Production safety accident liability investigation information extraction method and system

Abstract

The invention relates to the technical field of electric digital data processing, in particular to a production safety accident responsibility investigation information extraction method and system, and the method comprises the steps: obtaining a target rule specification and a target production safety accident investigation report set, and obtaining a target production safety accident responsibility association data set; executing data cleaning on the production safety accident responsibility associated data set to obtain a systematized responsibility associated data set; annotating the systematized responsibility associated data set based on an annotation tool; a multi-task information extraction model based on a base model ERNIE is constructed; optimizing the multi-task information extraction model; and performing regularization by using a weight attenuation control model, setting an early stop method, and executing production safety accident liability investigation information extraction. According to the method provided by the invention, information such as persons in charge, units and places in the accident report can be identified, and structured output is performed, so that the extraction efficiency and accuracy of production safety accident liability investigation information are improved.

Description

Technical Field

[0001] The present application relates to the technical field of electronic digital data processing, and in particular to a method for extracting information on production safety accident responsibility investigation, and also to a system. Background Art

[0002] At present, the accountability analysis of production safety accidents is mostly done by manual reading methods to identify information such as the responsible parties and handling suggestions described in the text. This method is time-consuming and laborious, and the extraction results are affected by personal subjective and psychological factors. Based on this, in recent years, the Transformer model has gradually replaced manual reading methods, recurrent neural networks, and convolutional neural networks with its powerful context understanding ability, suitability for multi-task learning, pre-training advantages, and processing of complex text structures, becoming the standard configuration of natural language processing models. Summary of the invention

[0003] The inventor found through research that information extraction is a classic natural language processing task, and its main research contents include entity extraction, relationship extraction, and event extraction. The use of information extraction technology based on the Transformer model to obtain key text information has been applied in the fields of electricity, medicine, meteorology, etc., but there is no case study on the accountability of production safety accidents. Research on how to extract information on the accountability of production safety accidents is of great significance to strengthening and implementing the main responsibilities of production and operation units and the regulatory responsibilities of the government.

[0004] The purpose of this application is to provide a method and system for extracting information on accountability for production safety accidents. By constructing a multi-task information extraction model based on the base model ERNIE based on labeled data and a pre-trained Transformer architecture, the technical problem that the prior art cannot provide a method and system that can improve the efficiency and accuracy of extracting information on accountability for production safety accidents.

[0005] According to one aspect of the present application, a method for extracting production safety accident responsibility information is provided, the method being executed by a processor, comprising: Obtain the target rule regulations and target production safety accident investigation report set, and obtain the target production safety accident responsibility association data set; Perform data cleaning on the production safety accident responsibility association data set to obtain a systematic responsibility association data set; Perform annotation on the systematic responsibility association dataset based on annotation tools; Based on the labeled data and the ERNIE model of the pre-trained Transformer architecture, a multi-task information extraction model based on the base model ERNIE is constructed; Optimize the multi-task information extraction model; Use weight decay to control model regularization, set early stopping method, and extract production safety accident responsibility information.

[0006] In some embodiments, the data cleaning of the production safety accident responsibility association data set is performed to obtain a systematic responsibility association data set, specifically: by collating and summarizing the punishment methods and expression contents stipulated in different types of target rules, the production safety accident responsibility association data set is at least screened, eliminated and classified to obtain a systematic responsibility association data set.

[0007] In some embodiments, the annotation of the systematic responsibility association data set based on the annotation tool is specifically performed as follows: Based on Label Studio, a small sample of the target production safety accident responsibility association data set is labeled to extract key information.

[0008] In some embodiments, the labeling of a small sample of target production safety accident responsibility association data set based on Label Studio is specifically as follows: Configure the annotation format for named entity recognition, annotate the names of persons responsible for production safety accidents, and annotate units and positions with enterprise or government agencies; Based on the systematic responsibility association data set, the disciplinary information is annotated and the relationship between entities is annotated.

[0009] In some embodiments, the ERNIE model based on the labeled data and the pre-trained Transformer architecture constructs a multi-task information extraction model based on the base model ERNIE, specifically: The labeled systematic responsibility association dataset is divided into a training set, a validation set, and a test set in proportion, where the ratio of the training set, the validation set, and the test set is 7:1.5:1.5; Based on the training set, a multi-task information extraction model based on the base model ERNIE is obtained.

[0010] In some embodiments, the optimized multi-task information extraction model is specifically: Based on the optimizer, configure the grid search method, at least adjust the learning rate and batch size; Perform evaluation based on the validation set to obtain the best parameters; At least configure synonym replacement and random masking methods to perform augmentation on the training data. In some embodiments, the weight decay control model is regularized and an early stopping method is set to perform production safety accident responsibility investigation information extraction, specifically: Use weight decay to control model regularization, and set early stopping to stop training when the validation set loss does not improve within a few consecutive epochs.

[0011] According to another aspect of the present application, a system for extracting information on responsibility for production safety accidents is provided. The system is controlled by a processor and includes: A data acquisition module, the data acquisition module is used to acquire target rule provisions and target production safety accident investigation report set, and obtain a target production safety accident responsibility association data set; A text preprocessing module, wherein the text preprocessing module is used to perform data cleaning on a production safety accident responsibility association data set to obtain a systematic responsibility association data set; An information extraction module, which is used to annotate the systematic responsibility association data set based on an annotation tool, and to construct a multi-task information extraction model based on the base model ERNIE based on the annotated data and the ERNIE model of the pre-trained Transformer architecture; An execution module, the execution module is used to optimize the multi-task information extraction model, use weight decay to control model regularization, and set an early stopping method to perform production safety accident responsibility investigation information extraction; The result output module is used to display the extraction results.

[0012] Compared with the prior art, this application has the following advantages and beneficial effects: the method of this application uses the multi-task information extraction model of the base model ERNIE to perform text analysis and information extraction of production safety accident reports. Through model fine-tuning, it can automatically identify the responsible person, unit, and punishment information in the accident report, and output it in a structured manner; at the same time, the system of this application can handle various complex language patterns, especially long texts and overlapping responsibilities in accident reports. Based on this, this application has the characteristics of strong context understanding ability, automatic and efficient, adaptable to complex scenarios, good transfer learning ability, and structured output. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0014] Figure 1 is a flow chart of the processing method of this application; Figure 2 It is a schematic diagram of the processing method of this application; Figure 3It is a structural diagram of the processing system of this application. DETAILED DESCRIPTION

[0015] The following will be combined with the attached examples of the present application Figure 1-3 The technical solutions in the embodiments of the present application are clearly and completely described together. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments.

[0016] Application Overview In order to solve the problems that there is no accurate information extraction method and system for accountability of production safety accidents in China, the rule-based information extraction method has weak generalization ability, and the statistical model-based information extraction method has strong dependence on data, this application proposes a method and system for extracting information on accountability of production safety accidents, so as to improve the efficiency and accuracy of extracting information on accountability of production safety accidents. At the same time, this application has important application value and promotion prospects in promoting the implementation of production safety responsibilities.

[0017] Exemplary Methods A method for extracting production safety accident responsibility information is performed by a processor. The processor can be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc.

[0018] Specifically include: Obtain the target rule regulations and the target production safety accident investigation report set to obtain the target production safety accident responsibility association data set; perform data cleaning on the production safety accident responsibility association data set to obtain a systematic responsibility association data set; perform annotation on the systematic responsibility association data set based on annotation tools; based on the annotated data and the ERNIE model of the pre-trained Transformer architecture, build a multi-task information extraction model based on the base model ERNIE; optimize the multi-task information extraction model, use weight decay to control model regularization, and set the early stopping method to perform production safety accident responsibility information extraction. The method of this application has the characteristics of strong context understanding ability, automatic and efficient, adaptable to complex scenarios, good transfer learning ability, structured output, etc., which is of great significance in improving the extraction efficiency and accuracy of production safety accident responsibility information.

[0019] Below, each step is described in detail: In some implementations, data cleaning is performed on the production safety accident responsibility association data set to obtain a systematic responsibility association data set, specifically: by collating and summarizing the punishment methods and expression contents stipulated in different types of target rules, the production safety accident responsibility association data set is at least screened, eliminated and classified to obtain a systematic responsibility association data set. It can be understood that the different types of target rules stipulate at least: "Work Safety Law of the People's Republic of China", "Criminal Law of the People's Republic of China" and other laws and regulations, and the punishment methods are at least the types of responsibility investigation such as exemption from accountability, judicial measures and their possible expressions, some of which refer to the following Table 1: Table 1 Target rule data table In some implementations, the systematic responsibility association data set is annotated based on an annotation tool, specifically: Based on Label Studio, a small sample of the target production safety accident responsibility association data set is labeled to extract key information.

[0020] Specifically, based on Label Studio, a small sample of the target production safety accident responsibility association data set is labeled as follows: Configure the annotation format of named entity recognition, annotate the names of the persons responsible for production safety accidents, and annotate the units and positions with enterprises or government agencies; annotate the disciplinary information and the relationships between entities based on the systematic responsibility association data set. It can be understood that the above method can clarify the affiliation between the person responsible and his unit, as well as the disciplinary relationship between the person responsible and his disciplinary action. For example, annotate that a person responsible belongs to a specific enterprise or government unit, and annotate the specific disciplinary action he received for failing to perform his duties, such as warnings, demerits, etc. Through the above method, the annotation data set is gradually constructed.

[0021] In detail, based on the ERNIE model of the labeled data and pre-trained Transformer architecture, a multi-task information extraction model based on the base model ERNIE is constructed. It can be understood that: the aforementioned ERNIE model is used as the base model by using its powerful context representation ability after large-scale Chinese corpus pre-training, and entity extraction subtasks and relationship extraction subtasks are added. The subtasks share the encoding layer of the base model ERNIE; the entity extraction task uses the TokenClassification layer to accept the context representation output by the base model ERNIE and output the entity category of each token; the relationship extraction task uses the SequenceClassification layer to accept the context representation output by the base model ERNIE through the representation of entity pairs, and calculates the relationship between entity pairs through the vector representation of the start and end positions of the entities, and outputs the relationship category; the cross entropy loss function is used to calculate the losses of the entity extraction and relationship extraction subtasks, and the weighted sum of the losses of the two subtasks is the overall loss.

[0022] Specifically: The labeled systematic responsibility association dataset is divided into training set, validation set and test set in proportion, where the ratio of training set, validation set and test set is 7:1.5:1.5; based on the training set, a multi-task information extraction model based on the base model ERNIE is obtained.

[0023] In some implementations, the multi-task information extraction model is optimized, specifically: based on Optimizer, configure the grid search method, at least adjust the learning rate and batch size; perform evaluation based on the validation set to obtain the best parameters; at least configure synonym replacement and random masking methods to enhance the training data; In some embodiments, weight decay is used to control model regularization, and an early stopping method is set to perform production safety accident responsibility information extraction. Specifically, weight decay is used to control model regularization, and an early stopping method is set to stop training when the validation set loss has not improved within several consecutive epochs to avoid overfitting.

[0024] Finally, based on the previous steps, the accuracy, recall and weighted F1 score of the model in entity extraction and relationship extraction tasks are calculated through the test set, and the generalization ability and actual effect of the multi-task information extraction model based on the base model ERNIE are comprehensively evaluated, and the evaluation results are at least saved locally.

[0025] In order to better understand the method of the present application, the present application also provides a production safety accident responsibility investigation information extraction system, which is controlled by a processor. Exemplarily, the aforementioned method can be divided into one or more modules, one or more modules are stored in a memory and executed by a processor to complete the present application. One or more modules can be a series of computer program instruction segments that can perform specific functions, and the instruction segments are used to describe the execution process of the computer program in the terminal device. For example, the aforementioned method can be divided into a data acquisition module, a text preprocessing module, an information extraction module, an execution module and a result output module, and the specific functions of each module are as follows: the data acquisition module is used to acquire the target rule provisions and the target production safety accident investigation report set, and obtain the target production safety accident responsibility association data set; the text preprocessing module is used to perform data cleaning on the production safety accident responsibility association data set, and obtain a systematic responsibility association data set; the information extraction module is used to perform annotation on the systematic responsibility association data set based on annotation tools, and based on the annotated data and the pre-trained Transformer architecture ERNIE model, construct a multi-task information extraction model based on the base model ERNIE; the execution module is used to optimize the multi-task information extraction model, use weight decay to control model regularization, and set the early stopping method to perform production safety accident responsibility information extraction; the result output module is used to display the extraction results.

[0026] Among them, it can be understood that the text preprocessing module uses the re library to remove noise such as special characters, redundant spaces and irrelevant information in the text, and uses the jieba library to convert the text into a format acceptable to the information extraction model, generate a token list, and prepare the input for tokenization; the information extraction module deploys the trained ERNIE-based multi-task information extraction model, receives the preprocessed text data for information extraction; the result output module stores the extracted information such as the person in charge, unit or position, and punishment in JSON format locally, and embeds the chart of the extracted information statistics into the web page through the Pyecharts library, generates interactive charts and provides dynamic data update functions to ensure that users can intuitively view and analyze the extraction results.

[0027] The above shows and describes the basic principles and main features of the present application and the advantages of the present application. For those skilled in the art, it is obvious that the present application is not limited to the details of the above exemplary embodiments, and the present application can be implemented in other specific forms without departing from the spirit or basic features of the present application. Therefore, no matter from which point of view, the embodiments should be regarded as exemplary and non-restrictive. The scope of the present application is defined by the attached claims rather than the above description, and it is intended that all changes falling within the meaning and scope of the equivalent elements of the claims are included in the present application. Any figure mark in the claims should not be regarded as limiting the claims involved.

[0028] In addition, it should be understood that although the present specification is described according to implementation modes, not every implementation mode contains only one independent technical solution. This description of the specification is only for the sake of clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment may also be appropriately combined to form other implementation modes that can be understood by those skilled in the art.

Claims

1. A method for extracting production safety accident responsibility information, the method being executed by a processor, characterized in that: include: Obtain the target rule regulations and target production safety accident investigation report set, and obtain the target production safety accident responsibility association data set; Perform data cleaning on the production safety accident responsibility association data set to obtain a systematic responsibility association data set; Perform annotation on the systematic responsibility association dataset based on annotation tools; Based on the labeled data and the ERNIE model of the pre-trained Transformer architecture, a multi-task information extraction model based on the base model ERNIE is constructed; Optimize the multi-task information extraction model; Use weight decay to control model regularization, set early stopping method, and extract production safety accident responsibility information.

2. The extraction method according to claim 1, characterized in that: The data cleaning of the production safety accident responsibility association data set to obtain the systematic responsibility association data set is specifically: by arranging and summarizing the punishment methods and expression contents stipulated in different types of target rules, at least data screening, elimination and classification are performed on the production safety accident responsibility association data set to obtain the systematic responsibility association data set.

3. The extraction method according to claim 1, characterized in that: The method of labeling the systematic responsibility association data set based on the labeling tool is as follows: Based on Label Studio, a small sample of the target production safety accident responsibility association data set is labeled to extract key information.

4. The extraction method according to claim 3, characterized in that: The labeling of the small sample data of the target production safety accident responsibility association data set based on Label Studio is specifically as follows: Configure the annotation format for named entity recognition, annotate the names of persons responsible for production safety accidents, and annotate units and positions with enterprise or government agencies; Based on the systematic responsibility association data set, the disciplinary information is annotated and the relationship between entities is annotated.

5. The extraction method according to claim 4, characterized in that: The ERNIE model based on labeled data and pre-trained Transformer architecture constructs a multi-task information extraction model based on the base model ERNIE, specifically: The labeled systematic responsibility association dataset is divided into a training set, a validation set, and a test set according to the proportion, where the ratio of the training set, the validation set, and the test set is: 7:1.5:1.5； Based on the training set, a multi-task information extraction model based on the base model ERNIE is obtained.

6. The extraction method according to claim 1, characterized in that: The optimized multi-task information extraction model is specifically: based on The optimizer, configures the grid search method and performs at least tuning the learning rate and batch size; Perform evaluation based on the validation set to obtain the best parameters; At least synonym replacement and random masking methods are configured to perform augmentation on the training data.

7. The extraction method according to claim 6, characterized in that: The weight decay control model is used for regularization, and the early stopping method is set to perform production safety accident responsibility investigation information extraction, specifically: Use weight decay to control model regularization, and set early stopping to stop training when the validation set loss does not improve within a few consecutive epochs.

8. A production safety accident responsibility investigation information extraction system, the system is controlled by a processor, characterized in that: include: A data acquisition module, the data acquisition module is used to acquire target rule provisions and target production safety accident investigation report set, and obtain a target production safety accident responsibility association data set; A text preprocessing module, wherein the text preprocessing module is used to perform data cleaning on a production safety accident responsibility association data set to obtain a systematic responsibility association data set; An information extraction module, which is used to annotate the systematic responsibility association data set based on an annotation tool, and to construct a multi-task information extraction model based on the base model ERNIE based on the annotated data and the ERNIE model of the pre-trained Transformer architecture; An execution module, the execution module is used to optimize the multi-task information extraction model, use weight decay to control model regularization, and set an early stopping method to perform production safety accident responsibility investigation information extraction; The result output module is used to display the extraction results.

Images