WO2023050470A1

WO2023050470A1 - Event detection method and apparatus based on multi-layer graph attention network

Info

Publication number: WO2023050470A1
Application number: PCT/CN2021/123249
Authority: WO
Inventors: 包先雨; 吴共庆; 何俐娟; 柯培超; 陆振亚; 王歆; 程立勋; 蔡伊娜; 郑文丽; 慕容灏鼎; 蔡屹
Original assignee: 深圳市检验检疫科学研究院; 合肥工业大学
Priority date: 2021-09-30
Filing date: 2021-10-12
Publication date: 2023-04-06
Also published as: CN113887213A

Abstract

The present application provides an event detection method and apparatus based on a multi-layer graph attention network, the method comprising: acquiring a context word in event text information, and determining a syntactic information adjacency matrix and splicing vector corresponding to the context word; by using the adjacency matrix and the splicing vector as the input of an artificial neural network, acquiring an output vector; aggregating and generating aggregation information according to the splicing vector and the output vector; and, according to the aggregation information, determining a trigger word category of the context word. In the present application, by means of simultaneously combining syntactic information and context information of a context word, the problem that information loss and error propagation easily occur when using a syntactic analysis tool can be effectively solved; by means of combining a jump connection module into a graph attention network layer, the situation in which the classification of a final trigger word is not ideal due to excessive propagation of some short-distance syntactic information can be avoided, thereby effectively improving the accuracy, recall rate, and F1 value of trigger word classification.

Description

A method and device for event detection based on multi-layer graph attention network

technical field

The present application relates to the field of natural language processing, in particular to an event detection method and device based on a multi-layer graph attention network.

Background technique

Knowledge Graph describes concepts, entities and their relationships in the objective world in a structured form, and expresses Internet information in a form closer to the human cognitive world, providing a better way to organize, manage and understand the Internet. The capacity of massive amounts of information. The knowledge map was proposed by Google in 2012 and successfully applied to search engines. The knowledge map belongs to the important research field of artificial intelligence - the research category of knowledge engineering. It is a killer application of using knowledge engineering to build large-scale knowledge resources. Typical examples are the knowledge map launched by Google in 2012 after acquiring Freebase (a free knowledge database), the graph search of Facebook (social network service website), Microsoft Satori (Microsoft), and specific fields such as business, finance, and life sciences. knowledge base.

The event knowledge in the knowledge graph is implicit in Internet resources, including existing structured semantic knowledge, structured information in databases, semi-structured information resources, and unstructured resources. Different types of resources have different knowledge acquisition method. Event identification and extraction research is how to identify and extract event information from the text describing event information and present it in a structured form, including the time, place, participating roles, and related actions or states. Change.

Traditional event detection methods ignore the syntactic features contained between words in a sentence and only use sentence-level features. Event detection is prone to low recognition efficiency and classification accuracy of trigger words due to ambiguity in words. In recent years, the method of using syntactic information to improve event detection has been proved to be very effective. For example, the paper "No Trigger Word Event Detection Method Fused with Syntactic Information" proposes to use syntactic information and combine attention mechanism (ATTENTION) to realize the connection of scattered event information in sentences to improve the accuracy of event detection; the paper " Vietnamese News Event Detection by Fusion of Dependency Information and Convolutional Neural Network"Using the features between the convolutional codes of the fusion of dependency syntactic information to encode the features between discontinuous words, and then fusing the two parts of features as event codes to realize event detection.

Contents of the invention

In view of the problem, this application is proposed to provide a multi-layer graph attention network-based event detection method that overcomes the problem or at least partially solves the problem, comprising the steps of:

An event detection method based on a multi-layer graph attention network, comprising the steps of:

Acquiring the context words in the event text information, and determining the syntactic information adjacency matrix and splicing vector corresponding to the context words;

Using the adjacency matrix and the splicing vector as the input of the artificial neural network to obtain an output vector;

Aggregating and generating aggregation information according to the stitching vector and the output vector;

The trigger word category of the context word is determined according to the aggregation information.

Further, the step of acquiring the context words in the event text information and determining the syntactic information adjacency matrix and splicing vector corresponding to the context words includes:

determining syntactic information corresponding to the context word according to the context word;

generating the syntax information adjacency matrix according to the syntax information;

The stitching vector is generated according to the word embedding vector of the context word.

Further, the step of determining the syntactic information corresponding to the context word according to the context word includes:

The event text information is analyzed through syntactic dependence, and syntactic information corresponding to the context words is generated according to the analysis result of the event text information.

Further, the step of using the adjacency matrix and the splicing vector as the input of the artificial neural network to obtain the output vector includes:

Generate a tensor from the adjacency matrix of the same batch;

The tensor and the splicing vector are input into the artificial neural network for calculation, and the output vector is generated according to the calculation result of the artificial neural network.

Further, the step of determining the trigger word category of the context word according to the aggregation information includes:

The trigger words of the context words are determined according to the aggregation information, and the trigger words are classified according to the classifier module.

An event detection device based on a multi-layer graph attention network, comprising:

An acquisition module, configured to acquire context words in the event text information, and determine a syntactic information adjacency matrix and splicing vectors corresponding to the context words;

A calculation module, configured to use the adjacency matrix and the splicing vector as the input of the artificial neural network to obtain an output vector;

An aggregation module, configured to generate aggregation information according to the aggregation of the spliced vector and the output vector;

A classification module, configured to determine the trigger word category of the context word according to the aggregation information.

Further, the acquisition module includes:

An expression submodule, configured to determine syntactic information corresponding to the context word according to the context word;

A generating submodule, configured to generate the syntax information adjacency matrix according to the syntax information;

The splicing submodule is used to generate the splicing vector according to the word embedding vector of the context word.

Further, the expression submodule includes:

The dependency analysis sub-module is configured to analyze the event text information through syntactic dependencies, and generate syntactic information corresponding to the context words according to the analysis result of the event text information.

Further, the calculation module includes:

The array conversion submodule is used to generate a tensor from the adjacency matrix of the same batch;

The artificial neural network calculation sub-module is used to input the tensor and the stitching vector into the artificial neural network for calculation, and generate the output vector according to the calculation result of the artificial neural network.

Further, the classification module includes:

The trigger word processing submodule is configured to determine the trigger word of the context word according to the aggregation information, and classify the trigger word according to the classifier module.

This application has the following advantages:

In the embodiment of the present application, by obtaining the context words in the event text information, and determining the syntactic information adjacency matrix and splicing vector corresponding to the context words; using the adjacency matrix and the splicing vector as the artificial neural network Inputting, obtaining an output vector; generating aggregation information according to the splicing vector and the output vector; determining the trigger word category of the context word according to the aggregation information. This application can effectively solve the problem of easy information loss and error propagation when using syntactic analysis tools by combining the syntactic information and context information of context words at the same time; and by combining skip connection modules in the graph attention network layer, more can be retained The original feature avoids the unsatisfactory classification of the final trigger word due to the excessive spread of some short-distance syntactic information, and effectively improves the accuracy, recall rate and F1 value of the trigger word classification.

Description of drawings

In order to illustrate the technical solution of the present application more clearly, the accompanying drawings that need to be used in the description of the present application will be briefly introduced below. Obviously, the accompanying drawings in the following description are only some embodiments of the present application. Ordinary technicians can also obtain other drawings based on these drawings without paying creative labor.

Fig. 1 is a flow chart of the steps of an event detection method based on a multi-layer graph attention network provided by an embodiment of the present application;

FIG. 2 is a schematic diagram of a syntax dependency tree provided by an embodiment of the present application;

FIG. 3 is a schematic diagram of an adjacency matrix provided by an embodiment of the present application;

Fig. 4 is a schematic diagram of a graph attention network provided by an embodiment of the present application;

FIG. 5 is a schematic flow diagram of an event detection method based on a multi-layer graph attention network provided by an embodiment of the present application;

Fig. 6 is a structural block diagram of an event detection device based on a multi-layer graph attention network provided by an embodiment of the present application.

Detailed ways

In order to make the purpose, features and advantages of the present application more obvious and understandable, the present application will be further described in detail below in conjunction with the accompanying drawings and specific implementation methods. Apparently, the described embodiments are some of the embodiments of the present application, but not all of them. Based on the embodiments in this application, all other embodiments obtained by persons of ordinary skill in the art without creative efforts fall within the protection scope of this application.

Referring to Fig. 1, it shows a kind of event detection method based on multi-layer graph attention network that an embodiment of the application provides;

The methods include:

S110. Obtain context words in the event text information, and determine a syntactic information adjacency matrix and splicing vector corresponding to the context words;

S120. Using the adjacency matrix and the splicing vector as an input of the artificial neural network to obtain an output vector;

S130. Aggregate and generate aggregation information according to the stitching vector and the output vector;

S140. Determine the trigger word category of the context word according to the aggregation information.

In the following, the event detection method based on the multi-layer graph attention network in this exemplary embodiment will be further described.

As described in step S110, the context words in the event text information are obtained, and the syntactic information adjacency matrix and splicing vector corresponding to the context words are determined.

In an embodiment of the present application, the specific process of "obtaining the context words in the event text information and determining the syntactic information adjacency matrix and splicing vectors corresponding to the context words" in step S110 can be further described in conjunction with the following description.

As described in the following steps, determine the syntactic information corresponding to the context word according to the context word;

In an embodiment of the present application, the specific process of "determining syntactic information corresponding to the context word according to the context word" may be further described in conjunction with the following description.

As described in the following steps, the event text information is analyzed through syntactic dependencies, and syntactic information corresponding to the context words is generated according to the analysis result of the event text information.

It should be noted that the syntactic dependence is to reveal the syntactic structure by analyzing the interdependence relationship between the components in the language unit. The syntactic dependence analysis identifies the grammatical components such as "subject-verb-object" and "fixed complement" in the sentence, and emphasizes Analyze the relationship between words. The core of the sentence in syntactic dependency analysis is the predicate verb, and then find out other components around the predicate, and finally analyze the sentence into a syntactic dependency tree. The syntactic dependency tree can describe the dependency relationship between each word.

In a specific implementation, the event text information is obtained, the event text information is identified, and the syntax dependency analysis is performed using Stanford Core NLP (StandFord Natural Language Processing, Stanford Natural Language Processing Tool), and each sentence in the event text is analyzed and identified. The event trigger words in the sentence, and emphasize the analysis of the dependency relationship between the event trigger words and event parameters, and/or, the event parameters and event parameters, to form a syntactic dependency tree.

Among them, the event trigger word refers to the word that can best represent the occurrence of the event in an event. It is the projection of the concept of the event at the word and phrase levels. Noun; event parameters refer to information describing the time, place, and person of an event.

Referring to FIG. 2 , it shows a schematic diagram of a syntax dependency tree provided by an embodiment of the present application. As shown in the figure, the sentence "I went to Beijing Tiananmen Square to watch the sun rise", in the constructed syntactic dependency tree, we can see that the core predicate of the sentence is "go", which is the root of the syntactic dependency tree, and the subject of "go" is The object of "I" and "go" is "Beijing Tiananmen", and the object of another verb "look" is "sun". The syntactic dependency tree can describe the dependency relationship between context words.

As described in the following steps, generating the syntax information adjacency matrix according to the syntax information;

It should be noted that the adjacency matrix is a matrix representing the adjacency relationship between vertices. Suppose G=(V, E) is a graph, where V={v ₁ ,v ₂ ,…,v _n }, V is a vertex, E is an edge, use a one-dimensional array to store all vertex data in the graph, use a A two-dimensional array stores data of relationships (edges or arcs) between vertices, and this two-dimensional array is called an adjacency matrix. Adjacency matrix is divided into directed graph adjacency matrix and undirected graph adjacency matrix. The adjacency matrix of G is an nth-order square matrix with the following properties: For an undirected graph, the adjacency matrix must be symmetric, and the main diagonal must be zero, and the subdiagonal must not be 0, and the directed graph Not necessarily so. In an undirected graph, the degree of any vertex i is the number of all non-zero elements in the i-th column (or i-th row), and the out-degree of a vertex i in a directed graph is the number of all non-zero elements in the i-th row number, and the in-degree is the number of all non-zero elements in the i-th column, and the adjacency matrix of the directed graph is used to store the syntactic dependency between two event parameters.

As an example, each sentence forms a syntactic dependency tree through syntactic dependency analysis, and then generates a corresponding adjacency matrix according to the syntactic dependency tree.

In a specific implementation, refer to FIG. 3 , which shows a schematic diagram of an adjacency matrix provided by an embodiment of the present application. The adjacency matrix shown in FIG. 3 corresponds to the syntax dependency tree shown in FIG. 2 . The trigger word in Figure 2 is "go", "Beijing" and "Tiananmen" are parallel objects, so in the corresponding adjacency matrix, the intersection position of the row of "go" and the column value of "Beijing" and "Tiananmen" , the value is 1. Each word is used as a node, "I", "Go", "Beijing", "Tiananmen", "Look", "Sun", and "Rise" are seven words, so it is a 7X7 square matrix. If there is a syntactic arc between two words, the corresponding position of the matrix is 1, otherwise it is 0. The adjacency matrix of the directed graph is used to store the syntactic dependencies of the text. If there is a dependency between words, the corresponding adjacency matrix element value is 1, and between words without dependency relationship, the corresponding adjacency matrix element is 0. The dependency relationship between the context words can be represented by the adjacency matrix.

As described in the following steps, the stitching vector is generated according to the word embedding vector of the context word.

It should be noted that the word-level information in the sentence needs to be converted into a real-valued vector as the input of the artificial neural network. Let X = {x1,x2,x3,...,xn} be a sentence of length n, where xi is the ith word in the sentence. In natural language processing tasks, the semantic information of a word is related to its position in the sentence, and the part-of-speech and entity type information can improve the recognition of trigger words and the understanding of semantics. In this application, the concatenated vector formed by concatenating the meaning vector, entity vector, part-of-speech vector and position vector of the context word is used as the input of the artificial neural network.

In a specific implementation, four different word embedding vectors including the meaning vector, entity vector, part-of-speech vector and position vector of the context word are spliced into the first spliced vector, and then the first spliced vector is input to the Bi-LSTM neural network The network layer generates a second concatenated vector, which is used as one of the input vectors of the multi-layer graph attention network, and the concatenated vector can obtain semantic information between context words.

As described in step S120, the adjacency matrix and the concatenated vector are used as the input of the artificial neural network to obtain an output vector.

It should be noted that the artificial neural network is a multi-layer graph attention network (Graph Attention Networks). Due to the various limitations of the traditional graph convolutional network, it cannot handle directed graphs well, cannot be applied to inductive tasks (inductive tasks refer to: the graph structure that needs to be processed in the training phase and the testing phase is different) and cannot handle dynamic Graph, and the graph attention network can solve the defects of the graph convolutional network very well. For each node, the attention mechanism can be used to calculate the similarity coefficient of node j to node i, so that it does not need to rely entirely on the graph structure. It can also be applied to inductive tasks. Under the graph attention network, even if we change the structure of the graph during the prediction process, the impact on the graph attention network is not great, only need to adjust the parameters and recalculate. The operation method of the graph attention network is a vertex-by-vertex operation, and each operation needs to cycle through all the vertices on the graph to complete. Vertex-by-vertex operation means getting rid of the constraints of the Laplacian matrix in the original graph structure, so that the directed graph problem can be easily solved.

In an embodiment of the present application, the specific process of "using the adjacency matrix and the concatenated vector as the input of the artificial neural network to obtain the output vector" in step S120 can be further described in conjunction with the following description.

Generate a tensor from the adjacency matrix of the same batch as described in the following steps;

In a specific implementation, the sentences recognized at the same time in the event text information are a batch, and the adjacency matrix of the sentences of the same batch is formed into a tensor, and the adjacency matrix set is expressed as

Forming a tensor is expressed as A∈R ^N*N*K , where K=|T ^V |, N is the number of nodes.

As described in the following steps, the tensor and the concatenated vector are input to the artificial neural network for calculation, and the output vector is generated according to the calculation result of the artificial neural network.

As an example, referring to FIG. 4 , it shows a schematic diagram of a graph attention network provided by an embodiment of the present application, which is divided into two steps of calculating the attention coefficient and weighted summation. The tensor and the second concatenated vector are used as the input of the graph attention layer, expressed as

Where N is the number of nodes, F is the number of node features; the output is

where F' represents the new node feature vector dimension. Calculate the attention coefficient of node i and surrounding neighbor nodes j∈N _i , as shown on the left side of Figure 4, the calculation formula is as follows:

Among them, a is a mapping of ^RF′ ×RF ^′ →R, and W∈RF ^′×F is the weight matrix.

The graph attention network can use the attention mechanism to calculate the similarity coefficient weights between node i and neighbor node j for each node, so that it does not need to rely entirely on the graph structure.

The attention coefficient is normalized by softmax, and the calculation formula is expressed as follows:

Among them, || represents vector concatenation, both e _ij and α _ij are called "attention coefficients", and α _ij is normalized on the basis of e _ij .

After the attention coefficients of all nodes are normalized, the features of neighbor nodes are weighted and summed to generate an output vector. The calculation formula is as follows:

Among them, W is the weight matrix multiplied with the feature, σ is the nonlinear activation function, and the j traversed in j∈N _i represents all the nodes adjacent to i.

As shown on the right side of Figure 4, it is a three-layer graph attention network, and the multi-layer attention mechanism assigns different attention weights to different features. For a multi-layer graph attention network, the calculation formula is as follows:

If the multi-layer graph attention network is applied to the output layer, the calculation formula is expressed as follows:

As described in the step S130, aggregation information is generated according to the aggregation of the spliced vector and the output vector.

As an example, in the graph attention network of each layer, the aggregation of syntactic information is realized through the skip connection module (Skip-Connection), and the splicing vector is skipped through the graph attention of each layer through the skip connection module network, and perform an aggregation operation with the output vector. The over-spreading of short-distance syntactic information can be prevented through the skip connection module, more original syntactic information can be retained, and the final trigger word classification effect is avoided.

As described in the step S140, the trigger word category of the context word is determined according to the aggregation information.

In an embodiment of the present application, the specific process of "determining the trigger word category of the context word according to the aggregation information" in step S140 can be further described in conjunction with the following description.

As described in the following steps, the trigger word of the context word is determined according to the aggregation information, and the trigger word is classified according to the classifier module.

As an example, the trigger words of the context words are determined according to the aggregation information, the trigger words are classified according to the preset conditions of the classifier module, and the corresponding event sentences are determined according to the classification categories of the trigger words. event type. The event types are pre-defined different types.

Specifically, the preset condition of the classifier module is to aggregate the information of different modules, pass through a fully connected layer, and then map the output of multiple neurons to the (0,1) interval through the softmax function (softmax function, you can see into a probability to understand, so as to perform multi-classification) select the category with the largest category probability corresponding to each context word as the label of the current trigger word prediction.

The following is an experimental demonstration of an event detection method based on a multi-layer graph attention network proposed in the embodiment of the present application:

Experimental environment: Pytorch-1.8.0 (open source Python machine learning library), Nvidia GeForce RTX 3060 (graphics chip), Windows 10 (computer operating system), Inter i7-11700k, memory 16G, hard disk 1T.

The experimental data are shown in Table 1:

Table 1 Experimental comparison results

Experimental results: The experiment uses precision (Precision, P), recall (Recall, R), and F1 value (F1-score) as observation variables. The definitions of P, R, and F1 are as follows:

In order to ensure the accuracy of the experiment, the division of the data set in this experiment is consistent with the division of the data set of other event detection methods. The experimental results prove that the event detection method proposed in this embodiment is better than the traditional event detection method that only uses sentence-level features. In the detection method, the F1-score is about 8% higher; compared with the method based on the graph neural network, the event method proposed in this embodiment also achieved the highest values in F1-score and Recall.

Referring to FIG. 5 , a schematic flow diagram of an event detection method based on a multi-layer graph attention network is shown;

In a specific implementation, after the event text information is acquired, the event text information is analyzed by syntactic analysis technology to generate a syntactic dependency tree, and then an adjacency matrix corresponding to the context word is generated according to the syntactic dependency tree, and The adjacency matrix of the same batch of sentences generates a tensor; a total of 4 different word embedding vectors of the context words are spliced into the first spliced vector, and the first spliced vector is input to the Bi-LSTM neural network layer to generate The second splicing vector, inputting the adjacency matrix and the second splicing vector into a multi-layer graph attention network to generate an output vector to aggregate syntactic information of different depths; passing the splicing vector through a skip connection module Skip the multi-layer graph attention network to do the aggregation operation; aggregate the output vector and the splicing vector, and classify the trigger words of the context words through the classifier module to determine the event type corresponding to the event sentence.

As for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and for related parts, please refer to the part of the description of the method embodiment.

Referring to FIG. 6 , it shows an event detection device based on a multi-layer graph attention network provided by an embodiment of the present application;

Specifically include:

An acquisition module 610, configured to acquire context words in the event text information, and determine a syntactic information adjacency matrix and splicing vectors corresponding to the context words;

Calculation module 620, is used for using described adjacency matrix and described stitching vector as the input of artificial neural network, obtains output vector;

Aggregation module 630, configured to generate aggregation information according to the aggregation of the spliced vector and the output vector;

A classification module 640, configured to determine the trigger word category of the context word according to the aggregation information.

In an embodiment of the present application, the acquisition module 610 includes:

In an embodiment of the present application, the expression submodule includes:

In an embodiment of the present application, the calculation module 620 includes:

In an embodiment of the present application, the classification module 640 includes:

While the preferred embodiments of the embodiments of the present application have been described, additional changes and modifications can be made to these embodiments by those skilled in the art once the basic inventive concept is understood. Therefore, the appended claims are intended to be interpreted to cover the preferred embodiment and all changes and modifications that fall within the scope of the embodiments of the application.

Finally, it should also be noted that in this text, relational terms such as first and second etc. are only used to distinguish one entity or operation from another, and do not necessarily require or imply that these entities or operations, any such actual relationship or order exists. Furthermore, the term "comprises", "comprises" or any other variation thereof is intended to cover a non-exclusive inclusion such that a process, method, article, or terminal equipment comprising a set of elements includes not only those elements, but also includes elements not expressly listed. other elements identified, or also include elements inherent in such a process, method, article, or end-equipment. Without further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of additional identical elements in the process, method, article or terminal device comprising said element.

The event detection method and device based on a multi-layer graph attention network provided by this application have been introduced in detail above. In this paper, specific examples are used to illustrate the principle and implementation of this application. The description of the above embodiments It is only used to help understand the method of the present application and its core idea; at the same time, for those of ordinary skill in the art, according to the idea of the present application, there will be changes in the specific implementation and application scope. In summary, The contents of this specification should not be understood as limiting the application.

Claims

A kind of event detection method based on multi-layer graph attention network, it is characterized in that, comprising steps:

Acquiring the context words in the event text information, and determining the syntactic information adjacency matrix and splicing vector corresponding to the context words;

Using the adjacency matrix and the splicing vector as the input of the artificial neural network to obtain an output vector;

Aggregating and generating aggregation information according to the stitching vector and the output vector;

The trigger word category of the context word is determined according to the aggregation information.
The event detection method based on the multi-layer graph attention network according to claim 1, wherein the context words in the acquisition event text information are determined, and the syntactic information adjacency matrix and splicing vector corresponding to the context words are determined steps, including:

determining syntactic information corresponding to the context word according to the context word;

generating the syntax information adjacency matrix according to the syntax information;

The stitching vector is generated according to the word embedding vector of the context word.
The event detection method based on multi-layer graph attention network according to claim 2, wherein the step of determining the syntactic information corresponding to the context word according to the context word comprises:

The event text information is analyzed through syntactic dependence, and syntactic information corresponding to the context words is generated according to the analysis result of the event text information.
The event detection method based on the multi-layer graph attention network according to claim 1, wherein the step of using the adjacency matrix and the splicing vector as the input of the artificial neural network to obtain the output vector comprises:

Generate a tensor from the adjacency matrix of the same batch;

The tensor and the splicing vector are input into the artificial neural network for calculation, and the output vector is generated according to the calculation result of the artificial neural network.
The event detection method based on the multi-layer graph attention network according to claim 1, wherein the step of determining the trigger word category of the context word according to the aggregation information includes:

The trigger words of the context words are determined according to the aggregation information, and the trigger words are classified according to the classifier module.
An event detection device based on a multi-layer graph attention network, characterized in that it includes:

An acquisition module, configured to acquire context words in the event text information, and determine a syntactic information adjacency matrix and splicing vectors corresponding to the context words;

A calculation module, configured to use the adjacency matrix and the splicing vector as the input of the artificial neural network to obtain an output vector;

An aggregation module, configured to generate aggregation information according to the aggregation of the spliced vector and the output vector;

A classification module, configured to determine the trigger word category of the context word according to the aggregation information.
The event detection device based on the multi-layer graph attention network according to claim 6, wherein the acquisition module includes:

An expression submodule, configured to determine syntactic information corresponding to the context word according to the context word;

A generating submodule, configured to generate the syntax information adjacency matrix according to the syntax information;

The splicing submodule is used to generate the splicing vector according to the word embedding vector of the context word.
The event detection device based on multi-layer graph attention network according to claim 7, wherein the expression submodule comprises:

The dependency analysis sub-module is configured to analyze the event text information through syntactic dependencies, and generate syntactic information corresponding to the context words according to the analysis result of the event text information.
The event detection device based on the multi-layer graph attention network according to claim 6, wherein the calculation module includes:

The array conversion submodule is used to generate a tensor from the adjacency matrix of the same batch;

The artificial neural network calculation sub-module is used to input the tensor and the spliced vector into the artificial neural network for calculation, and generate the output vector according to the calculation result of the artificial neural network.
The event detection device based on multi-layer graph attention network according to claim 6, wherein the classification module includes:

The trigger word processing submodule is configured to determine the trigger word of the context word according to the aggregation information, and classify the trigger word according to the classifier module.