CN117909505A - Event argument extraction method and related equipment - Google Patents
Event argument extraction method and related equipment Download PDFInfo
- Publication number
- CN117909505A CN117909505A CN202410285755.4A CN202410285755A CN117909505A CN 117909505 A CN117909505 A CN 117909505A CN 202410285755 A CN202410285755 A CN 202410285755A CN 117909505 A CN117909505 A CN 117909505A
- Authority
- CN
- China
- Prior art keywords
- information
- event
- fusion
- module
- text
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000605 extraction Methods 0.000 title claims abstract description 76
- 238000000034 method Methods 0.000 claims abstract description 28
- 238000007781 pre-processing Methods 0.000 claims abstract description 9
- 238000012545 processing Methods 0.000 claims abstract description 8
- 238000003062 neural network model Methods 0.000 claims abstract description 7
- 230000004927 fusion Effects 0.000 claims description 77
- 230000014509 gene expression Effects 0.000 claims description 15
- 238000004590 computer program Methods 0.000 claims description 8
- 230000007246 mechanism Effects 0.000 claims description 6
- 238000001914 filtration Methods 0.000 claims description 5
- 238000004891 communication Methods 0.000 description 11
- 238000010586 diagram Methods 0.000 description 6
- 239000013598 vector Substances 0.000 description 6
- 238000002372 labelling Methods 0.000 description 5
- 238000012549 training Methods 0.000 description 5
- 238000001514 detection method Methods 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 4
- 230000006870 function Effects 0.000 description 4
- 230000009286 beneficial effect Effects 0.000 description 3
- 230000008901 benefit Effects 0.000 description 3
- 238000012986 modification Methods 0.000 description 3
- 230000004048 modification Effects 0.000 description 3
- 238000010606 normalization Methods 0.000 description 3
- 238000012217 deletion Methods 0.000 description 2
- 230000037430 deletion Effects 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 230000003068 static effect Effects 0.000 description 2
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 1
- 235000002198 Annona diversifolia Nutrition 0.000 description 1
- 241000282842 Lama glama Species 0.000 description 1
- 230000009471 action Effects 0.000 description 1
- 230000004913 activation Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000001737 promoting effect Effects 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
- 238000004904 shortening Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
Landscapes
- Machine Translation (AREA)
Abstract
The disclosure provides a method for extracting event arguments and related equipment. The extraction method comprises the following steps: acquiring an original text; preprocessing the original text to obtain an event text; processing the event text by using a preset language model to obtain abstract information, associated information and named entity information; splitting to obtain statement information according to the event text; determining an event argument corresponding to the original text by using an event argument extraction model according to the abstract information, the associated information, the named entity information and the sentence information; wherein the event argument extraction model is a trained neural network model. According to the technical scheme, the accuracy of event argument extraction is improved.
Description
Technical Field
The disclosure relates to the field of information technology, and in particular relates to a method for extracting event arguments and related equipment.
Background
With the rapid development of information technology and media such as news media and social media, information meets the daily information acquisition needs of people in various forms, such as images, texts, streaming media and the like. The problems of information redundancy, poor readability and the like caused by a large number of information groups are various, how to automatically extract structured important information from unstructured information is a problem to be solved, and how to quickly and effectively extract events as basic units for information representation is important for information extraction tasks and information understanding tasks.
Disclosure of Invention
Accordingly, an objective of the present disclosure is to provide a method and related apparatus for extracting event arguments.
Based on the above objects, the present disclosure provides a method for extracting event arguments, the method comprising:
acquiring an original text; preprocessing the original text to obtain an event text;
processing the event text by using a preset language model to obtain abstract information, associated information and named entity information;
Splitting to obtain statement information according to the event text;
determining an event argument corresponding to the original text by using an event argument extraction model according to the abstract information, the associated information, the named entity information and the sentence information; wherein the event argument extraction model is a trained neural network model.
Based on the same inventive concept, the embodiment of the disclosure further provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor, where the processor executes the program to implement the extraction method according to any one of the above.
Based on the same inventive concept, the embodiments of the present disclosure also provide a non-transitory computer-readable storage medium storing computer instructions for causing a computer to perform any one of the above-described extraction methods.
From the above, it can be seen that, according to the method and the related device for extracting an event argument provided by the embodiments of the present disclosure, an event text based on an original text is processed by using a preset language model, so as to obtain abstract information, associated information and named entity information; splitting to obtain statement information according to the event text; and determining the event argument corresponding to the original text by using an event argument extraction model according to the abstract information, the associated information, the named entity information and the statement information. According to the scheme, on the basis of fully utilizing the preset language model, the event argument is determined by combining the event argument extraction model, so that the global view of the event text and the local event association can be controlled at the same time, the accuracy of event argument extraction is improved, and particularly, the event argument extraction performance under the condition of low resources is improved.
Drawings
In order to more clearly illustrate the technical solutions of the present disclosure or related art, the drawings required for the embodiments or related art description will be briefly described below, and it is apparent that the drawings in the following description are only embodiments of the present disclosure, and other drawings may be obtained according to these drawings without inventive effort to those of ordinary skill in the art.
Fig. 1 is a schematic diagram illustrating an application scenario of a method for extracting event arguments according to an embodiment of the present disclosure;
FIG. 2 is a flow chart illustrating a method for extracting event arguments according to an embodiment of the present disclosure;
FIG. 3 is a schematic diagram of an event argument extraction model according to an embodiment of the present disclosure;
fig. 4 shows a schematic structural diagram of an electronic device provided in an embodiment of the disclosure.
Detailed Description
For the purposes of promoting an understanding of the principles and advantages of the disclosure, reference will now be made to the embodiments illustrated in the drawings and specific language will be used to describe the same.
It should be noted that unless otherwise defined, technical or scientific terms used in the embodiments of the present disclosure should be given the ordinary meaning as understood by one of ordinary skill in the art to which the present disclosure pertains. The terms "first," "second," and the like, as used in embodiments of the present disclosure, do not denote any order, quantity, or importance, but rather are used to distinguish one element from another. The word "comprising" or "comprises", and the like, means that elements or items preceding the word are included in the element or item listed after the word and equivalents thereof, but does not exclude other elements or items. The terms "connected" or "connected," and the like, are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "upper", "lower", "left", "right", etc. are used merely to indicate relative positional relationships, which may also be changed when the absolute position of the object to be described is changed.
As described in the background section, the information clusters are various and huge in number, which causes information redundancy and poor readability. Currently, there are two main ways of event extraction models: pipeline model and joint model, pipeline model means that each subtask (event detection and classification, argument detection and classification) in event extraction is carried out in a serial mode, the implementation of the mode is simple and easy to operate, and the completion of the former task can reduce the detection sample number of the latter task, but if the former task is wrong, the detection precision of the latter task can be affected, so that the problem of error propagation is caused. The joint model refers to the relation and the dependency relationship among all subtasks in the joint event extraction, and can be performed in a parallel mode, so that the problem of error propagation caused by a pipeline model can be relieved as much as possible during model training, but the problem still exists during model testing.
In addition, the event argument extraction is usually based on a generation model, and the extraction accuracy rate is not enough, and particularly, the event argument extraction method based on a large model can generate words outside the original text due to the phenomenon of 'illusion' when generating the argument, so that the event argument extraction accuracy rate is reduced.
In view of this, the embodiments of the present disclosure provide a method and related device for extracting event arguments, which processes an event text based on an original text by using a preset language model to obtain summary information, associated information, and named entity information; splitting to obtain statement information according to the event text; and determining the event argument corresponding to the original text by using an event argument extraction model according to the abstract information, the associated information, the named entity information and the statement information. According to the scheme, on the basis of fully utilizing the preset language model, the event argument is determined by combining the event argument extraction model, so that the global view of the event text and the local event association can be controlled at the same time, the accuracy of event argument extraction is improved, and particularly, the event argument extraction performance under the condition of low resources is improved.
Fig. 1 illustrates an application scenario 100 of a method for extracting event arguments according to an embodiment of the present disclosure. The application scenario 100 comprises a huge amount of text information, e.g. a first text information 101, a second text information 102. By using the extraction method of the event argument provided by the embodiment of the invention, the text information can be used for accurately extracting the event argument, thereby being beneficial to improving the efficiency of acquiring information by a user.
In order to enable the event argument extraction method provided by the embodiment of the present disclosure to be performed smoothly, a preset neural network model needs to be trained in advance to obtain an accurate event argument extraction model.
Next, from the perspective of training a preset neural network model, an event argument extraction model in the event argument extraction method provided by the embodiment of the present disclosure will be described in detail. Here, a chinese text is taken as an example for explanation.
Firstly, chinese text data is obtained from various news websites and public general field event extraction data sets (such as ACE 2005), and the obtained Chinese text data is preprocessed to obtain the Chinese text data set. Here, the chinese text data is generally long text data.
Text unrelated to the event is included in the chinese text data. Meanwhile, part of original data sources are related news websites in the general field, so html tags can appear in the data.
In view of this, in some embodiments, preprocessing includes filtering for irrelevant events in the above-mentioned chinese text data, deletion of html tags in websites, and the like.
The screening of the irrelevant events refers to setting a general field event key dictionary, setting a regular expression according to the key dictionary, and filtering the irrelevant events by using the regular expression, namely eliminating Chinese text data irrelevant to the events. In other words, if keywords in the general field event keyword dictionary appear in the Chinese text data, the data is reserved, otherwise the data is deleted, and the effect of filtering irrelevant events is achieved.
The deletion of the html tag in the website means that the html tag in the Chinese text data is deleted by using a regular expression according to the html tag, so that a standard Chinese text data set is obtained.
Next, event types, trigger words, and corresponding event arguments for typical events in the general field are defined. Wherein, the event type refers to: conflict, death, wedding, interview, transaction, etc. Trigger words refer to terms that trigger the occurrence of the typical event. An event argument refers to an entity or text capable of describing an event in detail, such as a person, place, time, short text having a specific meaning, and the like.
And manually labeling the Chinese text data set based on the defined event type, the trigger word and the corresponding event argument to obtain a labeled data set. Here, the annotation content in the annotation dataset includes the beginning and ending positions of the event type, trigger word, and argument in the text. It should be understood that the number of argument here may be plural, and in particular, the data of the argument may be different, for example, four, five, etc., depending on the specific event, which is not limited by the present disclosure. It should be noted that the labeling data set includes at least one labeling chinese text data.
And then, extracting information from the marked Chinese text data in the marked data set by using a preset language model. Here, the language model may be a large model of a dialogue type, such as ChatGPT, chatGLM, llama, etc., which is not limited by the present disclosure.
In some embodiments, the extracted information includes summary information, association information, and named entity information.
Illustratively, a Prompt (Prompt) SP designed for text summary generation includes the annotated chinese text to be extracted, the maximum length of the summary, and event type information, such as: "the following text contains conflict and death events, the following text is summarized and abstracted to no more than 150 words, 'annotated chinese text'. And then inputting ChatGPT the prompt SP to obtain abstract information of the text, wherein the abstract information can greatly delete unimportant information in the text while keeping the key information of the event, thereby playing a role in shortening the length of the document.
Illustratively, a Prompt (Prompt) RP designed for event-related information generation includes annotated chinese text to be extracted and event type information, such as: "the following text contains conflict and death events, the events in the text are analyzed step by step to determine whether the events are associated, and a conclusion is made," Chinese text is marked ". And then inputting ChatGPT the prompt RP to obtain the associated information of the text event, wherein the associated information can display the associated information between the events in the document in a natural language form, including the display association and the implicit association, so as to improve the capability of the model to learn the event characteristics.
Illustratively, a Prompt (Prompt) EP designed for named entity information generation, since the named entity may be independent of the specific event type, may only utilize the tagged chinese text to be extracted, for example: "you are an expert in the field of information extraction," labeled chinese text' "now requires you to extract named entity information that may be contained in the following document. The prompt EP is then entered ChatGPT to obtain the named entities contained in the text, such as time, place, person, etc. The named entity information can improve the sensitivity of the argument extraction model to identify argument boundaries.
Therefore, abstract information, associated information and named entity information can be obtained by using a preset language model and a preset prompt.
Considering that the text possibly contains a large number of trivial elements irrelevant to the events and has hidden and obscure association among the events, the abstract information, the association information and the named entity information of the text are generated by utilizing the powerful logical reasoning capability of the language model, so that the event argument extraction task can be effectively assisted.
And then splitting the marked Chinese text to obtain sentence information. Here, the method for splitting the labeling chinese text may be flexibly selected by a person skilled in the art, and this disclosure is not limited thereto.
In some embodiments, the pre-set neural network model includes a pre-training model BERT, a feature fusion module, and an argument boundary discrimination module.
Then, the summary information, the association information, the named entity information and the sentence information are input into a pre-training model BERT, so that rich semantic features, such as summary semantic features Fg (which may be represented as vectors), association semantic features Fr (which may be represented as vectors), entity semantic features Fe (which may be represented as vectors) and sentence semantic features Fs (which may be represented as vectors), are obtained.
Next, a feature fusion module constructed based on the attention mechanism may fuse the semantic features described above.
In some embodiments, the sentence semantic feature Fs is fused with the abstract semantic feature Fg and the associated semantic feature Fr respectively, to obtain an abstract fusion feature f cln g and an associated fusion feature f cln r.
Here, the method for fusion is exemplified as follows:
The, formula (1)
The, formula (2)
; (3)
Wherein,Is a linear layer, g and b are both learnable parameters,/>Is/>Mean value of/(I)Is thatIs a standard deviation matrix of (2).
Here, formula (3) may be split intoAnd/>; In other words,/>Representative/>(Feature vector after the fusion of the associated semantic features and the sentence semantic features, namely associated fusion features) and/>(Feature vectors after fusion of abstract semantic features and sentence semantic features, namely abstract fusion features).
Because the entity semantic feature Fe has a larger degree of contribution to the identification of event argument boundaries, the method adopts a self-attention mechanism to fuse Fe with sentence semantic features Fs and associated fusion featuresSummary fusion features/>Fusion is performed. Here, fusion was performed using the following formula (4):
Formula (4).
In some embodiments, as shown in formula (5), each will，/>，/>Substituting (4) to obtain the first fusion result/>Then, the/>, as in formula (6), will be repeated，/>，/>Substituting (4) to obtain a second fusion result/>Based on the residual network thought, pair/>And/>Layer normalization was performed and H was obtained using GeLU activation functions as shown in equation (7).
Formula (5);
Formula (6);
Formula (7).
Finally, in the argument boundary discriminating module, an event argument extracting task is defined as a discriminating task, and a start position probability and an end position probability of the event argument boundary are calculated respectively as follows:
Formula (8);
Formula (9).
Here, the Sigmoid function output probability dimension coincides with the length of the sentence information. The length of the sentence information is the number of words included in the sentence, for example, a sentence of 8 words, and the length is 8, and the corresponding probability dimension is also 8, that is, the probability that the position of each word in the sentence is the start position and the probability that the position is the end position in the Sigmoid function output sentence. It should be understood that when the event text includes a plurality of sentences, the start position probability and the end position probability of each sentence may be output.
Thus, based on the threshold value, the position exceeding the threshold value in p s and p e is acquired as the start position or end position of the argument, and then a plurality of start positions and end positions are matched according to the nearest neighbor principle, so that at least one argument can be obtained. And (3) adopting a supervised learning strategy, calculating the loss between the prediction probability and the real label by using a cross entropy loss function, updating model parameters such as g, b and a threshold value by using an Adam optimizer, and finally obtaining an event argument extraction model. Here, the threshold value may be set in advance, for example, 0.5, which is not limited by the present disclosure.
By adopting the mode, the event argument extraction task is defined as a discrimination task, and words outside the original text generated by the generating type event argument extraction method are avoided, so that the accuracy rate of text event argument extraction is improved.
The extraction method of the event argument provided by the embodiment of the disclosure highlights the hidden event association existing among the events, which is beneficial to the model to learn the event mode and master the essential characteristics of the event, reduces the dependency degree of the model on high-quality labeling data, and improves the performance of the event argument extraction task under the condition of low resources.
The embodiment of the disclosure provides a method for extracting event arguments. FIG. 2 is a flow chart illustrating a method for extracting event arguments according to an embodiment of the present disclosure; fig. 3 is a schematic structural diagram of an event argument extraction model according to an embodiment of the present disclosure. As shown in fig. 2 and 3, the method includes:
Step S202: acquiring an original text; here, the original text refers to the text to be extracted, and may come from a network; preprocessing the original text to obtain an event text; for the method of preprocessing, reference may be made to the foregoing, and no further description is given;
Step S204: processing the event text by using a preset language model to obtain abstract information, associated information and named entity information; here, the preset language model may be ChatGPT;
step S206: splitting to obtain statement information according to the event text;
Step S208: determining an event argument corresponding to the original text by using an event argument extraction model according to the abstract information, the associated information, the named entity information and the sentence information; wherein the event argument extraction model is a trained neural network model. Here, the event argument extraction model is trained by the above-described training method.
In some embodiments, step S204 includes: preprocessing based on preset regular expressions, wherein the preset regular expressions comprise regular expressions of irrelevant events and regular expressions of website labels; wherein the regular expression of the uncorrelated event is used for filtering the original text; the regular expression of the website label is used for deleting the website label. It should be noted that if the original text is filtered, it is irrelevant to the event, and a new original text needs to be retrieved.
