CN113221578A - Disease entity retrieval method, device, equipment and medium - Google Patents
Disease entity retrieval method, device, equipment and medium Download PDFInfo
- Publication number
- CN113221578A CN113221578A CN202110485329.1A CN202110485329A CN113221578A CN 113221578 A CN113221578 A CN 113221578A CN 202110485329 A CN202110485329 A CN 202110485329A CN 113221578 A CN113221578 A CN 113221578A
- Authority
- CN
- China
- Prior art keywords
- entity
- synonym
- representation
- layer
- disease
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/30—Semantic 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
- G06F16/36—Creation of semantic tools, e.g. ontology or thesauri
- G06F16/367—Ontology
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/20—Natural language analysis
- G06F40/237—Lexical tools
- G06F40/247—Thesauruses; Synonyms
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
- G16H50/70—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for mining of medical data, e.g. analysing previous cases of other patients
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- General Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Databases & Information Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Public Health (AREA)
- Medical Informatics (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Biomedical Technology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Animal Behavior & Ethology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Pathology (AREA)
- Evolutionary Biology (AREA)
- Epidemiology (AREA)
- Primary Health Care (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Machine Translation (AREA)
Abstract
The invention relates to the field of artificial intelligence, and provides a disease entity retrieval method, a device, equipment and a medium, which can simultaneously combine semantic information and map knowledge information to better realize a synonym matching task, adopt character-level features to replace traditional word-level features, effectively solve the OOV problem, solve the problem that a synonym expressed by spoken language and a medical entity expressed by formal professional language belong to different semantic spaces, adaptively integrate knowledge representation into corresponding semantic representation through a self-defined fusion mechanism, introduce more external knowledge through integrating the knowledge representation of a map, establish hidden relation between an entity and the synonym, effectively relieve the long-tail problem, further realize retrieval of a disease entity through a self-defined trained retrieval model, and effectively improve the accuracy of a retrieval result. In addition, the invention also relates to a block chain technology, and the retrieval model can be stored in the block chain node.
Description
Technical Field
The invention relates to the technical field of artificial intelligence, in particular to a disease entity retrieval method, a disease entity retrieval device, disease entity retrieval equipment and disease entity retrieval media.
Background
The entity synonym discovery task is to find synonyms for entity nodes in the knowledge graph, belongs to a part of knowledge graph construction, and plays an important role in graph-related downstream tasks, such as: entity link, information retrieval, knowledge map question and answer and the like in the disease entity synonym retrieval task.
The currently widely adopted synonym search scheme for disease entities mainly comprises the following steps: jaccard-based methods, Embedding-based methods, etc., i.e., entities and synonyms are typically linked by syntactic string matching or Embedding matching.
However, the existing method has the following problems:
1) due to the lack of medical entities in the general corpus, medical vocabularies usually belong to non-thesaurus (OOV), and synonyms expressed by spoken language and medical entities expressed by formal professional language belong to different semantic spaces and are difficult to match.
2) The relationship between entities and synonyms is obscure and not directly accessible.
3) The long tail effect causes some medical entities to rarely appear in the training dataset.
Disclosure of Invention
In view of the above, there is a need to provide a disease entity retrieval method, apparatus, device and medium, which can implement retrieval of disease entities through a custom-trained retrieval model, and effectively improve the accuracy of the retrieval result.
A disease entity retrieval method, the disease entity retrieval method comprising:
acquiring each entity in a pre-constructed disease knowledge graph and acquiring synonyms of each entity;
calling an initial network, wherein the structure of the initial network comprises a coding layer, a graph embedding layer, a fusion layer and a matching layer;
inputting each entity and synonyms of each entity into the coding layer, and converting each entity and synonyms of each entity by using a feature conversion layer of the coding layer to obtain a character vector of each entity and a character vector of each synonym;
constructing a subgraph of each entity in the disease knowledge graph, and inputting the subgraph of each entity into the graph embedding layer to obtain a knowledge vector of each entity;
based on an alignment mechanism of shared weight, mapping the character vector of each entity and the character vector of each synonym to the same semantic space by using a full connection layer of the coding layer to obtain a semantic representation of each entity and a semantic representation of each synonym;
carrying out semantic space mapping on the knowledge vector of each entity by using the full-connection layer of the coding layer to obtain the knowledge representation of each entity;
fusing the semantic representation of each entity and the knowledge representation of each entity by using the fusion layer to obtain the entity representation of each entity;
inputting the semantic representation of each synonym and the entity representation of each entity into the matching layer for matching, and acquiring the current value of the loss function after matching;
carrying out back transmission training on the initial network according to the current value until the value of the loss function is not reduced any more, and stopping training to obtain a retrieval model;
and acquiring the synonym to be retrieved, inputting the synonym to be retrieved into the retrieval model, and acquiring the output of the retrieval model as the disease entity of the synonym to be retrieved.
According to the preferred embodiment of the present invention, the obtaining synonyms of each entity includes one or more of the following ways:
acquiring a database associated with the disease knowledge graph, and acquiring synonyms of each entity from the database; and/or
And crawling the designated page by adopting a web crawler technology to obtain the synonym of each entity.
According to a preferred embodiment of the present invention, said constructing a sub-graph of each entity in said disease knowledge-graph comprises:
performing multi-hop processing on each entity in a preset dimension to obtain a connection diagram of each entity in each dimension;
and combining the connection graphs obtained by each entity under each dimension to obtain a subgraph of each entity.
According to the preferred embodiment of the present invention, the fusing the semantic representation of each entity and the knowledge representation of each entity by using the fusion layer to obtain the entity representation of each entity includes:
constructing a preset function according to the semantic representation of each entity and the corresponding knowledge representation of each entity;
constructing a convolution function according to the knowledge representation of each entity and the corresponding preset function;
the sum of the semantic representation of each entity and the corresponding convolution function is computed as the entity representation of each entity.
According to the preferred embodiment of the present invention, the constructing the preset function according to the semantic representation of each entity and the corresponding knowledge representation of each entity includes:
acquiring configured weight;
constructing a weight function according to the weight, the semantic representation of each entity and the corresponding knowledge representation of each entity;
and processing the weight function by utilizing a Softmax function to obtain the preset function.
According to a preferred embodiment of the invention, the method further comprises:
calculating the similarity between the synonym to be retrieved and each entity contained in the retrieval model through a noise comparison estimation algorithm;
sequencing all entities contained in the retrieval model according to the sequence of the similarity from high to low;
and outputting the first-ranked entities, and determining the output entities as the disease entities of the synonyms to be searched.
According to a preferred embodiment of the invention, the method further comprises:
when detecting that a new entity is added into the disease knowledge graph, acquiring synonyms of the new entity;
and performing supplementary training on the retrieval model according to the new entity and the synonym of the new entity.
A disease entity retrieval apparatus, the disease entity retrieval apparatus comprising:
the acquisition unit is used for acquiring each entity in a pre-constructed disease knowledge graph and acquiring synonyms of each entity;
the device comprises a calling unit, a matching unit and a processing unit, wherein the calling unit is used for calling an initial network, and the structure of the initial network comprises a coding layer, a graph embedding layer, a fusion layer and a matching layer;
the input unit is used for inputting each entity and synonyms of each entity into the coding layer, and converting each entity and synonyms of each entity by using the feature conversion layer of the coding layer to obtain a character vector of each entity and a character vector of each synonym;
the construction unit is used for constructing a sub-graph of each entity in the disease knowledge graph and inputting the sub-graph of each entity into the graph embedding layer to obtain a knowledge vector of each entity;
the mapping unit is used for mapping the character vector of each entity and the character vector of each synonym to the same semantic space by utilizing the full-connection layer of the coding layer based on the alignment mechanism of the shared weight to obtain the semantic representation of each entity and the semantic representation of each synonym;
the mapping unit is further configured to perform semantic space mapping on the knowledge vector of each entity by using a full-connection layer of the coding layer to obtain a knowledge representation of each entity;
the fusion unit is used for fusing the semantic representation of each entity and the knowledge representation of each entity by using the fusion layer to obtain the entity representation of each entity;
the input unit is further used for inputting the semantic representation of each synonym and the entity representation of each entity into the matching layer for matching, and acquiring the current value of the loss function after matching;
the training unit is used for carrying out return training on the initial network according to the current value until the value of the loss function is not reduced any more, and stopping training to obtain a retrieval model;
the input unit is further configured to acquire a synonym to be retrieved, input the synonym to be retrieved to the retrieval model, and acquire an output of the retrieval model as a disease entity of the synonym to be retrieved.
An electronic device, the electronic device comprising:
a memory storing at least one instruction; and
a processor executing instructions stored in the memory to implement the disease entity retrieval method.
A computer-readable storage medium having at least one instruction stored therein, the at least one instruction being executable by a processor in an electronic device to implement the disease entity retrieval method.
It can be seen from the above technical solutions that the present invention can obtain each entity in a pre-constructed disease knowledge graph, obtain synonyms of each entity, and invoke an initial network, wherein the structure of the initial network includes a coding layer, a graph embedding layer, a fusion layer, and a matching layer, and combines semantic information and graph knowledge information to better implement a synonym matching task, and inputs each entity and synonyms of each entity into the coding layer, and converts each entity and synonyms of each entity by using a feature conversion layer of the coding layer to obtain a character vector of each entity and a character vector of each synonym, and replaces the conventional term-level features with the character-level features, so as to effectively solve the OOV problem, construct sub-graphs of each entity in the disease knowledge graph, and input sub-graphs of each entity into the graph embedding layer, obtaining a knowledge vector of each entity, mapping the character vector of each entity and the character vector of each synonym to the same semantic space by using a full connection layer of a coding layer based on an alignment mechanism of shared weight, obtaining a semantic representation of each entity and a semantic representation of each synonym so as to solve the problem that the synonym expressed by spoken language and the medical entity expressed by formal professional language belong to different semantic spaces, performing semantic space mapping on the knowledge vector of each entity by using the full connection layer of the coding layer to obtain a knowledge representation of each entity, fusing the semantic representation of each entity and the knowledge representation of each entity by using a fusion layer to obtain an entity representation of each entity, adaptively integrating the knowledge representations into the corresponding semantic representations by a self-defined fusion mechanism, and integrating the knowledge representations by using a map, introducing more external knowledge, establishing a hidden relation between entities and synonyms, effectively relieving the long tail problem, inputting the semantic representation of each synonym and the entity representation of each entity into the matching layer for matching, acquiring the current value of a loss function after matching, performing return training on the initial network according to the current value until the value of the loss function is not reduced any more, stopping training, obtaining a retrieval model, continuously improving the training effect of the model through the auxiliary training of the loss function, acquiring the synonym to be retrieved, inputting the synonym to be retrieved into the retrieval model, acquiring the output of the retrieval model as the disease entity of the synonym to be retrieved, further realizing the retrieval of the disease entity through the self-defined trained retrieval model, and effectively improving the accuracy of the retrieval result.
Drawings
FIG. 1 is a flow chart of a preferred embodiment of the disease entity search method of the present invention.
FIG. 2 is a functional block diagram of a disease entity retrieving device according to a preferred embodiment of the present invention.
FIG. 3 is a schematic structural diagram of an electronic device for implementing a disease entity retrieving method according to a preferred embodiment of the invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention will be described in detail with reference to the accompanying drawings and specific embodiments.
FIG. 1 is a flow chart of the disease entity search method according to the preferred embodiment of the present invention. The order of the steps in the flow chart may be changed and some steps may be omitted according to different needs.
The disease entity retrieval method is applied to one or more electronic devices, which are devices capable of automatically performing numerical calculation and/or information processing according to preset or stored instructions, and the hardware thereof includes, but is not limited to, a microprocessor, an Application Specific Integrated Circuit (ASIC), a Programmable Gate Array (FPGA), a Digital Signal Processor (DSP), an embedded device, and the like.
The electronic device may be any electronic product capable of performing human-computer interaction with a user, for example, a Personal computer, a tablet computer, a smart phone, a Personal Digital Assistant (PDA), a game machine, an interactive Internet Protocol Television (IPTV), an intelligent wearable device, and the like.
The electronic device may also include a network device and/or a user device. The network device includes, but is not limited to, a single network server, a server group consisting of a plurality of network servers, or a Cloud Computing (Cloud Computing) based Cloud consisting of a large number of hosts or network servers.
The Network where the electronic device is located includes, but is not limited to, the internet, a wide area Network, a metropolitan area Network, a local area Network, a Virtual Private Network (VPN), and the like.
S10, obtaining each entity in the pre-constructed disease knowledge map and obtaining synonyms of each entity.
In at least one embodiment of the invention, the disease knowledge graph may be custom configured, and the invention is not limited.
In at least one embodiment of the present invention, the obtaining synonyms for each entity includes, but is not limited to, a combination of one or more of the following ways:
acquiring a database associated with the disease knowledge graph, and acquiring synonyms of each entity from the database; and/or
And crawling the designated page by adopting a web crawler technology to obtain the synonym of each entity.
Through the embodiment, the synonym of each entity can be fully acquired, so that the training effect of the model is improved, and the accuracy of the retrieval result is higher.
And S11, calling an initial network, wherein the structure of the initial network comprises an encoding layer, a graph embedding layer, a fusion layer and a matching layer.
In this embodiment, the initial network is a self-defined network structure, and integrates the coding layer, the graph embedding layer, the fusion layer, and the matching layer, so that semantic information and graph knowledge information can be simultaneously combined, and a synonym matching task can be better achieved.
And S12, inputting each entity and the synonym of each entity into the coding layer, and converting each entity and the synonym of each entity by using the feature conversion layer of the coding layer to obtain the character vector of each entity and the character vector of each synonym.
In this embodiment, the feature conversion layer may be obtained by a pre-trained method, that is, by training a corpus with tools such as word2vec or GloVe, which are not described herein again.
It should be noted that, in the prior art, word-level features are usually adopted for matching, but due to the fact that medical entities in a general corpus are relatively lacking, medical vocabularies usually belong to non-thesaurus (OOV), and synonyms expressed by spoken language and medical entities expressed by formal professional language belong to different semantic spaces and are difficult to match.
Therefore, the present embodiment can effectively solve the OOV problem by replacing the conventional word-level features with the character-level features.
S13, constructing a subgraph of each entity in the disease knowledge graph, and inputting the subgraph of each entity into the graph embedding layer to obtain a knowledge vector of each entity.
In at least one embodiment of the present invention, the constructing a sub-graph of each entity in the disease knowledge graph comprises:
performing multi-hop processing on each entity in a preset dimension to obtain a connection diagram of each entity in each dimension;
and combining the connection graphs obtained by each entity under each dimension to obtain a subgraph of each entity.
The multi-hop processing refers to executing the extraction and construction process of the feature subgraph corresponding to the preset dimensionality.
For example: when the preset dimension is 2 dimensions, the integers less than or equal to 2 include 1 and 2, namely, one hop and two hops are performed. The one-hop represents the acquisition of the nodes adjacent to each entity, namely: traversing from each entity node, acquiring a first node traversed in each connecting line direction, extracting each entity node and the traversed node, and acquiring a connection graph of a node adjacent to each entity and each entity, wherein similarly, the two hops represent acquiring nodes separated from each entity by one bit, that is: and traversing from each entity node, acquiring a second node traversed in each connecting line direction, separating the traversed node from the corresponding entity node by one node in each connecting line direction, and further extracting each entity node and the traversed node to obtain a node separated from each entity by one bit and a connection graph between each entity. And determining all the obtained connection graphs as corresponding subgraphs of each entity.
Through the implementation mode, the subgraph keeps the upper and lower relations among the entity nodes, so that the obtained knowledge vector of each entity also keeps the upper and lower relations among the entity nodes, and meanwhile, the robust modeling is carried out on various relations among the entities, so that accurate knowledge representation is provided.
And S14, based on the alignment mechanism of the shared weight, mapping the character vector of each entity and the character vector of each synonym to the same semantic space by using the full connection layer of the coding layer, and obtaining the semantic representation of each entity and the semantic representation of each synonym.
It should be noted that the fully-connected layer is a classifier in a convolutional neural network, and can be used to map the learned features to a sample label space.
In this embodiment, since the alignment mechanism of sharing weight is adopted in the full connection layer, the character vector of each entity and the character vector of each synonym can be mapped to the same semantic space, so as to solve the problem that the synonym expressed by spoken language and the medical entity expressed by formal professional language belong to different semantic spaces.
And S15, carrying out semantic space mapping on the knowledge vector of each entity by using the full connection layer of the coding layer to obtain the knowledge representation of each entity.
Similarly, the knowledge vector for each entity is semantically space mapped based on the unique properties of the fully-connected layer.
The semantic space mapping is executed through the full connection layer, the local features are reassembled into complete features, the influence of the position area of the features on classification can be effectively reduced, and the processing accuracy is improved.
And S16, fusing the semantic representation of each entity and the knowledge representation of each entity by using the fusion layer to obtain the entity representation of each entity.
In at least one embodiment of the present invention, the fusing the semantic representation of each entity and the knowledge representation of each entity by using the fusion layer to obtain the entity representation of each entity includes:
constructing a preset function according to the semantic representation of each entity and the corresponding knowledge representation of each entity;
constructing a convolution function according to the knowledge representation of each entity and the corresponding preset function;
the sum of the semantic representation of each entity and the corresponding convolution function is computed as the entity representation of each entity.
The formula for the entity characterization for each entity is as follows:
wherein S represents an entity representation of each entity, a represents a semantic representation of the entity corresponding to S, and b represents a knowledge representation of the entity corresponding to S.
Further, the constructing a preset function according to the semantic representation of each entity and the corresponding knowledge representation of each entity includes:
acquiring configured weight;
constructing a weight function according to the weight, the semantic representation of each entity and the corresponding knowledge representation of each entity;
and processing the weight function by utilizing a Softmax function to obtain the preset function.
Wherein the preset function g (a, b) is constructed;
where W represents a weight.
It should be noted that the weight is an optimal weight obtained after the model training is finished, and belongs to a configuration parameter of the model.
Through the implementation mode, the knowledge representation is adaptively integrated into the corresponding semantic representation through a self-defined fusion mechanism, more external knowledge is introduced through the knowledge representation of the integrated map, the hidden relation between the entity and the synonym is established, and the long tail problem is effectively relieved.
And S17, inputting the semantic representation of each synonym and the entity representation of each entity into the matching layer for matching, and acquiring the current value of the loss function after matching.
In this embodiment, in the matching layer, the semantic representation of each synonym and the similarity between the entity representations of each entity are calculated to match the two.
Further, since the accuracy and precision of the whole model are ensured, the model needs to be continuously adjusted through a loss function to ensure that the accuracy of the model is continuously improved in the training process.
And S18, performing return training on the initial network according to the current value until the value of the loss function is not reduced any more, and stopping training to obtain a retrieval model.
Specifically, the current value of the loss function is continuously returned, so that the model is continuously updated iteratively according to the loss function, when the value of the loss function is not reduced, the loss function is shown to be converged, and at this time, the training is stopped, and the retrieval model can be obtained.
It should be noted that the retrieval model is a model with a brand-new structure constructed by combining semantic information and atlas knowledge information, and the training effect of the model is continuously improved by the auxiliary training of the loss function in the training process of the model.
The type of the loss function may be configured by a user, which is not limited in the present invention.
S19, obtaining the synonym to be retrieved, inputting the synonym to be retrieved into the retrieval model, and obtaining the output of the retrieval model as the disease entity of the synonym to be retrieved.
By the embodiment, the retrieval of the disease entity can be realized through the custom-trained retrieval model, and the accuracy of the retrieval result is effectively improved.
In at least one embodiment of the invention, the method further comprises:
calculating the similarity between the synonym to be searched and each entity contained in the searching model by a Noise Contrast Estimation (NCE) algorithm;
sequencing all entities contained in the retrieval model according to the sequence of the similarity from high to low;
and outputting the first-ranked entities, and determining the output entities as the disease entities of the synonyms to be searched.
For example: when the synonym to be searched is "skin laxity", the "skin laxity" can be identified as the synonym of "ehler syndrome" through the search model, that is, the search model outputs "ehler syndrome" as the disease entity of "skin laxity". Specifically, the map entity and candidate entity corresponding to "skin laxity" include "ehler syndrome", "scald", "dermatitis", and the like. Through knowledge map information, the three entities are all found to belong to skin diseases, and although the 'skin laxity' is completely dissimilar to the 'Eiles syndrome' in word, the 'skin laxity' has a similar meaning to the symptom 'extremely elastic skin', and has no obvious similarity to the attributes of the other two entities. Thus, the search model predicts that "skin laxity" belongs to the synonym of "Ehlers syndrome".
It should be noted that, in order to solve the OOV problem, the present embodiment uses the character-level features instead of the conventional word-level features, but at the same time, it also brings a high calculation amount. Therefore, in order to avoid the operation burden on the system due to a high calculation amount, the noise contrast estimation algorithm is further adopted to calculate the similarity between the synonym and the entity, and the noise contrast estimation algorithm is characterized in that the problem of complex calculation of the neural network can be solved, so that an obvious neutralization effect is exerted on the calculation amount, and the influence on the system is reduced.
In at least one embodiment of the invention, the method further comprises:
when detecting that a new entity is added into the disease knowledge graph, acquiring synonyms of the new entity;
and performing supplementary training on the retrieval model according to the new entity and the synonym of the new entity.
Specifically, the supplementary training is equivalent to adding a new training sample on the basis of the original model, and the training mode is similar to that of the search model and is not repeated here.
It should be noted that, in order to further ensure the security of the data, the retrieval model may be deployed in the blockchain to avoid malicious tampering of the data.
It can be seen from the above technical solutions that the present invention can obtain each entity in a pre-constructed disease knowledge graph, obtain synonyms of each entity, and invoke an initial network, wherein the structure of the initial network includes a coding layer, a graph embedding layer, a fusion layer, and a matching layer, and combines semantic information and graph knowledge information to better implement a synonym matching task, and inputs each entity and synonyms of each entity into the coding layer, and converts each entity and synonyms of each entity by using a feature conversion layer of the coding layer to obtain a character vector of each entity and a character vector of each synonym, and replaces the conventional term-level features with the character-level features, so as to effectively solve the OOV problem, construct sub-graphs of each entity in the disease knowledge graph, and input sub-graphs of each entity into the graph embedding layer, obtaining a knowledge vector of each entity, mapping the character vector of each entity and the character vector of each synonym to the same semantic space by using a full connection layer of a coding layer based on an alignment mechanism of shared weight, obtaining a semantic representation of each entity and a semantic representation of each synonym so as to solve the problem that the synonym expressed by spoken language and the medical entity expressed by formal professional language belong to different semantic spaces, performing semantic space mapping on the knowledge vector of each entity by using the full connection layer of the coding layer to obtain a knowledge representation of each entity, fusing the semantic representation of each entity and the knowledge representation of each entity by using a fusion layer to obtain an entity representation of each entity, adaptively integrating the knowledge representations into the corresponding semantic representations by a self-defined fusion mechanism, and integrating the knowledge representations by using a map, introducing more external knowledge, establishing a hidden relation between entities and synonyms, effectively relieving the long tail problem, inputting the semantic representation of each synonym and the entity representation of each entity into the matching layer for matching, acquiring the current value of a loss function after matching, performing return training on the initial network according to the current value until the value of the loss function is not reduced any more, stopping training, obtaining a retrieval model, continuously improving the training effect of the model through the auxiliary training of the loss function, acquiring the synonym to be retrieved, inputting the synonym to be retrieved into the retrieval model, acquiring the output of the retrieval model as the disease entity of the synonym to be retrieved, further realizing the retrieval of the disease entity through the self-defined trained retrieval model, and effectively improving the accuracy of the retrieval result.
FIG. 2 is a functional block diagram of the disease entity retrieving device according to the preferred embodiment of the present invention. The disease entity retrieving device 11 includes an obtaining unit 110, a calling unit 111, an input unit 112, a constructing unit 113, a mapping unit 114, a fusing unit 115, and a training unit 116. The module/unit referred to in the present invention refers to a series of computer program segments that can be executed by the processor 13 and that can perform a fixed function, and that are stored in the memory 12. In the present embodiment, the functions of the modules/units will be described in detail in the following embodiments.
The obtaining unit 110 obtains each entity in the pre-constructed disease knowledge graph and obtains synonyms of each entity.
In at least one embodiment of the invention, the disease knowledge graph may be custom configured, and the invention is not limited.
In at least one embodiment of the present invention, the obtaining unit 110 obtains synonyms of each entity, including, but not limited to, one or more of the following combinations:
acquiring a database associated with the disease knowledge graph, and acquiring synonyms of each entity from the database; and/or
And crawling the designated page by adopting a web crawler technology to obtain the synonym of each entity.
Through the embodiment, the synonym of each entity can be fully acquired, so that the training effect of the model is improved, and the accuracy of the retrieval result is higher.
The calling unit 111 calls an initial network, wherein the structure of the initial network comprises an encoding layer, a graph embedding layer, a fusion layer and a matching layer.
In this embodiment, the initial network is a self-defined network structure, and integrates the coding layer, the graph embedding layer, the fusion layer, and the matching layer, so that semantic information and graph knowledge information can be simultaneously combined, and a synonym matching task can be better achieved.
The input unit 112 inputs each entity and the synonym of each entity into the coding layer, and converts each entity and the synonym of each entity by using the feature conversion layer of the coding layer to obtain a character vector of each entity and a character vector of each synonym.
In this embodiment, the feature conversion layer may be obtained by a pre-trained method, that is, by training a corpus with tools such as word2vec or GloVe, which are not described herein again.
It should be noted that, in the prior art, word-level features are usually adopted for matching, but due to the fact that medical entities in a general corpus are relatively lacking, medical vocabularies usually belong to non-thesaurus (OOV), and synonyms expressed by spoken language and medical entities expressed by formal professional language belong to different semantic spaces and are difficult to match.
Therefore, the present embodiment can effectively solve the OOV problem by replacing the conventional word-level features with the character-level features.
The construction unit 113 constructs a subgraph of each entity in the disease knowledge graph, and inputs the subgraph of each entity to the graph embedding layer to obtain a knowledge vector of each entity.
In at least one embodiment of the present invention, the constructing unit 113 constructs a sub-graph of each entity in the disease knowledge graph including:
performing multi-hop processing on each entity in a preset dimension to obtain a connection diagram of each entity in each dimension;
and combining the connection graphs obtained by each entity under each dimension to obtain a subgraph of each entity.
The multi-hop processing refers to executing the extraction and construction process of the feature subgraph corresponding to the preset dimensionality.
For example: when the preset dimension is 2 dimensions, the integers less than or equal to 2 include 1 and 2, namely, one hop and two hops are performed. The one-hop represents the acquisition of the nodes adjacent to each entity, namely: traversing from each entity node, acquiring a first node traversed in each connecting line direction, extracting each entity node and the traversed node, and acquiring a connection graph of a node adjacent to each entity and each entity, wherein similarly, the two hops represent acquiring nodes separated from each entity by one bit, that is: and traversing from each entity node, acquiring a second node traversed in each connecting line direction, separating the traversed node from the corresponding entity node by one node in each connecting line direction, and further extracting each entity node and the traversed node to obtain a node separated from each entity by one bit and a connection graph between each entity. And determining all the obtained connection graphs as corresponding subgraphs of each entity.
Through the implementation mode, the subgraph keeps the upper and lower relations among the entity nodes, so that the obtained knowledge vector of each entity also keeps the upper and lower relations among the entity nodes, and meanwhile, the robust modeling is carried out on various relations among the entities, so that accurate knowledge representation is provided.
The mapping unit 114 maps the character vector of each entity and the character vector of each synonym to the same semantic space by using the fully-connected layer of the coding layer based on the alignment mechanism of the shared weight, so as to obtain the semantic representation of each entity and the semantic representation of each synonym.
It should be noted that the fully-connected layer is a classifier in a convolutional neural network, and can be used to map the learned features to a sample label space.
In this embodiment, since the alignment mechanism of sharing weight is adopted in the full connection layer, the character vector of each entity and the character vector of each synonym can be mapped to the same semantic space, so as to solve the problem that the synonym expressed by spoken language and the medical entity expressed by formal professional language belong to different semantic spaces.
The mapping unit 114 performs semantic space mapping on the knowledge vector of each entity by using the fully-connected layer of the coding layer to obtain a knowledge representation of each entity.
Similarly, the knowledge vector for each entity is semantically space mapped based on the unique properties of the fully-connected layer.
The semantic space mapping is executed through the full connection layer, the local features are reassembled into complete features, the influence of the position area of the features on classification can be effectively reduced, and the processing accuracy is improved.
The fusion unit 115 fuses the semantic representation of each entity and the knowledge representation of each entity by using the fusion layer to obtain the entity representation of each entity.
In at least one embodiment of the present invention, the fusing the semantic representation of each entity and the knowledge representation of each entity by using the fusion layer to obtain the entity representation of each entity includes:
constructing a preset function according to the semantic representation of each entity and the corresponding knowledge representation of each entity;
constructing a convolution function according to the knowledge representation of each entity and the corresponding preset function;
the sum of the semantic representation of each entity and the corresponding convolution function is computed as the entity representation of each entity.
The formula for the entity characterization for each entity is as follows:
wherein S represents an entity representation of each entity, a represents a semantic representation of the entity corresponding to S, and b represents a knowledge representation of the entity corresponding to S.
Further, the constructing a preset function according to the semantic representation of each entity and the corresponding knowledge representation of each entity includes:
acquiring configured weight;
constructing a weight function according to the weight, the semantic representation of each entity and the corresponding knowledge representation of each entity;
and processing the weight function by utilizing a Softmax function to obtain the preset function.
Wherein the preset function g (a, b) is constructed;
where W represents a weight.
It should be noted that the weight is an optimal weight obtained after the model training is finished, and belongs to a configuration parameter of the model.
Through the implementation mode, the knowledge representation is adaptively integrated into the corresponding semantic representation through a self-defined fusion mechanism, more external knowledge is introduced through the knowledge representation of the integrated map, the hidden relation between the entity and the synonym is established, and the long tail problem is effectively relieved.
The input unit 112 inputs the semantic representation of each synonym and the entity representation of each entity into the matching layer for matching, and obtains the current value of the loss function after matching.
In this embodiment, in the matching layer, the semantic representation of each synonym and the similarity between the entity representations of each entity are calculated to match the two.
Further, since the accuracy and precision of the whole model are ensured, the model needs to be continuously adjusted through a loss function to ensure that the accuracy of the model is continuously improved in the training process.
The training unit 116 performs back-pass training on the initial network according to the current value until the value of the loss function is no longer reduced, and stops training to obtain a retrieval model.
Specifically, the current value of the loss function is continuously returned, so that the model is continuously updated iteratively according to the loss function, when the value of the loss function is not reduced, the loss function is shown to be converged, and at this time, the training is stopped, and the retrieval model can be obtained.
It should be noted that the retrieval model is a model with a brand-new structure constructed by combining semantic information and atlas knowledge information, and the training effect of the model is continuously improved by the auxiliary training of the loss function in the training process of the model.
The type of the loss function may be configured by a user, which is not limited in the present invention.
The input unit 112 obtains the synonym to be retrieved, inputs the synonym to be retrieved to the retrieval model, and obtains the output of the retrieval model as the disease entity of the synonym to be retrieved.
By the embodiment, the retrieval of the disease entity can be realized through the custom-trained retrieval model, and the accuracy of the retrieval result is effectively improved.
In at least one embodiment of the present invention, a similarity between the synonym to be retrieved and each entity included in the retrieval model is calculated through a Noise Contrast Estimation (NCE) algorithm;
sequencing all entities contained in the retrieval model according to the sequence of the similarity from high to low;
and outputting the first-ranked entities, and determining the output entities as the disease entities of the synonyms to be searched.
For example: when the synonym to be searched is "skin laxity", the "skin laxity" can be identified as the synonym of "ehler syndrome" through the search model, that is, the search model outputs "ehler syndrome" as the disease entity of "skin laxity". Specifically, the map entity and candidate entity corresponding to "skin laxity" include "ehler syndrome", "scald", "dermatitis", and the like. Through knowledge map information, the three entities are all found to belong to skin diseases, and although the 'skin laxity' is completely dissimilar to the 'Eiles syndrome' in word, the 'skin laxity' has a similar meaning to the symptom 'extremely elastic skin', and has no obvious similarity to the attributes of the other two entities. Thus, the search model predicts that "skin laxity" belongs to the synonym of "Ehlers syndrome".
It should be noted that, in order to solve the OOV problem, the present embodiment uses the character-level features instead of the conventional word-level features, but at the same time, it also brings a high calculation amount. Therefore, in order to avoid the operation burden on the system due to a high calculation amount, the noise contrast estimation algorithm is further adopted to calculate the similarity between the synonym and the entity, and the noise contrast estimation algorithm is characterized in that the problem of complex calculation of the neural network can be solved, so that an obvious neutralization effect is exerted on the calculation amount, and the influence on the system is reduced.
In at least one embodiment of the invention, when a new entity is detected to be added to the disease knowledge graph, synonyms for the new entity are obtained;
and performing supplementary training on the retrieval model according to the new entity and the synonym of the new entity.
Specifically, the supplementary training is equivalent to adding a new training sample on the basis of the original model, and the training mode is similar to that of the search model and is not repeated here.
It should be noted that, in order to further ensure the security of the data, the retrieval model may be deployed in the blockchain to avoid malicious tampering of the data.
It can be seen from the above technical solutions that the present invention can obtain each entity in a pre-constructed disease knowledge graph, obtain synonyms of each entity, and invoke an initial network, wherein the structure of the initial network includes a coding layer, a graph embedding layer, a fusion layer, and a matching layer, and combines semantic information and graph knowledge information to better implement a synonym matching task, and inputs each entity and synonyms of each entity into the coding layer, and converts each entity and synonyms of each entity by using a feature conversion layer of the coding layer to obtain a character vector of each entity and a character vector of each synonym, and replaces the conventional term-level features with the character-level features, so as to effectively solve the OOV problem, construct sub-graphs of each entity in the disease knowledge graph, and input sub-graphs of each entity into the graph embedding layer, obtaining a knowledge vector of each entity, mapping the character vector of each entity and the character vector of each synonym to the same semantic space by using a full connection layer of a coding layer based on an alignment mechanism of shared weight, obtaining a semantic representation of each entity and a semantic representation of each synonym so as to solve the problem that the synonym expressed by spoken language and the medical entity expressed by formal professional language belong to different semantic spaces, performing semantic space mapping on the knowledge vector of each entity by using the full connection layer of the coding layer to obtain a knowledge representation of each entity, fusing the semantic representation of each entity and the knowledge representation of each entity by using a fusion layer to obtain an entity representation of each entity, adaptively integrating the knowledge representations into the corresponding semantic representations by a self-defined fusion mechanism, and integrating the knowledge representations by using a map, introducing more external knowledge, establishing a hidden relation between entities and synonyms, effectively relieving the long tail problem, inputting the semantic representation of each synonym and the entity representation of each entity into the matching layer for matching, acquiring the current value of a loss function after matching, performing return training on the initial network according to the current value until the value of the loss function is not reduced any more, stopping training, obtaining a retrieval model, continuously improving the training effect of the model through the auxiliary training of the loss function, acquiring the synonym to be retrieved, inputting the synonym to be retrieved into the retrieval model, acquiring the output of the retrieval model as the disease entity of the synonym to be retrieved, further realizing the retrieval of the disease entity through the self-defined trained retrieval model, and effectively improving the accuracy of the retrieval result.
Fig. 3 is a schematic structural diagram of an electronic device for implementing a disease entity search method according to a preferred embodiment of the present invention.
The electronic device 1 may comprise a memory 12, a processor 13 and a bus, and may further comprise a computer program, such as a disease entity retrieving program, stored in the memory 12 and executable on the processor 13.
It will be understood by those skilled in the art that the schematic diagram is merely an example of the electronic device 1, and does not constitute a limitation to the electronic device 1, the electronic device 1 may have a bus-type structure or a star-type structure, the electronic device 1 may further include more or less hardware or software than those shown in the figures, or different component arrangements, for example, the electronic device 1 may further include an input and output device, a network access device, and the like.
It should be noted that the electronic device 1 is only an example, and other existing or future electronic products, such as those that can be adapted to the present invention, should also be included in the scope of the present invention, and are included herein by reference.
The memory 12 includes at least one type of readable storage medium, which includes flash memory, removable hard disks, multimedia cards, card-type memory (e.g., SD or DX memory, etc.), magnetic memory, magnetic disks, optical disks, etc. The memory 12 may in some embodiments be an internal storage unit of the electronic device 1, for example a removable hard disk of the electronic device 1. The memory 12 may also be an external storage device of the electronic device 1 in other embodiments, such as a plug-in mobile hard disk, a Smart Media Card (SMC), a Secure Digital (SD) Card, a Flash memory Card (Flash Card), and the like provided on the electronic device 1. Further, the memory 12 may also include both an internal storage unit and an external storage device of the electronic device 1. The memory 12 may be used not only to store application software installed in the electronic device 1 and various types of data such as codes of disease entity search programs, etc., but also to temporarily store data that has been output or is to be output.
The processor 13 may be composed of an integrated circuit in some embodiments, for example, a single packaged integrated circuit, or may be composed of a plurality of integrated circuits packaged with the same or different functions, including one or more Central Processing Units (CPUs), microprocessors, digital Processing chips, graphics processors, and combinations of various control chips. The processor 13 is a Control Unit (Control Unit) of the electronic device 1, connects various components of the electronic device 1 by various interfaces and lines, and executes various functions and processes data of the electronic device 1 by running or executing programs or modules (e.g., executing a disease entity search program, etc.) stored in the memory 12 and calling data stored in the memory 12.
The processor 13 executes an operating system of the electronic device 1 and various installed application programs. The processor 13 executes the application program to implement the steps in the above-mentioned embodiments of the disease entity retrieval method, such as the steps shown in fig. 1.
Illustratively, the computer program may be divided into one or more modules/units, which are stored in the memory 12 and executed by the processor 13 to accomplish the present invention. The one or more modules/units may be a series of computer program instruction segments capable of performing specific functions, which are used to describe the execution process of the computer program in the electronic device 1. For example, the computer program may be divided into an acquisition unit 110, a call unit 111, an input unit 112, a construction unit 113, a mapping unit 114, a fusion unit 115, a training unit 116.
The integrated unit implemented in the form of a software functional module may be stored in a computer-readable storage medium. The software functional module is stored in a storage medium and includes several instructions to enable a computer device (which may be a personal computer, a computer device, or a network device) or a processor (processor) to execute parts of the disease entity retrieval method according to the embodiments of the present invention.
The integrated modules/units of the electronic device 1 may be stored in a computer-readable storage medium if they are implemented in the form of software functional units and sold or used as separate products. Based on such understanding, all or part of the flow of the method according to the embodiments of the present invention may be implemented by a computer program, which may be stored in a computer-readable storage medium, and when the computer program is executed by a processor, the steps of the method embodiments may be implemented.
Wherein the computer program comprises computer program code, which may be in the form of source code, object code, an executable file or some intermediate form, etc. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, U-disk, removable hard disk, magnetic disk, optical disk, computer Memory, Read-Only Memory (ROM), random-access Memory, or the like.
Further, the computer-readable storage medium may mainly include a storage program area and a storage data area, wherein the storage program area may store an operating system, an application program required for at least one function, and the like; the storage data area may store data created according to the use of the blockchain node, and the like.
The block chain is a novel application mode of computer technologies such as distributed data storage, point-to-point transmission, a consensus mechanism, an encryption algorithm and the like. A block chain (Blockchain), which is essentially a decentralized database, is a series of data blocks associated by using a cryptographic method, and each data block contains information of a batch of network transactions, so as to verify the validity (anti-counterfeiting) of the information and generate a next block. The blockchain may include a blockchain underlying platform, a platform product service layer, an application service layer, and the like.
The bus may be a Peripheral Component Interconnect (PCI) bus, an Extended Industry Standard Architecture (EISA) bus, or the like. The bus may be divided into an address bus, a data bus, a control bus, etc. For ease of illustration, only one arrow is shown in FIG. 3, but this does not indicate only one bus or one type of bus. The bus is arranged to enable connection communication between the memory 12 and at least one processor 13 or the like.
Although not shown, the electronic device 1 may further include a power supply (such as a battery) for supplying power to each component, and preferably, the power supply may be logically connected to the at least one processor 13 through a power management device, so as to implement functions of charge management, discharge management, power consumption management, and the like through the power management device. The power supply may also include any component of one or more dc or ac power sources, recharging devices, power failure detection circuitry, power converters or inverters, power status indicators, and the like. The electronic device 1 may further include various sensors, a bluetooth module, a Wi-Fi module, and the like, which are not described herein again.
Further, the electronic device 1 may further include a network interface, and optionally, the network interface may include a wired interface and/or a wireless interface (such as a WI-FI interface, a bluetooth interface, etc.), which are generally used for establishing a communication connection between the electronic device 1 and other electronic devices.
Optionally, the electronic device 1 may further comprise a user interface, which may be a Display (Display), an input unit (such as a Keyboard), and optionally a standard wired interface, a wireless interface. Alternatively, in some embodiments, the display may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, an OLED (Organic Light-Emitting Diode) touch device, or the like. The display, which may also be referred to as a display screen or display unit, is suitable for displaying information processed in the electronic device 1 and for displaying a visualized user interface, among other things.
It is to be understood that the described embodiments are for purposes of illustration only and that the scope of the appended claims is not limited to such structures.
Fig. 3 only shows the electronic device 1 with components 12-13, and it will be understood by a person skilled in the art that the structure shown in fig. 3 does not constitute a limitation of the electronic device 1, and may comprise fewer or more components than shown, or a combination of certain components, or a different arrangement of components.
With reference to fig. 1, the memory 12 of the electronic device 1 stores a plurality of instructions to implement a disease entity retrieval method, and the processor 13 executes the plurality of instructions to implement:
acquiring each entity in a pre-constructed disease knowledge graph and acquiring synonyms of each entity;
calling an initial network, wherein the structure of the initial network comprises a coding layer, a graph embedding layer, a fusion layer and a matching layer;
inputting each entity and synonyms of each entity into the coding layer, and converting each entity and synonyms of each entity by using a feature conversion layer of the coding layer to obtain a character vector of each entity and a character vector of each synonym;
constructing a subgraph of each entity in the disease knowledge graph, and inputting the subgraph of each entity into the graph embedding layer to obtain a knowledge vector of each entity;
based on an alignment mechanism of shared weight, mapping the character vector of each entity and the character vector of each synonym to the same semantic space by using a full connection layer of the coding layer to obtain a semantic representation of each entity and a semantic representation of each synonym;
carrying out semantic space mapping on the knowledge vector of each entity by using the full-connection layer of the coding layer to obtain the knowledge representation of each entity;
fusing the semantic representation of each entity and the knowledge representation of each entity by using the fusion layer to obtain the entity representation of each entity;
inputting the semantic representation of each synonym and the entity representation of each entity into the matching layer for matching, and acquiring the current value of the loss function after matching;
carrying out back transmission training on the initial network according to the current value until the value of the loss function is not reduced any more, and stopping training to obtain a retrieval model;
and acquiring the synonym to be retrieved, inputting the synonym to be retrieved into the retrieval model, and acquiring the output of the retrieval model as the disease entity of the synonym to be retrieved.
Specifically, the processor 13 may refer to the description of the relevant steps in the embodiment corresponding to fig. 1 for a specific implementation method of the instruction, which is not described herein again.
In the embodiments provided in the present invention, it should be understood that the disclosed system, apparatus and method may be implemented in other ways. For example, the above-described apparatus embodiments are merely illustrative, and for example, the division of the modules is only one logical functional division, and other divisions may be realized in practice.
The modules described as separate parts may or may not be physically separate, and parts displayed as modules may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of the present embodiment.
In addition, functional modules in the embodiments of the present invention may be integrated into one processing unit, or each unit may exist alone physically, or two or more units are integrated into one unit. The integrated unit can be realized in a form of hardware, or in a form of hardware plus a software functional module.
It will be evident to those skilled in the art that the invention is not limited to the details of the foregoing illustrative embodiments, and that the present invention may be embodied in other specific forms without departing from the spirit or essential attributes thereof.
The present embodiments are therefore to be considered in all respects as illustrative and not restrictive, the scope of the invention being indicated by the appended claims rather than by the foregoing description, and all changes which come within the meaning and range of equivalency of the claims are therefore intended to be embraced therein. Any reference signs in the claims shall not be construed as limiting the claim concerned.
Furthermore, it is obvious that the word "comprising" does not exclude other elements or steps, and the singular does not exclude the plural. A plurality of units or means recited in the present invention may also be implemented by one unit or means through software or hardware. The terms first, second, etc. are used to denote names, but not any particular order.
Finally, it should be noted that the above embodiments are only for illustrating the technical solutions of the present invention and not for limiting, and although the present invention is described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that modifications or equivalent substitutions may be made on the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.
Claims (10)
1. A disease entity retrieval method, comprising:
acquiring each entity in a pre-constructed disease knowledge graph and acquiring synonyms of each entity;
calling an initial network, wherein the structure of the initial network comprises a coding layer, a graph embedding layer, a fusion layer and a matching layer;
inputting each entity and synonyms of each entity into the coding layer, and converting each entity and synonyms of each entity by using a feature conversion layer of the coding layer to obtain a character vector of each entity and a character vector of each synonym;
constructing a subgraph of each entity in the disease knowledge graph, and inputting the subgraph of each entity into the graph embedding layer to obtain a knowledge vector of each entity;
based on an alignment mechanism of shared weight, mapping the character vector of each entity and the character vector of each synonym to the same semantic space by using a full connection layer of the coding layer to obtain a semantic representation of each entity and a semantic representation of each synonym;
carrying out semantic space mapping on the knowledge vector of each entity by using the full-connection layer of the coding layer to obtain the knowledge representation of each entity;
fusing the semantic representation of each entity and the knowledge representation of each entity by using the fusion layer to obtain the entity representation of each entity;
inputting the semantic representation of each synonym and the entity representation of each entity into the matching layer for matching, and acquiring the current value of the loss function after matching;
carrying out back transmission training on the initial network according to the current value until the value of the loss function is not reduced any more, and stopping training to obtain a retrieval model;
and acquiring the synonym to be retrieved, inputting the synonym to be retrieved into the retrieval model, and acquiring the output of the retrieval model as the disease entity of the synonym to be retrieved.
2. The disease entity retrieval method of claim 1, wherein the obtaining synonyms for each entity comprises one or more of the following:
acquiring a database associated with the disease knowledge graph, and acquiring synonyms of each entity from the database; and/or
And crawling the designated page by adopting a web crawler technology to obtain the synonym of each entity.
3. The disease entity retrieval method of claim 1, wherein the constructing of the subgraph of each entity in the disease knowledge graph comprises:
performing multi-hop processing on each entity in a preset dimension to obtain a connection diagram of each entity in each dimension;
and combining the connection graphs obtained by each entity under each dimension to obtain a subgraph of each entity.
4. The disease entity retrieval method of claim 1, wherein the fusing the semantic representation of each entity and the knowledge representation of each entity using the fusion layer to obtain the entity representation of each entity comprises:
constructing a preset function according to the semantic representation of each entity and the corresponding knowledge representation of each entity;
constructing a convolution function according to the knowledge representation of each entity and the corresponding preset function;
the sum of the semantic representation of each entity and the corresponding convolution function is computed as the entity representation of each entity.
5. The disease entity retrieval method of claim 4, wherein the constructing of the predetermined function based on the semantic representation of each entity and the corresponding knowledge representation of each entity comprises:
acquiring configured weight;
constructing a weight function according to the weight, the semantic representation of each entity and the corresponding knowledge representation of each entity;
and processing the weight function by utilizing a Softmax function to obtain the preset function.
6. The disease entity retrieval method of claim 1, further comprising:
calculating the similarity between the synonym to be retrieved and each entity contained in the retrieval model through a noise comparison estimation algorithm;
sequencing all entities contained in the retrieval model according to the sequence of the similarity from high to low;
and outputting the first-ranked entities, and determining the output entities as the disease entities of the synonyms to be searched.
7. The disease entity retrieval method of claim 1, further comprising:
when detecting that a new entity is added into the disease knowledge graph, acquiring synonyms of the new entity;
and performing supplementary training on the retrieval model according to the new entity and the synonym of the new entity.
8. A disease entity retrieval apparatus, characterized by comprising:
the acquisition unit is used for acquiring each entity in a pre-constructed disease knowledge graph and acquiring synonyms of each entity;
the device comprises a calling unit, a matching unit and a processing unit, wherein the calling unit is used for calling an initial network, and the structure of the initial network comprises a coding layer, a graph embedding layer, a fusion layer and a matching layer;
the input unit is used for inputting each entity and synonyms of each entity into the coding layer, and converting each entity and synonyms of each entity by using the feature conversion layer of the coding layer to obtain a character vector of each entity and a character vector of each synonym;
the construction unit is used for constructing a sub-graph of each entity in the disease knowledge graph and inputting the sub-graph of each entity into the graph embedding layer to obtain a knowledge vector of each entity;
the mapping unit is used for mapping the character vector of each entity and the character vector of each synonym to the same semantic space by utilizing the full-connection layer of the coding layer based on the alignment mechanism of the shared weight to obtain the semantic representation of each entity and the semantic representation of each synonym;
the mapping unit is further configured to perform semantic space mapping on the knowledge vector of each entity by using a full-connection layer of the coding layer to obtain a knowledge representation of each entity;
the fusion unit is used for fusing the semantic representation of each entity and the knowledge representation of each entity by using the fusion layer to obtain the entity representation of each entity;
the input unit is further used for inputting the semantic representation of each synonym and the entity representation of each entity into the matching layer for matching, and acquiring the current value of the loss function after matching;
the training unit is used for carrying out return training on the initial network according to the current value until the value of the loss function is not reduced any more, and stopping training to obtain a retrieval model;
the input unit is further configured to acquire a synonym to be retrieved, input the synonym to be retrieved to the retrieval model, and acquire an output of the retrieval model as a disease entity of the synonym to be retrieved.
9. An electronic device, characterized in that the electronic device comprises:
a memory storing at least one instruction; and
a processor executing instructions stored in the memory to implement the disease entity retrieval method of any one of claims 1 to 7.
10. A computer-readable storage medium characterized by: the computer-readable storage medium has stored therein at least one instruction that is executable by a processor in an electronic device to implement the disease entity retrieval method of any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110485329.1A CN113221578B (en) | 2021-04-30 | 2021-04-30 | Disease entity retrieval method, device, equipment and medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110485329.1A CN113221578B (en) | 2021-04-30 | 2021-04-30 | Disease entity retrieval method, device, equipment and medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113221578A true CN113221578A (en) | 2021-08-06 |
| CN113221578B CN113221578B (en) | 2022-11-25 |
Family
ID=77090642
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110485329.1A Active CN113221578B (en) | 2021-04-30 | 2021-04-30 | Disease entity retrieval method, device, equipment and medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113221578B (en) |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109522465A (en) * | 2018-10-22 | 2019-03-26 | 国家电网公司 | The semantic searching method and device of knowledge based map |
| CN110287337A (en) * | 2019-06-19 | 2019-09-27 | 上海交通大学 | The system and method for medicine synonym is obtained based on deep learning and knowledge mapping |
| US20200242444A1 (en) * | 2019-01-30 | 2020-07-30 | Baidu Usa Llc | Knowledge-graph-embedding-based question answering |
| CN111475623A (en) * | 2020-04-09 | 2020-07-31 | 北京北大软件工程股份有限公司 | Case information semantic retrieval method and device based on knowledge graph |
| CN112084331A (en) * | 2020-08-27 | 2020-12-15 | 清华大学 | Text processing method, text processing device, model training method, model training device, computer equipment and storage medium |
| CN112119412A (en) * | 2018-05-18 | 2020-12-22 | 伯耐沃伦人工智能科技有限公司 | Neural network of map with attention |
-
2021
- 2021-04-30 CN CN202110485329.1A patent/CN113221578B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112119412A (en) * | 2018-05-18 | 2020-12-22 | 伯耐沃伦人工智能科技有限公司 | Neural network of map with attention |
| CN109522465A (en) * | 2018-10-22 | 2019-03-26 | 国家电网公司 | The semantic searching method and device of knowledge based map |
| US20200242444A1 (en) * | 2019-01-30 | 2020-07-30 | Baidu Usa Llc | Knowledge-graph-embedding-based question answering |
| CN110287337A (en) * | 2019-06-19 | 2019-09-27 | 上海交通大学 | The system and method for medicine synonym is obtained based on deep learning and knowledge mapping |
| CN111475623A (en) * | 2020-04-09 | 2020-07-31 | 北京北大软件工程股份有限公司 | Case information semantic retrieval method and device based on knowledge graph |
| CN112084331A (en) * | 2020-08-27 | 2020-12-15 | 清华大学 | Text processing method, text processing device, model training method, model training device, computer equipment and storage medium |
Non-Patent Citations (2)
| Title |
|---|
| YIYING YANG ET AL.: "KGSynNet: A Novel Entity Synonyms Discovery Framework with Knowledge Graph", 《DATABASE SYSTEMS FOR ADVANCED APPLICATIONS. DASFAA 2021》 * |
| 张紫璇 等: "面向水利信息资源的智能问答系统构建与应用", 《计算机与现代化》 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113221578B (en) | 2022-11-25 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110609902B (en) | Text processing method and device based on fusion knowledge graph | |
| WO2018144052A1 (en) | K-selection using parallel processing | |
| CN113707297A (en) | Medical data processing method, device, equipment and storage medium | |
| CN112541338A (en) | Similar text matching method and device, electronic equipment and computer storage medium | |
| CN111460797B (en) | Keyword extraction method and device, electronic equipment and readable storage medium | |
| CN112885478B (en) | Medical document retrieval method, medical document retrieval device, electronic device and storage medium | |
| CN111639153A (en) | Query method and device based on legal knowledge graph, electronic equipment and medium | |
| CN112149409B (en) | Medical word cloud generation method and device, computer equipment and storage medium | |
| CN113312461A (en) | Intelligent question-answering method, device, equipment and medium based on natural language processing | |
| CN113378970B (en) | Sentence similarity detection method and device, electronic equipment and storage medium | |
| CN111930962A (en) | Document data value evaluation method and device, electronic equipment and storage medium | |
| CN112507230B (en) | Webpage recommendation method and device based on browser, electronic equipment and storage medium | |
| CN111753089A (en) | Topic clustering method and device, electronic equipment and storage medium | |
| CN113706322A (en) | Service distribution method, device, equipment and storage medium based on data analysis | |
| CN113886708A (en) | Product recommendation method, device, equipment and storage medium based on user information | |
| CN113887941A (en) | Business process generation method and device, electronic equipment and medium | |
| WO2022227171A1 (en) | Method and apparatus for extracting key information, electronic device, and medium | |
| CN112860851B (en) | Course recommendation method, device, equipment and medium based on root cause analysis | |
| CN112651782B (en) | Behavior prediction method, device, equipment and medium based on dot product attention scaling | |
| CN113220820B (en) | Efficient SPARQL query response method, device and equipment based on graph | |
| CN112507728A (en) | Intelligent conversation method and device, electronic equipment and storage medium | |
| CN116821373A (en) | Map-based prompt recommendation method, device, equipment and medium | |
| CN113221578B (en) | Disease entity retrieval method, device, equipment and medium | |
| CN112330080B (en) | Factor screening method, device, equipment and medium based on connectivity graph | |
| CN115346095A (en) | Visual question answering method, device, equipment and storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |