Model training method, nursing record information intelligent classification method and device
By collecting labeled datasets and training them using the BERT language model and joint loss function, a nursing record classification model was generated, which solved the problem of messy nursing record information, and achieved efficient classification of nursing record information and improved its reference value.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- THE FIRST AFFILIATED HOSPITAL OF ANHUI MEDICAL UNIV
- Filing Date
- 2026-03-16
- Publication Date
- 2026-07-10
AI Technical Summary
The nursing record information generated by existing technologies is messy, difficult to identify, and has low reference value, making it difficult to effectively utilize the nursing record information.
By collecting a dataset of nursing records labeled with classification tags, and using the BERT language model and joint loss function for iterative training, a target nursing record classification model is generated to achieve intelligent classification of nursing records.
This improved the recognizability and reference value of nursing record information, and enhanced the accuracy and readability of nursing record information classification.
Smart Images
Figure CN122365036A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of deep learning technology, and in particular to a model training method, a method and apparatus for intelligent classification of nursing record information. Background Technology
[0002] Currently, hospitals often use speech-to-text models to quickly generate nursing records based on nurses' observations and verbal descriptions of patients (or their medical and medication records), saving time in obtaining the written records. However, these records are often disorganized, with text information crammed into one place, making them difficult to decipher and limiting their reference value. Summary of the Invention
[0003] In view of the above problems, embodiments of this application provide a model training method, a method for intelligent classification of nursing record information, an apparatus, an electronic device, and a readable storage medium to overcome or at least partially solve the above problems.
[0004] In a first aspect, embodiments of this application provide a model training method, the method comprising: Collect a dataset of hospital nursing records, which includes multiple nursing records labeled with classification tags; The nursing record information in the nursing record dataset is classified once to obtain sample nursing entries; The sample nursing entries are input into the first language model to obtain the classification information of the nursing record entries output by the first language model. The classification information indicates the nursing record category to which the nursing record entries belong. Based on classification information and classification labels, the model parameters of the first language model are adjusted through iterative training to obtain the target nursing record classification model. The iterative training adopts a joint loss function, which includes a combination of semantic similarity loss and cross-entropy loss.
[0005] Optionally, the nursing record dataset is classified once to generate training input sequences, including: Determine the position of the target separator in the nursing record information; Based on the separation location, sample nursing entries are identified from nursing record information.
[0006] Optionally, based on the separation location, sample nursing entries are determined from nursing record information, including: Based on the separation location, the nursing record information is divided into multiple initial nursing record entries; Generate a knowledge graph of initial nursing record entries; Based on the shared entities between knowledge graphs, nursing record entries are clustered to obtain sample nursing entries.
[0007] Optionally, a knowledge graph of the initial nursing record entries is generated, including: Extract semantic entities from nursing record entries; A knowledge graph is constructed based on semantic entities and the relationships between them.
[0008] Optionally, the first language model includes a BERT language encoder module, a fully connected network module, and a convolutional neural network module; the fully connected network module connects the BERT language encoder module and the convolutional neural network module; the sample nursing entries are input into the first language model to obtain the classification information of the nursing record entries output by the first language model, including: The BERT language encoder module is used to determine the encoding vector of the sample care item; Using a convolutional neural network module, classification information for sample care items is determined based on encoded vectors.
[0009] Optionally, the specific calculation method for the joint loss function includes: Determine the first weight value for semantic similarity loss and the second weight value for cross-entropy loss respectively; The function value of the joint loss function is determined based on the first weight value, semantic similarity loss, second weight value, and cross-entropy loss.
[0010] Optionally, it also includes: data augmentation of sample care entries, including random mask word replacement or knowledge graph-based entity synonym expansion.
[0011] Secondly, embodiments of this application provide a method for classifying hospital nursing records, comprising the following steps: Obtain the nursing record information of the target patient, which includes multiple nursing record entries; The nursing record information is categorized once to obtain the target nursing record items; Multiple target nursing record entries are input into the target nursing record classification model to obtain the classification information of the target nursing record entries output by the target nursing record classification model. The classification information indicates the nursing record category of the nursing record entry. The target nursing record classification model is obtained based on the model training method described in any of the above items. Nursing record information is categorized and labeled based on classification information.
[0012] Thirdly, embodiments of this application provide a model training apparatus, the apparatus comprising: The data acquisition module is used to collect hospital nursing record datasets, which include multiple nursing record information labeled with classification tags; The first classification module is used to classify the nursing record information in the nursing record dataset to obtain sample nursing entries. The first input / output module is used to input sample nursing items into the first language model and obtain the classification information of nursing record items output by the first language model. The classification information indicates the nursing record category to which the nursing record item belongs. The adjustment module is used to adjust the model parameters of the first language model through iterative training based on classification information and classification labels to obtain the target nursing record classification model. The iterative training adopts a joint loss function, which includes a combination of semantic similarity loss and cross-entropy loss.
[0013] Fourthly, embodiments of this application provide a hospital nursing record classification device, the device comprising: The acquisition module is used to acquire the nursing record information of the target patient, which includes multiple nursing record entries. The second classification module is used to classify nursing record information to obtain target nursing record entries. The third input / output module is used to input multiple target nursing record items into the target nursing record classification model to obtain the classification information of the target nursing record items output by the target nursing record classification model. The classification information indicates the nursing record category of the nursing record item. The target nursing record classification model is obtained based on the model training method described in any of the above items. The third classification module is used to classify and label nursing record information based on classification information.
[0014] Fifthly, embodiments of this application provide an electronic device, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the model training method as described above, or the intelligent classification method for nursing record information.
[0015] Sixthly, embodiments of this application provide a readable storage medium storing a program or instructions, which, when executed by a processor, implement the model training method described above, or the intelligent classification method for nursing record information.
[0016] The specific beneficial effects are as follows: This application embodiment collects a hospital nursing record dataset, which includes multiple nursing record information labeled with classification tags. Each nursing record in the dataset is classified once to obtain sample nursing entries. These sample nursing entries are input into a first language model, which outputs classification information indicating the nursing record category to which the entry belongs. Based on the classification information and classification tags, the model parameters of the first language model are adjusted through iterative training to obtain a target nursing record classification model. The iterative training employs a joint loss function, which includes a combination of semantic similarity loss and cross-entropy loss. Deep learning methods can be used to train a target nursing record classification model suitable for nursing information classification. When using this target nursing record classification model to classify nursing records, classification information corresponding to each nursing record entry can be obtained, which can effectively improve the recognizability and reference value of nursing record information to a certain extent. Attached Figure Description
[0017] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the description of the embodiments of this application will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0018] Figure 1 This is a schematic flowchart of a model training method provided in an embodiment of this application; Figure 2 The loss value decrease curve provided in the embodiments of this application; Figure 3 A flowchart illustrating an intelligent classification method for nursing record information provided in this application embodiment; Figure 4 This is a logic block diagram of a model training device provided in an embodiment of this application; Figure 5 This is a logic block diagram of an intelligent classification device for nursing record information provided in an embodiment of this application; Figure 6 This is a schematic diagram of an electronic device provided in an embodiment of this application. Detailed Implementation
[0019] Exemplary embodiments of this application will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of this application are shown in the drawings, it should be understood that this application may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of this application and to fully convey the scope of this application to those skilled in the art.
[0020] Reference Figure 1 , Figure 1 This is a flowchart illustrating a model training method provided in an embodiment of this application. The method may include: Step S101: Collect a hospital nursing record dataset, which includes multiple nursing record information labeled with classification tags.
[0021] In the embodiments of this application, in actual nursing scenarios, hospitals typically use speech recognition models to quickly generate nursing records. Nurses dictate the nursing records, and the speech recognition model outputs the text of the records. Hospitals also usually compile standardized nursing records for patients after discharge. These two types of nursing records can be categorized and labeled by professionals, allowing the labeled records to be collected into a hospital nursing record dataset. Thus, the hospital nursing record dataset includes both standardized nursing records and records obtained through speech recognition, which contain some noise.
[0022] Step S102: Perform a classification process on each nursing record information in the nursing record dataset to obtain sample nursing entries.
[0023] In the embodiments of this application, each nursing record in the nursing record dataset typically corresponds to different classification attributes. For example, nursing record information may include the patient's daily symptom status, body temperature, medication use, sleep patterns, blood pressure, etc. Fixed delimiters are usually used to separate different types of nursing records in the nursing record information. For example, in Chinese nursing record information, semicolons ";" or periods "." are used to separate different types of nursing records. Therefore, a symbol detection method can be used to classify the nursing record information by recognizing the delimiters, obtaining nursing entries belonging to different categories.
[0024] Step S103: Input the sample nursing item into the first language model to obtain the classification information of the nursing record item output by the first language model. The classification information indicates the nursing record category to which the nursing record item belongs.
[0025] In embodiments of this application, the first language model can be a model based on Bidirectional Encoder Representations from Transformers (BERT) or itself, such as BERT, RoBERTa, ALBERT, etc. Sample nursing entries can be used as input to the first language model. After processing by the first language, classification information of the nursing record entries is obtained. This classification information can indicate the nursing record category to which the nursing record entry belongs, such as "patient," "symptom," "medication," "blood pressure," "blood sugar," "daytime," etc.
[0026] Step S104: Based on the classification information and the classification label, the model parameters of the first language model are adjusted through iterative training to obtain the target nursing record classification model; wherein, the iterative training adopts a joint loss function, which includes a combination of semantic similarity loss and cross-entropy loss.
[0027] In the embodiments of this application, the model parameters of the first language model are adjusted through iterative training using the classification information output by the first language model and the classification labels of the sample nursing entries, ultimately yielding a target nursing record classification model. The number of iterations can be set by the developer, for example, 1000 times (10 epochs, each epoch containing 100 training iterations). During iterative training, the reference data used can be a joint loss function composed of semantic similarity loss and cross-entropy loss. Both semantic similarity loss and cross-entropy loss can be calculated using the classification information and classification labels. The semantic similarity loss value can be calculated using the difference between the cosine similarity and 1 between the encoding vector of the classification information and the encoding vector of the classification label, expressed as: In Equation 1 above, A and B are the encoding vectors for classification information and classification labels, respectively, and the symbol "·" represents the dot product. The calculation method for cross-entropy loss is fairly standard and will not be elaborated here.
[0028] In the embodiments of this application, a hospital nursing record dataset is collected, which includes multiple nursing record information labeled with classification tags. Each nursing record in the dataset is classified once to obtain sample nursing entries. These sample nursing entries are input into a first language model to obtain the classification information of the nursing record entries output by the first language model. The classification information indicates the nursing record category to which the nursing record entry belongs. Based on the classification information and classification tags, the model parameters of the first language model are adjusted through iterative training to obtain a target nursing record classification model. The iterative training uses a joint loss function, which includes a combination of semantic similarity loss and cross-entropy loss. This allows for a classification of the nursing record information before model training, improving the model's learning efficiency and enabling the final target nursing record classification model to have high classification performance. The target nursing record classification model is then used to classify nursing records.
[0029] In some embodiments, step S102 includes: Step S121: Determine the separation position of the target separator in the nursing record information.
[0030] In embodiments of this application, the target separator may include common punctuation marks used in the language of nursing records, such as periods and semicolons. The separation position of the target separator in the nursing record information can be determined based on a pre-set target separator and by performing precise detection on the target separator.
[0031] Step S122: Based on the separation location, determine the sample nursing entries from the nursing record information.
[0032] In the embodiments of this application, the nursing records before and after the segmentation position are usually nursing records of different categories. Therefore, based on the aforementioned segmentation position, multiple sample nursing entries can be segmented from the nursing record information. Each sample nursing entry corresponds to a nursing record category.
[0033] Based on the above embodiments disclosed in this application, nursing record information can be accurately segmented according to a pre-set target delimiter to obtain sample nursing items belonging to different categories. This can improve the information uniformity of sample nursing items to a certain extent and enhance the information accuracy and training reliability of sample nursing items in model training.
[0034] In some embodiments, step S122 includes: Step S1221: Based on the separation location, the nursing record information is divided into multiple initial nursing record entries.
[0035] In the embodiments of this application, segmenting nursing record information based on a target delimiter is a relatively general segmentation method. In some practical scenarios, nursing information before and after the delimiter may belong to the same category. In this scenario, the nursing information after the delimiter is usually used to provide additional interpretation of the nursing information before the delimiter. To make the segmentation of nursing record information more accurate, the nursing record information can first be initially segmented according to the separation position of the target delimiter to obtain multiple initial nursing record entries.
[0036] Step S1222: Generate a knowledge graph of the initial nursing record entries.
[0037] In the embodiments of this application, a knowledge graph is a structured semantic knowledge base used to describe concepts and their relationships in the physical world (or virtual world) in symbolic form. Its basic building blocks are: entity-relation-entity triples (also called SPO triples: subject-verb-object), and attribute-value pairs of entities. In other words, a knowledge graph is a highly simplified semantic representation that embodies the key information of the original language text. In the embodiments of this application, knowledge graphs for each initial nursing record entry can be generated using methods such as instruction-based fine-tuning of a dedicated extraction model, differentiable end-to-end extraction methods, and multimodal knowledge graph extraction methods.
[0038] Step S1223: Based on the shared entities between knowledge graphs, cluster the nursing record entries to obtain sample nursing entries.
[0039] In the embodiments of this application, if different knowledge graphs have the same entity information, it can be considered that these knowledge graphs share the same entity. In this case, the nursing record entries corresponding to these knowledge graphs with the same shared entity can be clustered and integrated into a sample nursing entry.
[0040] Based on the above embodiments disclosed in this application, knowledge graphs can be used to cluster initial nursing record entries to obtain sample nursing entries belonging to the same category. This allows sample nursing entries to possess different information under the same entity, which can improve the information richness of sample nursing entries to a certain extent, further enhance the data richness of sample nursing entries in model training, and improve model training efficiency.
[0041] In some embodiments, step S1222 may include: Step A1: Extract semantic entities from nursing record entries; Step A2: Construct a knowledge graph based on semantic entities and the relationships between them.
[0042] In the embodiments of this application, a semantic entity refers to a node that has a clear semantic meaning, can be "understood" by a computer, and participates in reasoning. It typically represents something with independent meaning in the real world or conceptual world, and usually carries a type (category), attributes, and relationships with other entities. Examples include people, places, organizations, products, concepts, events, and domain-specific entities (e.g., Streptococcus pneumoniae and aspirin in the medical field; various laws, etc.). Based on the definition of a semantic entity, some entity recognition methods can be used to extract semantic entities from nursing record entries. Typically, a semantic entity can represent the category information of the nursing record or a subcategory under that category. For example, blood pressure is a category, blood pressure value is a subcategory, and diastolic pressure is also a subcategory. Then, a knowledge graph can be constructed based on the semantic entities and the relationships between them. Examples include "blood pressure - value - 120" and "patient - sleep time - 8h".
[0043] Based on the embodiments disclosed in this application, semantic entities in nursing record entries can be obtained by extracting semantic entities. Then, based on the relationships between semantic entities, a knowledge graph of nursing record entries can be constructed using semantic entities, which can improve the accuracy and reliability of the knowledge graph to a certain extent.
[0044] In some embodiments, the first language model includes a BERT language encoder module, a fully connected network module, and a convolutional neural network module; the fully connected network module is used to connect the BERT language encoder module and the convolutional neural network module; step S103 may include: Step S131: Use the BERT language encoder module to determine the encoding vector of the sample care item.
[0045] Step S132: Using a convolutional neural network module, the classification information of the sample care items is determined based on the encoded vector.
[0046] In embodiments of this application, the BERT language encoder module can be used to generate encoded vectors for sample care items, which include attention information. Then, utilizing the classification function of a convolutional neural network (CNN), the CNN module processes the encoded vectors to ultimately output classification information for the sample care items. This classification information includes the classification probability for each category.
[0047] Based on the embodiments disclosed in this application, the natural language processing function of the BERT language encoder and the classification function of the convolutional neural network module can be used to realize the classification information of the output sample nursing items, which can improve the classification efficiency of the first language model and improve the accuracy of the classification information to a certain extent.
[0048] In some embodiments, the specific calculation method of the joint loss function includes: Step S21: Determine the first weight value of semantic similarity loss and the second weight value of cross-entropy loss respectively.
[0049] Step S22: Determine the function value of the joint loss function based on the first weight value, semantic similarity loss, second weight value, and cross-entropy loss.
[0050] In the embodiments of this application, the calculation method of the joint loss function is shown in Equation 2 below: In equation 2 above, The value of the joint loss function. As the first weight value, For semantic similarity loss, As the second weight value, This is the cross-entropy loss. For ease of calculation, we can... .
[0051] Based on the embodiments disclosed in this application, the proportions of semantic similarity loss and cross-entropy loss in the joint loss function can be adjusted using a first weight value and a second weight value. This can, to some extent, improve the joint loss function to better meet user expectations and enhance its accuracy and reliability.
[0052] like Figure 2 As shown, Figure 2 The diagram illustrates the decrease in the loss value during model training. In the figure, "Training Loss Over Time" represents the decrease in the "Loss Value" of the joint loss function as the "Training Steps" progress. It can be seen that the joint loss function shows an overall decreasing trend, indicating that the trained target nursing record classification model has good nursing record classification performance.
[0053] In some embodiments, the above method may further include: data augmentation of sample care entries, the data augmentation including random mask word replacement or knowledge graph-based entity synonym expansion.
[0054] In the embodiments of this application, random masked word replacement refers to randomly replacing words in nursing record information with other words. Knowledge graph-based entity synonym expansion refers to replacing entity words in nursing record information with synonyms or near-synonyms. The data-augmented sample augmented nursing entries can be used as input to a first language model for training the first language model.
[0055] Based on the above embodiments disclosed in this application, the sample nursing entries can have higher data richness, thereby further improving the ability of the trained model to process messy nursing record information.
[0056] Reference Figure 3 , Figure 3 This application provides a flowchart illustrating an intelligent classification method for nursing record information, which may include: Step S301: Obtain the nursing record information of the target patient, which includes multiple nursing record entries.
[0057] Step S302: Perform a classification process on the nursing record information to obtain the target nursing record item.
[0058] Step S303: Input multiple target nursing record entries into the target nursing record classification model to obtain the classification information of the target nursing record entries output by the target nursing record classification model. The classification information indicates the nursing record category of the nursing record entry. The target nursing record classification model is obtained based on the model training method described in any of the above items. Step S304: Based on the classification information, classify and label the nursing record information.
[0059] In the embodiments of this application, a target nursing record classification model can be obtained based on the model training method described in any of the preceding claims. After obtaining the target nursing record classification model, the model can be used to classify nursing records. First, the nursing record information of the target patient can be obtained. Then, the nursing record information can be classified once, similar to the classification process during model training, to obtain target nursing record entries. Then, multiple target nursing record entries can be input into the target nursing record classification model to classify multiple target nursing record entries and output the classification information corresponding to each target nursing record entry. Finally, the classification information can be labeled on the nursing information record to achieve classification labeling. The labeling format can be set by the developer, for example, using the symbol "[ ]" to label the classification information.
[0060] Based on the above embodiments disclosed in this application, nursing record information of target patients can be classified, and the nursing record information can be categorized and labeled according to the classification information, thereby improving the readability and referenceability of nursing record information to a certain extent.
[0061] refer to Figure 4 , Figure 4 This application provides a logic block diagram of a model training apparatus 400, which may include: The data acquisition module 401 is used to acquire a hospital nursing record dataset, which includes multiple nursing record information labeled with classification tags. The first classification module 402 is used to perform a classification process on each nursing record information in the nursing record dataset to obtain sample nursing entries. The first input / output module 403 is used to input sample nursing items into the first language model and obtain the classification information of nursing record items output by the first language model. The classification information indicates the nursing record category to which the nursing record item belongs. The adjustment module 404 is used to adjust the model parameters of the first language model through iterative training based on classification information and classification labels to obtain the target nursing record classification model; wherein, the iterative training adopts a joint loss function, which includes a combination of semantic similarity loss and cross-entropy loss.
[0062] Optionally, the first classification module 402 includes: The first determination submodule is used to determine the separation position of the target separator in the nursing record information; The second determination submodule is used to determine sample nursing entries from nursing record information based on the separation location.
[0063] Optionally, the second determining submodule includes: A segmentation unit is used to divide nursing record information into multiple initial nursing record entries based on the segmentation location; A generation unit is used to generate a knowledge graph of initial nursing record entries; Clustering units are used to cluster nursing record entries based on the same entities shared between knowledge graphs to obtain sample nursing entries.
[0064] Optionally, the generating unit includes: Extraction sub-units are used to extract semantic entities from nursing record entries; Sub-units are constructed to build knowledge graphs based on semantic entities and the relationships between them.
[0065] Optionally, the first language model includes a BERT language encoder module, a fully connected network module, and a convolutional neural network module; the fully connected network module is used to connect the BERT language encoder module and the convolutional neural network module; the first input / output module 403 includes: The third determination submodule is used to determine the encoding vector of the sample care item using the BERT language encoder module; The fourth determination submodule is used to determine the classification information of sample care items based on the encoding vector using the convolutional neural network module.
[0066] Optionally, the specific calculation method for the joint loss function includes: The fifth determination submodule is used to determine the first weight value of semantic similarity loss and the second weight value of cross-entropy loss, respectively; The sixth determination submodule is used to determine the function value of the joint loss function based on the first weight value, semantic similarity loss, second weight value, and cross-entropy loss.
[0067] Optionally, the device 400 further includes an enhancement module for data augmentation of sample care entries, the data augmentation including random mask word replacement or knowledge graph-based entity synonym expansion.
[0068] The model training device in this application embodiment can be integrated into an electronic device, or a component in an electronic device, such as an integrated circuit or a chip. The electronic device can be a terminal, or other devices besides a terminal. For example, the electronic device can be a GPUbox, mobile phone, tablet computer, robot, ultra-mobile personal computer (UMPC), or personal digital assistant (PDA), etc., and can also be a server, personal computer (PC), etc. This application embodiment does not specifically limit the specific implementation.
[0069] The model training device in this application embodiment can be a device with an operating system. This operating system can be Android, Linux, Windows, or other possible operating systems; this application embodiment does not specifically limit it.
[0070] The model training device provided in this application embodiment can achieve... Figure 1 The various processes implemented in the method embodiments will not be described again here to avoid repetition. Reference Figure 5 , Figure 5 This application provides a logic block diagram of an intelligent classification device for nursing record information. The device 500 may include: The acquisition module 501 is used to acquire the nursing record information of the target patient, which includes multiple nursing record entries. The second classification module 502 is used to perform a classification process on the nursing record information to obtain the target nursing record item; The third input / output module 503 is used to input multiple target nursing record items into the target nursing record classification model to obtain the classification information of the target nursing record items output by the target nursing record classification model. The classification information indicates the nursing record category of the nursing record item. The target nursing record classification model is obtained based on the model training method described in any of the above items. The third classification module 504 is used to classify and label nursing record information based on classification information.
[0071] The intelligent classification device for nursing record information in this application embodiment can be integrated into an electronic device, or a component of an electronic device, such as an integrated circuit or a chip. The electronic device can be a terminal, or other devices besides a terminal. For example, the electronic device can be a GPUBOX, mobile phone, tablet computer, laptop computer, PDA, robot, ultra-mobile personal computer (UMPC), or personal digital assistant (PDA), etc., and can also be a server, personal computer (PC), etc. This application embodiment does not specifically limit the specific implementation.
[0072] The intelligent classification device for nursing record information in this application embodiment can be a device with an operating system. This operating system can be Android, Linux, Windows, or other possible operating systems; this application embodiment does not specifically limit it.
[0073] The intelligent classification device for nursing record information provided in this application embodiment can achieve… Figure 3 The various processes implemented in the method implementation examples will not be described again here to avoid repetition.
[0074] This application provides an electronic device, see [link to relevant documentation] Figure 6 The electronic device 60 includes: a processor 601, a memory 602, and a computer program 6021 stored in the memory 602 and executable on the processor 601. When the processor 601 executes the program, it implements the model training method of the foregoing embodiments, or the intelligent classification method for nursing record information.
[0075] This application also provides a computer-readable storage medium storing a computer program / instructions thereon, which, when executed by a processor, implements the steps in the model training method or the intelligent classification method for nursing record information disclosed in this application.
[0076] This application also provides a computer program product that, when run on an electronic device, enables the processor to implement the steps in the model training method or the intelligent classification method for nursing record information disclosed in this application.
[0077] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.
[0078] This application describes embodiments with reference to flowchart illustrations and / or block diagrams of methods, apparatuses, electronic devices, and computer program products according to embodiments of this application. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing terminal device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing terminal device, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0079] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing terminal device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.
[0080] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal equipment, causing a series of operational steps to be performed on the computer or other programmable terminal equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable terminal equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.
[0081] Although preferred embodiments of the present application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of the embodiments of the present application.
[0082] 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 terminal device 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 terminal device. 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 terminal device that includes said element.
[0083] The above provides a detailed description of the model training method, intelligent classification method and device for nursing record information provided in this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The description of the above embodiments is only for the purpose of helping to understand the method and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.
Claims
1. A model training method, characterized in that, The method includes: Collect a dataset of hospital nursing records, which includes multiple nursing records labeled with classification tags; The nursing record information in the nursing record dataset is classified once to obtain sample nursing entries; The sample nursing entries are input into the first language model to obtain the classification information of the nursing record entries output by the first language model. The classification information indicates the nursing record category to which the nursing record entries belong. Based on the classification information and the classification labels, the model parameters of the first language model are adjusted through iterative training to obtain the target nursing record classification model; wherein, the iterative training adopts a joint loss function, which includes a combination of semantic similarity loss and cross-entropy loss.
2. The method according to claim 1, characterized in that, The step of performing a classification process on the nursing record dataset to generate a training input sequence includes: Determine the separation position of the target separator in the nursing record information; Based on the separation location, sample nursing entries are determined from the nursing record information.
3. The method according to claim 2, wherein determining the sample nursing item from the nursing record information based on the separation position includes: Based on the separation position, the nursing record information is divided into multiple initial nursing record entries; Generate a knowledge graph of the initial nursing record entries; Based on the shared entities among the knowledge graphs, the nursing record entries are clustered to obtain sample nursing entries.
4. The method according to claim 3, characterized in that, The knowledge graph used to generate the initial nursing record entries includes: Extract semantic entities from the nursing record entries; A knowledge graph is constructed based on the semantic entities and the relationships between them.
5. The method according to claim 1, characterized in that, The first language model includes a BERT language encoder module, a fully connected network module, and a convolutional neural network module; the fully connected network module is used to connect the BERT language encoder module and the convolutional neural network module. The step of inputting the sample nursing entries into a first language model to obtain the classification information of the nursing record entries output by the first language model includes: The BERT language encoder module is used to determine the encoding vector of the sample care item; Using the convolutional neural network module, the classification information of the sample care items is determined based on the encoded vector.
6. The method according to claim 1, characterized in that, The specific calculation method for the joint loss function includes: Determine the first weight value of the semantic similarity loss and the second weight value of the cross-entropy loss respectively; The function value of the joint loss function is determined based on the first weight value, the semantic similarity loss, the second weight value, and the cross-entropy loss.
7. The method according to claim 1, characterized in that, Also includes: The sample care entries are augmented with data including random mask word replacement or knowledge graph-based entity synonym expansion.
8. A method for intelligent classification of nursing record information, characterized in that, Includes the following steps: Obtain the nursing record information of the target patient, wherein the nursing record information includes multiple nursing record entries; The nursing record information is classified once to obtain the target nursing record item; Multiple target nursing record entries are input into a target nursing record classification model to obtain classification information of the target nursing record entries output by the target nursing record classification model. The classification information indicates the nursing record category of the nursing record entry. The target nursing record classification model is obtained based on the model training method according to any one of claims 1 to 7. Based on the classification information, the nursing record information is classified and labeled.
9. A model training device, characterized in that, The device includes: The data acquisition module is used to collect hospital nursing record datasets, which include multiple nursing record information labeled with classification tags; The first classification module is used to perform a classification process on each nursing record information in the nursing record dataset to obtain sample nursing entries. The first input / output module is used to input the sample nursing item into the first language model and obtain the classification information of the nursing record item output by the first language model. The classification information indicates the nursing record category to which the nursing record item belongs. An adjustment module is used to adjust the model parameters of the first language model through iterative training based on the classification information and the classification labels to obtain a target nursing record classification model; wherein, the iterative training adopts a joint loss function, which includes a combination of semantic similarity loss and cross-entropy loss.
10. A nursing record classification device, characterized in that, The device includes: The acquisition module is used to acquire nursing record information of the target patient, wherein the nursing record information includes multiple nursing record entries; The second classification module is used to perform a classification process on the nursing record information to obtain the target nursing record item; The third input / output module is used to input multiple target nursing record entries into the target nursing record classification model to obtain the classification information of the target nursing record entries output by the target nursing record classification model, wherein the classification information indicates the nursing record category of the nursing record entry; The third classification module is used to classify and label the nursing record information based on the classification information.