CN114385793A - Event extraction method and related device - Google Patents
Event extraction method and related device Download PDFInfo
- Publication number
- CN114385793A CN114385793A CN202210287573.1A CN202210287573A CN114385793A CN 114385793 A CN114385793 A CN 114385793A CN 202210287573 A CN202210287573 A CN 202210287573A CN 114385793 A CN114385793 A CN 114385793A
- Authority
- CN
- China
- Prior art keywords
- event
- vector
- embedding vector
- event type
- word
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/33—Querying
- G06F16/3331—Query processing
- G06F16/334—Query execution
- G06F16/3344—Query execution using natural language analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/33—Querying
- G06F16/3331—Query processing
- G06F16/334—Query execution
- G06F16/3346—Query execution using probabilistic model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/35—Clustering; Classification
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/10—Text processing
- G06F40/12—Use of codes for handling textual entities
- G06F40/126—Character encoding
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/30—Semantic analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Data Mining & Analysis (AREA)
- Computational Linguistics (AREA)
- Databases & Information Systems (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Software Systems (AREA)
- Probability & Statistics with Applications (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- Medical Informatics (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Machine Translation (AREA)
Abstract
The application discloses an event extraction method and a related device, wherein the method comprises the steps of obtaining an event type corresponding to a statement to be extracted; carrying out word coding and position coding on each word in the sentence to be extracted to obtain a word embedded vector and a position embedded vector corresponding to the sentence to be extracted; and adding the word embedding vector, the position embedding vector and the event type embedding vector corresponding to the event type to obtain a target embedding vector, and extracting the event parameters corresponding to the event types based on the target embedding vector. The method and the device have the advantages that the word embedding vector, the position embedding vector and the event type embedding vector corresponding to the event type of the statement are added to obtain the target embedding vector, the event argument of each event type is determined based on the target embedding vector, the event type is used as indicating information to be fused with statement information, semantic information carried by the target embedding vector is improved, and accordingly the extraction accuracy of the event parameters can be improved.
Description
Technical Field
The present application relates to the field of natural language processing technologies, and in particular, to an event extraction method and a related device.
Background
With the development of artificial intelligence and deep neural networks, natural language processing techniques are widely used, wherein event extraction in natural language processing techniques can automatically structure the daily generated primitive events on network media. An event refers to a thing which occurs in a specific time slice and a region range, is participated by one or more roles and consists of one or more actions, and is generally sentence-level. The goal of structuring is to determine the event type to which the event belongs and to extract event arguments for the event, e.g., participants, related entities, related times, and related values. However, due to the complexity and diversity of language words, a plurality of events may be involved in a sentence, however, the existing event extraction method generally needs trigger words to perform event extraction, and lacks attention to event type information, thereby affecting the accuracy of extracted event arguments.
Thus, the prior art has yet to be improved and enhanced.
Disclosure of Invention
The present application provides an event extraction method and a related device, aiming at the deficiencies of the prior art.
In order to solve the above technical problem, a first aspect of the embodiments of the present application provides an event extraction method, where the method includes:
acquiring an event type corresponding to a statement to be extracted;
performing word coding and position coding on each word in the sentence to be extracted to obtain a word embedding vector and a position embedding vector corresponding to the sentence to be extracted;
and adding the word embedding vector, the position embedding vector and the event type embedding vector corresponding to the event type to obtain a target embedding vector, and extracting an event parameter corresponding to the event type based on the target embedding vector.
The event extraction method, wherein the obtaining of the event type corresponding to the statement to be extracted specifically includes:
performing word coding and position coding on each word in the sentence to be extracted to obtain a word embedding vector and a position embedding vector corresponding to the sentence to be extracted;
adding the word embedding vector and the position embedding vector to obtain an input vector;
inputting the input vector into an event type encoder in a pre-trained event classification model, and outputting a first contextualized expression of the statement to be extracted through the event type encoder;
inputting the contextualized representation into an event type classifier in the event classification model, and determining an event type corresponding to the statement to be extracted through the event type classifier.
The event extraction method comprises the steps that the vector dimension of a word vector and the vector dimension of a position vector corresponding to each word in the statement to be extracted are both equal to the vector dimension of an event type embedding vector, and when the event type corresponding to the statement to be extracted comprises a plurality of event types, the vector dimensions of the event type embedding vector corresponding to each event type in the event types are all equal.
The event extraction method, wherein the event type corresponding to the statement to be extracted includes a plurality of event types, the adding the word embedding vector, the position embedding vector, and the event type embedding vector corresponding to the event type to obtain a target embedding vector, and extracting the event parameter corresponding to the event type based on the target embedding vector specifically includes:
for each event type, respectively adding each word vector of the word embedding vector, a position vector corresponding to each word vector in the position embedding vector and the event type embedding vector of the event type point by point to obtain a target embedding vector;
performing information fusion on the target embedding vector based on an attention mechanism to obtain a second contextualized expression;
and determining an event argument corresponding to the target embedded vector based on the second contextualized expression, and acquiring an event parameter corresponding to the event type based on the determined event argument.
The event extraction method, wherein the event type corresponding to the statement to be extracted includes a plurality of event types, the adding the word embedding vector, the position embedding vector, and the event type embedding vector corresponding to the event type to obtain a target embedding vector, and extracting the event parameter corresponding to the event type based on the target embedding vector specifically includes:
weighting the event type embedding vector corresponding to each event type in the event types to obtain a target event type embedding vector;
respectively adding each word vector of the word embedding vectors, the position vector corresponding to each word vector in the position embedding vectors and the target event type embedding vector point by point to obtain a target embedding vector;
performing information fusion on the target embedding vector based on an attention mechanism to obtain a second contextualized expression;
and determining event arguments corresponding to the target embedded vector based on the second contextualized expression, and acquiring event parameters corresponding to the event types based on the determined event arguments.
The event extraction method, wherein weighting the event type embedding vector corresponding to each event type in the multiple event types to obtain the target event type embedding vector specifically includes:
acquiring event probability corresponding to each event type, wherein the event probability is determined when the event type corresponding to the statement to be extracted is acquired;
and weighting the event type embedding vector based on the weighting coefficient of each event type embedding vector to obtain a target event type embedding vector.
