CN114925186A - Knowledge graph question generation method, device, equipment and storage medium - Google Patents
Knowledge graph question generation method, device, equipment and storage medium Download PDFInfo
- Publication number
- CN114925186A CN114925186A CN202210566962.8A CN202210566962A CN114925186A CN 114925186 A CN114925186 A CN 114925186A CN 202210566962 A CN202210566962 A CN 202210566962A CN 114925186 A CN114925186 A CN 114925186A
- Authority
- CN
- China
- Prior art keywords
- question
- vector
- graph
- word
- external
- 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 37
- 239000013598 vector Substances 0.000 claims abstract description 238
- 230000006870 function Effects 0.000 claims abstract description 20
- 238000003062 neural network model Methods 0.000 claims abstract description 17
- 230000009466 transformation Effects 0.000 claims abstract description 15
- 230000011218 segmentation Effects 0.000 claims description 27
- 238000013528 artificial neural network Methods 0.000 claims description 13
- 238000004364 calculation method Methods 0.000 claims description 13
- 238000004590 computer program Methods 0.000 claims description 13
- 230000007246 mechanism Effects 0.000 claims description 13
- 239000011159 matrix material Substances 0.000 claims description 12
- 150000001875 compounds Chemical class 0.000 claims description 9
- 230000004927 fusion Effects 0.000 claims description 8
- 239000003550 marker Substances 0.000 claims description 4
- 238000005094 computer simulation Methods 0.000 claims description 3
- 238000010606 normalization Methods 0.000 claims 1
- 238000003058 natural language processing Methods 0.000 abstract description 2
- 238000012549 training Methods 0.000 description 10
- 238000010586 diagram Methods 0.000 description 6
- 238000005516 engineering process Methods 0.000 description 4
- 230000004044 response Effects 0.000 description 4
- 238000013135 deep learning Methods 0.000 description 3
- 230000006872 improvement Effects 0.000 description 3
- 230000008569 process Effects 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 2
- 238000001514 detection method Methods 0.000 description 2
- 238000002372 labelling Methods 0.000 description 2
- 238000013507 mapping Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 238000013473 artificial intelligence Methods 0.000 description 1
- 238000004422 calculation algorithm Methods 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/33—Querying
- G06F16/332—Query formulation
- G06F16/3329—Natural language query formulation or dialogue systems
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/36—Creation of semantic tools, e.g. ontology or thesauri
- G06F16/367—Ontology
-
- 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
- G06N3/044—Recurrent networks, e.g. Hopfield 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Databases & Information Systems (AREA)
- Human Computer Interaction (AREA)
- Animal Behavior & Ethology (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The embodiment of the invention provides a method, a device, equipment and a storage medium for generating question sentences of a knowledge graph, and relates to the technical field of natural language processing. The question generation method includes steps S1 through S7. And S1, acquiring the knowledge graph. And S2, obtaining sub-graph vectors of each sub-graph through graph transformation network models according to the knowledge graph. And S3, acquiring a question data set. And S4, acquiring external question sentences of each subgraph based on the similarity according to the knowledge graph and the question data set. And S5, acquiring the parameters of the five question types. And S6, obtaining an external enhancement vector through a Bi LSTM neural network model according to the five question type parameters and the external question. And S7, generating a network model through the pointer according to the sub-graph vector and the external enhancement vector, and acquiring the question. The invention has important guidance and promotion functions for generating questions with various types, rich semantic knowledge and natural language expression required in the teaching scene.
Description
Technical Field
The invention relates to the technical field of natural language processing, in particular to a question generating method, a question generating device, question generating equipment and a storage medium of a knowledge graph.
Background
In the text generation task, the artificial intelligence technology can extract knowledge by analyzing chapters, and then extract question sentences. However, most of the current researches from discourse to question generation are directly mapped from the discourse to the question, and the amount of data required for model training is huge. Meanwhile, the essence from chapters to question sentences is statistical learning and a deep learning model is trained by using large-scale data, which lacks knowledge guidance, and the quality of the generated question sentences cannot well sense the inference relationship and the hierarchical relationship among knowledge points.
In view of the above, the applicant has specifically proposed the present application after studying the existing technologies.
Disclosure of Invention
The invention provides a question generation method, a question generation device, question generation equipment and a storage medium of a knowledge graph, which aim to solve the technical problems.
First aspect,
The embodiment of the invention provides a question generation method of a knowledge graph, which comprises the following steps of S1 to S7:
and S1, acquiring the knowledge graph. The knowledge-graph includes a set of sub-graphs, each sub-graph including a set of entities, a set of relationships, and a set of triples.
And S2, obtaining sub-graph vectors of each sub-graph through graph transformation network models according to the knowledge graph.
And S3, acquiring a question data set.
And S4, acquiring external question sentences of each subgraph based on the similarity according to the knowledge graph and the question data set.
And S5, acquiring five problem type parameters.
And S6, obtaining an external enhancement vector through a BilSTM neural network model according to the five question type parameters and the external question.
And S7, generating a network model through the pointer according to the sub-graph vector and the external enhancement vector, and acquiring the question.
The second aspect,
The embodiment of the invention provides a question generating device of a knowledge graph, which comprises:
and the knowledge graph acquisition module is used for acquiring a knowledge graph. The knowledge-graph includes a set of sub-graphs, each sub-graph including a set of entities, a set of relationships, and a set of triples.
And the sub-graph vector acquisition module is used for acquiring the sub-graph vector of each sub-graph through the graph transformation network model according to the knowledge graph.
And the question set acquisition module is used for acquiring a question data set.
And the external question acquisition module is used for acquiring the external question of each subgraph based on the similarity according to the knowledge graph and the question data set.
And the five-question acquisition module is used for acquiring the five-question type parameters.
And the external vector acquisition module is used for acquiring an external enhancement vector through a BilSTM neural network model according to the five question type parameters and the external question.
And the question generation module is used for generating a network model through a pointer according to the sub-graph vector and the external enhancement vector to obtain a question.
The third aspect,
An embodiment of the present invention provides a question generation apparatus for a knowledge graph, which includes a processor, a memory, and a computer program stored in the memory. A computer program capable of being executed by a processor to implement the method for generating a question of a knowledge-graph as described in any of the paragraphs above.
The fourth aspect,
An embodiment of the present invention provides a computer-readable storage medium. The computer-readable storage medium comprises a stored computer program, wherein when the computer program runs, the apparatus in which the computer-readable storage medium is located is controlled to execute the question generation method for a knowledge-graph according to any one of the first aspect.
By adopting the technical scheme, the invention can obtain the following technical effects:
the knowledge map can well organize huge education knowledge points, and the knowledge map is combined with a deep learning technology to generate knowledge question and answer data with higher reasoning performance and clearer hierarchy, so that the user is helped to perform reasoning learning of reading understanding, and the reading understanding capability of the user is better improved.
Mapping from a knowledge-graph to a question is more effective than in the case of an equivalent data set. The knowledge graph can better strengthen the relation between knowledge points, and can open multi-dimensional information such as the meanings of important words and sentences in the text, the main viewpoints of articles and the like into a huge knowledge network, thereby assisting the improvement of the comprehension capability and the analysis comprehensive capability of the user on the chapters.
The invention has important guidance and promotion functions for generating questions with various types, rich semantic knowledge and natural language expression required in the teaching scene.
In order to make the aforementioned and other objects, features and advantages of the present invention comprehensible, preferred embodiments accompanied with figures are described in detail below.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings needed to be used in the embodiments will be briefly described below, it should be understood that the following drawings only illustrate some embodiments of the present invention and therefore should not be considered as limiting the scope, and for those skilled in the art, other related drawings can be obtained according to the drawings without inventive efforts.
Fig. 1 is a schematic flowchart of a question generation method according to a first embodiment of the present invention.
Fig. 2 is a logic block diagram of a question generation method according to a first embodiment of the present invention.
Fig. 3 is an example of a subgraph.
FIG. 4 is a result format of the labeling of the sentence text by the "answer" pre-training model;
fig. 5 is a schematic structural diagram of a question generation apparatus according to a second embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be obtained by a person skilled in the art without making any creative effort based on the embodiments in the present invention, belong to the protection scope of the present invention.
For better understanding of the technical solutions of the present invention, the following detailed descriptions of the embodiments of the present invention are provided with reference to the accompanying drawings.
The terminology used in the embodiments of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. As used in the examples of the present invention and the appended claims, the singular forms "a," "an," and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise.
It should be understood that the term "and/or" as used herein is merely one type of association that describes an associated object, meaning that three relationships may exist, e.g., a and/or B may mean: a exists alone, A and B exist simultaneously, and B exists alone. In addition, the character "/" herein generally indicates that the former and latter related objects are in an "or" relationship.
The word "if," as used herein, may be interpreted as "at … …" or "when … …" or "in response to a determination" or "in response to a detection," depending on the context. Similarly, the phrases "if determined" or "if detected (a stated condition or event)" may be interpreted as "when determined" or "in response to a determination" or "when detected (a stated condition or event)" or "in response to a detection (a stated condition or event)", depending on the context.
In the embodiments, the references to "first \ second" are merely to distinguish similar objects and do not represent a specific ordering for the objects, and it is to be understood that "first \ second" may be interchanged with a specific order or sequence, where permitted. It should be understood that "first \ second" distinguishing objects may be interchanged under appropriate circumstances such that the embodiments described herein may be implemented in sequences other than those illustrated or described herein.
The invention is described in further detail below with reference to the following detailed description and accompanying drawings:
the first embodiment is as follows:
referring to fig. 1 to 4, a first embodiment of the present invention provides a method for generating knowledge-graph question sentences, which can be executed by a knowledge-graph question sentence generating device. In particular, execution by one or more processors in the question generating device to implement steps S1-S7:
and S1, acquiring a knowledge graph. The knowledge-graph includes a set of sub-graphs, each sub-graph including a set of entities, a set of relationships, and a set of triples.
It can be understood that: it is understood that the question sentence generating device may be an electronic device with computing capability, such as a notebook computer, a desktop computer, a server, a smart phone, or a tablet computer.
It is noted that a knowledge-graph G is given. The knowledge graph G includes an entity set E ═ { E ═ E 1 ,e 2 ,…,e N And a set of relationship types R ═ R 1 ,r 2 ,…,r M }. Wherein, N and M are the entity number and the relation category number of the knowledge graph G respectively. Preferably, N-14253 and M-423.
Knowledge-graph G contains at least one subgraph. Any subgraph is g, and the subgraph g is one sub-graph<e i ，r，e j >Formal triple data. e.g. of the type i ，e j Respectively representing two different entitiesR represents e i Direction e j The relationship (c) in (c). Sub graph g contains entity set E g ＝{e 1 ,e 2 ,…,e n The relation classes are collected as R g ＝{r 1 ,r 2 ,…,r m }, and a triplet set Triad g ＝{triad 1 ,triad 2 ,…,triad tn }. n, m and tn respectively represent the entity number, the relation type number and the triple number of the subgraph g.
The specific format of sub-graph g is shown in fig. 3. Wherein, the array E in the figure represents the entity set E g The array R represents a set R of relationship classes g Array link represents triple set Triad g rId represents the id corresponding to the relation r, and source and target are both the id corresponding to the entity e.
And S2, obtaining sub-graph vectors of each sub-graph through graph transformation network models according to the knowledge graph.
Specifically, sub-graphs in the knowledge graph are obtained first, then the graph is transformed into a network model, the sub-graphs are transformed into a neural network model, and a vector can be recognized, so that prediction is carried out. Preferably, the Graph transformation network model is a Graph Transformer neural network model, in other embodiments, other existing graphs may also be used to represent the learning neural network model, and the present invention is not limited in this respect.
On the basis of the above embodiment, in an alternative embodiment of the present invention, step S2 specifically includes steps S21 to S24.
And S21, inputting the entity set of the subgraph into the word vector model to obtain the entity vector set of the subgraph.
In this embodiment, the word vector model in step S21 is a GloVe pre-training model, and in other embodiments, other word vector models may be adopted, which is not specifically limited in the present invention.
Specifically, the entity e and the relation r of the sub-graph g are subjected to initial vector representation through a GloVe pre-training model to obtain an entity vector set of the sub-graph g
And offSet of system vectors
v e And v r Respectively, an initial vector representation of any entity e and any relation r of the sub-graph g. Preferably, the word vector dimension is 200 dimensions.
S22, collecting sub-graph g entity vectors
Inputting a graph attention network model, and acquiring intermediate layer vectors of each entity e in the sub-graph g
Wherein, the calculation model of the intermediate layer vector is as follows:
in the formula (I), the compound is shown in the specification,
intermediate level vector, v, being entity e e An entity vector of an entity e, V represents the information of N attention heads in the attention network model,
Representing the attention parameter at the nth attention head
Is an entity e in the graph attention network model i With entity e j Attention to (1)The force parameters,
Is a feature matrix in the nth attention head,
As entity e j The entity vector of,
As entity e i Entity vector of, W Q And W K For calculating the feature matrix of the attention parameter,
As entity e i A set of neighbor nodes in the knowledge-graph.
In the present embodiment, the Graph of steps S22 and S23 is aware that the force network model is a Graph Attention neural network model.
S23, according to the intermediate layer vector of each entity e in the subgraph g
Obtaining output vector representations of the entities through a regularization layer and a feedforward neural network layer of the graph attention network model
Wherein the computational model of the output vector representation is:
FFN(x)＝W 1 ×(f(W 2 ×x+b 1 )+b 2 ) In the formula (I), the compound is shown in the specification,
is an output vector representation of entity e, LayerNorm is a normalizing function, v' e Intermediate layer vector for entity e
Hidden layer vector W obtained through LayerNorm function and feed-forward neural network layer FFN 1 And W 2 Is a feature matrix in the feedforward neural network, x represents the input information of the feedforward neural network layer, b 1 And b 2 Is an offset value.
S24, obtaining sub-graph vector of sub-graph g according to output vector representation of each entity
Thus obtaining a sub-picture vector for each sub-picture.
And S3, acquiring a question data set.
Specifically, the question data set is an encyclopedic question data set and is a published question data set.
And S4, acquiring external question sentences of each subgraph based on the similarity according to the knowledge graph and the question data set.
On the basis of the above embodiment, in an alternative embodiment of the present invention, step S4 specifically includes steps S41 to S45.
S41, inputting the relation set of the subgraph into a PaddleNLP 'language interpretation' model, and acquiring a relation part-of-speech tagging information set.
And S42, inputting the question data set into a PaddleNLP 'language solving' model, and acquiring a question word class labeling information set.
Specifically, the relationship set R of the last hop of the graph topology in each sub-graph g is set g ′＝{r 1 ,r 2 ,…,r m ' } and encyclopedia data set BQ ═ BQ 1 ,bq 2 ,bq 3 ,…,bq I Sending each encyclopedia question bq into a PaddleNLP 'language solving' pre-training model to obtain R g′ Relation part of speech tagging information set corresponding to BQ
And question word class label information set
The output result format is shown in figure 4. Where m' is the number of relationships in the last hop relationship set, I ═ 1502012 is the number of questions in the encyclopedic question data set BQ, text in fig. 4 is the input text, and items is the set of part-of-speech tagging information after parsing the text.
S43, calculating a question word class tagging information set P BQ Each part of speech tagging information p in (1) bq And relation part of speech tagging information set
Degree of similarity of (2)
Wherein, the similarity calculation model is as follows:
in the formula (I), the compound is shown in the specification,
representing relation part of speech tagging information
And question word class label information p bq The number of the part-of-speech tagging information of the intersection,
Representing relation part of speech tagging information
Word class tagging information p with question bq The number of the part of speech tagging information of the union.
And S44, acquiring a similarity set of the subgraph g and the question data set BQ according to the similarity.
Wherein, the similarity set
And S45, taking the question corresponding to the maximum similarity in the similarity set as the external question of the subgraph, thereby obtaining the external question of each subgraph.
Specifically, SIMILAR g Maximum mean max (SIMILAR) g ) The corresponding encyclopedia question is used as the external question ext of the subgraph g g 。
And S5, acquiring the parameters of the five question types.
Specifically, the five question type parameters include what, why, how, what, and why. Which belongs to the prior art, the invention is not repeated here.
And S6, obtaining an external enhancement vector through a BilSTM neural network model according to the five question type parameters and the external question.
Specifically, by introducing five question type parameters and external question sentences, knowledge can be guided, so that the language expression of the question sentences is more natural, semantic knowledge is richer, and the types are more various.
On the basis of the above embodiment, in an alternative embodiment of the present invention, step S6 specifically includes steps S61 to S63.
And S61, acquiring five question type vectors and a first participle vector set through a word vector model according to the five question type parameters and the external question. Specifically, the word vector model can convert the five question type parameters and the external question into vectors which can be identified by the neural network model, and has good practical significance.
On the basis of the foregoing embodiment, step S61 specifically includes step S611 to step S613 in an optional embodiment of the present invention.
S611, inputting the five question type parameters into the oneHot word vector model, and obtaining the five question type vectors. Preferably, the dimension of the five question type vector is 5 dimensions.
S612, inputting the external question into a PaddleNLP 'language interpretation' model, and acquiring a first participle set of the external question so as to acquire the first participle set of each subgraph. Wherein the first set of terms is represented as: extSplit g ＝{w 1 ,w 2 ,…,w eh In the formula, eh is extSplit g The number of words.
S613, inputting the first participle set into a GloVe word vector model to obtain a first participle vector set. Wherein the first set of tokenized vectors is represented as:
preferably, the word vector dimension of the first word segmentation vector adopts 200 dimensions.
Specifically, the five question type parameters are different from the format of the external question, so that different word vector models are adopted for targeted conversion, vectors which can be identified by the neural network model are obtained, and the method has a good practical significance.
And S62, carrying out vector longitudinal splicing on each word vector in the first word segmentation vector set and the five problem type vectors to obtain a spliced word vector set.
And S63, inputting the spliced word vector set into a BilSTM neural network model to obtain an external enhanced vector set.
In particular, the five problem type vector
First set of word vectors associated with an external question
Each word vector v in w Performing vector vertical stitching, wherein
Is 200DThe vector of the word or words,
the word vectors are 5-dimensional vectors, 205-dimensional word vectors are obtained after longitudinal splicing, and the spliced word vectors are sent into a BilSTM network to obtain external enhancement vectors
And S7, generating a network model through the pointer according to the sub-graph vector and the external enhancement vector, and acquiring the question.
Specifically, first, the entity set and the sub-graph vector of the sub-graph are input into Attention mechanism models of N multi-head Attention blocks, and a first context vector of the sub-graph is obtained.
Then, inputting the first participle set of the external question and the external enhancement vector into the Attention mechanism models of the N multi-head Attention blocks, and acquiring five question type parameters and a second context vector fused with the external question.
And then, splicing the first context vector and the second context vector to obtain a fusion context vector.
Then, the context vector and the hidden state h at time t of the time sequence in Attention mechanism are fused t And inputting a pointer to generate a network model, and acquiring the probability P (w) that the word generated at the time t is w.
And finally, taking the word w corresponding to the maximum probability P (w) at the time t as a prediction output word at the time t. And when the w is an end marker, ending the question generation task to obtain the generated question.
In this embodiment, the Pointer-generating Network model is Pointer-Generator Network.
Specifically, the knowledge map can well organize huge educational knowledge points, and the knowledge map is combined with a deep learning technology to generate knowledge question and answer data with higher reasoning performance and clearer hierarchy, so that the user is helped to perform reasoning learning of reading understanding, and the reading understanding ability of the user is better improved. Mapping from a knowledge-graph to a question is more effective than in the case of an equivalent data set. The knowledge map can better strengthen the connection between knowledge points, and can open multi-dimensional information such as the meanings of important words and sentences in the text, the main viewpoints of articles and the like into a huge knowledge network, thereby assisting the improvement of the comprehension ability and the analysis comprehensive ability of the user to chapters.
The question generation method has important guidance and promotion functions on generating questions with various types, rich semantic knowledge and natural language expression required in the teaching scene.
FIG. 2 shows a framework model of a Question generation method of Knowledge Graph, namely a T-KGQG (transform based Knowledge Graph Question Generator) model, which mainly comprises five parts, namely a BilSTM Network module, a Graph Transformer module, an Attention mechanism and a Pointer-Generator Network.
In the above embodiment, the T-KGQG is a trained model, and the question corresponding to the subgraph can be generated only by inputting the subgraph and the five question type parameters into the T-KGQG model. In other embodiments, the T-KGQG may be a model that has not yet been trained. The training procedure of T-KGQG is specifically from step S701 to step S711.
S701, obtaining target question queries corresponding to sub-graphs g Inputting the target question into a jieba word segmentation tool to obtain a second word segmentation set questinonsplit g ＝{w 1 ,w 2 ,…,w u }. Wherein u is a quetionsplit g The number of words.
Specifically, in this step, the subgraphs are training sets, and each subgraph has a corresponding target question sentence labeled manually.
S702, inputting the second participle set into the word vector model to obtain a second participle vector set
And the word vector dimension of the second participle vector adopts 200 dimensions. In this embodiment, the word vector model of step S702 is a GloVe pre-training model. Other existing word vector models may be used in other embodiments, and the invention is not specific to thisAnd (4) limiting.
S703, extracting all words in the entity set, the relation set and the second word segmentation set of the knowledge graph, and obtaining a dictionary.
Specifically, E ═ E in the knowledge graph 1 ，e 2 ，...，e N R } and R ═ R 1 ，r 2 ，...，r M }, and a second set of words questinonssplit g All the words in the dictionary are extracted to form a dictionary Dict ═ w 1 ，w 2 ，...，w DN }. Wherein DN 21035 is the number of words in dictionary Dict.
S704, inputting the entity set and the sub-image vector of the sub-image into Attention mechanism models of N multi-head Attention blocks, and obtaining a first context vector of the sub-image. Preferably, N ═ 5.
S705, inputting the first participle set of the external question and the external enhancement vector into Attention mechanism models of N multi-head Attention blocks, and acquiring a second context vector fused with five question type parameters and the external question. Preferably, N ═ 5.
S706, splicing the first context vector and the second context vector to obtain a fusion context vector.
S707, dictionary and fusion context vector, and hidden state h at time t of time series in Attention mechanism t And inputting a pointer to generate a network model, and acquiring the probability P (w) that the word generated at the time t is w. Wherein, the calculation model of the probability is as follows:
P(w)＝p*α copy +(1-p)*α vocab
p＝σ(W copy [h t ||c t ]+b copy )
α copy ＝a([h t ||c t ]，w)
α vocab ＝softmax(W vocab [h t ||c t ]+b vocab )
wherein p is a soft switch for selecting a word from a subgraph or an external question or selecting a word from a dictionary, and alpha copy For computing fromProbability distribution, alpha, of duplicate words in subgraphs or external questions vocab For calculating the probability distribution of words copied from a dictionary, sigma-representing sigmoid function, W copy And W vocab Generating a feature matrix in a network model for pointers, b copy And b vocab Is an offset value.
In this embodiment, the Pointer-generating Network model is Pointer-Generator Network.
And S708, taking the word w corresponding to the maximum probability P (w) at the time t as a prediction output word at the time t. And when w is an end marker, the question generation task is ended, so that a prediction word set for generating a sentence and a prediction word vector set corresponding to the word set are obtained.
And S709, calculating the loss L of the graph transformation network model according to the second participle vector set, all sub-graphs of the knowledge graph and the predicted word vector set. Wherein, the calculation model of the loss is as follows:
wherein G is a knowledge graph, G is a subgraph of the knowledge graph, NLLLoss represents a negative log-likelihood loss function, V Pre Is a set of vectors for the second word,
is a first set of word vectors.
And S710, with the minimum loss as a target, updating the network parameters of the graph transformation network model until the maximum iteration number T is reached, and obtaining a question generation model of the trained knowledge graph.
Specifically, a loss function L is minimized, and an algorithm model is built by using a Pythrch deep learning frame in an experiment. In order to prevent the neural network from being overfitting, a gradient clipping clip _ gradient method is introduced during training, the threshold value of gradient clipping is set to be 4, and gradient truncation operation is carried out before the optimizer is updated during training. The optimizer employs SGD random gradient descent with an initial learning rate of 1e-3 and a batch size of 32. The Dropout rate is set to be 0.1, the LSTM hidden vector size related to the invention is set to be 500, the network parameters are updated until the maximum iteration number T is 20, and a well-trained T-KGQG model is obtained.
And S711, inputting the subgraph and the five question type parameters into a question generation model of the knowledge graph to obtain a question.
Example II,
The embodiment of the invention provides a question generating device of a knowledge graph, which comprises:
the knowledge graph acquisition module 1 is used for acquiring a knowledge graph. The knowledge-graph includes a set of sub-graphs, each sub-graph including a set of entities, a set of relationships, and a set of triples.
And the sub-graph vector acquisition module 2 is used for acquiring sub-graph vectors of all sub-graphs through graph transformation network models according to the knowledge graph.
And the question set acquisition module 3 is used for acquiring a question data set.
And the external question acquiring module 4 is used for acquiring the external question of each subgraph based on the similarity according to the knowledge graph and the question data set.
And the five-question acquisition module 5 is used for acquiring the five-question type parameters.
And the external vector acquisition module 6 is used for acquiring external enhancement vectors through a BilSTM neural network model according to the five question type parameters and the external question.
And the question generation module 7 is used for generating a network model through a pointer according to the sub-graph vector and the external enhancement vector to obtain a question.
In an optional embodiment, the sub-picture vector obtaining module 2 specifically includes:
and the entity vector set acquisition unit is used for inputting the entity set of the subgraph into the word vector model and acquiring the entity vector set of the subgraph.
An intermediate layer vector obtaining unit for inputting the entity vector set into the attention network model of the graph to obtain the intermediate layer vector of each entity
Wherein, the calculation model of the intermediate layer vector is as follows:
in the formula (I), the compound is shown in the specification,
intermediate level vector, v, being entity e e The information of N attention heads in the graph attention network model is serially connected for an entity vector of an entity e and a V-shaped representation,
Representing the attention parameter at the nth attention head
Is entity e in the graph attention network model i With entity e j Attention parameters of,
Is a feature matrix in the nth attention head,
As entity e j The entity vector of (a),
As entity e i Entity vector of, W Q And W K For calculating the feature matrix of the attention parameter,
As entity e i A set of neighbor nodes in the knowledge-graph.
An output vector representation obtaining unit for obtaining the output vector representation of each entity through the regularization layer and the feedforward neural network layer of the graph attention network model according to the intermediate layer vector of each entity
Wherein the computational model of the output vector representation is:
FFN(x)＝W 1 ×(f(W 2 ×x+b 1 )+b 2 ) In the formula (I), the compound is shown in the specification,
is an output vector representation of entity e, LayerNorm is a normalizing function, v' e Intermediate layer vector of entity e
Hidden layer vector, W, obtained by LayerNorm function and feedforward neural network layer FFN 1 And W 2 Is a feature matrix in the feedforward neural network, x represents the input information of the feedforward neural network layer, b 1 And b 2 Is an offset value.
And the sub-image vector acquisition unit is used for obtaining the sub-image vectors of the sub-images according to the output vector representation of each entity so as to acquire the sub-image vectors of each sub-image.
In an optional embodiment, the external question obtaining module 4 specifically includes:
and the relational word class tagging information set acquisition unit is used for inputting the relational set of the subgraph into a PaddleNLP 'language interpretation' model and acquiring the relational word class tagging information set.
And the question word class tagging information set acquisition unit is used for inputting the question data set into the PaddleNLP 'language interpretation' model and acquiring the question word class tagging information set.
A similarity obtaining unit for calculating the similarity of each part of speech tagging information in the question part of speech tagging information set and the relation part of speech tagging information set
Wherein, the similarity calculation model is as follows:
in the formula (I), the compound is shown in the specification,
representing relation part of speech tagging information
Word class tagging information p with question bq The number of the intersection part of the part of speech tagging information,
Representing relation part of speech tagging information
Word class tagging information p with question bq The number of the part of speech tagging information of the union.
And the similarity set acquisition unit is used for acquiring a similarity set of the subgraph and the question data set according to the similarity.
And the external question acquiring unit is used for taking the question corresponding to the maximum similarity in the similarity set as the external question of the subgraph so as to acquire the external question of each subgraph.
In an optional embodiment, the external vector obtaining module 6 specifically includes:
and the vector set acquisition unit is used for acquiring the five question type vectors and the first word segmentation vector set through the word vector model according to the five question type parameters and the external question.
And the first vector splicing unit is used for carrying out vector longitudinal splicing on each word vector in the first word segmentation vector set and the five problem type vector sets to obtain a spliced word vector set.
And the external enhancement vector set acquisition unit is used for inputting the spliced word vector set into the BilSTM neural network model to acquire an external enhancement vector set.
In an optional embodiment, the vector set obtaining unit specifically includes:
and the five-question type vector obtaining subunit is used for inputting the five-question type parameters into the oneHot word vector model to obtain the five-question type vectors.
And the first participle vector acquisition subunit is used for inputting the external question into the PaddleNLP 'language solving' model, acquiring a first participle set of the external question and acquiring the first participle set of each subgraph.
And the first segmentation vector set acquisition subunit is used for inputting the first segmentation set into the GloVe word vector model to acquire a first segmentation vector set.
In an optional embodiment, the question generating module 7 specifically includes:
and the second word segmentation set acquisition unit is used for acquiring the target question corresponding to each subgraph, inputting the target question into a jieba word segmentation tool and acquiring a second word segmentation set.
And the second word segmentation vector set acquisition unit is used for inputting the second word segmentation set into the word vector model to acquire the second word segmentation vector set.
And the dictionary acquisition unit is used for extracting all words in the entity set, the relation set and the second participle set of the knowledge graph to acquire a dictionary.
And the first context vector acquisition unit is used for inputting the entity set of the sub-graph and the sub-graph vector into the Attention mechanism models of the N multi-head Attention blocks and acquiring the first context vector of the sub-graph.
And the second context vector acquisition unit is used for inputting the first word segmentation set of the external question and the external enhancement vector into the Attention mechanism models of the N multi-head Attention blocks and acquiring a second context vector fused with the five question type parameters and the external question.
And the fusion context vector acquisition unit is used for splicing the first context vector and the second context vector to acquire a fusion context vector.
A probability obtaining unit for combining the dictionary and the fusion context vector, and the hidden state h at the time of time series t in the Attention mechanism t And inputting a pointer to generate a network model, and acquiring the probability P (w) that the word generated at the time t is w. Wherein, the calculation model of the probability is as follows:
P(w)＝p*α copy +(1-p)*α vocab
p＝σ(W copy [h t ||c t ]+b copy )
α copy ＝a([h t ||c t ]，w)
α vooab ＝softmax(W vocab [h t ||c t ]+b vocab )
wherein p is a soft switch for selecting a word from a generated vocabulary from a subgraph or an external question, or selecting a word from a dictionary, and α copy For calculating the probability distribution, alpha, of words copied from subgraphs or external questions vocab For calculating the probability distribution of words copied from a dictionary, sigma-representing sigmoid function, W copy And W vocab Generating a feature matrix in a network model for pointers, b copy And b vocab Is an offset value.
And the predicted word set acquisition unit is used for taking the word w corresponding to the maximum probability P (w) at the time t as a predicted output word at the time t. And when w is an end marker, the question generation task is ended, so that a prediction word set for generating a sentence and a prediction word vector set corresponding to the word set are obtained.
And the loss calculation unit is used for calculating the loss L of the graph transformation network model according to the second participle vector set, all subgraphs of the knowledge graph and the predicted word vector set. Wherein, the calculation model of the loss is as follows:
wherein G is a knowledge graph, G is a subgraph of the knowledge graph, NLLLoss represents a negative log-likelihood loss function, V Pre Is a set of vectors for the second word,
is a first set of word vectors.
And the model updating unit is used for updating the network parameters of the graph transformation network model until the maximum iteration time T is reached by taking the minimum loss as a target so as to obtain a question generation model of the trained knowledge graph.
And the question generation unit is used for inputting the subgraph and the parameters of the types of the five questions into a question generation model of the knowledge graph to acquire the question.
Example III,
An embodiment of the present invention provides a question generation apparatus for a knowledge graph, which includes a processor, a memory, and a computer program stored in the memory. The computer program can be executed by a processor to implement the method for question generation of a knowledge-graph as described in any of the paragraphs above with reference to the embodiments.
Example four,
An embodiment of the present invention provides a computer-readable storage medium. The computer-readable storage medium includes a stored computer program, wherein when the computer program is executed, the apparatus in the computer-readable storage medium is controlled to execute the question generation method for a knowledge graph according to any one of the embodiments.
In the embodiments provided in the embodiments of the present invention, it should be understood that the disclosed apparatus and method may be implemented in other manners. The apparatus and method embodiments described above are merely illustrative and, for example, the flowchart and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of apparatus, methods and computer program products according to various embodiments of the present invention. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of code, which comprises one or more executable instructions for implementing the specified logical function(s). It should also be noted that, in some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams and/or flowchart illustration, and combinations of blocks in the block diagrams and/or flowchart illustration, can be implemented by special purpose hardware-based systems which perform the specified functions or acts, or combinations of special purpose hardware and computer instructions.
In addition, the functional modules in the embodiments of the present invention may be integrated together to form an independent part, or each module may exist separately, or two or more modules may be integrated to form an independent part.
The functions may be stored in a computer-readable storage medium if they are implemented in the form of software functional modules and sold or used as separate products. Based on such understanding, the technical solution of the present invention may be embodied in the form of a software product, which is stored in a storage medium and includes instructions for causing a computer device (which may be a personal computer, an electronic device, or a network device) to perform all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk, or an optical disk, and various media capable of storing program codes. It should be noted that, in this document, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus 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 apparatus. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other identical elements in a process, method, article, or apparatus that comprises the element.
The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention, and various modifications and changes may be made by those skilled in the art. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims (9)
1. A question generation method of a knowledge graph is characterized by comprising the following steps:
acquiring a knowledge graph; the knowledge graph comprises a set of subgraphs, each subgraph comprising an entity set, a relationship set and a triple set;
obtaining sub-graph vectors of each sub-graph through a graph transformation network model according to the knowledge graph;
acquiring a question data set;
acquiring external question sentences of each subgraph based on similarity according to the knowledge graph and the question data sets;
acquiring parameters of the types of the five problems;
acquiring an external enhancement vector through a BilSTM neural network model according to the five question type parameters and the external question;
and generating a network model through a pointer according to the sub-graph vector and the external enhancement vector to obtain a question.
2. The method for generating question sentences of a knowledge graph according to claim 1, wherein obtaining sub-graph vectors of each sub-graph through a graph transformation network model according to the knowledge graph, comprises:
inputting the entity set of the subgraph into a word vector model to obtain an entity vector set of the subgraph;
inputting the entity vector set into a graph attention network model to obtain intermediate layer vectors of all entities
Wherein, the calculation model of the intermediate layer vector is as follows:
in the formula (I), the compound is shown in the specification,
intermediate layer vector, v, being entity e e The information of N attention heads in the graph attention network model is serially connected for an entity vector of an entity e and a V-shaped representation,
Representing the attention parameter at the nth attention head
Is entity e in the graph attention network model i With entity e j Attention parameters of,
Is as followsA feature matrix in the n attention heads,
As entity e j The entity vector of (a),
As entity e i Entity vector, W Q And W K For calculating the feature matrix of the attention parameter,
As entity e i A set of neighbor nodes in the knowledge graph;
according to the intermediate layer vectors of all the entities, through the regularization layer and the feedforward neural network layer of the graph attention network model, the output vector representation of all the entities is obtained
Wherein the computational model of the output vector representation is:
FFN(x)＝W 1 ×(f(W 2 ×x+b 1 )+b 2 )
in the formula (I), the compound is shown in the specification,
is an output vector representation of entity e, LayerNorm is a normalization function, v' e Intermediate layer vector for entity e
Through layerHidden layer vector, W, obtained by rNorm function and feedforward neural network layer FFN 1 And W 2 Is a feature matrix in the feedforward neural network, x represents the input information of the feedforward neural network layer, b 1 And b 2 Is a bias value;
and obtaining the sub-image vector of each sub-image according to the output vector representation of each entity, thereby obtaining the sub-image vector of each sub-image.
3. The method for generating question sentences of a knowledge graph according to claim 1, wherein obtaining external question sentences of each subgraph based on similarity according to the knowledge graph and the question data set comprises:
inputting the relation set of the subgraph into a PaddleNLP 'language interpretation' model to obtain a relation part-of-speech tagging information set;
inputting the question data set into a PaddleNLP 'language interpretation' model to obtain a question word class tagging information set;
calculating the similarity between each part of speech tagging information in the question part of speech tagging information set and the relation part of speech tagging information set
Wherein, the similarity calculation model is as follows:
in the formula (I), the compound is shown in the specification,
representing relation part of speech tagging information
Word class tagging information p with question bq The number of the part-of-speech tagging information of the intersection,
Representing relation part of speech tagging information
Word class tagging information p with question bq The number of the part of speech tagging information of the union set;
acquiring a similarity set of a subgraph and the question data set according to the similarity;
and taking the question corresponding to the maximum similarity in the similarity set as an external question of the subgraph, thereby obtaining the external question of each subgraph.
4. The method for generating question sentences of a knowledge graph according to claim 1, wherein obtaining external enhancement vectors through a BilSTM neural network model according to the five question type parameters and the external question sentences comprises:
acquiring a five-question type vector and a first word segmentation vector set through a word vector model according to the five-question type parameters and the external question;
carrying out vector longitudinal splicing on each word vector in the first word segmentation vector set and the five problem type vectors to obtain a spliced word vector set;
and inputting the spliced word vector set into a BilSTM neural network model to obtain an external enhancement vector set.
5. The method for generating question sentences of a knowledge graph according to claim 4, wherein obtaining five question type vectors and a first word segmentation vector set by a word vector model according to the five question type parameters and the external question sentences comprises:
inputting the five question type parameters into an oneHot word vector model to obtain five question type vectors;
inputting the external question into a PaddleNLP 'language interpretation' model, and acquiring a first participle set of the external question so as to acquire a first participle set of each subgraph;
and inputting the first word segmentation set into a GloVe word vector model to obtain a first word segmentation vector set.
6. The method for generating question sentences of a knowledge graph according to claim 5, wherein generating a network model through a pointer according to the sub-graph vectors and the external enhanced vectors to obtain question sentences comprises:
acquiring a target question corresponding to each subgraph, inputting the target question into a jieba word segmentation tool, and acquiring a second word segmentation set;
inputting the second word segmentation set into a word vector model to obtain a second word segmentation vector set;
extracting all words in the entity set and the relation set of the knowledge graph and the second participle set to obtain a dictionary;
inputting the entity set and the sub-image vector of the sub-image into Attention mechanism models of N multi-head Attention blocks to obtain a first context vector of the sub-image;
inputting the first participle set of the external question and the external enhancement vector into Attention mechanism models of N multi-head Attention blocks, and acquiring five question type parameters and a second context vector fused with the external question;
splicing the first context vector and the second context vector to obtain a fusion context vector;
the dictionary and the fusion context vector are combined with the hidden state h at the time t of the time sequence in the Attention mechanism t Inputting a pointer generation network model, and acquiring the probability P (w) that a word generated at the moment t is w; wherein, the calculation model of the probability is as follows:
P(w)＝p*α copy +(1-p)*α vocab
p＝σ(W copy [h t ||c t ]+b copy )
α copy ＝a([h t ||c t ]，w)
α vocab ＝softmax(W vocab [h t ||c t ]+b vocab )
wherein p is a soft switch for selecting a word from a generated vocabulary from a subgraph or an external question, or selecting a word from a dictionary, and α copy For calculating the probability distribution, alpha, of words copied from subgraphs or external questions vocab For calculating the probability distribution of words copied from a dictionary, sigma-representing sigmoid function, W copy And W vocab Generating a feature matrix in a network model for pointers, b copy And b vocab Is a bias value;
taking the word w corresponding to the maximum probability P (w) at the time t as a prediction output word at the time t; when w is an end marker, the question generation task is ended, so that a prediction word set for generating a sentence and a prediction word vector set corresponding to the word set are obtained;
calculating the loss L of the graph transformation network model according to the second word segmentation vector set, all sub-graphs of the knowledge graph and the predicted word vector set; wherein the calculation model of the loss is:
wherein G is a knowledge graph, G is a subgraph of the knowledge graph, NLLLoss represents a negative log-likelihood loss function, V Pre Is a set of vectors for the second word,
is a first word vector set;
with the loss minimized as a target, updating the network parameters of the graph transformation network model until the maximum iteration times T is reached to obtain a question generation model of the trained knowledge graph;
and inputting the subgraph and the parameters of which question types into a question generation model of the knowledge graph to obtain a question.
7. A question generation device for a knowledge graph, comprising:
the knowledge graph acquisition module is used for acquiring a knowledge graph; the knowledge graph comprises a set of subgraphs, each subgraph comprising an entity set, a relationship set and a triple set;
the sub-image vector acquisition module is used for acquiring sub-image vectors of all sub-images through the image transformation network model according to the knowledge map;
the question set acquisition module is used for acquiring a question data set;
the external question acquisition module is used for acquiring external questions of all the subgraphs based on the similarity according to the knowledge graph and the question data set;
the five-question acquisition module is used for acquiring the five-question type parameters;
the external vector acquisition module is used for acquiring an external enhancement vector through a BilSTM neural network model according to the five question type parameters and the external question;
and the question generation module is used for generating a network model through a pointer according to the sub-graph vector and the external enhanced vector to obtain a question.
8. A knowledge-graph question generating apparatus comprising a processor, a memory, and a computer program stored in the memory; the computer program is executable by the processor to implement a method of knowledge-graph question generation as claimed in any one of claims 1 to 6.
9. A computer-readable storage medium, comprising a stored computer program, wherein the computer program, when executed, controls an apparatus on which the computer-readable storage medium is located to perform a method for generating a question of a knowledge-graph according to any one of claims 1 to 6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210566962.8A CN114925186A (en) | 2022-05-24 | 2022-05-24 | Knowledge graph question generation method, device, equipment and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210566962.8A CN114925186A (en) | 2022-05-24 | 2022-05-24 | Knowledge graph question generation method, device, equipment and storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114925186A true CN114925186A (en) | 2022-08-19 |
Family
ID=82810034
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210566962.8A Pending CN114925186A (en) | 2022-05-24 | 2022-05-24 | Knowledge graph question generation method, device, equipment and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114925186A (en) |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106897273A (en) * | 2017-04-12 | 2017-06-27 | 福州大学 | A kind of network security dynamic early-warning method of knowledge based collection of illustrative plates |
| CN107748757A (en) * | 2017-09-21 | 2018-03-02 | 北京航空航天大学 | A kind of answering method of knowledge based collection of illustrative plates |
| CN107766483A (en) * | 2017-10-13 | 2018-03-06 | 华中科技大学 | The interactive answering method and system of a kind of knowledge based collection of illustrative plates |
| CN112015868A (en) * | 2020-09-07 | 2020-12-01 | 重庆邮电大学 | Question-answering method based on knowledge graph completion |
| WO2021184311A1 (en) * | 2020-03-19 | 2021-09-23 | 中山大学 | Method and apparatus for automatically generating inference questions and answers |
| CN114254093A (en) * | 2021-12-17 | 2022-03-29 | 南京航空航天大学 | Multi-space knowledge enhanced knowledge graph question-answering method and system |
-
2022
- 2022-05-24 CN CN202210566962.8A patent/CN114925186A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106897273A (en) * | 2017-04-12 | 2017-06-27 | 福州大学 | A kind of network security dynamic early-warning method of knowledge based collection of illustrative plates |
| CN107748757A (en) * | 2017-09-21 | 2018-03-02 | 北京航空航天大学 | A kind of answering method of knowledge based collection of illustrative plates |
| CN107766483A (en) * | 2017-10-13 | 2018-03-06 | 华中科技大学 | The interactive answering method and system of a kind of knowledge based collection of illustrative plates |
| WO2021184311A1 (en) * | 2020-03-19 | 2021-09-23 | 中山大学 | Method and apparatus for automatically generating inference questions and answers |
| CN112015868A (en) * | 2020-09-07 | 2020-12-01 | 重庆邮电大学 | Question-answering method based on knowledge graph completion |
| CN114254093A (en) * | 2021-12-17 | 2022-03-29 | 南京航空航天大学 | Multi-space knowledge enhanced knowledge graph question-answering method and system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113761936B (en) | Multi-task chapter-level event extraction method based on multi-head self-attention mechanism | |
| CN108628823B (en) | Named entity recognition method combining attention mechanism and multi-task collaborative training | |
| CN108363743B (en) | Intelligent problem generation method and device and computer readable storage medium | |
| CN110245229B (en) | Deep learning theme emotion classification method based on data enhancement | |
| CN111858944B (en) | Entity aspect level emotion analysis method based on attention mechanism | |
| CN110807320B (en) | Short text emotion analysis method based on CNN bidirectional GRU attention mechanism | |
| CN112364660B (en) | Corpus text processing method, corpus text processing device, computer equipment and storage medium | |
| CN112541356B (en) | Method and system for recognizing biomedical named entities | |
| CN111274790B (en) | Chapter-level event embedding method and device based on syntactic dependency graph | |
| CN106778878B (en) | Character relation classification method and device | |
| CN113779211A (en) | Intelligent question-answer reasoning method and system based on natural language entity relationship | |
| CN111476038A (en) | Long text generation method and device, computer equipment and storage medium | |
| CN113128232B (en) | Named entity identification method based on ALBERT and multiple word information embedding | |
| CN113168499A (en) | Method for searching patent document | |
| US11727915B1 (en) | Method and terminal for generating simulated voice of virtual teacher | |
| CN113196277A (en) | System for retrieving natural language documents | |
| CN111274829A (en) | Sequence labeling method using cross-language information | |
| CN112905736A (en) | Unsupervised text emotion analysis method based on quantum theory | |
| CN111507093A (en) | Text attack method and device based on similar dictionary and storage medium | |
| CN114756681A (en) | Evaluation text fine-grained suggestion mining method based on multi-attention fusion | |
| CN114897167A (en) | Method and device for constructing knowledge graph in biological field | |
| CN114254645A (en) | Artificial intelligence auxiliary writing system | |
| CN114048314A (en) | Natural language steganalysis method | |
| Seilsepour et al. | Self-supervised sentiment classification based on semantic similarity measures and contextual embedding using metaheuristic optimizer | |
| CN116680407A (en) | Knowledge graph construction 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 |