CN112633483B - Four-tuple gate diagram neural network event prediction method, device, equipment and medium - Google Patents

Abstract

The embodiment of the application relates to a four-element portal map neural network event prediction method, device, equipment and medium, and aims to improve the traditional event prediction precision. The method comprises the following steps: forming a rational map by a plurality of initial background events and a plurality of events to be selected; representing vectors of all events in the event map in a four-element form to obtain an initial background event vector and an initial candidate event vector; performing graph network calculation on the event map by using a four-element portal graph neural network to obtain a plurality of new background event vectors and a plurality of new event vectors to be selected; calculating the vector of the event by using the attention neural network to obtain the whole vector of the background event; and scoring the whole vector and each new candidate event vector, and taking the candidate event corresponding to one candidate event vector with the highest score as a prediction result.

Description

Four-tuple gate diagram neural network event prediction method, device, equipment and medium

technical field

The embodiments of the present application relate to the technical field of data processing, and in particular, to a method, device, device and medium for predicting events in a quaternary gate diagram neural network.

Background technique

Scripted event prediction is an important research direction in the field of artificial intelligence. Understanding scripted events is an important step in realizing real artificial intelligence. Specifically, the scripted event prediction task is to choose one from multiple possible answers based on the context that has emerged. Standard answer, script event prediction can be applied in reading comprehension, intent recognition, and dialogue management. In the existing script event prediction methods, the context is mainly modeled, and the event prediction is realized by using the model.

In the existing technology, one problem is that the different components of the event are directly spliced, and the relationship between the different components within the event cannot be well captured. Another problem is that modeling based on events or event chains cannot It captures the interplay between different events very well.

Contents of the invention

Embodiments of the present application provide a quadruple gate diagram neural network event prediction method, device, equipment and medium. , which aims to improve the accuracy of traditional event prediction.

The first aspect of the embodiment of the present application provides a quadruple gate diagram neural network event prediction method, the method comprising:

Constitute multiple initial background events and multiple candidate events to form an event map;

Representing the vectors of the plurality of initial background events and the plurality of initial candidate events in the event map in the form of quadruples to obtain an initial background event vector and an initial candidate event vector;

Using the quaternion gate graph neural network to perform graph network calculation on the event map to obtain a plurality of new background event vectors and a plurality of new candidate event vectors;

Using the attention neural network to calculate the multiple new background event vectors and the multiple new candidate event vectors to obtain the overall vector of the background event;

Score the overall vector and each of the new candidate event vectors, and use the candidate event with the highest score as the prediction result.

Optionally, multiple initial background events and multiple candidate events constitute an event map, including:

Set the relationship between multiple initial background events and multiple events to be selected;

Taking multiple initial background events and multiple candidate events as nodes, and taking the relationship between multiple initial background events and multiple candidate events as edges, an event map is formed.

Optionally, the vectors of the plurality of initial background events and the plurality of initial candidate events in the event map are represented in the form of quadruples to obtain the initial background event vector and the initial candidate event vector, including:

Express the vectors of all events including the multiple initial background events and multiple initial candidate events in the form of v( _es , e _o , e _p ), where v represents a predicate verb, and e _s represents a subject , e _o represents the object, and e _p represents an entity that has a prepositional relationship with the predicate verb.

Optionally, before using the quaternion gate graph neural network to perform graph network calculation on the event map to obtain multiple new background event vectors and multiple new candidate event vectors, the method further includes:

Collecting a plurality of related events, marking a part of the plurality of related events as a background event, and marking the other part as a candidate event as a training set;

The training set is input into the quadruple gate diagram neural network to train the quadruple gate diagram neural network to obtain a trained quadruple gate diagram neural network.

Optionally, use the quaternion gate graph neural network to perform graph network calculation on the event map to obtain multiple new background event vectors and multiple new candidate event vectors, including:

Inputting the representation vector of the event in the event map and the adjacency matrix representing the relationship between events into the quadruple gate graph neural network;

The quaternion gating neural network calculates the representation vector of the event and the adjacency matrix to obtain the multiple new background event vectors and the multiple new candidate event vectors.

Optionally, the attention neural network is used to calculate the multiple new background event vectors and the multiple new candidate event vectors to obtain the overall vector of the background event, including:

Inputting the plurality of new background event vectors and the plurality of new candidate event vectors into the attention neural network;

For each new event vector to be selected in the plurality of new event vectors to be selected, perform an attention mechanism operation on each new background event vector in the plurality of new background event vectors, and obtain The weight coefficient of each new background event vector in the plurality of new background event vectors relative to each new candidate event vector;

According to the weight coefficient, the overall vector of the background event is obtained through calculation.

Optionally, scoring the overall vector and each of the new candidate event vectors, and using the candidate event corresponding to the highest-scoring candidate event vector as the prediction result includes:

According to the overall vector of the background event, calculate the Euclidean distance between each new event vector to be selected and the overall vector of the background event to obtain a plurality of Euclidean distance values;

Selecting the new candidate event vector corresponding to the minimum value among the values of multiple Euclidean distances as the candidate event vector with the highest score, and taking the candidate event corresponding to the candidate event vector with the highest score as the prediction result .

The second aspect of the embodiment of the present application provides a quadruple gate graph neural network event prediction device, the device comprising:

An event map constituting module, which is used to form an event map with multiple initial background events and multiple events to be selected;

The quadruple event representation module is used to represent the vectors of the plurality of initial background events and the plurality of initial candidate events in the event map in the form of quadruples, and obtain the initial background event vector and the initial candidate events event vector;

The quadruple gate diagram neural network module is used to use the quadruple gate diagram neural network to perform graph network calculations on the event map to obtain a plurality of new background event vectors and a plurality of new candidate event vectors;

The background fusion module based on the attention mechanism is used to calculate the multiple new background event vectors and the multiple new candidate event vectors by using the attention neural network to obtain the overall vector of the background event;

The background event and candidate event scoring module is configured to score the overall vector and each of the new candidate event vectors, and use the candidate event corresponding to the candidate event vector with the highest score as the prediction result.

Optionally, the constituting module of the event map includes:

A relationship setting submodule, used to set the relationship between multiple initial background events and multiple events to be selected;

The event map constitutes a sub-module, which is used to form an event map with multiple initial background events and multiple candidate events as nodes, and with relationships between multiple initial background events and multiple candidate events as edges.

Optionally, the quadruple event representation module includes:

The four-tuple event representation submodule is used to represent the vectors of all events including the multiple initial background events and multiple initial candidate events in the form of v( _es , e _o , e _p ), Among them, v represents the predicate verb, _es represents the subject, e _o represents the object, and e _p represents an entity that has a prepositional relationship with the predicate verb.

Optionally, the device also includes:

An event collection module, configured to collect a plurality of related events, mark a part of the plurality of related events as background events, and mark the other part as candidate events, as a training set;

The quadruple gate diagram neural network training module is used to input the training set into the quadruple gate diagram neural network to train the quadruple gate diagram neural network to obtain the trained quadruple gate Graph neural network.

Optionally, the quadruple gate graph neural network module includes:

The first vector input submodule is used to input the representation vector of the event in the event map and the adjacency matrix representing the relationship between the events into the quadruple gate graph neural network;

The four-atom gate graph neural network calculation submodule is used for the quadruple gate graph neural network to calculate the representation vector and the adjacency matrix of the event, and obtain the multiple new background event vectors and the multiple a new event vector to be selected.

Optionally, the background fusion module based on the attention mechanism includes:

The second vector input submodule is used to input the multiple new background event vectors and the multiple new candidate event vectors into the attention neural network;

The weight coefficient calculation sub-module is used for combining each new background event vector among the multiple new background event vectors with each new candidate event vector among the multiple new candidate event vectors Perform an attention mechanism operation to obtain the weight coefficient of each new background event vector in the plurality of new background event vectors relative to each of the new candidate event vectors;

The overall vector obtaining sub-module is used to calculate and obtain the overall vector of the background event according to the weight coefficient.

Optionally, the background event and candidate event scoring module includes:

The Euclidean distance calculation submodule is used to calculate the Euclidean distance between each new event vector to be selected and the overall vector of the background event according to the overall vector of the background event, and obtain a plurality of Euclidean distance values ;

The prediction result obtaining submodule selects the new candidate event vector corresponding to the minimum value among the values of multiple Euclidean distances as the candidate event vector with the highest score, and assigns the event vector corresponding to the highest score candidate event vector The event to be selected is used as the prediction result.

The third aspect of the embodiment of the present application provides a readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, the steps in the method described in the first aspect of the present application are implemented.

The fourth aspect of the embodiment of the present application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and operable on the processor. When the processor executes the computer program, the first aspect of the present application is realized. The steps of the method described in the aspect.

Using the quaternion gate graph neural network event prediction method provided by this application, firstly, the collected events constitute an event map, the collected events include background events and candidate events, and the event map of each event in the event map The four components in the vector correspond to the four components of the quadruple data, and the vector at each time is represented in the form of quadruples, and the events in the event map are analyzed by using the trained gate graph neural network model The vector is calculated to obtain a new event vector, which contains a new background event vector and a new candidate event vector. The attention mechanism is used to calculate the importance of each background event vector to each candidate event vector, according to The importance of each background event vector to each candidate event vector is obtained to obtain the overall vector of the background event, calculate the Euclidean distance between each candidate event vector and the overall background event vector, and select the candidate with the closest distance to the overall background event vector Event vectors as predictions. The present invention uses quadruples to represent events, which just correspond to the components of the event, and better capture the mutual influence between different components inside the event, and use the gate graph neural network to build the interaction between events The model can better capture the mutual influence between events, and use the attention mechanism to fuse the background events and calculate with the candidate events. The influence of each background event on the candidate events can be considered, and the prediction is more accurate.

Description of drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the accompanying drawings that need to be used in the description of the embodiments of the present application. Obviously, the accompanying drawings in the following description are only some embodiments of the present application , for those skilled in the art, other drawings can also be obtained according to these drawings without paying creative labor.

Fig. 1 is the flow chart of the quadruple gate graph neural network event prediction method that an embodiment of the present application proposes;

Fig. 2 is the calculation process of different components of the Hamiltonian operator operation based on the quaternion representation model proposed by an embodiment of the present invention;

FIG. 3 is a schematic diagram of an event representation based on a quadruple representation model proposed by an embodiment of the present invention;

Fig. 4 is a schematic diagram of an attention mechanism operation based on an attention mechanism background event and an event to be selected according to an embodiment of the present invention;

Fig. 5 is a schematic diagram of an attention mechanism operation based on an attention mechanism background event and a candidate event proposed by an embodiment of the present invention;

FIG. 6 is a schematic diagram of a scoring model for calculating the Euclidean distance based on weighted and summed background events and candidate events proposed by an embodiment of the present application;

Fig. 7 is the training flow diagram of the quadruple gate graph neural network that an embodiment of the present application proposes;

FIG. 8 is a schematic diagram of a quadruple gate graph neural network event prediction device proposed by an embodiment of the present application.

Detailed ways

The following will clearly and completely describe the technical solutions in the embodiments of the present application with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, 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 , FIG. 1 is a flow chart of a quadruple gate diagram neural network event prediction method proposed by an embodiment of the present application. As shown in Figure 1, the method includes the following steps:

S11: Construct a plurality of initial background events and a plurality of candidate events into an event map.

In this embodiment, the initial background event can be understood as an event that has already occurred, the initial candidate event is an event that may occur next corresponding to the initial background event, and the event map is composed of multiple events and the relationship between events Atlas.

In this embodiment, the specific steps of forming an event graph from multiple initial background events and multiple candidate events include:

S11-1: Setting the relationship between multiple initial background events and multiple events to be selected.

In this embodiment, multiple events including the initial background event and the initial candidate event are all related to each other and have a certain relationship, and the relationship between the events needs to be set in advance.

For example, one event is "entering a store" and the other event is "shopping", then the purpose of "entering a store" is "shopping", and "purpose" can be recorded as the relationship between "entering a store" and "shopping" .

S11-2: Taking multiple initial background events and multiple candidate events as nodes, and taking relationships between multiple initial background events and multiple candidate events as edges to form an event map.

In this embodiment, the event map is a knowledge base containing event logic, which describes the evolution rule and evolution mode between events. The structure of the event map is a directed and cyclic graph, in which nodes represent relationships, and directed edges represent logical relations between events, such as sequence, causality, conditions, and upper and lower levels.

In this embodiment, all events including multiple initial background events and multiple candidate events are used as the nodes of the event map, and the relationship between each event is used as the edge of the event map to form the event map, which can be clearly expressed The logical relationship between various events is more conducive to modeling and calculation of events.

S12: Represent the vectors of the plurality of initial background events and the plurality of initial candidate events in the event map in the form of quadruples to obtain an initial background event vector and an initial candidate event vector.

In this embodiment, the vectors of all events including the multiple initial background events and multiple initial candidate events are expressed in the form of v( _es , e _o , e _p ), where v represents a predicate verb , es _represents the subject, e _o represents the object, and e _p represents an entity that has a prepositional relationship with the predicate verb.

In this embodiment, the quadruple data structure is an extension of complex data, complex data includes a real part and an imaginary part, and the quadruple data structure has a real part and three imaginary parts, that is, Q1=a1+b1i+c1j +d1k, where a1 is the real part, b1, c1, and d1 are the imaginary parts, i, j, and k represent imaginary numbers, and the general model of event representation generally has four components, namely v( _es , e _o , e _p ), where v represents the predicate verb, _es represents the subject, e _o represents the object, and e _p represents an entity that has a prepositional relationship with the predicate verb. Let a1 correspond to v, b1, c1, d1 correspond to e _s , e _o , e _p respectively, by utilizing the similarity between the quadruple data structure and the event representation model in structure, use the quadruple data structure to model the event Relationships within events can be captured.

In this embodiment, as shown in Figure 2, Figure 2 is the calculation process of different components of the Hamiltonian operator operation based on the quadruple representation model proposed by an embodiment of the present invention, and the quadruple data is performed using the Hamiltonian operator Operation, the influence of the interaction between events can be well captured through weight sharing. In Figure 2, _Qin represents the input, Q _out represents the output, W represents the total weight, W _s , W _o , W _p are the weights corresponding to e _s , e _o , e _p , v', e' _s , e' _o , e' _p are the vector representations of v, e _s , e _o , e _p combined with the influence of each part respectively. The specific calculation process can be described by the following formula:

Where Q ₁ =a ₁ +b ₁ i+c ₁ j+d ₁ k, Q ₂ =a ₂ +b ₂ i+c ₂ j+d ₂ k.

In this embodiment, as shown in FIG. 3 , FIG. 3 is a schematic diagram of an event representation based on a quadruple representation model proposed by an embodiment of the present invention, where the shades of colors represent different components of each vector.

For example, there is an event in the event map as "I go to the mall to buy clothes" where "I" is the subject e _s , "clothes" is the object e _o , "shop" is the entity e _p that has a prepositional relationship with the predicate verb, "To buy" is the predicate verb v.

S13: Using the quaternion gate graph neural network to perform graph network calculation on the event map to obtain a plurality of new background event vectors and a plurality of new candidate event vectors.

In this embodiment, the quaternion gate graph neural network is a new type of neural network that combines the gate structure in the recurrent neural network with the graph neural network, wherein the gate structure is similar to that of the GRU, and the graph neural network is a A general neural network architecture for processing data with a graph structure. Different script events have multiple interactions. In order to combine the interaction of different events and the internal dependencies of events, in this embodiment, the quaternion and the gated neural network are combined with the graph neural network to obtain the quaternion gate graph neural network. The network can better learn the feature representation of events.

In this embodiment, the specific steps for obtaining multiple new background event vectors and multiple new candidate event vectors are:

S13-1: Input the representation vectors of the events in the event graph and the adjacency matrix representing the relationship between events into the gate graph neural network.

In this embodiment, using the quaternion gate graph neural network to perform graph network calculation on the event map, first, the vector h ⁽⁰⁾ and the adjacency matrix A of the initial background event and the initial candidate event need to be input into the neural network .

For example, suppose there are 8 background events and 5 candidate events, then h ⁽⁰⁾ represents the initial vectors of these 8 background events and 5 candidate events, and the adjacency matrix A∈R ^13×13 represents the 13 Interrelationships between events.

S13-2: The gate graph neural network calculates the event representation vector and the adjacency matrix to obtain the multiple new background event vectors and the multiple new candidate event vectors.

In this embodiment, as shown in FIG. 4 , FIG. 4 is a schematic diagram of an attention mechanism-based background event and a candidate event for an attention mechanism operation proposed by an embodiment of the present invention, in which nodes e ₁ , e ₁ . . . , _en represents different event vectors, and each edge represents the relationship between events.

The initial background event vector and the initial candidate event vector only express the events in the form of vectors, and do not model the relationship between them. The background event vector and the candidate event vector are input into the quadruple gate diagram neural network for Calculated, the new background event vector and the candidate event vector are obtained by fusing information between different nodes, and are new vectors fusing the influence between various events.

In this embodiment, the expression vector and adjacency matrix of the event are calculated by using the quaternion gate diagram neural network to obtain multiple new background event vectors and multiple new candidate event vectors. The specific calculation method is:

a ^(t) = A ^T h ^(t-1) + b (2)

h ^t ＝(1-z ^t )⊙h ^(t-1) +z ^t ⊙c ^t (6)

表示哈密顿算子，⊙表示元素级别的乘法，a^(t)是一个中间量，A^T代表邻接矩阵的转置，h^(t-1)代表第t个向量的上一个向量，b代表偏置量，/>

Q₁、Q₂都表示四元组数值，σ()代表四元组划分的sigmoid激活函数，tanh()代表四元组划分的tanh激活函数，z^t代表更新门，r^t代表重置门，c^t代表中间量，h^t代表新的背景事件向量或新的待选事件向量。in,

Indicates the Hamiltonian operator, ⊙ indicates element-level multiplication, a ^(t) is an intermediate quantity, A ^T represents the transpose of the adjacency matrix, h ^(t-1) represents the previous vector of the t-th vector, and b represents the bias set amount, />

Both Q ₁ and Q ₂ represent the quadruple value, σ() represents the sigmoid activation function of the quadruple division, tanh() represents the tanh activation function of the quadruple division, z ^t represents the update gate, and r ^t represents the reset gate , c ^t represents the intermediate quantity, h ^t represents the new background event vector or the new candidate event vector.

In the present embodiment, what formula (2) represented is the transfer process of information between the different nodes of the quadruple gate graph neural network, and what formula (3)-(6) represented is from remaining node and current node before The vectors are updated together to get a new vector. There is a certain number of steps K in the cyclic propagation process in the above formula, and the number of steps is determined by the number of vectors. After the calculation of the above formula, the quadruple gate graph neural network outputs a new background event vector and a new candidate event vector. The new background event vector can be represented by h _i , and the new candidate event vector can be represented by h _cj to express.

As an example, after inputting the vectors of 8 initial background events and 5 initial background events and the adjacency matrix representing the relationship between these 13 events into the quadruple gate graph neural network, the new vectors of these 8 background events can be obtained background event vectors h ₁ , h ₂ , ..., h ₈ , and new candidate event vectors h _c1 , h _c2 , ..., h _c5 of the five candidate events.

S14: Using the attention neural network to calculate the multiple new background event vectors and the multiple new candidate event vectors to obtain an overall vector of background events.

In this embodiment, after obtaining a plurality of new background event vectors and a plurality of new candidate event vectors, the attention mechanism is used to calculate the background event vectors and candidate event vectors, and it can be considered that each background event to be selected impact of the event. Merge multiple background events into an overall vector of background events, which is convenient for the next calculation.

In this embodiment, using the attention neural network to calculate the multiple new background event vectors and the multiple new candidate event vectors, the specific steps of obtaining the overall vector of the background event include:

S14-1: Input the multiple new background event vectors and the multiple new candidate event vectors into the attention neural network.

In this embodiment, the attention neural network is used to calculate the attention mechanism for the input new background event vector and the new candidate event vector.

S14-2: For each of the plurality of new event vectors to be selected, perform an attention mechanism on each of the plurality of new background event vectors with it operation to obtain the weight coefficient of each new background event vector in the plurality of new background event vectors relative to each new event vector to be selected.

In this embodiment, each background event has a different degree of influence on each candidate event. In order to reflect this influence in the vector calculation, this embodiment uses the attention neural network to the new background event vector Calculate with the new event vector to be selected.

In this embodiment, for each new event vector among the multiple new event vectors to be selected, attention is paid to each new background event vector among the multiple new background event vectors The operation of the mechanism, the specific method of obtaining the weight of each new background event vector in the plurality of new background event vectors relative to the weight of each new candidate event vector is:

u _ij ＝tanh(W _h h _i +W _c h _cj +b _u ) (7)

Among them, u _ij represents the score between the i-th new background event and the j-th new candidate event, the higher the score, the greater the degree of association, tanh() represents the tanh activation function of the quadruple division, W _h and W _c represent the weight, b _u represents the bias parameter, α _ij represents the weight coefficient of the i-th background event for the j-th candidate event, exp(u _ij ) represents the u _ij power of e, ∑ _k exp(u _kj ) represents the sum of e from u _1j power to u _kj power.

For example, as shown in FIG. 5, FIG. 5 is a schematic diagram of an attention mechanism operation based on background events and candidate events of the attention mechanism in an embodiment of the present invention. When 8 background events and 5 candidate events are input, Then calculate the weight coefficient of these 8 background events for each event to be selected.

S14-3: Calculate and obtain an overall vector of the background event according to the weight coefficient.

In this embodiment, according to the weight coefficient, the specific method for calculating the overall vector of the background event is:

Among them, h represents the overall vector of the background event, α _ij represents the weight coefficient of the i-th background event for the j-th candidate event, and h _i represents the i-th background vector.

For example, when 8 background events are input, the overall vector h is the weighted sum of the 8 background event vectors.

S15: Score the overall vector and each of the new candidate event vectors, and use the candidate event corresponding to the candidate event vector with the highest score as a prediction result.

In this embodiment, after the overall vector of the background event is obtained, the prediction result, that is, the most likely event can be determined by scoring each new candidate event vector and the background event vector.

In this embodiment, the steps of scoring the overall vector and each of the new candidate event vectors, and taking the candidate event corresponding to the candidate event vector with the highest score as the prediction result are as follows:

S15-1: According to the overall vector of the background event, calculate the Euclidean distance between each new event vector to be selected and the overall vector of the background event to obtain a plurality of Euclidean distance values.

In this embodiment, according to the overall vector of the background event, the Euclidean distance between each new event vector to be selected and the overall vector of the background event is calculated to obtain the values of multiple Euclidean distances. The calculation method It can be expressed as:

Among them, g() represents the Euclidean distance between two events, and s _j represents the Euclidean distance between two events, that is, the Euclidean distance.

As an example, the Euclidean distance between two events a and b can be expressed as:

g(a,b)=‖a-b‖ (11)

S15-2: Select the new candidate event vector corresponding to the minimum value among multiple Euclidean distance values as the candidate event vector with the highest score, and select the candidate event vector corresponding to the highest score candidate event vector events as predicted outcomes.

In this embodiment, the value of the Euclidean distance represents the distance between two vectors, the smaller the value of the Euclidean distance, the closer the distance between the two vectors, the higher the score, and the larger the value of the Euclidean distance , the lower the score, it can be seen that the candidate event corresponding to the candidate event vector closest to the overall background event vector is the predicted next most likely event, and also the candidate event with the highest score, that is, the prediction result .

For example, as shown in Figure 6, Figure 6 is a schematic diagram of a scoring model for calculating the Euclidean distance based on the weighted and summed background events and candidate events proposed by an embodiment of the present application, in which the overall vector of the background event is obtained respectively and the Euclidean distance between the 5 candidate events, 5 values are obtained, and 5 scores are given for the 5 candidate event vectors.

Exemplarily, the background events input into the quadruple gate diagram neural network are "I get up in the morning", "I wash my face and brush my teeth", "I have breakfast", "I carry my schoolbag". The events to be selected are "I go to school", "I go to work", "I go to the movies". After the quaternion gate diagram neural network fuses the vectors of these background events into an overall vector, it finds that the Euclidean distance between the vector corresponding to "I go to school" and the background event vector is the closest, so the predicted result is "I go to school".

As shown in FIG. 7 , FIG. 7 is a training flow chart of a quadruple gate graph neural network proposed by an embodiment of the present application. As shown in Figure 7, the method includes the following steps:

S21: Collect a plurality of related events, mark a part of the plurality of related events as background events, and mark the other part as candidate events, as a training set.

In this embodiment, it is necessary to collect multiple events as a training set to train the quadruple gate graph neural network, which includes background events and candidate events, and the collected events are divided into several groups, each group has multiple Background events and candidate events, the background events are marked, the correct candidate events corresponding to these background events are marked as correct candidate events, and the remaining candidate events are not marked.

For example, the collected events include several events such as "Xiao Li is out of class", "Xiao Li carries his schoolbag", "Xiao Li walks back to the dormitory", "Xiao Li took a basketball", and these events are divided into One group, marked as the background event, add candidate events "Xiao Li went to the cafeteria to eat", "Xiao Li went to the court to play basketball", "Xiao Li went to class", and marked "Xiao Li went to the court to play basketball" is the correct event to be selected.

S22: Input the training set into the quadruple gate diagram neural network to train the quadruple gate diagram neural network to obtain a trained quadruple gate diagram neural network.

In this embodiment, multiple groups of events in the training set are grouped and input into the quadruple gate graph neural network to train the neural network, wherein the objective optimization function is:

Among them, N represents the number of background events, k represents the number of candidate events, s _Ij represents the correlation score between the i-th background event and the j-th corresponding candidate event, y represents the index of the correct candidate event, margin Is the parameter of the margin loss function, λ is the parameter of L2 regularization, Θ represents the parameter of the model, and the optimization of the model parameter is based on the RMSprop optimizer.

Based on the same inventive concept, an embodiment of the present application provides a quadruple gate diagram neural network event prediction device 300 . Referring to FIG. 8 , FIG. 8 is a schematic diagram of a quadruple gate diagram neural network event prediction device proposed by an embodiment of the present application. As shown in Figure 8, the device includes:

An event map construction module 301, configured to form an event map with a plurality of initial background events and a plurality of candidate events;

The quadruple event representation module 302 is configured to represent the vectors of the plurality of initial background events and the plurality of initial candidate events in the event map in the form of quadruples, and obtain the initial background event vector and the initial candidate event vector. select event vector;

The quaternion gate diagram neural network module 303 is used to perform graph network calculation on the event map using the quadruple gate diagram neural network to obtain multiple new background event vectors and multiple new candidate event vectors;

The background fusion module 304 based on the attention mechanism is used to calculate the multiple new background event vectors and the multiple new candidate event vectors by using the attention neural network to obtain the overall vector of the background event;

The background event and candidate event scoring module 305 is configured to score the overall vector and each of the new candidate event vectors, and use the candidate event corresponding to the candidate event vector with the highest score as the prediction result.

Optionally, the constituting module of the event map includes:

A relationship setting submodule, used to set the relationship between multiple initial background events and multiple events to be selected;

The event map constitutes a sub-module, which is used to form an event map with multiple initial background events and multiple candidate events as nodes, and with relationships between multiple initial background events and multiple candidate events as edges.

Optionally, the quadruple event representation module includes:

The four-tuple event representation submodule is used to represent the vectors of all events including the multiple initial background events and multiple initial candidate events in the form of v( _es , e _o , e _p ), Among them, v represents the predicate verb, _es represents the subject, e _o represents the object, and e _p represents an entity that has a prepositional relationship with the predicate verb.

Optionally, the device also includes:

An event collection module, configured to collect a plurality of related events, mark a part of the plurality of related events as background events, and mark the other part as candidate events, as a training set;

The quadruple gate diagram neural network training module is used to input the training set into the quadruple gate diagram neural network to train the quadruple gate diagram neural network to obtain the trained quadruple gate Graph neural network.

Optionally, the quadruple gate graph neural network module includes:

The first vector input submodule is used to input the representation vector of the event in the event map and the adjacency matrix representing the relationship between the events into the quadruple gate graph neural network;

The four-atom gate graph neural network calculation submodule is used for the quadruple gate graph neural network to calculate the representation vector and the adjacency matrix of the event, and obtain the multiple new background event vectors and the multiple a new event vector to be selected.

Optionally, the background fusion module based on the attention mechanism includes:

The second vector input submodule is used to input the multiple new background event vectors and the multiple new candidate event vectors into the attention neural network;

The weight coefficient calculation sub-module is used for combining each new background event vector among the multiple new background event vectors with each new candidate event vector among the multiple new candidate event vectors Perform an attention mechanism operation to obtain the weight coefficient of each new background event vector in the plurality of new background event vectors relative to each of the new candidate event vectors;

The overall vector obtaining sub-module is used to calculate and obtain the overall vector of the background event according to the weight coefficient.

Optionally, the background event and candidate event scoring module includes:

The Euclidean distance calculation submodule is used to calculate the Euclidean distance between each new event vector to be selected and the overall vector of the background event according to the overall vector of the background event, and obtain a plurality of Euclidean distance values ;

The prediction result obtaining submodule selects the new candidate event vector corresponding to the minimum value among the values of multiple Euclidean distances as the candidate event vector with the highest score, and assigns the event vector corresponding to the highest score candidate event vector The event to be selected is used as the prediction result.

Based on the same inventive concept, another embodiment of the present application provides a readable storage medium, on which a computer program is stored, and when the program is executed by a processor, the quadruple gate diagram as described in any of the above-mentioned embodiments of the present application is realized. Steps in a neural network event prediction method.

Based on the same inventive concept, another embodiment of the present application provides an electronic device, including a memory, a processor, and a computer program stored on the memory and operable on the processor. The steps in the quadruple group gate diagram neural network event prediction method described in the embodiment.

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.

Each embodiment in this specification is described in a progressive manner, each embodiment focuses on the difference from other embodiments, and the same and similar parts of each embodiment can be referred to each other.

Those skilled in the art should understand that the embodiments of the embodiments of the present application may be provided as methods, devices, or computer program products. Therefore, the embodiment of the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, embodiments of the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein.

Embodiments of the present application are described with reference to flowcharts and/or block diagrams of methods, terminal devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each procedure and/or block in the flowchart and/or block diagram, and a combination of procedures and/or blocks in the flowchart and/or block diagram can be realized by computer program instructions. These computer program instructions may be provided to a general purpose computer, special purpose computer, embedded processor or processor of other programmable data processing terminal equipment to produce a machine such that instructions executed by the computer or processor of other programmable data processing terminal equipment Produce means for realizing the functions specified in one or more procedures of the flowchart and/or one or more blocks of the block diagram.

These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing terminal to operate in a specific manner, such that the instructions stored in the computer-readable memory produce an article of manufacture comprising instruction means, the The instruction means implements the functions specified in one or more procedures of the flowchart and/or one or more blocks of the block diagram.

These computer program instructions can also be loaded into a computer or other programmable data processing terminal equipment, so that a series of operational steps are performed on the computer or other programmable terminal equipment to produce computer-implemented processing, thereby The instructions executed above provide steps for implementing the functions specified in one or more procedures of the flowchart and/or one or more blocks of the block diagram.

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 terminal 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.

Above, a kind of quaternion gate diagram neural network event prediction method, device, equipment and medium provided by the application have been introduced in detail. In this paper, specific examples have been used to illustrate the principle and implementation of the application. The above implementation The description of the example 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 application, there will be changes in the specific implementation and scope of application. In summary As stated above, the content of this specification should not be construed as limiting the application.

Claims (9)

1. A four-tuple portal map neural network event prediction method, the method comprising:

forming a rational map by a plurality of initial background events and a plurality of events to be selected;

representing vectors of the initial background events and the initial candidate events in the event map in a four-tuple form to obtain initial background event vectors and initial candidate event vectors, wherein the representation forms of the initial background event vectors and the initial candidate event vectors are v (e) _s ,e _o ,e _p ) Wherein v represents predicate verb, e _s Representing subject, e _o Representing object, e _p Representing an entity having preposition relation with the predicate verb;

using a four-element gate graph neural network, and performing graph network calculation on the event map according to the initial background event vector and the initial candidate event vector to obtain a plurality of new background event vectors and a plurality of new candidate event vectors;

calculating the plurality of new background event vectors and the plurality of new candidate event vectors by using an attention neural network to obtain an overall vector of the background event;

and scoring the event to be selected according to Euclidean distance between each new event vector to be selected and the whole vector, and taking the event to be selected with the highest score as a prediction result.

2. The method of claim 1, wherein constructing a rational map of a plurality of initial background events with a plurality of candidate events, comprises:

setting a relation between a plurality of initial background events and a plurality of events to be selected;

and forming a rational map by taking a plurality of initial background events and a plurality of events to be selected as nodes and taking the relation between the plurality of initial background events and the plurality of events to be selected as edges.

3. The method of claim 1, wherein prior to performing a graph network calculation on the event map based on the initial background event vector and the initial candidate event vector using a four-tuple gate graph neural network to obtain a plurality of new background event vectors and a plurality of new candidate event vectors, the method further comprises:

collecting a plurality of related events, marking one part of the plurality of related events as background events, and marking the other part of the plurality of related events as events to be selected as a training set;

and inputting the training set into the four-element portal map neural network to train the four-element portal map neural network, so as to obtain a trained four-element portal map neural network.

4. The method of claim 1, wherein performing a graph network calculation on the rational graph using a four-tuple gate graph neural network to obtain a plurality of new background event vectors and a plurality of new candidate event vectors, comprising:

Inputting a representation vector of the event in the event map and an adjacency matrix representing a relationship between the events into the four-tuple portal map neural network;

and the four-element gate map neural network calculates the representation vector of the event and the adjacency matrix to obtain the plurality of new background event vectors and the plurality of new candidate event vectors.

5. The method of claim 1, wherein calculating the plurality of new background event vectors and the plurality of new candidate event vectors using an attention neural network to obtain an overall vector of background events comprises:

inputting the plurality of new background event vectors and the plurality of new candidate event vectors into the attention neural network;

for each new candidate event vector of the plurality of new candidate event vectors, performing attention mechanism operation on each new background event vector of the plurality of new background event vectors to obtain a weight coefficient of each new background event vector of the plurality of new background event vectors relative to each new candidate event vector;

and calculating to obtain the integral vector of the background event according to the weight coefficient.

6. The method of claim 1, wherein scoring the candidate events based on euclidean distance between each of the new candidate event vectors and the overall vector, and taking the candidate event with the highest score as a prediction result comprises:

calculating Euclidean distances between each new event vector to be selected and the whole vector of the background event according to the whole vector of the background event to obtain a plurality of Euclidean distance values;

and selecting a new candidate event vector corresponding to the minimum value in the plurality of Euclidean distances as the candidate event vector with the highest score, and taking the candidate event corresponding to the candidate event vector with the highest score as a prediction result.

7. A four-tuple portal map neural network event prediction apparatus, the apparatus comprising:

the event map construction module is used for constructing event maps from a plurality of initial background events and a plurality of events to be selected;

the quadruple event representation module is used for representing vectors of the initial background events and the initial candidate events in the event map in a quadruple mode to obtain initial background event vectors and initial candidate event vectors, wherein the representation forms of the initial background event vectors and the initial candidate event vectors are v (e _s ,e _o ,e _p ) Wherein v represents predicate verb, e _s Representing subject, e _o Representing object, e _p Representing an entity having preposition relation with the predicate verb;

the four-element gate map neural network module is used for performing map network calculation on the event map according to the initial background event vector and the initial event vector to be selected by using the four-element gate map neural network to obtain a plurality of new background event vectors and a plurality of new event vectors to be selected;

the background fusion module is used for calculating the plurality of new background event vectors and the plurality of new candidate event vectors by using the attention neural network to obtain an overall vector of the background event;

and the background event and event scoring module is used for scoring the event to be selected according to Euclidean distance between each new event vector to be selected and the whole vector, and taking the event to be selected with the highest score as a prediction result.

8. A computer readable storage medium, on which a computer program is stored, characterized in that the computer program, when being executed by a processor, implements the steps of the method according to any of claims 1 to 6.

9. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the processor implements the steps of the method according to any of claims 1 to 6 when executing the computer program.

Images