In some embodiments, step S204: the step of processing the event text by using a preset language model to obtain abstract information, associated information and named entity information comprises the following steps:
acquiring and utilizing a preset language model to obtain abstract information according to a preset abstract information prompt;
Acquiring and utilizing a preset language model to obtain the associated information according to a preset associated information prompt;
And acquiring and obtaining the named entity information by using a preset language model according to a preset named entity information prompt.
Here, for the setting of each prompt, reference may be made to the foregoing, and no description is repeated.
In some embodiments, as shown in fig. 3, the event argument extraction model 300 includes a feature extraction module 301, a feature fusion module 302, and an argument boundary discrimination module 303; wherein the feature fusion module 302 is constructed based on an attention mechanism; the argument boundary determination module 303 is a discriminant module.
In some embodiments, the feature extraction module 301 is configured to obtain abstract semantic features, associated semantic features, entity semantic features, and sentence semantic features according to the abstract information, the associated information, the named entity information, and the sentence information.
Alternatively, the feature extraction module 301 may be a pre-trained model BERT model.
In some embodiments, the feature fusion module 302 includes a first feature fusion module and a second feature fusion module; wherein,
Referring to formulas (1) to (3), the first feature fusion module is used for obtaining summary fusion features according to fusion of the sentence semantic features and the summary semantic features, and obtaining associated fusion features according to fusion of the sentence semantic features and the associated semantic features;
Referring to the formula (4) to the formula (7), the second feature fusion module is used for: based on an attention mechanism, fusing the entity semantic features, the sentence semantic features, the abstract fusion features and the associated fusion features to obtain a second fusion result; and normalizing the second fusion result and the summary fusion characteristic based on a residual network idea to obtain fusion data.
In some embodiments, the second feature fusion module includes a third feature fusion module, a fourth feature fusion module, and a normalization module; wherein,
Referring to the step (5), the third feature fusion module is configured to fuse the entity semantic features, the sentence semantic features and the abstract fusion features to obtain a first fusion result;
referring to formula (6), the fourth feature fusion module is configured to fuse the entity semantic feature, the first fusion result and the associated fusion feature to obtain a second fusion result;
And (7) according to the reference formula, the normalization module is used for normalizing to obtain the fusion data H according to the second fusion result and the summary fusion characteristics.
In some embodiments, the feature fusion module is configured to obtain fusion data; referring to formulas (8) and (9), the argument boundary discriminating module 303 is configured to determine, according to the fusion data H, a start position probability and an end position probability of each position of the sentence information as an argument boundary, and determine a start position and an end position of the event argument based on the start position probability and the end position probability.
It should be noted that the extraction method according to the embodiments of the present disclosure may be performed by a single device, for example, a computer or a server. The extraction method of the embodiment can also be applied to a distributed scene, and is completed by mutually matching a plurality of devices. In the case of such a distributed scenario, one of the devices may perform only one or more steps of the extraction method of the embodiments of the present disclosure, and the devices may interact with each other to complete the extraction method.
It should be noted that the foregoing describes some embodiments of the present disclosure. Other embodiments are within the scope of the following claims. In some cases, the actions or steps recited in the claims may be performed in a different order than in the embodiments described above and still achieve desirable results. In addition, the processes depicted in the accompanying figures do not necessarily require the particular order shown, or sequential order, to achieve desirable results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.
Based on the same inventive concept, the present disclosure also provides an electronic device corresponding to the extraction method of any of the above embodiments, including a memory, a processor, and a computer program stored on the memory and executable on the processor, where the processor implements the extraction method of any of the above embodiments when executing the program.
Fig. 4 shows a more specific hardware architecture of an electronic device according to this embodiment, where the device may include: a processor 1010, a memory 1020, an input/output interface 1030, a communication interface 1040, and a bus 1050. Wherein processor 1010, memory 1020, input/output interface 1030, and communication interface 1040 implement communication connections therebetween within the device via a bus 1050.
The processor 1010 may be implemented by a general-purpose CPU (Central Processing Unit ), a microprocessor, an Application SPECIFIC INTEGRATED Circuit (ASIC), or one or more integrated circuits, etc. for executing related programs to implement the technical solutions provided in the embodiments of the present disclosure.
The memory 1020 may be implemented in the form of ROM (read only memory), RAM (Random Access Memory ), static storage, dynamic storage, etc. Memory 1020 may store an operating system and other application programs, and when the embodiments of the present specification are implemented in software or firmware, the associated program code is stored in memory 1020 and executed by processor 1010.
The input/output interface 1030 is used to connect with an input/output module for inputting and outputting information. The input/output module may be configured as a component in a device (not shown) or may be external to the device to provide corresponding functionality. Wherein the input devices may include a keyboard, mouse, touch screen, microphone, various types of sensors, etc., and the output devices may include a display, speaker, vibrator, indicator lights, etc.
Communication interface 1040 is used to connect communication modules (not shown) to enable communication interactions of the present device with other devices. The communication module may implement communication through a wired manner (such as USB, network cable, etc.), or may implement communication through a wireless manner (such as mobile network, WIFI, bluetooth, etc.).
Bus 1050 includes a path for transferring information between components of the device (e.g., processor 1010, memory 1020, input/output interface 1030, and communication interface 1040).
It should be noted that although the above-described device only shows processor 1010, memory 1020, input/output interface 1030, communication interface 1040, and bus 1050, in an implementation, the device may include other components necessary to achieve proper operation. Furthermore, it will be understood by those skilled in the art that the above-described apparatus may include only the components necessary to implement the embodiments of the present description, and not all the components shown in the drawings.
The electronic device of the foregoing embodiment is configured to implement the corresponding extraction method in any of the foregoing embodiments, and has the beneficial effects of the corresponding method embodiment, which is not described herein.
Based on the same inventive concept, corresponding to any of the above embodiments of the method, the present disclosure further provides a non-transitory computer readable storage medium storing computer instructions for causing the computer to perform the extraction method as described in any of the above embodiments.
The computer readable media of the present embodiments, including both permanent and non-permanent, removable and non-removable media, may be used to implement information storage by any method or technology. The information may be computer readable instructions, data structures, modules of a program, or other data. Examples of storage media for a computer include, but are not limited to, phase change memory (PRAM), static Random Access Memory (SRAM), dynamic Random Access Memory (DRAM), other types of Random Access Memory (RAM), read Only Memory (ROM), electrically Erasable Programmable Read Only Memory (EEPROM), flash memory or other memory technology, compact disc read only memory (CD-ROM), digital Versatile Discs (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices, or any other non-transmission medium, which can be used to store information that can be accessed by a computing device.
The storage medium of the foregoing embodiments stores computer instructions for causing the computer to perform the extraction method of any of the foregoing embodiments, and has the advantages of the corresponding method embodiments, which are not described herein.
Based on the same inventive concept, the present disclosure also provides a computer program product, corresponding to the extraction method described in any of the above embodiments, comprising computer program instructions. In some embodiments, the computer program instructions may be executed by one or more processors of a computer to cause the computer and/or the processor to perform the extraction method. Corresponding to the execution subject corresponding to each step in each embodiment of the extraction method, the processor executing the corresponding step may belong to the corresponding execution subject.
The computer program product of the above embodiment is configured to enable the computer and/or the processor to perform the method of any one of the above embodiments, and has the advantages of the corresponding method embodiments, which are not described herein again.
Those of ordinary skill in the art will appreciate that: the discussion of any of the embodiments above is merely exemplary and is not intended to suggest that the scope of the disclosure, including the claims, is limited to these examples; the technical features of the above embodiments or in the different embodiments may also be combined under the idea of the present disclosure, the steps may be implemented in any order, and there are many other variations of the different aspects of the embodiments of the present disclosure as described above, which are not provided in details for the sake of brevity.
Additionally, well-known power/ground connections to Integrated Circuit (IC) chips and other components may or may not be shown within the provided figures, in order to simplify the illustration and discussion, and so as not to obscure the embodiments of the present disclosure. Furthermore, the devices may be shown in block diagram form in order to avoid obscuring the embodiments of the present disclosure, and this also accounts for the fact that specifics with respect to implementation of such block diagram devices are highly dependent upon the platform on which the embodiments of the present disclosure are to be implemented (i.e., such specifics should be well within purview of one skilled in the art). Where specific details (e.g., circuits) are set forth in order to describe example embodiments of the disclosure, it should be apparent to one skilled in the art that embodiments of the disclosure can be practiced without, or with variation of, these specific details. Accordingly, the description is to be regarded as illustrative in nature and not as restrictive.
While the present disclosure has been described in conjunction with specific embodiments thereof, many alternatives, modifications, and variations of those embodiments will be apparent to those skilled in the art in light of the foregoing description. For example, other memory architectures (e.g., dynamic RAM (DRAM)) may use the embodiments discussed.
The disclosed embodiments are intended to embrace all such alternatives, modifications and variances which fall within the broad scope of the appended claims. Accordingly, any omissions, modifications, equivalents, improvements, and the like, which are within the spirit and principles of the embodiments of the disclosure, are intended to be included within the scope of the disclosure.
Claims (10)
1. A method for extracting event arguments, the method comprising:
acquiring an original text; preprocessing the original text to obtain an event text;
processing the event text by using a preset language model to obtain abstract information, associated information and named entity information;
Splitting to obtain statement information according to the event text;
determining an event argument corresponding to the original text by using an event argument extraction model according to the abstract information, the associated information, the named entity information and the sentence information; wherein the event argument extraction model is a trained neural network model.
2. The extraction method according to claim 1, wherein preprocessing the original text to obtain an event text includes:
preprocessing based on a preset regular expression; wherein,
The preset regular expressions comprise regular expressions of irrelevant events and regular expressions of website labels; wherein the regular expression of the uncorrelated event is used for filtering the original text; the regular expression of the website label is used for deleting the website label.
3. The extraction method according to claim 1, wherein the step of processing the event text by using a preset language model to obtain abstract information, associated information and named entity information includes:
acquiring and utilizing a preset language model to obtain abstract information according to a preset abstract information prompt;
Acquiring and utilizing a preset language model to obtain the associated information according to a preset associated information prompt;
And acquiring and obtaining the named entity information by using a preset language model according to a preset named entity information prompt.
4. The extraction method according to claim 1, wherein the event argument extraction model comprises a feature extraction module, a feature fusion module, and an argument boundary discrimination module; the feature fusion module is constructed based on an attention mechanism; the argument boundary discriminating module is a discriminant module.
5. The extraction method according to claim 4, wherein the feature extraction module is configured to obtain abstract semantic features, associated semantic features, entity semantic features, and sentence semantic features according to the abstract information, the associated information, the named entity information, and the sentence information.
6. The extraction method according to claim 5, wherein the feature fusion module comprises a first feature fusion module and a second feature fusion module; wherein,
The first feature fusion module is used for fusing the sentence semantic features and the abstract semantic features to obtain abstract fusion features, and fusing the sentence semantic features and the associated semantic features to obtain associated fusion features;
The second feature fusion module is used for: based on an attention mechanism, fusing the entity semantic features, the sentence semantic features, the abstract fusion features and the associated fusion features to obtain a second fusion result; and normalizing the second fusion result and the summary fusion characteristic based on a residual network idea to obtain fusion data.
7. The extraction method according to claim 6, wherein the second feature fusion module includes a third feature fusion module and a fourth feature fusion module; wherein,
The third feature fusion module is used for fusing the entity semantic features, the sentence semantic features and the abstract fusion features to obtain a first fusion result;
The fourth feature fusion module is used for obtaining a second fusion result through fusion according to the entity semantic features, the first fusion result and the association fusion features.
8. The extraction method according to claim 4, wherein the feature fusion module is configured to obtain fusion data; the argument boundary judging module is used for determining the starting position probability and the ending position probability of each position of the statement information as an argument boundary according to the fusion data, and determining the starting position and the ending position of the event argument based on the starting position probability and the ending position probability.
9. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable by the processor, characterized in that the processor implements the extraction method according to any one of claims 1 to 8 when the computer program is executed.
10. A non-transitory computer readable storage medium storing computer instructions for causing a computer to perform the extraction method according to any one of claims 1 to 8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410285755.4A CN117909505A (en) | 2024-03-13 | 2024-03-13 | Event argument extraction method and related equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410285755.4A CN117909505A (en) | 2024-03-13 | 2024-03-13 | Event argument extraction method and related equipment |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117909505A true CN117909505A (en) | 2024-04-19 |
Family
ID=90692446
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202410285755.4A Pending CN117909505A (en) | 2024-03-13 | 2024-03-13 | Event argument extraction method and related equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117909505A (en) |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113486667A (en) * | 2021-07-26 | 2021-10-08 | 辽宁工程技术大学 | Medical entity relationship joint extraction method based on entity type information |
| CN113591483A (en) * | 2021-04-27 | 2021-11-02 | 重庆邮电大学 | Document-level event argument extraction method based on sequence labeling |
| CN114297394A (en) * | 2022-03-11 | 2022-04-08 | 中国科学院自动化研究所 | Method and electronic equipment for extracting event arguments in text |
| CN114637827A (en) * | 2021-12-04 | 2022-06-17 | 电子科技大学长三角研究院(湖州) | Carbon transaction text event extraction method based on graph neural network |
| US20230127652A1 (en) * | 2021-10-25 | 2023-04-27 | Adobe Inc. | Event understanding with deep learning |
| US20230259808A1 (en) * | 2022-02-16 | 2023-08-17 | International Business Machines Corporation | Active learning for event extraction from heterogeneous data sources |
| CN117056501A (en) * | 2023-07-03 | 2023-11-14 | 北京邮电大学 | Method and device for extracting argument, electronic equipment and storage medium |
-
2024
- 2024-03-13 CN CN202410285755.4A patent/CN117909505A/en active Pending
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113591483A (en) * | 2021-04-27 | 2021-11-02 | 重庆邮电大学 | Document-level event argument extraction method based on sequence labeling |
| CN113486667A (en) * | 2021-07-26 | 2021-10-08 | 辽宁工程技术大学 | Medical entity relationship joint extraction method based on entity type information |
| US20230127652A1 (en) * | 2021-10-25 | 2023-04-27 | Adobe Inc. | Event understanding with deep learning |
| CN114637827A (en) * | 2021-12-04 | 2022-06-17 | 电子科技大学长三角研究院(湖州) | Carbon transaction text event extraction method based on graph neural network |
| US20230259808A1 (en) * | 2022-02-16 | 2023-08-17 | International Business Machines Corporation | Active learning for event extraction from heterogeneous data sources |
| CN114297394A (en) * | 2022-03-11 | 2022-04-08 | 中国科学院自动化研究所 | Method and electronic equipment for extracting event arguments in text |
| CN117056501A (en) * | 2023-07-03 | 2023-11-14 | 北京邮电大学 | Method and device for extracting argument, electronic equipment and storage medium |
Non-Patent Citations (2)
| Title |
|---|
| JI ZHOU 等: "EACE: A document-level event argument extraction model with argument constraint enhancement", INFORMATION PROCESSING AND MANAGEMENT, 31 October 2023 (2023-10-31) * |
| 朱少华;李培峰;朱巧明;: "基于MLN的中文事件论元推理方法", 计算机科学, no. 03, 15 March 2016 (2016-03-15) * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US11948058B2 (en) | Utilizing recurrent neural networks to recognize and extract open intent from text inputs | |
| JP2019008778A (en) | Captioning region of image | |
| CN113011186B (en) | Named entity recognition method, named entity recognition device, named entity recognition equipment and computer readable storage medium | |
| CN114298417A (en) | Anti-fraud risk assessment method, anti-fraud risk training method, anti-fraud risk assessment device, anti-fraud risk training device and readable storage medium | |
| CN112632226B (en) | Semantic search method and device based on legal knowledge graph and electronic equipment | |
| US10915756B2 (en) | Method and apparatus for determining (raw) video materials for news | |
| CN112100401B (en) | Knowledge graph construction method, device, equipment and storage medium for science and technology services | |
| CN116304748B (en) | Text similarity calculation method, system, equipment and medium | |
| CN112183030A (en) | Event extraction method and device based on preset neural network, computer equipment and storage medium | |
| CN110750297A (en) | Python code reference information generation method based on program analysis and text analysis | |
| Petukhova et al. | TextCL: A Python package for NLP preprocessing tasks | |
| CN112906368A (en) | Industry text increment method, related device and computer program product | |
| CN109902162B (en) | Text similarity identification method based on digital fingerprints, storage medium and device | |
| CN116974554A (en) | Code data processing method, apparatus, computer device and storage medium | |
| CN112100364A (en) | Text semantic understanding method and model training method, device, equipment and medium | |
| US20180260476A1 (en) | Expert stance classification using computerized text analytics | |
| US11423228B2 (en) | Weakly supervised semantic entity recognition using general and target domain knowledge | |
| CN117909505A (en) | Event argument extraction method and related equipment | |
| CN111199170B (en) | Formula file identification method and device, electronic equipment and storage medium | |
| CN113807920A (en) | Artificial intelligence based product recommendation method, device, equipment and storage medium | |
| CN113536773A (en) | Commodity comment sentiment analysis method and system, electronic equipment and storage medium | |
| CN111967253A (en) | Entity disambiguation method and device, computer equipment and storage medium | |
| CN117874261B (en) | Question-answer type event extraction method based on course learning and related equipment | |
| WO2019246252A1 (en) | Systems and methods for identifying and linking events in structured proceedings | |
| KR102621869B1 (en) | Device and method for constructing dataset in korean document-level relation extraction |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |