CN111414750B - Synonym distinguishing method, device, equipment and storage medium - Google Patents
Synonym distinguishing method, device, equipment and storage medium Download PDFInfo
- Publication number
- CN111414750B CN111414750B CN202010190072.2A CN202010190072A CN111414750B CN 111414750 B CN111414750 B CN 111414750B CN 202010190072 A CN202010190072 A CN 202010190072A CN 111414750 B CN111414750 B CN 111414750B
- Authority
- CN
- China
- Prior art keywords
- entry
- synonymous
- layer
- characteristic information
- pairs
- 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.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
- Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation
Abstract
The application discloses a method, a device, equipment and a storage medium for distinguishing synonyms of terms, and relates to the technical field of knowledge maps. The specific implementation scheme is as follows: acquiring characteristic information of entry pairs to be subjected to synonymous judgment; inputting the characteristic information of the entry pairs into a trained neural network model to obtain synonymous discrimination results of the entry pairs; according to the embodiment, the structure, the parameters and the characteristics of the pre-training layer are directly subjected to knowledge migration, the pre-training layer is adopted to learn the characteristic information of the entry pairs, the labeling amount in the training process is reduced, a large amount of resources and manpower are saved, the synonymous judging efficiency is further improved, and the accuracy of synonymous judging results is improved.
Description
Technical Field
The application relates to the computer technology, in particular to the technical field of knowledge maps.
Background
In some industries, the terms describing entities within the industry are more aliases, and the spoken language descriptions are also quite different. For example, in the medical field, the standard names and aliases are more for entities of examination, surgery, clinic, medicine, diseases, etc., for example, common cold and upper respiratory tract infections describe the same diseases, pregnancy-induced hypertension and gestational hypertension.
When intelligent items are landed, standard names and aliases of the same entity need to be unified so that the items can run successfully. At present, the synonymous relation of the entries is mainly confirmed in a manual auditing and labeling mode, and synonymous entries are unified.
The mode of manual auditing and labeling requires a large amount of manpower, is long in time consumption, and can influence the landing of related intelligent projects.
Disclosure of Invention
The embodiment of the application provides a synonym distinguishing method, a synonym distinguishing device, a synonym distinguishing equipment and a storage medium, which are used for improving distinguishing efficiency of the vocabulary entries, saving manpower resources and accelerating the landing of related intelligent projects.
In a first aspect, an embodiment of the present application provides a method for determining synonyms of terms, including:
acquiring characteristic information of entry pairs to be subjected to synonymous judgment;
inputting the characteristic information of the entry pairs into a trained neural network model to obtain synonymous discrimination results of the entry pairs;
the neural network model comprises a pre-training layer, a fine-tuning layer and an output layer; the pre-training layer performs training of language understanding tasks by adopting natural language training samples in advance and is used for learning the characteristic information of the entry pairs to obtain the language understanding information of the entry pairs; the fine tuning layer is used for carrying out feature extraction and fusion on the language understanding information to obtain feature representation of whether the entry pairs are synonymous or not; and the output layer is used for obtaining the synonymous judging result according to the characteristic representation.
According to the embodiment of the application, the feature information of the entry pairs is input into the trained neural network model to carry out synonym discrimination, so that the synonym discrimination is carried out by combining the deep learning function of the neural network model on the feature level, the synonym discrimination efficiency can be improved, the manpower resources are saved, and the landing of related intelligent projects is accelerated; in the embodiment, the pre-training layer is obtained by training a language understanding task by adopting a natural language training sample in advance, and the natural language training sample and the vocabulary entry pair have the same feature distribution, so that the structure, the parameters and the features of the pre-training layer are directly subjected to knowledge migration, the feature information of the vocabulary entry pair is learned by adopting the pre-training layer, the labeling amount in the training process is reduced, a large amount of resources and manpower are saved, and the synonymous discrimination efficiency is further improved; by adopting the language understanding information, whether the entry pair is synonymous or not can be accurately reflected, the feature extraction and fusion are carried out on the language understanding information through the fine adjustment layer, the accurate feature representation of whether the entry pair is synonymous or not is obtained, the synonymous judging result is further obtained through the output layer according to the accurate feature representation, and the accuracy of the synonymous judging result is improved.
Optionally, the pre-training layer is a neural network structure of multiple language understanding tasks; the multilingual understanding task includes a lexical level task, a grammar level task, and a semantic level task, and the language understanding information includes lexical information, grammar information, and semantic information.
In an optional implementation manner in the above application, the pre-training layer better understands the information contained in the vocabulary entry pairs from 3 layers of morphology, grammar and semantics, so that the general semantic representation capability is greatly enhanced, and a more accurate synonymous discrimination result is obtained through a neural network model.
Optionally, the fine tuning layer includes: a convolution layer, a pooling layer and a full connection layer;
the convolution layer is used for extracting features of the language understanding information, the pooling layer is used for reducing dimensions of the extracted features, and the full-connection layer is used for fusing the features subjected to the dimension reduction to obtain feature representation of whether the entry pairs are synonymous.
In an optional implementation manner in the above application, the feature extraction and fusion functions of the fine tuning layer are realized through the convolution layer, the pooling layer and the full connection layer, and the implementation manner is simple and effective. When the neural network model is trained, different features can be flexibly extracted by adjusting parameters of a convolution layer, a pooling layer and a full connection layer, and fusion of different dimensions is carried out on the features, so that whether synonymous feature representation can be accurately obtained for each type of entry pair in each field.
Optionally, the output layer includes a classification layer, configured to perform synonym and heteronym classification on the feature representation of whether the entry is synonymous, so as to obtain confidence degrees of the synonymous type and the heteronym type.
An alternative embodiment of the above application performs synonymous and nonsense classification of the feature representation by the classification layer, and performs the classification operation instead of normalizing the feature representation, where the classification operation has a higher accuracy relative to normalization.
Optionally, the output layer further includes an intervention layer, configured to determine whether the feature information of the entry pair meets a set rule of dissimilarity or synonym; if the characteristic information of the entry pair meets the set dissimilarity rule, reducing the confidence coefficient of the synonymous type and/or improving the confidence coefficient of the dissimilarity type; if the characteristic information of the entry pair meets the set synonym rule, the confidence of the synonym type is improved and/or the confidence of the heteronym type is reduced;
wherein the synonym rule includes: a first template of a pair of synonyms, the first template comprising a first fixed vocabulary and a plurality of first candidate vocabularies, the rules comprising: the second template of the pair of the heteronyms comprises a second fixed vocabulary and a plurality of second candidate vocabularies.
In an optional implementation manner in the above application, considering that for some special term pairs, the output of the classification layer may have an error, according to whether the feature information satisfies the synonymous rule or the nonsense rule, the opposite confidence is interfered, so that the embodiment is also suitable for classifying special words; further, the synonym rule and the heteronym rule comprise corresponding templates, fixed vocabularies and candidate vocabularies, are suitable for synonym judgment of entry pairs which accord with the templates and contain the fixed vocabularies and any candidate vocabularies, and judge whether characteristic information meets the corresponding rules through template comparison.
Optionally, the obtaining the feature information of the entry pair to be synonymously judged includes at least one of the following operations:
acquiring plain text characteristic information of entry pairs to be subjected to synonymous judgment;
acquiring natural word segmentation characteristic information of entry pairs to be subjected to synonymous judgment;
acquiring part characteristic information of entry pairs to be subjected to synonymous discrimination;
acquiring degree characteristic information of entry pairs to be subjected to synonymous judgment;
obtaining directional characteristic information of entry pairs to be subjected to synonymous judgment;
obtaining frequency characteristic information of entry pairs to be subjected to synonymous judgment;
Acquiring quantity characteristic information of entry pairs to be subjected to synonymous judgment;
and obtaining sensory characteristic information of the entry pairs to be subjected to synonymous judgment.
In an optional implementation manner in the above application, feature information of the entry pairs is constructed from multiple dimensions, so that features of the entry pairs are comprehensively and deeply mined, and more accurate discrimination results are facilitated.
Optionally, before inputting the feature information of the term pair into the trained neural network model to obtain the synonym discrimination result of the term pair, the method further includes:
acquiring a neural network model to be trained, wherein the neural network model to be trained comprises the pre-training layer, a fine-tuning layer to be trained and an output layer to be trained;
and training the neural network model to be trained by taking the characteristic information of the synonymous entry pairs as a positive sample and the characteristic information of the dissimilarity entry pairs as a negative sample, and keeping the parameters and the characteristic representation of the pre-training layer unchanged in the training process.
In an optional implementation manner in the above application, parameters and characteristic representations of the pre-training layer are kept unchanged when the neural network model is trained, so that strong language understanding capability of the pre-training layer is ensured; and training the fine tuning layer and the output layer to enable the fine tuning layer to extract synonymous features from language understanding information and fuse the synonymous features to obtain synonymous feature representation or not, and enabling the output layer to obtain synonymous judging results according to the feature representation.
Optionally, the entry pair is an entity name in the medical field; the entity names include entity common names and entity aliases.
An optional implementation manner in the above application defines the fields and types of entry pairs, and by the method provided by the embodiment, the unification of entity common names and entity aliases in the medical field can be realized, so that the landing of medical items is quickened.
In a second aspect, an embodiment of the present application further provides a synonym determining device for an entry, including:
the acquisition module is used for acquiring characteristic information of entry pairs to be subjected to synonymous judgment;
the judging module is used for inputting the characteristic information of the entry pairs into the trained neural network model to obtain synonymous judging results of the entry pairs;
the neural network model comprises a pre-training layer, a fine-tuning layer and an output layer; the pre-training layer performs training of language understanding tasks by adopting natural language training samples in advance and is used for learning the characteristic information of the entry pairs to obtain the language understanding information of the entry pairs; the fine tuning layer is used for carrying out feature extraction and fusion on the language understanding information to obtain feature representation of whether the entry pairs are synonymous or not; and the output layer is used for obtaining the synonymous judging result according to the characteristic representation.
In a third aspect, an embodiment of the present application further provides an electronic device, including:
at least one processor; and
a memory communicatively coupled to the at least one processor; wherein, the liquid crystal display device comprises a liquid crystal display device,
the memory stores instructions executable by the at least one processor to enable the at least one processor to perform a synonym determination for an entry as provided by the embodiments of the first aspect.
In a fourth aspect, embodiments of the present application further provide a non-transitory computer-readable storage medium storing computer instructions for causing the computer to perform a synonym distinction method for an entry as provided by the embodiments of the first aspect.
Other effects of the above alternative will be described below in connection with specific embodiments.
Drawings
The drawings are included to provide a better understanding of the present application and are not to be construed as limiting the application. Wherein:
FIG. 1 is a flowchart of a method for determining synonym of an entry according to the first embodiment of the present disclosure;
FIG. 2 is a schematic diagram of a neural network model according to a second embodiment of the present application;
FIG. 3 is a flowchart of a method for determining synonym of an entry in a third embodiment of the present disclosure;
Fig. 4 is a block diagram of a synonym determination device for an entry as in the fourth embodiment of the present application;
fig. 5 is a block diagram of an electronic device for implementing a synonym distinction method for terms of an embodiment of the present disclosure.
Detailed Description
Exemplary embodiments of the present application will now be described with reference to the accompanying drawings, in which various details of the embodiments of the present application are included to facilitate understanding, and are to be considered merely exemplary. Accordingly, those of ordinary skill in the art will recognize that various changes and modifications of the embodiments described herein can be made without departing from the scope and spirit of the application. Also, descriptions of well-known functions and constructions are omitted in the following description for clarity and conciseness.
Example 1
Fig. 1 is a flowchart of a method for determining synonym of an entry in a first embodiment of the present application, where the embodiment of the present application is applicable to a case where a pair of entries is determined to be synonymous or ambiguous, the method is performed by a synonym determining device of the entry, and the device is implemented by software and/or hardware and is specifically configured in an electronic device having a certain data computing capability.
The synonym distinguishing method of the entry as shown in fig. 1 comprises the following steps:
s101, obtaining characteristic information of entry pairs to be subjected to synonymous judgment.
The term pair to be subjected to the synonym discrimination includes two terms, and the length, language, field and type of the term pair to be subjected to the synonym discrimination are not limited in this embodiment. For example, entry pairs include review opinion notifications and overrule notifications, claims and claims in the intellectual property field; or entry pairs including novel coronavirus pneumonia and pneumonia, atypical pneumonia and SARS in the medical field.
The feature information of the term pair is feature information of the term pair in terms of natural language understanding, for example, feature information capable of reflecting a lexical, a grammatical and a semantic. Optionally, the present operation includes at least one of: acquiring plain text characteristic information of entry pairs to be subjected to synonymous judgment; acquiring natural word segmentation characteristic information of entry pairs to be subjected to synonymous judgment; acquiring part characteristic information of entry pairs to be subjected to synonymous discrimination; acquiring degree characteristic information of entry pairs to be subjected to synonymous judgment; obtaining directional characteristic information of entry pairs to be subjected to synonymous judgment; obtaining frequency characteristic information of entry pairs to be subjected to synonymous judgment; acquiring quantity characteristic information of entry pairs to be subjected to synonymous judgment; and obtaining sensory characteristic information of the entry pairs to be subjected to synonymous judgment.
Wherein the plain text characteristic information of the entry pairs is the text of each entry. The natural word segmentation characteristic information of the entry pairs is word segmentation obtained by segmenting each word according to word composition. The location feature information of the entry pair is text, such as a shoulder, a hand, etc., indicating a location, which each entry contains. The degree characteristic information of the entry pairs is text which is contained in each entry and indicates the degree, such as light weight and heavy weight. The directional characteristic information of the entry pairs is text indicating the direction contained in each entry, such as up and down. The frequency characteristic information of the entry pairs is text which is contained in each entry and indicates the frequency, such as recurrence, origination and secondary. The number characteristic information of the entry pairs is text which is contained in each entry and represents the number, such as double lung and five fracture. The sensory characteristic information of the entry pairs is sensory text, such as sound, smell, color, etc., contained in each entry.
Exemplary, characteristic information of the term for "ulcerative chronic colitis" in "ulcerative chronic colitis and chronic ulcerative colitis" includes: ulcerative chronic colitis, ulcers, chronic, colitis, colon, ulcerative and chronic. The characteristic information of "chronic ulcerative colitis" includes: chronic ulcerative colitis, chronic, ulcerative, sexual, colitis, colon, chronic and ulcerative.
S102, inputting the characteristic information of the entry pairs into the trained neural network model to obtain the synonymous discrimination result of the entry pairs.
The neural network model comprises a pre-training layer, a fine-tuning layer and an output layer. The pre-training layer, the fine tuning layer and the output layer are connected in sequence, the pre-training layer receives entry pairs input by a user, the output layer outputs synonymous judging results of the entry pairs, and the method comprises the following steps: synonymous and heterologous.
Specifically, the pre-training layer is training of performing language understanding tasks by adopting a natural language training sample in advance, and because the natural language training sample and the vocabulary entry pairs have the same feature distribution, the structure, the parameters and the features of the pre-training layer are directly subjected to knowledge migration, and the feature information of the vocabulary entry pairs is learned by adopting the pre-training layer, so that the language understanding information is obtained. Optionally, the pre-training layer is a neural network structure of multiple language understanding tasks; the multilingual understanding task includes a lexical level task, a grammar level task, and a semantic level task, and the language understanding information includes lexical information, grammar information, and semantic information. Based on the research results of the present stage, the structure, parameters and characteristics of the pretraining layer in ERNIE 2.0 are migrated to the pretraining layer in this embodiment, that is, the whole framework of the pretraining layer in ERNIE 2.0 is directly applied. The ERNIE 2.0 acquires the natural language information of multiple dimensions such as morphology, grammar, semantics and the like from the training data, and greatly enhances the universal semantic representation capability. And ERNIE 2.0 trains the model by using encyclopedia, news information and forum dialogue data, so that the semantic expression capability is enhanced. Making it more powerful in chinese semantic representation. Language understanding tasks of the pretraining layer in ERNIE 2.0 include: knowledge masking, tag-document relationships, capitalization prediction, sentence reordering, sentence distance, utterance relationships, and relevance.
The fine tuning layer is used for extracting and fusing the characteristics of the language understanding information to obtain the characteristic representation of whether the entry pairs are synonymous. The characteristic representation may be a probability value, the size of which indicates whether the entry pairs are synonymous. And the output layer is used for obtaining a synonymous judging result according to the characteristic representation. For example, if the characteristic representation is greater than or equal to a set threshold, such as 50%, then determining that the entry pairs are synonymous; if the characteristic representation is less than the set threshold, the entry pair is determined to be ambiguous.
According to the embodiment of the application, the feature information of the entry pairs is input into the trained neural network model to carry out synonym discrimination, so that the synonym discrimination is carried out by combining the deep learning function of the neural network model on the feature level, the synonym discrimination efficiency can be improved, the manpower resources are saved, and the landing of related intelligent projects is accelerated; in the embodiment, the structure, the parameters and the characteristics of the pre-training layer are directly subjected to knowledge migration, the pre-training layer is adopted to learn the characteristic information of the entry pairs, the labeling amount in the training process is reduced, a large amount of resources and manpower are saved, and the synonymous distinguishing efficiency is further improved; by adopting the language understanding information, whether the entry pair is synonymous or not can be accurately reflected, the feature extraction and fusion are carried out on the language understanding information through the fine adjustment layer, the accurate feature representation of whether the entry pair is synonymous or not is obtained, the synonymous judging result is further obtained through the output layer according to the accurate feature representation, and the accuracy of the synonymous judging result is improved.
Furthermore, feature information of the entry pairs is constructed from a plurality of dimensions, so that features of the entry pairs are comprehensively and deeply excavated, and more accurate judging results are obtained.
Furthermore, the pre-training layer better understands the information contained in the vocabulary entry pairs from 3 layers of morphology, grammar and semantics, so that the general semantic representation capability is greatly enhanced, and a more accurate synonymous discrimination result is obtained through a neural network model.
Example two
Fig. 2 is a schematic structural diagram of a neural network model in a second embodiment of the present application, where the structures of the fine tuning layer and the output layer are defined based on the technical solutions of the foregoing embodiments.
As shown in fig. 2, the neural network model includes: a pre-training layer 10, a fine tuning layer 20 and an output layer 30. Wherein the trim layer 20 includes a convolutional layer 21, a pooling layer 22, and a full connection layer 23.
The convolution layer 21 is configured to perform feature extraction on language understanding information to obtain a plurality of one-dimensional feature vectors. The pooling layer 22 may be a maximum pooling layer or an average pooling layer, and is used for performing dimension reduction on the extracted features, for example, taking a maximum value or an average value for each one-dimensional feature vector, and then splicing the features together to serve as an output value of the layer. The full connection layer 23 is used for fusing the features after dimension reduction to obtain a feature representation of whether the entry pairs are synonymous. The number of the full connection layers 23 is at least one, and depends on the feature dimension after dimension reduction and the feature dimension to be output, so as to improve the learning capability of the network. For the situation that the feature dimension after dimension reduction is one-dimensional, only one 23 full-connection layer is needed, and the feature after dimension reduction is added to obtain a probability value.
As shown in fig. 2, the output layer 30 includes a classification layer 31 for performing synonym and heteronym classification on the feature representation of whether the entry is synonymous, so as to obtain the confidence of the synonymous type and the heteronym type. The classification layer 31 is a neural network structure, such as a commonly used softmax layer. The feature representations are classified synonymously and dissimilarly by the classification layer 31, and the normalization of the feature representations is replaced by a classification operation with a higher accuracy than the normalization.
Considering that the output of the classification layer 31 may be incorrect for some specific entry pairs, such as hepatitis b and viral hepatitis b are synonyms, but classified as being heterologous by the classification layer 31. Respiratory tract infections and viral respiratory tract infections are not synonymous, but are classified as synonymous by classification layer 31. Thus requiring intervention on specific entry pairs.
Based on this, the output layer 30 further includes an intervention layer 32 connected after the classification layer 31, for judging whether the feature information of the entry pair satisfies a set rule of dissimilarity or synonym. Wherein, the synonymous rule includes: the first template of the synonym entry pair, the first template comprising a first fixed vocabulary and a plurality of first candidate vocabularies, the heterology rule comprising: the second template of the pair of the heteronyms comprises a second fixed vocabulary and a plurality of second candidate vocabularies.
Illustratively, the synonym rule includes: the template of liver and viral hepatitis, the fixed vocabulary is liver and viral hepatitis, and the candidate vocabulary includes A, B and C. If the input entry pairs are hepatitis B and viral hepatitis B, the input entry pairs are matched with the template, namely, the preset synonym rule is met, the confidence of the synonym type is improved and/or the confidence of the heteronym type is reduced, the improvement and the reduction can be the same or different, and the improvement and the reduction can be custom-set according to the synonym discrimination precision, such as 5%.
Illustratively, the rules of the dissimilarity include: the template "respiratory tract infection and respiratory tract infection", the fixed vocabulary is respiratory tract infection, and the candidate vocabulary comprises viruses and bacteria. If the input entry pairs are respiratory tract infection and viral respiratory tract infection, the input entry pairs are matched with the template, that is, meet the set dissimilarity rule, reduce the confidence of the synonymous type and/or improve the confidence of the dissimilarity type, the improvement and the reduction can be the same or different, and the amplitude can be custom set according to the synonymous discrimination precision, such as 5%.
In the embodiment, the feature extraction and fusion functions of the fine adjustment layer are realized through the convolution layer, the pooling layer and the full connection layer, and the realization mode is simple and effective. When the neural network model is trained, different features can be flexibly extracted by adjusting parameters of a convolution layer, a pooling layer and a full connection layer, and fusion of different dimensions is carried out on the features, so that whether synonymous feature representation can be accurately obtained for each type of entry pair in each field.
Furthermore, according to whether the feature information meets a synonymy rule or an dissimilarity rule, the opposite credibility is interfered, so that the embodiment is also suitable for classifying special words; further, the synonym rule and the heteronym rule comprise corresponding templates, fixed vocabularies and candidate vocabularies, are suitable for synonym judgment of entry pairs which accord with the templates and contain the fixed vocabularies and any candidate vocabularies, and judge whether characteristic information meets the corresponding rules through template comparison.
Example III
Fig. 3 is a flowchart of a synonym determining method for terms according to the third embodiment of the present invention, where the present embodiment is further optimized based on the above embodiments, and specifically, a training operation of a neural network model is added. The synonym distinguishing method of the entry shown in fig. 3 comprises the following steps:
s301, acquiring a neural network model to be trained, wherein the neural network model to be trained comprises a pre-training layer, a fine-tuning layer to be trained and an output layer to be trained.
S302, feature information of synonymous entry pairs is used as a positive sample, feature information of heterogenous entry pairs is used as a negative sample, a neural network model to be trained is trained, and parameters and feature representation of a pre-training layer are kept unchanged in the training process.
S303, obtaining characteristic information of entry pairs to be subjected to synonymous judgment.
S304, inputting the characteristic information of the entry pairs into the trained neural network model to obtain the synonymous discrimination result of the entry pairs.
Optionally, the term pair is an entity name in the medical field, where the entity name includes an entity common name and an entity alias. Specifically, the entry to be processed and the universal name of each entity are adopted to form entry pairs. If a certain entry pair is synonymous, establishing a corresponding relation between the entry to be processed and the common name of the other entity in the entry pair so as to realize unification. The embodiment limits the field and the type of entry pairs, and the method provided by the embodiment can realize the unification of entity common names and entity aliases in the medical field and quicken the landing of medical items.
In the neural network model to be trained, a pre-training layer adopts a natural language training sample in advance to train a language understanding task, wherein parameters and characteristics in the pre-training layer adopt pre-training results, and parameters and characteristics in a fine-tuning layer and an output layer are initial values and need to be trained.
Taking the medical field as an example, a plurality of synonymous entry pairs and a plurality of ambiguous entry pairs are extracted from medical records, prescriptions and prescriptions of various hospitals, for example, the synonymous entry pairs comprise: SARS and SARS, new coronapneumonia and new coronavirus pneumonia. The pair of ambiguous entries includes: hypertension and hyperlipidemia, recurrent tuberculosis and secondary tuberculosis. And extracting the characteristic information of the synonymous entry pairs and the characteristic information of the nonsense entry pairs, and training the whole neural network model by taking the characteristic information of the synonymous entry pairs and the characteristic information of the nonsense entry pairs as training samples. In the training process, the parameters of the fine tuning layer and the output layer are iterated continuously, and the parameters and the characteristic representation of the pre-training layer are kept unchanged until the model reaches the preset precision.
Note that S301 and S302 may be executed before S304, may be executed before S303, or may be executed after S303.
In the embodiment, when the neural network model is trained, parameters and characteristic representation of the pre-training layer are kept unchanged, so that strong language understanding capability of the pre-training layer is ensured; and training the fine tuning layer and the output layer to enable the fine tuning layer to extract synonymous features from language understanding information and fuse the synonymous features to obtain synonymous feature representation or not, and enabling the output layer to obtain synonymous judging results according to the feature representation.
Example IV
Fig. 4 is a block diagram of a synonym determining device for an entry in a fourth embodiment of the present application, where the embodiment of the present application is applicable to a case where a pair of entries is determined to be synonymous or ambiguous, the device is implemented in software and/or hardware, and is specifically configured in an electronic device having a certain data computing capability.
The synonym determination device 400 of the term shown in fig. 4 comprises: an acquisition module 401 and a discrimination module 402; wherein, the liquid crystal display device comprises a liquid crystal display device,
the obtaining module 401 is configured to obtain feature information of an entry pair to be subjected to synonym discrimination.
The judging module 402 is configured to input feature information of the term pair into the trained neural network model, and obtain a synonymous judging result of the term pair;
The neural network model comprises a pre-training layer, a fine-tuning layer and an output layer; the pre-training layer is used for training a language understanding task by adopting a natural language training sample in advance and is used for learning the characteristic information of the entry pairs to obtain the language understanding information of the entry pairs; the fine tuning layer is used for extracting and fusing the characteristics of the language understanding information to obtain the characteristic representation of whether the entry pairs are synonymous or not; and the output layer is used for obtaining a synonymous judging result according to the characteristic representation.
According to the embodiment of the application, the feature information of the entry pairs is input into the trained neural network model to carry out synonym discrimination, so that the synonym discrimination is carried out by combining the deep learning function of the neural network model on the feature level, the synonym discrimination efficiency can be improved, the manpower resources are saved, and the landing of related intelligent projects is accelerated; in the embodiment, the pre-training layer is obtained by training a language understanding task by adopting a natural language training sample in advance, and the natural language training sample and the vocabulary entry pair have the same characteristic distribution, so that the structure, the parameters and the characteristics of the pre-training layer are directly subjected to knowledge migration, the feature information of the vocabulary entry pair is learned by adopting the pre-training layer, the labeling amount in the training process is reduced, a large amount of resources and manpower are saved, and the synonymous distinguishing efficiency is further improved; by adopting the language understanding information, whether the entry pair is synonymous or not can be accurately reflected, the feature extraction and fusion are carried out on the language understanding information through the fine adjustment layer, the accurate feature representation of whether the entry pair is synonymous or not is obtained, the synonymous judging result is further obtained through the output layer according to the accurate feature representation, and the accuracy of the synonymous judging result is improved.
Further, the pre-training layer is a neural network structure of multiple language understanding tasks; the multilingual understanding task includes a lexical level task, a grammar level task, and a semantic level task, and the language understanding information includes lexical information, grammar information, and semantic information.
Further, the trimming layer includes: a convolution layer, a pooling layer and a full connection layer; the convolution layer is used for extracting features of language understanding information, the pooling layer is used for reducing dimensions of the extracted features, and the full-connection layer is used for fusing the features subjected to dimension reduction to obtain feature representation of whether the entry pairs are synonymous.
Further, the output layer comprises a classification layer for performing synonymy and heterology classification on the feature representation of the synonymy or not by the entry to obtain the confidence of the synonymy type and the heterology type.
Further, the output layer further comprises an intervention layer for judging whether the characteristic information of the entry pair meets a set rule or a synonym rule; if the characteristic information of the entry pair meets the set dissimilarity rule, the confidence of the synonymous type is reduced and/or the confidence of the dissimilarity type is improved; if the characteristic information of the entry pair meets the set synonym rule, the confidence of the synonym type is improved and/or the confidence of the heteronym type is reduced; wherein, the synonymous rule includes: the first template of the synonym entry pair, the first template comprising a first fixed vocabulary and a plurality of first candidate vocabularies, the heterology rule comprising: the second template of the pair of the heteronyms comprises a second fixed vocabulary and a plurality of second candidate vocabularies.
Further, the obtaining module 401 is specifically configured to perform at least one of the following operations: acquiring plain text characteristic information of entry pairs to be subjected to synonymous judgment; acquiring natural word segmentation characteristic information of entry pairs to be subjected to synonymous judgment; acquiring part characteristic information of entry pairs to be subjected to synonymous discrimination; acquiring degree characteristic information of entry pairs to be subjected to synonymous judgment; obtaining directional characteristic information of entry pairs to be subjected to synonymous judgment; obtaining frequency characteristic information of entry pairs to be subjected to synonymous judgment; acquiring quantity characteristic information of entry pairs to be subjected to synonymous judgment; and obtaining sensory characteristic information of the entry pairs to be subjected to synonymous judgment.
Further, the device also comprises a training module, which is specifically used for: acquiring a neural network model to be trained, wherein the neural network model to be trained comprises a pre-training layer, a fine-tuning layer to be trained and an output layer to be trained; and training the neural network model to be trained by taking the characteristic information of the synonymous entry pairs as a positive sample and the characteristic information of the ambiguous entry pairs as a negative sample, and keeping the parameters and the characteristic representation of the pre-training layer unchanged in the training process.
Further, the entry pairs are entity names in the medical field; the entity names include entity common names and entity aliases.
The synonym judging device for the vocabulary entries can execute the synonym judging method for the vocabulary entries provided by any embodiment of the application, and has the corresponding functional modules and beneficial effects of executing the synonym judging method for the vocabulary entries.
Example five
According to an embodiment of the present application, the present application also provides an electronic device and a readable storage medium.
As shown in fig. 5, a block diagram of an electronic device implementing the method for determining synonym of an entry according to the embodiment of the present application is shown. Electronic devices are intended to represent various forms of digital computers, such as laptops, desktops, workstations, personal digital assistants, servers, blade servers, mainframes, and other appropriate computers. The electronic device may also represent various forms of mobile devices, such as personal digital processing, cellular telephones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions, are meant to be exemplary only, and are not meant to limit implementations of the applications described and/or claimed herein.
As shown in fig. 5, the electronic device includes: one or more processors 501, memory 502, and interfaces for connecting components, including high-speed interfaces and low-speed interfaces. The various components are interconnected using different buses and may be mounted on a common motherboard or in other manners as desired. The processor may process instructions executing within the electronic device, including instructions stored in or on memory to display graphical information of the GUI on an external input/output device, such as a display device coupled to the interface. In other embodiments, multiple processors and/or multiple buses may be used, if desired, along with multiple memories and multiple memories. Also, multiple electronic devices may be connected, each providing a portion of the necessary operations (e.g., as a server array, a set of blade servers, or a multiprocessor system). One processor 501 is illustrated in fig. 5.
Memory 502 is a non-transitory computer readable storage medium provided by the present application. The memory stores instructions executable by the at least one processor to cause the at least one processor to execute the synonym distinguishing method for the entry. The non-transitory computer-readable storage medium of the present application stores computer instructions for causing a computer to execute the synonym distinction method of the term provided by the present application.
The memory 502 is used as a non-transitory computer readable storage medium, and may be used to store a non-transitory software program, a non-transitory computer executable program, and a module, such as program instructions/modules (e.g., including the acquisition module 401 and the determination module 402 shown in fig. 4) corresponding to a method for synonym determination of terms in the embodiments of the present application. The processor 501 executes various functional applications of the server and data processing, i.e., a method for implementing synonym discrimination of terms in the above-described method embodiments, by running non-transitory software programs, instructions, and modules stored in the memory 502.
Memory 502 may include a storage program area that may store an operating system, at least one application program required for functionality, and a storage data area; the storage data area may store data created by the use of the electronic device implementing the synonym distinction method of the term, and the like. In addition, memory 502 may include high-speed random access memory, and may also include non-transitory memory, such as at least one magnetic disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory 502 may optionally include memory located remotely from processor 501, which may be connected via a network to an electronic device that performs the synonym-distinguishing method of the term. Examples of such networks include, but are not limited to, the internet, intranets, local area networks, mobile communication networks, and combinations thereof.
The electronic device that performs the synonym distinction method of the term may further include: an input device 503 and an output device 504. The processor 501, memory 502, input devices 503 and output devices 504 may be connected by a bus or otherwise, for example in fig. 5.
The input device 503 may receive input numeric or character information and generate key signal inputs related to user settings and function control of the electronic device performing the synonym distinction method of the term, such as input devices for a touch screen, a keypad, a mouse, a track pad, a touch pad, a pointer stick, one or more mouse buttons, a track ball, a joystick, etc. The output devices 504 may include a display device, auxiliary lighting devices (e.g., LEDs), and haptic feedback devices (e.g., vibration motors), among others. The display device may include, but is not limited to, a Liquid Crystal Display (LCD), a Light Emitting Diode (LED) display, and a plasma display. In some implementations, the display device may be a touch screen.
Various implementations of the systems and techniques described here can be realized in digital electronic circuitry, integrated circuitry, application specific ASIC (application specific integrated circuit), computer hardware, firmware, software, and/or combinations thereof. These various embodiments may include: implemented in one or more computer programs, the one or more computer programs may be executed and/or interpreted on a programmable system including at least one programmable processor, which may be a special purpose or general-purpose programmable processor, that may receive data and instructions from, and transmit data and instructions to, a storage system, at least one input device, and at least one output device.
These computing programs (also referred to as programs, software applications, or code) include machine instructions for a programmable processor, and may be implemented in a high-level procedural and/or object-oriented programming language, and/or in assembly/machine language. As used herein, the terms "machine-readable medium" and "computer-readable medium" refer to any computer program product, apparatus, and/or device (e.g., magnetic discs, optical disks, memory, programmable Logic Devices (PLDs)) used to provide machine instructions and/or data to a programmable processor, including a machine-readable medium that receives machine instructions as a machine-readable signal. The term "machine-readable signal" refers to any signal used to provide machine instructions and/or data to a programmable processor.
To provide for interaction with a user, the systems and techniques described here can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to a user; and a keyboard and pointing device (e.g., a mouse or trackball) by which a user can provide input to the computer. Other kinds of devices may also be used to provide for interaction with a user; for example, feedback provided to the user may be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user may be received in any form, including acoustic input, speech input, or tactile input.
The systems and techniques described here can be implemented in a computing system that includes a background component (e.g., as a data server), or that includes a middleware component (e.g., an application server), or that includes a front-end component (e.g., a user computer having a graphical user interface or a web browser through which a user can interact with an implementation of the systems and techniques described here), or any combination of such background, middleware, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: local Area Networks (LANs), wide Area Networks (WANs), the internet, and blockchain networks.
The computer system may include a client and a server. The client and server are typically remote from each other and typically interact through a communication network. The relationship of client and server arises by virtue of computer programs running on the respective computers and having a client-server relationship to each other.
It should be appreciated that various forms of the flows shown above may be used to reorder, add, or delete steps. For example, the steps described in the present application may be performed in parallel, sequentially, or in a different order, provided that the desired results of the disclosed embodiments are achieved, and are not limited herein.
The above embodiments do not limit the scope of the present application. It will be apparent to those skilled in the art that various modifications, combinations, sub-combinations and alternatives are possible, depending on design requirements and other factors. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present application should be included in the scope of the present application.
Claims (11)
1. A synonym distinguishing method of an entry is characterized by comprising the following steps:
acquiring characteristic information of entry pairs to be subjected to synonymous judgment;
inputting the characteristic information of the entry pairs into a trained neural network model to obtain synonymous discrimination results of the entry pairs;
the neural network model comprises a pre-training layer, a fine-tuning layer and an output layer; the pre-training layer performs training of language understanding tasks by adopting natural language training samples in advance and is used for learning the characteristic information of the entry pairs to obtain the language understanding information of the entry pairs; the fine tuning layer is used for carrying out feature extraction and fusion on the language understanding information to obtain feature representation of whether the entry pairs are synonymous or not; the output layer is used for obtaining the synonymous judging result according to the characteristic representation;
Wherein the natural language training sample and the entry pair have the same feature distribution.
2. The method of claim 1, wherein the pre-training layer is a neural network structure of multilingual understanding tasks;
the multilingual understanding task includes a lexical level task, a grammar level task, and a semantic level task, and the language understanding information includes lexical information, grammar information, and semantic information.
3. The method of claim 1, wherein the trim layer comprises: a convolution layer, a pooling layer and a full connection layer;
the convolution layer is used for extracting features of the language understanding information, the pooling layer is used for reducing dimensions of the extracted features, and the full-connection layer is used for fusing the features subjected to the dimension reduction to obtain feature representation of whether the entry pairs are synonymous.
4. The method of claim 1, wherein the output layer includes a classification layer for synonymously and heterologously classifying whether the entry is synonymous for the feature representation, resulting in confidence levels for synonymous types and heterologous types.
5. The method of claim 4, wherein the output layer further comprises an intervention layer for determining whether the characteristic information of the entry pair meets a set rule of dissimilarity or synonym; if the characteristic information of the entry pair meets the set dissimilarity rule, reducing the confidence coefficient of the synonymous type and/or improving the confidence coefficient of the dissimilarity type; if the characteristic information of the entry pair meets the set synonym rule, the confidence of the synonym type is improved and/or the confidence of the heteronym type is reduced;
Wherein the synonym rule includes: a first template of a pair of synonyms, the first template comprising a first fixed vocabulary and a plurality of first candidate vocabularies, the rules comprising: the second template of the pair of the heteronyms comprises a second fixed vocabulary and a plurality of second candidate vocabularies.
6. The method according to claim 1, wherein the obtaining the feature information of the entry pair to be synonymously discriminated includes at least one of the following operations:
acquiring plain text characteristic information of entry pairs to be subjected to synonymous judgment;
acquiring natural word segmentation characteristic information of entry pairs to be subjected to synonymous judgment;
acquiring part characteristic information of entry pairs to be subjected to synonymous discrimination;
acquiring degree characteristic information of entry pairs to be subjected to synonymous judgment;
obtaining directional characteristic information of entry pairs to be subjected to synonymous judgment;
obtaining frequency characteristic information of entry pairs to be subjected to synonymous judgment;
acquiring quantity characteristic information of entry pairs to be subjected to synonymous judgment;
and obtaining sensory characteristic information of the entry pairs to be subjected to synonymous judgment.
7. The method of claim 1, further comprising, before inputting the feature information of the term pair into the trained neural network model to obtain a synonym discrimination result for the term pair:
Acquiring a neural network model to be trained, wherein the neural network model to be trained comprises the pre-training layer, a fine-tuning layer to be trained and an output layer to be trained;
and training the neural network model to be trained by taking the characteristic information of the synonymous entry pairs as a positive sample and the characteristic information of the dissimilarity entry pairs as a negative sample, and keeping the parameters and the characteristic representation of the pre-training layer unchanged in the training process.
8. The method of any one of claims 1-7, wherein the entry pairs are entity names of medical fields;
the entity names include entity common names and entity aliases.
9. A synonym determination device for an entry, comprising:
the acquisition module is used for acquiring characteristic information of entry pairs to be subjected to synonymous judgment;
the judging module is used for inputting the characteristic information of the entry pairs into the trained neural network model to obtain synonymous judging results of the entry pairs;
the neural network model comprises a pre-training layer, a fine-tuning layer and an output layer; the pre-training layer performs training of language understanding tasks by adopting natural language training samples in advance and is used for learning the characteristic information of the entry pairs to obtain the language understanding information of the entry pairs; the fine tuning layer is used for carrying out feature extraction and fusion on the language understanding information to obtain feature representation of whether the entry pairs are synonymous or not; the output layer is used for obtaining the synonymous judging result according to the characteristic representation;
Wherein the natural language training sample and the entry pair have the same feature distribution.
10. An electronic device, comprising:
at least one processor; and
a memory communicatively coupled to the at least one processor; wherein, the liquid crystal display device comprises a liquid crystal display device,
the memory stores instructions executable by the at least one processor to enable the at least one processor to perform a synonym determination for an entry as claimed in any of claims 1-8.
11. A non-transitory computer readable storage medium storing computer instructions for causing a computer to perform a synonym identification method for an entry as claimed in any one of claims 1-8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010190072.2A CN111414750B (en) | 2020-03-18 | 2020-03-18 | Synonym distinguishing method, device, equipment and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010190072.2A CN111414750B (en) | 2020-03-18 | 2020-03-18 | Synonym distinguishing method, device, equipment and storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111414750A CN111414750A (en) | 2020-07-14 |
| CN111414750B true CN111414750B (en) | 2023-08-18 |
Family
ID=71492982
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010190072.2A Active CN111414750B (en) | 2020-03-18 | 2020-03-18 | Synonym distinguishing method, device, equipment and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111414750B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111949764B (en) * | 2020-08-18 | 2021-06-29 | 桂林电子科技大学 | Knowledge graph completion method based on bidirectional attention mechanism |
| CN112183088B (en) * | 2020-09-28 | 2023-11-21 | 云知声智能科技股份有限公司 | Word level determining method, model building method, device and equipment |
| CN116167455B (en) * | 2022-12-27 | 2023-12-22 | 北京百度网讯科技有限公司 | Model training and data deduplication method, device, equipment and storage medium |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US4724523A (en) * | 1985-07-01 | 1988-02-09 | Houghton Mifflin Company | Method and apparatus for the electronic storage and retrieval of expressions and linguistic information |
| CN101777042A (en) * | 2010-01-21 | 2010-07-14 | 西南科技大学 | Neural network and tag library-based statement similarity algorithm |
| CN106339369A (en) * | 2016-08-30 | 2017-01-18 | 广东医科大学 | Method and system for recognizing synonyms of data set |
| CN106844346A (en) * | 2017-02-09 | 2017-06-13 | 北京红马传媒文化发展有限公司 | Short text Semantic Similarity method of discrimination and system based on deep learning model Word2Vec |
| CN107145910A (en) * | 2017-05-08 | 2017-09-08 | 京东方科技集团股份有限公司 | Performance generation system, its training method and the performance generation method of medical image |
| CN107316015A (en) * | 2017-06-19 | 2017-11-03 | 南京邮电大学 | A kind of facial expression recognition method of high accuracy based on depth space-time characteristic |
| CN107797985A (en) * | 2017-09-27 | 2018-03-13 | 百度在线网络技术（北京）有限公司 | Establish synonymous discriminating model and differentiate the method, apparatus of synonymous text |
| EP3528263A1 (en) * | 2018-02-15 | 2019-08-21 | Siemens Healthcare GmbH | Providing a trained virtual tissue atlas and a synthetic image |
| CN110442684A (en) * | 2019-08-14 | 2019-11-12 | 山东大学 | A kind of class case recommended method based on content of text |
| CN110674314A (en) * | 2019-09-27 | 2020-01-10 | 北京百度网讯科技有限公司 | Sentence recognition method and device |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9646266B2 (en) * | 2012-10-22 | 2017-05-09 | University Of Massachusetts | Feature type spectrum technique |
| US20150095017A1 (en) * | 2013-09-27 | 2015-04-02 | Google Inc. | System and method for learning word embeddings using neural language models |
| JP6583686B2 (en) * | 2015-06-17 | 2019-10-02 | パナソニックＩｐマネジメント株式会社 | Semantic information generation method, semantic information generation device, and program |
| JP6301427B1 (en) * | 2016-10-11 | 2018-03-28 | 株式会社日本総合研究所 | Natural language processing apparatus, natural language processing method, and natural language processing program |
-
2020
- 2020-03-18 CN CN202010190072.2A patent/CN111414750B/en active Active
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US4724523A (en) * | 1985-07-01 | 1988-02-09 | Houghton Mifflin Company | Method and apparatus for the electronic storage and retrieval of expressions and linguistic information |
| CN101777042A (en) * | 2010-01-21 | 2010-07-14 | 西南科技大学 | Neural network and tag library-based statement similarity algorithm |
| CN106339369A (en) * | 2016-08-30 | 2017-01-18 | 广东医科大学 | Method and system for recognizing synonyms of data set |
| CN106844346A (en) * | 2017-02-09 | 2017-06-13 | 北京红马传媒文化发展有限公司 | Short text Semantic Similarity method of discrimination and system based on deep learning model Word2Vec |
| CN107145910A (en) * | 2017-05-08 | 2017-09-08 | 京东方科技集团股份有限公司 | Performance generation system, its training method and the performance generation method of medical image |
| CN107316015A (en) * | 2017-06-19 | 2017-11-03 | 南京邮电大学 | A kind of facial expression recognition method of high accuracy based on depth space-time characteristic |
| CN107797985A (en) * | 2017-09-27 | 2018-03-13 | 百度在线网络技术（北京）有限公司 | Establish synonymous discriminating model and differentiate the method, apparatus of synonymous text |
| EP3528263A1 (en) * | 2018-02-15 | 2019-08-21 | Siemens Healthcare GmbH | Providing a trained virtual tissue atlas and a synthetic image |
| CN110442684A (en) * | 2019-08-14 | 2019-11-12 | 山东大学 | A kind of class case recommended method based on content of text |
| CN110674314A (en) * | 2019-09-27 | 2020-01-10 | 北京百度网讯科技有限公司 | Sentence recognition method and device |
Non-Patent Citations (1)
| Title |
|---|
| 面向知识库问答的自然语言语义特征表示研究;王智昊;《万方数据库》;全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111414750A (en) | 2020-07-14 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112560912B (en) | Classification model training method and device, electronic equipment and storage medium | |
| CN111967268B (en) | Event extraction method and device in text, electronic equipment and storage medium | |
| CN111414750B (en) | Synonym distinguishing method, device, equipment and storage medium | |
| EP3859562A2 (en) | Method, apparatus, electronic device, storage medium and computer program product for generating information | |
| EP3851975A1 (en) | Method and apparatus for generating text topics, and electronic device | |
| CN111079442B (en) | Vectorization representation method and device of document and computer equipment | |
| EP3923159A1 (en) | Method, apparatus, device and storage medium for matching semantics | |
| US11907671B2 (en) | Role labeling method, electronic device and storage medium | |
| US20210200963A1 (en) | Machine translation model training method, apparatus, electronic device and storage medium | |
| WO2022095563A1 (en) | Text error correction adaptation method and apparatus, and electronic device, and storage medium | |
| CN111177355B (en) | Man-machine conversation interaction method and device based on search data and electronic equipment | |
| KR102456535B1 (en) | Medical fact verification method and apparatus, electronic device, and storage medium and program | |
| US11507751B2 (en) | Comment information processing method and apparatus, and medium | |
| CN111078878B (en) | Text processing method, device, equipment and computer readable storage medium | |
| CN112489637A (en) | Speech recognition method and device | |
| CN110705460A (en) | Image category identification method and device | |
| US20220129448A1 (en) | Intelligent dialogue method and apparatus, and storage medium | |
| CN111522967A (en) | Knowledge graph construction method, device, equipment and storage medium | |
| CN110717340B (en) | Recommendation method, recommendation device, electronic equipment and storage medium | |
| CN111738015B (en) | Article emotion polarity analysis method and device, electronic equipment and storage medium | |
| CN111783861A (en) | Data classification method, model training device and electronic equipment | |
| CN111538815A (en) | Text query method, device, equipment and storage medium | |
| CN114444462B (en) | Model training method and man-machine interaction method and device | |
| CN112270169B (en) | Method and device for predicting dialogue roles, electronic equipment and storage medium | |
| CN111241302B (en) | Position information map generation method, device, equipment and 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 |