A method for extracting conflict and dispute data information based on UIE model

Abstract

The present invention discloses a method for extracting contradiction and dispute data information based on a UIE model, S1, forming a complete contradiction and dispute data set; S2, generating a general word embedding model; S3, constructing a structured information extraction language, and uniformly converting the structured representations of various extraction tasks into a standardized coding format; S4, converting unstructured text information into structured text information through a structured information extraction language; S5, constructing a structured pattern director, and outputting structured information; S6, extracting triple information containing entities, relationships, and attributes, and uniformly representing the extracted information through a structured information extraction language; S7, constructing a contradiction and dispute thematic knowledge graph using triple information, a contradiction and dispute data set, and a word embedding model. The present invention significantly improves the efficiency and accuracy of contradiction and dispute data information extraction, can adapt to diverse task requirements, and has strong model adaptability and extensibility.

Description

Technical Field

[0001] The present invention relates to the technical field of contradictions and disputes, and in particular to a method for extracting contradiction and dispute data information based on a UIE model. Background Art

[0002] Existing technologies mainly rely on preset rules and manually annotated supervised learning models for information extraction. They perform well when processing structured or semi-structured data, but are unable to cope with large amounts of unstructured text data in the field of conflicts and disputes. Unstructured texts usually contain complex semantic structures and a large number of domain-specific terms, making it difficult for traditional rule-driven methods to fully identify and extract useful entities, relationships, and events. In addition, they lack sufficient flexibility when dealing with ever-changing case types and legal environments, and often require a lot of time and manpower to redesign and adjust the rules.

[0003] When dealing with large-scale conflict and dispute data, the reliance of existing supervised learning models on large amounts of labeled data has become a significant bottleneck. The cost of obtaining labeled data is high, and in some areas, the labeling process is not only time-consuming, but also easily affected by the subjective judgment of the labeler, resulting in poor data consistency, which further affects the generalization ability and accuracy of the model. In addition, the models in the existing technology are usually only optimized for a single task and lack support for multi-task information extraction. This means that in practical applications, multiple models need to be developed and maintained separately for different tasks, increasing the complexity and maintenance cost of the system.

[0004] In summary, the existing technology has the following significant shortcomings in the field of conflict and dispute data information extraction: first, the traditional rule-driven method performs poorly when faced with complex unstructured text data; second, the existing supervised learning model is highly dependent on labeled data, the data acquisition cost is high, and the quality of annotation is difficult to guarantee; finally, the existing technology lacks support for multi-task information extraction and cannot efficiently process the diverse information in conflict and dispute data. The above defects make it difficult for the existing technology to meet the growing demand for conflict and dispute data processing. A more intelligent, efficient and flexible information extraction method is urgently needed to solve the problems existing in the existing technology. Summary of the invention

[0005] One purpose of the present invention is to propose a method for extracting conflict and dispute data information based on the UIE model. The present invention significantly improves the efficiency and accuracy of conflict and dispute data information extraction, can adapt to diverse task requirements, and has strong model adaptability and scalability.

[0006] According to an embodiment of the present invention, a method for extracting conflict and dispute data information based on a UIE model comprises the following steps:

[0007] S1. Obtain text data related to conflicts and disputes, integrate and clean the text data, remove duplicate, erroneous and incomplete data, and desensitize personal privacy information in the text data to form a complete conflict and dispute data set;

[0008] S2. Using part of the data in the conflict and dispute dataset, the UIE model is self-supervised pre-trained from open source corpus to optimize the encoding ability of the UIE model for text data and generate a general word embedding model.

[0009] S3, build a structured information extraction language and convert the structured representation of various extraction tasks into a standardized coding format;

[0010] S4. Locate the key information of the unstructured text in the conflict and dispute dataset, associate and combine the located text segments using the word embedding model, and convert the unstructured text information into structured text information through the structured information extraction language;

[0011] S5. Construct a structured pattern guide to determine the elements that need to be identified in the information extraction process of the conflict and dispute dataset, including noun information and the names associated with it. Combine the structured information extraction language with the prompt words of the structured pattern guide, the text information in the conflict and dispute dataset, and the generated word embedding model, and input them into the encoder of the UIE model to generate a hidden layer representation, and decode the input text into a linear SEL format through an automatic regression method to output structured information;

[0012] S6. Construct prompt word information according to the conflict and dispute task and bring it into the UIE model. Use the conflict and dispute dataset to train the parameter-optimized UIE model, extract triple information containing entities, relations, and attributes, and represent the extracted information in a unified manner through a structured information extraction language.

[0013] S7. Use triple information, conflict and dispute datasets, and word embedding models to build a conflict and dispute-specific knowledge graph.

[0014] Optionally, S1 includes the following sub-steps:

[0015] S11, obtaining text data D related to conflicts and disputes, the text data including judgment documents, police data and case data;

[0016] S12, integrate the text data D, and define the integrated text data set as :

[0017] ;

[0018] in, represents each individual text data entry, and n represents the total number of text data entries;

[0019] S13. The integrated text dataset Clean and remove text data sets The erroneous data and incomplete data in the text dataset are desensitized to the personal privacy information in the text dataset, and the desensitized text dataset is represented as ;

[0020] S14. Finally, a complete conflict and dispute dataset is formed :

[0021] .

[0022] Optionally, S2 includes the following sub-steps:

[0023] S21. From the conflict and dispute dataset Select some data as training data set , define the selection function ,in Indicates selection conditions, which can be time range or specific case type;

[0024] S22. Obtain open source corpus datasets The open source corpus dataset is used to enhance the general text encoding capability of the UIE model and is combined with the training dataset to form an extended dataset. ;

[0025] S23, based on the extended dataset , perform self-supervised pre-training on the UIE model and define the loss function In the following form:

[0026] ;

[0027] in, is the input text of the i-th data entry in the extended dataset, is the encoding representation generated by the UIE model, and is the weight coefficient, is a regularization term used to avoid overfitting of the model. are model parameters;

[0028] S24. During the pre-training process, optimize the UIE model’s ability to encode text data and generate a general word embedding model :

[0029] ;

[0030] in, To input text, is the representation of the kth hidden layer, is the activation function, is the weight coefficient of each layer, are the parameters of the corresponding layer.

[0031] Optionally, S3 includes the following sub-steps:

[0032] S31. Based on the specific needs of the conflict and dispute field, construct a structured information extraction language SEL and define the structured representation of SEL :

[0033] ;

[0034] Among them, E represents the entity set, R represents the relationship set, and T represents the event set;

[0035] S32. Convert the structured representation of various extraction tasks into a standardized encoding format and define the conversion function ,in represents the structured representation of the input task, Indicates the conversion condition. During the conversion process, according to the structured representation The entity, relationship and event definitions in the , make the encoding format of each type of extraction task consistent, and generate a standardized encoding format , and standardize the encoding format through the UIE model to verify.

[0036] Optionally, S4 includes the following sub-steps:

[0037] S41. In the conflict and dispute dataset In this paper, we use a pre-trained general word embedding model Encode unstructured text and obtain the embedded representation of the text :

[0038] ;

[0039] in, represents the i-th word in the input text segment, and are the weight matrices for word embedding and context embedding, respectively. and is the bias term, and is a nonlinear activation function, is the hidden state representation of the previous layer, and is the weight coefficient of different levels in the model, n is the text length, and m is the number of hidden layers;

[0040] S42. Embedding representation based on encoded text , using the positioning function Locate key information:

[0041] ;

[0042] in, is the weight coefficient, is the positioning matrix, is the position offset, is the positioning threshold, To prevent a tiny constant with a zero denominator, is the activation function;

[0043] S43, the text segment after positioning Using the Correlation Function To associate a combination:

[0044] ;

[0045] in, and is the weight coefficient in the combination rule, is the hyperbolic tangent activation function, is the parameter that controls the strength of the combination. is the number of text segments to be combined;

[0046] S44, through the structured information extraction language SEL, the associated combined unstructured text information Convert to structured text information :

[0047] ;

[0048] in, is the weight coefficient of the structured representation, is the relationship matrix in the structured representation, is the transformation function, and t is the number of relationship matrices.

[0049] Optionally, S5 includes the following sub-steps:

[0050] S51. Build a structured pattern guide SSI to define the set of elements that need to be identified in the conflict and dispute dataset , the element set includes noun information N and the name L associated with it, that is, ,in , , p and q are the number of nouns and names respectively. During the construction process, the following formula is used to determine the association between each element and the corresponding name:

[0051] ;

[0052] in, and Noun and name The word embedding vector of and is the weight coefficient, is a nonlinear activation function, To prevent a tiny constant with a denominator of zero, K is the number of activation functions;

[0053] S52, the structured information extraction language SEL and the structured pattern guide SSI prompt words , text information in the conflict and dispute dataset And the generated word embedding model Combined, define the combination function , generating a synthetic representation for input into the UIE model encoder ;

[0054] S53, the comprehensive representation Input to the encoder of the UIE model to generate hidden layer representation ;

[0055] S54, use the automatic regression method to represent the hidden layer Decode into the linear information extraction language SEL format, define the decoding function, and the decoded structured information is represented as :

[0056] ;

[0057] in, is the decoding matrix, is the decoding weight matrix, and is the weight coefficient of the decoding level, is the decoding function, A tiny constant to prevent the denominator from being zero.

[0058] The beneficial effects of the present invention are:

[0059] (1) The present invention adopts the UIE model as the core, and realizes the parallel processing of multiple tasks such as entity recognition, relationship extraction and event extraction in conflict and dispute data by constructing a unified structured information extraction language SEL and a structured pattern instructor SSI. Different from the limitation of traditional methods that require a separate model to be designed for each specific task, the method of the present invention can handle multiple information extraction tasks under a unified architecture, greatly simplifying the complexity of task processes and model management, while improving the overall efficiency and stability of the model.

[0060] (2) By performing self-supervised pre-training on open source corpora and conflict and dispute datasets, the present invention significantly improves the UIE model's ability to encode text data. The generated universal word embedding model can more accurately capture the contextual semantic information in the text, reducing the dependence on vertical field annotated data. The improved loss function and optimization algorithm effectively reduce the risk of model overfitting, allowing the model to more accurately locate and extract key information when facing unstructured text, thereby improving the accuracy and robustness of information extraction.

[0061] (3) By designing association functions and positioning functions, the present invention can effectively identify and maintain the contextual association between text segments when processing unstructured text in conflict and dispute data. The located text segments are associated and combined through a word embedding model. Combined with a structured information extraction language, the associated text segments are converted into structured information with high semantic consistency. In complex conflict and dispute cases, characters, events and their relationships can be extracted more accurately, thereby constructing a more complete and accurate conflict and dispute knowledge graph.

[0062] (4) During the decoding process, the present invention decodes the encoded text representation into a linear SEL format through an automatic regression method. By iteratively generating outputs, the decoding results of each stage can make full use of the previously generated information, thereby ensuring that the structured information of the final output has a high degree of integrity and coherence. Especially when processing long texts or complex sentences, the automatic regression decoding method can significantly reduce the situation of information loss or structural errors, ensuring that the full picture of the contradictory and dispute data information can be fully presented. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0064] Figure 1 This is a general framework diagram of a method for extracting conflict and dispute data information based on a UIE model proposed by the present invention;

[0065] Figure 2This is a schematic diagram of the implementation process of a method for extracting conflict and dispute data information based on the UIE model proposed in the present invention. DETAILED DESCRIPTION

[0066] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, which only illustrate the basic structure of the present invention in a schematic manner, and therefore only show the components related to the present invention.

[0067] refer to Figure 1-2 , a method for extracting conflict and dispute data information based on UIE model, comprising the following steps:

[0068] S1. Obtain text data related to conflicts and disputes, integrate and clean the text data, remove duplicate, erroneous and incomplete data, and desensitize personal privacy information in the text data to form a complete conflict and dispute data set;

[0069] S2. Using part of the data in the conflict and dispute dataset, the UIE model is self-supervised pre-trained from open source corpus to optimize the encoding ability of the UIE model for text data and generate a general word embedding model.

[0070] S3, build a structured information extraction language and convert the structured representation of various extraction tasks into a standardized coding format;

[0071] S4. Locate the key information of the unstructured text in the conflict and dispute dataset, associate and combine the located text segments using the word embedding model, and convert the unstructured text information into structured text information through the structured information extraction language;

[0072] S5. Construct a structured pattern guide to determine the elements that need to be identified in the information extraction process of the conflict and dispute dataset, including noun information and the names associated with it. Combine the structured information extraction language with the prompt words of the structured pattern guide, the text information in the conflict and dispute dataset, and the generated word embedding model, and input them into the encoder of the UIE model to generate a hidden layer representation, and decode the input text into a linear SEL format through an automatic regression method to output structured information;

[0073] S6. Construct prompt word information according to the conflict and dispute task and bring it into the UIE model. Use the conflict and dispute dataset to train the parameter-optimized UIE model, extract triple information containing entities, relations, and attributes, and represent the extracted information in a unified manner through a structured information extraction language.

[0074] S7. Use triple information, conflict and dispute datasets, and word embedding models to build a conflict and dispute-specific knowledge graph.

[0075] In this implementation, S1 includes the following sub-steps:

[0076] S11, obtaining text data D related to conflicts and disputes, the text data including judgment documents, police data and case data;

[0077] S12, integrate the text data D, and define the integrated text data set as :

[0078] ;

[0079] in, represents each individual text data entry, and n represents the total number of text data entries;

[0080] S13. The integrated text dataset Clean and remove text data sets The erroneous data and incomplete data in the text dataset are desensitized to the personal privacy information in the text dataset, and the desensitized text dataset is represented as ;

[0081] S14. Finally, a complete conflict and dispute dataset is formed :

[0082] .

[0083] In this implementation, S2 includes the following sub-steps:

[0084] S21. From the conflict and dispute dataset Select some data as training data set , define the selection function ,in Indicates selection conditions, which can be time range or specific case type;

[0085] S22. Obtain open source corpus datasets The open source corpus dataset is used to enhance the general text encoding capability of the UIE model and is combined with the training dataset to form an extended dataset. ;

[0086] S23, based on the extended dataset , perform self-supervised pre-training on the UIE model and define the loss function In the following form:

[0087] ;

[0088] in, is the input text of the i-th data entry in the extended dataset, is the encoding representation generated by the UIE model, and is the weight coefficient, is a regularization term used to avoid overfitting of the model. are model parameters;

[0089] S24. During the pre-training process, optimize the UIE model’s ability to encode text data and generate a general word embedding model :

[0090] ;

[0091] in, To input text, is the representation of the kth hidden layer, is the activation function, is the weight coefficient of each layer, are the parameters of the corresponding layer.

[0092] In this implementation, S3 includes the following sub-steps:

[0093] S31. Based on the specific needs of the conflict and dispute field, construct a structured information extraction language SEL and define the structured representation of SEL :

[0094] ;

[0095] Among them, E represents the entity set, R represents the relationship set, and T represents the event set;

[0096] S32. Convert the structured representation of various extraction tasks into a standardized encoding format and define the conversion function ,in represents the structured representation of the input task, Indicates the conversion condition. During the conversion process, according to the structured representation The entity, relationship and event definitions in the , make the encoding format of each type of extraction task consistent, and generate a standardized encoding format , and standardize the encoding format through the UIE model to verify.

[0097] In this implementation, S4 includes the following sub-steps:

[0098] S41. In the conflict and dispute dataset In this paper, we use a pre-trained general word embedding model Encode unstructured text and obtain the embedded representation of the text :

[0099] ;

[0100] in, represents the i-th word in the input text segment, and are the weight matrices for word embedding and context embedding, respectively. and is the bias term, and is a nonlinear activation function, is the hidden state representation of the previous layer, and is the weight coefficient of different levels in the model, n is the text length, and m is the number of hidden layers;

[0101] S42. Embedding representation based on encoded text , using the positioning function Locate key information:

[0102] ;

[0103] in, is the weight coefficient, is the positioning matrix, is the position offset, is the positioning threshold, To prevent a tiny constant with a zero denominator, is the activation function;

[0104] S43, the text segment after positioning Using the Correlation Function To associate a combination:

[0105] ;

[0106] in, and is the weight coefficient in the combination rule, is the hyperbolic tangent activation function, is the parameter that controls the strength of the combination. is the number of text segments to be combined;

[0107] S44, through the structured information extraction language SEL, the associated combined unstructured text information Convert to structured text information :

[0108] ;

[0109] in, is the weight coefficient of the structured representation, is the relationship matrix in the structured representation, is the transformation function, and t is the number of relationship matrices.

[0110] In this implementation, S5 includes the following sub-steps:

[0111] S51. Build a structured pattern guide SSI to define the set of elements that need to be identified in the conflict and dispute dataset , the element set includes noun information N and the name L associated with it, that is, ,in , , p and q are the number of nouns and names respectively. During the construction process, the following formula is used to determine the association between each element and the corresponding name:

[0112] ;

[0113] in, and Noun and name The word embedding vector of and is the weight coefficient, is a nonlinear activation function, To prevent a tiny constant with a denominator of zero, K is the number of activation functions;

[0114] S52, the structured information extraction language SEL and the structured pattern guide SSI prompt words , text information in the conflict and dispute dataset And the generated word embedding model Combined, define the combination function , generating a synthetic representation for input into the UIE model encoder ;

[0115] S53, the comprehensive representation Input to the encoder of the UIE model to generate hidden layer representation ;

[0116] S54, use the automatic regression method to represent the hidden layer Decode into the linear information extraction language SEL format, define the decoding function, and the decoded structured information is represented as :

[0117] ;

[0118] in, is the decoding matrix, is the decoding weight matrix, and is the weight coefficient of the decoding level, is the decoding function, A tiny constant to prevent the denominator from being zero.

[0119] The present invention adopts the UIE model as the core, and realizes the parallel processing of multiple tasks such as entity recognition, relationship extraction and event extraction in conflict and dispute data by constructing a unified structured information extraction language SEL and a structured pattern instructor SSI. Different from the limitation of traditional methods that require a separate model to be designed for each specific task, the method of the present invention can handle multiple information extraction tasks under a unified architecture, greatly simplifying the complexity of task processes and model management, while improving the overall efficiency and stability of the model.

[0120] By performing self-supervised pre-training on open source corpora and conflict and dispute datasets, the present invention significantly improves the UIE model's encoding ability for text data. The generated universal word embedding model can more accurately capture the contextual semantic information in the text, reducing the dependence on vertical field annotated data. The improved loss function and optimization algorithm effectively reduce the risk of model overfitting, allowing the model to more accurately locate and extract key information when facing unstructured text, thereby improving the accuracy and robustness of information extraction.

[0121] By designing association functions and positioning functions, the present invention can effectively identify and maintain the contextual association between text segments when processing unstructured text in conflict and dispute data, associate and combine the located text segments through a word embedding model, and combine with a structured information extraction language to convert the associated text segments into structured information with high semantic consistency, so that in complex conflict and dispute cases, people, events and their relationships can be extracted more accurately, thereby constructing a more complete and accurate conflict and dispute knowledge graph.

[0122] During the decoding process, the present invention decodes the encoded text representation into a linear SEL format through an automatic regression method, and generates outputs iteratively, so that the decoding results of each stage can make full use of the previously generated information, thereby ensuring that the structured information finally output has a high degree of integrity and coherence. Especially when processing long texts or complex sentences, the automatic regression decoding method can significantly reduce information loss or structural errors, ensuring that the full picture of contradictory and dispute data information is fully presented.

[0123] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for extracting conflict and dispute data information based on UIE model, characterized in that: The steps include: S1. Obtain text data related to conflicts and disputes, integrate and clean the text data, remove duplicate, erroneous and incomplete data, and desensitize personal privacy information in the text data to form a complete conflict and dispute data set; S2. Using part of the data in the conflict and dispute dataset, the UIE model is self-supervised pre-trained from open source corpus to optimize the encoding ability of the UIE model for text data and generate a general word embedding model. S3, build a structured information extraction language and convert the structured representation of various extraction tasks into a standardized coding format; S4. Locate the key information of the unstructured text in the conflict and dispute dataset, associate and combine the located text segments using the word embedding model, and convert the unstructured text information into structured text information through the structured information extraction language; The S4 comprises the following sub-steps: S41. In the conflict and dispute dataset In this paper, we use a pre-trained general word embedding model Encode unstructured text and obtain the embedded representation of the text : ； in, represents the i-th word in the input text segment, and are the weight matrices for word embedding and context embedding, respectively. and is the bias term, and is a nonlinear activation function, is the hidden state representation of the previous layer, and is the weight coefficient of different levels in the model, n is the text length, and m is the number of hidden layers; S42. Embedding representation based on encoded text , using the positioning function Locate key information: ； in, is the weight coefficient, is the positioning matrix, is the position offset, is the positioning threshold, To prevent a tiny constant with a zero denominator, is the activation function; S43, the text segment after positioning Using the Correlation Function To associate a combination: ； in, and is the weight coefficient in the combination rule, is the hyperbolic tangent activation function, is the parameter that controls the strength of the combination. is the number of text segments to be combined; S44, through the structured information extraction language SEL, the associated combined unstructured text information Convert to structured text information : ； in, is the weight coefficient of the structured representation, is the relationship matrix in the structured representation, is the conversion function, t is the number of relationship matrices; S5. Construct a structured pattern guide to determine the elements that need to be identified in the information extraction process of the conflict and dispute dataset, including noun information and the names associated with it. Combine the structured information extraction language with the prompt words of the structured pattern guide, the text information in the conflict and dispute dataset, and the generated word embedding model, and input them into the encoder of the UIE model to generate a hidden layer representation, and decode the input text into a linear SEL format through an automatic regression method to output structured information; S6. Construct prompt word information according to the conflict and dispute task and bring it into the UIE model. Use the conflict and dispute dataset to train the parameter-optimized UIE model, extract triple information containing entities, relations, and attributes, and represent the extracted information in a unified manner through a structured information extraction language. S7. Use triple information, conflict and dispute datasets, and word embedding models to build a conflict and dispute-specific knowledge graph.

2. According to the method for extracting conflict and dispute data information based on UIE model in claim 1, it is characterized in that: The S1 comprises the following sub-steps: S11, obtaining text data D related to conflicts and disputes, the text data including judgment documents, police data and case data; S12, integrate the text data D, and define the integrated text data set as : ； in, represents each individual text data entry, and n represents the total number of text data entries; S13. The integrated text dataset Clean and remove text data sets The erroneous data and incomplete data in the text dataset are desensitized to the personal privacy information in the text dataset, and the desensitized text dataset is represented as ; S14. Finally, a complete conflict and dispute dataset is formed : 。 3. According to the method for extracting conflict and dispute data based on UIE model in claim 1, it is characterized in that: The S2 comprises the following sub-steps: S21. From the conflict and dispute dataset Select some data as training data set , define the selection function ,in Indicates selection conditions, which can be time range or specific case type; S22. Obtaining open source corpus datasets The open source corpus dataset is used to enhance the general text encoding capability of the UIE model and is combined with the training dataset to form an extended dataset. ; S23, based on the extended dataset , perform self-supervised pre-training on the UIE model and define the loss function In the following form: ； in, is the input text of the i-th data entry in the extended dataset, is the encoding representation generated by the UIE model, and is the weight coefficient, is a regularization term used to avoid overfitting of the model. are model parameters; S24. During the pre-training process, optimize the UIE model’s ability to encode text data and generate a general word embedding model : ； in, To input text, is the representation of the kth hidden layer, is the activation function, is the weight coefficient of each layer, are the parameters of the corresponding layer.

4. According to the method for extracting conflict and dispute data information based on UIE model in claim 1, it is characterized in that: The S3 comprises the following sub-steps: S31. Based on the specific needs of the conflict and dispute field, construct a structured information extraction language SEL and define the structured representation of SEL : ； Among them, E represents the entity set, R represents the relationship set, and T represents the event set; S32. Convert the structured representation of various extraction tasks into a standardized encoding format and define the conversion function ,in represents the structured representation of the input task, Indicates the conversion condition. During the conversion process, according to the structured representation The entity, relationship and event definitions in the , make the encoding format of each type of extraction task consistent, and generate a standardized encoding format , and standardize the encoding format through the UIE model to verify.

5. According to the method for extracting conflict and dispute data based on UIE model in claim 1, it is characterized in that: The S5 comprises the following sub-steps: S51. Build a structured pattern guide SSI to define the set of elements that need to be identified in the conflict and dispute dataset , the element set includes noun information N and the name L associated with it, that is, ,in , , p and q are the number of nouns and names respectively. During the construction process, the following formula is used to determine the association between each element and the corresponding name: ； in, and Noun and name The word embedding vector of and is the weight coefficient, is a nonlinear activation function, To prevent a tiny constant with a denominator of zero, K is the number of activation functions; S52, the structured information extraction language SEL and the structured pattern guide SSI prompt words , text information in the conflict and dispute dataset And the generated word embedding model Combined, define the combination function , generating a synthetic representation for input into the UIE model encoder ; S53, the comprehensive representation Input to the encoder of the UIE model to generate hidden layer representation ; S54, use the automatic regression method to represent the hidden layer Decode into the linear information extraction language SEL format, define the decoding function, and the decoded structured information is represented as : ； in, is the decoding matrix, is the decoding weight matrix, and is the weight coefficient of the decoding level, is the decoding function, A tiny constant to prevent the denominator from being zero.

Images