CN111145903A

CN111145903A - Method and device for acquiring vertigo inquiry text, electronic equipment and inquiry system

Info

Publication number: CN111145903A
Application number: CN201911312648.1A
Authority: CN
Inventors: 张淑蕾; 王超; 柴东; 杨英健; 范智渊; 康雁
Original assignee: Northeastern University China
Current assignee: Northeastern University China
Priority date: 2019-12-18
Filing date: 2019-12-18
Publication date: 2020-05-12

Abstract

The invention discloses a method, a device, electronic equipment and an inquiry system for acquiring a vertigo inquiry text, wherein the method for acquiring the inquiry text comprises the following steps: collecting in real time the questions posed by the doctor for vertigo and the patient's responses to the questions; converting the questions and responses into an interrogation text; storing the inquiry text and/or sending the inquiry text to a preset text analysis unit; wherein the question and the answer are voice signals. The problems that the disease information of the patients in the doctor-patient history conversation is automatically extracted, real-time interaction between doctors and the patients is lacked, questions and answers are designed for the patients to select, the expression of the patients is limited, some patients have difficulty in reading, understanding and writing characters, the inquiry information is easy to lose, and diagnosis is affected are solved.

Description

Method and device for acquiring vertigo inquiry text, electronic equipment and inquiry system

Technical Field

The invention relates to the field of medicine, in particular to a method, a device, electronic equipment and an inquiry system for acquiring a vertigo inquiry text.

Background

With the rise and development of artificial intelligence, the intelligent inquiry system plays an important role in the doctor-patient inquiry part. The interrogation is the first step in clinical work and is critical to the diagnosis of disease. When a patient is in a visit, a doctor can see a certain point of disease development, and the thorough understanding of the disease, particularly the development and evolution process of the disease, influencing factors and the like, needs to be solved through inquiry. Because many conditions of the disease, such as the occurrence, development, course of change and treatment course, subjective symptoms, past medical history, life history and family history of the patient, etc., can be obtained only by inquiry. The above-mentioned information related to diseases is the reliable basis for the physician to analyze the disease condition and differentiate the syndrome. In addition, the inquiry method can provide a general disease examination scope for other diagnosis methods, and the thought condition of the patient can be known through the inquiry method, so that the diagnosis and treatment of the disease can be facilitated. Therefore, the inquiry is one of the important methods for doctors to diagnose diseases. And because of the imbalance of medical resources, professional doctors are fewer, so that the problems of large workload of doctors, difficult patient treatment and the like are caused.

The current intelligent inquiry system mainly comprises a diagnosis model, a doctor-patient interaction module, a diagnosis module and a diagnosis module, wherein the diagnosis model is used for automatically extracting disease information, treatment medication information and diagnosis suggestion information of a patient in doctor-patient historical conversation, performing semantic representation according to the extracted information, converting the extracted information into numerical vectors, and performing training and learning on the numerical vectors to obtain the relation between the disease information and the doctor diagnosis suggestions existing in the doctor-patient historical interaction; obtaining inquiry content through the answers of the selected questions of the patient, performing semantic recognition on the inquiry content, extracting keywords, matching corresponding answers according to the extracted keywords by using a diagnosis model, and sending the answers to the client.

Currently developed intelligent interrogation systems, such as: an artificial intelligent inquiry method (application number: 201711269908.2) automatically extracts disease information of patients in doctor-patient historical conversations, real-time interaction between doctors and the patients is lacked, or inquiry contents based on an inquiry and answer method (application number: 201610854270.8) of inquiry of an intelligent terminal are obtained by designing questions and answers for the patients to select, more like questionnaires, the expression of the patients is limited, some patients have difficulty in reading, understanding and writing characters, the inquiry information is easy to lose, diagnosis is affected, doctors have the functions of soothing and soothing when communicating with the patients, the tension of the patients cannot be relieved due to a fixed inquiry mode, and humanization is not considered in the aspect of patient experience.

Vertigo is a syndrome of shaking, rotating, moving or dizziness of surrounding scenery reflected by cerebral cortex of a patient, inclination of body, blurred vision and the like caused by balance disorder caused by positional or motor illusion. It relates to multiple disciplines, most people experience the vertigo in life, the vertigo has complex etiology and difficult diagnosis and treatment, and the clinical research is relatively slow because of the multiple disciplines. Moreover, the number of professional doctors for treating vertigo is small, doctors can make a prediction through the inquiry of patients, and the work is repeated and heavy. Because there are many reasons for dizziness, patients will go to various examinations when seeing a doctor, such as neurology, cardiovascular department, etc., and vertigo patients often go to clinical multiple departments and cannot get clear and effective diagnosis and treatment, wasting time, energy and money. The patient generates fear psychology during vertigo attack, the patient is threatened greatly, doctors often diagnose according to subjective symptoms, objective physical signs are few, and if the doctors are dark or bright, the doctors are difficult to capture, and difficulties are brought to diagnosis and differential diagnosis. In order to reduce the workload of doctors and help the doctors to distinguish vertigo patients from non-vertigo patients, inquiry audio data provided by the doctors are recorded and analyzed to obtain a systematic inquiry mode, and an intelligent inquiry system is designed, so that the diagnosis and treatment efficiency is improved, and the patients are treated in time.

Diagnosis and treatment of vertigo is a few diseases which do not depend on CT and other image examination seriously, and is characterized in that the vertigo is manifested during the onset of diseases and no symptoms exist during the normal conditions, so the inquiry has an extremely important meaning for the diagnosis of vertigo.

Disclosure of Invention

In view of the above, the invention provides a method, a device, an electronic device and an inquiry system for acquiring an inquiry text of vertigo, so as to solve the problems that the disease condition information of a patient in a doctor-patient historical conversation is automatically extracted, real-time interaction between a doctor and the patient is lacked, the inquiry content is obtained by designing questions and answers for the patient to select, the expression of the patient is limited, some patients have difficulty in reading, understanding and writing characters, the inquiry information is easy to lose, the diagnosis is influenced, and the doctor has the functions of soothing and soothing when communicating with the patient, the tension emotion of the patient cannot be relieved due to a fixed inquiry mode, and humanization is not considered in the aspect of patient experience.

In a first aspect, a method for obtaining an interrogation text for vertigo, comprising:

collecting in real time the questions posed by the doctor for vertigo and the patient's responses to the questions;

converting the questions and responses into an interrogation text;

storing the inquiry text and/or sending the inquiry text to a preset text analysis unit;

wherein the question and the answer are voice signals.

Preferably, the method for converting the question and the response into an inquiry text is:

respectively determining the beginning and the end of the question and the beginning and the end of the response, and respectively carrying out voice recognition and speaker recognition on the question and the response with the beginning and the end; the voice recognition is used for converting the questions and the responses into inquiry texts; the speaker identification identifies the question and the speaker tag of the response;

wherein the speaker tag is an interrogation mark of the interrogation text, and the interrogation mark is used for distinguishing the question from the response.

Preferably, the method for acquiring the problems posed by the doctors for vertigo in real time and the responses of the patients for the problems is as follows:

obtaining in real time a spoken question of a doctor and a spoken response of a patient to the spoken question in real time; or

Presetting the questions proposed by a doctor, and acquiring the responses of the patients to the questions in real time.

Preferably, characterized in that the patient's response to the question is judged before the question and the response are converted into an interrogation text; if the answer rule is met, the question and the answer are reserved, otherwise, the question and the answer are deleted.

Preferably, the method for determining whether the set response rule is satisfied is: extracting key features of the question and key features of the response respectively, calculating the similarity between the key features of the response and the key features of the question, and judging the size of the similarity and the set value;

if the similarity is larger than a set value, the set response rule is considered to be met, otherwise, the set response rule is considered not to be met.

In a second aspect, an apparatus for obtaining a text for an interrogation of vertigo, comprising: the device comprises an acquisition unit, a conversion unit and a storage or storage sending unit;

the output end of the acquisition unit is connected with the input end of the conversion unit, and the output end of the conversion unit is connected with the storage or storage sending unit;

the acquisition unit acquires the problems proposed by doctors for vertigo and the responses of patients for the problems in real time;

the conversion unit is used for converting the questions and the responses into inquiry texts;

the storage or storage sending unit is used for storing the inquiry text and/or sending the inquiry text to a text analysis unit;

wherein the question and the answer are voice signals.

Preferably, the conversion unit includes: the device comprises a determining unit, a voice recognition unit and a speaker recognition unit;

the input end of the determining unit is connected with the output end of the collecting unit, the output end of the determining unit is respectively connected with the input ends of the voice recognition unit and the speaker recognition unit, and the output ends of the voice recognition unit and the speaker recognition unit are respectively connected with the storage or storage sending unit;

the determining unit is used for respectively determining the beginning and the end of the question and the beginning and the end of the response, and respectively sending the question with the beginning and the end and the response to the voice recognition unit and the speaker recognition unit for voice recognition and speaker recognition;

the voice recognition unit converts the questions and the responses into inquiry texts;

the speaker identification unit identifies the question and the speaker tag of the response;

Preferably, the acquisition unit includes: a recording or speech recognition unit;

the recording or speech recognition unit obtains the dictation questions of the doctor in real time and obtains the dictation responses of the patient to the dictation questions in real time.

Preferably, a judging unit is arranged between the acquisition unit and the conversion unit;

the judging unit judges the response of the patient to the question; if the answer rule is met, the question and the answer are reserved, otherwise, the question and the answer are deleted.

Preferably, the judging unit includes: a first extraction unit, a second extraction unit and a calculation unit;

the input ends of the first extraction unit and the second extraction unit are respectively connected with the output end of the acquisition unit, and the output ends of the first extraction unit and the second extraction unit are respectively connected with the input end of the calculation unit;

the first extraction unit extracts key features of the problem;

the second extraction unit extracts key features of the response;

the calculating unit is used for calculating the similarity between the key features of the response and the key features of the question and judging the size of the similarity and the set value; if the similarity is larger than a set value, the set response rule is considered to be met, otherwise, the set response rule is considered not to be met.

In a third aspect, an electronic device includes:

a processor;

a memory for storing processor-executable instructions;

wherein the processor is configured to: the method described above is performed.

In a fourth aspect, a vertigo interrogation system, comprising: the method as described above; or

The above-described apparatus; or

The electronic device described above;

sending the inquiry text to a preset text analysis unit, identifying the inquiry text through the preset text analysis unit to obtain the intention and word groove of each text in the inquiry text, searching the position information of the intention and the word groove in a preset database according to the intention and the word groove, matching preset key information by using the position information to generate a medical record report, sending the medical record report to a preset diagnosis model, and giving a diagnosis result by the preset diagnosis model according to the medical record report; and/or

The preset text analysis unit is an intention word groove model, the intention word groove model is constructed through a training text of the inquiry text, and the inquiry text is identified to obtain the intention and a word groove of each text in the inquiry text; and/or

Respectively recording the intention and the word groove of the inquiry text according to the speaker label, constructing the preset database according to the recording result, searching the position information of the intention and the word groove in the preset database according to the intention and the word groove, and generating a medical record report by matching preset key information with the position information; and/or

The construction process of the preset diagnosis model comprises the following steps: segmenting words of a given medical record report with diagnosis, training the given medical record report with the segmented words to obtain a text word vector, representing the text vector according to the inquiry text word vector, and training the diagnosis model by using the text vector to obtain the preset diagnosis model; sending the medical record report to the preset diagnosis model, wherein the preset diagnosis model gives a diagnosis result according to the medical record report; wherein the text vector is the sum of text word vectors/text length; or the text vector representation method comprises the following steps: acquiring a medical record text, and converting the medical record text into a word list; converting each word in the word list into a word vector by using at least 2 preset word vector models; respectively converting the word vectors into text vectors; synthesizing the text vector to obtain a synthesized text vector; the synthetic text vector is input into a classification model, and the classification model completes classification of the medical record text; or the text vector is generated by adopting a device for representing the medical record text vector, and the device for representing the medical record text vector comprises the following steps: the system comprises a word list conversion unit, at least 2 preset word vector models, a text vector conversion unit and a synthesis unit, wherein the output of the word list conversion unit is connected with the input of the preset word vector models respectively, the output of the preset word vector models is connected with the input of the text vector conversion unit, and the output of the text vector conversion unit is connected with the input of the synthesis unit; the word list conversion unit is used for acquiring a medical record text and converting the medical record text into a word list; the preset word vector model converts each word in the word list into a word vector respectively; the text vector conversion unit is used for respectively converting the word vectors into text vectors; the synthesis unit synthesizes the text vectors to obtain synthesized text vectors; the synthetic text vector is input into a classification model, and the classification model completes classification of the medical record text; or the text vector is generated by adopting a device for representing the medical record text vector, and the device for representing the medical record text vector comprises the following steps: a processor; a memory for storing processor-executable instructions; wherein the processor is configured to perform the steps of: acquiring a medical record text, and converting the medical record text into a word list; converting each word in the word list into a word vector by using at least 2 preset word vector models; respectively converting the word vectors into text vectors; synthesizing the text vector to obtain a synthesized text vector; and the synthetic text vector is input into a classification model, and the classification model completes the classification of the medical record text.

The invention has at least the following beneficial effects:

the invention provides a method, a device, electronic equipment and an inquiry system for acquiring an inquiry text of vertigo, which aim to solve the problems that the disease information of a patient in doctor-patient historical conversation is automatically extracted, the real-time interaction between a doctor and the patient is lacked, the inquiry content is obtained by designing questions and answers for the patient to select, the expression of the patient is limited, some patients have difficulty in reading, understanding and writing characters, the inquiry information is easy to lose, the diagnosis is influenced, and the doctor has the functions of soothing and soothing when communicating with the patient, the tension of the patient cannot be relieved due to a fixed inquiry mode, and humanization is not considered in the aspect of patient experience.

Because the patient generates fear psychology during vertigo attack, the patient is threatened greatly, doctors often diagnose the vertigo by subjective symptoms, objective physical signs are few, and the vertigo is difficult to capture if the vertigo is dark or bright, so that the diagnosis and differential diagnosis are difficult. The invention can reduce the workload of doctors and help the doctors to distinguish vertigo patients from non-vertigo patients, thereby improving the diagnosis and treatment efficiency and leading the patients to be treated in time.

Drawings

The above and other objects, features and advantages of the present invention will become more apparent from the following description of the embodiments of the present invention with reference to the accompanying drawings, in which:

FIG. 1 is a flow chart of a method for obtaining an inquiry text of vertigo according to an embodiment of the invention;

FIG. 2 is a schematic diagram of an apparatus for obtaining a text for an interrogation of vertigo according to an embodiment of the present invention;

fig. 3 is a flowchart illustrating a method for representing a medical record text vector according to an embodiment of the present invention.

Detailed Description

The present invention will be described below based on examples, but it should be noted that the present invention is not limited to these examples. In the following detailed description of the present invention, certain specific details are set forth. However, the present invention may be fully understood by those skilled in the art for those parts not described in detail.

Furthermore, those skilled in the art will appreciate that the drawings are provided solely for the purposes of illustrating the invention, features and advantages thereof, and are not necessarily drawn to scale.

Also, unless the context clearly requires otherwise, throughout the description and the claims, the words "comprise", "comprising", and the like are to be construed in an inclusive sense as opposed to an exclusive or exhaustive sense; that is, the meaning of "includes but is not limited to".

Fig. 1 is a flowchart illustrating a method for obtaining an inquiry text of vertigo according to an embodiment of the present invention. As shown in fig. 1, a method of obtaining an inquiry text includes: step 101, collecting the problems proposed by doctors for vertigo and the responses of patients for the problems in real time; step 102, converting the questions and responses into inquiry texts; step 103, storing the inquiry text and/or sending the inquiry text to a preset text analysis unit; wherein the question and the answer are speech signals. The problems that the disease information of the patient in the doctor-patient historical conversation is automatically extracted, real-time interaction between a doctor and the patient is lacked, the inquiry content is obtained by designing questions and answers for the patient to select, the expression of the patient is limited, some patients have difficulty in reading, understanding and writing characters, the inquiry information is easy to lose, and diagnosis is affected are solved, the doctor has the effects of soothing and soothing when communicating with the patient, the tension of the patient cannot be relieved due to the fixed inquiry mode, and humanization is not considered in the aspect of patient experience.

In fig. 1, the specific method for acquiring the problems posed by the doctor for vertigo and the responses of the patient for the problems in real time in step 101 is as follows: the method for collecting the questions posed by the doctor and the responses of the patient to the questions comprises the following steps: the physician's spoken questions are obtained in real time and the patient's spoken responses to the spoken questions are obtained in real time.

Specifically, in the embodiment disclosed in step 101 of the present invention, the specific method for collecting the questions posed by the doctor and the patient responses to the questions is as follows: recording the questions posed by the doctor and the responses of the patient to the questions in real time by using a recording function of a mobile phone end or a PC (personal computer) end in an inquiry room; in order to ensure the recording quality so that the voice recognition is accurate, an auxiliary recording device can be added, comprising: microphones, two-microphone or six-microphone array microphones, etc.

In fig. 1, the specific method for converting the questions and responses into the inquiry text in step 102 is as follows: determining the beginning and the end of the question and the beginning and the end of the response by utilizing a voice endpoint detection method respectively, and performing voice recognition and speaker recognition on the question and the response with the beginning and the end respectively; the voice recognition is used for converting the questions and the responses into inquiry texts; the speaker identification identifies the question and the speaker tag of the response; wherein the speaker tag is an interrogation mark of the interrogation text, and the interrogation mark is used for distinguishing the question from the response.

Specifically, in the embodiment disclosed in step 102 of the present invention, the method for determining the beginning and the end of the question and the beginning and the end of the answer by using the voice endpoint detection method respectively is as follows: the method comprises the steps of collecting voice signals of a problem posed by a doctor for vertigo and a response of a patient for the problem, judging the beginning and the end of the problem posed by the doctor for vertigo and the response of the patient for the problem by using a voice endpoint detection technology, and then dividing the problem posed by the doctor for vertigo and the response of the patient for the problem into voice signal segments, wherein one voice signal segment is a sentence.

Specifically, in the embodiment disclosed in step 102 of the present invention, the question and the response having a beginning and an end are subjected to speech recognition and speaker recognition, respectively, the speech recognition converting the question and the response into an inquiry text; the method for identifying the speaker and identifying the speaker label of the question and the response comprises the following steps:

the voice signal segments divided into individual voice signal segments are converted into words (inquiry text) by utilizing a voice recognition technology, wherein the voice signals of the problems proposed by doctors for vertigo and the responses of patients for the problems are converted into words, and meanwhile, the voice signals of the problems proposed by doctors for vertigo and the responses of patients for the problems are recognized by utilizing a speaker recognition technology to obtain speaker tags of the inquiry text. The speaker recognition technique herein mainly recognizes the speaker tag of the doctor and the speaker tag of the patient responding to the question, and further distinguishes the question from the response. The specific method comprises the following steps: the label identified by the speaker is the voice fragment of the doctor, and the speaker label is the doctor; the speaker-identified tags are not the doctor's voice segment, and the speaker tags are all patients.

The questions posed by the doctor for vertigo and the text of the interview of the patient's response to the questions after speech recognition and speaker recognition are given as follows: a doctor: how did you sleep recently? The patients: recently, insomnia has occurred. Wherein, the doctor and the patient are speaker tags.

In fig. 1, the patient's response to the question is determined before the question and response are converted into an interrogation text at step 102; if the answer rule is met, the question and the answer are reserved, otherwise, the question and the answer are deleted.

Specifically, the method for determining whether the set response rule is satisfied is as follows: extracting key features of the question and key features of the response respectively, calculating the similarity between the key features of the response and the key features of the question, and judging the size of the similarity and the set value; if the similarity is larger than a set value, the set response rule is considered to be met, otherwise, the set response rule is considered not to be met.

Specifically, in the embodiment disclosed in step 102 of the present invention, the specific method for extracting the key feature of the question and the key feature of the answer respectively is as follows: the method comprises the steps of performing word segmentation on a question text and a response text (inquiry text) by using a word segmentation tool (such as a jieba word segmentation tool) and a self-defined dictionary mode in the word segmentation tool, setting a stop word list, and filtering stop words of the inquiry text by using the stop word list, wherein the stop words can be added with words to be removed according to specific scenes. Stop words such as mood words, auxiliary words and/or punctuation marks. The query text gets key features of questions and responses after word segmentation and word stop.

Calculating the similarity between the key features of the response and the key features of the question, and judging the size of the similarity and the set value; if the similarity is greater than a set value, the set response rule is considered to be met, otherwise, the specific method for considering that the set response rule is not met comprises the following steps: and using the trained word2vec model, taking the key features of the question and the response as the input of the word2vec model, and outputting word vectors of the question and the response of each key feature by the word2vec model. The training corpus of the word2vec model is questions and responses, namely question-answer pairs (a group of questions and responses comprises questions and responses), and the question-answer pairs should cover all preset questions of vertigo. The corpus may be obtained by crawling on the web, or using open-source dialog corpus, or by human manufacture.

Key feature vectors for the questions and answers may then be computed. Specifically, the key feature vector of the question and the answer can be obtained by the following formula: the key feature vector of the question is the sum of word vectors of all key features of the question/the length of the question; the key feature vector of the response is the sum of the word vectors of the key features of the response/the length of the response.

Calculating cosine similarity of the key characteristic vector of the question and the key characteristic vector of the answer, wherein the larger the cosine similarity is, the more similar the key characteristic of the answer and the key characteristic of the question are. And if the cosine similarity of the key characteristic vector of the question and the key characteristic vector of the response is greater than a set value, considering that the response of the patient to the question conforms to a response rule, and keeping the question and the response. And if the cosine similarity of the key characteristic vector of the question and the key characteristic vector of the response is less than or equal to a set value, the response of the patient to the question is considered not to conform to the response rule, and the question and the response are deleted. Wherein, the set value may be 0.5.

In step 103, the inquiry text is stored and/or sent to a predetermined text analysis unit, which is described in detail in a vertigo inquiry system.

Fig. 2 is a schematic diagram of a device for obtaining an inquiry text of vertigo according to an embodiment of the invention. In fig. 2, an apparatus for obtaining an inquiry text for vertigo, comprising: the system comprises an acquisition unit 201, a conversion unit 202 and a storage or storage sending unit 203; the output end of the acquisition unit 201 is connected with the input end of the conversion unit 202, and the output end of the conversion unit 202 is connected with the storage or storage sending unit 203; an acquisition unit 201 that acquires, in real time, a question posed by a doctor for vertigo and a response of a patient to the question; a conversion unit 202 that converts the questions and responses into an inquiry text; a storage or storage transmitting unit 203 for storing the inquiry text and/or transmitting the inquiry text to the text analyzing unit; wherein the question and the answer are speech signals. The problems that the disease information of the patient in the doctor-patient historical conversation is automatically extracted, real-time interaction between a doctor and the patient is lacked, the inquiry content is obtained by designing questions and answers for the patient to select, the expression of the patient is limited, some patients have difficulty in reading, understanding and writing characters, the inquiry information is easy to lose, and diagnosis is affected are solved, the doctor has the effects of soothing and soothing when communicating with the patient, the tension of the patient cannot be relieved due to the fixed inquiry mode, and humanization is not considered in the aspect of patient experience. The details of an apparatus for vertigo examination text are described in detail in a method for vertigo examination text, and are not described in detail here.

In fig. 2, the conversion unit 202 includes: the device comprises a determining unit, a voice recognition unit and a speaker recognition unit; the input end of the determining unit is connected with the output end of the collecting unit 201, the output end of the determining unit is respectively connected with the input ends of the voice recognition unit and the speaker recognition unit, and the output ends of the voice recognition unit and the speaker recognition unit are respectively connected with the storage or storage sending unit 203; the determining unit is used for determining the beginning and the end of the question and the beginning and the end of the response by utilizing a voice endpoint detection method respectively, and sending the question with the beginning and the end and the response to the voice recognition unit and the speaker recognition unit for voice recognition and speaker recognition respectively; the speaker identification unit identifies the question and the speaker tag of the response; wherein the speaker tag is an interrogation mark of the interrogation text, and the interrogation mark is used for distinguishing the question from the response.

Specifically, in the embodiment disclosed in the present invention, the determining unit determines the start and the end of the question and the start and the end of the response by using a voice endpoint detection method, which is specifically implemented as follows: the method comprises the steps of collecting voice signals of a problem posed by a doctor for vertigo and a response of a patient for the problem, judging the beginning and the end of the problem posed by the doctor for vertigo and the response of the patient for the problem by using a voice endpoint detection technology, and then dividing the problem posed by the doctor for vertigo and the response of the patient for the problem into voice signal segments, wherein one voice signal segment is a sentence.

Specifically, in the embodiment disclosed in the present invention, the determination unit further performs speech recognition and speaker recognition on the question and the response having the beginning and the end, respectively, the speech recognition converting the question and the response into an inquiry text; the speaker identification identifies the question and the speaker label of the response, and the specific implementation mode is as follows: the voice signal segments divided into individual voice signal segments are converted into words (inquiry text) by utilizing a voice recognition technology, wherein the voice signals of the problems proposed by doctors for vertigo and the responses of patients for the problems are converted into words, and meanwhile, the voice signals of the problems proposed by doctors for vertigo and the responses of patients for the problems are recognized by utilizing a speaker recognition technology to obtain speaker tags of the inquiry text. The speaker recognition technique herein mainly recognizes the speaker tag of the doctor and the speaker tag of the patient responding to the question, and further distinguishes the question from the response. The specific method comprises the following steps: the label identified by the speaker is the voice fragment of the doctor, and the speaker label is the doctor; the speaker-identified tags are not the doctor's voice segment, and the speaker tags are all patients.

In fig. 2, the acquisition unit 201 includes: a recording or speech recognition unit; and the recording or voice recognition unit is used for acquiring the dictation questions of the doctor in real time and acquiring the dictation responses of the patient to the dictation questions in real time.

Specifically, in the disclosed embodiment of the present invention, the recording or speech recognition unit performs the following functions: recording the questions posed by the doctor and the responses of the patient to the questions in real time by using a recording function of a mobile phone end or a PC (personal computer) end in an inquiry room; in order to ensure the recording quality so that the voice recognition is accurate, an auxiliary recording device can be added, comprising: microphones, two-microphone or six-microphone array microphones, etc.

In fig. 2, a judgment unit is provided between the acquisition unit 201 and the conversion unit 202; the judging unit judges the response of the patient to the problem; if the answer rule is met, the question and the answer are reserved, otherwise, the question and the answer are deleted.

In fig. 2, the judging unit includes: a first extraction unit, a second extraction unit and a calculation unit; the input ends of the first extraction unit and the second extraction unit are respectively connected with the output end of the acquisition unit 201, and the output ends of the first extraction unit and the second extraction unit are respectively connected with the input end of the calculation unit; the first extraction unit is used for extracting key features of the problem; a second extraction unit for extracting key features of the response; the calculating unit is used for calculating the similarity between the key features of the response and the key features of the problem and judging the similarity and the set value; if the similarity is larger than the set value, the set response rule is considered to be met, otherwise, the set response rule is considered to be not met.

Specifically, the determination unit determines the response of the patient to the question; if the answer rule is met, the question and the answer are reserved, otherwise, the question and the answer are deleted, and the specific implementation mode is as follows: extracting key features of the question and key features of the response respectively, calculating the similarity between the key features of the response and the key features of the question, and judging the size of the similarity and the set value; if the similarity is larger than a set value, the set response rule is considered to be met, otherwise, the set response rule is considered not to be met.

Specifically, in the embodiment disclosed in the present invention, the specific implementation manner of respectively extracting the key features of the question and the key features of the answer is as follows: the method comprises the steps of performing word segmentation on a question text and a response text (inquiry text) by using a word segmentation tool (such as a jieba word segmentation tool) and a self-defined dictionary mode in the word segmentation tool, setting a stop word list, and filtering stop words of the inquiry text by using the stop word list, wherein the stop words can be added with words to be removed according to specific scenes. Stop words such as mood words, auxiliary words and/or punctuation marks. The query text gets key features of questions and responses after word segmentation and word stop.

The invention also discloses an electronic device, comprising: a processor; a memory for storing processor-executable instructions; wherein the processor is configured to: the above method is performed. The details can be seen in the description of fig. 1 or fig. 2.

The invention also discloses a vertigo interrogation system, which comprises: the above method; or the above-mentioned device; or the electronic device described above; sending the inquiry text to a preset text analysis unit, identifying the inquiry text through the preset text analysis unit to obtain the intention and the word groove of each text in the inquiry text, searching the position information of the intention and the word groove in a preset database according to the intention and the word groove, matching the preset key information by utilizing the position information to generate a medical record report, sending the medical record report to a preset diagnosis model, and sending a diagnosis result by the preset diagnosis model according to the medical record report.

Specifically, the inquiry text is sent to a preset text analysis unit, the inquiry text is identified through the preset text analysis unit to obtain the intention and word groove of each text in the inquiry text, the position information of the intention and the word groove is searched in a preset database according to the intention and the word groove, a medical record report is generated by matching preset key information with the position information, the medical record report is sent to a preset diagnosis model, and the preset diagnosis model gives a diagnosis result according to the medical record report.

Specifically, the preset text analysis unit is an intention word groove model, the intention word groove model is constructed through a training text of the inquiry text, and the inquiry text is identified to obtain the intention and the word groove of each text in the inquiry text. Analyzing according to the inquiry text, wherein the intention categories of the questions are 25 categories, such as the past history, the related medical history, dizziness and complications and the like; the response of the patients to the questions is about 80 types, such as response duration, headache history, car sickness history, and poor sleep. The greater the number of interview texts, the more sophisticated the categories of intentions the questions and patients respond to will be.

Specifically, the intention and the word groove of the inquiry text are recorded according to the speaker label, the preset database is constructed according to the recording result, the position information of the intention and the word groove is searched in the preset database according to the intention and the word groove, and a medical record report is generated by matching the preset key information with the position information.

Specifically, the preset diagnosis model is constructed by the following steps: segmenting words of a given medical record report with diagnosis, training the given medical record report with the segmented words to obtain a text word vector, representing the text vector according to the inquiry text word vector, and training the diagnosis model by using the text vector to obtain the preset diagnosis model; sending the medical record report to the preset diagnosis model, wherein the preset diagnosis model gives a diagnosis result according to the medical record report; wherein the text vector is the sum of text word vectors/text length; or the text vector representation method comprises the following steps: acquiring a medical record text, and converting the medical record text into a word list; converting each word in the word list into a word vector by using at least 2 preset word vector models; respectively converting the word vectors into text vectors; synthesizing the text vector to obtain a synthesized text vector; the synthetic text vector is input into a classification model, and the classification model completes classification of the medical record text; or the text vector is generated by adopting a device for representing the medical record text vector, and the device for representing the medical record text vector comprises the following steps: the system comprises a word list conversion unit, at least 2 preset word vector models, a text vector conversion unit and a synthesis unit, wherein the output of the word list conversion unit is connected with the input of the preset word vector models respectively, the output of the preset word vector models is connected with the input of the text vector conversion unit, and the output of the text vector conversion unit is connected with the input of the synthesis unit; the word list conversion unit is used for acquiring a medical record text and converting the medical record text into a word list; the preset word vector model converts each word in the word list into a word vector respectively; the text vector conversion unit is used for respectively converting the word vectors into text vectors; the synthesis unit synthesizes the text vectors to obtain synthesized text vectors; the synthetic text vector is input into a classification model, and the classification model completes classification of the medical record text; or the text vector is generated by adopting a device for representing the medical record text vector, and the device for representing the medical record text vector comprises the following steps: a processor; a memory for storing processor-executable instructions; wherein the processor is configured to perform the steps of: acquiring a medical record text, and converting the medical record text into a word list; converting each word in the word list into a word vector by using at least 2 preset word vector models; respectively converting the word vectors into text vectors; synthesizing the text vector to obtain a synthesized text vector; and the synthetic text vector is input into a classification model, and the classification model completes the classification of the medical record text.

Fig. 3 is a flowchart illustrating a method for representing a medical record text vector according to an embodiment of the present invention. As shown in fig. 3, a method of representing a medical record text vector includes: step 1001, acquiring a medical record text, and converting the medical record text into a word list; step 1002, converting each word in the word list into a word vector by using at least 2 preset word vector models; step 1003 converts the word vectors into text vectors respectively; step 1004, synthesizing all the text vectors to obtain synthesized text vectors; and the synthetic text vector is input into a classification model, the classification model completes the classification of the medical record text, and the classification model is a preset diagnosis model. That is to say, each word vector model obtains one text vector, and all the text vectors are synthesized to obtain a synthesized text vector. The method solves the problem that the traditional text vector representation method can not fully express text information, so that a classification model can not well classify the input text.

Step 1001 is a method for acquiring a medical record text and converting the medical record text into a word list:

the medical record text is a medical record report to be diagnosed, and the medical record report to be diagnosed has no diagnosis label and is a medical record report needing to be diagnosed. The medical record text is converted to a word list by word segmentation for a given medical record report with diagnosis. The method comprises the steps of utilizing a word segmentation tool (such as a jieba word segmentation tool) and a self-defined dictionary mode in the word segmentation tool to perform word segmentation on a medical record text, setting a stop word list, and utilizing the stop word list to filter stop words of the medical record text, wherein the stop words can be added with words which are set to be removed in the existing stop word list according to specific scenes. Stop words such as mood words, auxiliary words and/or punctuation marks. After the medical record text passes through the word segmentation tool, a word list is obtained, and the word list is a list with a plurality of words.

Step 1002 converts each word in the word list into a word vector using at least 2 preset word vector models:

the at least 2 preset word vector models are obtained by training the medical record text, and then each word in the word list is converted into a word vector by using the at least 2 preset word vector models, wherein the training method of the preset word vector models comprises the following steps: obtaining a training corpus of a word vector model according to the medical record text; respectively inputting the training corpora into a preset vector model, and training the preset vector model; the set of medical history texts for training the preset word vector models comprises medical history texts which are used for converting each word in the word list into a word vector by using at least 2 preset word vector models.

Specifically, the method for obtaining the training corpus of the word vector model according to the medical history text comprises the following steps: the medical record text is a medical record report, all collected medical record texts are subjected to word segmentation, the medical record texts are converted into word lists, and training linguistic data (word lists) of the word vector model are obtained. That is, all collected medical record texts are segmented by using a segmentation tool (such as a jieba segmentation tool) and a self-defined dictionary in the segmentation tool, a stop word list is set, stop words of the medical record texts are filtered by using the stop word list, and the stop words can be added into an existing stop word list according to specific scenes and are set to be removed. Stop words such as mood words, auxiliary words and/or punctuation marks. The medical record text is subjected to a word segmentation tool to obtain a word list.

Specifically, the method for inputting the training corpus into a preset vector model respectively and training the preset vector model comprises the following steps: the word vector model can select the existing models, such as a word2vec model and a GloVe model, and train the word2vec model and the GloVe model. And then converting each word in the word list into a word vector by utilizing at least 2 trained preset word vector models.

And the set of medical history texts for training the preset word vector models comprises medical history texts which are used for converting each word in the word list into a word vector by using at least 2 preset word vector models.

Step 1003, converting the word vectors into text vectors respectively:

the word vector is converted into a text vector by the above formula. Wherein, the

For all words in the word list

The set of word vectors of (a) is,

respectively representing a word vector for each word in a word list, n representing the number of words in said word list, (i) representing the number of word vector models, max () representing a maximum value, min () representing a minimum value, mean () representing a mean value, h () representing the operation of converting a word vector into a text vector,

a splice is indicated.

In particular, the first and second (c) substrates,

respectively obtaining all word minimum values, all word mean values and all word maximum values through minimum value taking, mean value taking and maximum value taking, and then splicing all word minimum values, all word mean values and all word maximum values to obtainA text vector.

The number of the text vectors is the same as that of the word vector models, that is, if 2 word vector models are adopted, each word vector model can obtain one text vector, and 2 word vector models can obtain 2 text vectors.

Step 1004 synthesizes all the text vectors to obtain synthesized text vectors:

and if 2 word vector models are adopted, splicing the 2 text vectors obtained by the 2 word vector models. And then the synthetic text vector is input into a classification model, the classification model completes the classification of the medical record text, and the classification model is a preset diagnosis model.

In the embodiment disclosed by the invention, the number of the preset word vector models is 2, and/or the preset word vector models are respectively a word2vec model and a GloVe model. Specifically, a medical record text is obtained, and the medical record text is converted into a word list; converting each word in the word list into a first word vector and a second word vector by using a word2vec model and a GloVe model respectively; the first word vector and the second word vector are word vectors derived from word lists derived from the same medical history text. Then, synthesizing (for example, splicing) the first word vector and the second word vector to obtain a synthesized text vector; the synthetic text vector is input into a classification model, and the classification model completes classification of the medical record text.

The invention also provides a device for representing the medical record text vector, which comprises: the system comprises a word list conversion unit, at least 2 preset word vector models, a text vector conversion unit and a synthesis unit, wherein the output of the word list conversion unit is connected with the input of the preset word vector models respectively, the output of the preset word vector models is connected with the input of the text vector conversion unit, and the output of the text vector conversion unit is connected with the input of the synthesis unit; the word list conversion unit is used for acquiring a medical record text and converting the medical record text into a word list; the preset word vector model converts each word in the word list into a word vector respectively; the text vector conversion unit is used for respectively converting the word vectors into text vectors; the synthesis unit synthesizes the text vectors to obtain synthesized text vectors; and the synthetic text vector is input into a classification model, and the classification model completes the classification of the medical record text. The detailed description can be seen in a method for representing a medical record text vector.

The synthesis unit converts the word vector into a text vector, and the obtained text vector completes the following operations:

For all words in the word list

The set of word vectors of (a) is,

a splice is indicated.

An apparatus for representing medical history text vectors, further comprising a training unit; the at least 2 preset word vector models are obtained by training the medical record text, then each word in the word list is converted into a word vector by using the at least 2 preset word vector models, and the training unit completes the following operations: obtaining a training corpus of a word vector model according to the medical record text; respectively inputting the training corpora into a preset vector model, and training the preset vector model; the set of medical history texts for training the preset word vector models comprises medical history texts which are used for converting each word in the word list into a word vector by using at least 2 preset word vector models.

The number of the preset word vector models is 2, and/or the preset word vector models are respectively a word2vec model and a GloVe model.

The invention also provides a device for representing the medical record text vector, which comprises: a processor; a memory for storing processor-executable instructions; wherein the processor is configured to: the above-described method is performed.

The invention also provides an inquiry system, comprising: sending the inquiry text to a preset text analysis unit, identifying the inquiry text through the preset text analysis unit to obtain the intention and word groove of each text in the inquiry text, searching the position information of the intention and the word groove in a preset database according to the intention and the word groove, matching preset key information by using the position information to generate a medical record report, sending the medical record report to a preset diagnosis model, and giving a diagnosis result by the preset diagnosis model according to the medical record report; and/or

The construction process of the preset diagnosis model comprises the following steps: segmenting words of a given medical record report with diagnosis, training the given medical record report with the segmented words to obtain a text word vector, representing the text vector according to the inquiry text word vector, and training the diagnosis model by using the text vector to obtain the preset diagnosis model; sending the medical record report to the preset diagnosis model, wherein the preset diagnosis model gives a diagnosis result according to the medical record report; the method for representing the text vector is the method for representing the medical record text vector; or the text vector is generated using the means for representing medical record text vectors as described above.

More specifically, the method or process for presetting a text analysis unit as an intention word slot model, constructing the intention word slot model through a training text of the inquiry text, and identifying the inquiry text to obtain the intention and word slot of each text in the inquiry text comprises the following steps: the inquiry text which accords with the set response rule is segmented, the inquiry text after the words are correctly segmented can be obtained by using an open source word segmentation tool (jieba word segmentation tool) and a self-defined dictionary mode in the word segmentation tool, and the segmentation can be specifically detailed in a method for obtaining the inquiry text by vertigo or a detailed description in a device for obtaining the inquiry text by vertigo. Marking intentions and word grooves in the inquiry text after word segmentation, wherein the intentions and word grooves are marked manually, then marking the inquiry text with the intention word grooves as a training text, training an intention word groove model through the training text, and enabling the intention word groove model to automatically identify the inquiry text, wherein the identification comprises the following steps: intent recognition and word slot filling.

The intent-Based RNN model of the present invention employs an Attention-Based RNN model to identify intent and slot filling (slot filling). The Attention-Based RNN model is used to classify each text into a corresponding intent category (intent recognition) and to sequence label (word slot filling) the words of each text, i.e., to label the keywords in a given sentence with corresponding labels. For example, a doctor: "do you have headache before? "this phrase is intended to be a history of illness and the word slot fill is headache. The patients: "none". In the patient's answer, the intention was to identify no headache and the word groove was not filled.

More specifically, the method of recognizing the speaker tag may be a method of obtaining an inquiry text for vertigo or a speaker tag obtained by an apparatus for obtaining an inquiry text for vertigo.

And respectively recording the intention and the word slot of the inquiry text according to the speaker label, and constructing the preset database according to the recording result by the method or the process comprising the following steps: in the process of constructing the intention word slot model through the training text of the inquiry text, the intention and the word slot marked by a doctor and a patient are obtained, the intention and the word slot marked by the doctor and the patient are recorded respectively, the question intention and the word slot, the response intention and the word slot are obtained respectively, and the recording result is the question intention and the word slot, the response intention and the word slot. The question intention and the word slot, the response intention and the word slot are respectively recorded in four tables and have corresponding position Information (ID), a key information matching table (table 5) is created, the first four columns of the key information matching table store the question intention and the 4 pieces of position Information (ID) of the word slot, the response intention and the word slot, the last column stores the key information to be output, and the key information is used for filling a medical record report to complete the construction of the preset database. Wherein, the four tables are a table 1 question intention recording table, a table 2 question word groove recording table, a table 3 answer intention recording table and a table 4 answer word groove recording table respectively.

And searching the position information of the intention and the word slot in a preset database according to the intention and the word slot, and generating a medical record report by matching preset key information with the position information, wherein the method or the process comprises the following steps: the intention word slot model identifies each text in the inquiry text to obtain the intention and the word slot of each text, and inquires key information (matching preset key information) according to the intention and 4 position Information (ID) of the word slot in the database to generate a medical record report. An example of the question-answering intention slot record table and the key information matching table in the database is shown below. Wherein, the intention and word slot are the intention of the question, the word slot of the question, the intention of the response and the word slot of the response.

TABLE 1 problem intention recording Table

TABLE 2 problem word groove record table

Word slot ID of question	Word slot of question
		2	Headache (headache)
3	Dizziness (lightheadedness)
		4	Hearing reduction
5	More recently, the development of new and more recently developed devices

TABLE 3 answer intention recording sheet

TABLE 4 answer word groove recording table

Word slot ID of the response	Word slot for response
		2	Worship
3	Insomnia
		4	For a few seconds
5	Vertigo (vertigo)

TABLE 5 Key information matching Table

A method or a process for generating a medical record report by matching the position information with preset key information is described in detail with reference to tables 1 to 5: tables 1-5 are stored in the preset database, and the position information of the intention of the question and the position information of the word slot of the question are obtained from the tables 1-2, and then the position information of the intention of the response and the position information of the word slot of the response are obtained from the tables 3-4. And matching according to the position information of the intention of the question, the position information of the word slot of the question, the position information of the answer intention, the position information of the answer word slot and the first four columns of the table 5, and if the matching is successful, outputting the preset key information of the last column of the table 5. If the preset key information of 'dizziness continuously attacks one worship' is output, the position information of the intention of the question, the position information of the word slot of the question, the position information of the intention of the response and the position information of the word slot of the response are 3, 5 and 2 in sequence, and the preset key information of 'dizziness continuously attacks one worship' is one item of a medical record report.

The construction process of the preset diagnosis model comprises the following steps: the method comprises the steps of segmenting words of a given medical record report with diagnosis, training the given medical record report with the segmented words to obtain a text word vector, representing the text vector according to the text word vector, and training a diagnosis model by using the text vector to obtain a preset diagnosis model. Text vector is the sum of text word vectors/text length.

Specifically, the disease diagnosis model is a classification model, i.e., a disease classification model, which belongs to a multi-classification model because of more than two types of diseases, and can make a disease diagnosis for medical record reports. The present invention trains 4 two classification models, each of which can classify one disease. For example, for otolithiasis, m medical record reports are selected as positive examples; n medical record reports are hierarchically sampled from medical record reports of three diseases of Meniere's disease, vestibular neuritis and vestibular migraine. Where m and n may be the same number, but the number is preferably not too large. Wherein, the layered sampling means that n/3 medical record reports are respectively extracted from three diseases of Meniere disease, vestibular neuritis and vestibular migraine, and are mixed into n parts in a disorderly way as negative examples; and (3) training the m + n medical record reports, wherein the label of the positive case is 1, the label of the negative case is 0, and the m + n medical record reports are used as training data to train a disease classification model capable of identifying otolithiasis diseases. The other three diseases were similarly operated as above.

In the input text processing of the disease classification model, characters cannot be directly input to the model, and the text needs to be converted into a vector form. Firstly, segmenting words and removing stop words from the text content of the medical record report, and training a Word2vec model by using the processed medical record report text, wherein the Word2vec is a tool for Word vector calculation which is open source of Google. Word2vec can train efficiently on millions of orders of magnitude dictionaries and billions of datasets; secondly, the training result obtained by the tool, word vector (word embedding), can well measure the similarity between words. The Word2vec model may derive a Word vector specifying a Word, and the dimensionality of the Word vector may be specified as either 50-or 100-dimensions at the time of training.

The text vector of each medical record report can be obtained by the following method: the text vector is the sum of text word vectors/text length; or the text vector representation method comprises the following steps: acquiring a medical record text, and converting the medical record text into a word list; converting each word in the word list into a word vector by using at least 2 preset word vector models; respectively converting the word vectors into text vectors; synthesizing the text vector to obtain a synthesized text vector; the synthetic text vector is input into a classification model, and the classification model completes classification of the medical record text; or the text vector is generated by adopting a device for representing the medical record text vector, and the device for representing the medical record text vector comprises the following steps: the system comprises a word list conversion unit, at least 2 preset word vector models, a text vector conversion unit and a synthesis unit, wherein the output of the word list conversion unit is connected with the input of the preset word vector models respectively, the output of the preset word vector models is connected with the input of the text vector conversion unit, and the output of the text vector conversion unit is connected with the input of the synthesis unit; the word list conversion unit is used for acquiring a medical record text and converting the medical record text into a word list; the preset word vector model converts each word in the word list into a word vector respectively; the text vector conversion unit is used for respectively converting the word vectors into text vectors; the synthesis unit synthesizes the text vectors to obtain synthesized text vectors; the synthetic text vector is input into a classification model, and the classification model completes classification of the medical record text; or the text vector is generated by adopting a device for representing the medical record text vector, and the device for representing the medical record text vector comprises the following steps: a processor; a memory for storing processor-executable instructions; wherein the processor is configured to perform the steps of: acquiring a medical record text, and converting the medical record text into a word list; converting each word in the word list into a word vector by using at least 2 preset word vector models; respectively converting the word vectors into text vectors; synthesizing the text vector to obtain a synthesized text vector; and the synthetic text vector is input into a classification model, and the classification model completes the classification of the medical record text.

And for medical record reports with different text lengths, text vectors with the same dimension can be obtained.

The disease classification model can be a machine learning stacking method, in the machine learning stacking method, a support vector machine, a Bayesian model and a decision tree model can be selected as a first basic learner model1, a second basic learner model2 and a third basic learner model3, and a linear regression model can be selected as a high-level learner. The training data set of m + n medical record reports is divided into two disjoint sets, namely a Train set and a Test set, and the Train set is divided into 5 sets of Train 1-Train 5. The remaining 4 of these were used to train the first base learner model 1; while predicting the test data, such a process is performed 5 times, and 5 copies of the trace data (original trace sample number/5) and 5 copies of the test data are generated. And then, longitudinally stacking the predicted values of the 5 train data to obtain a first training set P1 of the high-level learner, and averaging the results of test prediction to obtain a first test set P1 of the high-level learner. The training of the second basic learner model2 and the third basic learner model3 is the same as that of the training model1, and a second training set P2 of the high-level learner, a second Test set P2 of the high-level learner, a third training set P3 of the high-level learner and a third Test set P3 of the high-level learner are obtained, respectively, the first training set P1 of the predictive value learner from the Train set, the second training set P2 of the high-level learner and the third training set P3 of the high-level learner are transversely spliced for the training of the high-level learner, and the first Test set P1 of the predictive value learner from the Test set, the second Test set P2 of the high-level learner and the third Test set P3 of the high-level learner are transversely spliced and can be predicted by the high-level learner.

The 4 binary classification models can be trained by the method, and 4 diseases can be classified respectively. After the model is trained, the 4 models are used for predicting the input medical record text vectors respectively to obtain the prediction probability of each disease, and the prediction value with the maximum probability value is the diagnosed disease.

The invention directly utilizes the inquiry audio of doctors and patients to make the patients obtain diagnosis in the chat with the doctors. The existing mature voice recognition technology and the speaker recognition technology are organically combined and applied to an inquiry system, or the voice recognition technology is applied to the inquiry system, semantic analysis and key information extraction are carried out on the obtained inquiry text with speaker labels, and finally a standard medical record report is generated and diagnosis is given.

The above-mentioned embodiments are merely embodiments for expressing the invention, and the description is specific and detailed, but not construed as limiting the scope of the invention. It should be noted that, for those skilled in the art, various changes, substitutions of equivalents, improvements and the like can be made without departing from the spirit of the invention, and these are all within the scope of the invention. Therefore, the protection scope of the present patent shall be subject to the appended claims.

Claims

1. A method for obtaining an inquiry text for vertigo, comprising:

converting the questions and responses into an interrogation text;

wherein the question and the answer are voice signals.

2. The method of claim 1, wherein the questions and responses are converted into an interview text by:

determining a start and an end of the question and a start and an end of the response, respectively;

performing speech recognition and speaker recognition on the question and the response having a beginning and an end, respectively; the voice recognition is used for converting the questions and the responses into inquiry texts; the speaker identification identifies the question and the speaker tag of the response;

3. The method according to claim 1 or 2, wherein the method for acquiring the problems posed by doctors for vertigo and the responses of patients to the problems in real time is as follows:

obtaining in real time a spoken question of a doctor and a spoken response of a patient to the spoken question in real time; and/or

Determining the patient's response to the question prior to converting the question and the response to an interrogation text; if the answer rule is met, the question and the answer are reserved, otherwise, the question and the answer are deleted.

4. The method of claim 3, wherein the determination of whether the set response rule is satisfied is performed by: extracting key features of the question and key features of the response respectively, calculating the similarity between the key features of the response and the key features of the question, and judging the size of the similarity and the set value;

if the similarity is larger than the set value, the set response rule is considered to be met, otherwise, the set response rule is considered not to be met.

5. An apparatus for obtaining an interrogation text for vertigo, comprising: the device comprises an acquisition unit, a conversion unit and a storage or storage sending unit;

wherein the question and the answer are voice signals.

6. The apparatus of claim 5, wherein the conversion unit comprises: the device comprises a determining unit, a voice recognition unit and a speaker recognition unit;

7. The apparatus of claim 5 or 6, wherein the acquisition unit comprises: a recording or speech recognition unit;

the recording or voice recognition unit is used for acquiring the dictation questions of doctors in real time and acquiring the dictation responses of patients to the dictation questions in real time;

and/or a judging unit is arranged between the conversion units;

8. The apparatus of claim 7, wherein the determining unit comprises: a first extraction unit, a second extraction unit and a calculation unit;

the first extraction unit extracts key features of the problem;

the second extraction unit extracts key features of the response;

9. An electronic device, comprising:

a processor;

a memory for storing processor-executable instructions;

wherein the processor is configured to: performing the method of any one of claims 1 to 4.

10. A vertigo interrogation system, comprising: the method of any one of claims 1 to 4; or

The apparatus of any one of claims 5 to 8; or

The electronic device of claim 9;