CN114328687A - Event extraction model training method and device, event extraction method and device - Google Patents

Abstract

The present disclosure provides an event extraction model training method and device, which relate to the technical fields of artificial intelligence such as knowledge graphs and deep learning. Label data. The first sub-model is obtained by performing model training on the first training sample. Obtain a second training sample, the second training sample includes the second sample text, multiple second sample arguments existing in the second sample text, and every two second sample arguments in the multiple second sample arguments correspond to the same The sample probability of the event. The second sub-model is obtained by performing model training on the second training sample. It is determined that the event extraction model includes a first sub-model and a second sub-model. The technical solution provided by the present disclosure can effectively improve the accuracy of the event extraction model.

Description

Event extraction model training method and device, event extraction method and device

technical field

The present disclosure relates to the technical fields of artificial intelligence such as knowledge graphs and deep learning, and in particular, to an event extraction model training method and device, and an event extraction method and device.

Background technique

Event extraction refers to extracting the required event information from unstructured text and integrating it into a structured form.

At present, event extraction is usually achieved through an event extraction model, and because of insufficient labeling information, the accuracy of the event extraction model is poor.

SUMMARY OF THE INVENTION

The present disclosure provides an event extraction model training method and device, and an event extraction method and device.

According to a first aspect of the present disclosure, an event extraction model training method is provided, including:

Acquire a first training sample, where the first training sample includes first sample text and first labeling data, where the first labeling data includes: a plurality of samples corresponding to the plurality of sample arguments in the first sample text data packets, sample roles corresponding to each of the data packets, and sample event types corresponding to each of the data packets, wherein the sample arguments in any one of the data packets are the same;

Perform model training through the first training sample to obtain a first sub-model, where the first sub-model is used to determine the arguments existing in the text, the roles corresponding to the arguments, and the event types corresponding to the arguments;

acquiring a second training sample, where the second training sample includes second sample text, multiple sample events existing in the second sample text, and second sample arguments included in each of the sample events;

Perform model training through the second training sample to obtain a second sub-model, where the second sub-model is used to determine an event existing in the text and an argument corresponding to the event;

An event extraction model is determined based on the first sub-model and the second sub-model.

According to a second aspect of the present disclosure, an event extraction method is provided, comprising:

Get the first text to be processed;

The first text is processed by the first sub-model in the pre-trained event extraction model to obtain a first output result, where the first output result includes: arguments existing in the first text, the The role corresponding to the argument and the event type corresponding to the argument;

The first output result is processed by the second sub-model in the pre-trained event extraction model to obtain the event existing in the first text and the argument corresponding to the event.

According to a third aspect of the present disclosure, an event extraction model training device is provided, comprising:

An acquisition module, configured to acquire a first training sample, where the first training sample includes first sample text and first labeled data, and the first labeled data includes: a plurality of sample theories in the first sample text Multiple data packets corresponding to the element, sample roles corresponding to each of the data packets, and sample event types corresponding to each of the data packets, wherein the sample arguments in any one of the data packets are the same;

The first processing module is used to perform model training through the first training sample to obtain a first sub-model, and the first sub-model is used to determine the arguments existing in the text, the roles corresponding to the arguments and the arguments. The event type corresponding to the meta;

A second acquisition module, configured to acquire a second training sample, where the second training sample includes a second sample text, a plurality of sample events existing in the second sample text, and the first sample event included in each of the sample events. two-sample argument;

a second processing module, configured to perform model training through the second training sample to obtain a second sub-model, where the second sub-model is used to determine an event existing in the text and an argument corresponding to the event;

A determination module, configured to determine an event extraction model based on the first sub-model and the second sub-model.

According to a fourth aspect of the present disclosure, an event extraction apparatus is provided, comprising:

an acquisition module for acquiring the first text to be processed;

The first processing module is configured to process the first text through the first sub-model in the pre-trained event extraction model to obtain a first output result, where the first output result includes: The existing argument, the role corresponding to the argument, and the event type corresponding to the argument;

The second processing module is configured to process the first output result through the second sub-model in the pre-trained event extraction model to obtain the event existing in the first text and the argument corresponding to the event.

According to a fifth aspect of the present disclosure, there is provided an electronic device, comprising:

at least one processor; and

a memory communicatively coupled to the at least one processor; wherein,

The memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the method of the first aspect or the second aspect .

According to a sixth aspect of the present disclosure, there is provided a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to cause the computer to perform the method of the first aspect or the second aspect.

According to a seventh aspect of the present disclosure, there is provided a computer program product, the computer program product comprising: a computer program stored in a readable storage medium, from which at least one processor of an electronic device can Reading the storage medium reads the computer program, and executing the computer program by the at least one processor causes the electronic device to perform the method of the first aspect or the second aspect.

The technique according to the present disclosure solves the problem of poor accuracy of the event extraction model.

It should be understood that what is described in this section is not intended to identify key or critical features of embodiments of the disclosure, nor is it intended to limit the scope of the disclosure. Other features of the present disclosure will become readily understood from the following description.

Description of drawings

The accompanying drawings are used for better understanding of the present solution, and do not constitute a limitation to the present disclosure. in:

FIG. 1 provides a schematic diagram of the implementation of event extraction according to an embodiment of the present disclosure;

2 is a flowchart of an event extraction model training method provided by an embodiment of the present disclosure;

FIG. 3 is a schematic diagram of the implementation of labeling data provided by an embodiment of the present disclosure;

FIG. 4 is a second flowchart of the event extraction model training method provided by the embodiment of the present disclosure;

FIG. 5 is a schematic diagram of processing a first sub-model provided by an embodiment of the present disclosure;

FIG. 6 is a schematic diagram of an implementation of updating model parameters of a first sub-model provided by an embodiment of the present disclosure;

FIG. 7 is a third flowchart of a training method for an event extraction model provided by an embodiment of the present disclosure;

FIG. 8 is a schematic diagram of an implementation of determining a first probability according to an embodiment of the present disclosure;

FIG. 9 is a schematic diagram of the implementation of determining a candidate window according to an embodiment of the present disclosure;

FIG. 10 is a schematic diagram of the implementation of determining a target window provided by an embodiment of the present disclosure;

11 is a flowchart of an event extraction method provided by an embodiment of the present disclosure;

FIG. 12 is a schematic processing diagram of an event extraction method provided by an embodiment of the present disclosure;

13 is a schematic structural diagram of an event extraction model training apparatus according to an embodiment of the present disclosure;

14 is a schematic structural diagram of an event extraction apparatus according to an embodiment of the present disclosure;

FIG. 15 is a block diagram of an electronic device used to implement the event extraction model training method and the event extraction method according to the embodiment of the present disclosure.

Detailed ways

Exemplary embodiments of the present disclosure are described below with reference to the accompanying drawings, which include various details of the embodiments of the present disclosure to facilitate understanding and should be considered as exemplary only. Accordingly, those of ordinary skill in the art will recognize that various changes and modifications of the embodiments described herein can be made without departing from the scope and spirit of the present disclosure. Also, descriptions of well-known functions and constructions are omitted from the following description for clarity and conciseness.

In order to better understand the technical solutions of the present disclosure, the related technologies involved in the present disclosure will be further introduced in detail below.

As a form of information, an event is defined as the objective fact that a specific person or thing interacts at a specific time and a specific place, generally at the sentence level. In Topic Detection Tracking (TDT), an event refers to a set of related descriptions about a topic, which can be formed by classification or clustering.

Wherein, each element constituting an event may include at least one of the following: trigger word, event type, event argument, and argument role.

Among them, the event trigger word: the core word indicating the occurrence of the event, mostly verbs or nouns;

Event type: It can be preset for multiple event types, for example, 8 event types and 33 subtypes are defined in the ACE2005 corpus. Currently, most event extractions use 33 event types. It can be understood that event recognition is a word-based 34-category (33-category event type + None) multivariate classification task.

Event Argument: The participants of the event, which are mainly composed of entities, values, and time. A value is a non-physical event participant, such as a job;

Argument Role: The role the event argument plays in the event. There are 35 types of roles, such as attackers, victims, etc., which can also be preset. It can be understood that the role classification is a multi-classification task of 36 categories (35 types of role types + None) based on word pairs;

Event extraction technology extracts events of interest to users from unstructured information and presents them to users in a structured way. The event extraction task can be decomposed into four subtasks: trigger word recognition, event type classification, argument recognition and role classification tasks. Among them, trigger word recognition and event type classification can be combined into an event recognition task.

Event recognition judges the event type to which each word in a sentence belongs, and is a word-based multi-classification task. Argument identification and role classification can be combined into an argument role classification task. The role classification task is a multi-classification task based on word pairs, which judges the role relationship between any pair of trigger words and entities in the sentence.

For example, the event extraction can be understood in conjunction with FIG. 1 , which provides a schematic diagram of the implementation of event extraction according to an embodiment of the present disclosure.

Assuming that there is currently such an example sentence "In city A, a photographer died when a tank opened fire on Hotel P", the event extraction is performed for the current example sentence, for example, the two events shown in Figure 1 can be obtained.

Among them, the event type of event 1 is death, and its corresponding trigger word is "death" in the above example, and the event arguments in event 1 include photographer, tank, and city A, the arguments of these three event arguments The roles are the victim, the tool, and the place in that order.

And, the event type of event 2 is attack, the corresponding trigger word is "fire" in the above example, and the event arguments in event 2 include photographer, hotel P, tank, city A, these 4 event arguments The argument roles are, in order, goal, goal, tool, and place.

Based on the above examples, it can be determined that event extraction can extract the required event information from unstructured text and integrate it into a structured form.

On the basis of the event extraction described above, chapter-level event extraction refers to structuring all predefined types of events from a long text chapter. A structured event includes the event type, as well as all event arguments and its role. That is to say, chapter-level event extraction is to perform event extraction processing for long-text chapters, and chapter-level event extraction often faces the complex situation that the text is long and the text contains multiple events of the same type.

At present, the realization of event extraction is usually realized by the event extraction model. In the existing chapter-level event extraction model, it is usually necessary to give the supervision signal of each word in the chapter during training. Meta roles, event types, position in the text, etc.

It is understandable that the existing text-level event extraction models rely on the supervision signal of each word, and the text-level event arguments often appear multiple times in the text, but there may be insufficient labeled data, which leads to the model Training learning is easily misled by false supervision signals.

It can be explained here with a specific example, assuming that there is a text currently, including formula a in this text, and assuming that formula a appears 10 times in this text, that is, in this text 10 Formula a appears in different places. In the current prior art, the argument roles, event types, and positions appearing in the text corresponding to formula a are labeled. However, due to the huge workload of the labeling work, it is easy to cause insufficient labeling.

For example, in the current example, formula a appears 10 times in the text, but when labeling, only the formulas in 4 places are labeled, that is to say, the supervision signals are only labeled for the formulas in 4 places, and the remaining 6 places are labeled with supervision signals. The formulas in this place are not marked. Because the training of the model depends on the labeling of the supervision signal, for the model, because only 4 formulas are currently marked, the model will think that formula a only appears 4 times in this text, but The actual situation is not the case, then the four supervision signals marked above are misleading information for the model.

Therefore, based on the above introduction, it can be determined that if the supervision signal in the text is not sufficiently labeled, the labeled supervision signal is actually wrong, and it is easy to be misled by the wrong supervision signal when the model is trained and learned.

In addition, based on the above introduction, it can be determined that in chapter-level event extraction, because the text to be extracted is long and usually contains multiple complex events of the same type, in many cases, event extraction cannot Events of the same type are separated, resulting in an error in the final event extraction result. For example, event 1 and event 2 are mixed events of the same type, so it is possible to split the arguments in event 1 into events. 2 in the argument, resulting in an error in the result of event extraction.

In view of the problems in the prior art introduced above, the present disclosure proposes the following technical idea: by means of multi-instance learning, the event arguments that appear multiple times in the text are optimized as a package to prevent the model from erroneous learning, so that the Provides a model training method when the same event argument appears multiple times in a chapter and the annotation is insufficient. And, by determining the central argument, using the central argument as a method to distinguish multiple events of the same type, and at the same time building a window with the central argument as the center to match other arguments of the same event, which can effectively improve the accuracy of event extraction results. .

On the basis of the above-mentioned content, the following describes the event extraction model training method provided by the present disclosure with reference to specific embodiments. It should be noted that the executive body of each embodiment of the present disclosure may be a server, a processor, a microprocessor, etc. For a device with data processing function, the execution subject of each embodiment in this disclosure is not limited here, and it can be selected and set according to actual needs. Any device with data processing function can be used as the implementation of each embodiment in this disclosure. main body.

The following first introduces the event extraction model training method provided by the present disclosure with reference to FIG. 2 and FIG. 3 . FIG. 2 is a flowchart of the event extraction model training method provided by the embodiment of the present disclosure. Implementation schematic.

As shown in Figure 2, the method includes:

S201. Acquire a first training sample, where the first training sample includes first sample text and first labeled data, and the first labeled data includes: multiple data packets corresponding to multiple sample arguments in the first sample text, each The sample role corresponding to the data packet and the sample event type corresponding to each data packet, wherein the sample argument in any data packet is the same.

In this embodiment, the first training sample is a sample for training the first sub-model, where the first training sample may include the first sample text and the first annotation data.

It can be understood that the first sample text is a text chapter, such as a paper, a composition, an article, etc. This embodiment does not limit the format, word count, etc. of the first sample text. The specific implementation of this text can be selected according to actual needs.

And, the first annotation data in this embodiment is the annotation data for the first sample text, and the first annotation data may include: multiple data packets corresponding to multiple sample arguments in the first sample text, The respective sample roles corresponding to each data packet and the respective sample event types corresponding to each data packet, wherein the sample arguments in any data packet are the same. That is to say, in this embodiment, the arguments included in the first sample text, the argument roles corresponding to the arguments, and the event types corresponding to the arguments are all marked, so as to obtain the first marked data.

At the same time, it should be noted that in this embodiment, when the arguments in the first sample object are marked, a plurality of data packets are determined according to the plurality of sample arguments in the first sample text, wherein the One or more sample arguments may be included in any data packet, and the sample arguments included in any data packet are the same.

For example, it can be understood with reference to Figure 3. As shown in Figure 3, it is assumed that the sample text shown in Figure 3 currently exists. For the convenience of introduction, it is assumed that the sample text includes multiple formulas, such as formula a shown in Figure 3, Formula b, formula c, and in conjunction with FIG. 3 , it can be determined that in the sample text, both formula a and formula b appear multiple times.

In this embodiment, the same sample text is determined as one data packet, and three data packets shown on the right side of FIG. 3 are obtained. Since formula a appears 4 times in the sample text, 4 formula a are included in packet 1; and, formula b appears 2 times in the sample text, so 2 formula b is included in packet 2 ; and, the formula c appears only once in the sample text, so 1 formula c is included in packet 3. And for each data packet, it is marked with its corresponding sample role and sample event type.

Based on the introduction in Figure 3, it can be understood that the same argument role is actually included in any data packet, so for any data packet, the sample role corresponding to the data packet is actually the argument in the data packet. argument role. And the sample event type corresponding to the data packet is actually the sample event type of the argument in the data packet.

Based on the above introduction, it can be determined that, in this embodiment, for arguments that appear multiple times in the sample text, the same arguments in multiple places can be used as a data packet, and for arguments that appear only once, it can also be determined that Only packets that include this argument. The data packets are then marked, so that the model can be correctly informed of the number of occurrences of each argument in the sample text.

For example, for data packet 1 in Figure 3, the model can determine that formula a appears 4 times in the sample text. Although we did not tell the model the location of these 4 sample texts, we at least correctly told the model formula a. The number of occurrences to avoid misleading the model due to insufficient labeling. After that, the model can autonomously learn the specific positions of each argument. Therefore, in this embodiment, the same argument that appears multiple times can be regarded as a data packet, and the data packet can be marked, and the marked data can be provided to the model, so that the correct data can be provided to the model. The annotation information is provided to the model to avoid misleading the model due to insufficient annotation.

The implementation of data packet division described above is actually an application of multi-instance learning, wherein the specific implementation of multi-instance learning has been introduced in the above embodiments, and will not be repeated here.

In the actual implementation process, there may be multiple first sample texts included in the first training sample, and each first sample text has corresponding first annotation data, wherein the specific number of the first sample texts It can be selected and set according to actual needs.

S202 , performing model training on the first training sample to obtain a first sub-model, where the first sub-model is used to determine arguments existing in the text, roles corresponding to the arguments, and event types corresponding to the arguments.

After acquiring the first training data, model training may be performed using the first training data. For example, the initial first sub-model is trained through the first training sample to obtain the trained first sub-model. In this embodiment, the The first sub-model is used to determine the arguments existing in the text, the argument roles corresponding to the arguments, and the event types corresponding to the arguments.

S203. Obtain a second training sample, where the second training sample includes the second sample text, multiple second sample arguments existing in the second sample text, and every two second sample arguments in the multiple second sample arguments The sample probability corresponding to the same event.

And in this embodiment, there is also a second training sample, the second training sample is a sample used for training the second sub-model, wherein the second training sample may include the second sample text, the second sample text existing in the The multiple second sample arguments, and every two second sample arguments in the multiple second sample arguments correspond to the sample probability of the same event.

Based on the above introduction, it can be determined that for each event argument, it can belong to a certain event, then for any two event arguments, it can belong to the same event or different events, then this In the embodiment, the sample probability that every two second sample arguments correspond to the same event may be 0 (that is to say, they do not belong to the same event), or the sample probability that every two second sample arguments correspond to the same event may be 0. is 1 (that is, belongs to the same event),

And, in a possible implementation manner, for example, the training of the first sub-model and the second sub-model may have a contextual relationship, that is to say, the first sub-model is trained first, and then the first sub-model is trained according to the output data of the first sub-model. When the second sub-model is trained, the second sample text in the second training sample can be, for example, the same as the first sample text introduced above, and there are multiple first sample texts in the second training sample that exist in the second sample text. The two-sample argument, for example, can be the argument output by the first sub-model above. That is to say, based on the same sample text, the first sub-model is trained first, and then the second sub-model is trained according to the arguments extracted by the first sub-model.

Or, in another possible implementation manner, the training of the first sub-model and the second sub-model may also be independent of each other, that is, the process of training the first sub-model, the first training sample, and the training of the second sub-model There is no relationship between the process of the sub-model and the second training sample, which is a completely independent training process and completely independent sample data. In the actual implementation process, the specific implementation of the first training sample and the second training sample may be selected and set according to actual requirements, which is not particularly limited in this embodiment.

S204 , performing model training on the second training sample to obtain a second sub-model, where the second sub-model is used to determine an event existing in the text, an argument corresponding to the event, and a role corresponding to the argument.

After acquiring the second training data, model training can be performed by using the second training data. For example, the initial second sub-model is trained by using the second training sample to obtain the trained second sub-model. The second sub-model is used to determine the events existing in the text, the arguments corresponding to the events, and the roles corresponding to the arguments.

S205. Determine an event extraction model based on the first sub-model and the second sub-model.

After the above-mentioned first sub-model and second sub-model are trained, an event extraction model can be determined based on the first sub-model and the second sub-model. Therefore, the event extraction model in this embodiment includes the first sub-model described above. model and the second submodel. The event extraction model in this embodiment is used to process the input text, so as to output the events existing in the text, the arguments corresponding to the events, and the roles corresponding to the arguments.

The method for training an event extraction model provided by an embodiment of the present disclosure includes: acquiring a first training sample, where the first training sample includes first sample text and first labeling data, and the first labeling data includes: multiple data in the first sample text Multiple data packets corresponding to each sample argument, sample roles corresponding to each data packet, and sample event types corresponding to each data packet, wherein the sample argument in any data packet is the same. The first sub-model is obtained by performing model training on the first training sample, and the first sub-model is used to determine the arguments existing in the text, the roles corresponding to the arguments, and the event types corresponding to the arguments. Obtain a second training sample, the second training sample includes the second sample text, multiple second sample arguments existing in the second sample text, and every two second sample arguments in the multiple second sample arguments correspond to the same The sample probability of the event. The second sub-model is obtained by performing model training on the second training sample, and the second sub-model is used to determine the events existing in the text, the arguments corresponding to the events, and the roles corresponding to the arguments. An event extraction model is determined based on the first sub-model and the second sub-model. Through the first training sample and the second training sample, the first sub-model and the second sub-model in the event extraction model are processed respectively, wherein the first training data is to determine various sample arguments appearing in the sample text For data packets, the sample arguments included in any data packet are the same, and the roles and event types are marked for the data packets, so as to ensure that the correct annotation information is provided to the model, thereby avoiding the problem of insufficient annotation. The resulting misleading of the model can effectively improve the accuracy of the event extraction model.

On the basis of the above-mentioned embodiment, the implementation of the training of the first sub-model and the training of the second sub-model will be introduced separately below. First, in the event extraction model training method provided by the present disclosure with reference to FIG. 4 to FIG. The training process of a sub-model is described in detail. 4 is a second flowchart of an event extraction model training method provided by an embodiment of the present disclosure, FIG. 5 is a schematic diagram of processing a first sub-model provided by an embodiment of the present disclosure, and FIG. 6 is an update first sub-model provided by an embodiment of the present disclosure. A schematic diagram of the realization of the model parameters.

As shown in Figure 4, the method includes:

S401. Obtain a first training sample, where the first training sample includes first sample text and first label data, and the first label data includes: multiple data packets corresponding to multiple sample arguments in the first sample text, each The sample role corresponding to the data packet and the sample event type corresponding to each data packet, wherein the sample argument in any data packet is the same.

The implementation of S401 is similar to the implementation of S201 above, and details are not described herein again.

S402. Process the first sample text through the first sub-model to be trained to obtain first prediction data, where the first prediction data includes multiple prediction arguments, prediction roles corresponding to the prediction arguments, and predictions corresponding to the prediction arguments Event type.

In this embodiment, the untrained first sub-model is referred to as the to-be-trained first sub-model. When training the first sub-model according to the first training sample, for example, the first sub-model to be trained can be used to process the first sample text, because the first sub-model in this embodiment is used to determine The arguments present in the text, the roles of the arguments, and the types of events of the arguments.

Therefore, referring to FIG. 5 , the first sub-model can output first prediction data, and the first prediction data includes a plurality of prediction arguments, wherein the prediction arguments are obtained by the first sub-model performing argument extraction on the first sample text Arguments, for example, may include argument a, argument b, argument c, etc. shown in Figure 5 . And the first prediction data includes a prediction role corresponding to each prediction argument and a prediction event type corresponding to each prediction argument.

In a possible implementation manner, the first prediction data in this embodiment further includes prediction positions of multiple prediction arguments in the first sample text, and the first prediction data further includes multiple prediction arguments The probability of the predicted position in the first sample text. The probability can be understood as the confidence level output by the model itself, that is, how sure the model is about the predicted position of the current output.

Therefore, in this embodiment, although the first sub-model is not told the specific position of the same argument in the text, only the number of each argument in the text is told to the first sub-model, but the first sub-model can also be Based on the multi-instance learning method, each extracted prediction argument can be output, and at the same time, the predicted position of each extracted prediction argument in the text and the probability of the current predicted position can be output.

S403: Determine the first loss according to the first labeling data, the predicted role corresponding to the predicted argument, and the predicted event type corresponding to the predicted argument.

After the first sub-model outputs the first predicted data, since the first sample text is marked with the first marked data in this embodiment, the loss of the model can be determined according to the first marked data and the above-mentioned first predicted data.

In a possible implementation manner, based on the above description, it can be determined that the first labeled data includes multiple data packets, wherein the interior of each data packet includes the same argument, and for each data packet Packages are marked with argument roles and event types, so the argument roles and event types of arguments are actually marked in the marked data.

And the above-mentioned first prediction data includes a plurality of prediction arguments, as well as the prediction roles corresponding to the prediction arguments and the prediction event types corresponding to the prediction arguments.

Therefore, referring to FIG. 6 , the first loss can be determined according to the first labeled data, the predicted role corresponding to the predicted argument, and the event type corresponding to the predicted argument. In a possible implementation manner, for example, a first loss function may be preset, and then the first label data, the prediction role corresponding to the prediction argument, and the event type corresponding to the prediction argument are processed by the first loss function, Thereby the first loss is determined.

S404 , grouping multiple prediction arguments to obtain multiple groups of prediction arguments, and the arguments in each group of prediction arguments are the same.

And in this embodiment, multiple prediction arguments may be included in the first prediction data, and based on the above introduction, it can be determined that the same argument may appear multiple times in the text, because in this embodiment, the data The package tells the model how many sample arguments there are, so the model will output the predicted positions of multiple identical sample arguments.

Therefore, multiple prediction arguments can currently be grouped to obtain multiple groups of prediction arguments, and the arguments in each group of prediction arguments are the same. That is to say, the same arguments in the prediction arguments are used as a group, so as to obtain multiple groups of prediction arguments.

The set of prediction arguments here is actually similar to the data package in the labeled data described above, the difference is that the above data package is the data obtained by the pre-annotation, and the current set of prediction arguments is After the first sub-model is processed, the resulting data is grouped by the resulting plurality of prediction arguments.

S405. According to the probability of the predicted positions of the multiple prediction arguments in the first sample text, respectively determine the target prediction argument in the multiple groups of prediction arguments, wherein the target prediction argument in the group of prediction arguments is in The predicted position in the first sample text has the highest probability.

Based on the above description, it can be determined that the first prediction data in this embodiment includes the probability of the predicted positions of multiple prediction arguments in the first sample text, and the probability in this embodiment can be understood as the confidence output of the model Spend.

At the same time, it can be understood that because the position of each argument in the text is not notified in this disclosure, the model does not actually have reference information during the learning process, but learns independently. Then, in the process of starting training, The probability of the predicted position of the prediction argument output by the model in the first sample text is actually not very high, that is to say, the position of the predicted argument output by the model is not very certain.

Then, in order to improve the efficiency of model training, in this embodiment, a target prediction argument may be determined among multiple groups of prediction arguments, where the target prediction argument is the argument with the highest probability of the corresponding prediction position in a group of prediction arguments . That is to say, the loss is calculated according to the argument with the highest probability of the predicted position, which can effectively improve the effectiveness of model training.

S406. Determine the second loss according to the predicted position of the target argument and the actual position of the target argument in the first sample text.

It can be determined that the target argument is also an argument extracted from the first sample text, so based on the first sample text, the actual position of the target argument in the first sample text can be determined. In a possible implementation, in this embodiment, the actual position of each argument in the first sample text does not need to be marked in advance, but after the model outputs the predicted position of the predicted argument, the first text In this paper, the target argument near the predicted position is determined to determine the actual position of the target argument in the first sample text.

Referring to FIG. 6 , the second loss can then be determined according to the predicted position of the target argument and the actual position of the target argument in the first sample text. In a possible implementation manner, for example, the predicted position of the target argument and the actual position of the target argument in the first sample text may be processed through a preset second loss function to obtain the second loss function.

S407. Update the model parameters of the first sub-model according to the first loss and the second loss.

Among them, the first loss can optimize the accuracy of the arguments extracted by the first sub-model, as well as the argument roles and event types of the output arguments, while the second loss can optimize the accuracy of the positions of the output arguments of the first sub-model Therefore, in this embodiment, the model parameters of the first sub-model can be updated based on the first loss and the second loss determined above, so as to realize the training of the first sub-model.

And it can be understood that, in the actual implementation process, the above-mentioned operations of determining the loss and updating the model parameters of the first sub-model according to the loss can be iteratively performed for many times until the preset number of iterations is reached, or until the model converges. , the first sub-model after training can be obtained.

In the event extraction model training method provided by the embodiment of the present disclosure, the first sample text is processed by the first sub-model to be trained, so as to output the first prediction data, and then according to the theory of the prediction arguments included in the first prediction data The meta roles and event types, and the argument roles and event types of the arguments in the first labeled data, determine a first loss for optimizing the accuracy of determining the argument roles and event types. And it is also possible to select the target argument with the highest probability according to the predicted position of the predicted argument included in the first prediction data in the first sample text and the probability of the predicted position, and then determine the second loss based on the target argument. , which can effectively improve the speed and efficiency of model training. Specifically, the second loss of the accuracy of the position of the argument for optimizing the output is determined by the predicted position of the target argument and the actual position of the target argument in the first sample text. Then, the model parameters of the first sub-model are updated according to the first loss and the second loss, so that the training for the first sub-model can be effectively implemented, and the arguments and arguments output by the first sub-model can be effectively guaranteed. Argument roles, event types of arguments, and accuracy of argument locations.

The above-mentioned embodiments describe the implementation of training for the first sub-model, and the implementation of the training for the second sub-model will be described in detail below with reference to specific embodiments. For example, it can be described with reference to FIGS. 7 to 10 . FIG. 7 is a third flowchart of an event extraction model training method provided by an embodiment of the present disclosure, FIG. 8 is a schematic diagram of an implementation of determining a first probability provided by an embodiment of the present disclosure, and FIG. 9 is a A schematic diagram of implementation of determining a candidate window provided by an embodiment of the present disclosure, FIG. 10 is a schematic diagram of implementation of determining a target window provided by an embodiment of the present disclosure.

As shown in Figure 7, the method includes:

S701. Obtain a second training sample, where the second training sample includes the second sample text, multiple second sample arguments existing in the second sample text, and every two second sample arguments in the multiple second sample arguments The sample probability corresponding to the same event.

The implementation of S701 is similar to the implementation of S203, which is not repeated here.

S702. Determine multiple second sample roles corresponding to multiple second sample arguments.

In this embodiment, each sample argument corresponds to its own sample role. In a possible implementation manner, the current second sample argument may be, for example, the prediction argument output by the above-mentioned first sub-model, based on What can be determined from the above introduction is that the first submodel also outputs the argument roles and event types of the arguments. Therefore, in the current situation, the second sample roles corresponding to each of the plurality of second sample arguments can be obtained based on the output of the first sub-model.

S703. Determine the first probability that the second sample arguments under each second sample role correspond to the same event.

In this embodiment, the second sample role is the sample role corresponding to each second sample argument in the second sample text. It can be understood that each sample argument corresponds to its corresponding event, then this embodiment In can also determine the first probability that the second sample arguments under each second sample role correspond to the same event.

It is necessary to distinguish between event types and events here. It is understandable that there can be multiple different events for the same event type. For example, there are currently event a and event b, and the event types of these two events are both event type 1. .

In this embodiment, when determining the probability that the second sample arguments under each second sample role correspond to the same event, for example, analysis may be performed based on the second sample text, so as to determine the probability described above.

For example, it can be understood in conjunction with FIG. 8. As shown in FIG. 8, assuming that the second sample text is currently being analyzed, there are multiple arguments under the sample role 1 in the second sample text. Argument 1, Argument 2 and Argument 3, that is to say, the argument roles of these 3 arguments are sample role 1, and based on the above introduction, it can be determined that the same argument can appear multiple times in the text, Then the example in Figure 8 indicates that argument 2 appears three times in the second sample text.

Referring to Figure 8, where argument 1 belongs to event a, and the first argument 2 and the second argument 2 belong to event a, the third argument 2 belongs to event b, and the argument 3 belongs to event b. Then for the example of FIG. 8 , it can be determined that the probability of the second sample argument under the second sample text belonging to event a is 60% (3/5), and the probability of belonging to event b is 40% (2/5).

In this embodiment, the probability that the second sample argument under the second sample text belongs to the same event may be, for example, the probability that the second sample argument belongs to the same event with the largest number, that is to say, the probability with the largest probability is determined as the currently required probability, Therefore, for the example of FIG. 8 , it can be determined that the probability that the second sample argument under the second sample text belongs to the same event is 60%.

Figure 8 shows how each sample argument under the sample role corresponds to two different events. In the actual implementation process, each argument under the sample role may also correspond to more than two different events. The implementation manner is similar to that described above, and the probability corresponding to the event with the largest number of existing events may be determined as the currently required probability.

S704. Determine the role coefficient of each second sample role according to the recall rate and accuracy rate of each second sample role.

And, the second sample argument in this embodiment is the argument output by the first sub-model introduced above. Based on the above introduction, it can be determined that the first sub-model can output the prediction argument, the prediction of each prediction argument The role and the corresponding prediction event type for each prediction argument.

Because each prediction argument has its own prediction role, there are many prediction roles in the output of the first sub-model, and the second sample role in this embodiment is actually the prediction role output by the first sub-model.

Then, for any second sample role, the recall rate and accuracy rate corresponding to the second sample role output by the first sub-model can be counted, and the specific implementation of recall rate (Recall) and precision rate (Precision) can refer to the prior art The implementation in , will not be repeated here.

After the recall rate and accuracy rate of each second sample role are determined, the role coefficient of each second sample role can be determined according to the recall rate and accuracy rate of each second sample role. The role coefficient in this embodiment may be, for example, a comprehensive evaluation index (F-Measure).

In a possible implementation manner, for example, the recall rate and the accuracy rate of the second sample character may be processed according to a preset function to obtain the character coefficient of the second sample character.

Wherein, the preset function for determining the role coefficient, for example, may satisfy the following formula 1:

Among them, P is the accuracy rate, R is the recall rate, and F is the role coefficient. The role coefficient in this embodiment may specifically be a comprehensive evaluation index.

The above operations can be performed for each second sample character, so as to obtain the character coefficient F corresponding to each second sample character.

S705. Determine the priority of each second sample role according to the first probability corresponding to each second sample role and the role coefficient of each second sample role.

After determining the first probability corresponding to each second sample role and the role coefficient of each second sample role, for example, the priority of each second sample role may be determined according to the first probability and the role coefficient.

In a possible implementation manner, for any second sample character, for example, the product of the first probability P corresponding to the second sample character and the character coefficient F of the second sample character may be determined as the priority of the second sample character class. It can be understood that the greater the product of P and F, the greater the priority of identifying the second sample role.

S706: Determine the central role to be selected from among the plurality of second sample roles according to the priority of each second sample role.

After the respective priorities of the respective second sample roles are determined, the central role to be selected may be determined from among the plurality of second sample roles according to the priorities of the respective second sample roles.

In a possible implementation manner, for example, the priority of each second sample role may be compared with a preset threshold. If it is determined that the priority of the second sample role in the second sample role is greater than or equal to the preset threshold, then The second sample role whose priority is greater than or equal to the preset threshold may be determined as the central role to be selected. In this case, there may be multiple central roles to be selected in this embodiment.

In another possible implementation, if it is determined that the priorities of the multiple second sample roles are all smaller than the preset threshold, the second sample role with the highest priority among the multiple second sample roles may be determined as the center to be selected Role.

It should be noted here that the higher the priority of the sample role, the stronger the ability to distinguish events under the sample role. Therefore, in this embodiment, the central role to be selected is determined based on the priority, which actually determines the distinction. The stronger argument role of the event.

S707: Determine the argument corresponding to the central role to be selected as the central argument.

After the central role to be selected is determined, the argument corresponding to the central role to be selected can be determined as the central argument. Among them, the argument corresponding to the central role to be selected is actually the argument whose role is the central role to be selected.

S708. Determine a target window corresponding to the central argument in the second sample text, where the target window includes a preset number of characters.

It can be determined that there may be multiple central arguments determined in this embodiment, and any one central argument is used as an example for description below.

In this embodiment, the target window corresponding to the central argument may be determined in the second sample text, where the target window may include a preset number of characters, and the number of characters included in the target window is actually the number of characters in the window. length.

In a possible implementation manner, the target window in this embodiment satisfies the following conditions: the target window includes a preset number of characters, the target window includes a central argument, and on the basis of satisfying the above two conditions, the target window Include the most other arguments that are consistent with the event type of the central argument.

The possible implementations of determining the target window are described below:

For example, a plurality of candidate windows may be determined in the second sample text, the candidate windows include a preset number of characters, and the candidate windows include arguments corresponding to the central role to be selected.

It can be understood that, when determining the target window in this embodiment, because the length of the window is preset, and the target window must include the central argument, then in order to ensure that the target window includes the most events and the central argument For other arguments of the same type, you can first determine multiple candidate windows by sliding windows.

The window to be selected in this embodiment includes a preset number of characters, that is to say, the length of the window to be selected is a preset length, and the argument corresponding to the central role to be selected in the window to be selected, the central role to be selected here corresponds to The argument of is actually the central argument introduced above.

For example, it can be understood in conjunction with FIG. 9 . Referring to FIG. 9 , 901 is assumed to be the second sample text, and FIG. 9 illustrates part of the content of the second sample text. Assuming that the current central argument is "tens of millions of yuan" in Figure 9, for this central argument, for example, at least the candidate windows shown in 902, 903, and 904 in Figure 9 can be determined.

Based on the candidate window 902, the candidate window 903 and the candidate window 904, it can be determined that each candidate window includes the central argument "tens of millions of yuan", and the length of each candidate window is a preset length , where the preset length can be, for example, N words, and the specific setting of N can be selected according to actual needs.

And based on the above-mentioned FIG. 9 , it can be determined that in this embodiment, on the basis of including the central argument in the window, the window can be slid based on the preset window length, thereby determining multiple candidate windows.

After the candidate window is determined, the number of arguments corresponding to the first event type included in the candidate window may be determined, where the first event type is the event type corresponding to the central argument.

Based on the above description, it can be determined that the target window in this embodiment is the window that includes the most other arguments that are consistent with the event type of the central argument. Therefore, the number of arguments corresponding to the first event type included in each candidate window can be determined, where the first event type is the event type corresponding to the central argument.

For example, it can be understood with reference to FIG. 10 . As shown in FIG. 10 , it is assumed that window 1 and window 2 shown in FIG. 10 currently exist.

Among them, in window 1, in addition to the central argument, it also includes argument 1, argument 2, argument 3 and argument 4, where the event type of the central argument is event type a, and the event type of argument 1 The event type is event type b, the event type of argument 2 is event type a, the event type of argument 3 is event type b, and the event type of argument 4 is event type c. Then it can be determined that the number of other arguments included in the candidate window 1 that is consistent with the event type of the central argument is 1, that is, the argument 2 in the window 1.

And, in window 2, in addition to the central argument, it also includes argument 2, argument 3, argument 4, argument 1 and argument 5, where the event type of the central argument is event type a, The event type of argument 2 is event type a, the event type of argument 3 is event type b, the event type of argument 4 is event type a, the event type of argument 1 is event type a, the event type of argument 5 is event type is event type c. Then it can be determined that the number of other arguments included in the candidate window 1 that is consistent with the event type of the central argument is 3, that is, the argument 2, the argument 4 and the argument 1 in the window 1.

Afterwards, the target window may be determined according to the number of arguments corresponding to the first event type included in the window to be selected.

Because in this embodiment, it is necessary to determine the window that includes the most other arguments that are consistent with the event type of the central argument as the target window, in a possible implementation manner, the included first event type can be corresponding to The candidate window with the largest number of arguments is determined as the target window.

For example, in the example of FIG. 10 above, it is assumed that there are only candidate window 1 and candidate window 2, because in the example of FIG. 10, the candidate window includes other arguments that are consistent with the event type of the central argument Among the number of , the number of the window 2 to be selected is the largest, so the window 2 to be selected can be determined as the target window.

S709: Determine multiple first arguments existing in the target window.

Based on the above description, it can be determined that, in addition to the central argument, the target window can also include many other arguments. Therefore, in this embodiment, multiple first arguments existing in the target window can be determined, among which the The first argument can be, for example, the remaining arguments in the target window except the central argument.

S710. Obtain the predicted probability that the plurality of first arguments and the central argument correspond to the same event.

After multiple first arguments are determined, for example, the predicted probability that multiple first arguments and the central argument correspond to the same event can be obtained, which actually means that it is determined that the first argument and the central argument belong to the same event predicted probability.

The predicted probability in this embodiment can be obtained from the output of the second sub-model, that is to say, the second sub-model can process each of the first argument and the central argument, so as to output the first argument and the central argument corresponding to probability of the same event.

In a possible implementation manner, the current second sub-model processes the first argument and the central argument to output the probability that the first argument and the central argument correspond to the same event, which may be the second sub-model Part of the processing in the model, that is to say, the second sub-model includes the rest of the processing in addition to the processing of outputting the predicted probability that the first argument and the central argument belong to the same event. For example, the process of determining the target window and the central argument described above can also be processed by the second sub-model. In this case, the second training text can be directly input into the second sub-model. And the second sub-model will subsequently output the event existing in the text, the argument corresponding to the event, and the role corresponding to the argument.

Alternatively, the process of determining the target window and the central argument described above may not be processed by the second sub-model, but may be performed in a preprocessing stage before the second sub-model is processed. Then in this case, before the second training text is input into the second sub-model, the preprocessing process described above needs to be performed on the second training text to determine the target window and the central argument in the second training text, After that, in addition to inputting the second training sample into the second sub-model, it is also necessary to input the first text and center argument in the target window into the second sub-model, so that the second sub-model can output the first argument and center The predicted probability that arguments belong to the same event.

S711. According to the predicted probability corresponding to each first argument, determine the first argument whose corresponding predicted probability is greater than or equal to the probability threshold as the target argument.

In this embodiment, each first argument has a predicted probability corresponding to the central argument. In a possible implementation, if the predicted probability corresponding to the first argument is greater than or equal to the probability threshold, it indicates that the current The first argument and the central argument are more likely to belong to the same event, so the first argument whose corresponding predicted probability is greater than or equal to the probability threshold can be determined as the target argument. The current target argument is the argument that belongs to the same event as the central argument.

And in another possible implementation, if the predicted probability corresponding to the first argument is less than the probability threshold, it means that the current first argument and the central argument are less likely to belong to the same event, so this part of the first argument is less likely to belong to the same event. An argument is also not the target argument required in this embodiment.

And it is understandable that there will actually be multiple central arguments, and the above operations will be performed for each central argument, assuming that a first argument in the current target window and the current central argument do not belong to the same an event, but the first argument may belong to the same event as the remaining central arguments.

Therefore, it can be understood that in this embodiment, the first argument and the central argument are determined by determining the central argument and determining the predicted probability of each first argument and the central argument in the target window in the central argument Whether they belong to the same event is actually based on the central argument, which realizes the effective division of mixed events.

And in the actual implementation process, the specific setting of the probability threshold may be selected and set according to actual requirements, and the specific implementation of the probability threshold is not limited in this embodiment.

S712. Determine the predicted event corresponding to the central argument, wherein the predicted event includes the central argument and the target argument.

After the target argument is determined for the current central argument, the predicted event corresponding to the central argument can be determined. For the specific implementation of the event corresponding to the argument, refer to the implementation of event extraction in related technologies. This embodiment does not limit this.

After determining the predicted event corresponding to the central argument, because it is determined that the target argument and the central argument described above belong to the same event, it can be determined that the predicted event includes the central argument and each target argument.

And in the actual implementation process, because there are multiple central arguments, the operations described above are performed for each central argument, so that the corresponding predicted events can be determined for each central argument, and the predicted events also include and The central argument belongs to each target argument of an event, so multiple prediction events can be extracted from the sample text, and each prediction event includes at least one prediction argument.

S713. Determine a third loss according to the prediction argument in the prediction event and the second sample argument in the sample event.

After each prediction event is determined, because the second training sample also includes sample events, each sample event includes at least one second sample argument. Then the sample event is equivalent to the label information, and the predicted event is equivalent to the information output by the model, and then the third loss can be determined according to the prediction argument in the sample event and the second sample argument in the sample event.

In a possible implementation manner, for example, the third loss may be determined by processing the sample event and the predicted event through a third loss function. The third loss in this embodiment is used to optimize the accuracy of the predicted events output by the second sub-model. Further, it is used to optimize the division of predicted events output by the second sub-model, and the components included in the predicted events. The accuracy of the prediction arguments.

S714. Update the model parameters of the second sub-model according to the third loss.

The third loss can optimize the accuracy of the predicted events output by the second sub-model. Therefore, after the third loss is determined, the model parameters of the second sub-model can be updated according to the third loss to achieve Training of the second submodel.

And it can be understood that, in the actual implementation process, the above-mentioned operations of determining the loss and updating the model parameters of the second sub-model according to the loss can be iteratively performed for many times until the preset number of iterations is reached, or until the model converges. , the second sub-model after training can be obtained.

In the event extraction model training method provided by the embodiment of the present disclosure, the first probability corresponding to each second sample role in the second training sample is determined, and the respective role coefficients of each second sample role are determined, and then according to the role coefficient and the first probability of each second sample role A probability determines the priority of each second sample role, and then at least one central sample role is determined according to the priority, wherein the central sample role is a sample role with a strong ability to distinguish events, and then according to each central argument under the central sample role , determine the target window corresponding to each central argument. The target window in this embodiment is based on the central argument, and includes the most other arguments of the same event type as the central argument. window. Then, in the target window of the central argument, the predicted probability of each other argument and the central argument belonging to the same event is determined, and the predicted event corresponding to the central argument is determined according to the predicted probability, so that the central argument and the central argument can be effectively based on the prediction probability. Prediction probability, which divides multiple events in a complex sentence to output multiple predicted events, each of which includes at least one predicted argument. And according to the predicted events outputted by the model and the marked sample events, the model parameters of the second sub-model are updated, so that the training of the second sub-model can be accurately and effectively implemented, so that the second sub-model can accurately and effectively output including The predicted event for the corresponding argument.

The above describes the training process for the first sub-model and the second sub-model. Based on the above introduction, it can be determined that the first sub-model in the present disclosure can extract arguments and output the arguments corresponding to each argument. The meta-role, event type, and the second sub-model in the present disclosure can implement the arguments extracted according to the first sub-model to determine the respective events corresponding to each argument, that is, to realize the division of events. Then the first sub-model and the second sub-model are applied in combination to realize event extraction. Specifically, the events extracted from the text can be obtained, and the arguments included in each event, the argument roles corresponding to the arguments, and the event types corresponding to the arguments can be determined.

On the basis of the above embodiments, the event extraction method provided in the present disclosure will be further described in detail below with reference to FIG. 11 and FIG. 12 .

FIG. 11 is a flowchart of an event extraction method provided by an embodiment of the present disclosure, and FIG. 12 is a schematic processing diagram of the event extraction method provided by an embodiment of the present disclosure.

As shown in Figure 11, the method includes:

S1101. Acquire the first text to be processed.

In this embodiment, the first text to be processed is the text for which event extraction needs to be performed. This embodiment does not limit the specific content, length, format, etc. of the first text. It can be used as the first text in this embodiment.

S1102. Process the first text by using the first sub-model in the pre-trained event extraction model to obtain a first output result, where the first output result includes: arguments existing in the first text, roles corresponding to the arguments, and The event type to which the argument corresponds.

The pre-trained event extraction model in this embodiment may include a first sub-model and a second sub-model, wherein the first sub-model may process the first text, thereby outputting arguments, various arguments existing in the first text The corresponding argument roles and the corresponding event types of each argument.

Referring to FIG. 12 , the first sub-model of the model can be extracted from the first text input event, so that the first sub-model outputs a first output result, wherein the first output result includes the arguments existing in the first text and the arguments corresponding to the arguments. The argument role and the event type corresponding to the argument.

S1103: Process the first output result by using the second sub-model in the pre-trained event extraction model to obtain an event existing in the first text and an argument corresponding to the event.

And the pre-trained event extraction model in this embodiment also includes a second sub-model, wherein the second sub-model can process the first output result of the first sub-model, thereby outputting the events existing in the first text and the corresponding events of the events. Argument.

Based on the introduction of the above embodiments, it can be determined that, in fact, the processing of the second sub-model only requires the extracted arguments and the first text to be processed. Therefore, in a possible implementation manner, referring to FIG. 12 , Each argument in the first output result can be obtained.

And each argument and the first text are input into the second sub-model, so that the second sub-model can output the events existing in the first text and the arguments included in each event.

Then, referring to FIG. 12, it can be determined that the first sub-model can output each argument in the first text, the argument role corresponding to each argument, the event type corresponding to each argument, and the second sub-model Events existing in the first text and arguments corresponding to each event can be output.

Then, the extraction result shown in FIG. 12 can be obtained, that is, each event existing in the first text, the event type of each event, the arguments included in each event, and the argument role of each argument. And in an optional implementation manner, the second sub-model can also output trigger words of events, then the extraction results can also include trigger words of each event, so that events can be effectively extracted from the first text.

In this embodiment, based on the model training process described above, it can be understood that the pre-trained event extraction model in this embodiment can also achieve accurate and effective model training for arguments that appear multiple times in the text , and for the complex situation where multiple events are mixed, the division of events can be effectively realized. Therefore, the first text is processed based on the event extraction model obtained by the above training to obtain an event extraction result, thereby effectively ensuring the accuracy of the event extraction result.

FIG. 13 is a schematic structural diagram of an event extraction model training apparatus according to an embodiment of the present disclosure. As shown in FIG. 13 , the event extraction model training apparatus 1300 in this embodiment may include: an acquisition module 1301 , a first processing module 1302 , a second acquisition module 1303 , a second processing module 1304 , and a determination module 1305 .

The acquisition module 1301 is configured to acquire a first training sample, where the first training sample includes a first sample text and first labeled data, and the first labeled data includes: a plurality of samples in the first sample text The multiple data packets corresponding to the argument, the sample role corresponding to each of the data packets, and the sample event type corresponding to each of the data packets, wherein the sample argument in any one of the data packets is the same;

The first processing module 1302 is configured to perform model training through the first training sample to obtain a first sub-model, where the first sub-model is used to determine the arguments existing in the text, the roles corresponding to the arguments and the The event type corresponding to the argument;

The second obtaining module 1303 is configured to obtain a second training sample, where the second training sample includes the second sample text, a plurality of sample events existing in the second sample text, and the sample events included in each of the sample events. second sample argument;

The second processing module 1304 is configured to perform model training through the second training sample to obtain a second sub-model, where the second sub-model is used to determine an event existing in the text and an argument corresponding to the event;

A determination module 1305, configured to determine an event extraction model based on the first sub-model and the second sub-model.

In a possible implementation manner, the first processing module 1302 is specifically configured to:

The first sample text is processed by the first sub-model to be trained to obtain first prediction data, where the first prediction data includes a plurality of prediction arguments, the prediction roles corresponding to the prediction arguments, and the predicted event type corresponding to the predicted argument;

According to the first annotation data and the first prediction data, the model parameters of the first sub-model are updated.

In a possible implementation manner, the first prediction data further includes prediction positions of the plurality of prediction arguments in the first sample text;

The first processing module 1302 is specifically used for:

determining the first loss according to the first labeling data, the predicted role corresponding to the prediction argument, and the predicted event type corresponding to the prediction argument;

determining a second loss according to the predicted positions of the plurality of prediction arguments and the actual positions of the plurality of prediction arguments in the first sample text;

According to the first loss and the second loss, the model parameters of the first sub-model are updated.

In a possible implementation manner, the first prediction data further includes the probability of prediction positions of the plurality of prediction arguments in the first sample text;

The first processing module 1302 is specifically used for:

Grouping the plurality of prediction arguments to obtain multiple groups of prediction arguments, and the arguments in each group of prediction arguments are the same;

According to the probabilities of the predicted positions of the plurality of prediction arguments in the first sample text, target prediction arguments are respectively determined in the groups of prediction arguments, wherein the target in a group of prediction arguments The probability of the predicted position of the predicted argument in the first sample text is the highest;

The second loss is determined according to the predicted position of the target argument and the actual position of the target argument in the first sample text.

In a possible implementation manner, the second processing module 1304 is specifically configured to:

The second sample text and the plurality of second sample arguments are processed by the second sub-model to be trained to obtain at least one predicted event, and the predicted event includes at least one predicted argument;

determining a third loss based on the predicted argument in the predicted event and the second sample argument in the sample event;

The model parameters of the second sub-model are updated according to the third loss.

In a possible implementation manner, the second processing module 1304 is specifically configured to:

determining a central argument according to the second sample text;

determining a target window corresponding to the central argument in the second sample text, and the target window includes a preset number of characters;

determining multiple first arguments existing in the target window, and obtaining the predicted probability that the multiple first arguments and the central argument correspond to the same event;

At least one predicted event is determined according to the central argument, each of the first arguments, and the predicted probability corresponding to each of the first arguments.

In a possible implementation manner, the second processing module 1304 is specifically configured to:

determining a plurality of candidate windows in the second sample text, the to-be-selected windows include the preset number of characters, and the to-be-selected windows include arguments corresponding to the to-be-selected central role;

determining the number of arguments corresponding to the first event type included in the candidate window, where the first event type is the event type corresponding to the central argument;

The target window is determined according to the number of arguments corresponding to the first event type included in the candidate window.

In a possible implementation manner, the second processing module 1304 is specifically configured to:

A candidate window with the largest number of included arguments corresponding to the first event type is determined as the target window.

In a possible implementation manner, the second processing module 1304 is specifically configured to:

determining a plurality of second sample roles corresponding to the plurality of second sample arguments;

determining the first probability that the second sample arguments under each of the second sample roles correspond to the same event;

Determine the role coefficient of each of the second sample roles according to the recall rate and accuracy rate of each of the second sample roles;

determining the central character to be selected among the plurality of second sample characters according to the first probability corresponding to each of the second sample characters and the character coefficient of each of the second sample characters;

The argument corresponding to the central role to be selected is determined as the central argument.

In a possible implementation manner, for any second sample role; the second processing module 1304 is specifically used for:

The recall rate and the accuracy rate of the second sample character are processed according to a preset function, and the character coefficient of the second sample character is obtained.

In a possible implementation manner, the second processing module 1304 is specifically configured to:

determining the priority of each of the second sample characters according to the first probability corresponding to each of the second sample characters and the role coefficient of each of the second sample characters;

According to the priority of each of the second sample roles, the candidate central role is determined among the plurality of second sample roles.

In a possible implementation manner, for any second sample role; the second processing module 1304 is specifically used for:

The product of the first probability that the second sample character corresponds to an event and the character coefficient of the second sample character is determined as the priority of each second sample character.

In a possible implementation manner, the second processing module 1304 is specifically configured to:

If the priority of a second sample role in the plurality of second sample roles is greater than or equal to a preset threshold, determining the second sample role with a priority greater than or equal to the preset threshold as the central role to be selected;

If the priorities of the plurality of second sample characters are all smaller than the preset threshold, the second sample character with the highest priority among the plurality of second sample characters is determined as the central character to be selected.

In a possible implementation manner, the second processing module 1304 is specifically configured to:

According to the predicted probability corresponding to each of the first arguments, the corresponding first argument whose predicted probability is greater than or equal to the probability threshold is determined as the target argument;

A predicted event corresponding to the central argument is determined, wherein the predicted event includes the central argument and the target argument.

FIG. 14 is a schematic structural diagram of an event extraction apparatus according to an embodiment of the disclosure. As shown in FIG. 14 , the event extraction apparatus 1400 in this embodiment may include: an acquisition module 1401 , a first processing module 1402 , and a second acquisition module 1403 .

an acquisition module 1401, configured to acquire the first text to be processed;

The first processing module 1402 is configured to process the first text through the first sub-model in the pre-trained event extraction model to obtain a first output result, where the first output result includes: the first text The arguments present in the argument, the roles corresponding to the arguments, and the event types corresponding to the arguments;

The second processing module 1403 is configured to process the first output result through the second sub-model in the pre-trained event extraction model to obtain the event existing in the first text and the argument corresponding to the event.

In a possible implementation manner, the second processing module 1403 is specifically used for:

obtaining each argument in the first output result;

The respective arguments and the first text are input into the second sub-model, so that the second sub-model outputs events existing in the first text and arguments corresponding to the events.

The present disclosure provides an event extraction model training method and device, and an event extraction method and device, which are applied to artificial intelligence technology fields such as knowledge graphs and deep learning, so as to achieve the effect of improving the accuracy of the extraction results of the event extraction model.

It should be noted that the human head model in this embodiment is not a human head model for a specific user, and cannot reflect the personal information of a specific user. It should be noted that the two-dimensional face image in this embodiment comes from a public data set.

In the technical solutions of the present disclosure, the collection, storage, use, processing, transmission, provision, and disclosure of the user's personal information involved are all in compliance with relevant laws and regulations, and do not violate public order and good customs.

According to embodiments of the present disclosure, the present disclosure also provides an electronic device, a readable storage medium, and a computer program product.

According to an embodiment of the present disclosure, the present disclosure also provides a computer program product, the computer program product includes: a computer program, the computer program is stored in a readable storage medium, and at least one processor of the electronic device can read from the readable storage medium A computer program is taken, and at least one processor executes the computer program so that the electronic device executes the solution provided by any of the foregoing embodiments.

15 shows a schematic block diagram of an example electronic device 1500 that may be used to implement embodiments of the present disclosure. Electronic devices are intended to represent various forms of digital computers, such as laptops, desktops, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. Electronic devices may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are by way of example only, and are not intended to limit implementations of the disclosure described and/or claimed herein.

As shown in FIG. 15, the device 1500 includes a computing unit 1501 that can be executed according to a computer program stored in a read only memory (ROM) 1502 or a computer program loaded from a storage unit 1508 into a random access memory (RAM) 1503 Various appropriate actions and handling. In the RAM 1503, various programs and data necessary for the operation of the device 1500 can also be stored. The computing unit 1501 , the ROM 1502 , and the RAM 1503 are connected to each other through a bus 1504 . An input/output (I/O) interface 1505 is also connected to bus 1504 .

Various components in the device 1500 are connected to the I/O interface 1505, including: an input unit 1506, such as a keyboard, mouse, etc.; an output unit 1507, such as various types of displays, speakers, etc.; a storage unit 1508, such as a magnetic disk, an optical disk, etc. ; and a communication unit 1509, such as a network card, modem, wireless communication transceiver, and the like. The communication unit 1509 allows the device 1500 to exchange information/data with other devices through a computer network such as the Internet and/or various telecommunication networks.

Computing unit 1501 may be various general-purpose and/or special-purpose processing components with processing and computing capabilities. Some examples of computing units 1501 include, but are not limited to, central processing units (CPUs), graphics processing units (GPUs), various specialized artificial intelligence (AI) computing chips, various computing units that run machine learning model algorithms, digital signal processing processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 1501 performs the various methods and processes described above, such as an event extraction model training method or an event extraction method. For example, in some embodiments, the event extraction model training method or the event extraction method may be implemented as a computer software program tangibly embodied on a machine-readable medium, such as storage unit 1508 . In some embodiments, part or all of the computer program may be loaded and/or installed on device 1500 via ROM 1502 and/or communication unit 1509 . When the computer program is loaded into the RAM 1503 and executed by the computing unit 1501, one or more steps of the event extraction model training method or the event extraction method described above may be performed. Alternatively, in other embodiments, the computing unit 1501 may be configured to perform the event extraction model training method or the event extraction method by any other suitable means (eg, by means of firmware).

Various implementations of the systems and techniques described herein above may be implemented in digital electronic circuitry, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips system (SOC), complex programmable logic device (CPLD), computer hardware, firmware, software, and/or combinations thereof. These various embodiments may include being implemented in one or more computer programs executable and/or interpretable on a programmable system including at least one programmable processor that The processor, which may be a special purpose or general-purpose programmable processor, may receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device an output device.

Program code for implementing the methods of the present disclosure may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general purpose computer, special purpose computer or other programmable data processing apparatus, such that the program code, when executed by the processor or controller, performs the functions/functions specified in the flowcharts and/or block diagrams. Action is implemented. The program code may execute entirely on the machine, partly on the machine, partly on the machine and partly on a remote machine as a stand-alone software package or entirely on the remote machine or server.

In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in connection with the instruction execution system, apparatus or device. The machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. Machine-readable media may include, but are not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media would include one or more wire-based electrical connections, portable computer disks, hard disks, random access memory (RAM), read only memory (ROM), erasable programmable read only memory (EPROM or flash memory), fiber optics, compact disk read only memory (CD-ROM), optical storage, magnetic storage, or any suitable combination of the foregoing.

To provide interaction with a user, the systems and techniques described herein may be implemented on a computer having a display device (eg, a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user ); and a keyboard and pointing device (eg, a mouse or trackball) through which a user can provide input to the computer. Other kinds of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (eg, visual feedback, auditory feedback, or tactile feedback); and can be in any form (including acoustic input, voice input, or tactile input) to receive input from the user.

The systems and techniques described herein may be implemented on a computing system that includes back-end components (eg, as a data server), or a computing system that includes middleware components (eg, an application server), or a computing system that includes front-end components (eg, a user computer having a graphical user interface or web browser through which a user may interact with implementations of the systems and techniques described herein), or including such backend components, middleware components, Or any combination of front-end components in a computing system. The components of the system may be interconnected by any form or medium of digital data communication (eg, a communication network). Examples of communication networks include: Local Area Networks (LANs), Wide Area Networks (WANs), and the Internet.

A computer system can include clients and servers. Clients and servers are generally remote from each other and usually interact through a communication network. The relationship of client and server arises by computer programs running on the respective computers and having a client-server relationship to each other. The server can be a cloud server, also known as a cloud computing server or a cloud host. It is a host product in the cloud computing service system to solve the traditional physical host and VPS service ("Virtual Private Server", or "VPS" for short). , there are the defects of difficult management and weak business expansion. The server can also be a server of a distributed system, or a server combined with a blockchain.

It should be understood that steps may be reordered, added or deleted using the various forms of flow shown above. For example, the steps described in the present disclosure can be executed in parallel, sequentially, or in different orders. As long as the desired results of the technical solutions disclosed in the present disclosure can be achieved, there is no limitation herein.

The above-mentioned specific embodiments do not constitute a limitation on the protection scope of the present disclosure. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions may occur depending on design requirements and other factors. Any modifications, equivalent replacements, and improvements made within the spirit and principles of the present disclosure should be included within the protection scope of the present disclosure.

Claims (35)

1. An event extraction model training method comprises the following steps:

obtaining a first training sample, wherein the first training sample comprises a first sample text and first labeling data, and the first labeling data comprises: a plurality of data packets corresponding to a plurality of sample arguments in the first sample text, a sample role corresponding to each data packet, and a sample event type corresponding to each data packet, wherein the sample arguments in any data packet are the same;

performing model training through the first training sample to obtain a first sub-model, wherein the first sub-model is used for determining arguments existing in a text, roles corresponding to the arguments and event types corresponding to the arguments;

obtaining a second training sample, wherein the second training sample comprises a second sample text, a plurality of sample events existing in the second sample text, and a second sample argument included in each sample event;

performing model training through the second training sample to obtain a second sub-model, wherein the second sub-model is used for determining events existing in the text and arguments corresponding to the events;

determining an event extraction model based on the first sub-model and the second sub-model.

2. The method of claim 1, wherein model training through the first training sample results in a first sub-model comprising:

processing the first sample by the first sub-model to be trained to obtain first prediction data, wherein the first prediction data comprises a plurality of prediction arguments, a prediction role corresponding to the prediction arguments and a prediction event type corresponding to the prediction arguments;

and updating the model parameters of the first sub-model according to the first marking data and the first prediction data.

3. The method of claim 2, wherein the first prediction data further includes predicted locations of the plurality of prediction arguments in the first sample text;

updating model parameters of the first sub-model according to the first annotation data and the first prediction data, including:

determining a first loss according to the first labeled data, the predicted role corresponding to the predicted argument and the predicted event type corresponding to the predicted argument;

determining a second loss according to the predicted positions of the plurality of predicted arguments and the actual positions of the plurality of predicted arguments in the first sample text;

and updating the model parameters of the first submodel according to the first loss and the second loss.

4. The method of claim 3, wherein the first prediction data further includes probabilities of predicted locations of the plurality of prediction arguments in the first sample text;

determining a second loss based on the predicted locations of the plurality of predicted arguments and the actual locations of the plurality of predicted arguments in the first sample text, comprising:

grouping the multiple prediction arguments to obtain multiple groups of prediction arguments, wherein the arguments in each group of prediction arguments are the same;

determining target prediction arguments in the plurality of groups of prediction arguments respectively according to the probabilities of the prediction positions of the plurality of prediction arguments in the first sample text, wherein the probability of the prediction position of the target prediction argument in the group of prediction arguments in the first sample text is highest;

and determining the second loss according to the predicted position of the target argument and the actual position of the target argument in the first sample text.

5. The method of any of claims 1-4, wherein model training through the second training sample results in a second sub-model comprising:

processing the second sample text and the plurality of second sample arguments through the second submodel to be trained to obtain at least one predicted event, wherein the predicted event comprises at least one predicted argument;

determining a third loss according to a prediction argument in the prediction event and a second sample argument in the sample event;

and updating the model parameters of the second submodel according to the third loss.

6. The method of claim 5, wherein processing the second sample text and the plurality of second sample arguments with the second submodel to be trained to derive at least one predicted event comprises:

determining a central argument according to the second sample text;

determining a target window corresponding to the central argument in the second sample text, wherein the target window comprises a preset number of characters;

determining a plurality of first arguments existing in the target window, and acquiring the prediction probability that the plurality of first arguments and the central argument correspond to the same event;

and determining at least one predicted event according to the central argument, each first argument and the prediction probability corresponding to each first argument.

7. The method of claim 6, wherein determining a target window in the second sample text to which the central argument corresponds comprises:

determining a plurality of windows to be selected in the second sample text, wherein the windows to be selected comprise the characters with the preset number, and the windows to be selected comprise arguments corresponding to the central roles to be selected;

determining the number of arguments corresponding to a first event type included in the window to be selected, wherein the first event type is the event type corresponding to the central argument;

and determining the target window according to the number of arguments corresponding to the first event type in the window to be selected.

8. The method of claim 7, wherein determining the target window according to the number of arguments corresponding to the first event type included in the window to be selected comprises:

and determining the window to be selected with the maximum number of arguments corresponding to the first event type as the target window.

9. The method of any of claims 6-8, wherein determining a central argument from the second sample text comprises:

determining a plurality of second sample roles corresponding to the plurality of second sample arguments;

determining a first probability that a second sample argument under each second sample role corresponds to the same event;

determining the role coefficient of each second sample role according to the recall rate and the accuracy rate of each second sample role;

determining the center role to be selected in the plurality of second sample roles according to the first probability corresponding to each second sample role and the role coefficient of each second sample role;

and determining the argument corresponding to the central role to be selected as the central argument.

10. The method of claim 9, wherein for any one second sample role; determining the role coefficient of the second sample role according to the recall rate and the accuracy rate of the second sample role, wherein the role coefficient comprises the following steps:

and processing the recall rate and the accuracy rate of the second sample role according to a preset function to obtain the role coefficient of the second sample role.

11. The method according to claim 9 or 10, wherein determining the candidate center role among the plurality of second sample roles according to the first probability corresponding to each of the second sample roles and the role coefficient of each of the second sample roles includes:

determining the priority of each second sample role according to the corresponding first probability of each second sample role and the role coefficient of each second sample role;

and determining the role of the center to be selected in the plurality of second sample roles according to the priority of each second sample role.

12. The method of claim 11, wherein for any one second sample role; determining the priority of each second sample role according to the first probability corresponding to each second sample role and the role coefficient of each second sample role, including:

and determining the product of the first probability of the second sample role corresponding to one event and the role coefficient of the second sample role as the priority of each second sample role.

13. The method according to claim 11 or 12, wherein determining the candidate center role among the plurality of second sample roles according to the priority of each of the second sample roles comprises:

if the priority of a second sample role in the plurality of second sample roles is greater than or equal to a preset threshold, determining the second sample role with the priority greater than or equal to the preset threshold as the center role to be selected;

and if the priorities of the plurality of second sample roles are all smaller than the preset threshold, determining the second sample role with the highest priority in the plurality of second sample roles as the center role to be selected.

14. The method of any one of claims 6 to 13, wherein said determining at least one predicted event from said central argument, each of said first arguments, and a predicted probability corresponding to each of said first arguments comprises:

determining the first argument with the corresponding prediction probability larger than or equal to a probability threshold value as a target argument according to the prediction probability corresponding to each first argument;

and determining a predicted event corresponding to the central argument, wherein the predicted event comprises the central argument and the target argument.

15. An event extraction method, comprising:

acquiring a first text to be processed;

processing the first text through a first sub-model in a pre-trained event extraction model to obtain a first output result, wherein the first output result comprises: argument existing in the first text, role corresponding to the argument and event type corresponding to the argument;

and processing the first output result through a second sub-model in the pre-trained event extraction model to obtain an event existing in the first text and an argument corresponding to the event.

16. The method of claim 15, wherein the processing the first output result through a second submodel in the pre-trained event extraction model to obtain an event existing in the first text and an argument corresponding to the event comprises:

obtaining each argument in the first output result;

and inputting each argument and the first text into the second submodel, so that the second submodel outputs events existing in the first text and arguments corresponding to the events.

17. An event extraction model training device, comprising:

an obtaining module, configured to obtain a first training sample, where the first training sample includes a first sample text and first labeling data, and the first labeling data includes: a plurality of data packets corresponding to a plurality of sample arguments in the first sample text, a sample role corresponding to each data packet, and a sample event type corresponding to each data packet, wherein the sample arguments in any data packet are the same;

the first processing module is used for carrying out model training through the first training sample to obtain a first sub-model, and the first sub-model is used for determining arguments existing in a text, roles corresponding to the arguments and event types corresponding to the arguments;

a second obtaining module, configured to obtain a second training sample, where the second training sample includes a second sample text, a plurality of sample events existing in the second sample text, and a second sample argument included in each sample event;

the second processing module is used for carrying out model training through the second training sample to obtain a second sub-model, and the second sub-model is used for determining events existing in the text and arguments corresponding to the events;

a determination module to determine an event extraction model based on the first submodel and the second submodel.

18. The apparatus of claim 17, wherein the first processing module is specifically configured to:

processing the first sample by the first sub-model to be trained to obtain first prediction data, wherein the first prediction data comprises a plurality of prediction arguments, a prediction role corresponding to the prediction arguments and a prediction event type corresponding to the prediction arguments;

and updating the model parameters of the first sub-model according to the first marking data and the first prediction data.

19. The apparatus of claim 18, wherein the first prediction data further comprises predicted locations of the plurality of prediction arguments in the first sample text;

the first processing module is specifically configured to:

determining a first loss according to the first labeled data, the predicted role corresponding to the predicted argument and the predicted event type corresponding to the predicted argument;

determining a second loss according to the predicted positions of the plurality of predicted arguments and the actual positions of the plurality of predicted arguments in the first sample text;

and updating the model parameters of the first submodel according to the first loss and the second loss.

20. The apparatus of claim 19, wherein the first prediction data further comprises probabilities of predicted locations of the plurality of prediction arguments in the first sample text;

the first processing module is specifically configured to:

grouping the multiple prediction arguments to obtain multiple groups of prediction arguments, wherein the arguments in each group of prediction arguments are the same;

determining target prediction arguments in the plurality of groups of prediction arguments respectively according to the probabilities of the prediction positions of the plurality of prediction arguments in the first sample text, wherein the probability of the prediction position of the target prediction argument in the group of prediction arguments in the first sample text is highest;

and determining the second loss according to the predicted position of the target argument and the actual position of the target argument in the first sample text.

21. The apparatus according to any one of claims 17 to 20, wherein the second processing module is specifically configured to:

processing the second sample text and the plurality of second sample arguments through the second submodel to be trained to obtain at least one predicted event, wherein the predicted event comprises at least one predicted argument;

determining a third loss according to a prediction argument in the prediction event and a second sample argument in the sample event;

and updating the model parameters of the second submodel according to the third loss.

22. The apparatus of claim 21, wherein the second processing module is specifically configured to:

determining a central argument according to the second sample text;

determining a target window corresponding to the central argument in the second sample text, wherein the target window comprises a preset number of characters;

determining a plurality of first arguments existing in the target window, and acquiring the prediction probability that the plurality of first arguments and the central argument correspond to the same event;

and determining at least one predicted event according to the central argument, each first argument and the prediction probability corresponding to each first argument.

23. The apparatus of claim 22, wherein the second processing module is specifically configured to:

determining a plurality of windows to be selected in the second sample text, wherein the windows to be selected comprise the characters with the preset number, and the windows to be selected comprise arguments corresponding to the central roles to be selected;

determining the number of arguments corresponding to a first event type included in the window to be selected, wherein the first event type is the event type corresponding to the central argument;

and determining the target window according to the number of arguments corresponding to the first event type in the window to be selected.

24. The apparatus of claim 23, wherein the second processing module is specifically configured to:

and determining the window to be selected with the maximum number of arguments corresponding to the first event type as the target window.

25. The apparatus according to any one of claims 22-24, wherein the second processing module is specifically configured to:

determining a plurality of second sample roles corresponding to the plurality of second sample arguments;

determining a first probability that a second sample argument under each second sample role corresponds to the same event;

determining the role coefficient of each second sample role according to the recall rate and the accuracy rate of each second sample role;

determining the center role to be selected in the plurality of second sample roles according to the first probability corresponding to each second sample role and the role coefficient of each second sample role;

and determining the argument corresponding to the central role to be selected as the central argument.

26. The apparatus of claim 25, wherein for any one second sample role; the second processing module is specifically configured to:

and processing the recall rate and the accuracy rate of the second sample role according to a preset function to obtain the role coefficient of the second sample role.

27. The apparatus according to claim 25 or 26, wherein the second processing module is specifically configured to:

determining the priority of each second sample role according to the corresponding first probability of each second sample role and the role coefficient of each second sample role;

and determining the role of the center to be selected in the plurality of second sample roles according to the priority of each second sample role.

28. The apparatus of claim 27, wherein for any one second sample role; the second processing module is specifically configured to:

and determining the product of the first probability of the second sample role corresponding to one event and the role coefficient of the second sample role as the priority of each second sample role.

29. The apparatus according to claim 27 or 28, wherein the second processing module is specifically configured to:

if the priority of a second sample role in the plurality of second sample roles is greater than or equal to a preset threshold, determining the second sample role with the priority greater than or equal to the preset threshold as the center role to be selected;

and if the priorities of the plurality of second sample roles are all smaller than the preset threshold, determining the second sample role with the highest priority in the plurality of second sample roles as the center role to be selected.

30. The apparatus according to any one of claims 22-29, wherein the second processing module is specifically configured to:

determining the first argument with the corresponding prediction probability larger than or equal to a probability threshold value as a target argument according to the prediction probability corresponding to each first argument;

and determining a predicted event corresponding to the central argument, wherein the predicted event comprises the central argument and the target argument.

31. An event extraction device comprising:

the acquisition module is used for acquiring a first text to be processed;

a first processing module, configured to process the first text through a first sub-model in a pre-trained event extraction model to obtain a first output result, where the first output result includes: argument existing in the first text, role corresponding to the argument and event type corresponding to the argument;

and the second processing module is used for processing the first output result through a second sub-model in the pre-trained event extraction model to obtain the event existing in the first text and the argument corresponding to the event.

32. The apparatus of claim 31, wherein the second processing module is specifically configured to:

obtaining each argument in the first output result;

and inputting each argument and the first text into the second submodel, so that the second submodel outputs events existing in the first text and arguments corresponding to the events.

33. An electronic device, comprising:

at least one processor; and

a memory communicatively coupled to the at least one processor; wherein,

the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-14 or claims 15-16.

34. A non-transitory computer readable storage medium having stored thereon computer instructions for causing the computer to perform the method of any of claims 1-14 or claims 15-16.

35. A computer program product comprising a computer program which, when executed by a processor, carries out the steps of the method of any one of claims 1 to 14 or claims 15 to 16.

Images