The event extraction method, wherein weighting the event type embedding vector corresponding to each event type in the multiple event types to obtain the target event type embedding vector specifically includes:
reading preset weighting coefficients corresponding to the event types;
and weighting each event type embedding vector based on each read preset weighting coefficient to obtain a target event type embedding vector.
The event extraction method, wherein the obtaining of the event parameters corresponding to each event type based on the determined event arguments specifically includes:
the method comprises the steps of obtaining word attributes of each event argument in a plurality of event arguments, and forming a plurality of candidate event parameters based on the word attributes corresponding to the event arguments respectively, wherein each event parameter in the candidate event parameters comprises at least one event argument;
and event classification is carried out on the candidate event parameters to obtain the event parameters corresponding to the event types.
The event extraction method, wherein the obtaining of the event parameters corresponding to each event type based on the determined event arguments specifically includes:
the method comprises the steps of obtaining word attributes of each event argument in a plurality of event arguments, and forming a plurality of candidate event parameters based on the word attributes of the event arguments for drinking, wherein each event parameter in the candidate event parameters comprises at least one event argument;
for each event type, determining the vector distance between the event embedding vector of the event type and the parameter embedding vector of each candidate event parameter in a plurality of candidate event parameters;
and determining candidate event parameters corresponding to the event types based on the acquired vector distances to obtain the event parameters corresponding to the event types.
The event extraction method, wherein the information fusion of the target embedded vector based on the attention mechanism to obtain the second contextualized expression specifically includes:
and inputting the target embedded vector into an event argument coder in a pre-trained event extraction model, and performing information fusion on the target embedded vector through the event argument coder to obtain a second contextualized expression.
The event extraction method, wherein the determining, based on the second contextualized expression, an event argument corresponding to the target embedded vector specifically includes:
inputting the second contextualized expression into an event argument classifier in a pre-trained event extraction model, and determining an event argument corresponding to the target embedded vector through the event argument classifier.
A second aspect of the embodiments of the present application provides an event extraction device, where the event extraction device includes:
the acquisition module is used for acquiring the event type corresponding to the statement to be extracted;
the coding module is used for carrying out word coding and position coding on each word in the sentence to be extracted to obtain a word embedded vector and a position embedded vector corresponding to the sentence to be extracted;
and the extraction module is used for adding the word embedding vector, the position embedding vector and the event type embedding vector corresponding to the event type to obtain a target embedding vector, and extracting the event parameter corresponding to the event type based on the target embedding vector.
A third aspect of embodiments of the present application provides a computer readable storage medium storing one or more programs, which are executable by one or more processors to implement the steps in the event extraction method as described in any of the above.
A fourth aspect of the embodiments of the present application provides a terminal device, including: the device comprises a processor, a memory and a communication bus, wherein the memory is stored with a computer readable program which can be executed by the processor;
the communication bus realizes connection communication between the processor and the memory;
the processor, when executing the computer readable program, implements the steps in the event extraction method as described in any of the above.
Has the advantages that: compared with the prior art, the application provides an event extraction method and a related device, and the method comprises the steps of obtaining an event type corresponding to a statement to be extracted; performing word coding and position coding on each word in the sentence to be extracted to obtain a word embedding vector and a position embedding vector corresponding to the sentence to be extracted; and adding the word embedding vector, the position embedding vector and the event type embedding vector corresponding to the event type to obtain a target embedding vector, and extracting an event parameter corresponding to the event type based on the target embedding vector. The method and the device have the advantages that the word embedding vector of the statement, the position embedding vector and the event type embedding vector corresponding to the event type are added to obtain the target embedding vector, the event argument of each event type is determined based on the target embedding vector, the event type is used as indicating information and is fused with statement information, semantic information carried by the target embedding vector is improved, and accordingly the extraction accuracy of the event parameters can be improved. Meanwhile, the target embedded vector is determined by adding the word embedded vector, the position embedded vector and the event type embedded vector corresponding to the event type, so that the vector length of the target embedded vector is not increased when information carried by the event type is fused, the length of an input item is changed, and the expansibility of the extraction method is improved.
Drawings
In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed to be used in the description of the embodiments are briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present application, and it is obvious for those skilled in the art that other drawings can be obtained according to the drawings without any inventive work.
Fig. 1 is a flowchart of an event extraction method provided in the present application.
Fig. 2 is a schematic flow chart illustrating an event parameter obtained in a one-by-one obtaining manner in the event extraction method provided by the present application.
Fig. 3 is a schematic flow chart illustrating an example of acquiring event parameters in a one-time acquisition manner in the event extraction method provided by the present application.
Fig. 4 is a schematic flowchart of another example of acquiring event parameters in a one-time acquisition manner in the event extraction method provided by the present application.
Fig. 5 is a schematic structural diagram of an event extraction device provided in the present application.
Fig. 6 is a schematic structural diagram of a terminal device provided in the present application.
Detailed Description
In order to make the objects, technical solutions and effects of the present application clearer and clearer, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the present application and are not intended to limit the present application.
As used herein, the singular forms "a", "an", "the" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that the terms "comprises" and/or "comprising," when used in this specification, specify the presence of stated features, integers, steps, operations, elements, and/or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and/or groups thereof.
It will be understood by those within the art that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. It will be further understood that terms, such as those defined in commonly used dictionaries, should be interpreted as having a meaning that is consistent with their meaning in the context of the prior art and will not be interpreted in an idealized or overly formal sense unless expressly so defined herein.
It should be understood that, the sequence numbers and sizes of the steps in this embodiment do not mean the execution sequence, and the execution sequence of each process is determined by its function and inherent logic, and should not constitute any limitation on the implementation process of this embodiment.
The inventor finds that with the development of artificial intelligence and deep neural networks, natural language processing technology is widely applied, wherein event extraction in the natural language processing technology can automatically structure the daily generated native events on network media. An event refers to a thing which occurs in a specific time slice and a region range, is participated by one or more roles and consists of one or more actions, and is generally sentence-level. The goal of structuring is to determine the event type to which the event belongs and to extract event arguments for the event, e.g., participants, related entities, related times, and related values. However, due to the complexity and diversity of language words, a plurality of events may be involved in a sentence, however, the existing event extraction method generally needs trigger words to perform event extraction, and lacks attention to event type information, thereby affecting the accuracy of extracted event arguments.
In order to solve the above problem, in the embodiment of the present application, an event type corresponding to a statement to be extracted is obtained; performing word coding and position coding on each word in the sentence to be extracted to obtain a word embedding vector and a position embedding vector corresponding to the sentence to be extracted; and adding the word embedding vector, the position embedding vector and the event type embedding vector corresponding to the event type to obtain a target embedding vector, and extracting an event parameter corresponding to the event type based on the target embedding vector. The method and the device have the advantages that the word embedding vector of the statement, the position embedding vector and the event type embedding vector corresponding to the event type are added to obtain the target embedding vector, the event argument of each event type is determined based on the target embedding vector, the event type is used as indicating information and is fused with statement information, semantic information carried by the target embedding vector is improved, and accordingly the extraction accuracy of the event parameters can be improved. Meanwhile, the target embedding vector is determined by adding the word embedding vector, the position embedding vector and the event type embedding vector corresponding to the event type, so that the vector length of the target embedding vector is not increased when information carried by the event type is fused, and the length of the input item is changed.
The following further describes the content of the application by describing the embodiments with reference to the attached drawings.
The embodiment provides an event extraction method, as shown in fig. 1, the method includes:
and S10, acquiring the event type corresponding to the statement to be extracted.
Specifically, the statement to be extracted is a statement that needs to be event extracted, and the statement to be extracted may belong to one event type or a plurality of event types. Correspondingly, the event type corresponding to the statement to be extracted can be obtained by one event type, and a plurality of event types can also be obtained. In other words, acquiring the event type corresponding to the statement to be extracted refers to acquiring all event types corresponding to the statement to be extracted, and acquiring one event type when the statement to be extracted corresponds to one event type; and when the statement to be extracted corresponds to a plurality of event types, acquiring the plurality of event types. In a typical implementation manner, a statement to be extracted corresponds to a plurality of event types, and accordingly, the obtained event types are also a plurality of event types. For example, the event types corresponding to the statements to be extracted include attack events and death events, and then the extracted event types include attack event types and death event types.
In one implementation, the event type may be obtained through an event classification model, where the event classification model may include an event type encoder and an event type classifier, an output item of the event type encoder is an input item of the event type classifier, and the event type classifier outputs an event type corresponding to a statement to be extracted and an event probability corresponding to the event type. Based on this, the obtaining of the event type corresponding to the statement to be extracted specifically includes:
s11, performing word coding and position coding on each word in the sentence to be extracted to obtain a word embedding vector and a position embedding vector corresponding to the sentence to be extracted;
s12, adding the word embedding vector and the position embedding vector to obtain an input vector;
s13, inputting the input vector into an event type encoder in a pre-trained event classification model, and outputting a first contextualized expression of the statement to be extracted through the event type encoder;
and S14, inputting the contextualized representation into an event type classifier in the event classification model, and determining the event type corresponding to the statement to be extracted through the event type classifier.
Specifically, in step S11, the to-be-extracted sentence includes a plurality of words, each word forms a word vector after being subjected to word encoding, the position encoding is performed to encode position information of the word in the to-be-extracted sentence, and a position vector is formed after being subjected to position encoding. That is, after word encoding and position encoding are performed on each word in the sentence to be extracted, a word vector and a position vector are obtained. Further, it is worth to say that the position information refers to an appearance order of words in the sentence to be extracted, for example, the position information is an order in which words appear in the sentence to be extracted in an order from left to right. For example, the sentence to be extracted includes n words, which are sequentially denoted as w1, w2,. and wn from left to right; position information of n words in the order from left to right can be recorded as p1, p 2.,. pn, then performing word coding and position coding on the word w1 refers to performing word coding on the word w1, and performing position coding on the position information p1 corresponding to the word w1 to obtain a word vector (word embedding) and a position vector (position embedding) corresponding to the word w 1. In one implementation, the word code and the position code may be determined by a language model BERT, the sentence to be extracted passes through a word embedding layer of the BERT model to obtain a word vector corresponding to each word, and the position information corresponding to each word passes through the position embedding layer of the BERT model to obtain a position vector corresponding to each word.
After word vectors of respective drinking of all words are obtained, arranging the word vectors according to the sequencing sequence of all words in the sentence to be extracted to obtain a word embedding vector corresponding to the sentence to be extracted; and then, arranging the position vectors according to the sequence of the words in the statement to be extracted to obtain the position embedded vector corresponding to the statement to be extracted. In addition, it should be noted that the above words are a Chinese character, a word in foreign language, a character (e.g., Arabic numerals, etc.).
For example, the following steps are carried out: assuming that the sentence to be extracted is "d the y send him to Baghdad And kill", then performing word encoding And position encoding on each word in the "d the y send him to Baghdad And kill" to obtain a word vector And a position vector corresponding to each word, then ordering the word vectors of each word to obtain a word embedding vector corresponding to the sentence to be extracted, And ordering the position vectors of each word to obtain a position embedding vector.
In step S12, the word embedding vector and the position embedding vector are added to add each word vector in the word embedding vector and the position vector corresponding to each word vector, wherein the vector dimension of each word vector is equal to the vector dimension of the position vector corresponding to the word vector. Thus, the vector dimension of the word embedding vector and the vector dimension of the position embedding vector are both equal to the vector dimension of the input vector.
In step S13, the event classification model is a trained network model, and a first contextualized expression (contextualized expressions) of the sentence to be extracted can be determined by an event type encoder in the event classification model, so that the event type classifier can obtain semantic information of the sentence to be extracted, and thus can determine an event type corresponding to the sentence to be extracted. In one implementation, the event type encoder may employ a pre-trained Transformer language model, which is a machine learning model based on the attention mechanism, that is capable of processing all words or symbols in a text in parallel, while using the attention mechanism to combine the context with more distant words, by processing all words in parallel, and letting each word notice other words in the sentence in multiple processing steps. The input items of the Transformer language model are input vectors, and the output items are first contextualized expressions of the sentences to be extracted, wherein the first contextualized expressions of the sentences to be extracted comprise the first contextualized expressions of all the words in the sentences to be extracted. For example, as shown in FIG. 2, the input vector corresponding to the sentence to be extracted is "an d the y send him to Baghdad And killThe Transformer language model obtains a first contextualized expression corresponding to each word
Wherein, [ CLS]A start delimiter of a sentence is represented,
for pre-training specific flag bits [ CLS ] in language models]Is expressed in a first contextualization, [ CLS]A start delimiter is indicated;
indicates the first statement to be extracted
A first contextualized expression of an individual word.
In the step 14, the event type classifier may perform multi-label classification (multi-label classification), event types and event probabilities (probabilities) of the event types. In one implementation, the event type classifier may include a full link layer, where an input item of the full link layer is a first contextualized expression of the statement to be extracted, and an output item is each event type corresponding to the statement to be extracted and an event probability of each event type. For example, as shown in fig. 2, the event type classifier outputs the event types including an event type a, an event type b, an event type c, an event type d, and an event type e, where the event probability of the event type a is 0.1, the event probability of the event type b is 0.8, the event probability of the event type c is 0.6, the event probability of the event type d is 0.2, and the event probability of the event type e is 0.3.
In addition, after the event types determined by the event type classifiers and the event probabilities corresponding to the event types are obtained, multiple event types output by the event type classifiers can be screened based on a preset probability threshold, and the event types larger than the preset probability threshold are used as the event types corresponding to the statements to be extracted. For example, as shown in fig. 2, the preset probability threshold is set to 0.5, and if the event probability p _ b =0.8 of the event type b is greater than the threshold of 0.5, it is determined that the statement to be extracted corresponds to the event type b; and if the event probability p _ c =0.6 of the event type c is greater than the threshold value of 0.5, the statement to be extracted is considered to correspond to the event type c. Therefore, the statement to be extracted corresponds to both the event type class and the event type c.
And S20, performing word coding and position coding on each word in the sentence to be extracted to obtain a word embedding vector and a position embedding vector corresponding to the sentence to be extracted.
Specifically, the determining process of the word embedding vector and the position embedding vector is the same as the determining process of the word embedding vector and the position embedding vector in the process of obtaining the event type of the to-be-extracted statement, and is not repeated here. The event type determining process and the event argument extracting process are divided into two independent processes in the embodiment, and the two independent processes respectively and independently perform word embedding vector and position embedding vector acquisition, so that the method provided by the embodiment has better expansibility. It is to be noted that the word embedding vector and the position embedding vector may be directly determined in the event type obtaining process, and after the word embedding vector and the position embedding vector are determined in the event type obtaining process, the obtained word embedding vector and the obtained position embedding vector may be stored, and then directly read when extracting event arguments, so as to save the processes of word encoding and position encoding and improve the extraction speed of event arguments.
S30, adding the word embedding vector, the position embedding vector and the event type embedding vector corresponding to the event type to obtain a target embedding vector, and extracting the event parameter corresponding to the event type based on the target embedding vector.
Specifically, the vector dimension of the event type embedding vector of the event type, the vector dimension of the word vector corresponding to each word in the statement to be extracted, and the vector dimension of the position vector corresponding to each word are all the same. It can be understood that the vector dimension of the word vector of each word in the sentence to be extracted is equal to the vector dimension of the corresponding position vector, the vector dimension of the word vector of each word is equal, and the vector dimension of the event type embedding vector of the event type is equal to the vector dimension of the word vector of the word in the sentence to be extracted. For example, the statement to be extracted includes a word a and a word b, the statement to be extracted corresponds to an event type c, and then the vector dimension of the word vector w1 corresponding to the word a, the vector dimension of the word vector w2 corresponding to the word b, the vector dimension of the position vector p1 corresponding to the word a, the vector dimension of the position vector p1 corresponding to the word a, and the vector dimension of the event type embedding vector e1 corresponding to the event type c are all the same.
In addition, when the event type corresponding to the statement to be extracted is a multiple event type, the vector dimensions of the event type embedding vectors corresponding to each event type in the multiple event types are equal. For example, the statement to be extracted corresponds to the event type c1 and the event type c2, and then the vector dimension of the event type embedding vector e1 corresponding to the event type c1 is equal to the vector dimension of the event type embedding vector e2 corresponding to the event type c 2.
The event type corresponding to the statement to be extracted includes multiple event types, and determining the event parameter corresponding to the event type refers to determining the event parameter corresponding to each event type, where the event parameter corresponding to each event type in the multiple event types may be obtained in a one-by-one obtaining manner or obtained in a one-time obtaining manner.
In one implementation, the step S30: adding the word embedding vector, the position embedding vector and the event type embedding vector corresponding to the event type to obtain a target embedding vector, and extracting event parameters corresponding to the event type based on the target embedding vector specifically comprises:
s311, for each event type, respectively adding each word vector of the word embedding vector, a position vector corresponding to each word vector in the position embedding vector and the event type embedding vector of the event type point by point to obtain a target embedding vector;
s312, performing information fusion on the target embedded vector based on an attention mechanism to obtain a second contextualized expression;
s313, determining event arguments corresponding to the target embedded vector based on the second contextualized expression, and acquiring event parameters corresponding to the event types based on the determined event arguments.
Specifically, in step S311, the target embedding vector is obtained by adding the word embedding vector, the position embedding vector, and the event type embedding vector, and the vector dimension of the target embedding vector is equal to the vector position of the word embedding vector, so that the influence of the event type increase on the vector dimension of the target embedding vector can be avoided. Adding the word embedding vector, the position embedding vector and the event type embedding vector into each word vector in the word embedding vector, and performing point-by-point addition on the position vector and the event type embedding vector corresponding to each word vector. For example, as shown in FIG. 2, the to-be-extracted statement is "an the y send him to Baghdad And kill," And the word embedding vector, the location embedding vector, And the event type embedding vector are added to a word vector in which the words are embedded in the vectors, respectively
Position vector corresponding to word vector
And event type embedding type
The point-by-point addition is performed. I.e. word vectors where words are embedded in the vector
Position vector corresponding to word vector
And event type embedding type
A word vector in the word embedding vector is added point by point
Position vector corresponding to word vector
And event type embedding type
The point-by-point addition is performed. The event type embedding vector is added with the word embedding vector and the position embedding vector to determine a target embedding vector, and the target embedding vector is used as an input item for determining event parameters, so that the event type can be used as indicating information to assist in extracting the event parameters, the event type can be fused with information carried by the word embedding vector and the position embedding vector, and the accuracy of extracting the event parameters is improved. In addition, the target embedding vector is determined by adding the word embedding vector, the position embedding vector and the event type embedding vector, the dimensionality of an input item in the event parameter extraction process is unchanged, the vector dimensionality of the input item cannot be influenced by the increase of the random event type, and therefore the event parameter extraction process is made to have expansibility.
