114] In the aforementioned one-to-one matching process, if any medical charge item C in the standardized entity name of each medical charge item in the medical record to be reviewed fails to match any medical test item name in the set of medical test item names, it can be determined that there is no medical test item in the medical record that matches medical charge item C. Therefore, it can be determined that there is an illegal medical charge item in the medical record, namely medical charge item C.
[0115] The above-mentioned scheme, which matches the standardized entity names of various medical test items with the standardized entity names of various medical billing items to detect whether there are any illegal medical billing items in the medical records, achieves the matching of medical test items and medical billing items in the medical records in terms of name and quantity, which can improve the accuracy of the consistency detection of medical order billing and testing.
[0116] Exemplary device Accordingly, embodiments of this application also provide an entity information matching device, see [link to relevant documentation]. Figure 7 As shown, the device includes: Information acquisition unit 100 is used to acquire the name of the target entity; The entity matching unit 110 is used to match the name of the target entity with the entity names in a preset entity name dataset, and determine the first entity name that matches the name of the target entity from the entity name dataset; the entity name dataset includes the canonical names of each entity, as well as the hierarchical and attribution information between each entity; The information determination unit 120 is used to determine the first entity name as the canonical entity name of the target entity.
[0117] As an optional implementation, the entity name dataset includes an entity name tree, which is constructed using the canonical names of entities as tree nodes and according to the hierarchy and affiliation relationships between entities.
[0118] As an optional implementation, the name of the target entity is matched with entity names in a preset entity name dataset, and a first entity name matching the name of the target entity is determined from the entity name dataset, including: Extract the semantic features of the target entity's name, and extract the semantic features of the canonical names of each entity in the entity name dataset based on the hierarchy and attribution relationships between the entities; Based on the semantic features of the target entity's name and the semantic features of the canonical names of each entity in the entity name dataset, a first entity name that matches the target entity's name is determined from the entity name dataset.
[0119] As an optional implementation, when the entity name dataset includes an entity name tree, the step of extracting semantic features of the canonical names of each entity in the entity name dataset based on the hierarchy and attribution relationships between entities includes: Extract the semantic features of the canonical names of the first-level tree nodes; Based on the semantic features of the canonical names of the first-level tree nodes, extract the semantic features of the canonical names of the second-level tree nodes; In this context, the first-level tree node is a child node of the second-level tree node.
[0120] As an optional implementation, the semantic features of the canonical names of the first-level tree nodes are extracted; based on the semantic features of the canonical names of the first-level tree nodes, the semantic features of the canonical names of the second-level tree nodes are extracted, including: The entity name tree is input into a pre-trained tree-structured long short-term memory network, so that the tree-structured long short-term memory network extracts the semantic features of the canonical names of the first-level tree nodes and extracts the semantic features of the canonical names of the second-level tree nodes based on the semantic features of the canonical names of the first-level tree nodes, thereby obtaining the semantic features of the canonical names of each node in the entity name tree.
[0121] As an optional implementation, based on the semantic features of the target entity's name and the semantic features of the canonical names of each entity in the entity name dataset, a first entity name matching the target entity's name is determined from the entity name dataset, including: The semantic features of the target entity's name are fused with the semantic features of the canonical names of each entity in the entity name dataset to obtain fused features. Based on the fusion features, the name of the target entity is classified as the first entity name in the entity name dataset.
[0122] As an optional implementation, the semantic features of the target entity's name are fused with the semantic features of the canonical names of each entity in the entity name dataset to obtain fused features, including: The semantic features of the target entity's name are determined, and their relevance to the semantic features of the canonical names of each entity in the entity name dataset is calculated to obtain a relevance sequence; the dimension of the relevance sequence is the same as the number of canonical names in the entity name dataset. Based on the relevance sequence, the semantic features of the canonical names of each entity in the entity name dataset are weighted and fused to obtain reference features; The reference features are fused with the semantic features of the name of the target entity to obtain the fused features.
[0123] As an optional implementation, before extracting the semantic features of the target entity's name and the semantic features of the canonical names of each entity in the entity name dataset based on the hierarchy and attribution relationships between entities, the entity matching unit 110 is further configured to: The name of the target entity is matched with the regular expression corresponding to each entity name in the entity name dataset. If the regular expression corresponding to any entity name in the entity name dataset successfully matches the name of the target entity, the entity name corresponding to the regular expression that successfully matches the name of the target entity is determined as the first entity name that matches the name of the target entity.
[0124] As an optional implementation, the target entity includes a set of medical test items and a set of medical billing items in the same medical record; The device further includes: The screening and processing unit is used to match the standardized entity names of each medical test item with the standardized entity names of each medical billing item based on the entity name dataset, and to detect whether there are any illegal medical billing items in the medical record; wherein, the illegal medical billing items are medical billing items for which there is no matching medical test item.
[0125] As an optional implementation, the standardized entity names of each medical testing item are matched with the standardized entity names of each medical billing item to detect whether there are any illegal medical billing items in the medical record, including: For each standardized name of a medical testing item, based on the entity name dataset, a first associated standardized name and a second associated standardized name corresponding to the standardized name of that medical testing item are determined. The standardized names of each medical testing item and the first and second associated standardized names corresponding to each medical testing item are combined to form a set of medical testing item names for each medical testing item. Among them, the first associated standardized name is the standardized entity name of the medical testing item at the next higher level, and the second associated standardized name is the standardized entity name of the medical testing item at the next lower level. The standardized entity names of each medical charge item are matched one-to-one with the sets of names of each medical test item. If, among the standardized entity names of all medical charge items, there is a case where the standardized entity name of any medical charge item fails to match any medical test item name in the set of medical test item names, then it is determined that there is an illegal medical charge item in the medical record.
[0126] The entity information matching device provided in this embodiment belongs to the same application concept as the entity information matching method provided in the above embodiments of this application. It can execute the entity information matching method provided in any of the above embodiments of this application and has the corresponding functional modules and beneficial effects of executing the entity information matching method. Technical details not described in detail in this embodiment can be found in the specific processing content of the entity information matching method provided in the above embodiments of this application, and will not be repeated here.
[0127] Exemplary electronic devices Another embodiment of this application also provides an electronic device, see [link to relevant documentation] Figure 8 As shown, the device includes: Memory 200 and processor 210; The memory 200 is connected to the processor 210 and is used to store programs; The processor 210 is configured to implement the entity information matching method disclosed in any of the above embodiments by running the program stored in the memory 200.
[0128] Specifically, the aforementioned electronic device may also include: a bus, a communication interface 220, an input device 230, and an output device 240.
[0129] The processor 210, memory 200, communication interface 220, input device 230, and output device 240 are interconnected via a bus. Among them: A bus can include a pathway for transmitting information between various components of a computer system.
[0130] Processor 210 can be a general-purpose processor, such as a general-purpose central processing unit (CPU), a microprocessor, etc., or an application-specific integrated circuit (ASIC), or one or more integrated circuits used to control the execution of the program of the present invention. It can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), an off-the-shelf programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.
[0131] Processor 210 may include a main processor, as well as a baseband chip, modem, etc.
[0132] The memory 200 stores a program that executes the technical solution of this invention, and may also store an operating system and other key business functions. Specifically, the program may include program code, which includes computer operation instructions. More specifically, the memory 200 may include read-only memory (ROM), other types of static storage devices capable of storing static information and instructions, random access memory (RAM), other types of dynamic storage devices capable of storing information and instructions, disk storage, flash memory, etc.
[0133] Input device 230 may include a device for receiving user input data and information, such as a keyboard, mouse, camera, scanner, light pen, voice input device, touch screen, pedometer, or gravity sensor.
[0134] Output device 240 may include devices that allow information to be output to a user, such as a display screen, printer, speaker, etc.
[0135] The communication interface 220 may include a device that uses any transceiver to communicate with other devices or communication networks, such as Ethernet, Radio Access Network (RAN), Wireless Local Area Network (WLAN), etc.
[0136] The processor 210 executes the program stored in the memory 200 and calls other devices, which can be used to implement each step of any entity information matching method provided in the above embodiments of this application.
[0137] Exemplary computer program products and storage media In addition to the methods and devices described above, embodiments of this application may also be computer program products, which include computer program instructions that, when executed by a processor, cause the processor to perform the steps in the entity information matching method described in the "Exemplary Methods" section of this specification.
[0138] The computer program product can be written in any combination of one or more programming languages to perform the operations of the embodiments of this application. The programming languages include object-oriented programming languages such as Java and C++, as well as conventional procedural programming languages such as C or similar languages. The program code can be executed entirely on the user's computing device, partially on the user's computing device, as a standalone software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server.
[0139] Furthermore, embodiments of this application may also be storage media storing a computer program, which is executed by a processor in the entity information matching method described in the "Exemplary Methods" section above.
[0140] For the foregoing method embodiments, in order to simplify the description, they are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, because according to this application, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily essential to this application.
[0141] It should be noted that the various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For apparatus embodiments, since they are basically similar to method embodiments, the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments.
[0142] The steps in the methods of the various embodiments of this application can be adjusted, merged, or deleted in order according to actual needs, and the technical features described in each embodiment can be replaced or combined.
[0143] The modules and sub-modules in the various embodiments of the present application's devices and terminals can be merged, divided, and deleted according to actual needs.
[0144] It should be understood that the disclosed terminals, devices, and methods can be implemented in other ways, given the several embodiments provided in this application. For example, the terminal embodiments described above are merely illustrative. For instance, the division of modules or sub-modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple sub-modules or modules may be combined or integrated into another module, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices, or modules, and may be electrical, mechanical, or other forms.
[0145] The modules or submodules described as separate components may or may not be physically separate. The components that constitute a module or submodule may or may not be physical modules or submodules; that is, they may be located in one place or distributed across multiple network modules or submodules. Some or all of the modules or submodules can be selected to achieve the purpose of this embodiment's solution, depending on actual needs.
[0146] Furthermore, the functional modules or sub-modules in the various embodiments of this application can be integrated into one processing module, or each module or sub-module can exist physically separately, or two or more modules or sub-modules can be integrated into one module. The integrated modules or sub-modules described above can be implemented in hardware or in the form of software functional modules or sub-modules.
[0147] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.
[0148] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly by hardware, a software unit executed by a processor, or a combination of both. The software unit can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.
[0149] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.
[0150] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. A method for matching entity information, characterized in that, include: Get the name of the target entity; Semantic features of the target entity's name are extracted, and semantic features of the canonical names of each entity in the entity name dataset are extracted based on the hierarchy and attribution relationships between the entities. The entity name dataset includes the canonical names of each entity and information on the hierarchy and attribution relationships between the entities. Based on the semantic features of the target entity's name and the semantic features of the canonical names of each entity in the entity name dataset, the maximum probability of the target entity's name being classified as each entity name in the entity name dataset is determined, and the entity name corresponding to the maximum probability is determined as the first entity name matching the target entity. The first entity name is determined as the canonical entity name of the target entity.
2. The method according to claim 1, characterized in that, The entity name dataset includes an entity name tree, which is constructed using the canonical names of entities as tree nodes and according to the hierarchy and affiliation relationships between entities.
3. The method according to claim 1, characterized in that, When the entity name dataset includes an entity name tree, the step of extracting semantic features of the canonical names of each entity in the entity name dataset based on the hierarchy and attribution relationships between entities includes: Extract the semantic features of the canonical names of the first-level tree nodes; Based on the semantic features of the canonical names of the first-level tree nodes, extract the semantic features of the canonical names of the second-level tree nodes; In this context, the first-level tree node is a child node of the second-level tree node.
4. The method according to claim 3, characterized in that, Extract the semantic features of the canonical names of the first-level tree nodes; Based on the semantic features of the canonical names of the first-level tree nodes, extract the semantic features of the canonical names of the second-level tree nodes, including: The entity name tree is input into a pre-trained tree-structured long short-term memory network, so that the tree-structured long short-term memory network extracts the semantic features of the canonical names of the first-level tree nodes and extracts the semantic features of the canonical names of the second-level tree nodes based on the semantic features of the canonical names of the first-level tree nodes, thereby obtaining the semantic features of the canonical names of each node in the entity name tree.
5. The method according to claim 1, characterized in that, Based on the semantic features of the target entity's name and the semantic features of the canonical names of each entity in the entity name dataset, a first entity name matching the target entity's name is determined from the entity name dataset, including: The semantic features of the target entity's name are fused with the semantic features of the canonical names of each entity in the entity name dataset to obtain fused features. Based on the fusion features, the name of the target entity is classified as the first entity name in the entity name dataset.
6. The method according to claim 5, characterized in that, The semantic features of the target entity's name are fused with the semantic features of the canonical names of each entity in the entity name dataset to obtain fused features, including: The semantic features of the target entity's name are determined, and their relevance to the semantic features of the canonical names of each entity in the entity name dataset is calculated to obtain a relevance sequence; the dimension of the relevance sequence is the same as the number of canonical names in the entity name dataset. Based on the relevance sequence, the semantic features of the canonical names of each entity in the entity name dataset are weighted and fused to obtain reference features; The reference features are fused with the semantic features of the name of the target entity to obtain the fused features.
7. The method according to claim 1, characterized in that, Before extracting the semantic features of the target entity's name and the semantic features of the canonical names of each entity in the entity name dataset based on the hierarchy and attribution relationships between entities, the method further includes: The name of the target entity is matched with the regular expression corresponding to each entity name in the entity name dataset. If the regular expression corresponding to any entity name in the entity name dataset successfully matches the name of the target entity, the entity name corresponding to the regular expression that successfully matches the name of the target entity is determined as the first entity name that matches the name of the target entity.
8. The method according to any one of claims 1 to 7, characterized in that, The target entity includes a set of medical test items and a set of medical billing items in the same medical record; The method further includes: Based on the entity name dataset, the standardized entity names of each medical test item are matched with the standardized entity names of each medical billing item to detect whether there are any illegal medical billing items in the medical record; wherein, the illegal medical billing items are medical billing items for which there is no matching medical test item.
9. The method according to claim 8, characterized in that, The system matches the standardized entity names of each medical testing item with the standardized entity names of each medical billing item to detect whether there are any illegal medical billing items in the medical records, including: For each standardized name of a medical testing item, based on the entity name dataset, a first associated standardized name and a second associated standardized name corresponding to the standardized name of that medical testing item are determined. The standardized names of each medical testing item and the first and second associated standardized names corresponding to each medical testing item are combined to form a set of medical testing item names for each medical testing item. Among them, the first associated standardized name is the standardized entity name of the medical testing item at the next higher level, and the second associated standardized name is the standardized entity name of the medical testing item at the next lower level. The standardized entity names of each medical charge item are matched one-to-one with the sets of names of each medical test item. If, among the standardized entity names of all medical charge items, there is a case where the standardized entity name of any medical charge item fails to match any medical test item name in the set of medical test item names, then it is determined that there is an illegal medical charge item in the medical record.
10. An entity information matching device, characterized in that, include: The information acquisition unit is used to acquire the name of the target entity; The entity matching unit is used to extract the semantic features of the name of the target entity, and to extract the semantic features of the canonical names of each entity in the entity name dataset based on the hierarchy and attribution relationship between the entities; the entity name dataset includes the canonical names of each entity, as well as the hierarchy and attribution relationship information between the entities. Based on the semantic features of the target entity's name and the semantic features of the canonical names of each entity in the entity name dataset, the maximum probability of the target entity's name being classified as each entity name in the entity name dataset is determined, and the entity name corresponding to the maximum probability is determined as the first entity name matching the target entity. The information determination unit is used to determine the first entity name as the canonical entity name of the target entity.
11. An electronic device, characterized in that, include; Memory and processor; The memory is connected to the processor and is used to store programs; The processor is configured to implement the entity information matching method as described in any one of claims 1 to 9 by running a program in the memory.
12. A storage medium, characterized in that, The storage medium stores a computer program, which, when executed by a processor, implements the entity information matching method as described in any one of claims 1 to 9.
