CN116450783A - Method, system, storage medium and electronic equipment for extracting event facing chapter level - Google Patents
Method, system, storage medium and electronic equipment for extracting event facing chapter level Download PDFInfo
- Publication number
- CN116450783A CN116450783A CN202211479409.7A CN202211479409A CN116450783A CN 116450783 A CN116450783 A CN 116450783A CN 202211479409 A CN202211479409 A CN 202211479409A CN 116450783 A CN116450783 A CN 116450783A
- Authority
- CN
- China
- Prior art keywords
- event
- document
- model
- chapter
- training
- 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
- 238000000034 method Methods 0.000 title claims description 34
- 238000000605 extraction Methods 0.000 claims abstract description 64
- 238000003062 neural network model Methods 0.000 claims abstract description 23
- 238000010801 machine learning Methods 0.000 claims abstract description 19
- 238000012549 training Methods 0.000 claims description 47
- 230000008569 process Effects 0.000 claims description 18
- 238000009826 distribution Methods 0.000 claims description 7
- 238000004590 computer program Methods 0.000 claims description 6
- 238000005070 sampling Methods 0.000 claims description 6
- 230000009471 action Effects 0.000 claims description 5
- 230000006870 function Effects 0.000 claims description 5
- 230000002457 bidirectional effect Effects 0.000 claims description 3
- 238000006243 chemical reaction Methods 0.000 claims description 3
- 238000012790 confirmation Methods 0.000 claims description 3
- 230000003993 interaction Effects 0.000 claims description 3
- 230000009191 jumping Effects 0.000 claims description 3
- 238000010845 search algorithm Methods 0.000 claims description 3
- 238000012795 verification Methods 0.000 claims description 3
- 238000013507 mapping Methods 0.000 claims description 2
- 238000012545 processing Methods 0.000 claims description 2
- 238000004458 analytical method Methods 0.000 abstract description 5
- 238000003058 natural language processing Methods 0.000 abstract description 2
- 230000009466 transformation Effects 0.000 description 5
- 230000009286 beneficial effect Effects 0.000 description 4
- 238000013528 artificial neural network Methods 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 239000013598 vector Substances 0.000 description 3
- 230000007246 mechanism Effects 0.000 description 2
- 238000013459 approach Methods 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000001131 transforming effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/33—Querying
- G06F16/332—Query formulation
- G06F16/3329—Natural language query formulation or dialogue systems
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/35—Clustering; Classification
- G06F16/353—Clustering; Classification into predefined classes
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/20—Natural language analysis
- G06F40/279—Recognition of textual entities
- G06F40/289—Phrasal analysis, e.g. finite state techniques or chunking
- G06F40/295—Named entity recognition
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
Abstract
The invention provides a chapter-level-oriented event extraction method, a chapter-level-oriented event extraction system, a storage medium and electronic equipment, and relates to the technical field of natural language processing. Acquiring a document to be analyzed; acquiring event types based on a machine learning understanding model according to a document to be analyzed; acquiring a corresponding relation pair of the event parameter type-entity candidate object by adopting a learning search model or a deep neural network model according to the document marked with the event type; and filling a macroscopic event template frame according to the event type, the parameter type and the entity candidate object, and returning to the user as an event extraction result. By constraining the defined parameter fill set for each event type to a predefined, but highly interdependent set of categories containing key event information, events are presented to the user in a relatively compact but informative manner and further processed by downstream analysis methods. Can be generalized in a wide range of event types, and event structures have wide applicability.
Description
Technical Field
The invention relates to the technical field of natural language processing, in particular to a chapter-level-oriented event extraction method, a chapter-level-oriented event extraction system, a storage medium and electronic equipment.
Background
The goal of event extraction is to identify in the document an instance of a class of events and any parameters corresponding to roles in the event. The existing event extraction framework is divided into sentence-level event extraction and chapter-level event extraction. Sentence-level event extraction refers to identifying and extracting a single event from each sentence in a document, and any entity that plays a role in the demonstration of those events, but cannot summarize the content of the document. The chapter-level event extraction task is a challenging information extraction effort that requires inferring from the entire article, has obvious applicability in the real world, and allows users to quickly identify the person, content, location, and time of a document event without reading the entire document.
To date, event extraction methods have not provided a satisfactory solution, and most current event extraction systems rely on phrase or sentence level local information and do not consider articles as a whole. Since in most cases one event requires multiple sentences to be fully described, this approach cannot capture the relationships between the events mentioned at the document level, and event parameters in different sentences appear compared to the event triggers, limiting extraction performance. Meanwhile, the work of extracting the events of the documents is mainly concentrated in a highly specific field, generally depends on rules formulated by hands to a great extent, and cannot be well popularized to a new field.
Disclosure of Invention
(one) solving the technical problems
Aiming at the defects of the prior art, the invention provides a chapter-level-oriented event extraction method, a chapter-level-oriented event extraction system, a chapter-level-oriented event extraction storage medium and electronic equipment, and solves the technical problem that the relation among events mentioned at a document level cannot be captured.
(II) technical scheme
In order to achieve the above purpose, the invention is realized by the following technical scheme:
a chapter-level oriented event extraction method for pre-training a machine learning understanding model, a learning search model, or a deep neural network model, the extraction method comprising:
s1, acquiring a document to be analyzed;
s2, acquiring event types based on the machine learning understanding model according to the document to be analyzed;
s3, acquiring a corresponding relation pair of event parameter type-entity candidate object by adopting the learning search model or the deep neural network model according to the document marked with the event type;
and S4, filling a macroscopic event template frame according to the event type, the parameter type and the entity candidate object, and returning to the user as an event extraction result.
Preferably, the machine learning understanding model is a gated attention GA reader, and the training process includes:
obtaining event parameter type queries and training sets via a lookup tableWord embedding of document words is processed in a later K-layer network, and the K-layer network can acquire the previous document embedding from the K-1 layer network as input; document word embedding d for each layer i And parameter type query word embedding Q are respectively obtained by conversion through a bidirectional gating circulating unit GRU, and then are combined through a gating attention module to obtain input x of the next layer i The method comprises the steps of carrying out a first treatment on the surface of the Wherein each GA box represents a gating-attention module that applies attention of the attention parameter type query to the document representation;
after repeating this process beyond the K layers, scores are computed for the words in the document and converted to probability distributions over the words using the Softmax () function; the resulting probability distribution is used for the selection of the answers to the query.
Preferably, training the learning search model using a learning search algorithm includes:
s10, randomly extracting a document D from the marked document D and generating a sequence of entity and parameter type candidate relation pairs;
s20, executing the current learned strategy pi on the document d in the step length t of random sampling i ;
S30, executing the strategy pi in the step length t by the candidate relation pair corresponding to the document d i Thereafter, all possible policies of the remaining candidate relationship pairs are pi-compared to the best policy * Generating a new training example according to the strategy corresponding to the minimum loss of the training set, and adding the new training sample into the training set;
s40, after k new examples are generated, retraining a new strategy by using a new training setThen jumping back to the step 1;
s50, obtaining a final training strategy after circulating for N times
Wherein N is the number of custom training, m is the number of custom sampling, and the policy pi is to fill a specific parameter slot with a specific candidate entityAction map of pi * Is an optimal strategy for manual confirmation, beta i I=1,..n is a custom policy update weight.
Preferably, the deep neural network model comprises an input layer, two hidden layers and an output layer;
the input layer consists of text data features required by the model, contains information of a current candidate entity relation pair, and the output layer is a probability value of whether the current candidate entity relation pair is contained in a final template or not; more complex interaction information is learned from input features using a multi-layered network structure.
Preferably, the training process of the deep neural network model includes:
s100, using the current parameter W for each training sample x i Predicting, and comparing the predicted result with a real sample label y i Comparing and calculating loss;
and S200, updating the model parameter W through a back propagation process. After the parameter updating is completed, training the next sample;
s300, after execution of all training samples is completed, evaluating a current model by using a verification sample set (V, Z), and calculating an F1 score; if the current model F1 score is due to the historical highest score F1 best Then these parameters are stored as new optimal model parameters W best And update F1 best The current F1 fraction;
s400, after the end of k periods, returning to provide the model parameters W with optimal histories best 。
A chapter-level oriented event extraction system for pre-training a machine learning understanding model, a learning search model, or a deep neural network model, the extraction system comprising:
the acquisition module is used for acquiring the document to be analyzed;
the learning module is used for acquiring event types based on the machine learning understanding model according to the document to be analyzed;
the searching module is used for acquiring a corresponding relation pair of the event parameter type-entity candidate object by adopting the learning searching model or the deep neural network model according to the document marked with the event type;
and the filling module is used for filling the macroscopic event template frame according to the event type, the parameter type and the entity candidate object, and returning the macroscopic event template frame to the user as an event extraction result.
A storage medium storing a computer program for chapter-level oriented event extraction, wherein the computer program causes a computer to execute the chapter-level oriented event extraction method as described above.
An electronic device, comprising:
one or more processors;
a memory; and
one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the programs comprising instructions for performing a chapter-level oriented event extraction method as described above.
(III) beneficial effects
The invention provides a chapter-level-oriented event extraction method, a chapter-level-oriented event extraction system, a storage medium and electronic equipment. Compared with the prior art, the method has the following beneficial effects:
acquiring a document to be analyzed; acquiring event types based on the machine learning understanding model according to the document to be analyzed; acquiring a corresponding relation pair of event parameter type-entity candidate object by adopting the learning search model or the deep neural network model according to the document marked with the event type; and filling a macroscopic event template frame according to the event type, the parameter type and the entity candidate object, and returning to the user as an event extraction result. By constraining the defined parameter fill set for each event type to a predefined, but highly interdependent set of categories containing key event information, events are presented to the user in a relatively compact but informative manner and further processed by downstream analysis methods. Can be generalized in a wide range of event types, and event structures have wide applicability.
Drawings
In order to more clearly illustrate the embodiments of the invention or the technical solutions in the prior art, the drawings that are required in the embodiments or the description of the prior art will be briefly described, it being obvious that the drawings in the following description are only some embodiments of the invention, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
Fig. 1 is a block diagram of a chapter-level-oriented event extraction method according to an embodiment of the present invention;
FIG. 2 is a schematic diagram of a GA reader architecture according to an embodiment of the present invention;
fig. 3 is a schematic diagram of a deep neural network architecture for event extraction according to an embodiment of the present invention.
Detailed Description
For the purpose of making the objects, technical solutions and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention are clearly and completely described, and it is obvious that the described embodiments are some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
The embodiment of the application solves the technical problem that the relation between the events mentioned at the document level cannot be captured by providing the chapter-level-oriented event extraction method, the chapter-level-oriented event extraction system, the storage medium and the electronic equipment.
The technical scheme in the embodiment of the application aims to solve the technical problems, and the overall thought is as follows:
the method comprises the steps of obtaining a document to be analyzed; acquiring event types based on the machine learning understanding model according to the document to be analyzed; acquiring a corresponding relation pair of event parameter type-entity candidate object by adopting the learning search model or the deep neural network model according to the document marked with the event type; and filling a macroscopic event template frame according to the event type, the parameter type and the entity candidate object, and returning to the user as an event extraction result. By constraining the defined parameter fill set for each event type to a predefined, but highly interdependent set of categories containing key event information, events are presented to the user in a relatively compact but informative manner and further processed by downstream analysis methods. Can be generalized in a wide range of event types, and event structures have wide applicability.
In order to better understand the above technical solutions, the following detailed description will refer to the accompanying drawings and specific embodiments.
Examples:
as shown in fig. 1, an embodiment of the present invention provides a chapter-level-oriented event extraction method, in which a machine learning understanding model, a learning search model, or a deep neural network model is trained in advance, the extraction method includes:
s1, acquiring a document to be analyzed;
s2, acquiring event types based on the machine learning understanding model according to the document to be analyzed;
s3, acquiring a corresponding relation pair of event parameter type-entity candidate object by adopting the learning search model or the deep neural network model according to the document marked with the event type;
and S4, filling a macroscopic event template frame according to the event type, the parameter type and the entity candidate object, and returning to the user as an event extraction result.
The embodiment of the invention restricts the defined parameter filling set of each event type into a predefined but highly interdependent class set containing key event information, so that the event is represented to a user in a relatively simple but information-rich way and is further processed by a downstream analysis method. Can be generalized in a wide range of event types, and event structures have wide applicability.
The steps of the above scheme will be described in detail as follows:
in step S1, a document to be analyzed is acquired.
Any technical means of obtaining text in the prior art, such as a crawler technology, may be adopted, and details thereof are not described herein.
In step S2, according to the document to be analyzed, an event type is acquired based on the machine learning understanding model.
The method for training the general question-answering system by utilizing the existing large-scale machine reading and understanding corpus is used for extracting the events based on machine reading and understanding, and the trained model is directly applied to the macro event template filling task. The core of the process is a Gated Attention (GA) reader. As shown in FIG. 2, the GA reader is a deep learning model, which adopts a multi-hop structure in combination with an attention mechanism. The multi-hop structure simulates a person's understanding of the current text every time it is refined during several readings of the text. The role of the attention mechanism is to allow the reader to focus on a given question when reading text, and to make the model more focused on information related to the event when processing the text.
The training process of the GA reader comprises the following steps: the event parameter type query and the word embedding of the document words are acquired through the lookup table and are processed in the K-layer network, and the K-layer network can acquire the previous document embedding from the K-1 layer network as input. Using a multi-layer network structure, more complex embedded representations of input document words can be built for a model, each layer focusing attention on different aspects of a parameter type query. Document word embedding d for each layer i And parameter type query word embedding Q are respectively obtained by conversion through a bidirectional gating circulating unit (GRU), and then are combined through a gating attention module to obtain input x of the next layer i . Each "GA" box represents a gating-attention module that applies the attention of the attention parameter type query to the document representation. After repeating this process beyond the K layers, scores are computed for the words in the document and converted to probability distributions over the words using the Softmax () function. The resulting probability distribution may be used for selection of answers to the query.
Applying a gating-attention module to word embedding d in a document i The way of (2) is as follows:
α i =softmax(Q T d i )
where Q is the parameter type query embedded representation, +..
At the last layer, the document and query representation calculate a score for each word using the inner product, and then the probability distribution of the words in the document is obtained through the softmax layer. When a set of words occurs multiple times in a document, the probabilities of the words are summarized and re-normalized, and the final result is obtained by selecting the candidate word with the highest probability:
where C is the candidate result set, s is the softmax () probability vector for candidate word C,an index of the document d corresponding to the candidate word c is specified.
In step S3, according to the document after the event type is marked, the learning search model or the deep neural network model is adopted to obtain the corresponding relationship pair of the event parameter type-entity candidate object;
and training the labeled text data on the existing data set by using a learning search-based method or a deep neural network-based method, constructing a model, and completing filling of the macro event template.
First, macro event template parameter filling based on learning search:
for each document, the embodiment of the invention firstly collects all entity candidates by using a named entity recognition method, and pairs each entity with all possible parameter types in a macro event template structure. For the formed parameter type-entity relationship pairs, a decision needs to be made as to whether or not to be included in the final macro event template. For each decision, not only the local features of the current relationship pair, but also all decisions that the model made before need to be considered, so that when the template is filled, past decisions can be taken into context
Training the learning search model using a learning search algorithm, comprising:
s10, randomly extracting a document D from the marked document D and generating a sequence of entity and parameter type candidate relation pairs;
s20, executing the current learned strategy pi on the document d in the step length t of random sampling i ;
S30, executing the strategy pi in the step length t by the candidate relation pair corresponding to the document d i Thereafter, all possible policies of the remaining candidate relationship pairs are pi-compared to the best policy * Generating a new training example according to the strategy corresponding to the minimum loss of the training set, and adding the new training sample into the training set;
s40, after k new examples are generated, retraining a new strategy by using a new training setThen jumping back to the step 1;
s50, obtaining a final training strategy after circulating for N times
Wherein N is the number of custom training, m is the number of custom sampling, the policy pi is the action mapping of filling a specific parameter slot with a specific candidate entity, pi * Is an optimal strategy for manual confirmation, beta i I=1,..n is a custom policy update weight.
Second, deep neural network of event extraction:
to extract the main event in the text, for a given input text and set of entity candidate objects, embodiments of the present invention populate the macro event template by binary predicting whether each (entity candidate object, parameter type) relationship pair is included or excluded in the final template.
The model architecture is shown in fig. 3, and the deep neural network model comprises an input layer, two hidden layers and an output layer; the input layer consists of text data features required by the model, contains information of a current candidate entity relation pair, and the output layer is a probability value of whether the current candidate entity relation pair is contained in a final template or not; more complex interaction information is learned from input features using a multi-layered network structure.
The model first converts the input layer text data features to low-dimensional hidden vectors using a linear transformation W, after which the low-dimensional hidden vectors are processed by a nonlinear transformation, allowing the network to model more complex decisions. Common nonlinear transformation functions include sigmoid (), tanh (), and ReLU (), the transformation process is as follows:
ReLU(x)=max(0,x)
the nonlinear transformation process may be repeated multiple times. At the last layer, the data is transformed by a softmax () function, transforming the output decision score into a set of probabilities:
in particular, the training process of the deep neural network model comprises the following steps:
s100, using the current parameter W for each training sample x i Making predictions, andthe predicted result is compared with a real sample label y i Comparing and calculating loss;
and S200, updating the model parameter W through a back propagation process. After the parameter updating is completed, training the next sample;
s300, after execution of all training samples is completed, evaluating a current model by using a verification sample set (V, Z), and calculating an F1 score; if the current model F1 score is due to the historical highest score F1 best Then these parameters are stored as new optimal model parameters W best And update F1 best The current F1 fraction;
s400, after the end of k periods, returning to provide the model parameters W with optimal histories best 。
In step S4, a macroscopic event template frame is filled according to the event type, the parameter type and the entity candidate object, and the macroscopic event template frame is returned to the user as an event extraction result.
Aiming at the chapter-level text event extraction task, the embodiment of the invention provides a method for filling a template based on a macroscopic event template frame, and a true and useful chapter-level event extraction model is created. The event template form is shown in table 1 and consists of event type, event parameter type and event parameter filling entity, each parameter field can be filled with zero, one or more texts.
TABLE 1
The embodiment of the invention provides a chapter-level-oriented event extraction system, which is used for pre-training a machine learning understanding model, a learning search model or a deep neural network model, and comprises the following steps:
the acquisition module is used for acquiring the document to be analyzed;
the learning module is used for acquiring event types based on the machine learning understanding model according to the document to be analyzed;
the searching module is used for acquiring a corresponding relation pair of the event parameter type-entity candidate object by adopting the learning searching model or the deep neural network model according to the document marked with the event type;
and the filling module is used for filling the macroscopic event template frame according to the event type, the parameter type and the entity candidate object, and returning the macroscopic event template frame to the user as an event extraction result.
The embodiment of the invention provides a storage medium which stores a computer program for chapter-level-oriented event extraction, wherein the computer program causes a computer to execute the chapter-level-oriented event extraction method as described above.
The embodiment of the invention provides electronic equipment, which comprises:
one or more processors;
a memory; and
one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the programs comprising instructions for performing a chapter-level oriented event extraction method as described above.
It may be understood that the chapter-level-oriented event extraction system, the storage medium, and the electronic device provided by the embodiments of the present invention correspond to the chapter-level-oriented event extraction method provided by the embodiments of the present invention, and the explanation, the examples, and the beneficial effects of the relevant content may refer to the corresponding parts in the chapter-level-oriented event extraction method, which are not described herein again.
In summary, compared with the prior art, the method has the following beneficial effects:
the method comprises the steps of obtaining a document to be analyzed; acquiring event types based on the machine learning understanding model according to the document to be analyzed; acquiring a corresponding relation pair of event parameter type-entity candidate object by adopting the learning search model or the deep neural network model according to the document marked with the event type; and filling a macroscopic event template frame according to the event type, the parameter type and the entity candidate object, and returning to the user as an event extraction result. By constraining the defined parameter fill set for each event type to a predefined, but highly interdependent set of categories containing key event information, events are presented to the user in a relatively compact but informative manner and further processed by downstream analysis methods. Can be generalized in a wide range of event types, and event structures have wide applicability.
It is noted that relational terms such as first and second, and the like are used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Moreover, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one … …" does not exclude the presence of other like elements in a process, method, article, or apparatus that comprises the element.
The above embodiments are only for illustrating the technical solution of the present invention, and are not limiting; although the invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some technical features thereof can be replaced by equivalents; such modifications and substitutions do not depart from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims (8)
1. A chapter-level oriented event extraction method, characterized by pre-training a machine learning understanding model, a learning search model, or a deep neural network model, the extraction method comprising:
s1, acquiring a document to be analyzed;
s2, acquiring event types based on the machine learning understanding model according to the document to be analyzed;
s3, acquiring a corresponding relation pair of event parameter type-entity candidate object by adopting the learning search model or the deep neural network model according to the document marked with the event type;
and S4, filling a macroscopic event template frame according to the event type, the parameter type and the entity candidate object, and returning to the user as an event extraction result.
2. The chapter-level oriented event extraction method of claim 1 wherein said machine learning understanding model is a gated attention GA reader, the training process comprising:
acquiring event parameter type query and word embedding of document words in a training set through a lookup table, processing in a K-layer network, wherein the K-layer network can acquire previous document embedding from the K-1-layer network as input; document word embedding d for each layer i And parameter type query word embedding Q are respectively obtained by conversion through a bidirectional gating circulating unit GRU, and then are combined through a gating attention module to obtain input x of the next layer i The method comprises the steps of carrying out a first treatment on the surface of the Wherein each GA box represents a gating-attention module that applies attention of the attention parameter type query to the document representation;
after repeating this process beyond the K layers, scores are computed for the words in the document and converted to probability distributions over the words using the Softmax () function; the resulting probability distribution is used for the selection of the answers to the query.
3. The chapter-level oriented event extraction method of claim 1 or 2, wherein training the learning search model using a learning search algorithm comprises:
s10, randomly extracting a document D from the marked document D and generating a sequence of entity and parameter type candidate relation pairs;
s20, executing the current learned strategy pi on the document d in the step length t of random sampling i ;
S30, selecting a candidate corresponding to the document dIs to execute strategy pi in step length t i Thereafter, all possible policies of the remaining candidate relationship pairs are pi-compared to the best policy * Generating a new training example according to the strategy corresponding to the minimum loss of the training set, and adding the new training sample into the training set;
s40, after k new examples are generated, retraining a new strategy by using a new training setThen jumping back to the step 1;
s50, obtaining a final training strategy after circulating for N times
Wherein N is the number of custom training, m is the number of custom sampling, the policy pi is the action mapping of filling a specific parameter slot with a specific candidate entity, pi * Is an optimal strategy for manual confirmation, beta i I=1,..n is a custom policy update weight.
4. The chapter-level oriented event extraction method of claim 1 or 2, wherein the deep neural network model comprises an input layer, two hidden layers, and an output layer;
the input layer consists of text data features required by the model, contains information of a current candidate entity relation pair, and the output layer is a probability value of whether the current candidate entity relation pair is contained in a final template or not; more complex interaction information is learned from input features using a multi-layered network structure.
5. The chapter-level oriented event extraction method of claim 4 wherein said training process of said deep neural network model comprises:
s100, using the current parameter W for each training sample x i Predicting, and comparing the predicted result with a real sample label y i Comparing and calculating loss;
and S200, updating the model parameter W through a back propagation process. After the parameter updating is completed, training the next sample;
s300, after execution of all training samples is completed, evaluating a current model by using a verification sample set (V, Z), and calculating an F1 score; if the current model F1 score is due to the historical highest score F1 best Then these parameters are stored as new optimal model parameters W best And update F1 best The current F1 fraction;
s400, after the end of k periods, returning to provide the model parameters W with optimal histories best 。
6. A chapter-level oriented event extraction system that pre-trains a machine learning understanding model, a learning search model, or a deep neural network model, the extraction system comprising:
the acquisition module is used for acquiring the document to be analyzed;
the learning module is used for acquiring event types based on the machine learning understanding model according to the document to be analyzed;
the searching module is used for acquiring a corresponding relation pair of the event parameter type-entity candidate object by adopting the learning searching model or the deep neural network model according to the document marked with the event type;
and the filling module is used for filling the macroscopic event template frame according to the event type, the parameter type and the entity candidate object, and returning the macroscopic event template frame to the user as an event extraction result.
7. A storage medium storing a computer program for chapter-level-oriented event extraction, wherein the computer program causes a computer to execute the chapter-level-oriented event extraction method according to any one of claims 1 to 5.
8. An electronic device, comprising:
one or more processors;
a memory; and
one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the programs comprising instructions for performing the chapter-level oriented event extraction method of any one of claims 1-5.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202211479409.7A CN116450783A (en) | 2022-11-24 | 2022-11-24 | Method, system, storage medium and electronic equipment for extracting event facing chapter level |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202211479409.7A CN116450783A (en) | 2022-11-24 | 2022-11-24 | Method, system, storage medium and electronic equipment for extracting event facing chapter level |
Publications (1)
Publication Number | Publication Date |
---|---|
CN116450783A true CN116450783A (en) | 2023-07-18 |
Family
ID=87134381
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202211479409.7A Pending CN116450783A (en) | 2022-11-24 | 2022-11-24 | Method, system, storage medium and electronic equipment for extracting event facing chapter level |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN116450783A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116881851A (en) * | 2023-09-04 | 2023-10-13 | 成都无声讯通科技有限责任公司 | Internet of things data processing method and device based on machine learning and server |
-
2022
- 2022-11-24 CN CN202211479409.7A patent/CN116450783A/en active Pending
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116881851A (en) * | 2023-09-04 | 2023-10-13 | 成都无声讯通科技有限责任公司 | Internet of things data processing method and device based on machine learning and server |
CN116881851B (en) * | 2023-09-04 | 2023-12-19 | 成都无声讯通科技有限责任公司 | Internet of things data processing method and device based on machine learning and server |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN111078836B (en) | Machine reading understanding method, system and device based on external knowledge enhancement | |
CN110188358B (en) | Training method and device for natural language processing model | |
CN111695779B (en) | Knowledge tracking method, knowledge tracking device and storage medium | |
CN110188272B (en) | Community question-answering website label recommendation method based on user background | |
CN111191002B (en) | Neural code searching method and device based on hierarchical embedding | |
CN107562792A (en) | A kind of question and answer matching process based on deep learning | |
CN110096711B (en) | Natural language semantic matching method for sequence global attention and local dynamic attention | |
CN111738007B (en) | Chinese named entity identification data enhancement algorithm based on sequence generation countermeasure network | |
CN108845990A (en) | Answer selection method, device and electronic equipment based on two-way attention mechanism | |
CN113435203B (en) | Multi-modal named entity recognition method and device and electronic equipment | |
CN106202010A (en) | The method and apparatus building Law Text syntax tree based on deep neural network | |
CN111241807B (en) | Machine reading understanding method based on knowledge-guided attention | |
CN111985239A (en) | Entity identification method and device, electronic equipment and storage medium | |
CN110457714B (en) | Natural language generation method based on time sequence topic model | |
CN112883714B (en) | ABSC task syntactic constraint method based on dependency graph convolution and transfer learning | |
CN111666376B (en) | Answer generation method and device based on paragraph boundary scan prediction and word shift distance cluster matching | |
CN110019736A (en) | Question and answer matching process, system, equipment and storage medium based on language model | |
CN111274790A (en) | Chapter-level event embedding method and device based on syntactic dependency graph | |
CN109002519A (en) | Answer selection method, device and electronic equipment based on convolution loop neural network | |
CN116594768A (en) | Large-model-oriented universal tool collaboration and refinement learning system and method | |
CN116385937A (en) | Method and system for solving video question and answer based on multi-granularity cross-mode interaction framework | |
CN117648429B (en) | Question-answering method and system based on multi-mode self-adaptive search type enhanced large model | |
CN116450783A (en) | Method, system, storage medium and electronic equipment for extracting event facing chapter level | |
CN117377950A (en) | Accelerating document classification using machine learning | |
Shin et al. | End-to-end task dependent recurrent entity network for goal-oriented dialog learning |
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 |