An interactive triage terminal based on AI analysis

Abstract

The application relates to the field of data processing, in particular to an interactive triage terminal based on AI analysis, target entity extraction is carried out on collected symptom description information, so that symptom entities, sign entities and identity information are obtained. Then, the target entity is evaluated based on a pre-constructed triage database, so that it is determined whether there are symptoms and sign descriptions inconsistent with the identity information or there are unmatched or low-probability symptom and sign combinations in the description information input by the user. If the above conditions exist, the confidence is low. When the confidence is low, the application will construct a guide text based on the target entity and the triage database, so as to guide the user to express correctly. When correct expression with high confidence is obtained, it is input into the triage model to perform AI analysis, and a triage result can be obtained. The application can effectively avoid the triage error caused by incorrect expression of the user.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of data processing, and particularly relates to an interactive triage terminal based on AI analysis. BACKGROUND

[0002] Triage is a method used in medical environments to assess and prioritize patients, mainly used in emergency departments, disaster response scenarios, and situations with limited resources. Its main role is to ensure that patients who need emergency treatment the most can receive timely medical services, while reasonably allocating limited medical resources.

[0003] In the prior art, triage is mainly performed by personnel or a triage system. The existing triage system relies on user symptom information description, but in actual interaction, many users cannot accurately describe their own symptom information. Through entity extraction and recognition of user language, it is not like artificial correction of symptom description, and misclassification often occurs. SUMMARY

[0004] Therefore, the present application aims to provide an interactive triage terminal based on AI analysis to solve the problems in the background art.

[0005] In order to achieve the above-mentioned purpose, the present application adopts the following technical solutions:

[0006] The interactive triage terminal based on AI analysis comprises:

[0007] An acquisition module for acquiring current symptom description information of a user;

[0008] An entity extraction module for extracting a plurality of target entities from the current symptom description information, wherein the plurality of target entities include symptom entities, physical sign entities, and identity information;

[0009] An analysis module for analyzing the plurality of target entities based on a pre-constructed triage database to obtain the confidence of the plurality of target entities, wherein the triage database includes entity combinations corresponding to a plurality of triage results;

[0010] A result judgment module for inputting the plurality of target entities into a pre-constructed triage model when the confidence of the plurality of target entities is greater than or equal to a preset confidence threshold to obtain a triage result; and constructing a guide text based on the triage database and the plurality of target entities and sending the guide text to the user when the confidence of the plurality of target entities is less than the preset confidence threshold;

[0011] The loop module is used to take the symptom description information as the current symptom description information when it receives the user's symptom description information, and return to the entity extraction module until there are multiple target entities with a confidence level greater than or equal to a preset confidence level threshold. Then, the multiple target entities are input into the pre-built triage model to obtain the triage result. The symptom description information is generated based on the guidance text.

[0012] In one embodiment of this application, the method for constructing the triage database includes:

[0013] Retrieve historical triage records;

[0014] Extract the symptom entities, sign entities, identity information, and triage results recorded during multiple triage visits from the historical triage records. The symptom entities and the sign entities are then standardized to obtain standard symptom entities. and standard physical characteristics The standard symptom entity and the standard sign entity both include body parts and characteristic descriptions;

[0015] Based on identity information, the symptom entities, sign entities, and triage results recorded during multiple triage visits are divided into data units; and based on the triage results, the symptom entities and sign entities in each data unit are further divided into symptom data subunits and sign data subunits; and each data unit is directly divided based on the triage results to obtain symptom-sign data subunits.

[0016] Calculate each symptom entity in the symptom data subunit probability of occurrence Each vital sign entity in the vital sign data subunit probability of occurrence ,in, For entity serial number;

[0017] Extract symptom-sign entity combinations from the same time period from the symptom-sign data subunit. And calculate each symptom-sign entity combination probability of occurrence ;

[0018] Symptom entities with a probability less than a preset probability threshold Physical characteristics and symptom-sign combination Remove from the data unit and update the reference symptom entity in the symptom data subunit. probability of occurrence The reference vital sign entity in the vital sign data subunit probability of occurrence and updating the reference symptom-sign entity combination in the symptom-sign data subunit. probability of occurrence This allows access to the triage database.

[0019] In one embodiment of this application, the multiple target entities are analyzed based on a pre-built triage database to obtain the confidence levels of the multiple target entities, including:

[0020] The target data unit in the triage database is determined based on the user's identity information;

[0021] The symptom entities in the current description information are compared with reference symptom entities in multiple symptom data sub-units; the sign entities in the current description information are compared with reference sign entities in multiple sign data sub-units; and the symptom-sign entity combination in the current description information is compared with the reference symptom-sign entity combination in the symptom-sign data sub-unit.

[0022] If there is no state entity consistent with the reference symptom entity in the current description information, or if there is no sign entity consistent with the reference sign entity, the confidence of multiple target entities is set to zero, and it is determined that the current description information does not match the user's identity information.

[0023] The current description information contains entities similar to the reference symptom entity. Consistent state entity and exists with the reference trait entity When the physical signs are consistent, the probability of occurrence is based on the same triage result. , and Perform a weighted summation to obtain the confidence scores of multiple target entities. .

[0024] In one embodiment of this application, constructing guidance text based on the triage database and the plurality of target entities includes:

[0025] When the confidence level of the multiple target entities is zero, guide text is constructed to indicate that the symptoms or signs do not match the identity information;

[0026] When the confidence levels of the multiple target entities are not zero, a confirmation guidance text is constructed based on the symptom and sign entities among the target entities; and a reference symptom entity in the target data unit is referenced based on the symptom entity or sign entity. Or refer to physical signs Rate And based on the top N reference symptom entities with the highest scores Or reference sign entities Constructing a reference guide text, wherein the score The mathematical expression is:

[0027]

[0028] In the formula, is the first weight, is the second weight, is the cosine similarity of the symptom entity and the reference symptom entity, is the cosine similarity of the sign entity and the reference sign entity.

[0029] In an embodiment of the present application, after the guide text is sent to the user, it further includes:

[0030] When the confirmation description is received from the user, the confidence of the plurality of target entities is set to a target value greater than or equal to a preset confidence threshold.

[0031] In an embodiment of the present application, the reference symptom entity The mathematical expression of the appearance probability The appearance probability of the reference sign entity The appearance probability of the reference symptom-sign entity combination

[0032]

[0033]

[0034]

[0035] In the formula, is the total number of reference symptom entities in the symptom data subunit, is the total number of reference sign entities in the sign data subunit, is the total number of reference symptom-sign entity combinations in the symptom-sign data subunit.

[0036] In an embodiment of the present application, the method for constructing the triage model includes:

[0037] Obtaining a triage record;

[0038] Extracting a plurality of triage sample data from the triage record, wherein the triage sample data includes standard symptom description entities, sign conclusion entities, identity vector information, and triage conclusions;

[0039] ​​​The training data sample is constructed with the triage conclusion as a label to obtain a training data set;

[0040] The RNN convolutional neural network is trained based on the training data set to minimize a set multi-class cross-entropy loss function, so as to obtain a triage model.

[0041] In an embodiment of the present application, a plurality of triage sample data are extracted from the triage records, including:

[0042] Symptom information, sign information and identity information in each triage record are extracted;

[0043] The symptom information is standardized and vectorized to obtain a standard symptom description entity; the sign information is labeled and vectorized to obtain a sign conclusion entity; the triage conclusion is standardized and vectorized to obtain a triage conclusion; and the identity information is vectorized to obtain identity vector information.

[0044] In an embodiment of the present application, the multi-class cross-entropy loss function has a mathematical expression as follows:

[0045]

[0046] In the formula, is a sample number, is a sample serial number, is a category number, is a category serial number, is a label, is a probability that a model predicted sample belongs to a category.

[0047] In an embodiment of the present application, current symptom description information of a user is obtained, including:

[0048] When a triage starting command from an external source is received, a guide voice or initial guide text is outputted;

[0049] When a voice reply or text reply from an external source is received, the voice reply or the text reply is subjected to preliminary entity verification, and when the voice reply or the text reply contains a predetermined entity, the voice reply or the text reply is taken as symptom description information.

[0050] ​The beneficial effects of the present application are: the interactive triage terminal based on AI analysis of the present application extracts target entities from the collected symptom description information, thereby obtaining symptom entities, sign entities and identity information. Then, based on the pre-constructed triage database, the target entities are evaluated for confidence, thereby determining whether there are symptoms, sign descriptions inconsistent with the identity information, or there are unmatched or low probability symptom and sign combinations in the user input description information. If the above conditions exist, the confidence will be low. When the confidence is low, the present application will construct a guide text based on the target entity and the triage database, thereby guiding the user to express correctly. When the correct expression with high confidence is obtained, it is input into the triage model to perform AI analysis, and the triage result can be obtained. The present application can effectively avoid the triage error caused by the user's incorrect expression. BRIEF DESCRIPTION OF DRAWINGS

[0051] The present application will be further described below in conjunction with the drawings and embodiments:

[0052] Figure 1 is the application scenario of the interactive triage terminal based on AI analysis shown in an embodiment of the present application;

[0053] Figure 2 is a structural diagram of an interactive triage terminal based on AI analysis shown in an embodiment of the present application;

[0054] Figure 3 is the construction process of the collaborative association database in an embodiment of the present application;

[0055] Figure 4 is a reaction time calculation process diagram in an embodiment of the present application;

[0056] Figure 5 is a process diagram of the network based on collaborative verification and detection in the present application. DETAILED DESCRIPTION

[0057] The embodiments of the present application will be described below through specific concrete examples, and those skilled in the art can easily understand other advantages and effects of the present application from the disclosure of the present specification. The present application can also be implemented or applied by different specific embodiments, and the details in the present specification can be modified or changed based on different views and applications without departing from the spirit of the present application. It should be noted that the following embodiments and features in the embodiments can be combined with each other without conflict.

[0058] It is to be noted that the drawings provided in the following embodiments only schematically illustrate the basic concepts of the present application, and only the layers related to the present application are shown in the drawings, rather than being drawn according to the number, shape and size of the layers in actual implementation. The actual implementation of each layer can be a random change, and the layer layout pattern can also be more complex.

[0059] In the following description, a large number of details are discussed to provide a more thorough explanation of the embodiments of the present application, however, it is obvious to those skilled in the art that the embodiments of the present application can be implemented without these specific details.

[0060] Figure 1 is a hardware schematic diagram of an interactive triage terminal based on AI analysis shown in an embodiment of the present application, as shown in Figure 1 The interactive triage terminal in the present application is installed on the ground through the base. It is generally used in the waiting area outside the hospital clinic. The upper part of the terminal is provided with a touch screen 110 as an information input / output part. In addition, the terminal is provided with a microphone 120 and a loudspeaker 130. The microphone 120 is used to collect the voice data of the user, and the loudspeaker 130 is used for voice interaction with the user. In addition, a computer is arranged inside the terminal for executing the interactive logic in the present application.

[0061] Figure 2 is a logic structure diagram of an interactive triage terminal based on AI analysis shown in an embodiment of the present application, as shown in Figure 2 The interactive triage terminal based on AI analysis of the present embodiment comprises:

[0062] The acquisition module 210 is configured to acquire the current symptom description information of the user.

[0063] The current symptom description information in the present application can be the text directly input by the user, or the audio data recorded by the user through the microphone. If it is audio data, the automatic speech recognition (ASR) technology needs to be used to convert the voice data into text.

[0064] Specifically, the present application uses a guided process to acquire the current symptom description information of the user, and the process comprises:

[0065] S211, when receiving the triage starting command from the outside, outputting a guided voice or an initial guided text;

[0066] If the user touches the corresponding area on the terminal touch screen, the triage starting command can be input. At the start, in order to avoid the user not knowing how to input the current symptom description information, the application pre-sets an initial guide text as a template to guide the user to correctly input the description information.

[0067] For example, the initial guide text can be: "Hello, please input your triage information by speaking or keyboard input, refer to the example XXX". At the same time, the terminal opens the microphone and opens the virtual keyboard in the display screen. The user can input by speaking or virtual keyboard input.

[0068] S212, when receiving the voice reply or text reply from the outside, performing preliminary entity verification on the voice reply or the text reply, and when the voice reply or the text reply contains a predetermined entity, taking the voice reply or the text reply as the symptom description information.

[0069] After getting the reply, the reply text needs to be preliminarily verified by using a pre-constructed entity library. If the reply contains a predetermined entity, it can be considered that the user's input contains valid information, and the next step can be processed.

[0070] Specifically, when verifying, the reply text needs to be segmented. The segmentation algorithm can be forward maximum matching, reverse maximum matching, bidirectional maximum matching, etc. The obtained word is vectorized and encoded, and is quickly matched with the entity vector template in the entity library. The entity templates in the entity library include common symptom or sign description entities such as "fever", "headache", "high blood pressure", etc.

[0071] The entity extraction module 220 is configured to extract a plurality of target entities from the current symptom description information, wherein the plurality of target entities include symptom entities, sign entities, and identity information.

[0072] The process of target entity extraction is similar to the foregoing, that is, segmentation and matching, which will not be described here.

[0073] The analysis module 230 is configured to analyze the plurality of target entities based on a pre-constructed triage database to obtain a confidence degree of the plurality of target entities, wherein the triage database includes entity combinations corresponding to a plurality of triage results.

[0074] In the application, in order to verify whether the extracted target entities are reasonable, the triage database is used for verification. The big data of past triage is used to verify the plurality of target entities.

[0075] When users make mistakes in their descriptions, they often exhibit common symptoms that do not match their identity information, or more subtle symptoms that do not match their physical signs. For example, a female user may describe having prostate pain, or a male user may describe having a fever, but through preliminary physical examination before initial triage, the body temperature is normal. At this time, big data needs to be used to verify this information. Find entities that may have abnormalities, and further guide and confirm these abnormal entities.

[0076] The construction method of the triage database in this application includes:

[0077] (1) Obtain historical triage records;

[0078] Triage records are an important part of medical documents, which record the key information of patients when they first contact medical services. This record helps medical staff quickly assess the urgency and health of patients, and accordingly arrange appropriate care paths.

[0079] Generally, triage records contain the following information:

[0080] Patient basic information, such as name, age or date of birth, gender, contact information, and visit time;

[0081] Chief complaint (Chief Complaint): The main symptoms or problems described by the patient, such as "chest pain" or "dyspnea"

[0082] Symptoms and signs: detailed symptom descriptions, including but not limited to the nature, location, duration and intensity of pain; fever, dyspnea and other significant signs.

[0083] Vital signs, such as blood pressure, heart rate, respiratory rate, body temperature, and blood oxygen saturation.

[0084] Medical history, such as past medical history, such as hypertension, diabetes, etc.

[0085] Surgical history: past surgical experience.

[0086] Initial assessment results, a quick assessment based on the patient's symptoms and signs, which may include using specific triage systems (such as CTAS, ESI) to determine the severity of the illness.

[0087] Triage decision, based on all the above information, the triage conclusion is made, which determines whether the patient needs immediate treatment or can wait, and should be sent to which department or receive which type of care.

[0088] In this application, user information, user-reported symptoms, and vital sign data from pre-triage rapid testing are mainly extracted.

[0089] (2) Extracting the symptom entities, sign entities, identity information and triage results recorded in multiple triage from the historical triage records , and standardizing the symptom entities and the sign entities to obtain standard symptom entities and standard sign entities , wherein the standard symptom entities and the standard sign entities both include body parts and feature descriptions;

[0090] When constructing the database, in order to accurately match in the subsequent process, the application standardizes the symptom entities and the sign entities, and converts different types of literal descriptions into a standard format. That is, body part + feature description; for example, "head + pain", "chest + difficulty breathing"; for sign data, there are numbers of body indicators, in order to simplify these numbers, the same is converted into a conclusive word, for example, "body temperature + high", "blood pressure + high".

[0091] In addition, the triage results can also be standardized, for example, to the department + emergency level.

[0092] (3) Based on the identity information, the symptom entities, the sign entities and the triage results recorded in multiple triage are divided to obtain data units; and based on the triage results, the symptom entities and the sign entities in each data unit are divided respectively to obtain symptom data subunits and sign data subunits; and based on the triage results, each data unit is directly divided to obtain symptom-sign data subunits;

[0093] Firstly, the above data is divided into multiple data units based on the identity information. When dividing, since the identity information is multi-dimensional, the obtained data units are also multi-dimensional. For example, based on three age groups, three data units are divided, and based on gender, two data units are divided. A total of five data units are obtained. The identity information of other dimensions is the same, and will not be repeated here.

[0094] Then, based on the triage results, the data units are further divided to obtain symptom data subunits containing only symptom entities, sign data subunits containing only signs, and symptom-sign data subunits containing both sign entities and symptom entities.

[0095] When dividing the symptom-sign data subunits, the symptom entities and the sign entities in each triage record need to be extracted, arranged and combined to obtain combinations containing only one symptom entity and sign entity.

[0096] (4) Calculating the occurrence probability of each symptom entity in the symptom data subunit , the appearance probability of each symptom entity in the symptom data subunit , wherein, is the entity serial number;

[0097] (5) extracting the symptom-entity entity combination of the same time from the symptom-sign data subunit , and calculating the appearance probability of each symptom-entity entity combination ;

[0098] The probability here indicates the appearance probability of different symptom entities or sign entities in the triage result corresponding to the identity information. In addition, in order to further verify the probability of the simultaneous appearance of symptoms and signs, the appearance probability of each symptom-entity entity combination

[0099] (6) removing the symptom entity , the sign entity and the symptom-entity entity combination with a probability less than a preset probability threshold from the data unit, and updating the appearance probability of the reference symptom entity in the symptom data subunit , the appearance probability of the reference sign entity in the sign data subunit , and the appearance probability of the reference symptom-entity entity combination in the symptom-sign data subunit to obtain a triage database.

[0100] In the present application, the entities or entity combinations with lower probability that are accidentally removed can reflect the reasonable degree of multiple entities or entity combinations in a certain user information group when different triage results appear. The higher the probability, the higher the reasonable degree of appearance.

[0101] , the mathematical expression of the appearance probability of the reference symptom entity , the appearance probability of the reference sign entity , and the appearance probability of the reference symptom-entity entity combination

[0102]

[0103]

[0104] ​​​​​​​​

[0105] wherein, is the total number of reference symptom entities in the symptom data sub-unit, is the total number of reference sign entities in the sign data sub-unit, is the total number of reference symptom-sign entity combinations in the symptom-sign data sub-unit.

[0106] Finally, the above entities and entity combinations also need to be converted into vectors. In this application, Word2Vec encoding is used to convert the above entities and entity combinations into vectors and stored in the database.

[0107] After the above triage database is constructed, the process of using the above triage database to analyze the confidence of multiple target entities is as follows:

[0108] S231, determining the target data unit in the triage database based on the identity information of the user;

[0109] The target data unit records all the corresponding symptom entities, sign entities, symptom and sign combinations and their respective probabilities in all historical triage conclusions.

[0110] S232, comparing the symptom entities in the current description information with the reference symptom entities in multiple symptom data sub-units, comparing the sign entities in the current description information with the reference sign entities in multiple sign data sub-units, and comparing the symptom-sign entity combinations in the current description information with the reference symptom-sign entity combinations in the symptom-sign data sub-unit;

[0111] Before comparison, multiple entity combinations also need to be constructed based on the symptom entities and sign entities extracted in the foregoing.

[0112] For example, from the current symptom description information "I feel nasal congestion, cold and headache", and the sign description "body temperature 38.2, heart rate 92", three symptom entities are extracted, nasal congestion, body cold, and head pain; sign entities: high body temperature, moderate heart rate.

[0113] Six entity combinations can be obtained, which are:

[0114] Nasal congestion, high body temperature;

[0115] Body cold, high body temperature;

[0116] Headache, high body temperature;

[0117] Nasal congestion, moderate heart rate;

[0118] Body cold, moderate heart rate;

[0119] Head + pain, heart beat moderate;

[0120] By comparing the six groups of entity combinations with the reference symptom-sign entity combinations in the database, the rationality of the occurrence of the combinations can be judged.

[0121] S233, when there is no state entity consistent with the reference symptom entity or no sign entity consistent with the reference sign entity in the current description information, the confidence of the multiple target entities is set to zero, and it is determined that the current description information is inconsistent with the identity information of the user;

[0122] If there is no entity in the database in the current description information, it means that the described symptoms or signs are probably inconsistent with the identity information. At this time, the confidence of the multiple target entities is set to zero.

[0123] S234, when there is a state entity consistent with the reference symptom entity and a sign entity consistent with the reference sign entity in the current description information, the confidence of the multiple target entities is obtained by weighted summation based on the occurrence probability 、 and corresponding to the same triage result. .

[0124] If there is an entity in the database in the current description information, the rationality of the current symptom description information is verified by calculating the confidence.

[0125] Since the rationality of the occurrence of multiple target entities in the same information description needs to be judged, the probabilities of the corresponding entities and entity combinations need to be found from the symptom data subunit, the sign data subunit, and the symptom-sign combination data subunit corresponding to the same triage result, i.e. 、 and . Then, the confidence is obtained by weighted summation; the mathematical expression of the confidence is:

[0126]

[0127] wherein, is the first confidence weight, is the second confidence weight, is the third confidence weight.

[0128] ​The result judgment module 240 is configured to input the plurality of target entities into a pre-constructed triage model to obtain a triage result when the confidence of the plurality of target entities is greater than or equal to a preset confidence threshold; and construct a guide text based on the triage database and the plurality of target entities and send the guide text to the user when the confidence of the plurality of target entities is less than the preset confidence threshold.

[0129] The confidence describes the reasonable degree (comprehensive probability) of the plurality of target entities appearing in the same description information, and thus if the confidence is relatively high, it indicates that the description of the user conforms to the historical triage rule, and in this case, the plurality of target entities extracted are input into the triage model, and a triage result can be obtained based on AI analysis.

[0130] If the confidence is relatively low or even 0, it indicates that the description of the user does not conform to the historical triage rule or the probability of conforming to the rule is low. In this case, further confirmation and guidance are needed, and the process is as follows:

[0131] S241, constructing a guide text for prompting that the symptoms or signs do not match the identity information when the confidence of the plurality of target entities is 0.

[0132] The confidence of 0 only occurs when the described symptoms or signs do not match the identity information. In this case, the main purpose is to prompt the user that the symptom description is incorrect, for example, generating a guide text:

[0133] “Symptom description may be incorrect, please re-input or contact the staff.”

[0134] S242, constructing a confirmation guide text based on the symptom entity and the sign entity in the target entity when the confidence of the plurality of target entities is not 0; and scoring the reference symptom entity or the reference sign entity in the target data unit based on the symptom entity or the sign entity, and constructing a reference guide text based on the top N reference symptom entities or the reference sign entities with the highest scores, wherein the score is mathematically expressed as:

[0135]

[0136] In the formula, is a first weight, is a second weight, is a cosine similarity between the symptom entity and the reference symptom entity, is a cosine similarity between the sign entity and the reference sign entity.

[0137] If the confidence level is not zero, but relatively low, then confirmation guidance text and reference guidance text will be generated separately.

[0138] For example, confirm the guiding text:

[0139] "Are you sure you have the following symptoms: (1) XXX, (2) XXX..."

[0140] At this point, if the user is still likely unable to organize the correct language to describe the situation, further guidance and reference are needed.

[0141] This application employs a comprehensive evaluation of reference entities using both similarity and probability. First, it matches the target data unit to identify the symptom or sign entity with the highest similarity. Then, it selects the highest probability from multiple triage results and weights it by summing the highest probability with the similarity of the symptom entities. The entity with the highest score is likely the symptom or sign entity the user wants to express. Reference guidance text is then constructed using the highest-scoring entity, for example:

[0142] "Are the symptoms or signs you wish to describe: (1) XXX, (2) XXX...";

[0143] The aforementioned confirmation and reference guidance texts can effectively prompt users. After reviewing the guidance text, if the user confirms again, it's possible that the user has indeed experienced a low-probability symptom or sign. Therefore, upon receiving confirmation from the user, the confidence level of the multiple target entities is set to a target value greater than or equal to a preset confidence threshold. Then, the multiple target entities are input into the triage model to complete the triage process.

[0144] The loop module 250 is used to, when receiving the user's symptom description information, use the symptom description information as the current symptom description information and return to the entity extraction module until there are multiple target entities with a confidence level greater than or equal to a preset confidence level threshold. Then, the multiple target entities are input into the pre-built triage model to obtain the triage result. The symptom description information is generated based on the guidance text.

[0145] If the user does not provide further confirmation but instead enters new symptom descriptions, the system returns to the entity extraction module for a complete round of entity extraction and analysis. If the user describes the symptoms again based on the reference guidance text, the second result will likely show multiple target entities with a confidence level greater than or equal to the preset confidence threshold. Otherwise, the process continues until multiple target entities with a confidence level greater than or equal to the preset confidence threshold are found. These target entities are then input into a pre-built triage model to obtain the triage results.

[0146] Through the above process of triage, the unreasonable description of the user can be found in time, the user is guided to describe correctly, and then the triage accuracy of the triage model is improved.

[0147] In the present application, AI analysis is used for triage, and the basis for triage is mainly the user's symptoms and signs. If the user is an old person or a child, the age also needs to be considered, so the triage can be regarded as a multi-classification problem of classifying samples containing user's symptoms, signs and user's samples. This multi-classification scenario is very suitable for being executed by an RNN convolutional neural network, therefore, the construction method of the triage model based on the RNN convolutional neural network includes:

[0148] (1) obtaining a triage record;

[0149] (2) extracting a plurality of triage sample data from the triage record, wherein the triage sample data includes a standard symptom description entity, a sign conclusion entity, an identity vector information and a triage conclusion;

[0150] The triage record is text, so it needs to be standardized to improve the training efficiency and the performance of the model,

[0151] The standardization processing includes: extracting the symptom information, the sign information and the identity information in each triage record; performing standardization processing and vectorization coding on the symptom information to obtain a standard symptom description entity; performing conclusion labeling and vectorization coding on the sign information to obtain a sign conclusion entity; performing standardization processing and vectorization coding on the triage conclusion to obtain a triage conclusion; and performing vectorization coding on the identity information to obtain an identity vector information.

[0152] The vectorization coding mode in the present application adopts Word2Vec.

[0153] (3) constructing a training data sample with the triage conclusion as a label to obtain a training data set;

[0154] (4) training the RNN convolutional neural network based on the training data set to minimize the set multi-classification cross-entropy loss function, and obtaining a triage model.

[0155] In an embodiment of the present application, the multi-classification cross-entropy loss function The mathematical expression is:

[0156]

[0157] In the formula, is the number of samples, is the sample serial number, is the number of categories, is the category serial number, For the label, For the model to predict samples Belonging to the class Probability.

[0158] The triage model obtained through the above training process can effectively use the plurality of target entities verified in the foregoing to perform triage.

[0159] The present application can not only perform automatic triage, but also can replace personnel to a certain extent to check and correct the user's expression problem. In actual application scenarios, it has strong practicability.

[0160] An interactive triage terminal based on AI analysis of the present application extracts target entities from the collected symptom description information, thereby obtaining symptom entities, sign entities and identity information. Then, based on the pre-constructed triage database, the confidence of the target entities is evaluated, thereby determining whether there are symptoms and sign descriptions inconsistent with the identity information, or there are unmatched or low-probability symptom and sign combinations in the description information input by the user. If the above situations exist, the confidence will be low. When the confidence is low, the present application will construct a guide text based on the target entities and the triage database, thereby guiding the user to express correctly. When the correct expression with high confidence is obtained, it is input into the triage model to perform AI analysis, and the triage result can be obtained. The present application can effectively avoid the triage error caused by the user's inability to express correctly.

[0161] The present application also provides an interactive triage method based on AI analysis, comprising the steps of:

[0162] S310, acquiring current symptom description information of a user;

[0163] S320, extracting a plurality of target entities from the current symptom description information, wherein the plurality of target entities include symptom entities, sign entities and identity information;

[0164] S330, analyzing the plurality of target entities based on a pre-constructed triage database to obtain the confidence of the plurality of target entities, wherein the triage database includes entity combinations corresponding to a plurality of triage results;

[0165] S340, when the confidence of the plurality of target entities is greater than or equal to a preset confidence threshold, inputting the plurality of target entities into a pre-constructed triage model to obtain a triage result; when the confidence of the plurality of target entities is less than the preset confidence threshold, constructing a guide text based on the triage database and the plurality of target entities, and sending the guide text to the user;

[0166] S350, when receiving the symptom description information of the user, taking the symptom description information as the current symptom description information, and returning to the entity extraction module, until there are multiple target entities with a confidence greater than or equal to a preset confidence threshold, inputting the multiple target entities into a pre-constructed triage model to obtain a triage result, the symptom description information being generated based on the guide text.

[0167] The interactive triage method based on AI analysis provided in the present application extracts target entities from collected symptom description information, thereby obtaining symptom entities, sign entities, and identity information. Then, the target entities are evaluated based on a pre-constructed triage database, thereby determining whether there are symptoms and sign descriptions inconsistent with the identity information, or there are inconsistent or low-probability symptom and sign combinations in the description information input by the user. If the above conditions exist, the confidence will be low. When the confidence is low, the present application will construct a guide text based on the target entities and the triage database, thereby guiding the user to express correctly. When the correct expression with high confidence is obtained, it is input into the triage model to perform AI analysis, thereby obtaining a triage result. The present application can effectively avoid the triage errors caused by incorrect expression of the user.

[0168] The present embodiment also provides a computer readable storage medium having a computer program stored thereon, the program being executed by a processor to implement any of the methods in the present embodiment, wherein the method is the execution logic of the present system.

[0169] The present embodiment also provides an electronic terminal, comprising a processor and a memory.

[0170] The memory is configured to store a computer program, and the processor is configured to execute the computer program stored in the memory to enable the terminal to execute any of the methods in the present embodiment.

[0171] The computer readable storage medium in the present embodiment can be understood by those skilled in the art that all or part of the steps of the above-mentioned method embodiments can be completed by a computer program related hardware. The aforementioned computer program can be stored in a computer readable storage medium. The program is executed to perform the steps of the above-mentioned method embodiments; and the aforementioned storage medium includes ROM, RAM, magnetic disc or optical disc and various storage medium that can store program code.

[0172] The electronic terminal provided in the present embodiment comprises a processor, a memory, a transceiver and a communication interface. The memory and the communication interface are connected with the processor and the transceiver and complete communication between each other. The memory is configured to store a computer program, the communication interface is configured to communicate, and the processor and the transceiver are configured to run the computer program to enable the electronic terminal to execute the steps of the above method.

[0173] In the present embodiment, the memory can include a random access memory (RAM) and can also include a non-volatile memory such as at least one disk memory.

[0174] The processor described above can be a general processor including a central processing unit (CPU), a network processor (NP), etc., and can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic device, a discrete gate or transistor logic, a discrete hardware component.

[0175] In the above-described embodiments, although the present application has been described in conjunction with specific embodiments thereof, it is evident that many alternatives, modifications and variations will be apparent to those skilled in the art in light of the foregoing description. Embodiments of this application are intended to embrace all such alternatives, modifications and variations as can fall within the scope of the appended claims.

[0176] The above-described embodiments are merely illustrative for the principles and effects of the present application, and are not intended to limit the present application. Any modification or change made by any person skilled in the art without departing from the spirit and scope of the present application shall be covered by the appended claims.

Claims

1. An interactive triage terminal based on AI analysis, characterized in that, The method comprises the following steps: An acquisition module is configured to acquire current symptom description information of a user; An entity extraction module is configured to extract a plurality of target entities from the current symptom description information, wherein the plurality of target entities comprise a symptom entity, a sign entity, and identity information; The analysis module is used to analyze the multiple target entities based on a pre-built triage database to obtain the confidence levels of the multiple target entities. The triage database includes entity combinations corresponding to various triage results. The method for constructing the triage database includes: acquiring historical triage records; and extracting symptom entities, sign entities, identity information, and triage results recorded during multiple triages from the historical triage records. The symptom entities and the sign entities are then standardized to obtain standard symptom entities. and standard physical characteristics The standard symptom entity and the standard sign entity both include body parts and feature descriptions. Based on identity information, the symptom entities, sign entities, and triage results recorded during multiple triages are divided into data units. Then, based on the triage results, the symptom entities and sign entities in each data unit are further divided into symptom data subunits and sign data subunits. Finally, each data unit is directly divided based on the triage results to obtain symptom-sign data subunits. Each symptom entity in the symptom data subunit is calculated. probability of occurrence Each vital sign entity in the vital sign data subunit probability of occurrence ,in, The entity number is used to extract symptom-sign entity combinations from the same time period from the symptom-sign data subunit. And calculate each symptom-sign entity combination probability of occurrence Symptom entities with a probability less than a preset probability threshold Physical characteristics and symptom-sign combination Remove from the data unit and update the reference symptom entity in the symptom data subunit. probability of occurrence The reference vital sign entity in the vital sign data subunit probability of occurrence and updating the reference symptom-sign entity combination in the symptom-sign data subunit. probability of occurrence This allows access to the triage database. A result judgment module is configured to input the plurality of target entities into a pre-constructed triage model when a confidence level of the plurality of target entities is greater than or equal to a preset confidence threshold, to obtain a triage result; and when the confidence level of the plurality of target entities is less than the preset confidence threshold, to construct a guide text based on the triage database and the plurality of target entities, and to send the guide text to the user; A loop module is configured to, when receiving symptom description information of a user, take the symptom description information as current symptom description information, and return to the entity extraction module, until there are a plurality of target entities with a confidence level greater than or equal to a preset confidence threshold, input the plurality of target entities into a pre-constructed triage model to obtain a triage result, and the symptom description information is generated based on the guide text. 2.The AI analysis-based interactive triage terminal of claim 1, wherein, The plurality of target entities are analyzed based on a pre-constructed triage database to obtain a confidence level of the plurality of target entities, comprising: Determining a target data unit in the triage database based on the identity information of the user; Comparing a symptom entity in the current description information with a reference symptom entity in a plurality of symptom data subunits, comparing a sign entity in the current description information with a reference sign entity in a plurality of sign data subunits, and comparing a symptom-sign entity combination in the current description information with a reference symptom-sign entity combination in the symptom-sign data subunit; When there is no state entity consistent with the reference symptom entity in the current description information, or there is no sign entity consistent with the reference sign entity, setting the confidence level of the plurality of target entities to zero, and determining that the current description information is inconsistent with the identity information of the user; there is a reference symptom entity in the current description information a consistent state entity and there is a reference sign entity consistent sign entity, the occurrence probability corresponding to the same triage result is weighted and summed based on 、 and the confidence of the plurality of target entities . 3.The AI analysis-based interactive triage terminal of claim 2, wherein, Constructing a guide text based on the triage database and the plurality of target entities, comprising: When the confidence level of the plurality of target entities is zero, constructing a guide text for prompting that the symptom or sign is inconsistent with the identity information; When the confidence of the plurality of target entities is not zero, constructing confirmation guidance text based on a symptom entity and a sign entity in the target entity; and scoring reference symptom entities or reference sign entities in the target data unit based on the symptom entity or the sign entity , and constructing reference guidance text based on the top N reference symptom entities or reference sign entities with the highest scores , wherein the score is mathematically expressed as:​ wherein, is a first weight, is a second weight, is a cosine similarity of a symptom entity to a reference symptom entity, is a cosine similarity of a sign entity to a reference sign entity. 4.The AI analysis-based interactive triage terminal of claim 1, wherein, After sending the guide text to the user, further comprising: When receiving a confirmation description from the user, setting the confidence level of the plurality of target entities to a target value greater than or equal to the preset confidence threshold. 5.The AI analysis-based interactive triage terminal of claim 1, wherein, the probability of occurrence of the reference symptom entity the mathematical expression of the reference symptom entity the probability of occurrence of the reference symptom entity the mathematical expression of the reference symptom entity the probability of occurrence of the reference symptom entity the probability of occurrence of the reference symptom entity respectively. wherein, is the total number of reference symptom entities in the symptom data subunit, is the total number of reference sign entities in the sign data subunit, is the total number of reference symptom-sign entity combinations in the symptom-sign data subunit. 6.The AI analysis-based interactive triage terminal of claim 1, wherein, The construction method of the triage model comprises: Acquiring triage records; Extracting a plurality of triage sample data from the triage records, wherein the triage sample data comprises a standard symptom description entity, a sign conclusion entity, identity vector information, and a triage conclusion; Constructing a training data sample with the triage conclusion as a label to obtain a training data set; Training an RNN convolutional neural network based on the training data set to minimize a set multi-classification cross-entropy loss function, to obtain a triage model. 7.The AI analysis-based interactive triage terminal of claim 6, wherein, Extracting a plurality of triage sample data from the triage records comprises: Extracting symptom information, sign information, and identity information in each triage record; The symptom information is standardized and vectorized to obtain a standard symptom description entity; the sign information is conclusively labeled and vectorized to obtain a sign conclusion entity; the triage conclusion is standardized and vectorized to obtain a triage conclusion; and the identity information is vectorized to obtain identity vector information. 8.The AI analysis-based interactive triage terminal of claim 6, wherein, The multi-classification cross-entropy loss function The mathematical expression is: where, is the number of samples, is the sample index, is the number of classes, is the class index, is the label, is the model's predicted probability that a sample belongs to a class . 9.The AI analysis-based interactive triage terminal of claim 1, wherein, The current symptom description information of the user is obtained, including: When a triage starting command from the outside is received, a guide voice or initial guide text is outputted; When a voice reply or text reply from the outside is received, the voice reply or the text reply is preliminarily entity-verified, and when the voice reply or the text reply contains a predetermined entity, the voice reply or the text reply is taken as symptom description information.

Images