WO2022041728A1 - 医学领域意图识别方法、装置、设备及存储介质 - Google Patents

医学领域意图识别方法、装置、设备及存储介质 Download PDF

Info

Publication number
WO2022041728A1
WO2022041728A1 PCT/CN2021/084653 CN2021084653W WO2022041728A1 WO 2022041728 A1 WO2022041728 A1 WO 2022041728A1 CN 2021084653 W CN2021084653 W CN 2021084653W WO 2022041728 A1 WO2022041728 A1 WO 2022041728A1
Authority
WO
WIPO (PCT)
Prior art keywords
target
preset
medical
sentence
text data
Prior art date
Application number
PCT/CN2021/084653
Other languages
English (en)
French (fr)
Inventor
贾声声
Original Assignee
康键信息技术(深圳)有限公司
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by 康键信息技术(深圳)有限公司 filed Critical 康键信息技术(深圳)有限公司
Publication of WO2022041728A1 publication Critical patent/WO2022041728A1/zh

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/30Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
    • G06F16/33Querying
    • G06F16/332Query formulation
    • G06F16/3329Natural language query formulation or dialogue systems
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/30Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
    • G06F16/35Clustering; Classification
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/30Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
    • G06F16/36Creation of semantic tools, e.g. ontology or thesauri
    • G06F16/367Ontology
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F40/00Handling natural language data
    • G06F40/20Natural language analysis
    • G06F40/279Recognition of textual entities
    • G06F40/289Phrasal analysis, e.g. finite state techniques or chunking
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F40/00Handling natural language data
    • G06F40/30Semantic analysis
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/04Architecture, e.g. interconnection topology
    • G06N3/045Combinations of networks
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/04Architecture, e.g. interconnection topology
    • G06N3/049Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/08Learning methods

Definitions

  • the present application relates to the field of medical data, and in particular, to a method, apparatus, device, and storage medium for identifying intent in the medical field.
  • Intent recognition is an important part of intelligent question answering, that is, the user's questions are classified into the corresponding medical intent categories through the classification method.
  • intent is the user's intention, that is, what the user wants to do, which is classified into previously defined intent categories according to the domain and intent of the user's expression.
  • the accuracy of intent recognition is related to the performance of semantic slot filling and is helpful for subsequent research on intelligent question answering.
  • the existing solutions mainly extract features through a segmentation algorithm. It is necessary to extract key features from the expected text to extract intent features such as characters and words. Commonly used algorithms include naive Bayes, support vector machines, and logistic regression.
  • the inventor realizes that in the medical question and answer, the expressions of users in the medical field are not standardized, the descriptions are short, and have certain ambiguity, and it is difficult to obtain richer intent features, resulting in inaccurate identification of the user's medical intent.
  • the present application provides a method, apparatus, device, and storage medium for identifying intent in the medical field, which are used to accurately identify the intent in the medical field consulted by a user, and improve the accuracy of an intent identification result in the medical field.
  • a first aspect of the present application provides a method for recognizing intent in the medical field, including: acquiring initial text data, where the initial text data includes question sentences and/or question words input by a target user; The model rewrites the initial text data to generate target question sentences, which are sentences with clear semantics and no redundancy; feature extraction is performed on the target question sentences based on a preset medical-specific feature thesaurus, Obtain multiple target features, multiple target feature weights and multiple target structured information; call the preset deep learning model, the multiple target feature weights and the multiple target structured information to evaluate the multiple target features The classification is performed to obtain a classification result, and the medical field intention of the target user is determined according to the classification result.
  • a second aspect of the present application provides an intent recognition device in the medical field, comprising a memory, a processor, and computer-readable instructions stored on the memory and executable on the processor, the processor executing the computer
  • the instruction is readable, the following steps are implemented: obtaining initial text data, the initial text data includes question sentences and/or question words input by the target user; calling a preset joint model to rewrite the initial text data to generate the target question sentence , the target question sentence is a sentence with clear semantics and no redundancy; feature extraction is performed on the target question sentence based on a preset medical-specific feature thesaurus, and multiple target features, multiple target feature weights, and multiple target feature weights are obtained.
  • Target structured information call the preset deep learning model, the multiple target feature weights, and the multiple target structured information to classify the multiple target features, obtain a classification result, and determine according to the classification result The medical domain intention of the target user.
  • a third aspect of the present application provides a computer-readable storage medium, where computer instructions are stored in the computer-readable storage medium, and when the computer instructions are executed on a computer, the computer is caused to perform the following steps: acquiring initial text data, The initial text data includes question sentences and/or question words input by the target user; a preset joint model is invoked to rewrite the initial text data to generate target question sentences, the target question sentences being semantically clear and without redundancy sentence; perform feature extraction on the target question sentence based on the preset medical-specific feature thesaurus to obtain multiple target features, multiple target feature weights and multiple target structured information; call the preset deep learning model, The plurality of target feature weights and the plurality of target structured information are used to classify the plurality of target features to obtain a classification result, and the medical domain intention of the target user is determined according to the classification result.
  • a fourth aspect of the present application provides an intention recognition device in the medical field, comprising: a data acquisition module for acquiring initial text data, where the initial text data includes question sentences and/or question words input by a target user; a data rewriting module, It is used to call the preset joint model to rewrite the initial text data, and generate the target question sentence, the target question sentence is a sentence with clear semantics and no redundancy; the feature extraction module is used for the preset medical-specific The feature thesaurus performs feature extraction on the target question sentence to obtain multiple target features, multiple target feature weights, and multiple target structured information; an intent determination module is used to call a preset deep learning model, the multiple The target feature weight and the plurality of target structured information are used to classify the plurality of target features to obtain a classification result, and the medical field intention of the target user is determined according to the classification result.
  • initial text data is obtained, and the initial text data includes question sentences and/or question words input by the target user; a preset joint model is invoked to rewrite the initial text data to generate the target question sentence and the target question sentence.
  • the question sentence is a sentence with clear semantics and no redundancy; feature extraction is performed on the target question sentence based on a preset medical-specific feature thesaurus to obtain multiple target features, multiple target feature weights, and multiple target structured information; call The preset deep learning model, multiple target feature weights and multiple target structured information classify multiple target features, obtain a classification result, and determine the medical field intention of the target user according to the classification result.
  • the embodiments of the present application can accurately identify the user's intention in the medical field, and improve the accuracy of the identification of the intention in the medical field.
  • FIG. 1 is a schematic diagram of an embodiment of a method for recognizing intent in the medical field in an embodiment of the present application
  • FIG. 2 is a schematic diagram of another embodiment of the method for recognizing intent in the medical field in the embodiment of the present application;
  • FIG. 3 is a schematic diagram of an embodiment of an intention identification device in the medical field in an embodiment of the present application
  • FIG. 4 is a schematic diagram of another embodiment of the device for recognizing intention in the medical field in the embodiment of the present application.
  • FIG. 5 is a schematic diagram of an embodiment of an intention recognizing device in the medical field in an embodiment of the present application.
  • the present application provides a method, apparatus, device and storage medium for identifying intent in the medical field, which are used to accurately identify the intent in the medical field of a user and improve the accuracy of intent identification in the medical field.
  • FIG. 1 a flowchart of a method for recognizing intent in the medical field provided by an embodiment of the present application, which specifically includes:
  • the server obtains initial text data, where the initial text data includes question sentences and/or question words input by the target user.
  • the question sentence can include various forms, and the question sentence can be a sentence with clear semantics and concise expression, such as: "How to treat gastroenteritis?" etc.; the question sentence can also be a sentence with ambiguous semantics, such as: “please take a look at me”, “what's wrong with me”, “feel unwell”, “always feel dizzy”, etc.; the question sentence can also be an expression Redundant sentences, for example, “My stomach is a little uncomfortable and I am not feeling well", “I have a little discomfort in my throat and difficulty swallowing, maybe I am getting angry", "I had a liver ultrasound yesterday, and the result of the B-ultrasound was echo-dense, Enhance, thicken”, the problem word may include only one keyword, for example, the problem word may be "diabetes", “migraine”, “gastritis”, “vaginitis” and so on.
  • the execution subject of the present application may be an intention identification device in the medical field, or may be a server, which is not specifically limited here.
  • the embodiments of the present application take the server as an execution subject as an example for description.
  • the server invokes the preset joint model to rewrite the initial text data, and generates the target question sentence.
  • the target question sentence is a sentence with clear semantics and no redundancy.
  • the server invokes the preset joint model to perform entity extraction and normalization processing on the initial text data to obtain multiple standard entities; the server searches the preset medical domain knowledge graph for the attribute information corresponding to each standard entity, Obtain multiple standard attribute information, each standard attribute information corresponds to a different standard entity; the server performs information completion on the initial text data according to the multiple standard attribute information and multiple standard entities, generates multiple candidate question statements and sends multiple
  • the candidate question sentence is sent to the terminal; the server determines the target question sentence according to the instruction fed back by the terminal, and the target question sentence is a sentence with clear semantics and no redundancy.
  • the server performs feature extraction on the target question sentence based on a preset medical-specific feature thesaurus, and obtains multiple target features, multiple target feature weights, and multiple target structured information. Specifically, the server invokes the preset joint model to perform entity feature extraction on the target question statement to obtain multiple candidate features; the server invokes the preset normalized knowledge base to normalize the multiple candidate features to obtain multiple standardized features ; The server queries the context corresponding to multiple standardized features and multiple target feature weights from the preset medical domain knowledge graph, and each standardized feature corresponds to a target feature weight; the server performs structured processing on the corresponding context to obtain multiple targets Structured information; the server screens multiple standardized features based on a preset medical-specific feature thesaurus to obtain multiple target features.
  • the target structured information may be sentence word vector features, position vectors, manually extracted features or entity weighted features, and may also include other structured information, which is not specifically limited here.
  • medical-specific feature thesaurus such as drug library, disease library, treatment library, etc.
  • Medical-specific feature thesaurus can be used to improve the accuracy of word segmentation and retrieve relevant information of medical entities.
  • the server invokes a preset deep learning model, multiple target feature weights, and multiple target structured information to classify multiple target features, obtains a classification result, and determines the medical domain intention of the target user according to the classification result.
  • the deep learning model uses the improved textcnn model.
  • the input layer includes sentence and word vector features, position vectors, manually extracted features, and entity weighted features.
  • Feature fusion is performed to obtain input layer features (fusion features), and the input layer features are transmitted.
  • Enter the convolution layer learn the sentence feature vector through the convolution layer, and use the sentence feature vector to complete the intent classification through the fully connected layer.
  • the size of the sliding step size of the feature map of the convolution layer is set according to the training data entity and medical text.
  • the average length of the medical information database constructed by the feature is set.
  • the preset deep learning model is a model that has been trained in advance.
  • data enhancement is performed for the limited labeled data, and the existing learning model is used to pre-label the unsupervised data. , select the data with higher confidence to add it to the labeled data, and obtain the enhanced labeled data, that is, the training data, so as to realize the enhancement of the training data.
  • deep learning is combined with the unique features of medical texts to rewrite the text input by the user, and combined with the knowledge graph in the medical field, the rewritten text is recognized for medical intent, which can accurately identify the information consulted by the user.
  • the medical domain intent improves the accuracy of the medical domain intent recognition results. And this solution can be applied in the field of smart medical care, thereby promoting the construction of smart cities.
  • FIG. 2 another flowchart of the method for recognizing intent in the medical field provided by the embodiment of the present application, which specifically includes:
  • initial text data where the initial text data includes question sentences and/or question words input by a target user.
  • the server obtains initial text data, where the initial text data includes question sentences and/or question words input by the target user.
  • the question sentence can include various forms, and the question sentence can be a sentence with clear semantics and concise expression, such as: "How to treat gastroenteritis?" etc.; the question sentence can also be a sentence with ambiguous semantics, such as: “please take a look at me”, “what's wrong with me”, “feel unwell”, “always feel dizzy”, etc.; the question sentence can also be an expression Redundant sentences, for example, “My stomach is a little uncomfortable and I am not feeling well", “I have a little discomfort in my throat and difficulty swallowing, maybe I am getting angry", "I had a liver ultrasound yesterday, and the result of the B-ultrasound was echo-dense, Enhance, thicken”, the problem word may include only one keyword, for example, the problem word may be "diabetes", “migraine”, “gastritis”, “vaginitis” and so on.
  • the execution subject of the present application may be an intention identification device in the medical field, or may be a server, which is not specifically limited here.
  • the embodiments of the present application take the server as an execution subject as an example for description.
  • step 201 it may further include:
  • the server builds a preset medical knowledge graph.
  • the server builds a preset medical knowledge graph specifically including:
  • the server obtains the medical training data and integrates the medical training data; the server performs entity extraction and relation extraction on the medical training data to generate knowledge triples; the server labels the knowledge triples to obtain the labeled data;
  • the labeling data trains and performs knowledge inference on the preset sequence labeling model, and generates a preset medical knowledge map.
  • the server performs entity extraction and relationship extraction on the medical training data to generate a knowledge triple, including: the server performs entity extraction on the medical training data to obtain multiple entities; the server performs entity extraction on the relationships between multiple entities Extract and obtain multiple entity relationships; the server generates knowledge triples according to multiple entity relationships and multiple entities.
  • Entity recognition refers to identifying entities with specific meanings in medical texts, including entities such as subjects, locations, inspection items, and diseases. Entity relationship extraction needs to infer the relationship between entities based on text semantic information; for example, a given sentence: "Pregnant women can eat Crab?", the entity of the sentence is "pregnant woman” pregnant woman and "crab", the preset joint extraction model can obtain the "inappropriate” relationship through semantics, and finally extract the knowledge ternary of (pregnant woman, inappropriate, crab) Group.
  • Knowledge triples are relational entity triples.
  • the preset joint model is used to extract entities and entity relationships, and the labeled data of "BIOES" mode will be used, where B means start, I means internal, O means non-entity, E entity tail, S means word change is an entity in itself.
  • the serialized labeling model is a BILSTM+CRF deep learning model.
  • the labeled data (labeled triples) are used to train and optimize the BILSTM+CRF deep learning model to obtain a medical knowledge map.
  • the server invokes the preset joint model to perform entity extraction on the initial text data to obtain multiple initial entities; the server invokes the preset normalized knowledge base to normalize the multiple initial entities to obtain multiple standard entities.
  • pregnancy test strips and “pregnancy test strips” are both standardized as “pregnancy test strips”.
  • motor neuron disease and “Charcot disease” are both standardized as “Amyotrophic Lateral Sclerosis”
  • AIDS and “AIDS” are both standardized as "Acquired Immune Deficiency Syndrome”.
  • entity extraction and normalization are performed on the user's initial text data, and the normalization of user text is helpful for subsequent intent recognition and avoids semantic inaccuracy and redundancy.
  • Different standard entities correspond to different standard attribute information.
  • the attribute information corresponding to "gastritis” is “treatment method” and “precautions”
  • the attribute information corresponding to "amoxicillin” is “indication” and “contraindication” .
  • the server matches the initial text data with a preset sentence pattern template to obtain a target sentence pattern template; the server adjusts the sentence pattern of the initial text data according to the target sentence pattern template to obtain an initial statement; The initial entities are replaced with corresponding multiple standard entities to generate intermediate sentences; the server completes the intermediate sentences according to multiple standard attribute information to generate multiple candidate question sentences; the server sends multiple candidate question sentences to the terminal.
  • the initial text data is "What is the difference between B ultrasound and color ultrasound?”
  • the sentence "What is the difference between a and b" is matched through the preset sentence template.
  • the template corresponding to the sentence is "((.*)( [ ⁇ ])(.*))—— ⁇ 0 ⁇ What is the difference between ⁇ 2 ⁇ ”
  • vaginitis For example, for the "vaginitis” input by the user, the corresponding standard entity is also “vaginitis”, and the attribute information of "vaginitis” is found through the knowledge map as “treatment method” and “precautions”, then the correct information after completing the information
  • treatment method For example, for the "vaginitis” input by the user, the corresponding standard entity is also “vaginitis”, and the attribute information of "vaginitis” is found through the knowledge map as “treatment method” and “precautions”, then the correct information after completing the information
  • the question asked by users is "how to treat vaginitis? What needs to be paid attention to in vaginitis?"
  • the attribute information is an attribute list constructed for the entity
  • the attributes of the drug include indications and contraindications
  • the attributes of the disease may include treatment drugs, precautions, etc.
  • amoxicillin is prohibited for those allergic to penicillin.
  • the target user selects “how to treat gastritis?”, then the question sentence "how to treat gastritis?” is determined as the target question sentence. Understandably, letting the target user make a choice is more conducive to understanding the user's needs.
  • the server performs feature extraction on the target question sentence based on a preset medical-specific feature thesaurus, and obtains multiple target features, multiple target feature weights, and multiple target structured information. Specifically, the server invokes the preset joint model to perform entity feature extraction on the target question statement to obtain multiple candidate features; the server invokes the preset normalized knowledge base to normalize the multiple candidate features to obtain multiple standardized features ; The server queries the context corresponding to multiple standardized features and multiple target feature weights from the preset medical domain knowledge graph, and each standardized feature corresponds to a target feature weight; the server performs structured processing on the corresponding context to obtain multiple targets Structured information; the server screens multiple standardized features based on a preset medical-specific feature thesaurus to obtain multiple target features.
  • the target structured information may be sentence word vector features, position vectors, manually extracted features or entity weighted features, and may also include other structured information, which is not specifically limited here.
  • medical-specific feature thesaurus such as drug library, disease library, treatment library, etc.
  • Medical-specific feature thesaurus can be used to improve the accuracy of word segmentation and retrieve relevant information of medical entities.
  • the server calls the preset deep learning model, multiple target feature weights and multiple target structured information to classify multiple target features, obtains the classification result, and determines the medical domain intention of the target user according to the classification result.
  • the deep learning model uses the improved textcnn model.
  • the input layer includes sentence and word vector features, position vectors, manually extracted features, and entity weighted features.
  • Feature fusion is performed to obtain input layer features (fusion features), and the input layer features are transmitted.
  • Enter the convolution layer learn the sentence feature vector through the convolution layer, and use the sentence feature vector to complete the intent classification through the fully connected layer.
  • the size of the sliding step size of the feature map of the convolution layer is set according to the training data entity and medical text.
  • the average length of the medical information database constructed by the feature is set.
  • the preset deep learning model is a model that has been trained in advance.
  • data enhancement is performed for the limited labeled data, and the existing learning model is used to pre-label the unsupervised data. , select the data with higher confidence to add it to the labeled data, and obtain the enhanced labeled data, that is, the training data, so as to realize the enhancement of the training data.
  • the server searches for answers to questions based on the medical domain intent of the target user.
  • the server when the medical field of the target user intends to inquire about the medicine for treating gastric ulcer, the server will determine the relevant information of the gastric medicine in the preset medical knowledge graph; Information about gastric medicines and information about gastric medicines for curing ulcers; the server generates answers to questions according to the information about gastric medicines for curing ulcers in a preset format, and feeds them back to the target terminal of the target user.
  • the medical domain intention in the user input question can be accurately identified, the answer required by the user input question can be retrieved more accurately, the answer retrieval time is reduced, and the answer retrieval efficiency is improved.
  • deep learning is combined with the unique features of medical texts to rewrite the text input by the user, and combined with the knowledge graph in the medical field, the rewritten text is recognized for medical intent, which can accurately identify the information consulted by the user.
  • the medical domain intent improves the accuracy of the medical domain intent recognition results. And this solution can be applied in the field of smart medical care, thereby promoting the construction of smart cities.
  • an embodiment of the device for recognizing intent in the medical field in the embodiment of the present application includes:
  • a data acquisition module 301 configured to acquire initial text data, where the initial text data includes question sentences and/or question words input by the target user;
  • the data rewriting module 302 is used for invoking a preset joint model to rewrite the initial text data to generate a target question sentence, and the target question sentence is a sentence with clear semantics and no redundancy;
  • a feature extraction module 303 configured to perform feature extraction on the target question sentence based on a preset medical-specific feature thesaurus to obtain multiple target features, multiple target feature weights, and multiple target structured information;
  • the intent determination module 304 is configured to call the preset deep learning model, the multiple target feature weights and the multiple target structured information to classify the multiple target features, obtain a classification result, and classify the multiple target features according to the classification The results determine the medical domain intent of the target user.
  • deep learning is combined with the unique features of medical texts to rewrite the text input by the user, and combined with the knowledge graph in the medical field, the rewritten text is recognized for medical intent, which can accurately identify the information consulted by the user.
  • the medical domain intent improves the accuracy of the medical domain intent recognition results. And this solution can be applied in the field of smart medical care, thereby promoting the construction of smart cities.
  • another embodiment of the device for recognizing intention in the medical field in the embodiment of the present application includes:
  • a data acquisition module 301 configured to acquire initial text data, where the initial text data includes question sentences and/or question words input by the target user;
  • the data rewriting module 302 is used for invoking a preset joint model to rewrite the initial text data to generate a target question sentence, and the target question sentence is a sentence with clear semantics and no redundancy;
  • Feature extraction module 303 for carrying out feature extraction to described target problem statement based on the preset medical-specific feature thesaurus, obtains multiple target features, multiple target feature weights and multiple target structured information;
  • the intent determination module 304 is configured to call the preset deep learning model, the multiple target feature weights and the multiple target structured information to classify the multiple target features, obtain a classification result, and classify the multiple target features according to the classification The results determine the medical domain intent of the target user.
  • the data rewriting module 302 includes:
  • the extraction unit 3021 is used to call the preset joint model to perform entity extraction and normalization processing on the initial text data to obtain a plurality of standard entities;
  • the searching unit 3022 is used to search the attribute information corresponding to each standard entity in the preset medical domain knowledge graph, and obtain a plurality of standard attribute information, and each standard attribute information corresponds to a different standard entity;
  • a completion unit 3023 configured to perform information completion on the initial text data according to the multiple standard attribute information and the multiple standard entities, generate multiple candidate question sentences, and send the multiple candidate question sentences to the terminal;
  • the determining unit 3024 is configured to determine a target question sentence according to an instruction fed back by the terminal, where the target question sentence is a sentence with clear semantics and no redundancy.
  • the extraction unit 3021 is specifically used for:
  • completion unit 3023 is specifically used for:
  • the multiple initial entities are replaced with corresponding multiple standard entities to generate intermediate statements; the intermediate statements are completed according to the multiple standard attribute information to generate multiple candidate question statements; the multiple candidate question statements sent to the terminal.
  • the feature extraction module 303 is specifically used for:
  • the device for recognizing intent in the medical field further includes:
  • the graph construction module 305 is used to construct a preset medical knowledge graph.
  • map building module 305 is specifically used for:
  • Obtain medical training data and integrate the medical training data; perform entity extraction and relationship extraction on the medical training data to generate a knowledge triplet; perform data labeling on the knowledge triplet to obtain labeled data;
  • the pre-labeled data is used for training and knowledge inference on the pre-set sequence labeling model, and a pre-set medical knowledge map is generated.
  • deep learning is combined with the unique features of medical texts to rewrite the text input by the user, and combined with the knowledge graph in the medical field, the rewritten text is recognized for medical intent, which can accurately identify the information consulted by the user.
  • the medical domain intent improves the accuracy of the medical domain intent recognition results. And this solution can be applied in the field of smart medical care, thereby promoting the construction of smart cities.
  • FIG. 5 is a schematic structural diagram of a medical field intention recognition device provided by an embodiment of the present application.
  • the medical field intention recognition device 500 may vary greatly due to different configurations or performances, and may include one or more processors (central processing units, CPU) 510 (eg, one or more processors) and memory 520, one or more storage media 530 (eg, one or more mass storage devices) storing application programs 533 or data 532.
  • the memory 520 and the storage medium 530 may be short-term storage or persistent storage.
  • the program stored in the storage medium 530 may include one or more modules (not shown in the figure), and each module may include a series of instruction operations in the medical field intention recognition device 500 .
  • the processor 510 may be configured to communicate with the storage medium 530 to execute a series of instruction operations in the storage medium 530 on the medical field intention recognition device 500 .
  • the medical domain intent recognition device 500 may also include one or more power supplies 540, one or more wired or wireless network interfaces 550, one or more input and output interfaces 560, and/or, one or more operating systems 531, such as Windows Serve, Mac OS X, Unix, Linux, FreeBSD, and more.
  • operating systems 531 such as Windows Serve, Mac OS X, Unix, Linux, FreeBSD, and more.
  • the processor 510 may perform the functions of the data acquisition module 301 , the data rewriting module 302 , the feature extraction module 303 , the intent determination module 304 and the graph construction module 305 in the above-mentioned embodiments.
  • the present application also provides a device for recognizing intent in the medical field, comprising: a memory and at least one processor, wherein instructions are stored in the memory, the memory and the at least one processor are interconnected by a line; the at least one processor The instructions in the memory are invoked, so that the medical domain intention recognition device executes the steps in the medical domain intention recognition method described above.
  • the present application also provides a computer-readable storage medium, and the computer-readable storage medium may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium.
  • the computer-readable storage medium stores computer instructions, and when the computer instructions are executed on the computer, the computer performs the following steps:
  • the initial text data includes question sentences and/or question words input by the target user
  • Medical field intent Invoke the preset deep learning model, the multiple target feature weights, and the multiple target structured information to classify the multiple target features, obtain a classification result, and determine the target user's profile according to the classification result. Medical field intent.
  • the integrated unit if implemented in the form of a software functional unit and sold or used as an independent product, may be stored in a computer-readable storage medium.
  • the technical solutions of the present application can be embodied in the form of software products in essence, or the parts that contribute to the prior art, or all or part of the technical solutions, and the computer software products are stored in a storage medium , including several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present application.
  • the aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), magnetic disk or optical disk and other media that can store program codes .

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Artificial Intelligence (AREA)
  • Mathematical Physics (AREA)
  • General Health & Medical Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Computing Systems (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Evolutionary Computation (AREA)
  • Software Systems (AREA)
  • Molecular Biology (AREA)
  • Databases & Information Systems (AREA)
  • Audiology, Speech & Language Pathology (AREA)
  • Animal Behavior & Ethology (AREA)
  • Human Computer Interaction (AREA)
  • Medical Treatment And Welfare Office Work (AREA)

Abstract

一种医学领域意图识别方法、装置、设备及存储介质,应用于智慧医疗领域中,用于精准识别用户的医学领域意图,提高了医学领域意图识别的准确度。该方法包括:获取初始文本数据,初始文本数据包括目标用户输入的问题句子和/或问题词语(101);调用预置的联合模型对初始文本数据进行改写,生成目标问题语句,目标问题语句为语义明确且无冗余的句子(102);基于预置的医学专有特征词库对目标问题语句进行特征提取,得到多个目标特征、多个目标特征权重和多个目标结构化信息(103);调用预置的深度学习模型、多个目标特征权重和多个目标结构化信息对多个目标特征进行分类,得到分类结果,并根据分类结果确定目标用户的医学领域意图(104)。

Description

医学领域意图识别方法、装置、设备及存储介质
本申请要求于2020年8月28日提交中国专利局、申请号为202010885018.X、发明名称为“医学领域意图识别方法、装置、设备及存储介质”的中国专利申请的优先权,其全部内容通过引用结合在申请中。
技术领域
本申请涉及医疗数据领域,尤其涉及一种医学领域意图识别方法、装置、设备及存储介质。
背景技术
意图识别是智能问答的重要组成部分,即通过分类的方法将用户的问题分到相应的医学意图种类。简单来说意图就是用户的意愿,即用户想要做什么,根据用户表达所涉及的领域和意图将其分类到先前定义好的意图类别中。意图识别的精准性关系到语义槽填充的性能并且有助于后续智能问答的研究。
现有方案主要是通过分法算法进行特征提取,需要对预料文本进行关键特征抽取,提取得到字符、词语等意图特征,常用的算法有朴素贝叶斯、支持向量机和逻辑回归等。
发明人意识到,在医疗问答中医学领域用户的表达不规范、描述较短,且具有一定的歧义性,很难获取较丰富的意图特征,导致对用户的医疗意图识别结果不准确。
发明内容
本申请提供了一种医学领域意图识别方法、装置、设备及存储介质,用于精准识别用户咨询的医学领域意图,提高了医学领域意图识别结果的准确度。
为实现上述目的,本申请第一方面提供了一种医学领域意图识别方法,包括:获取初始文本数据,所述初始文本数据包括目标用户输入的问题句子和/或问题词语;调用预置的联合模型对所述初始文本数据进行改写,生成目标问题语句,所述目标问题语句为语义明确且无冗余的句子;基于预置的医学专有特征词库对所述目标问题语句进行特征提取,得到多个目标特征、多个目标特征权重和多个目标结构化信息;调用预置的深度学习模型、所述多个目标特征权重和所述多个目标结构化信息对所述多个目标特征进行分类,得到分类结果,并根据所述分类结果确定所述目标用户的医学领域意图。
本申请第二方面提供了一种医学领域意图识别设备,包括存储器、处理器及存储在所述存储器上并可在所述处理器上运行的计算机可读指令,所述处理器执行所述计算机可读指令时实现如下步骤:获取初始文本数据,所述初始文本数据包括目标用户输入的问题句子和/或问题词语;调用预置的联合模型对所述初始文本数据进行改写,生成目标问题语句,所述目标问题语句为语义明确且无冗余的句子;基于预置的医学专有特征词库对所述目标问题语句进行特征提取,得到多个目标特征、多个目标特征权重和多个目标结构化信息;调用预置的深度学习模型、所述多个目标特征权重和所述多个目标结构化信息对所述多个目标特征进行分类,得到分类结果,并根据所述分类结果确定所述目标用户的医学领域意图。
本申请第三方面提供了一种计算机可读存储介质,所述计算机可读存储介质中存储有计算机指令,当所述计算机指令在计算机上运行时,使得计算机执行如下步骤:获取初始文本数据,所述初始文本数据包括目标用户输入的问题句子和/或问题词语;调用预置的联合模型对所述初始文本数据进行改写,生成目标问题语句,所述目标问题语句为语义明确且无冗余的句子;基于预置的医学专有特征词库对所述目标问题语句进行特征提取,得到多个目标特征、多个目标特征权重和多个目标结构化信息;调用预置的深度学习模型、所述多个目标特征权重和所述多个目标结构化信息对所述多个目标特征进行分类,得到分类结果,并根据所述分类结果确定所述目标用户的医学领域意图。
本申请第四方面提供了一种医学领域意图识别装置,包括:数据获取模块,用于获取 初始文本数据,所述初始文本数据包括目标用户输入的问题句子和/或问题词语;数据改写模块,用于调用预置的联合模型对所述初始文本数据进行改写,生成目标问题语句,所述目标问题语句为语义明确且无冗余的句子;特征提取模块,用于基于预置的医学专有特征词库对所述目标问题语句进行特征提取,得到多个目标特征、多个目标特征权重和多个目标结构化信息;意图确定模块,用于调用预置的深度学习模型、所述多个目标特征权重和所述多个目标结构化信息对所述多个目标特征进行分类,得到分类结果,并根据所述分类结果确定所述目标用户的医学领域意图。
本申请实施例提供的技术方案中,获取初始文本数据,初始文本数据包括目标用户输入的问题句子和/或问题词语;调用预置的联合模型对初始文本数据进行改写,生成目标问题语句,目标问题语句为语义明确且无冗余的句子;基于预置的医学专有特征词库对目标问题语句进行特征提取,得到多个目标特征、多个目标特征权重和多个目标结构化信息;调用预置的深度学习模型、多个目标特征权重和多个目标结构化信息对多个目标特征进行分类,得到分类结果,并根据分类结果确定目标用户的医学领域意图。本申请实施例,能精准识别用户的医学领域意图,提高了医学领域意图识别的准确度。
附图说明
图1为本申请实施例中医学领域意图识别方法的一个实施例示意图;
图2为本申请实施例中医学领域意图识别方法的另一个实施例示意图;
图3为本申请实施例中医学领域意图识别装置的一个实施例示意图;
图4为本申请实施例中医学领域意图识别装置的另一个实施例示意图;
图5为本申请实施例中医学领域意图识别设备的一个实施例示意图。
具体实施方式
本申请提供了一种医学领域意图识别方法、装置、设备及存储介质,用于精准识别用户的医学领域意图,提高了医学领域意图识别的准确度。
为了使本技术领域的人员更好地理解本申请方案,下面将结合本申请实施例中的附图,对本申请实施例进行描述。
本申请的说明书和权利要求书及上述附图中的术语“第一”、“第二”、“第三”、“第四”等(如果存在)是用于区别类似的对象,而不必用于描述特定的顺序或先后次序。应该理解这样使用的数据在适当情况下可以互换,以便这里描述的实施例能够以除了在这里图示或描述的内容以外的顺序实施。此外,术语“包括”或“具有”及其任何变形,意图在于覆盖不排他的包含,例如,包含了一系列步骤或单元的过程、方法、系统、产品或设备不必限于清楚地列出的那些步骤或单元,而是可包括没有清楚地列出的或对于这些过程、方法、产品或设备固有的其它步骤或单元。
请参阅图1,本申请实施例提供的医学领域意图识别方法的流程图,具体包括:
101、获取初始文本数据,初始文本数据包括目标用户输入的问题句子和/或问题词语。
服务器获取初始文本数据,初始文本数据包括目标用户输入的问题句子和/或问题词语。其中,问题句子可以包括多种形式,问题句子可以是语义明确且表达简洁的句子,例如:“肠胃炎怎么治疗?”、“肠胃炎需要吃什么药?”、“小孩发烧需要怎么办?”等;问题句子也可以是语义不明确的句子,例如:“你帮我看一下”、“我这是怎么了”、“身体感到不适”、“总感觉头晕”等;问题句子还可以是表达冗余的句子,例如,“我的胃有点难受,身体不适”、“我嗓子有点不舒服,吞咽困难,可能是上火了”、“昨天去做肝脏彩超了,B超结果为回声致密、增强、增粗”,问题词语可以只包括一个关键词,例如,问题词语可以为“糖尿病”、“偏头痛”、“胃炎”“阴道炎”等。
可以理解的是,本申请的执行主体可以为医学领域意图识别装置,还可以是服务器,具体此处不做限定。本申请实施例以服务器为执行主体为例进行说明。
102、调用预置的联合模型对初始文本数据进行改写,生成目标问题语句,目标问题语句为语义明确且无冗余的句子。
服务器调用预置的联合模型对初始文本数据进行改写,生成目标问题语句,目标问题语句为语义明确且无冗余的句子。具体的,服务器调用预置的联合模型对初始文本数据进行实体抽取并进行归一化处理,得到多个标准实体;服务器在预置的医学领域知识图谱中查找每个标准实体对应的属性信息,得到多个标准属性信息,每个标准属性信息对应一个不同的标准实体;服务器根据多个标准属性信息和多个标准实体对初始文本数据进行信息补全,生成多个候选问题语句并将多个候选问题语句发送至终端;服务器根据终端反馈的指令确定目标问题语句,目标问题语句为语义明确且无冗余的句子。
例如,用户输入“胃炎”,无法理解用户想要咨询关于“胃炎”的具体问题,通过医学领域知识图谱查找“胃炎”的属性,生成两个候选问题语句“胃炎怎么治疗?”和“胃炎需要注意什么?”,将上述两个候选问题语句发送到目标用户使用的终端,以使得目标用户在终端上进行选择,当获取到终端反馈的指令时,确定目标问题语句。假如,目标用户选择了“胃炎怎么治疗?”,那么该问题语句为目标问题语句。可以理解的是,让用户进行选择,更加有利于理解用户的需求。
103、基于预置的医学专有特征词库对目标问题语句进行特征提取,得到多个目标特征、多个目标特征权重和多个目标结构化信息。
服务器基于预置的医学专有特征词库对目标问题语句进行特征提取,得到多个目标特征、多个目标特征权重和多个目标结构化信息。具体的,服务器调用预置的联合模型对目标问题语句进行实体特征抽取,得到多个候选特征;服务器调用预置的归一知识库对多个候选特征进行归一化处理,得到多个标准化特征;服务器从预置的医学领域知识图谱中查询多个标准化特征对应的上下文以及多个目标特征权重,每个标准化特征对应一个目标特征权重;服务器对对应的上下文进行结构化处理,得到多个目标结构化信息;服务器基于预置的医学专有特征词库对多个标准化特征进行筛选,得到多个目标特征。
其中,目标结构化信息可以是句子词语向量特征、位置向量、人工提取的特征或实体加权特征,还可以包括其他结构化信息,具体此处不做限定。
需要说明的是,基于专业医学文本描述的规范性和严谨性,提前构建医疗专有特征词库,如药品库、疾病库、治疗库等。医疗专有特征词库可以用于提高问分分词的准确度以及检索医学实体的相关信息。
104、调用预置的深度学习模型、多个目标特征权重和多个目标结构化信息对多个目标特征进行分类,得到分类结果,并根据分类结果确定目标用户的医学领域意图。
服务器调用预置的深度学习模型、多个目标特征权重和多个目标结构化信息对多个目标特征进行分类,得到分类结果,并根据分类结果确定目标用户的医学领域意图。
其中,深度学习模型使用改进后的textcnn模型,输入层包括句子词语向量特征、位置向量、人工提取的特征、实体加权特征,进行特征融合,得到输入层特征(融合特征),将输入层特征传入卷积层,通过卷积层学习句子特征向量,使用句子特征向量通过全连接层完成意图分类,其中,卷积层的特征图的滑动步长的大小设置是根据训练数据实体和医学文本的特征构建的医疗信息库平均长度进行设定的。
需要说明的是,预置的深度学习模型为已经提前训练好的模型,为了提高模型的稳定性,对于有限的标注数据进行数据增强,使用已有的学习模型,对无监督数据进行预打标,选择置信度较高的数据加入到标注数据中,得到增强后的标注数据,即训练数据,从而实现训练数据的增强。
本申请实施例,将深度学习与医学文本专有的特征相结合,对用户输入的文本进行改写,并结合医学领域的知识图谱,对改写后的文本进行医学意图识别,能够精准识别用户 咨询的医学领域意图,提高了医学领域意图识别结果的准确度。并且本方案可应用于智慧医疗领域中,从而推动智慧城市的建设。
请参阅图2,本申请实施例提供的医学领域意图识别方法的另一个流程图,具体包括:
201、获取初始文本数据,初始文本数据包括目标用户输入的问题句子和/或问题词语。
服务器获取初始文本数据,初始文本数据包括目标用户输入的问题句子和/或问题词语。其中,问题句子可以包括多种形式,问题句子可以是语义明确且表达简洁的句子,例如:“肠胃炎怎么治疗?”、“肠胃炎需要吃什么药?”、“小孩发烧需要怎么办?”等;问题句子也可以是语义不明确的句子,例如:“你帮我看一下”、“我这是怎么了”、“身体感到不适”、“总感觉头晕”等;问题句子还可以是表达冗余的句子,例如,“我的胃有点难受,身体不适”、“我嗓子有点不舒服,吞咽困难,可能是上火了”、“昨天去做肝脏彩超了,B超结果为回声致密、增强、增粗”,问题词语可以只包括一个关键词,例如,问题词语可以为“糖尿病”、“偏头痛”、“胃炎”“阴道炎”等。
可以理解的是,本申请的执行主体可以为医学领域意图识别装置,还可以是服务器,具体此处不做限定。本申请实施例以服务器为执行主体为例进行说明。
可选的,在步骤201之前,还可以包括:
服务器构建预置的医学知识图谱。
可选的,服务器构建预置的医学知识图谱具体包括:
服务器获取医学训练数据,并对医学训练数据进行整合;服务器对医学训练数据进行实体抽取和关系抽取,生成知识三元组;服务器对知识三元组进行数据标注,得到已标注数据;服务器基于已标注数据对预置的序列标注模型进行训练和知识推理,生成预置的医学知识图谱。
其中,可选的,服务器对医学训练数据进行实体抽取和关系抽取,生成知识三元组,包括:服务器对医学训练数据进行实体抽取,得到多个实体;服务器对多个实体之间的关系进行抽取,得到多个实体关系;服务器根据多个实体关系和多个实体生成知识三元组。
实体识别是指识别医学文本中具有特定意义的实体,主要包括主体、地点、检查项目、疾病等实体,实体关系抽取需要根据文本语义信息推测实体间的关系;比如给定句子:“孕妇可以吃螃蟹吗”,该句实体为“孕妇”孕妇与“螃蟹”,预置的联合抽取模型可以通过语义得到“不适宜”的关系,并最终抽取出(孕妇,不适宜,螃蟹)的知识三元组。知识三元组为有关系的实体三元组。
其中,预置的联合模型用于对实体和实体关系进行抽取,将采用“BIOES”模式的标注数据,其中,B表示开始,I表示内部,O表示非实体,E实体尾部,S表示改词本身就是一个实体。序列化标注模型为BILSTM+CRF深度学习模型,使用已标注数据(已标注的三元组)对BILSTM+CRF深度学习模型进行训练和优化,得到医学知识图谱。
202、调用预置的联合模型对初始文本数据进行实体抽取并进行归一化处理,得到多个标准实体;
具体的,服务器调用预置的联合模型对初始文本数据进行实体抽取,得到多个初始实体;服务器调用预置的归一知识库对多个初始实体进行归一化处理,得到多个标准实体。
例如,“验孕纸”和“测孕纸”都标准化为“验孕试纸”。又例如,“运动神经元病”和“夏科病”都标准化为“肌萎缩侧硬化”,“艾滋病”和“AIDS”都标准化为“获得性免疫缺陷综合症”。
需要说明的是,对用户的初始文本数据进行实体抽取并归一化,用户文本的标准化有助于后续意图识别,避免出现语义不准确和冗余。
203、在预置的医学领域知识图谱中查找每个标准实体对应的属性信息,得到多个标准属性信息,每个标准属性信息对应一个不同的标准实体;
不同的标准实体对应不同的标准属性信息,例如,“胃炎”对应的属性信息为“治疗方法”和“注意事项”、“阿莫西林”对应的属性信息为“适应症”和“禁忌症”。
204、根据多个标准属性信息和多个标准实体对初始文本数据进行信息补全,生成多个候选问题语句,并将多个候选问题语句发送至终端;
具体的,服务器将初始文本数据与预置句式模板进行匹配,得到目标句式模板;服务器按照目标句式模板对初始文本数据的句式进行调整,得到初始语句;服务器将初始语句中的多个初始实体替换为对应的多个标准实体,生成中间语句;服务器根据多个标准属性信息对中间语句进行补全,生成多个候选问题语句;服务器将多个候选问题语句发送至终端。
例如,初始文本数据为“B超和彩超有什么不同呢”,通过预置句式模板匹配到“a和b有什么区别”句式,该句式对应的模板有“((.*)([和同跟与])(.*))——{0}{和}{2}{有什么区别}”,根据改预置句式模板将“B超和彩超有什么不同呢”最终修改为“B超和彩超有什么区别”。
例如,用户输入的“阴道炎”,对应的标准实体也是“阴道炎”,通过知识图谱查找到“阴道炎”的属性信息为“治疗方法”和“注意事项”,那么补全信息后的对用户提问问题为“阴道炎怎么治疗?阴道炎需要注意什么?”
需要说明的是,属性信息是为实体构造的属性列表,药品的属性包括适应症、禁忌症,疾病的属性可以包括治疗药物、注意事项等,例如,阿莫西林是青霉素过敏者禁用。
例如,用户输入“胃炎”,无法理解用户想要咨询关于“胃炎”的具体问题,通过医学领域知识图谱查找“胃炎”的属性,生成两个候选问题语句“胃炎怎么治疗?”和“胃炎需要注意什么?”,将上述两个候选问题语句发送到目标用户使用的终端,以使得目标用户在终端上进行选择,当获取到终端反馈的指令时,确定目标问题语句。
205、根据终端反馈的指令确定目标问题语句,目标问题语句为语义明确且无冗余的句子;
例如,对于问题“胃炎怎么治疗?”和“胃炎需要注意什么?”,目标用户选择了“胃炎怎么治疗?”,那么该问题语句“胃炎怎么治疗?”确定为目标问题语句。可以理解的是,让目标用户进行选择,更加有利于理解用户的需求。
206、基于预置的医学专有特征词库对目标问题语句进行特征提取,得到多个目标特征、多个目标特征权重和多个目标结构化信息。
服务器基于预置的医学专有特征词库对目标问题语句进行特征提取,得到多个目标特征、多个目标特征权重和多个目标结构化信息。具体的,服务器调用预置的联合模型对目标问题语句进行实体特征抽取,得到多个候选特征;服务器调用预置的归一知识库对多个候选特征进行归一化处理,得到多个标准化特征;服务器从预置的医学领域知识图谱中查询多个标准化特征对应的上下文以及多个目标特征权重,每个标准化特征对应一个目标特征权重;服务器对对应的上下文进行结构化处理,得到多个目标结构化信息;服务器基于预置的医学专有特征词库对多个标准化特征进行筛选,得到多个目标特征。
其中,目标结构化信息可以是句子词语向量特征、位置向量、人工提取的特征或实体加权特征,还可以包括其他结构化信息,具体此处不做限定。
需要说明的是,基于专业医学文本描述的规范性和严谨性,提前构建医疗专有特征词库,如药品库、疾病库、治疗库等。医疗专有特征词库可以用于提高问分分词的准确度以及检索医学实体的相关信息。
207、调用预置的深度学习模型、多个目标特征权重和多个目标结构化信息对多个目标特征进行分类,得到分类结果,并根据分类结果确定目标用户的医学领域意图。
服务器调用预置的深度学习模型、多个目标特征权重和多个目标结构化信息对多个目 标特征进行分类,得到分类结果,并根据分类结果确定目标用户的医学领域意图。
其中,深度学习模型使用改进后的textcnn模型,输入层包括句子词语向量特征、位置向量、人工提取的特征、实体加权特征,进行特征融合,得到输入层特征(融合特征),将输入层特征传入卷积层,通过卷积层学习句子特征向量,使用句子特征向量通过全连接层完成意图分类,其中,卷积层的特征图的滑动步长的大小设置是根据训练数据实体和医学文本的特征构建的医疗信息库平均长度进行设定的。
需要说明的是,预置的深度学习模型为已经提前训练好的模型,为了提高模型的稳定性,对于有限的标注数据进行数据增强,使用已有的学习模型,对无监督数据进行预打标,选择置信度较高的数据加入到标注数据中,得到增强后的标注数据,即训练数据,从而实现训练数据的增强。
208、根据目标用户的医学领域意图搜索问题答案。
服务器根据目标用户的医学领域意图搜索问题答案。
可选的,当目标用户的医学领域意图为查询治疗胃溃疡的药物时,服务器将在预置的医学知识图谱中确定胃药的相关信息;服务器对胃药的相关信息进行分类,得到制酸类胃药的信息和治愈溃疡类胃药的信息;服务器将治愈溃疡类胃药的信息按照预置格式生成问题答案,并反馈给目标用户的目标终端。
因为能够准确识别到用户输入问题中的医学领域意图,因此,可以更精确的检索用户输入问题所需的答案,减少了答案的检索时长,提高了答案检索效率。
本申请实施例,将深度学习与医学文本专有的特征相结合,对用户输入的文本进行改写,并结合医学领域的知识图谱,对改写后的文本进行医学意图识别,能够精准识别用户咨询的医学领域意图,提高了医学领域意图识别结果的准确度。并且本方案可应用于智慧医疗领域中,从而推动智慧城市的建设。
上面对本申请实施例中医学领域意图识别方法进行了描述,下面对本申请实施例中医学领域意图识别装置进行描述,请参阅图3,本申请实施例中医学领域意图识别装置的一个实施例包括:
数据获取模块301,用于获取初始文本数据,所述初始文本数据包括目标用户输入的问题句子和/或问题词语;
数据改写模块302,用于调用预置的联合模型对所述初始文本数据进行改写,生成目标问题语句,所述目标问题语句为语义明确且无冗余的句子;
特征提取模块303,用于基于预置的医学专有特征词库对所述目标问题语句进行特征提取,得到多个目标特征、多个目标特征权重和多个目标结构化信息;
意图确定模块304,用于调用预置的深度学习模型、所述多个目标特征权重和所述多个目标结构化信息对所述多个目标特征进行分类,得到分类结果,并根据所述分类结果确定所述目标用户的医学领域意图。
本申请实施例,将深度学习与医学文本专有的特征相结合,对用户输入的文本进行改写,并结合医学领域的知识图谱,对改写后的文本进行医学意图识别,能够精准识别用户咨询的医学领域意图,提高了医学领域意图识别结果的准确度。并且本方案可应用于智慧医疗领域中,从而推动智慧城市的建设。
请参阅图4,本申请实施例中医学领域意图识别装置的另一个实施例包括:
数据获取模块301,用于获取初始文本数据,所述初始文本数据包括目标用户输入的问题句子和/或问题词语;
数据改写模块302,用于调用预置的联合模型对所述初始文本数据进行改写,生成目标问题语句,所述目标问题语句为语义明确且无冗余的句子;
特征提取模块303,用于基于预置的医学专有特征词库对所述目标问题语句进行特征 提取,得到多个目标特征、多个目标特征权重和多个目标结构化信息;
意图确定模块304,用于调用预置的深度学习模型、所述多个目标特征权重和所述多个目标结构化信息对所述多个目标特征进行分类,得到分类结果,并根据所述分类结果确定所述目标用户的医学领域意图。
可选的,数据改写模块302包括:
抽取单元3021,用于调用预置的联合模型对所述初始文本数据进行实体抽取并进行归一化处理,得到多个标准实体;
查找单元3022,用于在预置的医学领域知识图谱中查找每个标准实体对应的属性信息,得到多个标准属性信息,每个标准属性信息对应一个不同的标准实体;
补全单元3023,用于根据所述多个标准属性信息和所述多个标准实体对所述初始文本数据进行信息补全,生成多个候选问题语句,并将所述多个候选问题语句发送至终端;
确定单元3024,用于根据所述终端反馈的指令确定目标问题语句,所述目标问题语句为语义明确且无冗余的句子。
可选的,抽取单元3021具体用于:
调用预置的联合模型对所述初始文本数据进行实体抽取,得到多个初始实体;调用预置的归一知识库对所述多个初始实体进行归一化处理,得到多个标准实体。
可选的,补全单元3023具体用于:
将所述初始文本数据与预置句式模板进行匹配,得到目标句式模板;按照所述目标句式模板对所述初始文本数据的句式进行调整,得到初始语句;将所述初始语句中的多个初始实体替换为对应的多个标准实体,生成中间语句;根据所述多个标准属性信息对所述中间语句进行补全,生成多个候选问题语句;将所述多个候选问题语句发送至终端。
可选的,特征提取模块303具体用于:
调用所述预置的联合模型对所述目标问题语句进行实体特征抽取,得到多个候选特征;调用预置的归一知识库对所述多个候选特征进行归一化处理,得到多个标准化特征;从预置的医学领域知识图谱中查询所述多个标准化特征对应的上下文以及多个目标特征权重,每个标准化特征对应一个目标特征权重;对所述对应的上下文进行结构化处理,得到多个目标结构化信息;基于预置的医学专有特征词库对所述多个标准化特征进行筛选,得到多个目标特征。
可选的,医学领域意图识别装置还包括:
图谱构建模块305,用于构建预置的医学知识图谱。
可选的,图谱构建模块305具体用于:
获取医学训练数据,并对医学训练数据进行整合;对所述医学训练数据进行实体抽取和关系抽取,生成知识三元组;对所述知识三元组进行数据标注,得到已标注数据;基于所述已标注数据对预置的序列标注模型进行训练和知识推理,生成预置的医学知识图谱。
本申请实施例,将深度学习与医学文本专有的特征相结合,对用户输入的文本进行改写,并结合医学领域的知识图谱,对改写后的文本进行医学意图识别,能够精准识别用户咨询的医学领域意图,提高了医学领域意图识别结果的准确度。并且本方案可应用于智慧医疗领域中,从而推动智慧城市的建设。
上面图3至图4从模块化功能实体的角度对本申请实施例中的医学领域意图识别装置进行详细描述,下面从硬件处理的角度对本申请实施例中医学领域意图识别设备进行详细描述。
图5是本申请实施例提供的一种医学领域意图识别设备的结构示意图,该医学领域意图识别设备500可因配置或性能不同而产生比较大的差异,可以包括一个或一个以上处理器(central processing units,CPU)510(例如,一个或一个以上处理器)和存储器520, 一个或一个以上存储应用程序533或数据532的存储介质530(例如一个或一个以上海量存储设备)。其中,存储器520和存储介质530可以是短暂存储或持久存储。存储在存储介质530的程序可以包括一个或一个以上模块(图示没标出),每个模块可以包括对医学领域意图识别设备500中的一系列指令操作。更进一步地,处理器510可以设置为与存储介质530通信,在医学领域意图识别设备500上执行存储介质530中的一系列指令操作。
医学领域意图识别设备500还可以包括一个或一个以上电源540,一个或一个以上有线或无线网络接口550,一个或一个以上输入输出接口560,和/或,一个或一个以上操作系统531,例如Windows Serve,Mac OS X,Unix,Linux,FreeBSD等等。本领域技术人员可以理解,图5示出的医学领域意图识别设备结构并不构成对医学领域意图识别设备的限定,可以包括比图示更多或更少的部件,或者组合某些部件,或者不同的部件布置。处理器510可以执行上述实施例中数据获取模块301、数据改写模块302、特征提取模块303、意图确定模块304和图谱构建模块305的功能。
本申请还提供一种医学领域意图识别设备,包括:存储器和至少一个处理器,所述存储器中存储有指令,所述存储器和所述至少一个处理器通过线路互连;所述至少一个处理器调用所述存储器中的所述指令,以使得所述医学领域意图识别设备执行上述医学领域意图识别方法中的步骤。
本申请还提供一种计算机可读存储介质,该计算机可读存储介质可以为非易失性计算机可读存储介质,也可以为易失性计算机可读存储介质。计算机可读存储介质存储有计算机指令,当所述计算机指令在计算机上运行时,使得计算机执行如下步骤:
获取初始文本数据,所述初始文本数据包括目标用户输入的问题句子和/或问题词语;
调用预置的联合模型对所述初始文本数据进行改写,生成目标问题语句,所述目标问题语句为语义明确且无冗余的句子;
基于预置的医学专有特征词库对所述目标问题语句进行特征提取,得到多个目标特征、多个目标特征权重和多个目标结构化信息;
调用预置的深度学习模型、所述多个目标特征权重和所述多个目标结构化信息对所述多个目标特征进行分类,得到分类结果,并根据所述分类结果确定所述目标用户的医学领域意图。
所属领域的技术人员可以清楚地了解到,为描述的方便和简洁,上述描述的系统,装置和单元的具体工作过程,可以参考前述方法实施例中的对应过程,在此不再赘述。
所述集成的单元如果以软件功能单元的形式实现并作为独立的产品销售或使用时,可以存储在一个计算机可读取存储介质中。基于这样的理解,本申请的技术方案本质上或者说对现有技术做出贡献的部分或者该技术方案的全部或部分可以以软件产品的形式体现出来,该计算机软件产品存储在一个存储介质中,包括若干指令用以使得一台计算机设备(可以是个人计算机,服务器,或者网络设备等)执行本申请各个实施例所述方法的全部或部分步骤。而前述的存储介质包括:U盘、移动硬盘、只读存储器(read-only memory,ROM)、随机存取存储器(random access memory,RAM)、磁碟或者光盘等各种可以存储程序代码的介质。
以上所述,以上实施例仅用以说明本申请的技术方案,而非对其限制;尽管参照前述实施例对本申请进行了详细的说明,本领域的普通技术人员应当理解:其依然可以对前述各实施例所记载的技术方案进行修改,或者对其中部分技术特征进行等同替换;而这些修改或者替换,并不使相应技术方案的本质脱离本申请各实施例技术方案的精神和范围。

Claims (20)

  1. 一种医学领域意图识别方法,包括:
    获取初始文本数据,所述初始文本数据包括目标用户输入的问题句子和/或问题词语;
    调用预置的联合模型对所述初始文本数据进行改写,生成目标问题语句,所述目标问题语句为语义明确且无冗余的句子;
    基于预置的医学专有特征词库对所述目标问题语句进行特征提取,得到多个目标特征、多个目标特征权重和多个目标结构化信息;
    调用预置的深度学习模型、所述多个目标特征权重和所述多个目标结构化信息对所述多个目标特征进行分类,得到分类结果,并根据所述分类结果确定所述目标用户的医学领域意图。
  2. 根据权利要求1所述的医学领域意图识别方法,其中,所述调用预置的联合模型对所述初始文本数据进行改写,生成目标问题语句,所述目标问题语句为语义明确且无冗余的句子,包括:
    调用预置的联合模型对所述初始文本数据进行实体抽取并进行归一化处理,得到多个标准实体;
    在预置的医学领域知识图谱中查找每个标准实体对应的属性信息,得到多个标准属性信息,每个标准属性信息对应一个不同的标准实体;
    根据所述多个标准属性信息和所述多个标准实体对所述初始文本数据进行信息补全,生成多个候选问题语句,并将所述多个候选问题语句发送至终端;
    根据所述终端反馈的指令确定目标问题语句,所述目标问题语句为语义明确且无冗余的句子。
  3. 根据权利要求2所述的医学领域意图识别方法,其中,所述调用预置的联合模型对所述初始文本数据进行实体抽取并进行归一化处理,得到多个标准实体,包括:
    调用预置的联合模型对所述初始文本数据进行实体抽取,得到多个初始实体;
    调用预置的归一知识库对所述多个初始实体进行归一化处理,得到多个标准实体。
  4. 根据权利要求3所述的医学领域意图识别方法,其中,所述根据所述多个标准属性信息和所述多个标准实体对所述初始文本数据进行信息补全,生成多个候选问题语句,并将所述多个候选问题语句发送至终端,包括:
    将所述初始文本数据与预置句式模板进行匹配,得到目标句式模板;
    按照所述目标句式模板对所述初始文本数据的句式进行调整,得到初始语句;
    将所述初始语句中的多个初始实体替换为对应的多个标准实体,生成中间语句;
    根据所述多个标准属性信息对所述中间语句进行补全,生成多个候选问题语句;
    将所述多个候选问题语句发送至终端。
  5. 根据权利要求1所述的医学领域意图识别方法,其中,所述基于预置的医学专有特征词库对所述目标问题语句进行特征提取,得到多个目标特征、多个目标特征权重和多个目标结构化信息,包括:
    调用所述预置的联合模型对所述目标问题语句进行实体特征抽取,得到多个候选特征;
    调用预置的归一知识库对所述多个候选特征进行归一化处理,得到多个标准化特征;
    从预置的医学领域知识图谱中查询所述多个标准化特征对应的上下文以及多个目标特征权重,每个标准化特征对应一个目标特征权重;
    对所述对应的上下文进行结构化处理,得到多个目标结构化信息;
    基于预置的医学专有特征词库对所述多个标准化特征进行筛选,得到多个目标特征。
  6. 根据权利要求1-5中任一项所述的医学领域意图识别方法,其中,在所述获取初始文本数据,所述初始文本数据包括目标用户输入的问题句子和/或问题词语之前,所述医学 领域意图识别方法还包括:
    构建预置的医学知识图谱。
  7. 根据权利要求6所述的医学领域意图识别方法,其中,所述构建预置的医学知识图谱,包括:
    获取医学训练数据,并对医学训练数据进行整合;
    对所述医学训练数据进行实体抽取和关系抽取,生成知识三元组;
    对所述知识三元组进行数据标注,得到已标注数据;
    基于所述已标注数据对预置的序列标注模型进行训练和知识推理,生成预置的医学知识图谱。
  8. 一种医学领域意图识别设备,包括存储器、处理器及存储在所述存储器上并可在所述处理器上运行的计算机可读指令,所述处理器执行所述计算机可读指令时实现如下步骤:
    获取初始文本数据,所述初始文本数据包括目标用户输入的问题句子和/或问题词语;
    调用预置的联合模型对所述初始文本数据进行改写,生成目标问题语句,所述目标问题语句为语义明确且无冗余的句子;
    基于预置的医学专有特征词库对所述目标问题语句进行特征提取,得到多个目标特征、多个目标特征权重和多个目标结构化信息;
    调用预置的深度学习模型、所述多个目标特征权重和所述多个目标结构化信息对所述多个目标特征进行分类,得到分类结果,并根据所述分类结果确定所述目标用户的医学领域意图。
  9. 根据权利要求8所述的医学领域意图识别设备,所述处理器执行所述计算机程序时还实现以下步骤:
    调用预置的联合模型对所述初始文本数据进行实体抽取并进行归一化处理,得到多个标准实体;
    在预置的医学领域知识图谱中查找每个标准实体对应的属性信息,得到多个标准属性信息,每个标准属性信息对应一个不同的标准实体;
    根据所述多个标准属性信息和所述多个标准实体对所述初始文本数据进行信息补全,生成多个候选问题语句,并将所述多个候选问题语句发送至终端;
    根据所述终端反馈的指令确定目标问题语句,所述目标问题语句为语义明确且无冗余的句子。
  10. 根据权利要求9所述的医学领域意图识别设备,所述处理器执行所述计算机程序时还实现以下步骤:
    调用预置的联合模型对所述初始文本数据进行实体抽取,得到多个初始实体;
    调用预置的归一知识库对所述多个初始实体进行归一化处理,得到多个标准实体。
  11. 根据权利要求10所述的医学领域意图识别设备,所述处理器执行所述计算机程序时还实现以下步骤:
    将所述初始文本数据与预置句式模板进行匹配,得到目标句式模板;
    按照所述目标句式模板对所述初始文本数据的句式进行调整,得到初始语句;
    将所述初始语句中的多个初始实体替换为对应的多个标准实体,生成中间语句;
    根据所述多个标准属性信息对所述中间语句进行补全,生成多个候选问题语句;
    将所述多个候选问题语句发送至终端。
  12. 根据权利要求8所述的医学领域意图识别设备,所述处理器执行所述计算机程序时还实现以下步骤:
    调用所述预置的联合模型对所述目标问题语句进行实体特征抽取,得到多个候选特征;
    调用预置的归一知识库对所述多个候选特征进行归一化处理,得到多个标准化特征;
    从预置的医学领域知识图谱中查询所述多个标准化特征对应的上下文以及多个目标特征权重,每个标准化特征对应一个目标特征权重;
    对所述对应的上下文进行结构化处理,得到多个目标结构化信息;
    基于预置的医学专有特征词库对所述多个标准化特征进行筛选,得到多个目标特征。
  13. 根据权利要求8-12中任一项所述的医学领域意图识别设备,所述处理器执行所述计算机程序时实现所述获取初始文本数据,所述初始文本数据包括目标用户输入的问题句子和/或问题词语之前,还包括以下步骤:
    构建预置的医学知识图谱。
  14. 根据权利要求13所述的医学领域意图识别设备,所述处理器执行所述计算机程序时还实现以下步骤:
    获取医学训练数据,并对医学训练数据进行整合;
    对所述医学训练数据进行实体抽取和关系抽取,生成知识三元组;
    对所述知识三元组进行数据标注,得到已标注数据;
    基于所述已标注数据对预置的序列标注模型进行训练和知识推理,生成预置的医学知识图谱。
  15. 一种计算机可读存储介质,所述计算机可读存储介质中存储计算机指令,当所述计算机指令在计算机上运行时,使得计算机执行如下步骤:
    获取初始文本数据,所述初始文本数据包括目标用户输入的问题句子和/或问题词语;
    调用预置的联合模型对所述初始文本数据进行改写,生成目标问题语句,所述目标问题语句为语义明确且无冗余的句子;
    基于预置的医学专有特征词库对所述目标问题语句进行特征提取,得到多个目标特征、多个目标特征权重和多个目标结构化信息;
    调用预置的深度学习模型、所述多个目标特征权重和所述多个目标结构化信息对所述多个目标特征进行分类,得到分类结果,并根据所述分类结果确定所述目标用户的医学领域意图。
  16. 根据权利要求15所述的计算机可读存储介质,当所述计算机指令在计算机上运行时,使得计算机还执行以下步骤:
    调用预置的联合模型对所述初始文本数据进行实体抽取并进行归一化处理,得到多个标准实体;
    在预置的医学领域知识图谱中查找每个标准实体对应的属性信息,得到多个标准属性信息,每个标准属性信息对应一个不同的标准实体;
    根据所述多个标准属性信息和所述多个标准实体对所述初始文本数据进行信息补全,生成多个候选问题语句,并将所述多个候选问题语句发送至终端;
    根据所述终端反馈的指令确定目标问题语句,所述目标问题语句为语义明确且无冗余的句子。
  17. 根据权利要求16所述的计算机可读存储介质,当所述计算机指令在计算机上运行时,使得计算机还执行以下步骤:
    调用预置的联合模型对所述初始文本数据进行实体抽取,得到多个初始实体;
    调用预置的归一知识库对所述多个初始实体进行归一化处理,得到多个标准实体。
  18. 根据权利要求17所述的计算机可读存储介质,当所述计算机指令在计算机上运行时,使得计算机还执行以下步骤:
    将所述初始文本数据与预置句式模板进行匹配,得到目标句式模板;
    按照所述目标句式模板对所述初始文本数据的句式进行调整,得到初始语句;
    将所述初始语句中的多个初始实体替换为对应的多个标准实体,生成中间语句;
    根据所述多个标准属性信息对所述中间语句进行补全,生成多个候选问题语句;
    将所述多个候选问题语句发送至终端。
  19. 根据权利要求15所述的计算机可读存储介质,当所述计算机指令在计算机上运行时,使得计算机还执行以下步骤:
    调用所述预置的联合模型对所述目标问题语句进行实体特征抽取,得到多个候选特征;
    调用预置的归一知识库对所述多个候选特征进行归一化处理,得到多个标准化特征;
    从预置的医学领域知识图谱中查询所述多个标准化特征对应的上下文以及多个目标特征权重,每个标准化特征对应一个目标特征权重;
    对所述对应的上下文进行结构化处理,得到多个目标结构化信息;
    基于预置的医学专有特征词库对所述多个标准化特征进行筛选,得到多个目标特征。
  20. 一种医学领域意图识别装置,所述医学领域意图识别装置包括:
    数据获取模块,用于获取初始文本数据,所述初始文本数据包括目标用户输入的问题句子和/或问题词语;
    数据改写模块,用于调用预置的联合模型对所述初始文本数据进行改写,生成目标问题语句,所述目标问题语句为语义明确且无冗余的句子;
    特征提取模块,用于基于预置的医学专有特征词库对所述目标问题语句进行特征提取,得到多个目标特征、多个目标特征权重和多个目标结构化信息;
    意图确定模块,用于调用预置的深度学习模型、所述多个目标特征权重和所述多个目标结构化信息对所述多个目标特征进行分类,得到分类结果,并根据所述分类结果确定所述目标用户的医学领域意图。
PCT/CN2021/084653 2020-08-28 2021-03-31 医学领域意图识别方法、装置、设备及存储介质 WO2022041728A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202010885018.X 2020-08-28
CN202010885018.XA CN112035637A (zh) 2020-08-28 2020-08-28 医学领域意图识别方法、装置、设备及存储介质

Publications (1)

Publication Number Publication Date
WO2022041728A1 true WO2022041728A1 (zh) 2022-03-03

Family

ID=73586083

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2021/084653 WO2022041728A1 (zh) 2020-08-28 2021-03-31 医学领域意图识别方法、装置、设备及存储介质

Country Status (2)

Country Link
CN (1) CN112035637A (zh)
WO (1) WO2022041728A1 (zh)

Cited By (13)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114842982A (zh) * 2022-07-06 2022-08-02 广东省科技基础条件平台中心 一种面向医疗信息系统的知识表达方法、装置及系统
CN114925181A (zh) * 2022-04-28 2022-08-19 支付宝(杭州)信息技术有限公司 数据处理方法及装置、计算机存储介质及终端
CN115270802A (zh) * 2022-09-29 2022-11-01 中科雨辰科技有限公司 一种问题语句的处理方法、电子设备及存储介质
CN116364296A (zh) * 2023-02-17 2023-06-30 中国人民解放军总医院 标准检查项目名称确认方法、装置、设备、介质及产品
CN116483982A (zh) * 2023-06-25 2023-07-25 北京中关村科金技术有限公司 知识问答方法、装置、电子设备及可读存储介质
CN116933800A (zh) * 2023-09-12 2023-10-24 深圳须弥云图空间科技有限公司 一种基于模版的生成式意图识别方法及装置
CN117171332A (zh) * 2023-11-02 2023-12-05 江西拓世智能科技股份有限公司 基于ai的智能问答方法及系统
CN117252211A (zh) * 2023-10-18 2023-12-19 中建八局第一数字科技有限公司 提问意图识别模型的构建、提问意图识别方法及装置
CN117725961A (zh) * 2024-02-18 2024-03-19 智慧眼科技股份有限公司 医疗意图识别模型训练方法、医疗意图识别方法及设备
CN117874240A (zh) * 2024-03-12 2024-04-12 天津电力工程监理有限公司 基于知识图谱的审计文本分类方法、系统和设备
CN118070909A (zh) * 2024-04-25 2024-05-24 支付宝(杭州)信息技术有限公司 医疗问答系统的答案生成方法及装置
CN118132729A (zh) * 2024-04-28 2024-06-04 支付宝(杭州)信息技术有限公司 基于医疗知识图谱的答案生成方法及装置
CN118469005A (zh) * 2024-07-10 2024-08-09 北方健康医疗大数据科技有限公司 基于大语言模型的医学知识图谱构建方法、系统、终端及存储介质

Families Citing this family (15)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112035637A (zh) * 2020-08-28 2020-12-04 康键信息技术(深圳)有限公司 医学领域意图识别方法、装置、设备及存储介质
CN112735543B (zh) * 2020-12-30 2024-07-19 杭州依图医疗技术有限公司 医学数据的处理方法、装置及存储介质
CN113256459A (zh) * 2021-04-30 2021-08-13 深圳市鹰硕教育服务有限公司 微课视频管理方法、装置、系统及存储介质
CN113255354B (zh) * 2021-06-03 2021-12-07 北京达佳互联信息技术有限公司 搜索意图识别方法、装置、服务器及存储介质
CN115146037A (zh) * 2021-08-09 2022-10-04 上海蓬海涞讯数据技术有限公司 基于知识图谱的问答方法及系统、电子设备及存储介质
CN113657102B (zh) * 2021-08-17 2023-05-30 北京百度网讯科技有限公司 信息抽取方法、装置、设备及存储介质
CN113642312A (zh) * 2021-08-19 2021-11-12 平安医疗健康管理股份有限公司 体检数据的处理方法、装置、设备及存储介质
CN113657114A (zh) * 2021-08-31 2021-11-16 平安医疗健康管理股份有限公司 疾病名称对码列表的生成方法、装置、设备及存储介质
CN113761899A (zh) * 2021-09-07 2021-12-07 卫宁健康科技集团股份有限公司 一种医疗文本生成方法、装置、设备及存储介质
CN113886400A (zh) * 2021-10-26 2022-01-04 卫宁健康科技集团股份有限公司 一种医疗数据的管理方法、装置、计算机设备及存储介质
CN114724078B (zh) * 2022-03-28 2023-06-23 西南交通大学 基于目标检测网络与知识推理的人员行为意图识别方法
CN115862896B (zh) * 2023-02-13 2023-05-09 深圳市汇健智慧医疗有限公司 基于围术期的医患协同管理方法及系统
CN116468396B (zh) * 2023-04-14 2023-11-03 浪潮智慧科技有限公司 一种基于意图分析模型的水利业务处理方法、设备及介质
CN116842145B (zh) * 2023-04-20 2024-02-27 海信集团控股股份有限公司 基于城市问答系统的领域识别方法及设备
CN117034957B (zh) * 2023-06-30 2024-05-31 海信集团控股股份有限公司 一种融合大模型的语义理解方法及设备

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107729493A (zh) * 2017-09-29 2018-02-23 北京创鑫旅程网络技术有限公司 旅行知识图谱的构建方法、装置及旅行问答方法、装置
CN109522544A (zh) * 2018-09-27 2019-03-26 厦门快商通信息技术有限公司 基于卡方检验的句向量计算方法、文本分类方法及系统
CN109657238A (zh) * 2018-12-10 2019-04-19 宁波深擎信息科技有限公司 基于知识图谱的上下文识别补全方法、系统、终端及介质
US20200151542A1 (en) * 2018-11-12 2020-05-14 Alibaba Group Holding Limited Question and answer matching method, system and storage medium
CN112035637A (zh) * 2020-08-28 2020-12-04 康键信息技术(深圳)有限公司 医学领域意图识别方法、装置、设备及存储介质

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109918489B (zh) * 2019-02-28 2021-02-02 上海乐言信息科技有限公司 一种多策略融合的知识问答方法和系统
CN110675944A (zh) * 2019-09-20 2020-01-10 京东方科技集团股份有限公司 分诊方法及装置、计算机设备及介质
CN111339777A (zh) * 2020-02-24 2020-06-26 中国科学院自动化研究所 基于神经网络的医学相关意图识别方法及系统

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107729493A (zh) * 2017-09-29 2018-02-23 北京创鑫旅程网络技术有限公司 旅行知识图谱的构建方法、装置及旅行问答方法、装置
CN109522544A (zh) * 2018-09-27 2019-03-26 厦门快商通信息技术有限公司 基于卡方检验的句向量计算方法、文本分类方法及系统
US20200151542A1 (en) * 2018-11-12 2020-05-14 Alibaba Group Holding Limited Question and answer matching method, system and storage medium
CN109657238A (zh) * 2018-12-10 2019-04-19 宁波深擎信息科技有限公司 基于知识图谱的上下文识别补全方法、系统、终端及介质
CN112035637A (zh) * 2020-08-28 2020-12-04 康键信息技术(深圳)有限公司 医学领域意图识别方法、装置、设备及存储介质

Cited By (17)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114925181A (zh) * 2022-04-28 2022-08-19 支付宝(杭州)信息技术有限公司 数据处理方法及装置、计算机存储介质及终端
CN114842982B (zh) * 2022-07-06 2022-09-23 广东省科技基础条件平台中心 一种面向医疗信息系统的知识表达方法、装置及系统
CN114842982A (zh) * 2022-07-06 2022-08-02 广东省科技基础条件平台中心 一种面向医疗信息系统的知识表达方法、装置及系统
CN115270802A (zh) * 2022-09-29 2022-11-01 中科雨辰科技有限公司 一种问题语句的处理方法、电子设备及存储介质
CN116364296A (zh) * 2023-02-17 2023-06-30 中国人民解放军总医院 标准检查项目名称确认方法、装置、设备、介质及产品
CN116364296B (zh) * 2023-02-17 2023-12-26 中国人民解放军总医院 标准检查项目名称确认方法、装置、设备、介质及产品
CN116483982A (zh) * 2023-06-25 2023-07-25 北京中关村科金技术有限公司 知识问答方法、装置、电子设备及可读存储介质
CN116483982B (zh) * 2023-06-25 2023-10-13 北京中关村科金技术有限公司 知识问答方法、装置、电子设备及可读存储介质
CN116933800B (zh) * 2023-09-12 2024-01-05 深圳须弥云图空间科技有限公司 一种基于模版的生成式意图识别方法及装置
CN116933800A (zh) * 2023-09-12 2023-10-24 深圳须弥云图空间科技有限公司 一种基于模版的生成式意图识别方法及装置
CN117252211A (zh) * 2023-10-18 2023-12-19 中建八局第一数字科技有限公司 提问意图识别模型的构建、提问意图识别方法及装置
CN117171332A (zh) * 2023-11-02 2023-12-05 江西拓世智能科技股份有限公司 基于ai的智能问答方法及系统
CN117725961A (zh) * 2024-02-18 2024-03-19 智慧眼科技股份有限公司 医疗意图识别模型训练方法、医疗意图识别方法及设备
CN117874240A (zh) * 2024-03-12 2024-04-12 天津电力工程监理有限公司 基于知识图谱的审计文本分类方法、系统和设备
CN118070909A (zh) * 2024-04-25 2024-05-24 支付宝(杭州)信息技术有限公司 医疗问答系统的答案生成方法及装置
CN118132729A (zh) * 2024-04-28 2024-06-04 支付宝(杭州)信息技术有限公司 基于医疗知识图谱的答案生成方法及装置
CN118469005A (zh) * 2024-07-10 2024-08-09 北方健康医疗大数据科技有限公司 基于大语言模型的医学知识图谱构建方法、系统、终端及存储介质

Also Published As

Publication number Publication date
CN112035637A (zh) 2020-12-04

Similar Documents

Publication Publication Date Title
WO2022041728A1 (zh) 医学领域意图识别方法、装置、设备及存储介质
WO2021000676A1 (zh) 问答方法、问答装置、计算机设备及存储介质
CN112131393B (zh) 一种基于bert和相似度算法的医疗知识图谱问答系统构建方法
WO2022041730A1 (zh) 医疗领域意图识别方法、装置、设备及存储介质
AU2019200437B2 (en) A method to build an enterprise-specific knowledge graph
US20220121824A1 (en) Method for determining text similarity, method for obtaining semantic answer text, and question answering method
US12131827B2 (en) Knowledge graph-based question answering method, computer device, and medium
WO2021139232A1 (zh) 基于医疗知识图谱的分诊方法、装置、设备及存储介质
CN111984851B (zh) 医学资料搜索方法、装置、电子装置及存储介质
CN112487202B (zh) 融合知识图谱与bert的中文医学命名实体识别方法、装置
CN110675944A (zh) 分诊方法及装置、计算机设备及介质
WO2023029512A1 (zh) 基于知识图谱的医疗问题解答方法、装置、设备及介质
CN112148851A (zh) 一种基于知识图谱的医药知识问答系统的构建方法
CN113569023B (zh) 一种基于知识图谱的中文医药问答系统及方法
WO2018153215A1 (zh) 一种自动生成语义相近句子样本的方法
US20190340503A1 (en) Search system for providing free-text problem-solution searching
WO2021169640A1 (zh) 一种问题查询装置、方法、设备及存储介质
CN113590783B (zh) 基于nlp自然语言处理的中医养生智能问答系统
CN114004237A (zh) 一种基于膀胱癌知识图谱的智能问答系统构建方法
CN115114420A (zh) 一种知识图谱问答方法、终端设备及存储介质
CN116737911A (zh) 基于深度学习的高血压问答方法及系统
CN114969459A (zh) 一种基于认知双通道的认知推理的视觉问答方法
Montenegro et al. The HoPE model architecture: A novel approach to pregnancy information retrieval based on conversational agents
CN118051598A (zh) 药品知识问答方法、装置、电子设备及存储介质
CN113157887A (zh) 知识问答意图识别方法、装置、及计算机设备

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 21859577

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 06.06.2023)

122 Ep: pct application non-entry in european phase

Ref document number: 21859577

Country of ref document: EP

Kind code of ref document: A1