CN112069818A - Triple prediction model generation method, relation triple extraction method and device - Google Patents
Triple prediction model generation method, relation triple extraction method and device Download PDFInfo
- Publication number
- CN112069818A CN112069818A CN202010785316.1A CN202010785316A CN112069818A CN 112069818 A CN112069818 A CN 112069818A CN 202010785316 A CN202010785316 A CN 202010785316A CN 112069818 A CN112069818 A CN 112069818A
- Authority
- CN
- China
- Prior art keywords
- relation
- target
- word
- words
- triple
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 57
- 238000000605 extraction Methods 0.000 title abstract description 13
- 238000012549 training Methods 0.000 claims abstract description 152
- 238000010276 construction Methods 0.000 claims description 7
- 238000002372 labelling Methods 0.000 abstract description 36
- 230000006870 function Effects 0.000 description 33
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 16
- 238000010586 diagram Methods 0.000 description 12
- 238000004590 computer program Methods 0.000 description 7
- 230000008569 process Effects 0.000 description 6
- 238000007405 data analysis Methods 0.000 description 5
- 238000012545 processing Methods 0.000 description 5
- 230000009471 action Effects 0.000 description 3
- 230000002457 bidirectional effect Effects 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000004075 alteration Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/20—Natural language analysis
- G06F40/279—Recognition of textual entities
- G06F40/289—Phrasal analysis, e.g. finite state techniques or chunking
- G06F40/295—Named entity recognition
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/33—Querying
- G06F16/3331—Query processing
- G06F16/334—Query execution
- G06F16/3344—Query execution using natural language analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/36—Creation of semantic tools, e.g. ontology or thesauri
- G06F16/367—Ontology
Abstract
The embodiment of the invention provides a triple prediction model generation method, a relation triple extraction method and a relation triple extraction device. The triple prediction model generation method comprises the steps of obtaining a training text containing a preset relation triple; the relation triples comprise main entity words, relation words and auxiliary entity words; the training text has the sub-entity word labeling information and the relation word labeling information corresponding to the relation triple; and taking the training text and the main entity word in the relation triple as a model input, taking the auxiliary entity word labeling information and the relation word labeling information as a model output, training a preset model, and generating a triple prediction model. Therefore, the predicted relation triples can contain various different relation types, the number and the relation types of the relation triples which can be extracted by the triple prediction model are expanded, and the extraction of the relation triples in the open field is realized.
Description
Technical Field
The present invention relates to the field of data processing technologies, and in particular, to a method for generating a triplet prediction model, a method for extracting a relational triplet, a device for generating a triplet prediction model, and a device for extracting a relational triplet.
Background
In the prior art, in order to meet the requirements of performing data analysis, information retrieval, establishing a question-answering system and the like on mass data, a plurality of relationship triples can be obtained, and a knowledge graph is further constructed or mass data analysis is performed based on the relationship triples. However, the existing extracting manner of the relationship triples is generally limited in the fixed domain, so that the number of the relationship triples that can be extracted and the relationship types of the relationship triples are limited, and the subsequent data analysis is affected.
Disclosure of Invention
In view of the above problems, embodiments of the present invention are provided to provide a method for generating a triplet prediction model, a method for extracting a relational triplet, an apparatus for generating a triplet prediction model, and an apparatus for extracting a relational triplet, which overcome or at least partially solve the above problems.
In order to solve the above problem, an embodiment of the present invention discloses a method for generating a triple prediction model, including:
acquiring a training text containing a preset relation triple; the relation triples comprise main entity words, relation words and auxiliary entity words; the training text comprises main entity words, auxiliary entity words and relation words corresponding to the relation triples;
and taking the training text and the main entity word as model input, taking the auxiliary entity word and the relation word as model output, training a preset model, and generating a triple prediction model.
Optionally, the step of obtaining a training text containing a preset relationship triple includes:
searching a candidate text which simultaneously contains a main entity word and an auxiliary entity word in a preset relation triple;
and searching a candidate text containing the relation words in the relation triple or the similar meaning words of the relation words as a training text.
Optionally, the step of training a preset model and generating a triple prediction model by using the training text and the primary entity word in the relation triple as model inputs and the secondary entity word and the relation word as model outputs includes:
taking the training text and the main entity word in the relation triple as model input, taking the auxiliary entity word and the relation word as model output, training a preset model, and acquiring auxiliary entity word prediction information and relation word prediction information output by the model;
determining a link loss function of the model according to the auxiliary entity word prediction information and the relation word prediction information; the link loss function is used for representing whether the auxiliary entity word prediction information and the relation word prediction information with the corresponding relation appear in pairs or not;
and adjusting model parameters of the model according to the link loss function, and training the model in an iterative manner until the link loss function reaches a preset condition, finishing the model training, and taking the model as a triple prediction model.
Optionally, the method further comprises:
extracting high-frequency relation words with the occurrence frequency higher than the predicted frequency from a preset knowledge graph;
and determining the similar meaning words corresponding to the high-frequency relation words.
The embodiment of the invention also discloses a method for extracting the relation triple, which comprises the following steps:
searching a text to be mined containing a target main entity word;
determining target auxiliary entity words and target relation words by adopting the target main entity words, the text to be mined and a preset triple prediction model; the triple prediction model acquires a training text containing a preset relation triple; taking the training text and the main entity word in the relation triple as model input, taking the auxiliary entity word and the relation word as model output, and training a preset model to generate;
and constructing a target relation triple comprising the target main entity word, the target relation word and the target auxiliary entity word.
Optionally, the step of constructing a target relationship triple including the target main entity word, the target relation word, and the target sub-entity word includes:
determining the positions of target auxiliary entity words and target relation words in the text to be mined;
and constructing a target relation triple by adopting the target main entity word, the target auxiliary entity word and the target relation word which are adjacent in position in the text to be mined.
Optionally, the step of constructing a target relationship triple by using the target primary entity word and the target secondary entity word and the target relationship word which are adjacent to each other in the text to be mined includes:
if the target subordinate entity words have at least two target relation words adjacent in position in the text to be mined, taking the target relation words with the least number of spaced characters between the target subordinate entity words as the target relation words corresponding to the target subordinate entity words;
and constructing a target relation triple by adopting the target main entity word, the target auxiliary entity word and the target relation word corresponding to the target auxiliary entity word.
The embodiment of the invention also discloses a generating device of the triple prediction model, which comprises the following steps:
the acquisition module is used for acquiring a training text containing a preset relation triple; the relation triples comprise main entity words, relation words and auxiliary entity words; the training text comprises main entity words, auxiliary entity words and relation words corresponding to the relation triples;
and the training module is used for inputting the training text and the main entity word as models, outputting the auxiliary entity word and the relation word as models, training a preset model and generating a triple prediction model.
Optionally, the obtaining module includes:
the candidate text searching sub-module is used for searching a candidate text which simultaneously contains the main entity words and the auxiliary entity words in the relation triple for a preset relation triple;
and the training text searching sub-module is used for searching a candidate text containing the relation words in the relation triple or the similar meaning words of the relation words as a training text.
Optionally, the training module comprises:
the prediction information acquisition sub-module is used for inputting the training text and the main entity words in the relation triples as models, outputting the auxiliary entity words and the relation words as models, training a preset model, and acquiring auxiliary entity word prediction information and relation word prediction information output by the models;
the link loss function determining submodule is used for determining a link loss function of the model according to the auxiliary entity word prediction information and the relation word prediction information; the link loss function is used for representing whether the auxiliary entity word prediction information and the relation word prediction information with the corresponding relation appear in pairs or not;
and the training submodule is used for adjusting the model parameters of the model according to the link loss function and training the model in an iterative manner until the link loss function reaches a preset condition, the model training is finished, and the model is used as a triple prediction model.
The embodiment of the invention also discloses a device for extracting the relation triple, which comprises the following components:
the searching module is used for searching a text to be mined, which contains the target main entity words;
the input module is used for determining target auxiliary entity words and target relation words by adopting the target main entity words, the text to be mined and a preset triple prediction model; the triple prediction model acquires a training text containing a preset relation triple; taking the training text and the main entity word in the relation triple as model input, taking the auxiliary entity word and the relation word as model output, and training a preset model to generate;
and the construction module is used for constructing a target relation triple comprising the target main entity word, the target relation word and the target auxiliary entity word.
Optionally, the building module comprises:
the position determining submodule is used for determining the positions of the target auxiliary entity words and the target relation words in the text to be mined;
and the construction sub-module is used for constructing a target relation triple by adopting the target main entity word, the target auxiliary entity word and the target relation word which are adjacent in position in the text to be mined.
Optionally, the building submodule comprises:
a target relation word determining unit, configured to, if the target subordinate entity word has at least two target relation words adjacent in position in the text to be mined, take the target relation word with the smallest number of characters spaced apart from the target subordinate entity word as the target relation word corresponding to the target subordinate entity word;
and the construction unit is used for constructing a target relation triple by adopting the main entity word, the target auxiliary entity word and the target relation word corresponding to the target auxiliary entity word.
The embodiment of the invention also discloses a device, which comprises:
one or more processors; and
one or more machine-readable media having instructions stored thereon, which when executed by the one or more processors, cause the apparatus to perform one or more methods as described in embodiments of the invention.
Embodiments of the invention also disclose one or more machine-readable media having instructions stored thereon, which when executed by one or more processors, cause the processors to perform one or more methods as described in embodiments of the invention.
The embodiment of the invention has the following advantages:
according to the method for generating the triple prediction model, provided by the embodiment of the invention, the training text containing the preset relation triple is obtained, the training text and the main entity word in the relation triple are used as model input, the auxiliary entity word and the relation word are used as model output, the preset model is trained, and the triple prediction model is generated. Therefore, the trained triple prediction model can predict the secondary entity words and the relation words of the training text based on the training text and the main entity words, so that the predicted relation triples can contain various different relation types, the quantity and the relation types of the relation triples which can be extracted by the triple prediction model are expanded, and the extraction of the relation triples in the open field is realized.
Drawings
FIG. 1 is a flowchart illustrating steps of a method for generating a triple prediction model according to an embodiment of the present invention;
FIG. 2 is a flow chart illustrating steps in a method for generating a triple predictive model according to an embodiment of the present invention;
FIG. 3 is a flowchart illustrating steps of a method for extracting a relational triple according to an embodiment of the present invention;
FIG. 4 is a flow chart of steps of another method for extracting relational triples, in accordance with an embodiment of the present invention;
fig. 5 is a block diagram of a triple prediction model generation apparatus according to an embodiment of the present invention;
fig. 6 is a block diagram of a structure of an apparatus for extracting a relational triple according to an embodiment of the present invention.
Detailed Description
In order to make the aforementioned objects, features and advantages of the present invention comprehensible, embodiments accompanied with figures are described in further detail below.
Referring to fig. 1, a flowchart illustrating steps of an embodiment of a method for generating a triple prediction model according to the present invention is shown, which may specifically include the following steps:
in the embodiment of the present invention, a plurality of relationship triples may be preset, where the relationship triples include a main entity word, a sub entity word, and a relationship word indicating a relationship between the main entity word and the sub entity word. For example, in the relationship triple "zhangao, zhuang, a company", zhangao "may be a main entity word, a" company "may be a side entity word, and" zhuang "may be a relationship between the main entity word" zhangao "and the side entity word" zhangao ". The entity words may refer to words having specific meanings in the text, such as names of people, places, organizations, proper nouns, and the like.
In the embodiment of the invention, in order to obtain the triple prediction model capable of labeling the triples in the text, the training text containing the main entity words, the auxiliary entity words and the relation words corresponding to the preset relation triples can be obtained, so that the model can be conveniently trained by adopting the training text subsequently.
And 102, inputting the training text and the main entity word as models, outputting the auxiliary entity word and the relation word as models, training a preset model, and generating a triple prediction model.
In the embodiment of the present invention, in order to implement extraction of a relationship triplet in an open field without limiting the extraction of the relationship triplet to a specified relationship type, the secondary entity word and the relationship word may be output as models, and the training text and the primary entity word in the relationship triplet may be input as models, a preset model is trained, and a triplet prediction model is generated.
In particular implementations, the triple prediction model may include a sequence annotation model. The main entity words in the training text can be labeled based on the training text and the main entity words, and the training text with the main entity word labeling information is obtained. And then, labeling the subordinate entity words and the relation words in the training text to obtain the training text with the labeling information of the subordinate entity words and the labeling information of the relation words. And then, training texts with main entity word labeling information can be used as model input, training texts with auxiliary entity word labeling information and relation word labeling information are used as model output, and the sequence labeling model in the triple prediction model is trained.
The main entity word labeling information may be used to obtain an entity word belonging to the main entity word type in the training text and obtain a position of the entity word in the training text. The sub-entity word labeling information can be used for obtaining the entity words belonging to the sub-entity word type in the training text and obtaining the positions of the entity words in the training text. The relation word labeling information can be used for obtaining the entity words belonging to the relation word types in the training text and obtaining the positions of the entity words in the training text.
The trained triple prediction model can label the main entity word in the text based on the text and the main entity word to obtain the text with the label information of the main entity word, and obtain the label information of the auxiliary entity word and the label information of the relation word based on the text with the label information of the main entity word. And then, obtaining the auxiliary entity words and the relation words in the text based on the auxiliary entity word tagging information and the relation word tagging information, and finally outputting the relation triples containing the main entity words, the relation words and the auxiliary entity words.
Therefore, the trained triple prediction model can predict the relation words and the auxiliary entity words in the text based on the main entity words, so that the relation triples predicted by the triple prediction model can not be limited to a plurality of specified relation types, but can predict new relation types, more different types of relation triples can be obtained, the extraction of the relation triples in the open field is realized, the accuracy is high, and the knowledge graph or mass data analysis can be conveniently constructed by adopting the relation triples subsequently.
The preset model may be a hidden markov model, a conditional random field model, a bidirectional long and short memory network (BiLSTM) model, or other models for sequence labeling, which is not limited in the present invention.
Through the generation method of the triple prediction model provided by the embodiment of the invention, the training text containing the preset relation triple is obtained; the relation triples comprise main entity words, relation words and auxiliary entity words; the training text comprises main entity words, auxiliary entity words and relation words corresponding to the relation triples; and taking the training text and the main entity word as model input, taking the auxiliary entity word and the relation word as model output, training a preset model, and generating a triple prediction model. Therefore, the trained triple prediction model can predict the secondary entity words and the relation words of the training text based on the training text and the main entity words, so that the predicted relation triples can contain various different relation types, the quantity and the relation types of the relation triples which can be extracted by the triple prediction model are expanded, and the extraction of the relation triples in the open field is realized.
Referring to fig. 2, a flowchart illustrating steps of an embodiment of a method for generating a triple prediction model according to the present invention is shown, and specifically, the method may include the following steps:
in the embodiment of the present invention, a plurality of relationship triples may be preset, where the relationship triples include a main entity word, a sub entity word, and a relationship word indicating a relationship between the main entity word and the sub entity word.
In the embodiment of the present invention, in order to train a triple prediction model that can label triples in a text, a training text including a preset relationship triple may be obtained. In order to increase the number of extracted training texts, for a preset relationship triple, candidate texts containing main entity words and auxiliary entity words in the relationship triple may be first searched. And then screening the candidate texts, and further searching training texts in the candidate texts.
In one embodiment of the invention, the method further comprises:
s11, extracting high-frequency relation words with the occurrence frequency higher than the preset frequency from a preset knowledge graph;
in the embodiment of the invention, in order to improve the training effect of the triple prediction model, high-frequency relation words with the occurrence frequency higher than the preset times can be extracted from the predicted knowledge graph.
The knowledge graph can be a database which is composed of massive data and contains a plurality of relation triples.
The preset times may be determined as 3 times, 5 times, 50 times, and the like according to actual needs, which is not limited in the present invention.
And S12, taking the relation triplets containing the high-frequency relation words as training relation triplets.
In the embodiment of the present invention, the relationship triplet including the high-frequency keyword may be used as a training relationship triplet for training the triplet prediction model. And training the model by adopting the training relation ternary group containing the high-frequency relation words, so that the model can be converged better, and a better training effect is obtained.
in the embodiment of the invention, in order to make the triple prediction model obtained after training available for labeling a plurality of relationship triples of different relationship types, the relationship term may be expanded during the training process. Therefore, the relation words in the relation triple can be extracted, the relation words are subjected to near meaning word expansion, and the near meaning words corresponding to the relation words are determined.
In the embodiment of the present invention, for a relationship triple, a relationship word including the relationship triple or a candidate text including a near-synonym corresponding to the relationship word of the relationship triple may be searched in a training text including a main entity word and a sub-entity word in the relationship triple, and the candidate text is used as a training text.
in the embodiment of the present invention, in order to implement extraction of a relationship triple in an open field without limiting the extraction of the relationship triple in a specified relationship type, the secondary entity words and the relationship words may be output as models, and the training text and the primary entity words in the relationship triple may be input as models to train a preset model. In the training process, the model can output assistant entity word prediction information of the model for predicting the assistant entity words and the positions of the assistant entity words, and relation word prediction information of the model for predicting the relation words and the positions of the relation words. Whether the model is predicted to be completed may be further determined based on the subordinate entity word prediction information and the relation word prediction information.
The preset model may be a hidden markov model, a conditional random field model, a bidirectional long and short memory network (BiLSTM) model, or other models for sequence labeling, which is not limited in the present invention.
in the embodiment of the present invention, since the subordinate entity words and the relation words in the training text need to appear in pairs, the main entity words, the relation words and the subordinate entity words may be used to form the relation triples. Moreover, a plurality of groups of subordinate entity words and relation words with corresponding relations can exist in the training text. For example, in the training text "lugu lake altitude 2690 meters and water area up to 58 square kilometers", the primary entity word may be "lugu lake", the secondary entity word having a correspondence relationship and the relationship word may include the relationship word "altitude" and the corresponding secondary entity word "2690 meters", and the relationship word "water area" and the corresponding secondary entity word "58 square kilometers". In this case, if the model only marks out one of the secondary entity word and the relation word having the corresponding relation, it may cause that there may be an error in the relation triple generated based on the model marking result. If the model labels the auxiliary entity words and/or the relation words incorrectly, errors may also exist in the relation triples generated based on the labeling result of the model.
Thus, the link loss function of the model can be determined based on the subordinate entity word prediction information and the relation word prediction information output by the model. The link loss function may be used to characterize whether the subordinate entity word prediction information and the relation word prediction information having a corresponding relationship occur in pairs.
If the subordinate entity word prediction information and the relation word prediction information with the corresponding relationship appear in pairs, the model can be considered to be correct, and the subordinate entity words and the relation words in the same relation triple are labeled at the same time, and at the moment, the link loss function can obtain a better result. If the sub-entity word prediction information and the relation word prediction information having the corresponding relationship do not appear in pairs, it may be considered that the model incorrectly labels the sub-entity word and/or the relation word in the same relation triple, or the model does not label the sub-entity word and the relation word in the same relation triple at the same time, and at this time, the link loss function may obtain a poor result.
In the embodiment of the present invention, whether the model reaches convergence may be evaluated by using the link loss function. Under the condition that the model does not reach convergence, parameters of the model can be adjusted according to the link loss function, the model is iteratively trained until the link loss function reaches a preset condition, the model can be considered to reach convergence, the model training is finished, and the model is used as a triple prediction model.
The preset condition may be determined according to actual needs, for example, iterative training is performed until the link loss function reaches an optimum. For another example, in the iterative training of the preset number of times, the model with the optimal link loss function is selected as the trained model, and the like, which is not limited in the present invention.
In the embodiment of the invention, the trained triple prediction model can predict the relation words and the auxiliary entity words in the text based on the main entity words, so that the relation triples predicted by the triple prediction model can not be limited in a plurality of specified relation types, but can predict new relation types, obtain more different types of relation triples, have higher accuracy and facilitate the follow-up construction of a knowledge graph or mass data analysis by adopting the relation triples.
According to the method for generating the triple prediction model, the candidate text which simultaneously contains the main entity word and the auxiliary entity word in the relation triple is searched for the preset relation triple; searching a candidate text containing the relation words in the relation triple or the similar meaning words of the relation words as a training text; taking the training text and the main entity word in the relation triple as model input, taking the auxiliary entity word and the relation word as model output, and training a preset model; determining a link loss function of the model according to the auxiliary entity word prediction information and the relation word prediction information, adjusting model parameters of the model, iteratively training the model until the link loss function reaches a preset condition, finishing the model training, and taking the model as a triple prediction model. Therefore, the trained triple prediction model can predict the secondary entity words and the relation words of the training text based on the training text and the main entity words, so that the predicted relation triples can contain various different relation types, the quantity and the relation types of the relation triples which can be extracted by the triple prediction model are expanded, and the extraction of the relation triples in the open field is realized.
Referring to fig. 3, a flowchart illustrating steps of an embodiment of a method for extracting a relationship triple according to the present invention is shown, and specifically, the method may include the following steps:
in the embodiment of the invention, under the condition that the relation triple associated with a target main entity word needs to be mined from mass data, a text to be mined containing the target main entity word can be searched in mass data such as internet data, document data and the like.
in the embodiment of the present invention, in order to search for the target subordinate entity word and the target relation word corresponding to the target main entity word in the text to be mined, the target subordinate entity word and the target relation word may be determined by using the target main entity word, the text to be mined, and a preset triple prediction model.
Specifically, the main entity words and the text to be mined may be input into a preset triple prediction model, the triple prediction model may label the main entity words in the text to be mined to obtain the text to be mined with the main entity word labeling information, the triple prediction model may label the auxiliary entity words and the relation words in the text to be mined based on the text to be mined with the main entity word labeling information to obtain target auxiliary entity labeling information and target relation word labeling information output by the triple prediction model, and determine the target auxiliary entity words and the target relation words based on the target auxiliary entity labeling information and the target relation word labeling information.
The triple prediction model acquires a training text containing a preset relation triple; and taking the training text and the main entity word in the relation triple as model input, taking the auxiliary entity word and the relation word as model output, and training a preset model to generate.
In the embodiment of the invention, the target relationship triplets can be constructed according to the target main entity words, the target auxiliary entity words output by the triple prediction model and the target relation words, so that the target relationship triplets related to the main entity words can be extracted from mass data in the open field, and the target relationship triplets have high accuracy and can reach more than 70%. And subsequently, a knowledge graph of the main entity word can be constructed by adopting the target relation triple, and mass data associated with the main entity word can be analyzed.
By the method for extracting the relation triple, the text to be mined, which contains the target main entity word, is searched; determining target auxiliary entity words and target relation words by adopting the target main entity words, the text to be mined and a preset triple prediction model; the triple prediction model acquires a training text containing a preset relation triple; taking the training text and the main entity word in the relation triple as model input, taking the auxiliary entity word and the relation word as model output, and training a preset model to generate; and constructing a target relation triple comprising the target main entity word, the target relation word and the target auxiliary entity word. Therefore, the target relation triples associated with the main entity words can be obtained by adopting the triple prediction model, the knowledge graph of the main entity words can be conveniently constructed by adopting the target relation triples subsequently, and mass data associated with the main entity words can be analyzed.
Referring to fig. 4, a flowchart illustrating steps of an embodiment of a method for extracting a relationship triple according to the present invention is shown, and specifically, the method may include the following steps:
in the embodiment of the invention, under the condition that the relation triple associated with a target main entity word needs to be mined from mass data, a text to be mined containing the target main entity word can be searched in mass data such as internet data, document data and the like.
in the embodiment of the present invention, in order to search for the target subordinate entity word and the target relation word corresponding to the target main entity word in the text to be mined, the target subordinate entity word and the target relation word may be determined by using the target main entity word, the text to be mined, and a preset triple prediction model.
Specifically, the main entity words and the text to be mined may be input into a preset triple prediction model, the triple prediction model may label the main entity words in the text to be mined to obtain the text to be mined with the main entity word labeling information, the triple prediction model may label the auxiliary entity words and the relation words in the text to be mined based on the text to be mined with the main entity word labeling information to obtain target auxiliary entity labeling information and target relation word labeling information output by the triple prediction model, and determine the target auxiliary entity words and the target relation words based on the target auxiliary entity labeling information and the target relation word labeling information.
The triple prediction model acquires a training text containing a preset relation triple; and taking the training text and the main entity word in the relation triple as model input, taking the auxiliary entity word and the relation word as model output, and training a preset model to generate.
in the embodiment of the present invention, the positions of the target subordinate entity words and the target relation words in the text to be mined may be determined based on the output result of the triple prediction model.
Specifically, the triple prediction model may label the target sub-entity word on the text to be mined to obtain target sub-entity word label information. Therefore, the target auxiliary entity words and the positions of the target auxiliary entity words in the text to be mined can be obtained according to the target auxiliary entity word labeling information. The triple prediction model can label the target relation words on the text to be mined to obtain labeling information of the target relation words. Therefore, the target relation words and the positions of the target relation words in the text to be mined can be obtained according to the target relation words.
As an example of the present invention, the text to be mined may be "lugu lake called" mother lake "by morgan, the altitude is 2690 m, the area of the water is 58 square kilometers, and the average water depth is 45 m", the primary entity word may be "lugu lake", the triple prediction model may output target secondary entity word tagging information, tag "2690 m", "58 square kilometers", and "45 m" in the text to be mined as secondary entity words, and output target relation word tagging information, and tag "altitude", "area of the water, and" average water depth "in the text to be mined as target relation words. Therefore, the positions of the target subordinate entity words and the target relation words in the text to be mined can be obtained according to the target subordinate entity word labeling information and the target relation word labeling information.
In the embodiment of the present invention, each relationship triple includes a main entity word, a relationship word, and a sub entity word having a corresponding relationship. Therefore, under the condition that the triple prediction model marks a plurality of target relation words and a plurality of target subordinate entity words in the text to be mined, the target subordinate entity words and the target relation words with corresponding relations need to be determined, and the target relation triples are constructed with the target main entity words.
In the embodiment of the present invention, it may be considered that, in the text to be mined, if the target subordinate entity word is adjacent to the target relation word, it may be considered that the target subordinate entity word and the target relation word have a corresponding relationship and belong to the same target relation triple. Therefore, the target relation triple can be constructed by adopting the target main entity word, the target auxiliary entity word and the target relation word which are adjacent in position in the text to be mined.
Specifically, the position of the target subordinate entity word is adjacent to the target relation word, which may mean that a first target relation word is found forward or backward in the text to be mined based on the position of the target subordinate entity word, and there is no other target subordinate entity word between the target subordinate entity word and the target relation word. The target subordinate entity words and the target relation words which are adjacent in position may not be separated by characters, or may be separated by at least one character, which is not limited in the present invention.
As an example of the present invention, the text to be mined may be "lugu lake called" mother lake "by morgan, with an altitude of 2690 meters, a water area of 58 square kilometers, and an average water depth of 45 meters," 2690 meters "," 58 square kilometers ", and" 45 meters "as subordinate entity words, and" altitude "," water area ", and" average water depth "as target relation words. Wherein the target relation word "altitude" is adjacent to the target subordinate entity word "2690 m", the target relation word "water area" is adjacent to the target subordinate entity word "58 square kilometer", the target relation word "average water depth" is adjacent to the target subordinate entity word "45 m", the target relation word "altitude" and the target subordinate entity word "2690 m" can be considered to have a corresponding relationship, the target relation word "water area" and the target subordinate entity word "58 square kilometer" have a corresponding relationship, the target relation word "average water depth" and the target subordinate entity word "45 m" have a corresponding relationship, the target relation triple is constructed by using the main entity word and the target subordinate entity word and the target relation word adjacent to each other in the position in the text to be mined, a target relation triple (luzhou lake, altitude, 2690 m), (staku lake, water area) is obtained, 58 square kilometers), and (lugu lake, average water depth, 45 meters).
In an embodiment of the present invention, the step of constructing a target relationship triple by using the target primary entity word, and the target secondary entity word and the target relationship word that are adjacent to each other in the text to be mined includes:
s11, if the target sub-entity words have at least two target relation words adjacent in position in the text to be mined, taking the target relation words with the minimum number of spaced characters between the target sub-entity words as the target relation words corresponding to the target sub-entity words;
in the embodiment of the present invention, when the text to be mined finds a plurality of target subordinate entity words and a plurality of target relation words, for a target subordinate entity word, at least two target relation words adjacent to each other in position may be found in the text to be mined. At this time, it is necessary to further determine a target relation word having a corresponding relationship with the target subordinate entity word.
In the embodiment of the present invention, it may be considered that, in the text to be mined, the smaller the number of interval characters between the target relation word and the target subordinate entity word is, the more likely the target relation word has a correspondence relationship with the target subordinate entity word. Therefore, if the target subordinate entity words have at least two target relation words adjacent to each other in position in the text to be mined, the target relation words with the least number of characters spaced apart from the target subordinate entity words can be used as the target relation words corresponding to the target subordinate entity words.
Alternatively, there may be a same number of characters spaced between the target subordinate entity words and at least two of the target relation words located adjacent to each other. In this case, it may be considered that, in the text to be mined, a target relation word adjacent to and before the target subordinate entity word is more likely to be a target relation word corresponding to the target subordinate entity word, so that the target relation word adjacent to and before the target subordinate entity word is considered as the target relation word corresponding to the target subordinate entity word.
Optionally, each text segment to be mined input into the triple prediction model may include one sentence of text or may include multiple sentences of text. Each sentence of text may or may not include at least one punctuation mark. Therefore, when the target sub-entity words respectively have the same number of interval characters with at least two adjacent target relation words, the target relation words corresponding to the target sub-entity words can be determined based on whether the interval characters contain punctuation marks and the types of the punctuation marks.
For example, it may be considered that, of the target relation words whose interval characters with the target subordinate entity words do not include punctuation marks and the target relation words whose interval characters with the target subordinate entity words include punctuation marks, the target relation words whose interval characters with the target subordinate entity words do not include punctuation marks are more likely to be the target relation words corresponding to the target subordinate entity words. It may be considered that, of the target relation words whose interval character between the target subordinate entity words includes a comma, and the target relation words whose interval character between the target subordinate entity words includes a period, the target relation words whose interval character between the target subordinate entity words includes a comma may be the target relation words corresponding to the target subordinate entity words.
And S12, constructing a target relation triple by adopting the target main entity word, the target auxiliary entity word and the target relation word corresponding to the target auxiliary entity word.
In the embodiment of the present invention, after the target relation word corresponding to the target secondary entity word is determined, the target relation triple may be constructed by using the main entity word, the target secondary entity word, and the target relation word corresponding to the target secondary entity word. Therefore, the target relation triple associated with the main entity word can be extracted from the mass data, the target relation triple is convenient to be subsequently adopted to construct the knowledge map of the main entity word, and the mass data associated with the main entity word can be analyzed.
By the method for extracting the relation triples, the text to be mined, which contains the main entity words, is searched; determining target auxiliary entity words and target relation words by adopting the target main entity words, the text to be mined and a preset triple prediction model; determining the positions of target auxiliary entity words and target relation words in the text to be mined; and constructing a target relation triple by adopting the main entity word, the target auxiliary entity word and the target relation word which are adjacent in position in the text to be mined. Therefore, the target relation triples associated with the main entity words can be obtained by adopting the triple prediction model, the knowledge graph of the main entity words can be conveniently constructed by adopting the target relation triples subsequently, and mass data associated with the main entity words can be analyzed.
It should be noted that, for simplicity of description, the method embodiments are described as a series of acts or combination of acts, but those skilled in the art will recognize that the present invention is not limited by the illustrated order of acts, as some steps may occur in other orders or concurrently in accordance with the embodiments of the present invention. Further, those skilled in the art will appreciate that the embodiments described in the specification are presently preferred and that no particular act is required to implement the invention.
Referring to fig. 5, a block diagram of a structure of an embodiment of an apparatus for generating a triple prediction model according to the present invention is shown, and the apparatus specifically includes the following modules:
an obtaining module 501, configured to obtain a training text including a preset relationship triple; the relation triples comprise main entity words, relation words and auxiliary entity words; the training text comprises main entity words, auxiliary entity words and relation words corresponding to the relation triples;
a training module 502, configured to take the training text and the primary entity word as model inputs, take the secondary entity word and the relation word as model outputs, train a preset model, and generate a triple prediction model.
In an embodiment of the present invention, the obtaining module 501 includes:
the candidate text searching sub-module is used for searching a candidate text which simultaneously contains the main entity words and the auxiliary entity words in the relation triple for a preset relation triple;
and the training text searching sub-module is used for searching a candidate text containing the relation words in the relation triple or the similar meaning words of the relation words as a training text.
In one embodiment of the present invention, the training module 502 comprises:
the prediction information acquisition sub-module is used for inputting the training text and the main entity words in the relation triples as models, outputting the auxiliary entity words and the relation words as models, training a preset model, and acquiring auxiliary entity word prediction information and relation word prediction information output by the models;
the link loss function determining submodule is used for determining a link loss function of the model according to the auxiliary entity word prediction information and the relation word prediction information; the link loss function is used for representing whether the auxiliary entity word prediction information and the relation word prediction information with the corresponding relation appear in pairs or not;
and the training submodule is used for adjusting the model parameters of the model according to the link loss function and training the model in an iterative manner until the link loss function reaches a preset condition, the model training is finished, and the model is used as a triple prediction model.
Referring to fig. 6, a block diagram of a structure of an embodiment of the apparatus for extracting a relationship triple in the present invention is shown, which may specifically include the following modules:
the searching module 601 is used for searching a text to be mined, which contains the target main entity words;
an input module 602, configured to determine a target subordinate entity word and a target relation word by using the target main entity word, the text to be mined, and a preset triple prediction model; the triple prediction model acquires a training text containing a preset relation triple; taking the training text and the main entity word in the relation triple as model input, taking the auxiliary entity word and the relation word as model output, and training a preset model to generate;
the constructing module 603 is configured to construct a target relationship triple including the target main entity word, the target relation word, and the target sub-entity word.
In one embodiment of the present invention, the building module 603 includes:
the position determining submodule is used for determining the positions of the target auxiliary entity words and the target relation words in the text to be mined;
and the construction sub-module is used for constructing a target relation triple by adopting the target main entity word, the target auxiliary entity word and the target relation word which are adjacent in position in the text to be mined.
In one embodiment of the invention, the building submodule comprises:
a target relation word determining unit, configured to, if the target subordinate entity word has at least two target relation words adjacent in position in the text to be mined, take the target relation word with the smallest number of characters spaced apart from the target subordinate entity word as the target relation word corresponding to the target subordinate entity word;
and the construction unit is used for constructing a target relation triple by adopting the main entity word, the target auxiliary entity word and the target relation word corresponding to the target auxiliary entity word.
For the device embodiment, since it is basically similar to the method embodiment, the description is simple, and for the relevant points, refer to the partial description of the method embodiment.
An embodiment of the present invention further provides an apparatus, including:
one or more processors; and
one or more machine-readable media having instructions stored thereon, which when executed by the one or more processors, cause the apparatus to perform methods as described in embodiments of the invention.
Embodiments of the invention also provide one or more machine-readable media having instructions stored thereon, which when executed by one or more processors, cause the processors to perform the methods described in embodiments of the invention.
The embodiments in the present specification are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other.
As will be appreciated by one skilled in the art, embodiments of the present invention may be provided as a method, apparatus, or computer program product. Accordingly, embodiments of the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, embodiments of the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
Embodiments of the present invention are described with reference to flowchart illustrations and/or block diagrams of methods, terminal devices (systems), and computer program products according to embodiments of the invention. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing terminal to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing terminal, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing terminal to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing terminal to cause a series of operational steps to be performed on the computer or other programmable terminal to produce a computer implemented process such that the instructions which execute on the computer or other programmable terminal provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
While preferred embodiments of the present invention have been described, additional variations and modifications of these embodiments may occur to those skilled in the art once they learn of the basic inventive concepts. Therefore, it is intended that the appended claims be interpreted as including preferred embodiments and all such alterations and modifications as fall within the scope of the embodiments of the invention.
Finally, it should also be noted that, herein, relational terms such as first and second, and the like may be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Also, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or terminal that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or terminal. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other like elements in a process, method, article, or terminal that comprises the element.
The method for generating a triple prediction model, the method for extracting a relational triple, the device for generating a triple prediction model, and the device for extracting a relational triple provided by the present invention have been described in detail, and a specific example is applied in the present document to explain the principle and the implementation of the present invention, and the description of the above embodiment is only used to help understand the method of the present invention and the core idea thereof; meanwhile, for a person skilled in the art, according to the idea of the present invention, there may be variations in the specific embodiments and the application scope, and in summary, the content of the present specification should not be construed as a limitation to the present invention.
Claims (10)
1. A method for generating a triple prediction model, comprising:
acquiring a training text containing a preset relation triple; the relation triples comprise main entity words, relation words and auxiliary entity words; the training text comprises main entity words, auxiliary entity words and relation words corresponding to the relation triples;
and taking the training text and the main entity word as model input, taking the auxiliary entity word and the relation word as model output, training a preset model, and generating a triple prediction model.
2. The method of claim 1, wherein the step of obtaining the training text containing the preset relationship triples comprises:
searching a candidate text which simultaneously contains a main entity word and an auxiliary entity word in a preset relation triple;
and searching a candidate text containing the relation words in the relation triple or the similar meaning words of the relation words as a training text.
3. The method according to claim 1, wherein the step of training a preset model by using the training text and the primary entity word in the relation triplet as model inputs and the secondary entity word and the relation word as model outputs to generate a triplet prediction model comprises:
taking the training text and the main entity word in the relation triple as model input, taking the auxiliary entity word and the relation word as model output, training a preset model, and acquiring auxiliary entity word prediction information and relation word prediction information output by the model;
determining a link loss function of the model according to the auxiliary entity word prediction information and the relation word prediction information; the link loss function is used for representing whether the auxiliary entity word prediction information and the relation word prediction information with the corresponding relation appear in pairs or not;
and adjusting model parameters of the model according to the link loss function, and training the model in an iterative manner until the link loss function reaches a preset condition, finishing the model training, and taking the model as a triple prediction model.
4. A method for extracting a relation triple is characterized by comprising the following steps:
searching a text to be mined containing a target main entity word;
determining target auxiliary entity words and target relation words by adopting the target main entity words, the text to be mined and a preset triple prediction model; the triple prediction model acquires a training text containing a preset relation triple; taking the training text and the main entity word in the relation triple as model input, taking the auxiliary entity word and the relation word as model output, and training a preset model to generate;
and constructing a target relation triple comprising the target main entity word, the target relation word and the target auxiliary entity word.
5. The method of claim 4, wherein the step of constructing a target relationship triplet including the target main entity word, the target relationship word, and the target sub-entity word comprises:
determining the positions of target auxiliary entity words and target relation words in the text to be mined;
and constructing a target relation triple by adopting the target main entity word, the target auxiliary entity word and the target relation word which are adjacent in position in the text to be mined.
6. The method according to claim 5, wherein the step of constructing a target relationship triple by using the target primary entity word and the target secondary entity word and the target relationship word which are adjacent to each other in the text to be mined comprises:
if the target subordinate entity words have at least two target relation words adjacent in position in the text to be mined, taking the target relation words with the least number of spaced characters between the target subordinate entity words as the target relation words corresponding to the target subordinate entity words;
and constructing a target relation triple by adopting the target main entity word, the target auxiliary entity word and the target relation word corresponding to the target auxiliary entity word.
7. An apparatus for generating a ternary prediction model, comprising:
the acquisition module is used for acquiring a training text containing a preset relation triple; the relation triples comprise main entity words, relation words and auxiliary entity words; the training text is provided with the subordinate entity words and the relation words corresponding to the relation triples;
and the training module is used for inputting the training text and the main entity words in the relation triples as models, outputting the auxiliary entity words and the relation words as models, training a preset model and generating a triple prediction model.
8. An apparatus for extracting a relational triplet, comprising:
the searching module is used for searching a text to be mined, which contains the main entity words;
the input module is used for determining target auxiliary entity words and target relation words by adopting the target main entity words, the text to be mined and a preset triple prediction model; the triple prediction model acquires a training text containing a preset relation triple; taking the training text and the main entity word in the relation triple as model input, taking the auxiliary entity word and the relation word as model output, and training a preset model to generate;
and the construction module is used for constructing a target relation triple comprising the target main entity word, the target relation word and the target auxiliary entity word.
9. An apparatus, comprising:
one or more processors; and
one or more machine-readable media having instructions stored thereon that, when executed by the one or more processors, cause the apparatus to perform the method of one or more of claims 1-3 or 4-6.
10. One or more machine readable media having instructions stored thereon that, when executed by one or more processors, cause the processors to perform the method of one or more of claims 1-3 or 4-6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010785316.1A CN112069818A (en) | 2020-08-06 | 2020-08-06 | Triple prediction model generation method, relation triple extraction method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010785316.1A CN112069818A (en) | 2020-08-06 | 2020-08-06 | Triple prediction model generation method, relation triple extraction method and device |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112069818A true CN112069818A (en) | 2020-12-11 |
Family
ID=73660900
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010785316.1A Pending CN112069818A (en) | 2020-08-06 | 2020-08-06 | Triple prediction model generation method, relation triple extraction method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112069818A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114595686A (en) * | 2022-03-11 | 2022-06-07 | 北京百度网讯科技有限公司 | Knowledge extraction method, and training method and device of knowledge extraction model |
| WO2024066045A1 (en) * | 2022-09-27 | 2024-04-04 | 深圳先进技术研究院 | Guarantee information extraction and value prediction method and system, terminal, and storage medium |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109902145A (en) * | 2019-01-18 | 2019-06-18 | 中国科学院信息工程研究所 | A kind of entity relationship joint abstracting method and system based on attention mechanism |
| WO2019174422A1 (en) * | 2018-03-16 | 2019-09-19 | 北京国双科技有限公司 | Method for analyzing entity association relationship, and related apparatus |
| CN111198932A (en) * | 2019-12-30 | 2020-05-26 | 北京明略软件系统有限公司 | Triple acquiring method and device, electronic equipment and readable storage medium |
| CN111444305A (en) * | 2020-03-19 | 2020-07-24 | 浙江大学 | Multi-triple combined extraction method based on knowledge graph embedding |
-
2020
- 2020-08-06 CN CN202010785316.1A patent/CN112069818A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2019174422A1 (en) * | 2018-03-16 | 2019-09-19 | 北京国双科技有限公司 | Method for analyzing entity association relationship, and related apparatus |
| CN109902145A (en) * | 2019-01-18 | 2019-06-18 | 中国科学院信息工程研究所 | A kind of entity relationship joint abstracting method and system based on attention mechanism |
| CN111198932A (en) * | 2019-12-30 | 2020-05-26 | 北京明略软件系统有限公司 | Triple acquiring method and device, electronic equipment and readable storage medium |
| CN111444305A (en) * | 2020-03-19 | 2020-07-24 | 浙江大学 | Multi-triple combined extraction method based on knowledge graph embedding |
Non-Patent Citations (1)
| Title |
|---|
| 李颖;郝晓燕;王勇;: "中文开放式多元实体关系抽取", 计算机科学, no. 1 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114595686A (en) * | 2022-03-11 | 2022-06-07 | 北京百度网讯科技有限公司 | Knowledge extraction method, and training method and device of knowledge extraction model |
| CN114595686B (en) * | 2022-03-11 | 2023-02-03 | 北京百度网讯科技有限公司 | Knowledge extraction method, and training method and device of knowledge extraction model |
| WO2024066045A1 (en) * | 2022-09-27 | 2024-04-04 | 深圳先进技术研究院 | Guarantee information extraction and value prediction method and system, terminal, and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107818085B (en) | Answer selection method and system for reading understanding of reading robot | |
| US20170193393A1 (en) | Automated Knowledge Graph Creation | |
| CN111783518A (en) | Training sample generation method and device, electronic equipment and readable storage medium | |
| CN106503231B (en) | Search method and device based on artificial intelligence | |
| US10831993B2 (en) | Method and apparatus for constructing binary feature dictionary | |
| CN110968695A (en) | Intelligent labeling method, device and platform based on active learning of weak supervision technology | |
| CN110929038A (en) | Entity linking method, device, equipment and storage medium based on knowledge graph | |
| US10789533B2 (en) | Generating a consistently labeled training dataset by automatically generating and displaying a set of most similar previously-labeled texts and their previously assigned labels for each text that is being labeled for the training dataset | |
| US11720481B2 (en) | Method, apparatus and computer program product for predictive configuration management of a software testing system | |
| CN110232112A (en) | Keyword extracting method and device in article | |
| CN111259262A (en) | Information retrieval method, device, equipment and medium | |
| CN112149386A (en) | Event extraction method, storage medium and server | |
| CN112069818A (en) | Triple prediction model generation method, relation triple extraction method and device | |
| US11379527B2 (en) | Sibling search queries | |
| CN110751234B (en) | OCR (optical character recognition) error correction method, device and equipment | |
| CN115526171A (en) | Intention identification method, device, equipment and computer readable storage medium | |
| CN116361306A (en) | Open domain science popularization-oriented question-answer library automatic updating method and device | |
| CN116795789B (en) | Method and device for automatically generating patent retrieval report | |
| CN113934834A (en) | Question matching method, device, equipment and storage medium | |
| CN115150354B (en) | Method and device for generating domain name, storage medium and electronic equipment | |
| CN115438655A (en) | Person gender identification method and device, electronic equipment and storage medium | |
| CN114491209A (en) | Method and system for mining enterprise business label based on internet information capture | |
| KR20220042103A (en) | Method and Apparatus for Providing Hybrid Intelligent Customer Consultation | |
| CN112115362B (en) | Programming information recommendation method and device based on similar code recognition | |
| CN109344254B (en) | Address information classification method and device |
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 |