The event parameters can be obtained through a pre-trained event extraction model, wherein the event extraction model can comprise an event argument encoder and an event argument classifier, an output item of the event type encoder is an input item of the event type classifier, and the event type classifier outputs event arguments corresponding to the event type to obtain the event parameters corresponding to the event type.
Based on this, the step S312: performing information fusion on the target embedded vector based on an attention mechanism to obtain a second contextualized expression specifically as follows:
and inputting the target embedded vector into an event argument coder in a pre-trained event extraction model, and performing information fusion on the target embedded vector through the event argument coder to obtain a second contextualized expression.
Specifically, the event extraction model is a trained network model, and semantic information carried by a word vector carried by a target embedded vector, semantic information carried by a position vector and semantic information carried by an event type embedded vector can be fused by an event argument encoder in the event extraction model, so that the event extraction classifier can obtain a second contextualized expression (to-be-extracted statement) of the statement to be extracted, and the second contextualized expression carries the semantic information carried by the event type embedded vector, so that the semantic information carried by the event type embedded vector can be used as indication information to assist in extracting event parameters. In one implementation, the event argument coder may employ a pre-trained Transformer language model, which is a machine learning model based on the attention mechanism, that is capable of processing all words or symbols in a text in parallel, while using the attention mechanism to combine the context with more distant words, by processing all words in parallel, and letting each word notice other words in the sentence in multiple processing steps. And the input items of the Transformer language model are target embedded vectors, and the output items are second contextualized expressions of the sentences to be extracted, wherein the second contextualized expressions of the sentences to be extracted comprise the second contextualized expressions of all the words in the sentences to be extracted.
The process of fusing the target embedded vector by the Transformer language model can be as follows: first, a query vector q, a key vector k, and a value vector v are obtained from a target embedded vector, and one is calculated for the query vector q, the key vector k, and the value vector v
(ii) a Second, the score is normalized by dividing the score by
Dim is the vector dimension; thirdly, applying a softmax activating function to the score, multiplying the softmax point by the Value v to obtain a weighted score v of each input vector, and finally adding the scores of the input vectors to obtain a final output result
。
The step S313: based on the second contextualized expression, determining an event argument corresponding to the target embedded vector specifically includes:
inputting the second contextualized expression into an event argument classifier in a pre-trained event extraction model, and determining an event argument corresponding to the target embedded vector through the event argument classifier.
Specifically, the event type classifier may perform sequential task labeling on each term to obtain an event argument corresponding to the target embedding vector. In one implementation, the event type classifier may include a full-link layer, where an input item of the full-link layer is a second contextualized expression of the statement to be extracted, and an output item is an event argument corresponding to the target embedded vector. For example, if the event type of the statement to be extracted is an attack event, the event argument corresponding to the target embedded vector includes time, place, person, and the like in the attack event. In addition, after the event arguments are acquired, all the acquired event arguments are used as the event parameters of the event type, so that the event parameters corresponding to the event types can be acquired by acquiring the event arguments corresponding to the event types one by one. In other words, the process of step S311 to step S313 is performed once for each event type corresponding to the statement to be extracted, so as to obtain the event parameter corresponding to each event type.
In one implementation, the step S30: the adding the word embedding vector, the position embedding vector and the event type embedding vector corresponding to the event type to obtain a target embedding vector, and extracting the event parameter corresponding to the event type based on the target embedding vector specifically includes:
s321, weighting the event type embedding vectors corresponding to the event types in the event types to obtain target event type embedding vectors;
s322, respectively adding each word vector of the word embedding vectors, the position vector corresponding to each word vector in the position embedding vectors and the target event type embedding vector point by point to obtain a target embedding vector;
s323, performing information fusion on the target embedded vector based on an attention mechanism to obtain a second contextualized expression;
s324, based on the second contextualized expression, determining event arguments corresponding to the target embedded vector, and acquiring event parameters corresponding to the event types based on the determined event arguments.
Specifically, the target event type embedding vector is obtained by weighting event embedding vectors of a plurality of event types, so that the target event type embedding vector carries semantic information carried by each event embedding vector, wherein a weighting coefficient of the event embedding vector of each event type in the weighting process may be preset or determined according to an event probability corresponding to each event type. In this embodiment, the target event type embedding vector is obtained by weighting the multiple event types, and the result of adding the target event type embedding vector, the word embedding vector, and the position embedding vector is used as the input of the encoder, so that the event parameters corresponding to the event types can be obtained in one step, and the extraction efficiency of the event parameters of the multiple event types can be improved.
Based on this, in one implementation, the step S321: weighting the event type embedding vector corresponding to each event type in the multiple event types to obtain a target event type embedding vector specifically comprises:
acquiring event probability corresponding to each event type, wherein the event probability is determined when the event type corresponding to the statement to be extracted is acquired;
and weighting the event type embedding vector based on the weighting coefficient of each event type embedding vector to obtain a target event type embedding vector.
Specifically, the event probability is determined when the event type corresponding to the statement to be extracted is obtained, that is, when the event type of the statement to be extracted is obtained, the event probabilities of the event types are obtained synchronously. For example, as shown in fig. 3, the event type corresponding to the statement to be extracted includes an event type b and an event type c, the event probability of the event type b is P _ b, the event type embedding vector of the event type b is E _ b, the event probability of the event type c is P _ c, and the event type embedding vector of the event type c is E _ c, and then the target event type embedding vector E = P _ b _ E _ b + P _ c _ E _ c.
In addition, the target event type embedding vector is obtained by weighting the event type embedding vectors of multiple event types, so that the vector dimension of the target event type embedding vector is equal to the vector dimension of the event type embedding vector of each event type, and the vector dimensions of the obtained target event type embedding vectors are equal no matter whether the number of the event types corresponding to each statement to be extracted is the same or not, so that the statement to be extracted with each multiple event type can be used by the event extraction model.
In one implementation, the step S321: weighting the event type embedding vector corresponding to each event type in the multiple event types to obtain a target event type embedding vector specifically comprises:
reading preset weighting coefficients corresponding to the event types;
and weighting each event type embedding vector based on each read preset weighting coefficient to obtain a target event type embedding vector.
Specifically, the preset weighting coefficient corresponding to each event type may be preset, for example, a preset weighting coefficient sequence is preset, and a preset weighting coefficient is randomly selected for each event type in the preset weighting coefficient sequence; or, a preset weighting coefficient is preset, and the preset weighting coefficient is used as a weighting coefficient of each event type, and the like. In a typical implementation, the preset weighting coefficients corresponding to the event types are equal. For example, as shown in fig. 4, the statement to be extracted corresponds to an event type b and an event type c, the event type embedding vector of the event type b is E _ b, the event type embedding vector of the event type c is E _ c, the weighting coefficient of the event type b and the weighting coefficient of the event type c are 0.5, and then the target event type embedding vector E =0.5 × E _ b +0.5 × E _ c.
In addition, the process of performing point-by-point addition on each word vector of the word embedding vectors, the position vector corresponding to each word vector in the position embedding vectors, and the target event type embedding vector in step S322 is the same as the process of performing point-by-point addition on each word vector of the word embedding vectors, the position vector corresponding to each word vector in the position embedding vectors, and the event type embedding vector in step S311, which is not repeated herein, and specific reference may be made to the specific description of step S311. Meanwhile, the execution process of step S323 is the same as the execution process of step S312, and the process of "determining the event argument corresponding to the target embedded vector based on the second contextualized expression" in step S324 is the same as the execution process of step S313, which may specifically refer to the above description, and is not repeated here.
Further, after the target event type embedding vector is adopted to determine the target embedding vector to determine the event argument, the determined event argument comprises event arguments of multiple event types, so that event parameters corresponding to the event types need to be selected from the determined event arguments. Based on this, when determining the event argument corresponding to the target embedded vector, it is necessary to obtain the event parameters corresponding to each event type based on the determined event argument.
In one implementation, the obtaining of the event parameter corresponding to each event type based on the determined event argument specifically includes:
the method comprises the steps of obtaining word attributes of each event argument in a plurality of event arguments, and forming a plurality of candidate event parameters based on the word attributes corresponding to the event arguments respectively, wherein each event parameter in the candidate event parameters comprises at least one event argument;
and event classification is carried out on the candidate event parameters to obtain the event parameters corresponding to the event types.
Specifically, each of the several event arguments is a word, and the word has a word attribute, e.g., subject, predicate, object, complement, or the like. After obtaining the statement attribute of each event argument, determining all candidate statements which can be formed by a plurality of event arguments according to the word attribute corresponding to each event argument, then performing event classification on each formed candidate statement to obtain a candidate event type corresponding to each candidate statement, finally matching the candidate event type corresponding to each candidate statement with each event type corresponding to the statement to be extracted, and taking the candidate statement corresponding to the candidate event type matched with the event type as the event parameter corresponding to the event type to obtain the event parameter corresponding to each event type. When event classification is performed on each formed candidate sentence, event classification and the like can be performed through a pre-trained event classification model.
In one implementation, the obtaining of the event parameter corresponding to each event type based on the determined event argument specifically includes:
the method comprises the steps of obtaining word attributes of each event argument in a plurality of event arguments, and forming a plurality of candidate event parameters based on the word attributes of the event arguments for drinking, wherein each event parameter in the candidate event parameters comprises at least one event argument;
for each event type, determining the vector distance between the event embedding vector of the event type and the parameter embedding vector of each candidate event parameter in a plurality of candidate event parameters;
and determining candidate event parameters corresponding to the event types based on the acquired vector distances to obtain the event parameters corresponding to the event types.
Specifically, the determination process of several candidate event parameters is the same as that of the candidate event parameters in the above embodiment, and is not repeated here. After the candidate event parameters are obtained, parameter embedding vectors of the candidate event parameters are obtained, then for each parameter embedding vector, the distance between the parameter embedding vector and each event type embedding vector is calculated, and the candidate event parameters are distributed to the event type with the nearest distance, so that the event parameters corresponding to each event type are obtained.
In summary, the embodiment provides an event extraction method, which includes acquiring an event type corresponding to a statement to be extracted; performing word coding and position coding on each word in the sentence to be extracted to obtain a word embedding vector and a position embedding vector corresponding to the sentence to be extracted; and adding the word embedding vector, the position embedding vector and the event type embedding vector corresponding to the event type to obtain a target embedding vector, and extracting an event parameter corresponding to the event type based on the target embedding vector. The method and the device have the advantages that the word embedding vector of the statement, the position embedding vector and the event type embedding vector corresponding to the event type are added to obtain the target embedding vector, the event argument of each event type is determined based on the target embedding vector, the event type is used as indicating information and is fused with statement information, semantic information carried by the target embedding vector is improved, and accordingly the extraction accuracy of the event parameters can be improved. Meanwhile, the target embedded vector is determined by adding the word embedded vector, the position embedded vector and the event type embedded vector corresponding to the event type, so that the vector length of the target embedded vector is not increased when information carried by the event type is fused, the length of an input item is changed, and the expansibility of the extraction method is improved.
Based on the above event extraction method, this embodiment provides an event extraction device, as shown in fig. 5, where the device includes:
an obtaining module 100, configured to obtain an event type corresponding to a statement to be extracted;
the encoding module 200 is configured to perform word encoding and position encoding on each word in the sentence to be extracted to obtain a word embedding vector and a position embedding vector corresponding to the sentence to be extracted;
an extracting module 300, configured to add the word embedding vector, the position embedding vector, and the event type embedding vector corresponding to the event type to obtain a target embedding vector, and extract an event parameter corresponding to the event type based on the target embedding vector.
Based on the above event extraction method, the present embodiment provides a computer-readable storage medium storing one or more programs, which are executable by one or more processors to implement the steps in the event extraction method according to the above embodiment.
Based on the above event extraction method, the present application further provides a terminal device, as shown in fig. 6, which includes at least one processor (processor) 20; a display screen 21; and a memory (memory) 22, and may further include a communication Interface (Communications Interface) 23 and a bus 24. The processor 20, the display 21, the memory 22 and the communication interface 23 can communicate with each other through the bus 24. The display screen 21 is configured to display a user guidance interface preset in the initial setting mode. The communication interface 23 may transmit information. The processor 20 may call logic instructions in the memory 22 to perform the methods in the embodiments described above.
Furthermore, the logic instructions in the memory 22 may be implemented in software functional units and stored in a computer readable storage medium when sold or used as a stand-alone product.
The memory 22, which is a computer-readable storage medium, may be configured to store a software program, a computer-executable program, such as program instructions or modules corresponding to the methods in the embodiments of the present disclosure. The processor 20 executes the functional application and data processing, i.e. implements the method in the above-described embodiments, by executing the software program, instructions or modules stored in the memory 22.
The memory 22 may include a storage program area and a storage data area, wherein the storage program area may store an operating system, an application program required for at least one function; the storage data area may store data created according to the use of the terminal device, and the like. Further, the memory 22 may include a high speed random access memory and may also include a non-volatile memory. For example, a variety of media that can store program codes, such as a usb disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk, or an optical disk, may also be transient storage media.
In addition, the specific processes loaded and executed by the storage medium and the instruction processors in the terminal device are described in detail in the method, and are not stated herein.
Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit the same; although the present application has been described in detail with reference to the foregoing embodiments, it should be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions in the embodiments of the present application.
Claims (14)
1. An event extraction method, characterized in that the method comprises:
acquiring an event type corresponding to a statement to be extracted;
performing word coding and position coding on each word in the sentence to be extracted to obtain a word embedding vector and a position embedding vector corresponding to the sentence to be extracted;
and adding the word embedding vector, the position embedding vector and the event type embedding vector corresponding to the event type to obtain a target embedding vector, and extracting an event parameter corresponding to the event type based on the target embedding vector.
2. The event extraction method according to claim 1, wherein the obtaining of the event type corresponding to the statement to be extracted specifically comprises:
performing word coding and position coding on each word in the sentence to be extracted to obtain a word embedding vector and a position embedding vector corresponding to the sentence to be extracted;
adding the word embedding vector and the position embedding vector to obtain an input vector;
inputting the input vector into an event type encoder in a pre-trained event classification model, and outputting a first contextualized expression of the statement to be extracted through the event type encoder;
inputting the contextualized representation into an event type classifier in the event classification model, and determining an event type corresponding to the statement to be extracted through the event type classifier.
3. The event extraction method according to claim 1, wherein a vector dimension of a word vector and a vector dimension of a position vector corresponding to each word in the sentence to be extracted are both equal to a vector dimension of the event type embedding vector, and wherein when the event type corresponding to the sentence to be extracted includes a plurality of event types, the vector dimensions of the event type embedding vector corresponding to each event type in the plurality of event types are both equal.
4. The event extraction method according to claim 1, wherein the event types corresponding to the to-be-extracted statement include a plurality of event types, and the adding the word embedding vector, the position embedding vector, and the event type embedding vector corresponding to the event type to obtain a target embedding vector, and extracting the event parameters corresponding to the event type based on the target embedding vector specifically includes:
for each event type, respectively adding each word vector of the word embedding vector, a position vector corresponding to each word vector in the position embedding vector and the event type embedding vector of the event type point by point to obtain a target embedding vector;
performing information fusion on the target embedding vector based on an attention mechanism to obtain a second contextualized expression;
and determining an event argument corresponding to the target embedded vector based on the second contextualized expression, and acquiring an event parameter corresponding to the event type based on the determined event argument.
5. The event extraction method according to claim 1, wherein the event types corresponding to the to-be-extracted statement include a plurality of event types, and the adding the word embedding vector, the position embedding vector, and the event type embedding vector corresponding to the event type to obtain a target embedding vector, and extracting the event parameters corresponding to the event type based on the target embedding vector specifically includes:
weighting the event type embedding vector corresponding to each event type in the event types to obtain a target event type embedding vector;
respectively adding each word vector of the word embedding vectors, the position vector corresponding to each word vector in the position embedding vectors and the target event type embedding vector point by point to obtain a target embedding vector;
performing information fusion on the target embedding vector based on an attention mechanism to obtain a second contextualized expression;
and determining event arguments corresponding to the target embedded vector based on the second contextualized expression, and acquiring event parameters corresponding to the event types based on the determined event arguments.
6. The event extraction method according to claim 5, wherein weighting the event type embedding vector corresponding to each of the plurality of event types to obtain the target event type embedding vector specifically comprises:
acquiring event probability corresponding to each event type, wherein the event probability is determined when the event type corresponding to the statement to be extracted is acquired;
and weighting the event type embedding vector based on the weighting coefficient of each event type embedding vector to obtain a target event type embedding vector.
7. The event extraction method according to claim 5, wherein weighting the event type embedding vector corresponding to each of the plurality of event types to obtain the target event type embedding vector specifically comprises:
reading preset weighting coefficients corresponding to the event types;
and weighting each event type embedding vector based on each read preset weighting coefficient to obtain a target event type embedding vector.
8. The event extraction method according to claim 5, wherein the obtaining of the event parameters corresponding to each event type based on the determined event arguments specifically comprises:
the method comprises the steps of obtaining word attributes of each event argument in a plurality of event arguments, and forming a plurality of candidate event parameters based on the word attributes corresponding to the event arguments respectively, wherein each event parameter in the candidate event parameters comprises at least one event argument;
and event classification is carried out on the candidate event parameters to obtain the event parameters corresponding to the event types.
9. The event extraction method according to claim 5, wherein the obtaining of the event parameters corresponding to each event type based on the determined event arguments specifically comprises:
the method comprises the steps of obtaining word attributes of each event argument in a plurality of event arguments, and forming a plurality of candidate event parameters based on the word attributes of the event arguments for drinking, wherein each event parameter in the candidate event parameters comprises at least one event argument;
for each event type, determining the vector distance between the event embedding vector of the event type and the parameter embedding vector of each candidate event parameter in a plurality of candidate event parameters;
and determining candidate event parameters corresponding to the event types based on the acquired vector distances to obtain the event parameters corresponding to the event types.
10. The event extraction method according to claim 4 or 5, wherein the information fusion of the target embedding vector based on the attention mechanism to obtain the second contextualized expression is specifically:
and inputting the target embedded vector into an event argument coder in a pre-trained event extraction model, and performing information fusion on the target embedded vector through the event argument coder to obtain a second contextualized expression.
11. The event extraction method according to claim 4 or 5, wherein the determining the event argument corresponding to the target embedding vector based on the second contextualized expression specifically comprises:
inputting the second contextualized expression into an event argument classifier in a pre-trained event extraction model, and determining an event argument corresponding to the target embedded vector through the event argument classifier.
12. An event extraction device, said device comprising:
the acquisition module is used for acquiring the event type corresponding to the statement to be extracted;
the coding module is used for carrying out word coding and position coding on each word in the sentence to be extracted to obtain a word embedded vector and a position embedded vector corresponding to the sentence to be extracted;
and the extraction module is used for adding the word embedding vector, the position embedding vector and the event type embedding vector corresponding to the event type to obtain a target embedding vector, and extracting the event parameter corresponding to the event type based on the target embedding vector.
13. A computer-readable storage medium storing one or more programs, the one or more programs being executable by one or more processors to perform the steps of the event extraction method as claimed in any one of claims 1-11.
14. A terminal device, comprising: a processor, a memory, and a communication bus; the memory has stored thereon a computer readable program executable by the processor;
the communication bus realizes connection communication between the processor and the memory;
the processor, when executing the computer readable program, implements the steps in the event extraction method of any of claims 1-11.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210287573.1A CN114385793B (en) | 2022-03-23 | 2022-03-23 | Event extraction method and related device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210287573.1A CN114385793B (en) | 2022-03-23 | 2022-03-23 | Event extraction method and related device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114385793A true CN114385793A (en) | 2022-04-22 |
| CN114385793B CN114385793B (en) | 2022-07-08 |
Family
ID=81205975
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210287573.1A Active CN114385793B (en) | 2022-03-23 | 2022-03-23 | Event extraction method and related device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114385793B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114757189A (en) * | 2022-06-13 | 2022-07-15 | 粤港澳大湾区数字经济研究院(福田) | Event extraction method and device, intelligent terminal and storage medium |
| CN115062615A (en) * | 2022-04-29 | 2022-09-16 | 中证信用增进股份有限公司 | Financial field event extraction method and device |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110781302A (en) * | 2019-10-23 | 2020-02-11 | 清华大学 | Method, device and equipment for processing event role in text and storage medium |
| CN111475617A (en) * | 2020-03-30 | 2020-07-31 | 招商局金融科技有限公司 | Event body extraction method and device and storage medium |
| US20220067030A1 (en) * | 2020-08-31 | 2022-03-03 | Microsoft Technology Licensing, Llc | Learning graph representations using hierarchical transformers for content recommendation |
-
2022
- 2022-03-23 CN CN202210287573.1A patent/CN114385793B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110781302A (en) * | 2019-10-23 | 2020-02-11 | 清华大学 | Method, device and equipment for processing event role in text and storage medium |
| CN111475617A (en) * | 2020-03-30 | 2020-07-31 | 招商局金融科技有限公司 | Event body extraction method and device and storage medium |
| US20220067030A1 (en) * | 2020-08-31 | 2022-03-03 | Microsoft Technology Licensing, Llc | Learning graph representations using hierarchical transformers for content recommendation |
Non-Patent Citations (1)
| Title |
|---|
| 魏优等: "基于深层语境词表示与自注意力的生物医学事件抽取", 《计算机工程与科学》, no. 09, 15 September 2020 (2020-09-15) * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115062615A (en) * | 2022-04-29 | 2022-09-16 | 中证信用增进股份有限公司 | Financial field event extraction method and device |
| CN114757189A (en) * | 2022-06-13 | 2022-07-15 | 粤港澳大湾区数字经济研究院(福田) | Event extraction method and device, intelligent terminal and storage medium |
| CN114757189B (en) * | 2022-06-13 | 2022-10-18 | 粤港澳大湾区数字经济研究院(福田) | Event extraction method and device, intelligent terminal and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114385793B (en) | 2022-07-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109299273B (en) | Multi-source multi-label text classification method and system based on improved seq2seq model | |
| WO2023024412A1 (en) | Visual question answering method and apparatus based on deep learning model, and medium and device | |
| KR102189688B1 (en) | Mehtod for extracting synonyms | |
| CN114385793B (en) | Event extraction method and related device | |
| CN109977416A (en) | A kind of multi-level natural language anti-spam text method and system | |
| CN107506414A (en) | A kind of code based on shot and long term memory network recommends method | |
| CN112084381A (en) | Event extraction method, system, storage medium and equipment | |
| CN112749274B (en) | Chinese text classification method based on attention mechanism and interference word deletion | |
| CN114490953B (en) | Method for training event extraction model, method, device and medium for extracting event | |
| CN114036950B (en) | Medical text named entity recognition method and system | |
| CN111241232A (en) | Business service processing method and device, service platform and storage medium | |
| CN113705196A (en) | Chinese open information extraction method and device based on graph neural network | |
| CN112183106A (en) | Semantic understanding method and device based on phoneme association and deep learning | |
| CN113609857A (en) | Legal named entity identification method and system based on cascade model and data enhancement | |
| CN114492661B (en) | Text data classification method and device, computer equipment and storage medium | |
| CN113095072B (en) | Text processing method and device | |
| CN110852066A (en) | Multi-language entity relation extraction method and system based on confrontation training mechanism | |
| CN117591543B (en) | SQL sentence generation method and device for Chinese natural language | |
| CN117558270B (en) | Voice recognition method and device and keyword detection model training method and device | |
| CN114266252A (en) | Named entity recognition method, device, equipment and storage medium | |
| US11941360B2 (en) | Acronym definition network | |
| CN115526176A (en) | Text recognition method and device, electronic equipment and storage medium | |
| KR20200141419A (en) | Mehtod for extracting synonyms | |
| CN116912579A (en) | Scene graph generation method based on multi-level attention mechanism | |
| CN116756605A (en) | ERNIE-CN-GRU-based automatic speech step recognition method, system, equipment and medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |