Method for extracting chapter-level fine-grained events in military news field

Abstract

The invention discloses a chapter-level fine-grained event extraction method in the military news field, and relates to the related technical field of military information extraction, and the method comprises the following steps: S1, data collection and preprocessing, S2, event trigger word recognition, S3, construction and training of an MRC-based PAIE event argument extraction model, S4, fine-grained event argument extraction, and S5, transfer learning application. S6, performing comprehensive optimization processing on the event argument; and S7, performing structured output and application, and outputting the event argument after comprehensive optimization processing in a structured form. According to the method, the efficiency and accuracy of event argument extraction in the chapter-level military news field are improved by introducing the PAIE model, the bipartite graph matching and reasoning stages are further introduced in the method, complexity of exhaustion threshold adjustment in a traditional method is avoided, and the series of innovative designs not only enhance the accuracy of the model, but also improve the efficiency of event argument extraction in the chapter-level military news field. And it is also ensured that the model can still effectively perform event argument extraction under the condition of data scarcity.

Description

Technical Field

[0001] The present invention relates to the technical field related to military information extraction, and in particular to a chapter-level fine-grained event extraction method in the field of military news. Background Art

[0002] With the rapid development of big data and natural language processing technology, information extraction technology has shown great application potential in many fields such as military intelligence analysis and public opinion monitoring. Information extraction can effectively transform unstructured text information into structured data by automatically identifying and extracting entities, relationships and events in text, thereby providing support for downstream analysis and decision-making. However, most current event extraction research in the field of military news focuses on the sentence level, ignoring the in-depth mining of event arguments in paragraph-level text. These methods often only focus on simple event relations within a sentence, and have difficulty in handling complex event arguments and long-distance dependencies across sentences.

[0003] As a key technology for building knowledge graphs and intelligent analysis systems, event extraction aims to automatically identify events and their related argument information from text. In the field of military news, fine-grained event extraction not only helps to accurately understand and analyze military situations, but also provides strong support for subsequent intelligence reasoning and decision-making. Traditional event extraction methods mostly rely on manually annotated features or rules, which is not only time-consuming and labor-intensive, but also difficult to cope with the complex event structure in paragraph-level text. In recent years, MRC-based event extraction methods have gradually become a research hotspot, especially in the field of fine-grained argument extraction. By combining event recognition with question-answering, the accuracy and breadth of event extraction have been significantly improved.

[0004] However, current MRC-based event extraction methods still have some limitations. On the one hand, most methods only perform event recognition and argument extraction at the sentence level, ignoring the complex event relationship across sentences in paragraph-level texts, resulting in limited extraction effects. On the other hand, event information in military news is usually highly professional and confidential. Contextual dependencies and noise data between different texts also pose challenges to event extraction. Therefore, how to effectively utilize paragraph-level contextual information and improve the accuracy and robustness of event argument extraction has become a key issue that needs to be urgently addressed in the current military news field. Summary of the invention

[0005] The purpose of the present invention is to provide a chapter-level fine-grained event extraction method in the field of military news to solve the problems raised in the above background technology.

[0006] To achieve the above object, the present invention provides the following technical solution: a chapter-level fine-grained event extraction method in the field of military news, comprising the following steps:

[0007] S1: Data collection and preprocessing: collect a large number of military news articles as sample data, clean and preprocess these data to prepare for subsequent event extraction;

[0008] S2: Identification of event trigger words: using an event detection model based on machine reading comprehension, the pre-processed military news chapters are analyzed sentence by sentence to accurately identify the event trigger words therein;

[0009] S3: Construction and training of the MRC-based PAIE event argument extraction model. The MRC-based PAIE event argument extraction model was constructed and trained using a large amount of annotated military news chapter data, and the parameters and structure of the model were continuously adjusted.

[0010] S4: Extraction of fine-grained event arguments. The trained MRC-based PAIE event argument extraction model is applied to military news articles to extract fine-grained event arguments associated with each event trigger word, thereby mining the complex interactions between trigger words, event types, and arguments, and capturing the long-range dependencies between event arguments in long texts.

[0011] S5: Application of transfer learning, using the transfer learning ability of the pre-trained language model to transfer the language model parameters trained on other large-scale text data to the event extraction task of military news chapters;

[0012] S6: Comprehensive optimization processing of event arguments, performing comprehensive optimization processing on the extracted event arguments;

[0013] S7: Structured output and application: output the event arguments after comprehensive optimization processing in a structured form.

[0014] Preferably, the pre-processing in step S1 comprises the following steps:

[0015] S11: Use Chinese word segmentation tools to segment military news articles into individual word units;

[0016] S12: Perform part-of-speech tagging to mark the part-of-speech information of each word;

[0017] S13: Build a vocabulary, summarize all the words that appear in the article, remove duplicate words, and give each word a unique identifier.

[0018] Preferably, the event detection technology of the event detection model in step S2 includes:

[0019] S21: event trigger word recognition stage, in which event trigger words are recognized through MRC question-answering templates;

[0020] S22: Event type classification stage, where the event type classification stage performs precise classification of event types based on the identified trigger words.

[0021] Preferably, the event argument extraction process in step S3 includes:

[0022] S31: Manual template, where all roles are manually connected using natural language;

[0023] S32: Soft prompt template, which combines different argument roles with role-specific pseudo-tokens that can be learned by the pre-trained language model;

[0024] S33: Connection-based template, which uses natural language to connect all role names belonging to an event type.

[0025] Preferably, the PAIE event argument extraction model in step S3 performs event argument extraction through model-created prompt templates, an event argument selection decoder part, and event argument prediction.

[0026] Preferably, the event argument prediction combines bipartite graph matching and reasoning to solve the optimal allocation problem of multiple similar role arguments in the same event:

[0027] By introducing bipartite graph matching technology, the Hungarian algorithm is used to insert multiple slots in the model template settings to achieve the global optimal allocation with the minimum cost matching. In the reasoning stage, a candidate event argument set is defined and scored;

[0028] To perform reasoning, the candidate event argument set is defined as:

[0029]

[0030] The candidate event argument set contains all spans shorter than the threshold L and the special span (0, 0) where no event argument is extracted. The model extracts the argument of each span selector θ by enumerating all candidate spans and scoring them; k and scores the argument of each span selector θ k to extract:

[0031]

[0032] And the predicted span of time slot k is given by the following formula:

[0033]

[0034] Preferably, the extraction of fine-grained event arguments in step S4 includes the following steps:

[0035] S41: Discourse analysis, conduct a detailed sentence-by-sentence analysis of military news discourses to clarify the semantics and structures of each sentence;

[0036] S42: Trigger word location, in the analyzed discourse, accurately locate each event trigger word, and accurately identify these keyword vocabularies through a model based on machine reading comprehension;

[0037] S43: Argument association, for each located event trigger word, according to its semantics and context information, associate the related fine-grained event arguments;

[0038] S44: Interaction mining, during the process of associating arguments, further mine the complex interaction situations among trigger words, event types, and arguments;

[0039] S45: Long-distance dependency capture, through the overall understanding of the discourse and the grasp of the context, capture the long-distance dependency relationships among event arguments in long texts.

[0040] Preferably, the application of transfer learning in step S5 includes the following steps:

[0041] S51: Select a pre-trained language model, according to the characteristics and requirements of military news discourses, select a suitable pre-trained language model;

[0042] S52: Model parameter transfer preparation, obtain the parameters of the pre-trained language model trained on other large-scale text data;

[0043] S53: Adapt to military news data, adapt the selected pre-trained language model so that it can adapt to the specific domain and language style of military news discourses;

[0044] S54: Parameter transfer and fine-tuning, transfer the parameters trained on other large-scale text data to the adapted military news discourse event extraction model, and then use the labeled data of military news discourses to fine-tune the model;

[0045] S55: Evaluation and optimization, evaluate the military news discourse event extraction model after transfer learning, evaluate the extraction effect of the model by comparing it with manually labeled data, and further optimize and adjust the model according to the evaluation results.

[0046] Preferably, the comprehensive optimization process of event arguments in step S6 includes the following steps:

[0047] S61: Duplicate removal processing, carefully check all the extracted event arguments, use specific algorithms and rules to find and remove the repeatedly occurring event arguments;

[0048] S62: Error correction, which conducts a detailed review of each event argument and corrects the error classification;

[0049] S63: Semantic consistency check, which checks and corrects the semantic consistency between event arguments;

[0050] S64: Quality assessment and screening, which formulates quality assessment criteria and conducts quality assessment on the event arguments after deduplication and error correction.

[0051] Preferably, the structured output and application in step S7 include the following steps:

[0052] S71: Determine the structured format, and clarify the structured output format to be adopted;

[0053] S72: Field extraction and filling, which sequentially extracts the event type, trigger word, and information of each argument from the event arguments after comprehensive optimization processing;

[0054] S73: Organize the output order, and determine the output order of event arguments according to specific application requirements and logic;

[0055] S74: Interface with the application system, and interface the structured output event arguments with the specific application system.

[0056] The technical effects and advantages of the present invention:

[0057] (1) By introducing the PAIE model, the present invention improves the efficiency and accuracy of event argument extraction in the field of chapter-level military news. This method applies the event argument extraction method based on MRC to chapter-level military news texts. In view of the characteristic that event arguments may be scattered in different sentences, a model creation prompt template part is designed. In this part, a set of prompts is created for each event type, and these prompts, through different forms of templates, such as manual templates, soft prompt templates, and connection-based templates, help the model better capture the implicit information and semantic associations between different roles. This design effectively improves the information extraction ability of the model in long texts, making event argument extraction more accurate;

[0058] (2) In order to further improve the generalization ability of the model in the few-shot case, the method also introduces bipartite graph matching and an inference stage. In the bipartite graph matching stage, the Hungarian algorithm is used to handle the problem of multiple argument roles of the same role, ensuring the globally optimal matching result, thereby reducing prediction errors. The inference stage scores the candidate event argument set and uses an enumeration strategy to select the best argument span, avoiding the complexity of exhaustive threshold adjustment in traditional methods. This series of innovative designs not only enhance the accuracy of the model but also ensure that the model can still effectively extract event arguments in the case of scarce data. Description of the Drawings

[0059] Figure 1 This is a flowchart of the method for extracting document-level fine-grained events in the field of military news according to the present invention.

[0060] Figure 2 This is an architecture diagram of event detection based on the QAED model of the present invention.

[0061] Figure 3 This is a flowchart of the preprocessing according to the present invention.

[0062] Figure 4 This is a flowchart of the event detection technology of the event detection model according to the present invention.

[0063] Figure 5 This is a flowchart of the event argument extraction process according to the present invention.

[0064] Figure 6 This is a flowchart of the extraction of fine-grained event arguments according to the present invention.

[0065] Figure 7 This is a flowchart of the application of transfer learning according to the present invention.

[0066] Figure 8 This is a flowchart of the comprehensive optimization process of event arguments according to the present invention.

[0067] Figure 9 This is a flowchart of the structured output and application according to the present invention. Detailed Embodiments

[0068] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0069] The present invention provides a method for extracting document-level fine-grained events in the field of military news as shown in Figures 1-9 and includes the following steps:

[0070] S1: Data collection and preprocessing. Collect a large number of military news articles as sample data, clean and preprocess this data to prepare for subsequent event extraction work;

[0071] S2: Identification of event trigger words. Use an event detection model based on machine reading comprehension to analyze each sentence of the preprocessed military news articles to accurately identify the event trigger words therein;

[0072] S3: Construction and training of the PAIE event argument extraction model based on MRC. Build the PAIE event argument extraction model based on MRC, and use a large amount of labeled military news text data for training. Continuously adjust the parameters and structure of the model to enable it to better understand the relationships and contexts between event arguments.

[0073] S4: Extraction of fine-grained event arguments. Apply the trained PAIE event argument extraction model based on MRC to military news texts, and extract the fine-grained event arguments associated with each event trigger word, so as to deeply explore the complex interactions between trigger words, event types, and arguments, and capture the long-distance dependencies between event arguments in long texts.

[0074] S5: Application of transfer learning. Utilize the transfer learning ability of pre-trained language models to transfer the parameters of the language model trained on other large-scale text data to the event extraction task of military news texts.

[0075] S6: Comprehensive optimization processing of event arguments. Conduct comprehensive optimization processing on the extracted event arguments.

[0076] S7: Structured output and application. Output the event arguments after comprehensive optimization processing in a structured form.

[0077] By introducing the PAIE model, the efficiency and accuracy of event argument extraction in the field of military news at the text level are improved. Apply the event argument extraction method based on MRC to military news texts at the text level. Considering the characteristic that event arguments may be scattered in different sentences, a model creation prompt template part is designed. In this part, a set of prompts are created for each event type. These prompts, through different forms of templates, such as manual templates, soft prompt templates, and connection-based templates, help the model better capture the implicit information and semantic associations between different roles. This design effectively improves the information extraction ability of the model in long texts and makes event argument extraction more accurate.

[0078] Specifically, the preprocessing in step S1 includes the following steps:

[0079] S11: Use a Chinese word segmentation tool to segment the military news text into individual word units, and make appropriate parameter adjustments and optimizations according to the characteristics of military news to improve the accuracy of word segmentation.

[0080] S12: Perform part-of-speech tagging to label the part-of-speech information of each word, such as noun, verb, adjective, etc. This helps to better understand the semantics and grammatical functions of words in the text and provides richer language information for subsequent event extraction.

[0081] S13: Construct a vocabulary by summarizing all the words that appear in the passage, removing duplicate words, and assigning a unique identifier to each word. The construction of the vocabulary helps improve the representation efficiency and processing speed of the data, and also provides a basic vocabulary resource for subsequent model training and event extraction.

[0082] Specifically, the event detection techniques of the event detection model in step S2 include:

[0083] S21: Event trigger word recognition stage, in which the event trigger word recognition stage identifies event trigger words through the MRC question-answering template;

[0084] S22: Event type classification stage, in which the event type classification stage performs precise classification of event types based on the identified trigger words;

[0085] In event detection, identifying event trigger words through the MRC question-answering template can accurately locate key information, provide a clear basis for subsequent event type classification, and perform precise classification of event types based on this. This can make event detection more targeted and accurate, effectively improve the recognition and classification efficiency of various events in fields such as military news, avoid vague and incorrect judgments, and lay a solid foundation for subsequent event analysis and processing.

[0086] Furthermore, the event argument extraction process in step S3 includes:

[0087] S31: Manual template, where all roles are manually connected using natural language;

[0088] S32: Soft prompt template, which combines different argument roles with learnable, role-specific pseudo-tokens of the pre-trained language model;

[0089] S33: Connection-based template, which uses natural language to connect all role names belonging to an event type;

[0090] Combining various types of prompt structures can guide and constrain event arguments from different perspectives. The manual template can provide clear rules and structures, the soft prompt template increases flexibility and adaptability, and the connection-based template can better capture the relationships between event arguments. In this way, the advantages of various templates can be fully utilized, the limitations of a single template can be avoided, event arguments can be extracted more comprehensively and accurately, the generalization ability of event argument extraction can be significantly enhanced, enabling it to better handle different types and styles of text, and improving the accuracy and stability of extraction.

[0091] Furthermore, in step S3, the PAIE event argument extraction model performs event argument extraction through model-created prompt templates, the event argument selection decoder part, and event argument prediction.

[0092] Furthermore, the event argument prediction combines bipartite graph matching and reasoning to solve the optimal allocation problem of multiple similar role arguments in the same event:

[0093] By introducing bipartite graph matching technology, the Hungarian algorithm is used to insert multiple slots in the model template to achieve the global optimal allocation with the minimum cost matching. In the reasoning stage, the candidate event argument set is defined and scored.

[0094] To perform reasoning, the candidate event argument set is defined as:

[0095]

[0096] The candidate event argument set contains all spans shorter than the threshold L and the special span (0, 0) where no event argument is extracted. The model extracts each span selector θ by enumerating all candidate spans and scoring them. k arguments of θ and scores them to extract each span selector θ k arguments:

[0097]

[0098] And the predicted span of time slot k is given by:

[0099]

[0100] Since each slot in the prompt predicts at most one span, this strategy avoids exhaustive threshold adjustment.

[0101] Figure 2 It is the architecture diagram of event detection based on the QAED model.

[0102] By introducing event detection technology based on the QAED (Question Answering Event Detection) model, the accuracy and generalization ability of document-level fine-grained event detection in the field of military news have been improved. First, this method transforms the traditional event detection task into a multi-turn machine reading comprehension (MRC) question-and-answer problem, which is divided into two stages: trigger word recognition and trigger word classification. Specifically, in the trigger word recognition stage, the task of extracting trigger words from unstructured text is formulated as an extractive MRC question, and a multi-round question-and-answer technology based on BERT is used to identify continuous trigger word phrases from sentences or paragraphs. In the trigger word classification stage, the trigger word classification is formulated as a yes / no MRC question, and further questions are constructed to determine whether the identified trigger words belong to a specific event type. This method effectively utilizes the semantic interaction between trigger words, event types, and event arguments, thereby gradually learning relevant knowledge of event types during rich training and improving the accuracy of event detection.

[0103] The method also introduces a strategy of converting event detection problems into two consecutive MRC questions through multi-turn question-and-answer, further solving the generalization problem of traditional methods when facing a small amount of training data. This not only enhances the scalability of the model in dealing with new event types but also avoids the limitations of traditional sequence labeling techniques and can maintain good detection performance for event types not seen by the model. By explicitly encoding event types as binary classification problems, the QAED model forms a closer association between trigger word recognition and classification, enabling the model to better meet the needs of complex and changing military news event detection with the support of rich corpora.

[0104] By introducing the PAIE model, the efficiency and accuracy of event argument extraction in the field of document-level military news have been improved. First, this method innovatively applies the MRC-based event argument extraction method to document-level military news texts. Considering the characteristic that event arguments may be scattered in different sentences, a part for creating prompt templates for the model is designed. In this part, a set of prompts is created for each event type, and these prompts, through different forms of templates, including manual templates, soft prompt templates, and connection-based templates, help the model better capture the implicit information and semantic associations between different roles. This design effectively improves the model's information extraction ability in long texts, making event argument extraction more accurate.

[0105] To further improve the generalization ability of the model in the few-shot scenario, bipartite graph matching and an inference stage are also introduced in the method. In the bipartite graph matching stage, the Hungarian algorithm is used to handle the problem of multiple argument roles of the same role, ensuring a globally optimal matching result, thereby reducing prediction errors. In the inference stage, the candidate event argument set is scored, and an enumeration strategy is used to select the best argument span, avoiding the complexity of exhaustive threshold adjustment in traditional methods. This series of innovative designs not only enhances the accuracy of the model but also ensures that the model can still effectively perform event argument extraction in the case of scarce data.

[0106] The data of the method for document-level fine-grained event extraction in the military news field comes from the publicly available CMNEE dataset. The CMNEE dataset is a large-scale open-source Chinese military news event extraction dataset at the file level, containing 17,000 documents and 29,223 events, and these events are all manually annotated according to the predefined military domain patterns, including 8 event types and 11 argument role types. This application utilizes the event and argument information of the CMNEE dataset for event extraction. In the evaluation stage, the dataset is generally divided into a training set and a test set. Among them, the training set is used for model training, while the test set is used for the final performance evaluation.

[0107] Based on the values of the evaluation metrics, the performances of various models are compared. The evaluation metrics used are Precision, Recall, and F1 value as the main evaluation metrics to test the performance and effectiveness of the named entity recognition model on three datasets. Among them, Precision and Recall are used as auxiliary evaluation metrics, while the calculation of the F1 value takes into account both accuracy and recall.

[0108] To evaluate the proposed model, some competitive entity alignment models are selected for comparison in this chapter. They mainly include: GLACIER, TIER, Multi-Granularity Reader. The following is an introduction to these three models, GLACIER, TIER, and Multi-Granularity Reader:

[0109] GLACIER uses a unified probability model for event extraction. It jointly considers sentence features and phrase features when extracting each role filling. The model consists of a sentence event classifier and a set of event role recognizers. The final extraction decision is based on the product of the normalized sentence and phrase probabilities.

[0110] TIER is a multi-layer architecture for event extraction. The documents adopt a pipeline processing flow and are analyzed at different text granularity levels, including document level, sentence level, and phrase level. In the absence of event detection, TIER aims to extract event arguments by using a document type classifier and a set of sentence classifiers specific to event roles.

[0111] The Muti-Granularity Reader aims to explore the impact of context length on the event role extraction model and proposes a multi-granularity reader that fuses sentence granularity and context granularity. The Muti-Granularity Reader has achieved the best results reported so far in the event role extraction task based on the MUC-4 dataset.

[0112] The experimental results show that the performance of the model in this chapter is almost better than all other comparison methods on the CMNEE dataset.

[0113] Table 1 Performance comparison of models

[0114] model Precision(%) Recall(%) F1(％) GLACIER 48.7 58.3 53.2 TIER 51.2 62.5 57.2 Multi-GranularityReader 58.3 60.4 59.2 model of this chapter 74.3 70.2 73.5

[0115] It can be seen that the performance of the model in this paper is better than that of other models. This is because the model effectively solves the limitations of traditional methods in label semantic modeling and generalization ability by transforming the event detection and argument extraction tasks into multi-turn question-answering questions based on machine reading comprehension (MRC). At the same time, by designing a prompt template and a bipartite graph matching algorithm, the complex semantic relationships and dependency information among event trigger words, event types, and arguments are deeply mined, especially performing well in long text processing and few-shot learning. In addition, the model utilizes the transfer learning ability of pre-trained language models, enabling efficient and accurate event extraction even in the case of scarce data, thus improving the overall performance.

[0116] Specifically, the extraction of fine-grained event arguments in step S4 includes the following steps:

[0117] S41: Discourse analysis, conducting a detailed sentence-by-sentence analysis of military news articles to clarify the semantics and structure of each sentence, laying a foundation for accurately extracting fine-grained event arguments in the following steps;

[0118] S42: Trigger word location, accurately locating each event trigger word in the analyzed article, and precisely identifying these key vocabulary through a model based on machine reading comprehension, laying a foundation for accurately extracting fine-grained event arguments in the following steps;

[0119] S43: Argument association, for each located event trigger word, associating the relevant fine-grained event arguments according to its semantics and context information, which requires an in-depth understanding of the internal logical relationship between the trigger word and the argument;

[0120] S44: Interactive mining. During the process of associating arguments, further mine the complex interaction situations among trigger words, event types, and arguments. For example, the differences in the impacts of different trigger words on arguments, the characteristics of arguments under the same event type, etc.;

[0121] S45: Long-distance dependency capture. Through the overall understanding of the text and the grasp of the context, capture the long-distance dependency relationships among event arguments in long texts, which may require spanning multiple sentences or paragraphs and comprehensively considering various semantic associations.

[0122] The extraction of fine-grained event arguments in step S4 includes text analysis to lay the foundation, accurate trigger word location for precise foundation laying, in-depth understanding of the internal logic of argument association, interactive mining to grasp complex situations, and long-distance dependency capture to comprehensively consider semantics. Its advantage is that it can carefully extract accurate fine-grained event arguments from military news texts. By comprehensively analyzing the text and mining various relationships, it improves the accuracy and comprehensiveness of event argument extraction, better understands the event details and internal logic in military news, and provides higher-quality event argument data for subsequent related research and applications.

[0123] Specifically, the application of transfer learning in step S5 includes the following steps:

[0124] S51: Select a pre-trained language model. According to the characteristics and requirements of military news texts, select a suitable pre-trained language model, such as models in the BERT, GPT, etc. series. These models have been trained on large-scale text data and have rich language knowledge and semantic understanding capabilities;

[0125] S52: Preparation for model parameter transfer. Obtain the parameters of the pre-trained language model trained on other large-scale text data. These parameters contain information such as the general patterns of language, lexical semantics, etc., and are the basis for transfer learning;

[0126] S53: Adapt to military news data. Adapt the selected pre-trained language model so that it can adapt to the specific domain and language style of military news texts. This may include adjusting the input-output format, vocabulary list, etc. of the model to better process content related to military news;

[0127] S54: Parameter transfer and fine-tuning. Transfer the parameters trained on other large-scale text data to the event extraction model of the adapted military news text, and then use the labeled data of the military news text to fine-tune the model. By adjusting the weights and structure of the model, make it better adapt to the military news event extraction task and improve the accuracy and performance of extraction;

[0128] S55: Evaluation and Optimization. Evaluate the military news article event extraction model after transfer learning. By comparing it with manually annotated data, assess the extraction effect of the model. According to the evaluation results, further optimize and adjust the model, such as adjusting parameters, improving the model structure, etc., to continuously enhance the performance of the model in the military news article event extraction task.

[0129] Step S5 utilizes the rich language knowledge and semantic understanding ability of the pre-trained language model. Through adaptation and fine-tuning, it can quickly make the model adapt to the military news field, improve the efficiency and quality of event extraction, save training time and costs, and at the same time enhance the application effect of the model in this specific field.

[0130] Specifically, the comprehensive optimization process of event arguments in step S6 includes the following steps:

[0131] S61: Duplicate Removal. Carefully check all the extracted event arguments. Using specific algorithms and rules, identify and remove duplicate event arguments. For example, by comparing the text content, semantic features, etc. of the arguments, merge exactly the same or similar arguments and only keep one copy to reduce redundant information.

[0132] S62: Error Correction. Conduct a detailed review of each event argument and correct misclassifications. For example, if an event argument is misclassified into the wrong event type, correct it according to relevant knowledge and rules to ensure that each event argument can be accurately assigned to the corresponding event type.

[0133] S63: Semantic Consistency Check. Check and correct the semantic consistency between event arguments. For example, for the relevant arguments in the same event, they should be semantically related and logically reasonable. If semantic inconsistencies are found, such as an argument being contradictory or unreasonable semantically with other relevant arguments, make adjustments and corrections to ensure the coherence and reasonableness of the entire set of event arguments semantically.

[0134] S64: Quality Evaluation and Screening. Establish quality evaluation criteria, evaluate the quality of event arguments after duplicate removal and error correction, and screen out event arguments with high quality and meeting the requirements according to the criteria. Remove those with poor quality and potential problems to further improve the quality and reliability of event arguments.

[0135] Step S6 reduces redundant information, ensures accurate argument classification, guarantees semantic coherence and reasonableness, improves the overall quality and reliability of event arguments, and provides a better data foundation for subsequent related work.

[0136] Specifically, the structured output and application in step S7 include the following steps:

[0137] S71: Determine the structured format, clarify the structured output format to be adopted. For example, a table structure containing fields such as event type, trigger word, and argument can be defined, or the JSON format can be used, etc. Determine the specific format specifications so that the event arguments can be organized according to this format later;

[0138] S72: Field extraction and filling. Sequentially extract the event type, trigger word, and information of each argument from the event arguments after comprehensive optimization processing, and accurately fill this information into the corresponding fields according to the previously determined structured format. For example, in the table structure, fill the event type into the corresponding event type field, the trigger word into the trigger word field, and the arguments into their respective argument fields;

[0139] S73: Organize the output order. According to specific application requirements and logic, determine the output order of the event arguments. It can be classified and output according to the event type, or sorted according to other logics such as time order or importance level. Ensure that the output order can meet the requirements of subsequent applications and facilitate the analysis and utilization of the event arguments;

[0140] S74: Interface with the application system. Interface the structured output event arguments with the specific application system. If it is a data analysis system, use the event arguments as data input for further statistics, analysis, and mining; if it is a visualization system, convert the event arguments into visual charts or graphs for more intuitive display of event information; if it is other related application systems, perform corresponding adaptation and interfacing according to the characteristics and requirements of the system so that the event arguments can play a role in this system;

[0141] Step S7 improves data standardization, provides a reliable data foundation, meets application requirements, enhances data value, expands the application scope, and improves business efficiency and collaboration.

[0142] Finally, it should be noted that the above are only the preferred embodiments of the present invention and are not used to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A chapter-level fine-grained event extraction method in the field of military news, characterized by: The following steps are involved: S1: Data collection and preprocessing: collect a large number of military news articles as sample data, clean and preprocess these data to prepare for subsequent event extraction; S2: Identification of event trigger words: using an event detection model based on machine reading comprehension, the pre-processed military news chapters are analyzed sentence by sentence to accurately identify the event trigger words therein; S3: Construction and training of the MRC-based PAIE event argument extraction model. The MRC-based PAIE event argument extraction model was constructed and trained using a large amount of annotated military news chapter data, and the parameters and structure of the model were continuously adjusted. S4: Extraction of fine-grained event arguments. The trained MRC-based PAIE event argument extraction model is applied to military news articles to extract fine-grained event arguments associated with each event trigger word, thereby mining the complex interactions between trigger words, event types, and arguments, and capturing the long-range dependencies between event arguments in long texts. S5: Application of transfer learning, using the transfer learning ability of the pre-trained language model to transfer the language model parameters trained on other large-scale text data to the event extraction task of military news chapters; S6: Comprehensive optimization processing of event arguments, performing comprehensive optimization processing on the extracted event arguments; S7: Structured output and application: output the event arguments after comprehensive optimization processing in a structured form.

2. The method for extracting fine-grained events at the chapter level in the field of military news according to claim 1 is characterized in that: The pre-processing in step S1 comprises the following steps: S11: Use Chinese word segmentation tools to segment military news articles into individual word units; S12: Perform part-of-speech tagging to mark the part-of-speech information of each word; S13: Build a vocabulary, summarize all the words that appear in the article, remove duplicate words, and give each word a unique identifier.

3. The method for extracting fine-grained events at the chapter level in the field of military news according to claim 1 is characterized in that: The event detection technology of the event detection model in step S2 includes: S21: event trigger word recognition stage, in which event trigger words are recognized through MRC question-answering templates; S22: Event type classification stage, in which the event type is accurately classified based on the identified trigger words.

4. The method for extracting fine-grained events at the chapter level in the field of military news according to claim 1 is characterized in that: The event argument extraction process in step S3 includes: S31: Manual template, all roles are manually connected using natural language; S32: Soft hint templates that combine different argument roles with role-specific pseudo-tokens learnable by a pre-trained language model; S33: Based on the connection template, all role names belonging to an event type are connected using natural language.

5. The method for extracting fine-grained events at the chapter level in the field of military news according to claim 1 is characterized in that: In step S3, the PAIE event argument extraction model extracts event arguments by creating a prompt template, an event argument selection decoder part, and event argument prediction.

6. The method for extracting fine-grained events at the chapter level in the field of military news according to claim 4 is characterized in that: The event argument prediction is combined with bipartite graph matching and reasoning to solve the problem of optimal allocation of multiple similar role arguments in the same event: By introducing bipartite graph matching technology, the Hungarian algorithm is used to insert multiple slots in the model template setting to achieve the global optimal allocation of matches with minimum cost. The reasoning stage defines the candidate event argument set and scores it; For inference, the candidate event argument set is defined as: The candidate event argument set contains all spans shorter than the threshold L and the special span (0,0) from which no event arguments are extracted. The model extracts each span selector θ by enumerating all candidate spans and scoring them. k The arguments of and score them to extract each span selector θ k The arguments: And the prediction span for time slot k is given by:

7. The method for extracting fine-grained events at the chapter level in the field of military news according to claim 1 is characterized in that: The extraction of fine-grained event arguments in step S4 includes the following steps: S41: Text analysis, conduct detailed sentence-by-sentence analysis of military news texts to clarify the semantics and structure of each sentence; S42: Trigger word location: in the analyzed text, accurately locate each event trigger word and accurately identify these key words through a model based on machine reading comprehension; S43: Argument association: for each located event trigger word, associate the related fine-grained event arguments according to its semantics and context information; S44: Interaction mining, in the process of associating arguments, further explore the complex interactions between trigger words, event types and arguments; S45: Long-distance dependency capture, through the overall understanding of the text and grasp of the context, capture the long-distance dependency relationship between event arguments in long texts.

8. The method for extracting fine-grained events at the chapter level in the field of military news according to claim 1 is characterized in that: The application of transfer learning in step S5 includes the following steps: S51: Select a pre-trained language model. According to the characteristics and requirements of the military news chapter, select a suitable pre-trained language model; S52: Model parameter migration preparation, obtaining the parameters of the pre-trained language model trained on other large-scale text data; S53: Adapting military news data, adapting the selected pre-trained language model to adapt it to the specific domain and language style of military news articles; S54: Parameter migration and fine-tuning: Migrate the parameters trained on other large-scale text data to the adapted military news chapter event extraction model, and then fine-tune the model using the annotated data of military news chapters; S55: Evaluation and optimization: Evaluate the military news chapter event extraction model after transfer learning, evaluate the extraction effect of the model by comparing it with manually annotated data, and further optimize and adjust the model based on the evaluation results.

9. The method for extracting fine-grained events at the chapter level in the field of military news according to claim 1 is characterized in that: The comprehensive optimization processing of event arguments in step S6 includes the following steps: S61: De-duplication processing, carefully checking all extracted event arguments, using specific algorithms and rules to find and remove repeated event arguments; S62: Error correction, each event argument is carefully reviewed and the incorrect classification is corrected; S63: semantic consistency check, check and correct the semantic consistency between event arguments; S64: Quality assessment and screening: Establish quality assessment standards and conduct quality assessment on event arguments after deduplication and error correction.

10. The method for extracting fine-grained events at the chapter level in the field of military news according to claim 1 is characterized in that: The structured output and application in step S7 includes the following steps: S71: Determine the structured format and specify the structured output format to be adopted; S72: field extraction and filling, extracting the event type, trigger word and information of each argument in turn from the event arguments after comprehensive optimization processing; S73: Organize the output sequence, and determine the output sequence of event arguments according to specific application requirements and logic; S74: Connect with the application system to connect the structured output event arguments with the specific application system.

Images