CN115186068A - Symptom question-answering method, device, equipment and storage medium based on knowledge graph - Google Patents
Symptom question-answering method, device, equipment and storage medium based on knowledge graph Download PDFInfo
- Publication number
- CN115186068A CN115186068A CN202210868668.2A CN202210868668A CN115186068A CN 115186068 A CN115186068 A CN 115186068A CN 202210868668 A CN202210868668 A CN 202210868668A CN 115186068 A CN115186068 A CN 115186068A
- Authority
- CN
- China
- Prior art keywords
- symptom
- concept
- determining
- diagnosis
- question
- 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.)
- Pending
Links
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
-
- 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
- G06F16/3331—Query processing
- G06F16/334—Query execution
- G06F16/3344—Query execution using natural language analysis
-
- 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
- G06F16/3331—Query processing
- G06F16/334—Query execution
- G06F16/3346—Query execution using probabilistic model
-
- 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
- G06F16/335—Filtering based on additional data, e.g. user or group profiles
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Artificial Intelligence (AREA)
- Human Computer Interaction (AREA)
- Probability & Statistics with Applications (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention relates to the field of artificial intelligence, in particular to a symptom question-answering method, a symptom question-answering device, a symptom question-answering equipment and a storage medium based on a knowledge graph, wherein the method is used for inquiring and matching symptom information of a target user according to a diagnosis path in the symptom knowledge graph and quickly positioning symptoms of the target user; screening out all similar candidate syndrome types in the diagnosis path according to a symptom semantic similarity matching principle, and avoiding inaccurate syndrome type diagnosis caused by different semantics; by inquiring the target user, the symptom attribute corresponding to the symptom information is determined, so that the diagnosis syndrome type is more meticulous and accurate to judge; the candidate syndrome type with the diagnosis probability larger than the preset threshold is taken as the diagnosis confirmation type through probability calculation of the candidate syndrome types with similar symptoms, so that misjudgment caused by similar semantics is avoided, the accuracy of diagnosis syndrome type is improved, and the technical problem of low accuracy of diagnosis syndrome type of the traditional Chinese medicine diagnosis system is solved.
Description
Technical Field
The invention relates to the technical field of artificial intelligence, in particular to a symptom question-answering method, a symptom question-answering device, a symptom question-answering equipment and a symptom question-answering storage medium based on a knowledge graph.
Background
The diagnosis of syndrome type in traditional Chinese medicine is a complex thinking and distinguishing process combined with the information of inquiry and hearing, wherein the clinical property of symptoms is the main basis of syndrome differentiation. In the diagnosis process of syndrome type in traditional Chinese medicine, the diagnosis results of syndrome type of similar symptoms may be completely different. However, the existing traditional Chinese medicine diagnosis system is not careful and comprehensive enough to excavate and define clinical properties of symptoms, so that the diagnosis result is not accurate enough, and the practicability is low.
Therefore, how to solve the problem of low accuracy of diagnosis syndrome of the traditional Chinese medicine diagnosis system becomes a technical problem to be solved urgently.
Disclosure of Invention
The invention mainly aims to provide a symptom question-answering method, a symptom question-answering device, a symptom question-answering equipment and a symptom question-answering storage medium based on a knowledge graph, and aims to solve the technical problem that the diagnosis syndrome type of the traditional Chinese medicine diagnosis system is low in accuracy.
In order to achieve the above object, the present invention provides a symptom question-answering method based on a knowledge graph, which comprises: acquiring symptom information of a target user, and inquiring and matching the symptom information based on a symptom knowledge graph and a symptom semantic similar matching principle to determine a candidate syndrome type corresponding to the symptom information; initiating a symptom inquiry to the target user based on the candidate syndrome type, and determining a symptom attribute corresponding to the symptom information based on feedback data of the target user to the symptom inquiry; and calculating the diagnosis probability of the candidate syndrome type based on the symptom attribute, and determining the target syndrome type based on the diagnosis probability and a preset threshold value.
In addition, to achieve the above object, the present invention also provides a knowledge-map-based symptom question answering device including: the candidate syndrome type determining module is used for acquiring symptom information of a target user, inquiring and matching the symptom information based on a symptom knowledge graph and a symptom semantic similar matching principle, and determining a candidate syndrome type corresponding to the symptom information; the symptom attribute determining module is used for initiating a symptom inquiry to the target user based on the candidate syndrome type and determining a symptom attribute corresponding to the symptom information based on feedback data of the target user to the symptom inquiry; and the diagnosis confirming type determining module is used for calculating the diagnosis confirming probability of the candidate syndrome type based on the symptom attribute and determining the target syndrome type based on the diagnosis confirming probability and a preset threshold value.
In addition, to achieve the above object, the present invention also provides a knowledge-map-based symptom question-answering apparatus including a processor, a memory, and a knowledge-map-based symptom question-answering program stored on the memory and executable by the processor, wherein the knowledge-map-based symptom question-answering program, when executed by the processor, implements the steps of the knowledge-map-based symptom question-answering method as described above.
In addition, to achieve the above object, the present invention also provides a computer-readable storage medium having a symptom question-answering program based on a knowledge graph stored thereon, wherein the symptom question-answering program based on the knowledge graph is executed by a processor to implement the steps of the symptom question-answering method based on the knowledge graph as described above.
The invention provides a symptom question-answering method based on a knowledge graph, which is used for acquiring symptom information of a target user, inquiring and matching the symptom information based on a symptom knowledge graph and a symptom semantic similar matching principle, and determining a candidate syndrome type corresponding to the symptom information; initiating a symptom inquiry to the target user based on the candidate syndrome type, and determining a symptom attribute corresponding to the symptom information based on feedback data of the target user to the symptom inquiry; and calculating the diagnosis probability of the candidate syndrome type based on the symptom attribute, and determining the target syndrome type based on the diagnosis probability and a preset threshold value. By the mode, the symptom information of the target user is inquired and matched according to the diagnosis path in the symptom knowledge graph, and the symptom of the target user can be quickly positioned; according to the symptom semantic similarity matching principle, all similar candidate syndrome types in the diagnosis path can be screened out, and the syndrome type diagnosis is prevented from being inaccurate due to different semantics; by inquiring the target user, the symptom attribute corresponding to the symptom information is determined, so that the diagnosis syndrome type is more meticulous and accurate to judge; the candidate syndrome type with the diagnosis confirming probability larger than the preset threshold is taken as the diagnosis confirming type through probability calculation of the candidate syndrome types with similar symptoms, so that misjudgment caused by similar semantics is avoided, the accuracy of diagnosis syndrome type is improved, and the technical problem of low accuracy of diagnosis syndrome type of the traditional Chinese medicine diagnosis system is solved.
Drawings
FIG. 1 is a diagram of a hardware configuration of a knowledge-map based symptom question answering apparatus according to an embodiment of the present invention;
FIG. 2 is a schematic flow chart of a first embodiment of a method for knowledge-graph based symptomatic question-answering according to the present invention;
FIG. 3 is a schematic flow chart of a second embodiment of the method for knowledge-graph based symptomatic question-answering according to the present invention;
FIG. 4 is a schematic flow chart of a third embodiment of a method for knowledge-graph based symptomatic question-answering according to the present invention;
FIG. 5 is a functional block diagram of a first embodiment of a knowledge-graph based symptomatic question-answering apparatus according to the present invention.
The implementation, functional features and advantages of the objects of the present invention will be further explained with reference to the accompanying drawings.
Detailed Description
It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
The symptom question-answering method based on the knowledge graph is mainly applied to symptom question-answering equipment based on the knowledge graph, and the symptom question-answering equipment based on the knowledge graph can be equipment with display and processing functions, such as a PC (personal computer), a portable computer, a mobile terminal and the like.
Referring to fig. 1, fig. 1 is a schematic diagram of a hardware structure of a knowledge-map-based symptom question answering apparatus according to an embodiment of the present invention. In an embodiment of the present invention, the knowledge-graph-based symptom question and answer apparatus may include a processor 1001 (e.g., a CPU), a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. The communication bus 1002 is used for realizing connection communication among the components; the user interface 1003 may include a Display screen (Display), an input unit such as a Keyboard (Keyboard); the network interface 1004 may optionally include a standard wired interface, a wireless interface (e.g., WI-FI interface); the memory 1005 may be a high-speed RAM memory, or may be a non-volatile memory (e.g., a magnetic disk memory), and optionally, the memory 1005 may be a storage device independent of the processor 1001.
Those skilled in the art will appreciate that the hardware configuration shown in FIG. 1 does not constitute a limitation of a knowledge-graph-based symptomatic question-answering apparatus, and may include more or less components than those shown, or some components may be combined, or a different arrangement of components.
With continued reference to fig. 1, the memory 1005 of fig. 1, which is one type of computer-readable storage medium, may include an operating system, a network communication module, and a knowledge-graph-based symptom question-answering program.
In fig. 1, the network communication module is mainly used for connecting a server and performing data communication with the server; and the processor 1001 may call the knowledgemap-based symptom question-answering program stored in the memory 1005 and execute the knowledgemap-based symptom question-answering method provided by the embodiment of the present invention.
The embodiment of the invention provides a symptom question-answering method based on a knowledge graph.
Referring to fig. 2, fig. 2 is a schematic flow chart of a first embodiment of the method for symptom question-answering based on knowledge-graph according to the present invention.
In this embodiment, the symptom question-answering method based on the knowledge graph includes the following steps:
step S10, acquiring symptom information of a target user, and inquiring and matching the symptom information based on a symptom knowledge graph and a symptom semantic similar matching principle to determine a candidate syndrome type corresponding to the symptom information;
in this embodiment, in the diagnosis path of the symptom knowledge graph, each symptom concept entity includes different attributes, and the node relationship between the symptom concept entities includes a hierarchical structure, a symptom property, a combined evolution relationship, and the like.
Specifically, in the symptom knowledge graph, the symptom concept entity is divided into a plurality of levels, including a central concept, a synonymous concept, a derived concept, an upper/lower concept, a mutually exclusive concept, and the like; in the symptom knowledge graph, a central concept is taken as a basic hierarchy, and other concepts are used as extensions and supplements to the central concept, so that a hierarchical structure is formed. When diagnosis is performed through the symptom knowledge base, the corresponding central concept is matched according to the symptom information of the user, and then further symptom diagnosis is performed according to the central concept. For example, the user symptom is "cough", the central concept is "cough", and after "cough", there are also derivative concepts such as "wind-heat cough" and "phlegm-damp cough", and the derivative concepts and the central concept form a hierarchical structure.
In particular, the symptom nature is a symptomatically derived attribute including duration, urgency, home relief, type of pain, definitional factors, and the like. Symptom nature is not a simple entity, but is composed of multiple attributes, entities, relationships together. Taking "cough with expectoration" as an example, the "cough" is a symptom entity, the "expectoration" is a symptom property, and the "expectoration" also includes symptom properties such as "color, viscosity, expectoration amount", etc.
Specifically, the combined evolutionary relationship refers to the establishment of an evolutionary relationship between symptoms and signs, which is divided into qualitative evolution and combined evolution. Where the hierarchy itself implies a qualitative evolution between symptoms, such as the central concept adding a specific attribute, it becomes a derivative attribute ("headache" → "headache frequency"). The combination evolution means that one symptom entity simultaneously comprises a plurality of sub-symptoms; or conversely, when a plurality of symptoms are simultaneously satisfied, automatically converting into a new combined symptom; for example, "distending pain in the head and eyes" includes "distending pain in the head" and "distending pain in the eyes" at the same time; accordingly, if both symptom entities of "distending pain in the head" and "distending eyes" are included, the two symptom entities can be converted into "distending pain in the head and eyes".
Specifically, when a disease diagnosis model based on a symptom knowledge graph is used for traditional Chinese medicine syndrome type, symptom information input by a patient is identified, such as 'cough', the disease input by the patient is 'cough and expectoration', and the disease diagnosis model returns two candidate cough syndrome types according to the hit of symptoms in a diagnosis rule: cough due to wind-heat, phlegm-dampness.
Step S20, based on the candidate syndrome type, initiating a symptom inquiry to the target user, and based on feedback data of the target user to the symptom inquiry, determining a symptom attribute corresponding to the symptom information;
in the embodiment, according to the symptom semantic similarity matching principle, a plurality of candidate syndrome types can be obtained from the query matching result of the symptom information, and according to the symptom information of the current target user, the disease diagnosis model cannot determine the real diagnosis syndrome type of the current target user, so that more information is needed, and the target syndrome type is determined in the plurality of candidate syndrome types; because the candidate syndrome types are obtained by inquiring the same symptom information, similar attributes generally exist among a plurality of candidate syndrome types, and the basis for distinguishing the similar candidate syndrome types is the attribute difference among the syndrome types, the corresponding problems can be generated according to the similar attributes among the plurality of candidate syndrome types, more detailed symptom information can be collected for the patient, the attributes of the candidate syndrome types can be inquired and matched according to the feedback of the patient, and then the diagnosis basis is increased.
In a specific embodiment, the disease diagnosis model returns two candidate cough syndrome types according to the hit of the symptom in the diagnosis rule and the query of the 'cough' symptom: cough due to wind-heat, phlegm-dampness. Because of the two types of symptoms hitting cough and expectoration, the diagnosis model has lower diagnosis probability for the pre-result and is lower than the preset empirical threshold. At this time, the symptoms can be deeply spread according to the diagnostic rules of wind-heat cough and phlegm-damp cough. Such as: finding three dimensions of degree attribute, observation and condition factor, which can further expand the inquiry, and further generating the symptom inquiry to be inquired in the next round: "is there much sputum? "," what color sputum is? "," is expectoration in the morning? ", confirmation is performed to the patient. And determining attributes of the patient's symptoms based on the patient's feedback.
And S30, calculating the diagnosis probability of the candidate syndrome type based on the symptom attribute, and determining the target syndrome type based on the diagnosis probability and a preset threshold value.
In this embodiment, the attributes of the symptoms of the patient are determined by continuously querying the patient, and the more the attributes of the symptoms of the patient are determined, the more the disease diagnosis model can perform more accurate matching on the candidate syndrome types according to the attributes. When the diagnosis probability of one of the candidate syndrome types is higher than a preset threshold value or is obviously higher than the diagnosis probability of other candidate syndrome types, the syndrome type can be used as a confirmed syndrome type.
The embodiment provides a symptom question-answering method based on a knowledge graph, which comprises the steps of obtaining symptom information of a target user, inquiring and matching the symptom information based on a symptom knowledge graph diagnosis path and a symptom semantic similar matching principle, and determining a candidate syndrome type corresponding to the symptom information; initiating a symptom inquiry to the target user based on the inquiry result of the candidate certificate type, and determining a symptom attribute corresponding to the symptom information based on feedback data of the target user to the symptom inquiry; and judging the diagnosis confirmation probability of the candidate syndrome type based on the symptom attribute, and determining the diagnosis confirmation type based on the diagnosis confirmation probability and a preset threshold value. By the mode, the symptom information of the target user is inquired and matched according to the diagnosis path in the symptom knowledge graph, and the symptom of the target user can be quickly positioned; according to the symptom semantic similarity matching principle, all similar candidate syndrome types in the diagnosis path can be screened out, and the syndrome type diagnosis is prevented from being inaccurate due to different semantics; by inquiring the target user, the symptom attribute corresponding to the symptom information is determined, so that the diagnosis syndrome type is more meticulous and accurate to judge; the candidate syndrome type with the diagnosis confirming probability larger than the preset threshold is taken as the diagnosis confirming type through probability calculation of the candidate syndrome types with similar symptoms, so that misjudgment caused by similar semantics is avoided, the accuracy of diagnosis syndrome type is improved, and the technical problem of low accuracy of diagnosis syndrome type of the traditional Chinese medicine diagnosis system is solved.
Referring to fig. 3, fig. 3 is a schematic flow chart of a second embodiment of the symptom question-answering method based on the knowledge-graph of the present invention.
Based on the foregoing embodiment shown in fig. 2, in this embodiment, before the step S10, the method further includes:
step S01, extracting symptom concept entities based on a pre-input symptom knowledge data source, and constructing a symptom concept prefix tree based on the node relation of the symptom concept entities;
s02, determining a symptom center concept based on the weight calculation of each path in the symptom concept prefix tree;
and S03, determining symptom derivative concepts based on the symptom center concepts and the semantic similarity principle, and establishing a connection relation between the symptom derivative concepts and the symptom center concepts to obtain the symptom knowledge graph.
In this embodiment, for the construction of the symptom knowledge base, the symptom concept ontology needs to be defined first.
In this embodiment, the basis of the symptom concept ontology mainly includes a central concept and derivative concepts.
Among them, the central concept is the smallest unit of symptoms, such as "obesity", "headache", "cough", and the like;
alternatively, the central concept may have different expressions, for example, "fat" may be described as "fat", "full body" or "fat body", etc.
A derived concept is an extension to a central concept, with more symptom attributes than the central concept. A derivative concept such as "headache" may be "headache frequency", an attribute of which indicates the frequency of attacks of "headache"; the "headache" can also be "the headache gets worse after strain" as its attribute indicates "headache getting worse" due to "strain".
Optionally, the basic composition of the symptom concept ontology may also include mutually exclusive concepts, that is, one symptom is established, and the other symptom is not necessarily established; such as: the concept of mutual exclusion is "expectoration" and "no sputum".
Further, on the basis of the constitution of the symptom concept ontology, the construction of the symptom knowledge graph can add related symptom characteristic entities around the symptom concept.
In particular, symptom characteristic entities may be location, function and nature.
Wherein, the part refers to the anatomical position where symptoms occur, for example, the part corresponding to the headache is the head;
the function refers to the functional classification corresponding to the symptoms, corresponding to the five zang-organs and six fu-organs. Similarly, it is "headache", and in the theory of traditional Chinese medicine, the corresponding dysfunction of five zang organs is "liver", which is different from anatomical location.
The properties are derived attributes of the symptoms, including duration, severity, relief, pain type, definitional factors, and the like. A property is not a simple entity, but is composed of multiple classes of attributes, entities, and relationships.
In particular, the properties may include:
the degree, which represents the degree of severity of the symptoms, is: mild, moderate, severe, etc.;
duration, meaning the duration of the symptoms, such as: minutes/hour/day, etc.;
frequency, indicating the frequency of onset of symptoms, such as: recurrent attacks, multiple attacks, etc.;
urgency, which refers to the degree of urgency of a symptom, such as: "sudden epigastric pain" refers to a sudden and aggravated pain with an acute degree.
Pain, which indicates the type of pain, such as: cold pain, dull pain, etc.
The condition factors, which indicate the occurrence of symptoms, include various categories, including the causative factors: seasonal time, such as nighttime attacks, autumn attacks, etc.; physical movements, such as sedentary attacks, bed-rising attacks, labor attacks, etc.; external stimuli such as onset in rainy days, onset due to wind-cold, etc.; physiological periods, such as menstruation, pregnancy (premenstrual or menstrual onset, pregnancy onset, etc.); eating and drinking, such as hunger attack, cold drink attack, etc.; sleep emotions such as restlessness attacks, sleep attacks, etc.; accompanied by other symptoms such as the onset of emesis, etc., in which case the condition factor "emesis" is also a symptom entity.
Aggravation and relief factors, which can be established simultaneously with the condition factors. Such as: the 'cough aggravated at night' is an aggravating factor and simultaneously meets the condition factors that: seasonal hours-attacks at night. The 'vomiting and stomachache alleviation' is not only an alleviation factor, but also satisfies the following conditional factors: with other symptoms. In addition, failure to mitigate is also a factor. Such as: "stomach ache is not relieved when lying down.
Specifically, when a symptom knowledge map is constructed, an evolutionary relationship needs to be constructed between symptoms and symptoms, and the evolutionary relationship is divided into the evolution of properties and the combined evolution. Wherein, if specific attributes are added to the central concept, derivative concepts are formed (such as from the central concept of 'headache' to 'headache frequency'), namely, the property evolution between symptoms is realized; the combination means that one symptom entity simultaneously comprises a plurality of sub-symptoms, or when a plurality of symptoms are simultaneously satisfied, the combination automatically converts into a new combination symptom, for example, "head distending pain" and "eye distending" can be developed into "head distending pain".
In this embodiment, based on the defined symptom ontology structure, models such as entity identification, concept classification, and relationship extraction are used, and expert rules are combined to extract symptom relationships from a data source, so as to construct a symptom knowledge graph.
In this embodiment, the symptom concept entity is the most basic element for constructing the symptom relationship. The method comprises the steps of taking unstructured text corpora such as traditional Chinese medical book books, textbooks and traditional Chinese medicine encyclopedia websites as a symptom knowledge data source (denoted as S), inputting the symptom knowledge data source S into a construction model of a symptom knowledge graph, and automatically extracting symptom concept entities related to traditional Chinese medicine by using the construction model.
In a specific embodiment, through concept heuristic search, a text related to diagnosis (implying more symptom information) is extracted from a symptom knowledge data source S, and the text is divided into sentences to obtain a training set S' composed of a plurality of short sentences. Applying a preset symptom dictionary D (such as ' stomach ache gets hot and relieves ') and a concept word template T (such as ' stomach XX ache ', ' abdomen X full ' and the like) to the training set S ', and matching symptom concepts contained in the short sentences:
S′＝{(s i ，c i )} (1)
such as: there is a phrase "epigastric-abdominal fullness and fullness in the training set S' for more than 2 years, and stomach pain is dull and stomachache is relieved with heat after eating in the near day. "then:
s i = epigastric and abdominal fullness and fullness for more than 2 years, dull pain in stomach after eating in the near day, stomach pain gets hot and relieves
c i = { epigastric fullness and distension over 2 years, dull pain in stomach after eating in the near day, stomach pain relieved with heat }
Specifically, in the conventional method, the symptom concept is extracted by training an entity recognition model based on a large number of symptom concept entity labels, and in this embodiment, in order to reduce the labeling cost, a technique combining unsupervised and supervised is adopted.
Specifically, unsupervised learning is to compose training samples by screening sentences that inspire concepts not empty:
Wherein, the noise adding stage is based on the concept word template T to all inspiring concepts c i The concept of (1) adding noise. If the abdominal fullness X is matched with the abdominal fullness, the corresponding positions of X are randomly exchanged: ' gastric X abdominal fullness>"gastric distention and abdominal fullness".
In the specific embodiment, the symptom concept text containing missing characters and disorder is used as an input sequence and input into an Encoder (Encoder) of a bidirectional sequence for training, the trained symptom concept text is guided into a Decoder (Decoder) through a hidden layer vector h in the Encoder, the Decoder decodes the trained symptom concept text, and the symptom concept text with complete positive sequence is output. For example, it is input with "epigastric fullness and distention" and output with "epigastric fullness and distention".
Wherein Encoder may use a pre-trained model based on a Transformer, such as Bert; the Decoder may then use a pre-trained language model, such as GPT.
Specifically, supervised learning is to use an Encoder trained in unsupervised learning to interface with a downstream character classification model, and only a small amount of symptom concept entity sample labels are needed at the moment.
In the specific embodiment, sentences in a training set are input, the input sentences are trained through an Encoder, then the trained characters are predicted through a character classification model, the output characters are marked, and the predicted characters are distinguished into symptom concept words and non-symptom concept words. For example, inputting 'headache attacks for two weeks, empty head pain, dizziness and headache aggravated by wind cold', the concept word characters are spliced to obtain the predicted symptom concept entities of 'headache', 'empty head pain' and 'headache aggravated by cold'. After the supervised learning is finished, the predicted symptom concept entity is rechecked by the expert, and the symptom concept entity passing the rechecking is added into the preset symptom dictionary D.
Specifically, the extraction process of the symptom concept entity can be performed circularly, and in the circulating process, the trained symptom concept entity can be added into a preset symptom dictionary D, and a new template is generated and added into a concept word template T; and extracting the symptom knowledge data source by the supplemented preset symptom dictionary D and the supplemented concept word template T until a certain control turn is reached or the model symptom identification reaches a certain accuracy rate, and stopping circulation.
Further, the step S01 specifically includes:
obtaining at least one candidate template based on the identified symptom concept entities;
performing confusion prediction on the candidate template based on the preset training model, determining a seed template, and determining a root node of the symptom concept prefix tree based on the seed template;
and constructing the symptom concept prefix tree based on the node relation between the root node and the symptom concept entity.
In this embodiment, after the symptom concept entity is extracted, the symptom center concept needs to be further screened out. Before extracting the symptom center concept, a symptom concept prefix tree needs to be constructed, and the root node of the symptom concept prefix tree can be extracted according to a seed template generated in pre-training.
Specifically, based on the identified symptom concept entities, a new template is generated and added to the concept word template T.
In a specific embodiment, assuming that the identified symptom concept entity is "head distending pain," masking N ∈ N consecutive characters, such as N = [1,2], then a list of candidate templates L is obtained: "X distending pain", "head X", "X distending pain", and "head X distending pain" …. And inputting the elements in the L into an Encoder-Decoder model, and enabling the Encoder-Decoder model to automatically fill the content of the X. Note: here the model is allowed to generate padding of length > n.
Where, for sequences starting with X, it is necessary to invert them and then input them into the model. For example, "XX distending pain", is fed into the model after being inverted to "distending XX".
Calculating the confusion of Decoder prediction:
it will be appreciated that the lower the degree of confusion, the more confident the model is about the prediction of the output sequence. In this embodiment, the output sequence with higher confusion (i.e., less confident in model prediction) is selected as the seed template — indicating that the obscured content has higher flexibility, selectivity, and thus the template is more generic. For example, the degree of freedom of "XX distending pain" is high: "head distending pain", "abdomen distending pain", "chest distending pain", "menstrual breast distending pain" and the like.
In this embodiment, the seed template is an output sequence with a higher degree of confusion (i.e., less confident in model prediction); the template is high in confusion, and the masked content has higher flexibility and selectivity, namely the unmasked character has stronger wildcard property. By using the unmasked characters as the root nodes of the symptom concept prefix tree, more symptom concept entities can be covered, and more paths can be formed.
Further, the step S02 specifically includes:
calculating information entropy of each path based on the symptom concept prefix tree;
calculating the weight of each path based on the text word frequency of the symptom concept entity corresponding to each path and the information entropy;
determining the symptom center concept based on the weights of the paths.
In this embodiment, a symptom center concept is a smallest non-resegmentable semantic unit in a set of symptom concepts. For example, "cough with phlegm" can be divided into "cough" and "phlegm with phlegm", wherein "cough" cannot be subdivided. Thus, "cough" is a central concept, while "expectoration" is an extension of "cough".
Specifically, starting from the root node of the symptom concept prefix tree, determining all paths in the prefix tree which can form symptom concept entities, and calculating the information entropy of all the paths:
wherein X ∈ X is the continuation path of all symptom concept entities C.
Specifically, for all prefix tree paths i ∈ Path (R) from the root node, the information entropy E corresponding to the sub-Path can be calculated i ＝E(C i ). The larger the entropy, the more the symptom concept entity C is represented i The richer the suffix succession of (A), then (C) i The higher the probability of becoming a central concept.
Calculating the weight of each path based on the information entropy corresponding to each path of the prefix tree:
W(C i )＝E i ·IDF(C i ) (4)
wherein, IDF (C) i ) Concept of an entity C for symptoms i Inverse text word frequency in the symptom knowledge data source S. The higher the word frequency of the inverse text is, the more common the word is represented and the higher the recognition degree is. The common and present paths can be effectively identified by comparing the weights of all pathsMore successive possible symptom concept entities as symptom-centric concepts. Such as: compared with "cough with phlegm" and "cough with heavy turbid sound", there are more possibilities of succession, such as "cough with hoarse voice" and "cough with frequent action", but "cough with phlegm" has no succession, so "cough" can be regarded as the central concept, and "cough with phlegm" and "cough with heavy turbid sound" can be regarded as the derivative concepts.
It can be understood that, by performing weight calculation on each path of each symptom concept prefix tree, all symptom center concepts can be screened out.
Further, the step S03 specifically includes:
matching the symptom derivative concepts with the symptom center concepts based on the semantic similarity principle to obtain a concept word pair set;
training the concept word pair set based on a multi-label classification model to obtain the symptom attributes of the symptom derivative concepts;
establishing a binding relationship between the symptom derivative concept and the symptom center concept based on the symptom attribute.
In this embodiment, for candidate derived concepts existing in the symptom knowledge data source S and not mined, a symptom concept entity not included in a symptom concept prefix tree in the symptom knowledge data source S is identified by a sequence character classification model Encoder-NER, and a concept word list is obtained; and in the concept word list, find all word lists containing symptom-centric concepts. Such as: "the aggravation of the stomachache caused by cold" is a candidate derivative concept which is identified by a model and is not included, wherein the "stomachache" is a known central concept word; or the central concept can not be directly matched, for example, "the stomach pain is aggravated by cold" can not be directly matched with "the stomach pain", at the moment, the candidate derived concepts are matched with the known symptom central concept by using a semantic similarity matching method, and thus, a concept word pair set of the symptom central concept and the candidate derived concepts is obtained.
Correspondingly, combining the symptom derivative concepts which are already included in the symptom concept prefix tree with the corresponding symptom center concepts, and finally outputting a concept word pair set L = { (c, c) p ) Where c is a central concept word, c p Are candidate derived concept words.
In this embodiment, a multi-label classification model of attributes is trained based on the concept word pair set L. Wherein the attribute tag includes: mild/heavy, high/low frequency, aggravation, remission, seasonal hours, diet, defecation, physical movement, sleep emotion, menstruation, abortion, external stimulation, etc. A binary model is built for each attribute tag, but L is still encoded using a shared Encoder in the feature representation portion. Therefore, the binary classification model only needs to use a simple classifier (such as an SVM), so that the training and deployment of the model are relatively light.
Specifically, the attribute with the multi-label classification model prediction result of 1 is used as the attribute of the corresponding symptom derivation concept; and establishing connection between the attribute-filled symptom-derived concepts and corresponding symptom center concepts so as to obtain a center-derived concept relation. Therefore, a more detailed and comprehensive traditional Chinese medicine syndrome type knowledge map is constructed by taking the symptom center concept as a core and taking the symptom derivative concept as a supplement.
Referring to fig. 4, fig. 4 is a schematic flow chart of a third embodiment of the symptom question-answering method based on the knowledge-graph of the present invention.
Based on the embodiment shown in fig. 3, in this embodiment, the step S03 further includes:
step S031, based on the symptom concept prefix tree and the symptom knowledge data source, extracting candidate derived concepts;
step S032, calculating the semantic similarity between the candidate derived concepts and the symptom center concept based on the semantic similarity principle;
step 033, determining the symptom derivative concept based on a preset similarity threshold and the semantic similarity.
In this embodiment, for the symptom concept entity that has been incorporated into the symptom concept prefix tree, except for the symptom center concept, the rest of the symptom concept entities are the symptom derivative concepts associated with the symptom center concept. However, there will still be a large number of unrecognized symptom concept entities in the symptom knowledge data source S, which need to be further mined.
Specifically, after segmenting the sentences in the symptom knowledge data source S, a candidate derived concept list C' composed of N phrases is extracted by an N-gram model. For example, after the sentence "cough due to wind-cold evil and swollen sore throat" is segmented, the sentence "cough with swollen sore throat due to wind-cold evil" can be obtained, and if 3-gram is taken, a list of candidate derived concepts is available: 'wind-cold due to exopathogen', 'cough due to exopathogen, wind-cold cough', 'cough and throat swelling', 'throat swelling and pain', 'and swelling pain'.
Specifically, after obtaining the candidate derived concept list, it is necessary to associate the candidate derived concept with the symptom concept entity that has been included in the symptom concept prefix tree according to the semantic similarity principle. Since the number of candidate derived concept and symptom concept entities is large, the calculation amount is too large if one-to-one matching correlation is performed. Therefore, in the embodiment, the association is realized by matching the candidate derived concepts with the symptom center concept and then performing fine matching.
In a specific embodiment, the average edit distance D between candidate derived concepts and symptom concept entities is first calculated e Semantic similarity between the candidate derived concepts and the symptom concept entities in the prefix tree is then computed.
In a specific embodiment, based on a trained Encode-Decoder model, the candidate derived concepts and the symptom concept entities in the symptom concept prefix tree are respectively encoded by the Encode. When the symptom concept prefix tree is coded, the symptom center concept and other symptom concept entities are respectively coded, weighted average is taken, and semantic similarity D is obtained s ：
D s ＝cosSim(E c′ ，α·E c +β·E C ) (5)
Wherein, E c′ Coded output representing candidate derived concepts, E c Coded output representing the concept of symptom center, E C Representing the encoded output of other symptom concept entities in the symptom concept prefix tree.
In a specific embodiment, according to D = D e +D s Deriving concepts and symptoms for each candidateAnd the shape concept prefix tree is used for carrying out comprehensive semantic similarity scoring and evaluating the comprehensive semantic similarity according to a preset similarity threshold value.
In a specific embodiment, if the comprehensive semantic similarity reaches a preset similarity threshold, it is considered that the corresponding candidate derived concept can be incorporated into the corresponding symptom concept prefix tree, and then the candidate derived concept is finely matched with other symptom concept entities in the symptom concept prefix tree; and if the comprehensive semantic similarity fails to reach a preset similarity threshold, the comprehensive semantic similarity is judged by an expert.
In a specific embodiment, if the expert determines that the corresponding candidate derived concept can be used as the symptom concept entity, the candidate derived concept is added to the prefix tree of the existing symptom concept, and if there is no symptom concept prefix tree that can be included, a symptom concept prefix tree belonging to the candidate derived concept is separately created. Thereby, candidate derived concepts are determined as symptom derived concepts.
Further, based on the embodiment described in fig. 4, the step S03 specifically further includes:
determining a target prefix tree corresponding to the symptom derivative concept based on the semantic similarity, and determining a target symptom concept entity based on the target prefix tree;
and determining a target symptom concept corresponding to the maximum semantic similarity based on the semantic similarity between the symptom derivative concept and the target symptom concept entity, and establishing a connection relation between the target symptom concept and the symptom derivative concept.
In this embodiment, when it is determined that a candidate derived concept can be incorporated into a certain symptom concept prefix tree, the candidate derived concept is subjected to similarity accurate matching with a symptom concept entity in the symptom concept prefix tree according to a semantic similarity matching principle, and the symptom concept entity with the highest semantic similarity is associated with the candidate derived concept, so that the candidate derived concept is incorporated into the symptom concept prefix tree as a symptom derived concept. For example, assuming that "sore throat" has been located to a prefix tree with "sore throat" as a symptom center concept and has the highest semantic similarity with "sore throat", the "sore throat" is synonymous with "sore throat", and the two are associated. Therefore, the extraction of symptom derivative concepts and the establishment of the association relationship between the symptom derivative concepts and the symptom center concepts are realized.
In addition, the embodiment of the invention also provides a symptom question-answering device based on the knowledge graph.
Referring to fig. 5, fig. 5 is a functional block diagram of a first embodiment of the knowledge-graph-based symptom question answering apparatus according to the present invention.
In this embodiment, the symptom question-answering device based on the knowledge graph includes:
the candidate syndrome type determining module 10 is configured to acquire symptom information of a target user, query and match the symptom information based on a symptom knowledge graph and a symptom semantic similarity matching principle, and determine a candidate syndrome type corresponding to the symptom information;
a symptom attribute determining module 20, configured to initiate a symptom query to the target user based on the candidate syndrome type, and determine a symptom attribute corresponding to the symptom information based on feedback data of the target user to the symptom query;
and a confirmation type determining module 30, configured to calculate a diagnosis probability of the candidate syndrome type based on the symptom attribute, and determine a target syndrome type based on the diagnosis probability and a preset threshold.
Further, the symptom question-answering device based on the knowledge graph comprises a symptom knowledge graph building module, wherein the symptom knowledge graph building module comprises:
the concept prefix tree construction submodule is used for extracting a symptom concept entity based on a pre-input symptom knowledge data source and constructing a symptom concept prefix tree based on the node relation of the symptom concept entity;
the symptom center concept determining submodule is used for determining a symptom center concept based on weight calculation of each path in the symptom concept prefix tree;
and the symptom knowledge map obtaining submodule is used for determining symptom derivative concepts based on the symptom center concepts and the semantic similarity principle, establishing a connection relation between the symptom derivative concepts and the symptom center concepts, and obtaining the symptom knowledge map.
Further, the concept prefix tree construction sub-module specifically includes:
a candidate template obtaining unit for obtaining at least one candidate template based on the identified symptom concept entity;
the seed template determining unit is used for predicting the confusion degree of the candidate template based on the preset training model, determining a seed template and determining a root node of the symptom concept prefix tree based on the seed template;
and the prefix tree construction unit is used for constructing the symptom concept prefix tree based on the node relation between the root node and the symptom concept entity.
Further, the symptom center concept determination submodule specifically includes:
the information entropy calculation unit is used for calculating the information entropy of each path based on the symptom concept prefix tree;
a path weight calculation unit, configured to calculate a weight of each path based on the text word frequency and the information entropy of the symptom concept entity corresponding to each path;
a symptom center concept determination unit configured to determine the symptom center concept based on the weight of each path.
Further, the symptom knowledge base acquisition submodule specifically includes:
a candidate derived concept extracting unit configured to extract candidate derived concepts based on the symptom concept prefix tree and the symptom knowledge data source;
a semantic similarity calculation unit, configured to calculate a semantic similarity between the candidate derived concepts and the symptom center concept based on the semantic similarity principle;
and the symptom derived concept determining unit is used for determining the symptom derived concept based on a preset similarity threshold and the semantic similarity.
Further, the symptom knowledge mapping obtaining submodule specifically further comprises:
a target prefix tree determining unit, configured to determine, based on the semantic similarity, a target prefix tree corresponding to the symptom derivative concept, and determine, based on the target prefix tree, a target symptom concept entity;
and the association relationship establishing unit is used for determining a target symptom concept corresponding to the maximum semantic similarity based on the semantic similarity between the symptom derivative concept and the target symptom concept entity, and establishing the association relationship between the target symptom concept and the symptom derivative concept.
Further, the symptom knowledge mapping obtaining submodule specifically further comprises:
a concept word pair set obtaining unit, configured to match the symptom derivative concept with the symptom center concept based on the semantic similarity principle to obtain a concept word pair set;
a symptom attribute obtaining unit, configured to train the concept word pair set based on a multi-label classification model, and obtain a symptom attribute of the symptom-derived concept;
and the association relation establishing subunit is used for establishing an association relation between the symptom derivative concept and the symptom center concept based on the symptom attribute.
Each module in the symptom question-answering device based on the knowledge graph corresponds to each step in the embodiment of the symptom question-answering method based on the knowledge graph, and the functions and the implementation process of the symptom question-answering device based on the knowledge graph are not described in detail here.
In addition, the embodiment of the invention also provides a computer readable storage medium.
The computer-readable storage medium of the present invention stores a knowledgemap-based symptom question-answering program, wherein the knowledgemap-based symptom question-answering program, when executed by a processor, implements the steps of the above-described knowledgemap-based symptom question-answering method.
The method implemented when the symptom question-answering program based on the knowledge graph is executed may refer to each embodiment of the symptom question-answering method based on the knowledge graph of the present invention, and will not be described herein again.
It should be noted that, in this document, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or system. Without further limitation, an element defined by the phrase "comprising a … …" does not exclude the presence of another identical element in a process, method, article, or system that comprises the element.
The above-mentioned serial numbers of the embodiments of the present invention are merely for description and do not represent the merits of the embodiments.
The application is operational with numerous general purpose or special purpose computing system environments or configurations. For example: personal computers, server computers, hand-held or portable devices, tablet-type devices, multiprocessor systems, microprocessor-based systems, set top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, distributed computing environments that include any of the above systems or devices, and the like. The application may be described in the general context of computer-executable instructions, such as program modules, being executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform particular tasks or implement particular abstract data types. The application may also be practiced in distributed computing environments where tasks are performed by remote processing devices that are linked through a communications network. In a distributed computing environment, program modules may be located in both local and remote computer storage media including memory storage devices.
Through the description of the foregoing embodiments, it is clear to those skilled in the art that the method of the foregoing embodiments may be implemented by software plus a necessary general hardware platform, and certainly may also be implemented by hardware, but in many cases, the former is a better implementation. Based on such understanding, the technical solution of the present invention may be embodied in the form of a software product, which is stored in a storage medium (e.g., ROM/RAM, magnetic disk, optical disk) as described above and includes instructions for enabling a terminal device (e.g., a mobile phone, a computer, a server, an air conditioner, or a network device) to execute the method according to the embodiments of the present invention.
The above description is only a preferred embodiment of the present invention, and not intended to limit the scope of the present invention, and all modifications of equivalent structures and equivalent processes, which are made by using the contents of the present specification and the accompanying drawings, or directly or indirectly applied to other related technical fields, are included in the scope of the present invention.
Claims (10)
1. A symptom question-answering method based on a knowledge graph is characterized by comprising the following steps:
acquiring symptom information of a target user, and inquiring and matching the symptom information based on a symptom knowledge graph and a symptom semantic similar matching principle to determine a candidate syndrome type corresponding to the symptom information;
initiating a symptom inquiry to the target user based on the candidate syndrome type, and determining a symptom attribute corresponding to the symptom information based on feedback data of the target user to the symptom inquiry;
and calculating the diagnosis probability of the candidate syndrome type based on the symptom attribute, and determining the target syndrome type based on the diagnosis probability and a preset threshold value.
2. The method of claim 1, wherein before obtaining symptom information of the target user, the method further comprises:
extracting symptom concept entities based on a pre-input symptom knowledge data source, and constructing a symptom concept prefix tree based on the node relation of the symptom concept entities;
determining a symptom center concept based on the weight calculation of each path in the symptom concept prefix tree;
and determining symptom derivative concepts based on the symptom center concept and a semantic similarity principle, and establishing a connection relation between the symptom derivative concepts and the symptom center concept to obtain the symptom knowledge graph.
3. The method for symptom question-answering based on the knowledge graph of claim 2, wherein the constructing a symptom concept prefix tree based on the node relation of the symptom concept entity comprises:
obtaining at least one candidate template based on the identified symptom concept entities;
performing confusion prediction on the candidate templates based on the preset training model, determining a seed template, and determining a root node of the symptom concept prefix tree based on the seed template;
and constructing the symptom concept prefix tree based on the node relation between the root node and the symptom concept entity.
4. The method of claim 2, wherein determining a symptom center concept based on a weight calculation for each path in the symptom concept prefix tree comprises:
calculating the information entropy of each path based on the symptom concept prefix tree;
calculating the weight of each path based on the text word frequency of the symptom concept entity corresponding to each path and the information entropy;
determining the symptom center concept based on the weights of the paths.
5. The method for symptom questioning and answering based on knowledge graph according to claim 2, wherein said determining symptom derivative concepts based on said symptom center concept and semantic similarity principle comprises:
extracting candidate derived concepts based on the symptom concept prefix tree and the symptom knowledge data source;
calculating semantic similarity between the candidate derived concepts and the symptom center concept based on the semantic similarity principle;
determining the symptom derivative concept based on a preset similarity threshold and the semantic similarity.
6. The knowledgeable map-based symptom questioning and answering method according to claim 5, wherein said establishing of the association relationship between said symptom-derived concepts and said symptom-centered concepts comprises:
determining a target prefix tree corresponding to the symptom derived concept based on the semantic similarity, and determining a target symptom concept entity based on the target prefix tree;
and determining a target symptom concept corresponding to the maximum semantic similarity based on the semantic similarity between the symptom derivative concept and the target symptom concept entity, and establishing a connection relation between the target symptom concept and the symptom derivative concept.
7. The knowledgegraph-based symptom question-answering method according to any one of claims 1-6, wherein the associating of the symptom derivative concepts with the symptom center concepts further comprises:
matching the symptom derivative concepts with the symptom center concepts based on the semantic similarity principle to obtain a concept word pair set;
training the concept word pair set based on a multi-label classification model to obtain symptom attributes of the symptom derivative concepts;
establishing a binding relationship between the symptom derivative concept and the symptom center concept based on the symptom attribute.
8. A knowledge-graph-based symptomatic question-answering apparatus, comprising:
the candidate syndrome type determining module is used for acquiring symptom information of a target user, inquiring and matching the symptom information based on a symptom knowledge graph and a symptom semantic similar matching principle, and determining a candidate syndrome type corresponding to the symptom information;
the symptom attribute determining module is used for initiating a symptom inquiry to the target user based on the candidate syndrome type and determining a symptom attribute corresponding to the symptom information based on feedback data of the target user to the symptom inquiry;
and the diagnosis confirming type determining module is used for calculating the diagnosis confirming probability of the candidate syndrome type based on the symptom attribute and determining the target syndrome type based on the diagnosis confirming probability and a preset threshold value.
9. A knowledgemap-based symptom question-answering apparatus comprising a processor, a memory, and a knowledgemap-based symptom question-answering program stored on the memory and executable by the processor, wherein the knowledgemap-based symptom question-answering program, when executed by the processor, implements the steps of the knowledgemap-based symptom question-answering method of any one of claims 1 to 7.
10. A computer-readable storage medium, characterized in that the computer-readable storage medium has stored thereon a knowledgemap-based symptom question-answering program, wherein the knowledgemap-based symptom question-answering program, when executed by a processor, implements the steps of the knowledgemap-based symptom question-answering method according to any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210868668.2A CN115186068A (en) | 2022-07-22 | 2022-07-22 | Symptom question-answering method, device, equipment and storage medium based on knowledge graph |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210868668.2A CN115186068A (en) | 2022-07-22 | 2022-07-22 | Symptom question-answering method, device, equipment and storage medium based on knowledge graph |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115186068A true CN115186068A (en) | 2022-10-14 |
Family
ID=83520426
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210868668.2A Pending CN115186068A (en) | 2022-07-22 | 2022-07-22 | Symptom question-answering method, device, equipment and storage medium based on knowledge graph |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115186068A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116340544A (en) * | 2023-04-03 | 2023-06-27 | 浙江大学 | Visual analysis method and system for ancient Chinese medicine books based on knowledge graph |
| CN117476163A (en) * | 2023-12-27 | 2024-01-30 | 万里云医疗信息科技（北京）有限公司 | Method, apparatus and storage medium for determining disease conclusion |
-
2022
- 2022-07-22 CN CN202210868668.2A patent/CN115186068A/en active Pending
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116340544A (en) * | 2023-04-03 | 2023-06-27 | 浙江大学 | Visual analysis method and system for ancient Chinese medicine books based on knowledge graph |
| CN116340544B (en) * | 2023-04-03 | 2024-02-23 | 浙江大学 | Visual analysis method and system for ancient Chinese medicine books based on knowledge graph |
| CN117476163A (en) * | 2023-12-27 | 2024-01-30 | 万里云医疗信息科技（北京）有限公司 | Method, apparatus and storage medium for determining disease conclusion |
| CN117476163B (en) * | 2023-12-27 | 2024-03-08 | 万里云医疗信息科技（北京）有限公司 | Method, apparatus and storage medium for determining disease conclusion |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110069631B (en) | Text processing method and device and related equipment | |
| CN115186068A (en) | Symptom question-answering method, device, equipment and storage medium based on knowledge graph | |
| CN111709233B (en) | Intelligent diagnosis guiding method and system based on multi-attention convolutional neural network | |
| WO2020252919A1 (en) | Resume identification method and apparatus, and computer device and storage medium | |
| CN110737763A (en) | Chinese intelligent question-answering system and method integrating knowledge map and deep learning | |
| CN109871538A (en) | A kind of Chinese electronic health record name entity recognition method | |
| CN112035637A (en) | Medical field intention recognition method, device, equipment and storage medium | |
| CN112131393A (en) | Construction method of medical knowledge map question-answering system based on BERT and similarity algorithm | |
| US10943075B2 (en) | Translating a first language phrase into a second language phrase | |
| WO2021139229A1 (en) | Text rhetorical sentence generation method, apparatus and device, and readable storage medium | |
| CN110032728A (en) | The standardized conversion method of disease name and device | |
| CN109599187A (en) | A kind of online interrogation point examines method, server, terminal, equipment and medium | |
| WO2023029502A1 (en) | Method and apparatus for constructing user portrait on the basis of inquiry session, device, and medium | |
| US20220121824A1 (en) | Method for determining text similarity, method for obtaining semantic answer text, and question answering method | |
| WO2023241410A1 (en) | Data processing method and apparatus, and device and computer medium | |
| CN114897060B (en) | Training method and device for sample classification model, and sample classification method and device | |
| CN115438674B (en) | Entity data processing method, entity linking method, entity data processing device, entity linking device and computer equipment | |
| CN111444715A (en) | Entity relationship identification method and device, computer equipment and storage medium | |
| CN112256827A (en) | Sign language translation method and device, computer equipment and storage medium | |
| CN113821587A (en) | Text relevance determination method, model training method, device and storage medium | |
| US11544320B2 (en) | Image processing utilizing an entigen construct | |
| CN114611529B (en) | Intention recognition method and device, electronic equipment and storage medium | |
| CN111190998A (en) | Question-answering robot system based on hybrid model and question-answering robot | |
| CN115759052A (en) | Text error correction method and device, electronic equipment and storage medium | |
| CN113190659B (en) | Language and language machine reading understanding method based on multi-task joint training |
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 |