First aid information processing method and device, electronic equipment and storage medium

Abstract

The present disclosure discloses an emergency information processing method and device, an electronic device and a storage medium. The emergency information processing method can include: correcting a first emergency text to obtain a second emergency text; obtaining an emergency key question text based on the second emergency text and a target model; wherein the target model is obtained by training a third emergency text and an emergency expert knowledge text.

Description

Technical Field

[0001] This disclosure relates to the field of information technology, and in particular to a method and apparatus for processing emergency medical information, an electronic device, and a storage medium. Background Technology

[0002] With the development of technology, machine learning algorithms can be used to process emergency texts, thereby outputting preliminary emergency guidance.

[0003] However, machine learning algorithms for emergency scenarios in related technologies have problems such as incomplete or incorrect emergency information collection and / or inaccurate emergency guidance when collecting emergency information and / or providing emergency guidance.

[0004] In view of this, the ability of machine learning algorithms based on emergency text to collect emergency information and / or provide correct emergency guidance needs to be further improved. Summary of the Invention

[0005] This disclosure provides an emergency information processing method and apparatus, electronic device and storage medium, providing more suitable or more complete emergency key issue texts.

[0006] This disclosure provides an emergency medical information processing method, including:

[0007] The second emergency response text is obtained by correcting errors in the first emergency response text.

[0008] Based on the second emergency text and the target model, a text on key emergency issues is obtained; wherein, the target model is trained using the third emergency text and emergency expert knowledge text.

[0009] A second aspect of this disclosure provides a method for training a target model, the method comprising:

[0010] Obtain third-party emergency text and emergency expert knowledge text; train a preset network using the third-party emergency text and the emergency expert knowledge text to obtain a target model;

[0011] The target model is used at least to determine the first emergency text after error correction, the key issue text of the emergency text, and / or the emergency guidance text, and to obtain the third emergency text and the emergency expert knowledge text.

[0012] A third aspect of this disclosure provides an emergency medical information processing device, comprising:

[0013] The module is used to correct errors in the first emergency text to obtain the second emergency text;

[0014] The output module is used to obtain a text of key emergency questions based on the second emergency text and the target model; wherein the target model is trained using the third emergency text and emergency expert knowledge text.

[0015] A fourth aspect of this disclosure provides a target model training apparatus, the apparatus comprising:

[0016] The acquisition module is used to acquire third-party emergency text and emergency expert knowledge text;

[0017] The training module is used to train a preset network using the third first aid text and the first aid expert knowledge text to obtain the target model;

[0018] The target model is used at least to determine the first emergency text after error correction, the key issue text of the emergency text, and / or the emergency guidance text, and to obtain the third emergency text and the emergency expert knowledge text.

[0019] A fifth aspect of this disclosure provides an electronic device, comprising:

[0020] Memory;

[0021] A processor, connected to the memory, is configured to implement the emergency information processing method provided by any of the first and / or second aspects of the foregoing technical solutions by executing computer-executable instructions stored in the memory.

[0022] A sixth aspect of this disclosure provides a computer storage medium, characterized in that the computer storage medium stores computer-executable instructions; after being executed by a processor, the computer-executable instructions are able to implement the emergency information processing method provided by any of the aforementioned first and / or second aspects.

[0023] The technical solution provided in this disclosure first corrects the received first emergency text before using the target model to output the key question set text, thus obtaining the corrected second emergency text. In this way, the target model can provide a more complete key question set text that is more adapted to the current emergency scenario based on the correct second emergency text, thereby improving the ability to collect emergency information. Attached Figure Description

[0024] Figure 1 A flowchart illustrating an emergency medical information processing method provided in this embodiment of the present disclosure;

[0025] Figure 2 A flowchart illustrating an emergency medical information processing method provided in this embodiment of the present disclosure;

[0026] Figure 3This is a schematic diagram of emergency text processing provided in an embodiment of the present disclosure;

[0027] Figure 4 A schematic diagram of an emergency medical expert knowledge text and an emergency medical knowledge text provided for embodiments of this disclosure;

[0028] Figure 5 A schematic diagram of an emergency medical expert knowledge text provided in an embodiment of this disclosure;

[0029] Figure 6 A schematic diagram of a first aid instruction text provided in an embodiment of this disclosure;

[0030] Figure 7 This is a schematic diagram of a target model training method provided in an embodiment of this disclosure;

[0031] Figure 8 This is a schematic diagram illustrating how a hypothetical sentence is transformed into a selective question, as provided in an embodiment of this disclosure.

[0032] Figure 9 This is a schematic diagram of a preset model provided in an embodiment of this disclosure;

[0033] Figure 10 This is a schematic diagram of a preset model provided in an embodiment of this disclosure;

[0034] Figure 11 This is a schematic diagram of the structure of an emergency information processing device provided in an embodiment of this disclosure;

[0035] Figure 12 This is a schematic diagram of the structure of an emergency information processing device provided in an embodiment of this disclosure;

[0036] Figure 13 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this disclosure. Detailed Implementation

[0037] To gain a more detailed understanding of the features and technical content of this disclosure, the implementation of this disclosure will be described in detail below with reference to the accompanying drawings. The accompanying drawings are for reference and illustration only and are not intended to limit this disclosure.

[0038] The emergency medical plan database is a core component of the intelligent emergency medical assistance system, serving as the basis for dispatchers to provide remote guidance to emergency callers. Its main function is to assist dispatchers at the dispatch center in collecting emergency call information and to provide remote assistance guidance to callers as much as possible before the ambulance arrives at the scene. The application scenario begins when a caller dials the number and the dispatcher at the dispatch center answers the phone; therefore, the emergency medical plan database is primarily intended for dispatcher use.

[0039] Current emergency response plan databases include an entry question set, a key emergency response question set, post-dispatch emergency guidance, and / or auxiliary tools. The key emergency response question set primarily comes from the knowledge of emergency response experts. An emergency response expert knowledge base is constructed through manual screening of this knowledge. However, building such a knowledge base is often time-consuming and inefficient. Furthermore, because it relies on a limited pool of experts, the source is relatively singular, resulting in an incomplete key question set that may not cover certain emergency scenarios. This leads to problems such as insufficient usability of the key question set or incomplete emergency response information.

[0040] like Figure 1 As shown, this disclosure provides an emergency medical information processing method, including:

[0041] S1110: Correct the first emergency text to obtain the second emergency text;

[0042] S1120: Based on the second emergency text and the target model, obtain the emergency key issue text; wherein, the target model is trained using the third emergency text and emergency expert knowledge text.

[0043] This emergency information processing method can be executed by various electronic devices, which may include: devices of various types of terminals and / or servers, etc.

[0044] In some embodiments, the first emergency text may be text collected at the emergency scene, such as emergency information entered by the victim's family or friends via voice and / or text. In other embodiments, the first emergency text may be online text related to emergency response.

[0045] In one embodiment, the first emergency text may include information about the basic symptoms of the person receiving emergency care.

[0046] The basic symptoms may include: the basic symptoms exhibited by the person receiving emergency care, or the visualized symptoms, or information about the basic symptoms collected through a smart device worn by the person receiving emergency care.

[0047] If an emergency text is received and it is determined that there is no basic medical information, a prompt message is output asking for basic medical information. This reminds the emergency personnel at the scene to provide the first emergency text that the target model can process.

[0048] In some embodiments, if a received emergency text does not contain basic information about the person requiring emergency care, the target model can prompt for the basic information of the person requiring emergency care while processing the first emergency text. After obtaining the basic information, the first emergency text will be corrected, and the corrected first emergency text will become the second emergency text. This second emergency text, compared to the first emergency text, at least reduces the number of non-medical terms or incorrect words given by non-professionals.

[0049] For example, in one embodiment, the first emergency text may include information about at least one basic symptom, so that the target model can conveniently process the information and output the corresponding question set.

[0050] In another embodiment, the first emergency text may further include: basic information describing the person receiving emergency care. This basic information may include, but is not limited to: the person receiving emergency care's gender, age, medical history, location of the emergency scene, and / or basic assistance available at the emergency scene.

[0051] The error correction here refers to one method of preprocessing the first aid text. In specific implementations, the preprocessing of the first aid text may also include: word segmentation, word segmentation encoding, and / or removal of punctuation marks or spaces, etc.

[0052] In this embodiment of the disclosure, after obtaining the second emergency text, an emergency critical questions text is generated.

[0053] The critical questions text for first aid here may include: at least one question text related to first aid.

[0054] In another embodiment, if the emergency critical question text includes multiple emergency questions, it also includes a sequential text outlining the jump relationships between the multiple emergency questions. This sequential text facilitates emergency operators or emergency equipment in outputting one or more emergency questions in order. Further emergency information can then be collected through the emergency caller's responses to the emergency questions.

[0055] It is worth noting that this emergency critical question text can be used to construct the emergency critical question set in the emergency plan library, or it can be used after the plan library is established. When an emergency call is received, the emergency critical question set established in the plan library can be exported based on the first emergency text corresponding to the emergency call, and emergency-related questions can be given to collect more information related to the patient.

[0056] First, the third-party emergency text can be any text related to emergency care that differs from the knowledge text of emergency care experts. This third-party emergency text can be a text related to emergency care provided by ordinary patients, patients' families, and / or relatives of patients who are not emergency care experts. For example, this third-party emergency text can be a non-professional emergency text.

[0057] First aid expert knowledge book can be a professional text provided by first aid experts, and the first aid expert instruction text can be a first aid text provided by first aid experts based on their own prior knowledge and / or medical professional books.

[0058] In summary, the third emergency text may include: emergency text provided by on-site personnel in actual emergency scenarios collected from previous emergency calls, and / or emergency text collected from the internet. This allows the target model to utilize emergency text from actual emergency scenarios in its training, and also incorporates expert knowledge text for emergency care, enabling the target model to correct errors in the first emergency text. Furthermore, the introduction of the third emergency text broadens and increases the amount of training data available for the target model, further enhancing its information processing capabilities. These capabilities include, but are not limited to, information collection capabilities and / or the ability to accurately output emergency guidance text.

[0059] The target model can be any type of machine learning model, for example, it can be any type of neural network model.

[0060] In one embodiment, preprocessing such as error correction of the first emergency text can be performed in a separate preprocessing model, or error correction can be achieved through a non-machine learning model. In short, there are multiple ways to correct errors in the first emergency text, and no specific implementation is limited.

[0061] In another embodiment, the first first aid text is input into a target model, which may include one or more preprocessing layers that can preprocess the first first aid text.

[0062] The preprocessing here may include:

[0063] Text segmentation;

[0064] Encoding of characters, words, phrases, sentences, etc.;

[0065] The location and / or encoding of key information in the first emergency text, etc.

[0066] It is worth noting that in this embodiment of the disclosure, the preprocessing may further include: correcting errors in the first emergency text.

[0067] Error correction here may include: correcting typos and / or correcting non-medical terms to medical terms, etc.

[0068] In this embodiment, because the preprocessing layer of the target model has error correction capabilities, it can resolve some of the errors and incompleteness caused by inaccurate initial emergency text provided by non-professionals at the emergency scene, thereby improving the ability to collect emergency information. Simultaneously, by collecting more complete emergency information and reducing errors in the initial emergency text, the accuracy of the emergency guidance text provided by the target model can be improved, thus also enhancing the accuracy of the emergency guidance text.

[0069] In some embodiments, S1110 may include:

[0070] The first emergency text is segmented using the target model to obtain the first segmentation sequence;

[0071] The erroneous words and / or non-medical terms in the first word segmentation sequence are replaced with the medical terms to obtain the second word segmentation sequence, and the second emergency text is obtained based on the second word segmentation sequence.

[0072] The second word segmentation sequence here is the same as the word segmentation sequence of the second emergency text. Therefore, the second word segmentation sequence can be converted into the second emergency text.

[0073] If the preprocessing layer of the target model completes the error correction of the first emergency text and obtains the second word segmentation sequence representing the second emergency text, it is not necessary to convert the second word segmentation sequence into the second emergency text. The information processing layer of the target model can directly perform subsequent processing on the second word segmentation sequence until the emergency key issue text is output.

[0074] In this embodiment of the disclosure, the first text is segmented, for example, using the segmentation layer of a target model or preprocessing model to segment the first emergency text as natural language, thereby obtaining a first segmented sequence after segmenting the first emergency text. Then, after segmentation, erroneous words and / or non-medical terms in the first segmented sequence are replaced with medical terms mentioned in the emergency expert knowledge text. These medical terms can be emergency / medical terminology. Thus, the second segmented sequence can be an accurate sequence after error correction.

[0075] Furthermore, after generating the second word segmentation sequence, the second emergency text is generated.

[0076] In some embodiments, such as Figure 2 As shown, S1130 may include:

[0077] S1131: Perform an embedding operation on the second word segmentation sequence to obtain the first embedding sequence;

[0078] S1132: When the first embedded sequence exceeds the preset length corresponding to the target model, the first and last elements of the embedded text sequence are truncated to obtain a second embedded sequence equal to the preset length;

[0079] S1130: Output the problem set according to the second embedding sequence.

[0080] After obtaining the corrected second word segment sequence, the second word segment is embedded using the embedding layer of the target model or preprocessing model. This will convert the second word segment sequence into an embedding feature encoded with the embedding feature (i.e., the first embedding sequence).

[0081] Thus, in this embodiment of the disclosure, if the first embedded sequence obtained after the embedding operation exceeds a preset length that the target model can process at one time, the first embedded sequence will be truncated. During truncation, the middle portion of the first embedded sequence will be removed, retaining only the first and last portions. The preset length here can be the maximum length of the longest embedded sequence that the target model or preprocessing model can process.

[0082] Generally, the first and last elements of emergency rescue text contain more information. Therefore, by using this sequence truncation method, we can retain as much information as possible from the first emergency rescue text, so that the question set can be output more accurately based on the truncated word segmentation sequence.

[0083] In some embodiments, outputting the second emergency text based on the second word segmentation sequence includes:

[0084] Based on the second word segmentation sequence, construct word vectors, sentence vectors, and the positions of preset words in the second word segmentation sequence, and determine the position vectors;

[0085] The key emergency response text is obtained by weighting the word vectors, sentence vectors, and position vectors using an attention mechanism.

[0086] In one embodiment, the target model may include a preprocessing layer.

[0087] For example, the preprocessing layer may consist of one or more convolutional networks.

[0088] If the target model includes a preprocessing layer, then S1110 may include:

[0089] The first emergency text is input into the preprocessing layer for error correction to obtain the second emergency text.

[0090] The target model itself has the function of preprocessing the first aid text, such as error correction, which makes it simpler to implement and eliminates the need for additional preprocessing models, thus reducing the required tools.

[0091] In some embodiments, the target model further includes an information processing layer connected to the back end of the preprocessing layer. For example, the output of the preprocessing layer is connected to the input of the information processing layer. The information processing layer may include an encoding layer and / or a decoding layer, etc. Of course, this is merely an example of an information processing layer, and the specific implementation is not limited to the above example.

[0092] In this embodiment of the disclosure, in order to implement the target model in a specific process, the encoding layer can encode the second word segmentation sequence. By encoding from the word dimension (i.e., word vectors), sentence dimension (i.e., forming sentence vectors), and / or position vector encoding, comprehensive information can be extracted to output complete emergency key issue text.

[0093] In some embodiments, the critical first aid question text includes at least one optional question related to first aid.

[0094] The questions described in the first aid key issues text are all optional questions, and the answers to optional questions can include "yes" and "no".

[0095] In this way, when selective questions are presented to personnel at the emergency scene, even those who are not medical professionals are more likely to provide answers, thus simplifying the acquisition of emergency information. For example, the stem of a selective question can prompt emergency personnel to identify which aspects of the patient's information they need to focus on and provide corresponding responses.

[0096] In this way, the device can combine the stem of the selective question with the received answers to obtain emergency information. This emergency information may include descriptions of various aspects of the patient and / or environmental information of the emergency scene.

[0097] In some embodiments, the critical emergency text includes a question text and a sequence text;

[0098] The selective questions described in the question text include: initial selective questions, alternative selective questions, and cutoff selective questions;

[0099] The sequential text is used to output the (m+1)th selective question based on the answer to the m-th selective question; where m is less than M, and M is the total number of selective questions.

[0100] For example, this initial selective question can be used in emergency rescue scenarios, such as drowning, heart failure, sudden death, bleeding, and / or fracture scenarios. When the first emergency rescue text contains keywords such as "drowning" or "choking," a key emergency rescue question text related to drowning can be invoked. This initial selective question could be "Has a person drowned?"

[0101] Of course, the above are just examples. The sequential text refers to the instruction information to jump to the target selective question to be output based on the answer to the previous output selective question.

[0102] In this way, the operator can provide the selective question based on the caller's answer and the sequential text. Alternatively, the server can select the target selective question to be output based on the received answer to the previous selective question and the sequential text.

[0103] The process continues until all optional questions in the identified critical emergency question set have been output or the system jumps to the cutoff optional question position. If there are still optional questions that have not been output in the corresponding critical emergency question set, the system jumps to the cutoff optional question position. Upon receiving the cutoff optional question, the output of optional questions stops, indicating that the emergency information has been obtained or the critical emergency question set needs to be changed.

[0104] For example, the set of critical questions for first aid includes: initiation selection questions, alternative selection questions, and termination selection questions related to first aid;

[0105] The method further includes:

[0106] Based on the text of critical emergency questions, output the initial selective questions;

[0107] Based on the response to the nth output selective problem, select the (n+1)th selective problem to be output from the candidate selective problems and the cutoff selective problems, until the cutoff selective problem or all selective problems in the emergency set have been output; where n is any positive integer;

[0108] Based on the response results of the selected questions, the fourth emergency text is obtained.

[0109] The fourth emergency text is the response result generated for the selective question.

[0110] For example, the fourth first aid text may include: first aid keywords that are output as selective questions and answered with preset answers. The preset answer may be "yes" or "no".

[0111] As another example, the fourth emergency text may include: all the output selective questions and their responses.

[0112] In summary, the fourth emergency text is generated based on the output emergency key question text and / or the response results of the emergency key question text. The fourth emergency text generated in this way can provide relatively complete information about the person being rescued at the emergency scene and / or information about the environment at the emergency scene.

[0113] In some embodiments, the step of outputting first aid guidance text based on the fourth first aid text includes:

[0114] Based on the fourth emergency text, determine the type of emergency response that reflects the emergency scenario;

[0115] The first aid guidance text is determined based on the pre-established mapping relationship between the first aid type and the guidance content.

[0116] Based on the basic information of the emergency responder and the fourth emergency text collected from the question set, an emergency guidance text is provided. If it is determined that the emergency responder's condition is serious and requires dispatching an ambulance, the system can automatically dispatch an ambulance from the nearest location based on the emergency responder's location, so that professional medical personnel can arrive at the scene as quickly as possible to provide emergency treatment.

[0117] The first aid instructions text may include:

[0118] Instruct other personnel at the emergency scene to perform emergency procedures on the person in need, such as turning them over and / or artificial respiration.

[0119] In summary, the aforementioned first aid guidance text can be used as a guide for dispatching various emergency first aid measures.

[0120] The output set of questions may include one or more questions that further collect symptom information of the first aid recipient and / or environmental information at the first aid scene that may provide emergency measures.

[0121] In some embodiments, determining the emergency response type reflecting the emergency response scenario based on the fourth emergency response text may include:

[0122] Based on the fourth emergency text and the second emergency text, the emergency type reflecting the emergency scenario is determined.

[0123] For example, different types of first aid require different first aid procedures. Specifically, the first aid types can be divided into at least one of the following: drowning first aid scenario, diving first aid scenario, traffic accident scenario, cardiac or respiratory arrest scenario, fainting scenario, chest pain scenario, limb convulsion scenario and / or limb numbness scenario.

[0124] Different types of emergency medical services can have different emergency medical service instructions before dispatch, thus enabling the provision of accurate emergency medical service instructions before dispatch.

[0125] In some embodiments, certain emergency scenarios do not require dispatching medical vehicles for professional medical personnel, and therefore do not require dispatching professional medical personnel for emergency treatment. For example, if a fainting scenario is determined to be caused by only a brief syncope and recovery is quick, then there is no need to dispatch professional medical personnel for emergency treatment.

[0126] In some embodiments, the method further includes: outputting an entry question set text when the first emergency text does not contain preset information.

[0127] The entry question set text includes: one or more entry questions.

[0128] The entry question is used to inquire about the gender, age, and any medical history of the person being treated. The entry question may also include: the emergency scene and / or multiple emergency contacts, etc. In some embodiments,

[0129] If the first-aid text lacks basic information about the first-aid recipients mentioned in the entry question text, then the output entry question set text will contain...

[0130] In one embodiment, multiple entry questions in the entry question set text are parallel and independent, and all require information to be understood.

[0131] In another embodiment, the entry question set text also includes a sequence text, which may indicate the order in which answers to the entry questions are collected, based on the importance and / or necessity of the entry questions.

[0132] In some embodiments, before correcting the first emergency text to obtain the second emergency text, the method further includes:

[0133] The first emergency text is matched and filtered with the emergency guidance set to obtain the target emergency text;

[0134] Based on the target emergency text, candidate emergency question texts are constructed; wherein, the key emergency question texts are derived from the candidate emergency question texts.

[0135] In this embodiment of the disclosure, in order to improve the contingency plan library and make the candidate question texts of the emergency key question set in the contingency plan library more accurate, the first emergency text is matched with the emergency guidance set to obtain accurate and usable texts that can be used to construct candidate emergency question texts.

[0136] The step of matching and filtering the first emergency text with the emergency guidance set may include:

[0137] The keywords of the first aid text are matched with the keywords of the first aid guidance set. The first aid text that successfully matches is identified as the target first aid text.

[0138] The keywords may include: medical terms related to first aid and / or non-medical terms, etc.

[0139] like Figure 7 As shown, a target model training method includes:

[0140] S2110: Obtain the third-party emergency text and the emergency expert knowledge text respectively;

[0141] S2120: Train the preset network using the third first aid text and the first aid expert knowledge text to obtain the target model;

[0142] The target model is used at least to determine the first emergency text after error correction, the key issue text of the emergency text, and / or the emergency guidance text.

[0143] In this embodiment of the disclosure, third-party emergency text and emergency expert knowledge text will be acquired respectively, and the target model will be trained simultaneously using standard emergency expert knowledge text and third-party emergency text, thereby enabling the target model to have error correction function. Figure 5 The image shown is a schematic diagram of a first aid expert knowledge text.

[0144] The third first aid text can be text collected from various actual first aid scenarios and / or various first aid texts collected from the internet. Such first aid texts may contain typos, colloquial language, and / or contain non-medical professional terminology.

[0145] The first aid expert knowledge text may include at least one of the following:

[0146] Knowledge text specifically provided by emergency medical experts for model training;

[0147] Actual emergency medical records generated by emergency medical experts during their rounds.

[0148] In some embodiments, the preset network may include: one or more preprocessing layers and / or a natural language processing neural network. This neural network may include, but is not limited to, the BERT network. Figure 4 As shown, after converting first aid expert knowledge text and common first aid text into embedding features (i.e. through embedding), they are input into the BERT network to obtain the similarity output by the BERT network.

[0149] The preset network may include: a preprocessing layer and an information processing layer connected to the preprocessing layer.

[0150] In one embodiment, the preprocessing layer has an error correction function. In another embodiment, the preprocessing layer does not have an error correction function. The error correction function can be used to replace erroneous words and / or non-medical terms with medical terms.

[0151] In practical implementation, the preprocessing layer is not limited to error correction; it can also have functions such as word segmentation, filtering out meaningless words, and / or converting word segments into embedding vectors.

[0152] If the preprocessing layer has an error correction function, the target model obtained by training the preset network simultaneously with the actual third-party emergency text containing non-medical professional terms or errors and emergency expert knowledge text has the function of correcting emergency text and can output a suitable set of questions based on the corrected emergency text, thereby improving the emergency information collection ability of the trained target model and / or the ability to provide correct emergency guidance text.

[0153] like Figure 3 As shown, two emergency texts, "difficulty breathing, coma, and arrhythmia" and "difficulty breathing, coma, and cough," are input into the target model and then converted into word vectors, protocol vectors, and position embeddings. The position embedding vector here is one type of position vector mentioned above, but in specific implementations, it represents the position of the preset emergency keywords in the second word segmentation sequence and is not limited to the position embedding vectors mentioned above.

[0154] For example, keywords describing the patient's current condition such as "coma" and / or "cough" are the aforementioned preset emergency keywords.

[0155] For example, if a preset emergency rescue keyword table is configured in the contingency plan database, all preset emergency rescue keywords can be derived from this table. In this case, matching methods such as regular expressions can be used to extract the aforementioned preset emergency rescue keywords from the second emergency rescue text.

[0156] For example, the preset emergency keywords may be entity words such as nouns or verbs that characterize the patient's current condition, while function words such as modal particles or adverbs are not used as preset emergency keywords.

[0157] In some embodiments, training a preset network using the third first aid text and the first aid expert knowledge text to obtain a target model includes:

[0158] When the similarity between a third-party emergency text and an emergency expert knowledge text reaches a preset threshold, a training text team is constructed.

[0159] The preset model is trained based on the training text, and the preset model is used to replace the professional words in the emergency medical expert knowledge text with erroneous words and / or non-medical professional words in the training text.

[0160] In this embodiment of the disclosure, during the training process, the third-party emergency text and the emergency expert knowledge text are matched. If the similarity between the two reaches a preset threshold, a training text team is constructed to train the error correction capability of one or more preprocessing layers in the model.

[0161] In some embodiments, training a preset network using the third first aid text and the first aid expert knowledge text to obtain a target model includes:

[0162] When the first aid expert knowledge text contains hypothetical sentences, the input text and selective questions asking about the symptoms of the person receiving first aid are constructed based on the distribution of hypothetical conjunctions in the hypothetical sentences.

[0163] When training the model, the correspondence between the input text and the question is constructed based on the hypothetical conjunctions of the hypothetical sentences. The input text and the question with this correspondence are the direct sample data of the preset network.

[0164] For example, the hypothetical sentence "If there are no symptoms of headache or dizziness, then confirm whether there is vomiting" can construct training text pairs for "headache and dizziness" and "presence or absence of vomit." Here, "presence or absence of vomit" is the aforementioned selective text. "Headache and dizziness" is the aforementioned "input text."

[0165] Figure 8 The diagram illustrates how an encoder and decoder convert a hypothetical sentence into a selective question.

[0166] Figure 9 This is a schematic diagram of a preset model provided in an embodiment of the present disclosure. The preset model may include: an encoder and a decoder; and a recurrent neural network element between the encoder and the decoder.

[0167] The emergency medical plan database is a core component of the intelligent emergency medical assistance system, serving as the basis for dispatchers to provide remote guidance to emergency callers. Its main function is to assist dispatchers at the dispatch center in collecting emergency call information and to provide remote assistance guidance to callers as much as possible before the ambulance arrives at the scene. The application scenario begins when a caller dials the number and the dispatcher at the dispatch center answers the phone; therefore, the emergency medical plan database is primarily intended for dispatcher use.

[0168] The emergency response plan database of the intelligent emergency assistance system includes: an entry question set, a key question set, post-dispatch guidance text, and / or information on rescue auxiliary tools. The entry question set here can be one of the aforementioned second question set. The key question set here can be one of the aforementioned first question set.

[0169] The entry question set can be used to collect basic information such as location, gender, age, level of consciousness, breathing status, and chief complaint (patient condition, scene situation) to identify and recommend appropriate emergency response plans.

[0170] A single emergency response plan (such as for drowning or diving accidents or traffic accidents) includes a set of key questions, post-dispatch instructions, and auxiliary tools. The set of key questions is used to further gather crucial information about the patient within the current emergency response plan. Post-dispatch instructions are used by the dispatcher to remotely guide on-site personnel in providing assistance before the ambulance arrives. Auxiliary tools are used to assist in assessing the patient's condition or in implementing rescue procedures.

[0171] The entry questions in the emergency response plan database are relatively general questions, while the key question set consists of specific questions raised by dispatchers based on the patient's situation at the scene, under different emergency response categories.

[0172] The key questions set includes key questions under various emergency response plans (e.g., drowning and diving accidents, traffic accidents, cardiac or respiratory arrest or death, chest pain, etc.). Due to the urgency of the emergency response scenarios, the key question set for each emergency response category is essentially a set of sequential yes / no questions. For example, the key questions for a drowning and diving accident plan include whether the patient is conscious, whether the patient is breathing normally, and whether the patient has been removed from the water.

[0173] This disclosure provides a method for extracting a set of key questions from data such as guidance from emergency medical experts and common emergency medical procedures collected online, including the following steps:

[0174] Step 1: Preprocess the text data, including expert guidance texts and common first aid measures collected online. Preprocessing includes error detection, correction, and medical terminology conversion of data from different sources, as well as using machine learning methods to preserve the maximum semantic meaning of long texts.

[0175] Step 2: Perform named entity recognition and part-of-speech tagging on the preprocessed text data.

[0176] Step 3: Input the labeled words and entities into the encoder and decoder, output the key question set, and establish the mapping relationship.

[0177] Step 4: Following steps 1-3, perform data preprocessing and model training. When extracting the key question set from the text data, input the labeled words and entities into the encoder and decoder, and the output is the sequential key question set required by the contingency plan.

[0178] Specifically, this disclosure provides a method for extracting a set of key questions from data such as guidance from emergency medical experts and common emergency medical procedures collected online, including the following steps:

[0179] 1. First aid-related data from different sources undergoes error detection, correction, and medical terminology processing. All first aid-related data is text-based. Because some common first aid measures are sourced online, they may differ from expert knowledge and contain non-medical terminology. For keyword (key phrase) extraction in the first aid scenario of this disclosure embodiment, in addition to conventional text preprocessing (e.g., word segmentation, cleaning special symbols and stop words), machine learning techniques can be used for error correction and medical terminology preprocessing. The method may include:

[0180] Step 1: Segment the common online emergency rescue measures into words, and then use a medical knowledge graph to find and replace the non-medical terms in the segmented words with the corresponding medical terms.

[0181] Step 2: Perform embedding operations on the first aid expert knowledge text and the terminated online first aid common measures text separately, and then input the embedding results into the BERT network. Assume the longest text sequence in BERT is 512. The first aid expert knowledge text and the terminated online first aid common measures text may exceed 512. The general practice is to truncate the excess portion. However, in the first aid expert knowledge text, the later parts of the text are more important than the earlier parts. For example... Figure 5 This is an excerpt from a professional book regarding routine emergency treatment for drowning. For example... Figure 5 The items listed are arranged in the following order: clinical manifestations, etiology and pathogenesis, pre-hospital care, emergency examination, diagnosis, differential diagnosis, emergency treatment, indications for observation, indications for hospitalization, and indications for discharge. For collecting the key issues in the drowning emergency plan, the contents of clinical manifestations, emergency examination, diagnosis, differential diagnosis, and emergency treatment can relatively completely express the key issues and post-dispatch guidance.

[0182] Therefore, this embodiment proposes a method for concatenating the beginning and end of the text to obtain more balanced information and avoid omitting important first aid content. The processed first aid expert knowledge text and common online first aid measures text are input into the network, and the similarity between the two texts is compared. Text containing common online first aid measures with a similarity greater than a threshold of 0.8 is retained. This ensures that the retained common online first aid measures text accurately represents professional knowledge in first aid scenarios, improving the effectiveness and accuracy of the selection process.

[0183] For pre-processed emergency medical expert knowledge texts and online emergency medical common measures texts, a certain filtering strategy can be employed based on the correlation with the guidance content in post-dispatch instructions. This enhances the correspondence between the key question set and the post-dispatch instructions, resulting in better association and higher content matching. For example, filtering can be based on the emergency type (i.e., emergency scenario, such as drowning, traffic accidents, respiratory arrest, etc.) in the post-dispatch instructions. If the post-dispatch instructions include guidance content corresponding to emergency type 1, emergency type 2, and emergency type 3, the corresponding emergency texts for these three types can be selected from the pre-processed emergency medical expert knowledge texts and online emergency medical common measures texts, or these texts can be prioritized. Another example is content association and matching filtering by extracting emergency keywords. In this way, by associating and filtering online emergency medical common measures with emergency medical expert knowledge, the key question set subsequently obtained based on these measures and knowledge can be better associated with the post-dispatch instructions, resulting in higher matching.

[0184] The pre-processed and screened first aid texts (including first aid expert knowledge texts and texts on common online measures) were uniformly organized and used to... Figure 5 Taking drowning first aid knowledge as an example, we first use named entity recognition methods (such as Bert+LSTM+CRF) to identify the above entities and retain the entity words as key information.

[0185] Then, part-of-speech tagging methods (such as BERT) are used to tag hypothetical conjunctions (such as if, if, if, supposing, as long as, unless, suppose, even if, if, for example, etc.), nouns, verbs, etc. in the text. However, when using BERT for part-of-speech tagging, because emergency medical texts often contain many long medical terms and sentences, a single entity cannot fully express the meaning of the emergency medical text.

[0186] This embodiment uses word vectors, sentence vectors, and positional encoding weighted as input to the BERT network. Specifically, the clinical manifestations of drowning include difficulty breathing, coma, and arrhythmia, while the clinical manifestations of airway obstruction also include difficulty breathing, coma, and choking. As can be seen, a single entity cannot distinguish the type of emergency response plan. It is necessary to add the information of the entire sentence to accurately reflect the emergency response scenario category, as shown in the figure below.

[0187] Figure 5 Taking drowning first aid knowledge as an example, information such as entities, nouns, and verbs near hypothetical conjunctions in the text is extracted to obtain short texts such as whether there is chest movement, whether there is no breathing or heartbeat, and whether there is vomiting. The construction for "vomiting" is: hypothetical conjunction (if present) + entity / verb / medical term (vomiting). Using such short texts as input, the corresponding output text "whether there is vomiting" is constructed, and a mapping relationship is established. In other words, during part-of-speech tagging, hypothetical statements can be converted into selective questions, directly outputting the key question set corresponding to the first aid plan. Simultaneously, based on the positional encoding at the time of input, sequential yes / no selective questions can be directly output.

[0188] like Figure 10 As shown, an encoder and decoder are designed based on a self-attention mechanism. The internal structure and connection methods of the encoder and decoder are illustrated. The decoder contains both a self-attention layer and an attention layer. The self-attention layer processes features using its own attention mechanism, while the attention layer performs attention processing, followed by forward processing in the feed-forward layer.

[0189] like Figure 9 As shown, there are 8 encoder layers and 8 decoder layers, all fully connected, with a recurrent neuron in the middle. This recurrent neuron efficiently utilizes information from the encoder layers, retaining valid information and filtering out invalid information. The internal structures of the encoder and decoder are as follows: Figure 9 As shown, the output of the encoder / decoder is connected to a softmax layer and a linear layer. The loss is calculated using the cross-entropy loss function with the label, and the weight parameters are optimized through backpropagation to train the network.

[0190] This embodiment incorporates a recurrent neuron, which can retain previously input short text information containing hypothetical conjunctions. The formula is as follows:

[0191] O t =W·E t +λ·(W·Et-1 +b t-1 )+b t

[0192] Among them, O t It is the output at time t, W is the fully connected weights of the 8 encoder layers, and E is the output at time t. t b is the output of the encoder layer at time t. t This is the bias term at time t. λ can be a pre-set coefficient. t It can be a pre-set bias value.

[0193] The above steps complete the training of the codec network and persist the codec model. When text data such as first aid expert knowledge and common first aid measures from different sources are collected, after preprocessing, named entity recognition, and part-of-speech tagging, short texts of hypothetical relation conjunctions + entities / verbs / medical professional terms are extracted. These short texts are then input into the persisted codec model, which outputs the sequential key questions required for the first aid plan.

[0194] This disclosure uses machine learning methods to perform error detection, error correction, and medical terminology preprocessing on data from different sources, which can improve the accuracy of extracting key question sets in emergency scenarios.

[0195] This embodiment of the disclosure uses word vectors, sentence vectors, and positional encoding weighted as input to the BERT network for part-of-speech tagging, which can more closely approximate the text information expressed in emergency rescue scenarios. Furthermore, by converting hypothetical statements into selective questions during the part-of-speech tagging process, it is more conducive to quickly collecting emergency scene information in emergency rescue scenarios.

[0196] This disclosure proposes a method for concatenating the beginning and end of text, which, compared to the commonly used method of truncating parts of the text, can obtain more balanced information from emergency texts and improve the accuracy of semantic recognition of emergency texts. Furthermore, it uses natural language processing technology to retain only common emergency measures texts collected online that are semantically similar to expert knowledge texts, thereby improving the effectiveness and professionalism of emergency text recognition.

[0197] When optimizing the key question set, post-dispatch guidance was fully considered, which means linking the optimization of the key question set with the post-dispatch guidance. This allows for the sequential output of the key question set, while also enriching its content. This enables targeted and rapid collection of emergency scene information. Based on the correlation between the key question set and the guidance content in the post-dispatch guidance, certain strategies are used for filtering to enhance the correspondence between the key question set and the post-dispatch guidance, making the two more closely related and with higher content matching.

[0198] This disclosure proposes an encoder-decoder that fits a network model of the input text and the target text. The encoder uses an Encoders and Decoders structure and inserts recurrent neurons in the middle of the structure, which can retain short text information in the preceding text and output the sequential key questions required for the emergency response plan.

[0199] This disclosed embodiment of the emergency intelligent assistance system provides clear and easy-to-follow instructions before the dispatcher's assessment of the caller's condition and remote guidance arrives, enabling on-site personnel to take appropriate self-rescue and mutual rescue measures to stabilize the patient's (i.e., the emergency recipient's) condition and advance the rescue of the (i.e., the emergency recipient's) condition.

[0200] The method proposed in this disclosure can accurately assess the patient's condition at the scene and provide correct rescue guidance. This method collects first aid knowledge from multiple sources and uses natural language processing to intelligently extract key question sets from the contingency plan database, effectively linking these key question sets with post-dispatch guidance.

[0201] Using word vectors, sentence vectors, and positional encoding as weighted inputs to a BERT network for part-of-speech tagging more closely approximates the textual information expressed in emergency situations. Furthermore, by converting hypothetical statements into selective questions during the tagging process, it is more effective in rapidly collecting information from the emergency scene.

[0202] Based on the specific characteristics of emergency medical texts, this embodiment of the disclosure uses a method of splicing the beginning and end of the text, which, compared to the commonly used truncation method, can obtain more balanced information in the text.

[0203] Inserting recurrent neurons between the Encoders and Decoders structures allows for efficient use of information from the encoder layer, retention of valid encoder layer information, and filtering of invalid encoder layer information.

[0204] like Figure 11 As shown, this disclosure provides an emergency medical information processing device, including:

[0205] Error correction module 110 is used to correct errors in the first emergency text to obtain the second emergency text;

[0206] The module 120 is used to obtain a text of key emergency questions based on the second emergency text and the target model; wherein the target model is trained using the third emergency text and emergency expert knowledge text.

[0207] The emergency information processing device can be any type of electronic device.

[0208] In one embodiment, the error correction module and the obtaining module can be program modules; after the program module is executed by the processor, it can perform the above operations.

[0209] In another embodiment, the error correction module and the obtaining module may be hardware or software modules, which may include, but are not limited to, various programmable arrays; the programmable arrays include, but are not limited to, field-programmable arrays and / or complex programmable arrays.

[0210] In some embodiments, the error correction module and the obtaining module may be pure hardware modules; the pure hardware modules include, but are not limited to, application-specific integrated circuits.

[0211] In one embodiment, the error correction module is specifically used to segment the first emergency text to obtain a first segmentation sequence; replace erroneous words and / or non-medical words in the first segmentation sequence with medical words to obtain a second segmentation sequence, and obtain the second emergency text based on the second segmentation sequence.

[0212] In one embodiment, the error correction module is specifically used to perform an embedding operation on the second word segmentation sequence to obtain a first embedded sequence; when the length of the first embedded sequence exceeds a preset length, the first embedded sequence is truncated and the last subsequence is spliced ​​together to obtain a second embedded sequence equal to the preset length; the emergency critical issue text is obtained based on the second embedded sequence.

[0213] In one embodiment, the obtaining module is specifically used to construct word vectors and sentence vectors corresponding to the second word segmentation sequence; determine position vectors based on the positions of preset emergency keywords in the second word segmentation sequence; and obtain the emergency key question text based on the encoding information obtained by weighting the word vectors, sentence vectors, and position vectors using an attention mechanism.

[0214] In one embodiment, the target model includes: a preprocessing layer;

[0215] The error correction module is specifically used to input the first emergency text into the preprocessing layer for error correction, thereby obtaining the second emergency text.

[0216] In one embodiment, the target model further includes: an information processing layer; the information processing layer is connected to the back end of the preprocessing layer;

[0217] The obtaining module is specifically used to input the second emergency text into the information processing layer to obtain the emergency key question text output by the information processing layer.

[0218] In one embodiment, the critical first aid question text includes at least one optional question related to first aid.

[0219] In one embodiment, the critical emergency text includes a question text and a sequence text;

[0220] The selective questions described in the question text include: initial selective questions, alternative selective questions, and cutoff selective questions;

[0221] The sequential text is used to output the (m+1)th selective question based on the answer to the m-th selective question; where m is less than M, and M is the total number of selective questions.

[0222] In one embodiment, the set of critical first aid questions includes: initiation selection questions, alternative selection questions, and termination selection questions related to first aid;

[0223] The device further includes:

[0224] The first output module is used to output the initial selective questions based on the emergency critical questions text;

[0225] The second output module is used to select the (n+1)th selective problem to be output from the candidate selective problems and the cutoff selective problems based on the response result of the nth output selective problem, until the cutoff selective problem or all selective problems in the emergency problem set have been output; wherein, n is any positive integer;

[0226] The text module is used to obtain the fourth emergency text based on the response results of the output selective questions.

[0227] In some embodiments, the apparatus further includes:

[0228] The third output module is used to output first aid guidance text based on the fourth first aid text.

[0229] In some embodiments, the third output module is configured to determine the emergency type reflecting the emergency scenario based on the fourth emergency text; and to determine the emergency guidance text based on a pre-established mapping relationship between the emergency type and the guidance content.

[0230] In some embodiments, before correcting the first emergency text to obtain the second emergency text, the apparatus further includes:

[0231] The matching and filtering module is used to match and filter the first emergency text with the emergency guidance set to obtain the target emergency text;

[0232] A construction module is used to construct candidate first aid question texts based on the target emergency text; wherein the first aid key question texts are derived from the candidate first aid question texts.

[0233] like Figure 12 As shown, this disclosure provides a target model training apparatus, the apparatus comprising:

[0234] Module 210 is used to acquire third-party emergency text and emergency expert knowledge text respectively;

[0235] Training module 220 is used to train a preset network using the third first aid text and the first aid expert knowledge text to obtain a target model;

[0236] The target model is used at least to determine the first emergency text after error correction, the key issue text of the emergency text, and / or the emergency guidance text.

[0237] In some embodiments, the acquisition module 210 and the training module 220 may be program modules; after being executed by the processor, the program modules can perform the above operations.

[0238] In another embodiment, the acquisition module 210 and the training module 220 may be hardware or software modules, which may include, but are not limited to, various programmable arrays; the programmable arrays include, but are not limited to, field-programmable arrays and / or complex programmable arrays.

[0239] In some embodiments, the acquisition module 210 and the training module 220 may be pure hardware modules; the pure hardware modules include, but are not limited to, application-specific integrated circuits.

[0240] In some embodiments, the training module 220 can be used to construct a training text team when there is a third first aid text and a first aid expert knowledge text with a similarity that reaches a preset threshold; and train the preset model according to the training text team, so that the preset model can replace erroneous words and / or non-medical professional words in the third first aid text in the training text team with medical professional words.

[0241] In some embodiments, the training module 220 is further configured to construct input text and selective questions about the symptoms of the person receiving first aid, based on the distribution of hypothetical conjunctions in the hypothetical sentences when the first aid expert knowledge text contains hypothetical sentences.

[0242] like Figure 13 As shown, this disclosure provides an electronic device, the electronic device comprising:

[0243] Memory;

[0244] A processor, connected to the memory, is configured to implement the information processing method provided in any of the foregoing embodiments by executing computer-executable instructions stored in the memory, such as performing... Figures 1 to 7 Any text processing method and / or model training method shown.

[0245] The electronic device may be a terminal device and / or a server in a service platform.

[0246] like Figure 13 As shown, the electronic device may also include a network interface, which can be used to interact with peer devices via a network.

[0247] This disclosure provides a computer storage medium storing computer-executable instructions; when executed by a processor, these computer-executable instructions can implement the text processing method provided in any of the foregoing embodiments, for example, performing... Figures 1 to 7 Any text processing method and / or model training method shown.

[0248] The computer's storage medium is a non-instantaneous storage medium.

[0249] The technical solutions described in the embodiments of this disclosure can be combined arbitrarily without conflict.

[0250] In the several embodiments provided in this disclosure, it should be understood that the disclosed methods and smart devices can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods, such as: multiple units or components may be combined, or integrated into another system, or some features may be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the various components shown or discussed may be through some interfaces, and the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0251] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the units may be selected to achieve the purpose of this embodiment according to actual needs.

[0252] In addition, each functional unit in the various embodiments of this disclosure can be integrated into a second processing unit, or each unit can be a separate unit, or N or more units can be integrated into a unit; the integrated unit can be implemented in hardware or in the form of hardware plus software functional units.

[0253] The above description is merely a specific embodiment of this disclosure, but the scope of protection of this disclosure is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A method for processing emergency medical information, characterized in that, include: The second emergency response text is obtained by correcting errors in the first emergency response text. Based on the second emergency text and the target model, a text of key emergency questions is obtained; wherein, the target model is obtained by training a preset network based on a training text team; the training text team is constructed using a third emergency text and an emergency expert knowledge text, when there exists a third emergency text and an emergency expert knowledge text whose similarity reaches a preset threshold; the preset network can replace erroneous words and / or non-medical professional words in the third emergency text in the training text team with medical professional words; The step of correcting errors in the first emergency text to obtain the second emergency text includes: The first emergency text is segmented using the target model to obtain a first segmentation sequence; Replace erroneous words and / or non-medical terms in the first word segmentation sequence with medical terms to obtain a second word segmentation sequence, and obtain the second emergency text based on the second word segmentation sequence; The process of obtaining key emergency response question text based on the second emergency response text and the target model includes: Based on the target model, the second word segmentation sequence is embedded to obtain the first embedded sequence; When the length of the first embedded sequence exceeds the preset length, the first embedded sequence is truncated and the tail subsequence is spliced ​​together to obtain a second embedded sequence equal to the preset length. The emergency response key question text is obtained based on the second embedded sequence.

2. The method according to claim 1, characterized in that, The process of obtaining key emergency response question text based on the second emergency response text and the target model includes: Construct word vectors and sentence vectors corresponding to the second word segmentation sequence; Determine the position vector based on the position of the preset emergency keywords in the second word segmentation sequence; The key emergency response text is obtained by weighting the word vectors, sentence vectors, and position vectors using an attention mechanism.

3. The method according to claim 1, characterized in that, The target model includes: a preprocessing layer; The step of correcting errors in the first emergency text to obtain the second emergency text includes: The first emergency text is input into the preprocessing layer for error correction to obtain the second emergency text.

4. The method according to claim 3, characterized in that, The target model further includes: an information processing layer; the information processing layer is connected to the back end of the preprocessing layer; The process of obtaining key emergency response question text based on the second emergency response text and the target model includes: The second emergency text is input into the information processing layer to obtain the emergency key question text output by the information processing layer.

5. The method according to claim 1, characterized in that, The first aid key questions text includes at least one optional question related to first aid.

6. The method according to claim 5, characterized in that, The critical emergency response text includes both question text and sequence text; The selective questions described in the question text include: initial selective questions, alternative selective questions, and cutoff selective questions; The sequential text is used to output the (m+1)th selective question based on the answer to the m-th selective question; where m is less than M, and M is the total number of selective questions.

7. The method according to claim 5 or 6, characterized in that, The critical questions set for first aid includes: initiation selection questions, alternative selection questions, and termination selection questions related to first aid; The method further includes: Based on the text of critical emergency questions, output the initial selective questions; Based on the response to the nth output selective problem, select the (n+1)th selective problem to be output from the candidate selective problems and the cutoff selective problems, until the cutoff selective problem or all selective problems in the emergency set have been output; where n is any positive integer; Based on the response results of the selected questions, the fourth emergency text is obtained.

8. The method according to claim 7, characterized in that, The method further includes: Based on the fourth emergency text, output the emergency guidance text.

9. The method according to claim 8, characterized in that, The step of outputting first aid guidance text based on the fourth first aid text includes: Based on the fourth emergency text, determine the type of emergency response that reflects the emergency scenario; The first aid guidance text is determined based on the pre-established mapping relationship between the first aid type and the guidance content.

10. The method according to claim 1, characterized in that, Before correcting the first emergency text to obtain the second emergency text, the method further includes: The first emergency text is matched and filtered with the emergency guidance set to obtain the target emergency text; Based on the target first aid text, candidate first aid question texts are constructed; wherein, the first aid key question texts are derived from the candidate first aid question texts.

11. A model training method, characterized in that, The method includes: Obtain third-party emergency text and emergency expert knowledge text; train a preset network using the third-party emergency text and the emergency expert knowledge text to obtain a target model; The step of training a preset network using the third first aid text and the first aid expert knowledge text to obtain a target model includes: When the similarity between a third-party emergency text and an emergency expert knowledge text reaches a preset threshold, a training text team is constructed. The preset network is trained based on the training text set, enabling the preset network to replace erroneous words and / or non-medical terms in the third first aid text set with medical terms; wherein, the target model is at least used to determine the first aid key issue text and / or first aid guidance text; the first aid key issue text is obtained based on the second first aid text and the target model; the second first aid text is obtained by correcting errors in the first first aid text; The second emergency text is obtained based on a second word segmentation sequence, which is obtained by replacing erroneous words and / or non-medical words in the first word segmentation sequence with medical words. The first word segmentation sequence is obtained by segmenting the first emergency text using the target model. The emergency critical issue text is obtained based on a second embedding sequence. The second embedding sequence is obtained by splicing together the first embedding sequence by truncating the first embedding sequence's front and rear sub-sequences when the length of the first embedding sequence exceeds a preset length. The length of the second embedding sequence is the preset length. The first embedding sequence is obtained by embedding the second word segmentation sequence based on the target model.

12. The method according to claim 11, characterized in that, The step of training a preset network using the third first aid text and the first aid expert knowledge text to obtain a target model includes: When the first aid expert knowledge text contains hypothetical sentences, the input text and selective questions inquiring about the symptoms of the first aid recipient are constructed based on the distribution of hypothetical conjunctions in the hypothetical sentences.

13. An emergency medical information processing device, characterized in that, include: The error correction module is used to correct errors in the first emergency text to obtain the second emergency text; The module is used to obtain key emergency response question text based on the second emergency text and the target model; wherein, the target model is obtained by training a preset network based on a training text team; the training text team is constructed using a third emergency text and emergency expert knowledge text when there is a third emergency text and an emergency expert knowledge text whose similarity reaches a preset threshold; the preset network can replace erroneous words and / or non-medical professional words in the third emergency text in the training text team with medical professional words; The error correction module is used to segment the first emergency text using the target model to obtain a first segmentation sequence; replace erroneous words and / or non-medical words in the first segmentation sequence with medical words to obtain a second segmentation sequence; and obtain the second emergency text based on the second segmentation sequence. The error correction module is used to perform an embedding operation on the second word segmentation sequence based on the target model to obtain a first embedding sequence; when the length of the first embedding sequence exceeds a preset length, the first embedding sequence is truncated and the last subsequence is spliced ​​together to obtain a second embedding sequence equal to the preset length; the emergency critical issue text is obtained according to the second embedding sequence.

14. A target model training device, characterized in that, The device includes: The acquisition module is used to acquire the third-party emergency text and the emergency expert knowledge text, respectively. The training module is used to train a preset network using the third first aid text and the first aid expert knowledge text to obtain the target model; The training module is used to construct a training text team when the similarity between a third first aid text and a first aid expert knowledge text reaches a preset threshold; and to train the preset network based on the training text team, so that the preset network can replace erroneous words and / or non-medical professional words in the third first aid text in the training text team with medical professional words; The target model is used at least to determine the critical first aid question text and / or first aid guidance text; the critical first aid question text is obtained based on the second first aid text and the target model; the second first aid text is obtained by correcting errors in the first first aid text. The second emergency text is obtained based on a second word segmentation sequence, which is obtained by replacing erroneous words and / or non-medical words in the first word segmentation sequence with medical words. The first word segmentation sequence is obtained by segmenting the first emergency text using the target model. The emergency critical issue text is obtained based on a second embedding sequence. The second embedding sequence is obtained by splicing together the first embedding sequence by truncating the first embedding sequence's front and rear sub-sequences when the length of the first embedding sequence exceeds a preset length. The length of the second embedding sequence is the preset length. The first embedding sequence is obtained by embedding the second word segmentation sequence based on the target model.

15. An electronic device, characterized in that, The electronic device includes: Memory; A processor, connected to the memory, is configured to implement the method provided by any one of claims 1 to 10, 11 to 12 by executing computer-executable instructions stored on the memory.

16. A computer storage medium, characterized in that, The computer storage medium stores computer-executable instructions; when executed by a processor, the computer-executable instructions can implement the method provided by any one of claims 1 to 10 and 11 to 12.

Images