CN111221947A - Multi-round conversation realization method and device of ophthalmologic pre-inquiry device - Google Patents
Multi-round conversation realization method and device of ophthalmologic pre-inquiry device Download PDFInfo
- Publication number
- CN111221947A CN111221947A CN201911365885.4A CN201911365885A CN111221947A CN 111221947 A CN111221947 A CN 111221947A CN 201911365885 A CN201911365885 A CN 201911365885A CN 111221947 A CN111221947 A CN 111221947A
- Authority
- CN
- China
- Prior art keywords
- information
- question
- history information
- patient
- ophthalmic
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 56
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 claims abstract description 54
- 201000010099 disease Diseases 0.000 claims abstract description 53
- 230000008569 process Effects 0.000 claims description 14
- 238000013461 design Methods 0.000 abstract description 25
- 208000024891 symptom Diseases 0.000 description 37
- 238000012549 training Methods 0.000 description 22
- 238000002372 labelling Methods 0.000 description 6
- 238000011161 development Methods 0.000 description 4
- 230000006870 function Effects 0.000 description 4
- 229940023490 ophthalmic product Drugs 0.000 description 4
- 206010027146 Melanoderma Diseases 0.000 description 3
- 230000004456 color vision Effects 0.000 description 3
- 238000012938 design process Methods 0.000 description 3
- 238000013507 mapping Methods 0.000 description 3
- 230000008961 swelling Effects 0.000 description 3
- 206010061818 Disease progression Diseases 0.000 description 2
- 206010061218 Inflammation Diseases 0.000 description 2
- 206010027145 Melanocytic naevus Diseases 0.000 description 2
- 208000007256 Nevus Diseases 0.000 description 2
- 206010047571 Visual impairment Diseases 0.000 description 2
- 230000005856 abnormality Effects 0.000 description 2
- 210000004087 cornea Anatomy 0.000 description 2
- 230000005750 disease progression Effects 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 230000004054 inflammatory process Effects 0.000 description 2
- 230000003993 interaction Effects 0.000 description 2
- 201000004614 iritis Diseases 0.000 description 2
- 239000000463 material Substances 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 208000035824 paresthesia Diseases 0.000 description 2
- 208000029257 vision disease Diseases 0.000 description 2
- 230000000007 visual effect Effects 0.000 description 2
- 230000004393 visual impairment Effects 0.000 description 2
- 201000004384 Alopecia Diseases 0.000 description 1
- 206010011033 Corneal oedema Diseases 0.000 description 1
- 206010030113 Oedema Diseases 0.000 description 1
- 206010047642 Vitiligo Diseases 0.000 description 1
- 230000009471 action Effects 0.000 description 1
- 208000004631 alopecia areata Diseases 0.000 description 1
- 230000004075 alteration Effects 0.000 description 1
- 210000004556 brain Anatomy 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 201000004778 corneal edema Diseases 0.000 description 1
- 238000012217 deletion Methods 0.000 description 1
- 230000037430 deletion Effects 0.000 description 1
- 238000003745 diagnosis Methods 0.000 description 1
- 230000037213 diet Effects 0.000 description 1
- 235000005911 diet Nutrition 0.000 description 1
- 208000035475 disorder Diseases 0.000 description 1
- 208000030533 eye disease Diseases 0.000 description 1
- 208000024963 hair loss Diseases 0.000 description 1
- 230000003676 hair loss Effects 0.000 description 1
- 210000003128 head Anatomy 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 230000001717 pathogenic effect Effects 0.000 description 1
- 239000000049 pigment Substances 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 230000037380 skin damage Effects 0.000 description 1
- 206010040882 skin lesion Diseases 0.000 description 1
- 231100000444 skin lesion Toxicity 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000003813 thin hair Effects 0.000 description 1
- 230000002087 whitening effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/33—Querying
- G06F16/332—Query formulation
- G06F16/3329—Natural language query formulation or dialogue systems
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Theoretical Computer Science (AREA)
- Artificial Intelligence (AREA)
- Human Computer Interaction (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Eye Examination Apparatus (AREA)
Abstract
The invention discloses a multi-round conversation realization method and a multi-round conversation realization device of an ophthalmic pre-inquiry device, wherein the method comprises the following steps: sequentially collecting basic information, current medical history information and past history information of a patient; predicting disease information of the patient according to the basic information, the current medical history information, the past medical history information and the ophthalmologic medical knowledge base; collecting the question-following information of the patient, generating question-answering information according to the question-following information and the spoken language expression table of the user, and generating an information list. The method adopts the whole serial and partial parallel dialogue design flow, thereby better meeting the scene requirement, greatly improving the flow realization speed and reducing the marking cost.
Description
Technical Field
The invention relates to the technical field of voice recognition, in particular to a method and a device for realizing multi-turn conversation of an ophthalmic pre-inquiry device.
Background
Related art, there are two main ways to realize multi-turn dialog:
1. task-based multi-round dialogue realization based on NLU DM NLG
The task type robot core module mainly comprises three modules: natural language understanding module (NLU) dialog management module natural language generation module (NLG). Among them, natural language understanding and dialogue management are core. Natural language understanding is used to understand user utterances, meaning that the user utterances are parsed in different semantic representations. When the user language passes through the natural language understanding module, namely the field identification, the user intention identification and the slot extraction are needed. The method comprises the following steps of (1) field identification, namely identifying whether a statement belongs to a task scene, wherein when a plurality of robots are generally integrated, such as a chatting robot, a question and answer robot and the like, the field identification needs to be judged and distributed before entering a task robot; intention identification, namely identifying the intention of a user, and subdividing the sub-scenes in the task-based scene; and entity identification and slot filling are used for inputting the dialogue management module. The dialogue management module corresponds to the brain of the task robot and is divided into dialogue state maintenance DST + action candidate ordering Policy. The two parts form a multi-turn dialog experience between the human and the machine. DM is largely divided into two functions, one to record user historical utterances and another to generate system decisions. The triplet output of the natural language understanding module will be the input to the dialog management system. The state tracking module includes various information of the continuous conversation, and updates the current conversation state according to the old state, the user state (i.e. the above-mentioned triplets) and the system state (i.e. through the inquiry condition with the database). The conversation strategy is closely related to the scene of the task, and is usually used as the output of a conversation management module, such as a question-back strategy for missing slots in the scene.
2. End-to-end method for constructing dialogue model
Another technical solution for implementing a medical inquiry session is to use a model method of end-to-end generation (Seq2 Seq). Such as: an end-to-end task-type dialog system comprises a data preprocessing module, a named body extraction module, a compiling module, a dialog history encoder module and a decoding output module. Although such an end-to-end dialogue system can reduce the tedious manual rules and reduce the data volume of the training model, it is difficult to cope with the dialogue logic of the medical guide. The adoption of an end-to-end system introduces a certain uncertainty and inexplicability, so that each round of conversation cannot be strictly based on the previously designed flow, and problems such as logic disorder or repeated inquiry are caused.
Disclosure of Invention
The present invention is directed to solving, at least to some extent, one of the technical problems in the related art.
Therefore, one objective of the present invention is to provide a multi-round dialog implementation method for an ophthalmic pre-interrogation apparatus, which adopts an overall serial and partially parallel dialog design process, so as to better meet the requirements of the scene, greatly increase the process implementation speed and reduce the labeling cost.
Another object of the present invention is to provide a multi-turn dialogue-implementing apparatus of an ophthalmic pre-interrogation apparatus.
In order to achieve the above object, an embodiment of an aspect of the present invention provides a method for implementing multiple rounds of dialog of an ophthalmic pre-interrogation apparatus, including the following steps: sequentially collecting basic information, current medical history information and past history information of a patient; predicting disease information of the patient according to the basic information, the current medical history information, the past medical history information and an ophthalmologic medical knowledge base; and acquiring the question-following information of the patient, generating question-answering information according to the question-following information and a spoken language expression table of the user, and generating an information list.
According to the multi-round conversation implementation method of the ophthalmic pre-inquiry device, a set of conversation design flows for medical pre-inquiry are provided by fusing a model and template decision scheme, and a set of model data design specifications for medical inquiry are designed from the conversation flows to training data of each sub-model module, so that the ophthalmic pre-inquiry device adopts an overall serial and partially parallel conversation design flow, and the system is enabled to be more efficient and better suitable for medical inquiry tasks; the data format of each part is designed aiming at the whole inquiry flow, and the complete system presents various designs required by the whole ophthalmic inquiry dialogue; the method for generating the training data is innovatively introduced, the training of the algorithm module can be completed under the limited labeled data, and the method has applicability and high efficiency.
In addition, the multi-turn dialogue implementation method of the ophthalmic pre-interrogation apparatus according to the above embodiment of the present invention may further have the following additional technical features:
further, in an embodiment of the present invention, the generating answer information according to the question-following information and the user spoken language expression table includes: obtaining a question-following corpus from the question-following information; inputting the question-chasing corpus into a trained NL2SQL model so as to directly convert the question into an SQL query statement; and determining the question answering information according to the SQL query statement.
Further, in an embodiment of the present invention, the method further includes: matching an ending keyword from the question information; and if the matching is successful, ending the conversation process.
Further, in an embodiment of the present invention, the predicting the disease information of the patient based on the basic information, the present history information, the past history information, and the ophthalmic medicine knowledge base includes: all intents are obtained, and for each intention, a pre-written slot value is selected and filled into the data template.
Further, in an embodiment of the present invention, the predicting the disease information of the patient based on the basic information, the present history information, the past history information, and the ophthalmic medicine knowledge base includes: and matching the information with the content in the ophthalmic medical knowledge base by using a Rank model.
In order to achieve the above object, according to another embodiment of the present invention, there is provided a multi-turn dialogue realization apparatus for an ophthalmic pre-interrogation apparatus, including: the acquisition module is used for sequentially acquiring basic information, current medical history information and past history information of a patient; a prediction module for predicting disease information of the patient based on the basic information, the present history information, the past history information and an ophthalmic medical knowledge base; and the generating module is used for acquiring the question-following information of the patient, generating answer information according to the question-following information and the spoken language expression table of the user and generating an information list.
The multi-round conversation realization device of the ophthalmic pre-inquiry device provided by the embodiment of the invention integrates a model and template decision scheme, provides a set of conversation design flow aiming at medical pre-inquiry, and designs a set of model data design specifications of medical inquiry from the conversation flow to the training data of each sub-model module, so that the ophthalmic pre-inquiry device adopts the whole serial and partially parallel conversation design flow, and the system is more efficient and better suitable for medical inquiry tasks; the data format of each part is designed aiming at the whole inquiry flow, and the complete system presents various designs required by the whole ophthalmic inquiry dialogue; the method for generating the training data is innovatively introduced, the training of the algorithm module can be completed under the limited labeled data, and the method has applicability and high efficiency.
In addition, the multi-turn dialogue realization device of the ophthalmic pre-interrogation apparatus according to the above-described embodiment of the present invention may further have the following additional technical features:
further, in an embodiment of the present invention, the generating module is further configured to obtain a query corpus from the query information; inputting the question-chasing corpus into a trained NL2SQL model so as to directly convert the question into an SQL query statement; and determining the question answering information according to the SQL query statement.
Further, in an embodiment of the present invention, the method further includes: and the matching module is used for matching the finishing keywords from the question tracing information and finishing the conversation process when the matching is successful.
Further, in an embodiment of the present invention, the prediction module is further configured to obtain all intents, and for each intention, select and fill in the data template from the pre-written slot values.
Further, in an embodiment of the present invention, the prediction module is further configured to use a Rank model to perform short sentence matching between the information and the content in the ophthalmic medical knowledge base.
Additional aspects and advantages of the invention will be set forth in part in the description which follows and, in part, will be obvious from the description, or may be learned by practice of the invention.
Drawings
The foregoing and/or additional aspects and advantages of the present invention will become apparent and readily appreciated from the following description of the embodiments, taken in conjunction with the accompanying drawings of which:
FIG. 1 is a flow chart of a method for implementing a multi-turn dialog of an ophthalmic pre-interrogation apparatus in accordance with an embodiment of the present invention;
FIG. 2 is a flow chart of an interrogation according to an embodiment of the present invention;
FIG. 3 is a flow chart of disease prediction according to an embodiment of the present invention;
fig. 4 is a schematic structural view of a multi-turn dialogue realization apparatus of an ophthalmic pre-interrogation apparatus according to an embodiment of the present invention.
Detailed Description
Reference will now be made in detail to embodiments of the present invention, examples of which are illustrated in the accompanying drawings, wherein like or similar reference numerals refer to the same or similar elements or elements having the same or similar function throughout. The embodiments described below with reference to the drawings are illustrative and intended to be illustrative of the invention and are not to be construed as limiting the invention.
The present application is based on the recognition and discovery by the inventors of the following problems:
the ophthalmic pre-inquiry device achieves the purpose of collecting the data of the patient before diagnosis by carrying out multiple rounds of dialogue interaction with the user. The device needs to design a set of overall multi-round inquiry modes, including basic information inquiry, chief complaint inquiry, disease history inquiry, disease development inquiry, user answer and the like. These procedures place high demands on model design and training data construction.
In order to solve the problems that a single dialogue model is difficult to process complex interaction in the pre-inquiry process and the like, the embodiment of the invention provides a whole serial and partially parallel dialogue design process. The whole dialogue logic is based on the flow of basic information, chief complaint information, physiological status, medical history information, disease development, disease prediction and related questions, user's answering and information list generation. This way, the information of the complete user can be effectively collected and the conversation process can be controlled. Meanwhile, in order to ensure the conversation diversity, a parallel module is used in each independent part for process management, for example, in a disease development module, the method determines which aspect to inquire according to the context, thereby improving the diversity of the conversation process.
Meanwhile, training the labeled data of the model is a large cost in the real land project. In order to more effectively utilize the existing labeling data, the embodiment of the invention designs a set of data formats aiming at the ophthalmic pre-inquiry problem, thereby completing model training with the minimum labeling cost. The method comprises the following three parts: 1) sequence annotation data; 2) short text similarity data; 3) NL2SQL data. As will be explained in detail below in the implementation section.
Aiming at two problems of difficult flow design and difficult annotation data acquisition in the ophthalmic pre-inquiry device, the embodiment of the invention provides a method and a device for realizing multi-turn conversation of the ophthalmic pre-inquiry device, which greatly improve the flow realization speed and reduce the annotation cost, can be used for ophthalmic inquiry, and can be easily transferred to other medical inquiry scenes after some simple deletion and modification.
Hereinafter, a method and an apparatus for implementing a multi-turn dialog of an ophthalmic pre-interrogation apparatus according to an embodiment of the present invention will be described with reference to the accompanying drawings, and first, a method for implementing a multi-turn dialog of an ophthalmic pre-interrogation apparatus according to an embodiment of the present invention will be described with reference to the accompanying drawings.
Fig. 1 is a flowchart of a multi-session implementation method of an ophthalmic pre-interrogation apparatus according to an embodiment of the present invention.
As shown in fig. 1, the multi-turn dialogue implementation method of the ophthalmic pre-interrogation apparatus includes the following steps:
in step S101, basic information, present history information, and past history information of the patient are sequentially collected.
It can be understood that the inquiry dialogue has the characteristic of fixed large-problem flow based on the collection and arrangement of the common inquiry records of the ophthalmology department. As shown in fig. 2, the inquiry process mainly includes five modules, i.e., basic information, current medical history, past medical history, disease prediction, and patient inquiry, and the inquiry is ordered, and the sub-module sequence of the specific inquiry in the inquiry is not regular. Therefore, the embodiment of the invention adopts an external serial-internal parallel mode to design a man-machine inquiry dialogue, namely the whole flow is ordered, and the inquiry sequence of the module is determined according to the information missing condition provided by the user.
In step S102, disease information of the patient is predicted based on the basic information, the present history information, the past history information, and the ophthalmic medicine knowledge base.
It is understood that, aiming at the specialty and complexity of the ophthalmic medical data, the embodiment of the present invention constructs the ophthalmic medical knowledge base, which is divided into 11 fields and corresponding values, as shown in table 1, wherein table 1 is the ophthalmic disease knowledge base.
TABLE 1
In addition, embodiments of the invention perform entity matching data for the relevant symptoms, wherein the relevant symptoms are the focus of the predicted disease. To extract relevant symptoms from the chief complaints and stages of development entered by the user, embodiments of the present invention label 509 eye disease symptoms in the knowledge base with a common spoken user expression for each symptom. As shown in table 2, table 2 is a spoken language expression table of the relevant symptoms.
TABLE 2
Further, in one embodiment of the present invention, predicting disease information of a patient based on the basic information, the present history information, the past history information, and the ophthalmic medicine knowledge base comprises: acquiring all intents, selecting each intention from the pre-written slot values, and filling the selected intention into a data template; and matching the information with the content in the ophthalmologic medical knowledge base by using a Rank model.
It can be understood that, in order to meet the requirements on the algorithm function in the dialog system, the embodiment of the present invention designs named entity recognition (NLU), a phrase matching model and an NL2SQL model as the main functions of the algorithm engine. Because the acquisition cost of the labeled data is too high, the training data is generated by adopting a mode based on an external knowledge template, and the training cost is greatly reduced. The NLU model training data and the phrase matching model will be described in detail below, and the NL2SQL model will be described later.
Specifically, (1) NLU model training data
The NLU model is responsible for extracting the desired bin values from the user's input, and training an entity recognition model requires a large amount of sequence labeling data. In order to reduce the cost of manually labeling data, the embodiment of the invention trains data by adopting a data generation mode based on a data template, and as shown in table 3, table 3 is an NLU model training data table. Where for each intent, fill-in to the data template is chosen from the pre-written slot values (here,% represents the fill-in location).
TABLE 3
(2) Short sentence matching model
Since the expression of the disease name or the related symptom by the user is not very accurate, a phrase matching model is required to perform phrase matching between the expression of the user and the content in the content library. The embodiment of the invention adopts a Rank model and uses a Pairwise mode to make positive and negative sample to carry out similarity calculation. Each user expression is used as a keyword, the corresponding related symptom is used as a positive example, one related symptom is randomly selected as a negative example, and a triple is made to train the short sentence matching model, as shown in table 4, table 4 is a short sentence matching model table.
TABLE 4
Associated symptoms | Expression 1 | Expression 2 | Expression 3 | Expression 4 | Expression 5 |
Corneal edema | Swelling of the cornea | Swelling of the eye | Corneal swelling | Inflammation of the cornea | Edema of the eyes |
Pigment skin damage | Vitiligo | White spot | White spot | White skin lesions | Whitening of skin |
Blind spot | Invisible dots | Point of missing object | Blind area | Point of visual field loss | Blind spot of visual field |
Black nevus | Black nevus | Black spot | Black plaque | Black spot | Black spot |
Inflammation of iris | Inflammation of the eye | Iris inflammation | Red and swollen iris | Iris pain | Iris pain |
Pillow for baby | Hair loss | The hair on the pillow part is less | Thinning hair | Alopecia areata | Lack of hair in the head |
Loss of colour vision | Invisible color | Can not see color | Loss of color vision | No color is present | Without color perception |
Further, based on the mapping data of a large number of diseases and related symptoms, the embodiment of the present invention may use the following two strategies for prediction, specifically as follows:
strategy 1: selecting the most specific symptom
Assuming that the probability of disease is significantly different for each related symptom and labeled with a distinguishing mark, for example, one symptom can set the probability of disease to be divided into three scores of 3, 2 and 1 from high to low.
Diseases containing symptom(s) are selected as candidates by extracting symptom descriptions in the chief complaints, disease progression, visual impairment, paresthesia, and appearance abnormalities and mapping their corresponding associated symptoms.
And selecting the relevant symptoms with the highest pathogenic probability from the relevant symptoms of the candidate diseases to ask. If the user answers yes, adding a score corresponding to the disease rate for the disease with the symptom, and removing the disease without the symptom from the candidate set; if not, the disease with the symptom is filtered out, so that the condition is not obvious and the patient can make wrong judgment.
Finally, the three diseases are filtered out as the prediction results, wherein the higher the score, the higher the disease prevalence probability.
Strategy 2: selecting the symptoms with the most frequent appearance
The probability of causing disease for each relevant symptom of the disease is assumed to be similar.
Diseases containing symptom(s) are selected as candidates by extracting symptom descriptions in the chief complaints, disease progression, visual impairment, paresthesia, and appearance abnormality and mapping their corresponding associated symptoms, including one associated symptom plus 1.
The symptoms with the largest number of related symptoms of the candidate diseases are selected and asked. If the user answers yes, adding 1 point to the disease with the symptom, and removing the disease without the symptom from the candidate set; if not, the disease with the symptom is filtered out, so that the condition is not obvious and the patient can make wrong judgment.
Finally, the three diseases are filtered out as the prediction results, wherein the higher the score, the higher the disease prevalence probability.
The first is based on the differentiated disease symptom data, and the second is applicable to the undifferentiated disease symptom data. Therefore, the embodiment of the invention can tentatively adopt the strategy two to predict the diseases.
In addition, the flow of disease prediction is shown in fig. 3, wherein the specific pseudo code of the main program of the disease prediction strategy algorithm is as follows:
traversing the candidate question-hunting symptom dictionary S to find the symptom pmax with the maximum key value array length and not belonging to userSym for question hunting
In step S103, the question-following information of the patient is collected, and the answer information and the information list are generated according to the question-following information and the spoken language expression table of the user.
Further, in an embodiment of the present invention, generating the question answering information according to the question answering information and the spoken language expression table of the user includes: obtaining a question-following corpus from the question-following information; inputting the query corpus into a trained NL2SQL model so as to directly convert the question into an SQL query statement; and determining the question answering information according to the SQL query statement.
It will be appreciated that after the disease is predicted, the user will make relevant queries for the predicted disease. The user's query corpus is used to train the NL2SQL model, i.e., the question is directly converted into an SQL query statement, as shown in Table 5, where Table 5 is the patient query statement table.
TABLE 5
User question field | Corpus sample |
Disease prediction | Ask me what is this? |
Introduction to diseases | If i am to be this, explain this? |
Care method | How to care? |
Diet contraindication | What is there a need to pay attention to eat? |
Preliminary examination items | How do there are checks to be made in advance? |
Infectivity of | Is the disease transmitted? |
Cause/cause of disease | How can you get ill? |
Further, the last user query link in the system uses NL2SQL to select answers to questions from the database. The embodiment of the invention adopts a model based on the Seq2Tree to convert each intention shown in the table 5 into the SQL statement, and trains the model in an end-to-end mode, such as 7 SQL statement major classes shown in the table 6. After receiving user input, the model makes a forward-sequence traversal decision from the SQL syntax tree broad class formed by the following 7 SQL statements until the model sequentially moves to the root node. Wherein, table 6 is a SQL statement category table.
TABLE 6
Further, in an embodiment of the present invention, the method further includes: matching an ending keyword from the question tracing information; and if the matching is successful, ending the conversation process.
It will be understood that the end phrase is divided into an end of query module and an end of the entire session. The language sample is shown in Table 7. Wherein, table 7 is an end keyword matching table.
TABLE 7
Ending wording | Corpus sample |
Term of end of stage | Kah-then; i know of |
End of dialog wording | Is not provided with |
To sum up, the multi-round dialogue implementation method of the ophthalmic pre-interrogation device provided by the embodiment of the invention integrates a model and template decision scheme to provide a set of dialogue design flow aiming at medical pre-interrogation, and designs a set of model data design specifications of medical interrogation from the dialogue flow to the training data of each sub-model module, so that the ophthalmic pre-interrogation device adopts an overall serial and partially parallel dialogue design flow, and the system is more efficient and better suitable for medical interrogation tasks; the data format of each part is designed aiming at the whole inquiry flow, and the complete system presents various designs required by the whole ophthalmic inquiry dialogue; the method for generating the training data is innovatively introduced, the training of the algorithm module can be completed under the limited labeled data, and the method has applicability and high efficiency.
Next, a multi-turn dialogue realization device of the ophthalmic pre-interrogation apparatus proposed according to an embodiment of the present invention will be described with reference to the accompanying drawings.
Fig. 4 is a schematic structural view of a multi-turn dialogue realization device of the ophthalmic pre-interrogation apparatus according to an embodiment of the present invention.
As shown in fig. 4, the multi-turn dialogue realization device 10 of the ophthalmic pre-interrogation apparatus includes: an acquisition module 100, a prediction module 200, and a generation module 300.
The acquisition module 100 is used for sequentially acquiring basic information, current medical history information and past history information of a patient; the prediction module 200 is used for predicting the disease information of the patient according to the basic information, the current medical history information, the past medical history information and the ophthalmology medical knowledge base; the generating module 300 is configured to collect question-following information of a patient, generate question-answering information according to the question-following information and a spoken language expression table of the user, and generate an information list. The device 10 of the embodiment of the invention adopts the whole serial and partially parallel conversation design process, thereby better meeting the scene requirement, greatly improving the process realization speed and reducing the labeling cost.
Further, in an embodiment of the present invention, the generating module is further configured to obtain the query corpus from the query information; inputting the query corpus into a trained NL2SQL model so as to directly convert the question into an SQL query statement; and determining the question answering information according to the SQL query statement.
Further, in an embodiment of the present invention, the method further includes: and the matching module is used for matching the finishing keywords from the question tracing information and finishing the conversation process when the matching is successful.
Further, in an embodiment of the present invention, the prediction module 200 is further configured to obtain all intents, and for each intention, select and fill in the data template from the pre-written slot values.
Further, in an embodiment of the present invention, the prediction module 200 is further configured to use a Rank model to perform short sentence matching between the information and the content in the ophthalmic medical knowledge base.
It should be noted that the foregoing explanation of the embodiment of the multi-round dialog implementation method for the ophthalmic pre-interrogation apparatus is also applicable to the multi-round dialog implementation apparatus for the ophthalmic pre-interrogation apparatus of this embodiment, and details are not repeated here.
According to the multi-round conversation realization device of the ophthalmic pre-inquiry device provided by the embodiment of the invention, a set of conversation design flows aiming at medical pre-inquiry are provided by combining a model and a template decision scheme, and a whole set of model data design specifications of medical inquiry are designed from the conversation flows to training data of each sub-model module, so that the ophthalmic pre-inquiry device adopts an overall serial and partially parallel conversation design flow, and the system is more efficient and better suitable for medical inquiry tasks; the data format of each part is designed aiming at the whole inquiry flow, and the complete system presents various designs required by the whole ophthalmic inquiry dialogue; the method for generating the training data is innovatively introduced, the training of the algorithm module can be completed under the limited labeled data, and the method has applicability and high efficiency.
Furthermore, the terms "first", "second" and "first" are used for descriptive purposes only and are not to be construed as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one such feature. In the description of the present invention, "a plurality" means at least two, e.g., two, three, etc., unless specifically limited otherwise.
In the present invention, unless otherwise expressly stated or limited, the first feature "on" or "under" the second feature may be directly contacting the first and second features or indirectly contacting the first and second features through an intermediate. Also, a first feature "on," "over," and "above" a second feature may be directly or diagonally above the second feature, or may simply indicate that the first feature is at a higher level than the second feature. A first feature being "under," "below," and "beneath" a second feature may be directly under or obliquely under the first feature, or may simply mean that the first feature is at a lesser elevation than the second feature.
In the description herein, references to the description of the term "one embodiment," "some embodiments," "an example," "a specific example," or "some examples," etc., mean that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the invention. In this specification, the schematic representations of the terms used above are not necessarily intended to refer to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples. Furthermore, various embodiments or examples and features of different embodiments or examples described in this specification can be combined and combined by one skilled in the art without contradiction.
Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention, and that variations, modifications, substitutions and alterations can be made to the above embodiments by those of ordinary skill in the art within the scope of the present invention.
Claims (10)
1. A multi-round conversation realization method of an ophthalmic pre-inquiry device is characterized by comprising the following steps:
sequentially collecting basic information, current medical history information and past history information of a patient;
predicting disease information of the patient according to the basic information, the current medical history information, the past medical history information and an ophthalmologic medical knowledge base; and
and acquiring the question-following information of the patient, generating question-answering information according to the question-following information and a spoken language expression table of the user, and generating an information list.
2. The method of claim 1, wherein generating answer information based on the question-following information and a spoken language expression form of the user comprises:
obtaining a question-following corpus from the question-following information;
inputting the question-chasing corpus into a trained NL2SQL model so as to directly convert the question into an SQL query statement;
and determining the question answering information according to the SQL query statement.
3. The method of claim 1, further comprising:
matching an ending keyword from the question information;
and if the matching is successful, ending the conversation process.
4. The method of claim 1, wherein predicting disease information for the patient based on the base information, the present history information, the past history information, and an ophthalmic medical knowledge base comprises:
all intents are obtained, and for each intention, a pre-written slot value is selected and filled into the data template.
5. The method of claim 1, wherein predicting disease information for the patient based on the base information, the present history information, the past history information, and an ophthalmic medical knowledge base comprises:
and matching the information with the content in the ophthalmic medical knowledge base by using a Rank model.
6. A multi-turn dialog implementation device for an ophthalmic pre-interrogation apparatus, comprising:
the acquisition module is used for sequentially acquiring basic information, current medical history information and past history information of a patient;
a prediction module for predicting disease information of the patient based on the basic information, the present history information, the past history information and an ophthalmic medical knowledge base; and
and the generating module is used for acquiring the question-following information of the patient, generating answer information according to the question-following information and the spoken language expression table of the user and generating an information list.
7. The apparatus according to claim 6, wherein the generating module is further configured to obtain a query corpus from the query information; inputting the question-chasing corpus into a trained NL2SQL model so as to directly convert the question into an SQL query statement; and determining the question answering information according to the SQL query statement.
8. The apparatus of claim 6, further comprising:
and the matching module is used for matching the finishing keywords from the question tracing information and finishing the conversation process when the matching is successful.
9. The apparatus of claim 6, wherein the prediction module is further configured to obtain all intents, and for each intention, select and fill in a data template from pre-written slot values.
10. The apparatus of claim 6, wherein the prediction module is further configured to use a Rank model to phrase match information with content in the ophthalmic medical knowledge base.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201911365885.4A CN111221947B (en) | 2019-12-26 | 2019-12-26 | Multi-round dialogue realization method and device of ophthalmologic pre-consultation device |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201911365885.4A CN111221947B (en) | 2019-12-26 | 2019-12-26 | Multi-round dialogue realization method and device of ophthalmologic pre-consultation device |
Publications (2)
Publication Number | Publication Date |
---|---|
CN111221947A true CN111221947A (en) | 2020-06-02 |
CN111221947B CN111221947B (en) | 2023-11-24 |
Family
ID=70825974
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201911365885.4A Active CN111221947B (en) | 2019-12-26 | 2019-12-26 | Multi-round dialogue realization method and device of ophthalmologic pre-consultation device |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN111221947B (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112948599A (en) * | 2021-05-13 | 2021-06-11 | 北京富通东方科技有限公司 | Disease diagnosis method based on path reasoning of medical knowledge graph |
CN114300160A (en) * | 2021-11-16 | 2022-04-08 | 北京左医科技有限公司 | Inquiry dialogue method and system |
CN117093679A (en) * | 2023-06-19 | 2023-11-21 | 无码科技(杭州)有限公司 | Large language model intelligent inquiry dialogue method, system, equipment and medium |
Families Citing this family (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
KR20230079729A (en) * | 2021-11-29 | 2023-06-07 | 주식회사 포티투마루 | Method for converting natural language query to sql and device thereof |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20060136142A1 (en) * | 2000-11-02 | 2006-06-22 | Kurt Berlin | Systems, methods and computer program products for guiding selection of a therapeutic treatment regimen based on the methylation status of the DNA |
JP2008293294A (en) * | 2007-05-24 | 2008-12-04 | Fuji Xerox Co Ltd | Medical information processing system, information processing method, and computer program |
CN105912846A (en) * | 2016-04-07 | 2016-08-31 | 南京小网科技有限责任公司 | Intelligent medical aid decision making system on basis of cloud computing technique and medical knowledge base technique |
CN107247868A (en) * | 2017-05-18 | 2017-10-13 | 深思考人工智能机器人科技(北京)有限公司 | A kind of artificial intelligence aids in interrogation system |
CN109065172A (en) * | 2018-07-04 | 2018-12-21 | 平安科技(深圳)有限公司 | Illness information acquisition method, device, computer equipment and storage medium |
CN109408526A (en) * | 2018-10-12 | 2019-03-01 | 平安科技(深圳)有限公司 | SQL statement generation method, device, computer equipment and storage medium |
CN109522393A (en) * | 2018-10-11 | 2019-03-26 | 平安科技(深圳)有限公司 | Intelligent answer method, apparatus, computer equipment and storage medium |
-
2019
- 2019-12-26 CN CN201911365885.4A patent/CN111221947B/en active Active
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20060136142A1 (en) * | 2000-11-02 | 2006-06-22 | Kurt Berlin | Systems, methods and computer program products for guiding selection of a therapeutic treatment regimen based on the methylation status of the DNA |
JP2008293294A (en) * | 2007-05-24 | 2008-12-04 | Fuji Xerox Co Ltd | Medical information processing system, information processing method, and computer program |
CN105912846A (en) * | 2016-04-07 | 2016-08-31 | 南京小网科技有限责任公司 | Intelligent medical aid decision making system on basis of cloud computing technique and medical knowledge base technique |
CN107247868A (en) * | 2017-05-18 | 2017-10-13 | 深思考人工智能机器人科技(北京)有限公司 | A kind of artificial intelligence aids in interrogation system |
CN109065172A (en) * | 2018-07-04 | 2018-12-21 | 平安科技(深圳)有限公司 | Illness information acquisition method, device, computer equipment and storage medium |
CN109522393A (en) * | 2018-10-11 | 2019-03-26 | 平安科技(深圳)有限公司 | Intelligent answer method, apparatus, computer equipment and storage medium |
CN109408526A (en) * | 2018-10-12 | 2019-03-01 | 平安科技(深圳)有限公司 | SQL statement generation method, device, computer equipment and storage medium |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112948599A (en) * | 2021-05-13 | 2021-06-11 | 北京富通东方科技有限公司 | Disease diagnosis method based on path reasoning of medical knowledge graph |
CN114300160A (en) * | 2021-11-16 | 2022-04-08 | 北京左医科技有限公司 | Inquiry dialogue method and system |
CN117093679A (en) * | 2023-06-19 | 2023-11-21 | 无码科技(杭州)有限公司 | Large language model intelligent inquiry dialogue method, system, equipment and medium |
CN117093679B (en) * | 2023-06-19 | 2024-04-02 | 无码科技(杭州)有限公司 | Large language model intelligent inquiry dialogue method, system, equipment and medium |
Also Published As
Publication number | Publication date |
---|---|
CN111221947B (en) | 2023-11-24 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN111221947B (en) | Multi-round dialogue realization method and device of ophthalmologic pre-consultation device | |
CN107731269A (en) | Disease code method and system based on raw diagnostic data and patient file data | |
CN110209791A (en) | It is a kind of to take turns dialogue intelligent speech interactive system and device more | |
CN110489566A (en) | A kind of hospital guide's method of intelligence hospital guide's service robot | |
CN109446306A (en) | A kind of intelligent answer method of more wheels dialogue of task based access control driving | |
CN106357942A (en) | Intelligent response method and system based on context dialogue semantic recognition | |
CN110472017A (en) | A kind of analysis of words art and topic point identify matched method and system | |
CN105468468B (en) | Data error-correcting method towards question answering system and device | |
JP2021096873A (en) | Communication system, communication control method, and program | |
CN109344240A (en) | A kind of data processing method, server and electronic equipment | |
CN109378077B (en) | Method for acquiring pre-diagnosis medical history and machine-readable storage medium for executing method | |
US20230394247A1 (en) | Human-machine collaborative conversation interaction system and method | |
CN110347811A (en) | A kind of professional knowledge question and answer robot system based on artificial intelligence | |
CN113871020B (en) | Health management method and system after critical illness diagnosis based on AI machine learning | |
CN112164477A (en) | Intelligent question-answering system for hypertension patient based on knowledge graph and establishing method thereof | |
CN108647229B (en) | Virtual person model construction method based on artificial intelligence | |
CN110008308A (en) | For the method and apparatus of user's question sentence supplemental information | |
CN111400453A (en) | Intelligent interaction system and method based on legal consultation | |
CN114299068A (en) | Intelligent decision-making system for evaluating psoriasis severity degree based on skin image | |
CN113506625A (en) | Diagnosis and treatment suggestion scoring system based on csco guide | |
CN109410074A (en) | Intelligent core protects method and system | |
CN117370525A (en) | Intelligent diagnosis guiding method based on fine tuning large model | |
CN107798137B (en) | A kind of multi-source heterogeneous data fusion architecture system based on additive models | |
US20230098296A1 (en) | Method and system for generating data set relating to facial expressions, and non-transitory computer-readable recording medium | |
CN114155962A (en) | Data cleaning method and method for constructing disease diagnosis by using knowledge graph |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |