Text processing method and device, electronic equipment, medium and program product

Abstract

The invention provides a text processing method and device, electronic equipment, a medium and a program product, and can be applied to the technical field of big data and the technical field of artificial intelligence. The method comprises the steps of obtaining a to-be-processed unstructured text; performing event abstract classification on the unstructured text to obtain a plurality of semantic event units; performing modular analysis on the semantic event unit to obtain corresponding module attribute information; and performing structured conversion and format verification based on the module attribute information to obtain target structured data corresponding to the unstructured text.

Description

Technical Field

[0001] This application relates to the fields of big data technology and artificial intelligence technology, and more specifically to a text processing method, apparatus, device, medium and program product. Background Technology

[0002] As the digital transformation of the financial industry continues to deepen, financial institutions have accumulated a large amount of unstructured text data in scenarios such as risk management, compliance review, contract analysis, and customer profiling. They need to use efficient structuring technologies to extract information and transform its value.

[0003] Currently, existing processing methods mainly include direct conversion based on large-scale language models, parsing based on manual rule engines, and hybrid solutions combining rules and statistical learning. While direct conversion based on language models offers some automation, it often results in fabricated content that does not match the original text due to a lack of deep understanding and verification mechanisms for specialized financial semantics. Rule engine methods rely on manually written parsing rules, resulting in a large number of rules that are complex to maintain, leading to a significant performance drop in large-scale business scenarios. Hybrid methods, while combining models and rules, still require manual template design, lacking flexibility and failing to meet regulatory and auditing requirements in complex credit texts and scenarios with multiple semantic intersections. Summary of the Invention

[0004] In view of the above problems, this application provides text processing methods, apparatus, devices, media and program products.

[0005] According to a first aspect of this application, a text processing method is provided, the method comprising: acquiring unstructured text to be processed; performing event summary classification on the unstructured text to obtain multiple semantic event units; performing modular analysis on the semantic event units to obtain corresponding module attribute information; and performing structured transformation and format verification based on the module attribute information to obtain target structured data corresponding to the unstructured text.

[0006] According to embodiments of this application, the classification of event summaries for the unstructured text includes:

[0007] Based on a preset prompt template, the large model is guided to extract a preliminary event summary of the unstructured text according to the sequence number identification rule; and the preliminary event summary is delineated based on contextual logical relationship or semantic similarity to obtain the semantic event unit.

[0008] According to embodiments of this application, the method further includes: performing structural and content checks on the semantic event unit; and, in response to detecting an abnormal unit in the semantic event unit, performing local regeneration or local content correction on the abnormal unit.

[0009] According to an embodiment of this application, the modular analysis of the semantic event unit to obtain the corresponding module attribute information includes: obtaining a module template based on a preset module configuration file, wherein the module configuration file is used to define the module template and to define the mapping relationship between the module name and attribute fields; and matching the corresponding module template based on the semantic identifier information of the semantic event unit, and filling the attribute fields according to the mapping relationship to generate the module attribute information.

[0010] According to an embodiment of this application, the method further includes: performing association analysis on the module attribute information based on the temporal order or participating entity relationship of the semantic event units; identifying attribute conflicts between different semantic event units based on the association analysis results; and adjusting the corresponding module attribute information in response to the identification of attribute conflicts.

[0011] According to an embodiment of this application, the structured conversion based on the module attribute information includes: parsing an intermediate data file containing the module attribute information to extract key-value pairs corresponding to the module attribute information; and converting the key-value pairs into structured data based on a preset mapping configuration file.

[0012] According to an embodiment of this application, the format verification includes: performing format verification on the structured data based on preset structured data verification rules; locating target module attribute information in response to a failed format verification; and regenerating or structurally adjusting the target module attribute information until the format verification is passed, thereby obtaining the target structured data.

[0013] According to an embodiment of this application, the step of matching the corresponding module template based on the semantic identifier information of the semantic event unit and filling the attribute fields according to the mapping relationship to generate the module attribute information includes: matching the corresponding module template based on the event type or keyword of the semantic event unit; in response to no matching module template, obtaining the added or modified module template and the updated mapping relationship; and generating the module attribute information using the updated mapping relationship.

[0014] A second aspect of this application provides a text processing apparatus, the apparatus comprising: a data acquisition module for acquiring unstructured text to be processed; an event summary classification module for performing event summary classification on the unstructured text to obtain multiple semantic event units; a modular analysis module for performing modular analysis on the semantic event units to obtain corresponding module attribute information; and a structured conversion and format verification module for performing structured conversion and format verification based on the module attribute information to obtain target structured data corresponding to the unstructured text.

[0015] A third aspect of this application provides an electronic device comprising: one or more processors; and a memory for storing one or more computer programs, wherein the one or more processors execute the one or more computer programs to implement the steps of the method described above.

[0016] A fourth aspect of this application also provides a computer-readable storage medium having a computer program or instructions stored thereon, which, when executed by a processor, implement the steps of the above-described method.

[0017] The fifth aspect of this application also provides a computer program product, including a computer program or instructions that, when executed by a processor, implement the steps of the above-described method.

[0018] According to embodiments of this application, by dividing the processing of unstructured text into multiple stages such as event summary classification, modular analysis, and structured transformation and format verification, step-by-step controllability and process traceability of text structure generation are achieved. This embodiment can automatically identify multiple semantic event units in complex semantic text and generate corresponding attribute information based on modular configuration, thereby ensuring the accuracy and hierarchical clarity of data extraction. Simultaneously, by performing format verification and automatic correction after structured transformation, the generation error rate is reduced, and the stability and consistency of the system are improved. Embodiments of this application not only improve the accuracy and reliability of converting unstructured text to structured data but also possess good scalability and versatility, and can be widely applied to automated analysis and knowledge extraction scenarios for various types of text. Attached Figure Description

[0019] The above-mentioned contents, other objects, features and advantages of this application will become clearer from the following description of embodiments with reference to the accompanying drawings, in which:

[0020] Figure 1 The illustrations depict application scenarios of text processing methods, apparatuses, devices, media, and program products according to embodiments of this application.

[0021] Figure 2 A flowchart illustrating a text processing method according to an embodiment of this application is shown schematically.

[0022] Figure 3 A flowchart illustrating a method for performing modular analysis according to some exemplary embodiments of this application is shown schematically;

[0023] Figure 4 The illustration shows a schematic diagram of a method for structural transformation based on module attribute information according to some exemplary embodiments of this application;

[0024] Figure 5 A schematic diagram illustrating the structure of a text processing apparatus according to an embodiment of this application is shown; and

[0025] Figure 6 A block diagram schematically illustrates an electronic device suitable for implementing a text processing method according to an embodiment of this application. Detailed Implementation

[0026] The embodiments of this application will now be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of this application. In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the embodiments of this application for ease of explanation. However, it will be apparent that one or more embodiments may be implemented without these specific details. Furthermore, descriptions of well-known structures and technologies are omitted in the following description to avoid unnecessarily obscuring the concepts of this application.

[0027] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of this application. The terms “comprising,” “including,” etc., as used herein indicate the presence of the stated features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0028] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein are to be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0029] When using expressions such as "at least one of A, B and C", they should generally be interpreted in accordance with the meaning that is commonly understood by those skilled in the art (e.g., "a system having at least one of A, B and C" should include, but is not limited to, a system having A alone, a system having B alone, a system having C alone, a system having A and B, a system having A and C, a system having B and C, and / or a system having A, B and C, etc.).

[0030] As the digital transformation of the financial industry deepens, financial institutions have accumulated a massive amount of unstructured text data in scenarios such as risk management, compliance review, contract analysis, and customer profiling. This data comes from complex sources and is diverse in content, including credit contracts, review reports, transaction records, customer communication records, regulatory documents, and internal decision-making documents. Due to the high compliance, accuracy, and timeliness characteristics of financial business scenarios, it is necessary to efficiently and accurately extract key elements from these unstructured texts and transform them into computable and analyzable structured data.

[0031] Currently, the mainstream technical approaches to financial text structuring can be broadly categorized into three types: direct conversion methods based on large-scale language models, parsing methods based on human-engineered rules, and hybrid methods combining rules and statistical learning. Each of these approaches has its own advantages and disadvantages in terms of automation, scalability, and accuracy, but they still have significant limitations in handling complex semantics, adapting to multiple domains, and ensuring traceability and control.

[0032] The first category is direct conversion methods based on large-scale language models. This method relies on a general-purpose language model to achieve end-to-end conversion from raw text to structured data through a single prompt word, such as directly outputting tabular or annotated results. Its advantages lie in its simplicity and flexible deployment, enabling rapid tasks such as text summarization, feature extraction, and entity recognition even without specialized rule configuration. In some financial scenarios, such as compliance announcement summarization and internal report generation, this approach can indeed significantly improve information extraction efficiency. However, this method has a fundamental problem: the general-purpose language model lacks a deep understanding of financial domain knowledge. Especially in texts involving complex logic, numerical calculations, and cross-sentence dependencies, the model is prone to generating fictitious content or semantic illusions—that is, the generated results appear reasonable but are factually incorrect. Due to the lack of intermediate verification mechanisms and interpretable processes, such errors are often difficult to detect, leading to discrepancies between the generated structured results and the original data, thereby affecting the reliability of critical business decisions such as downstream risk assessment and compliance review.

[0033] The second category is parsing methods based on human-developed rule engines. This type of solution drives a rule engine to perform structured extraction through expert-defined text parsing rules (such as regular expressions, syntax tree matching, or keyword templates). This method was widely adopted in the early stages of informatization in the financial industry, particularly suitable for scenarios with standardized formats and fixed semantics, such as parsing credit contract terms, entering accounting vouchers, and standardizing audit reports. Its advantages include strong controllability, stable output, and ease of audit tracing. However, with increasing business complexity and changing regulatory requirements, systems based on static rules have revealed significant shortcomings. Each time a new contract template, report style, or regulatory format is added, the rule base must be manually maintained and updated. After hundreds or thousands of rules are added, system performance significantly declines, and loading and execution efficiency drops drastically. More seriously, lagging rule updates struggle to cope with rapid changes in business models and regulatory policies, leading to exponentially increasing maintenance costs. Manually written rules are often limited to surface feature matching, lacking semantic understanding capabilities and struggling to handle complex logical relationships across sentences and paragraphs, thus limiting the adaptability of this solution in dynamic financial environments.

[0034] The third category is a hybrid approach combining rule-based and statistical learning. This approach integrates the controllability of rules with the adaptability of machine learning models, typically employing a two-stage structure of rule preprocessing and model recognition. The system first extracts potentially relevant fragments through rule engines or keyword filtering, then uses statistical or neural network models (such as sequence labeling models and context encoding models) to identify entities and relationships, generating semi-structured results. This type of method performs well in tasks such as customer risk identification, balancing accuracy and versatility. However, it still heavily relies on manually designed rules and templates, lacking flexibility. Especially in financial scenarios, due to the diverse formats of contract terms, transaction descriptions, and review opinions, hybrid methods often require designing specific preprocessing rules and model structures for each document type, leading to high system maintenance complexity. Furthermore, in credit texts with multiple forms and overlapping semantics, the accuracy of this method still falls short of regulatory-grade application requirements. For example, in credit approval documents, the semantics of a field may span multiple sentences or table cells, making it difficult for traditional hybrid approaches to correctly understand their contextual dependencies.

[0035] In addition to the inherent limitations of the three schemes mentioned above, existing financial text structuring technologies also generally suffer from the following common problems:

[0036] First, there is a lack of intermediate result management and verification mechanisms. Most systems directly output the final structured results without explicitly recording and verifying the intermediate extraction, identification, and transformation processes. Once the output results deviate, it is difficult to trace the source error and to effectively correct errors or optimize the model.

[0037] Secondly, the output format is rigid and cannot flexibly adapt to the structural requirements of different business systems. Traditional solutions are often bound to fixed templates or database table structures, making it difficult to dynamically adjust the output format according to different task scenarios.

[0038] Third, there is a lack of a unified scheduling and feedback mechanism between the rule system and the model modules. Currently, most systems use the rule engine and machine learning model separately, with rules only playing a static filtering role and unable to achieve self-optimization based on model feedback, resulting in a lack of closed-loop control in the entire structuring process.

[0039] Based on this, embodiments of this application provide a text processing method, the method comprising: acquiring unstructured text to be processed; performing event summary classification on the unstructured text to obtain multiple semantic event units; performing modular analysis on the semantic event units to obtain corresponding module attribute information; and performing structured conversion and format verification based on the module attribute information to obtain target structured data corresponding to the unstructured text. According to embodiments of this application, by dividing the unstructured text processing into multiple stages such as event summary classification, modular analysis, and structured conversion and format verification, step-by-step controllability and process traceability of text structure generation are achieved. This embodiment can automatically identify multiple semantic event units in complex semantic text and generate corresponding attribute information based on modular configuration, thereby ensuring the accuracy and hierarchical clarity of data extraction; simultaneously, by performing format verification and automatic correction after structured conversion, the generation error rate is reduced, and the stability and consistency of the system are improved. Embodiments of this application not only improve the accuracy and reliability of converting unstructured text to structured data, but also possess good scalability and versatility, and can be widely applied to automated analysis and knowledge extraction scenarios of various types of text.

[0040] It should be noted that the text processing methods, apparatus, devices, media, and program products defined in this application can be used in the fields of big data technology and artificial intelligence technology, as well as in the field of fintech, and in a variety of other fields besides big data technology, artificial intelligence technology, and fintech. The application fields of the text processing methods, apparatus, devices, media, and program products provided in the embodiments of this application are not limited.

[0041] In the technical solution of this application, the user information (including but not limited to user personal information, user image information, user device information, such as location information) and data (including but not limited to data used for analysis, stored data, and displayed data) involved are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, storage, use, processing, transmission, provision, disclosure, and application of related data all comply with relevant laws, regulations, and standards, take necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation entry points for users to choose to authorize or refuse.

[0042] In scenarios where personal information is used for automated decision-making, the methods, devices, and systems provided in this application all offer users corresponding operation entry points, allowing them to choose to agree to or reject the automated decision results; if the user chooses to reject, the process proceeds to expert decision-making. Here, "expert decision-making" refers to the decision-making activities of personnel who specialize in a particular field, possess specialized experience, knowledge, and skills, and have reached a certain level of professional expertise.

[0043] Figure 1 The illustrations depict application scenarios of text processing methods, apparatuses, devices, media, and program products according to embodiments of this application.

[0044] like Figure 1 As shown, application scenario 100 according to this embodiment may include a first terminal device 101, a second terminal device 102, a third terminal device 103, a network 104, and a server 105. The network 104 serves as a medium for providing a communication link between the first terminal device 101, the second terminal device 102, the third terminal device 103, and the server 105. The network 104 may include various connection types, such as wired or wireless communication links, or fiber optic cables, etc.

[0045] Users can use the first terminal device 101, the second terminal device 102, and the third terminal device 103 to interact with the server 105 via the network 104 to receive or send messages, etc. Various communication client applications can be installed on the first terminal device 101, the second terminal device 102, and the third terminal device 103, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, social media platform software, etc. (for example only).

[0046] In the embodiments of this application, the first terminal device 101 can be an example of the first device, and the second terminal device 102 and / or the third terminal device 103 can be an example of at least one second device. The first device and the second device can communicate collaboratively through an internal client mechanism to implement the data distribution and rendering logic described in the text processing method.

[0047] In some embodiments, the first device and at least one second device may be different display modules, windows or screens on the same computing terminal (such as a host), or multiple physical devices that work together through a network, such as different client instances deployed on a desktop computer, tablet terminal or mobile device respectively.

[0048] The first terminal device 101, the second terminal device 102, and the third terminal device 103 can be various electronic devices with displays and support web browsing, including but not limited to smart mobile terminals, tablet computers, laptop computers, and desktop computers.

[0049] Server 105 can be a server that provides various services, such as a backend management server that supports websites browsed by users using the first terminal device 101, the second terminal device 102, and the third terminal device 103 (this is just an example). The backend management server can analyze and process data such as received user requests, and feed back the processing results (such as web pages, information, or data obtained or generated according to user requests) to the terminal devices.

[0050] It should be noted that the text processing method provided in this application embodiment can generally be executed by server 105. Correspondingly, the text processing device provided in this application embodiment can generally be located in server 105. The text processing method provided in this application embodiment can also be executed by a server or server cluster that is different from server 105 and capable of communicating with the first terminal device 101, the second terminal device 102, the third terminal device 103, and / or server 105. Correspondingly, the text processing device provided in this application embodiment can also be located in a server or server cluster that is different from server 105 and capable of communicating with the first terminal device 101, the second terminal device 102, the third terminal device 103, and / or server 105.

[0051] It should be understood that Figure 1 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.

[0052] The following will be based on Figure 1 The described scene, through Figures 2-4 The text processing method of the disclosed embodiments will be described in detail.

[0053] Figure 2 A flowchart illustrating a text processing method according to an embodiment of this application is shown schematically.

[0054] like Figure 2 As shown, the text processing method 200 of this embodiment includes operations S210 to S240.

[0055] In operation S210, the unstructured text to be processed is obtained.

[0056] In the embodiments of this application, the acquisition of unstructured text to be processed can be achieved through automatic collection of multi-source text data from within or outside financial institutions. To ensure data integrity and accuracy, the system can simultaneously record data source, timestamp, and file type information during the collection phase, and perform basic character encoding unification and format standardization processing on the original text. For example, for scanned contract text, optical character recognition technology can be used to convert image data into editable text content, providing structured input for subsequent semantic analysis.

[0057] In some embodiments, the system can interface with business process systems to capture unstructured text in real time during document generation. For example, when a credit approval system generates a customer funds review report, the system can automatically input a copy of the report into the text processing module as an input source for subsequent structured analysis, thereby achieving a "generate and structure simultaneously" processing mechanism, reducing manual processing steps, and ensuring the timeliness and consistency of data. For text information from external channels, the system can also collect it synchronously in real time through web crawling engines or message queues.

[0058] In some embodiments, to adapt to the characteristics of different types of text, the system can perform multi-layered preprocessing strategies during the data input stage. For example, for texts with relatively regular formats (such as contract terms and audit reports), the system can perform chapter segmentation and tag recognition based on preset document templates to extract logical units such as titles, clause numbers, and body text; for free texts (such as customer descriptions and manual approval opinions), the system can use natural language processing tools to perform sentence segmentation and paragraphing operations, and identify key information such as dates, amounts, and organization names. Through this type of layered preprocessing, the system can improve the accuracy of semantic unit extraction without changing the meaning of the original text.

[0059] In the embodiments of this application, when acquiring unstructured text, the system can also perform data anonymization and security screening on the text according to business needs. For example, when processing documents containing customer privacy or sensitive financial information, the system can automatically identify sensitive fields such as ID card numbers, bank card numbers, and contact information during the input stage, and process them by masking or replacing identifiers, thereby ensuring the security of subsequent structured analysis and complying with data compliance management requirements.

[0060] In some embodiments, the system can support multilingual text input to meet the document processing needs of cross-border financial transactions or international cooperation projects. When the system detects the language type of the input text, it can automatically load the corresponding language's word segmentation dictionary and semantic model to achieve unified parsing of multilingual text. For multilingual mixed text, the system can also employ sentence-level language recognition and segment-by-segment translation to ensure the accuracy of semantic analysis.

[0061] In some embodiments, acquiring unstructured text may also include extracting text content from files of different formats. The system can support reading and parsing multiple file types. For documents containing tables and embedded images, the system can use layout analysis and table reconstruction algorithms to extract table structure information and save the text content along with the position index for subsequent semantic structure analysis.

[0062] In some embodiments, to improve system performance in large-scale text processing scenarios, the process of acquiring unstructured text may also include a distributed data scheduling and parallel loading mechanism. The system can distribute the text to be processed to different nodes for asynchronous loading and cleaning, and achieve efficient data flow through message queues or caching mechanisms. When the amount of text input is large, the data channels can be dynamically expanded to ensure throughput and stability during the data acquisition phase.

[0063] In some embodiments, the system can also incorporate contextual business metadata to perform contextual annotation during the text input stage. Specifically, when reading text, the system can append information such as the document's business origin (e.g., credit, compliance), generating department, document version number, and approval process nodes. Contextual business metadata can serve as important context for subsequent semantic analysis and structured mapping, enabling the system to automatically adjust its parsing strategy based on differences in different business lines.

[0064] In operation S220, event summarization and classification are performed on the unstructured text to obtain multiple semantic event units.

[0065] In the embodiments of this application, after obtaining unstructured text, event summarization and classification can be performed on the unstructured text to transform long and complex natural language content into multiple semantically complete event units. This process can be based on a pre-trained language model or a self-developed semantic understanding model, and can identify events with independent semantic boundaries in the text through prompt templates. For example, when the input text is a credit approval report, the system can automatically identify independent events such as "customer submits funding application", "risk control department completes preliminary review", and "credit committee approves", and extract them into multiple semantic event units.

[0066] In some embodiments, the system can combine contextual logical relationships and semantic similarity calculations when performing event summary classification to improve the accuracy of event segmentation. Specifically, after initially segmenting the text into sentences, the system calculates the semantic similarity and logical coherence between adjacent sentences. When a significant semantic shift or topic change is detected, that position can be used as the event boundary. For example, for a customer due diligence report, the system can identify the semantic shift point between "customer background investigation" and "financial status assessment," thereby dividing them into two independent events. This effectively reduces event confusion and ensures the semantic independence of event unit segmentation.

[0067] In some embodiments, event summary classification can be aided by combining domain-specific dictionaries and rule templates. For example, based on model predictions, the system can utilize pre-defined financial business dictionaries or rule templates, such as "meeting minutes events," "contract signing events," and "approval operation events," to perform semantic annotation and secondary screening of candidate events. This allows for the introduction of domain knowledge constraints while maintaining model flexibility, making the event classification results more consistent with financial business logic. For instance, when keywords such as "review comments" or "approval decisions" appear in the text, the system can automatically identify that part as an approval event unit, thereby improving the professionalism and consistency of event classification.

[0068] To further improve the quality of event summaries, the system can perform summary generation and tag verification operations after event identification. Specifically, after identifying several semantic event units, the system can provide a brief natural language summary for each event, generated in formats such as "Event 1: Customer submits funding application," "Event 2: Risk control review passed," and "Event 3: Credit approval passed." Simultaneously, the system can use regular expressions or a syntax tree analyzer to verify the legality of the generated tag format, ensuring that each event has correct sequence labels, boundary closure markers, and content legality.

[0069] In another embodiment, the event summary classification process can also incorporate a multi-model collaborative mechanism. The system can call different types of language models or semantic extraction models in parallel to perform multi-perspective analysis on the same text, and then fuse the results of each model based on a voting mechanism or confidence weighting strategy to improve the stability of event recognition. For example, when processing long reports or texts containing multiple descriptions, the main model is responsible for identifying the main event sequence, while the auxiliary model can specifically identify detailed events such as time nodes and changes in participants. The combination of the two can generate a more complete event structure.

[0070] In some embodiments, event summary classification can be validated and optimized by combining it with a visual semantic network approach. After extracting multiple semantic event units, the system can construct a semantic relationship graph between events, using nodes to represent events and edges to represent the temporal or logical relationships between events, thereby displaying the event logic of the entire text in a graph structure. Users or the system can identify potential missing or redundant events by analyzing the density and correlation of nodes in the event network, and perform add, delete, and modify operations through an interactive interface to achieve human-computer collaborative event classification optimization.

[0071] In another embodiment, the system can adopt customized event classification strategies based on different scenario types. For example, for customer service communication records, the system focuses more on event types such as "customer inquiry," "problem handling," and "follow-up feedback"; for contract texts, it mainly identifies events such as "contract signing," "contract modification," and "contract termination."

[0072] In some embodiments, when the input text involves multilingual content, the system can first perform language detection and sentence-level translation, and then uniformly extract semantic events. For multiple documents on the same topic, the system can use semantic clustering to group similar events from different documents into the same semantic event unit, thereby achieving cross-document event integration and unified expression.

[0073] In operation S230, the semantic event unit is subjected to modular analysis to obtain the corresponding module attribute information.

[0074] In the embodiments of this application, after acquiring multiple semantic event units, modular analysis can be performed on these event units to extract key information related to each event and generate module attribute information. For example, this process can be implemented based on a preset module configuration file, which defines the module name, attribute fields, and the mapping relationship between each field and semantic features. The system first identifies the event type to which the semantic event unit belongs, such as "meeting," "product launch," "risk assessment," and "contract signing," and then automatically matches the corresponding module template to classify and fill in the attributes of the elements contained in the event. For example, when a "meeting" type event is detected, the system can automatically extract attribute fields such as meeting time, participants, and discussion topics; when a "contract signing" type event is detected, information such as contract number, signatory, and signing date is extracted. Through this modular analysis, the original event text is transformed into a structured set of module attributes, providing a clear semantic framework for subsequent data transformation.

[0075] In some embodiments, modular analysis can be combined with a dynamic template allocation mechanism to enhance the system's adaptability across multiple scenarios. When analyzing semantic event units, if the system finds that their semantic features cannot fully match existing predefined modules, it can dynamically generate or expand new module templates based on the event content. For example, when a new type of event, such as a "strategic cooperation announcement," appears that was not previously covered, the system can automatically construct a new module template based on key elements appearing in the event (such as partners, scope of cooperation, and signing time) and update the module configuration file.

[0076] In some embodiments, modular analysis can be performed using a hybrid approach combining language models and rule templates. The system first uses the language model to identify entities and attribute relationships within semantic event units, and then calibrates the identification results according to rules in the configuration file. For example, after the system identifies two pieces of information, "Project Start Time: March 2025" and "Responsible Department: Risk Management Department," it can verify the legality and contextual consistency of the fields according to rules, ensuring that the attribute values ​​meet the format and semantic requirements.

[0077] In some embodiments, modular analysis can leverage contextual information to achieve cross-event attribute sharing. When multiple semantic event units are associated in terms of time or participating entities, the system can identify shared attributes during modular analysis and inherit or reference them in relevant modules. For example, for three consecutive events—"customer fund application," "fund approval," and "fund disbursement"—the system can identify common attribute fields such as "customer name" and "fund amount" and automatically reuse this information in subsequent modules, thereby avoiding duplicate extraction and maintaining data consistency.

[0078] In some embodiments, the modular analysis process can also be enhanced semantically by incorporating a business knowledge graph. When generating module attribute information, the system can call upon a built-in knowledge base to perform semantic mapping and relationship completion on the extracted entities and attributes. For example, when the text mentions "customer company A," the system can query the company's registration information, industry category, and historical risk records through the knowledge graph, and supplement this information as extended attribute fields.

[0079] In some embodiments, the system can introduce a confidence evaluation mechanism during the modular analysis phase to label the generated module attribute information with reliability. Specifically, the system calculates the semantic matching confidence score each time an attribute is extracted and stores the confidence score along with the attribute result. When the confidence score is lower than a preset threshold, the system can mark the attribute as "pending confirmation" for manual review or trigger an automatic re-analysis operation.

[0080] In another embodiment, the system can perform attribute consistency checks during modular analysis to ensure logical coordination between attributes of different modules. For example, if the time field extracted from "Event 1" is "March 2025" while the time field extracted from "Event 2" is "December 2024", the system can automatically detect the time sequence anomaly and mark it as a potential conflict, then automatically correct it based on the context logic or prompt for manual review. Similarly, for the role descriptions of the same participating entity in different modules (such as "Party A", "Client", "Applicant"), the system can ensure consistent referencing through semantically unified mapping.

[0081] In another embodiment, modular analysis can also be combined with a visual interface to enable interactive verification and optimization. After generating module attribute information, the system can display each module and its attributes to the user in a table or tree structure. The user can intuitively view the source text, extraction results, and confidence score of each attribute through the interface. When the user modifies or confirms attribute values ​​on the interface, the system can use the feedback results for model optimization and configuration file updates, forming a closed-loop mechanism.

[0082] In operation S240, a structured conversion and format verification are performed based on the module attribute information to obtain the target structured data corresponding to the unstructured text.

[0083] In the embodiments of this application, after obtaining the module attribute information, the attributes can be structurally transformed based on a preset mapping configuration file, and the semantically clear module attribute information can be transformed into machine-processable data objects.

[0084] In some embodiments, a template adaptive mechanism can be introduced during the structured transformation process to support flexible expansion for different business scenarios. When the system detects that the scenario to which the input text belongs does not completely match the existing template, it can automatically and dynamically adjust the output structure based on module attribute information. For example, when processing an investment analysis report, if new fields such as "profit forecast" and "risk preference" appear in the module attributes, the system can expand the hierarchical relationship of the target structured data according to preset rules, embedding the new attributes into the corresponding module nodes, thereby achieving adaptive template updates.

[0085] In some embodiments, to ensure the reliability of the structured data conversion results, the system may introduce a multi-layered validation mechanism during the conversion process. After generating preliminary structured data, the system automatically checks field types, hierarchical relationships, and required fields using structured data validation rules. When format errors, missing fields, or type mismatches are detected, the system can automatically locate the corresponding module attribute information and trigger a backtracking correction process.

[0086] In some embodiments, structured transformation can also be performed in conjunction with industry-standard data models to improve the universality and exchangeability of data. When generating target structured data, the system can map module attributes to fields in industry standards to ensure that the output conforms to standardized interface specifications.

[0087] In some embodiments, to improve system performance in large-scale text processing scenarios, structured transformation can be implemented through a distributed computing architecture. The system can distribute attribute information from different modules across multiple computing nodes for parallel mapping and verification, ultimately merging them at the master node to generate the complete target structured data. When the input text volume is large or the number of modules is numerous, this architecture can shorten processing time and reduce redundant computation overhead through a distributed caching mechanism. Simultaneously, the system can perform global consistency checks during the merging phase to ensure that the data structures generated across nodes are unified and conflict-free.

[0088] In some embodiments, to improve data quality and interpretability, the system can record the source and transformation path of each attribute during the structured transformation stage. Specifically, the system can record metadata such as "source event unit identifier," "original text location," and "transformation rule number" for each piece of structured data, and generate a corresponding mapping table in the final output. Thus, users or system administrators can trace the generation process of any field, with each step from the original text to the final structured result clearly recorded.

[0089] In some embodiments, after the structured conversion is completed, the system can also perform automatic formatting and persistent storage operations. The system can select the output format according to business needs and automatically perform formatting, encoding standardization, and storage path generation.

[0090] In some embodiments, the system can also be combined with a visualization verification module to intuitively display and manually review the transformation results of structured data. The system can generate a hierarchical view of the structured data, presenting each module and its attributes in a tree structure. Users can click on any field to view its source, confidence level, and corresponding text fragment. When users confirm or modify attribute values ​​on the interface, the system can automatically update the corresponding structured data and record a revision log, thereby achieving human-machine collaborative verification and optimization.

[0091] Furthermore, to achieve continuous optimization, the system can feed back the format validation results to the model training and rule configuration stages. The error types and correction results found in each validation will be recorded in the system's knowledge base, which will be used to update field mapping rules or adjust the model's prompt templates, enabling the system to self-improve when processing similar text.

[0092] According to embodiments of this application, a complete text structuring process is constructed by dividing the unstructured text processing into consecutive steps such as input text acquisition, event summary classification, modular analysis, and structured transformation and verification. This embodiment can automatically extract valuable information units from multi-source, multi-format raw text and achieve logical segmentation of content through semantic event classification, enabling the accurate decomposition of long and complex texts. Subsequently, the system generates attributed semantic structures based on modular analysis, allowing key information of different event types to be mapped and managed within a unified framework, significantly improving the clarity and scalability of data organization. Through structured transformation and format verification, the system further ensures that the output results conform to preset standards and hierarchical logic, achieving high-precision transformation from natural language text to computable and storable data. The entire process forms a closed-loop mechanism from semantic extraction to structure verification, supporting error backtracking, dynamic correction, and template adaptation, effectively reducing illusions and format errors during the generation process.

[0093] The text processing method of this application will be specifically described below by way of preferred embodiments.

[0094] In the embodiments of this application, the process of performing event summary classification on unstructured text can be completed based on a preset prompt template to guide the large model to extract preliminary event summaries according to the sequence identification rules. Specifically, a prompt template can be loaded first, which is used to standardize the output format and semantic organization of the large model when processing input text. The template can include instructional statements, such as "Please summarize the following text content in the order of events, and identify each event with a sequence number (such as event 1, event 2, etc.)," ​​and require the model to maintain label closure and semantic integrity in the output. In this way, the system can constrain the output structure during the model generation stage to avoid semantic confusion or format errors. For example, when the input text is "The company holds an annual meeting and releases new products," the system can generate a preliminary event summary of "Event 1: Holding an annual meeting; Event 2: Releasing new products" under the guidance of the template, thereby realizing the semantic decomposition and orderly expression of the original text.

[0095] After obtaining a preliminary event summary, the system can further delineate the boundaries of the summary content based on contextual logical relationships or semantic similarity to generate semantically complete event units. This process can be implemented through a semantic similarity calculation model or a context dependency analysis module. The system performs quantitative analysis on the semantic similarity between adjacent event summaries. When a significant semantic shift, time change, or topic switch is detected, that position can be used as the event boundary, thereby ensuring that event units are semantically independent and complete. For example, when processing credit review reports, the system can distinguish between two logically different behavioral events: "customer submits application materials" and "risk department completes preliminary review," avoiding conflating the two.

[0096] In the embodiments of this application, to further improve the accuracy of event classification, the system can combine semantic dependency analysis and contextual reasoning algorithms to perform secondary optimization on the preliminary summary. When the system detects causal or logical subordinate relationships between certain event summaries, it can automatically merge or split them to maintain the rationality of the event hierarchy. For example, when the original text contains "The company held a meeting, during which it decided to establish a new subsidiary," the system can divide it into two event units based on semantic dependency relationships: "held a meeting" and "decided to establish a subsidiary," thereby more accurately reflecting the business logic contained in the text.

[0097] Furthermore, a tag verification mechanism can be introduced during the event summary classification process to check the format and logical consistency of the generated event summary results. The system can check the continuity, closure, and content validity of the sequence number tags through regular expressions or a syntax parser. If incorrect tags or missing sequence numbers are detected, the system can automatically trigger a regeneration process to ensure that the output event unit set meets both semantic and structural specifications.

[0098] According to the embodiments of this application, while maintaining a high degree of automation, the accuracy and stability of event segmentation can be guaranteed, providing a reliable semantic foundation for the subsequent modular analysis stage.

[0099] In the embodiments of this application, after generating the semantic event units, structural and content checks can be performed on the semantic event units to ensure the completeness and semantic accuracy of the event segmentation. Structural checks are mainly used to verify the correctness of the event units at the format level, such as checking whether event numbers are consecutive, whether tags are paired and closed, and whether event boundaries overlap or are missing. For example, the structure of events can be automatically reviewed through regular expression matching, syntax tree parsing, or template comparison. If format abnormalities are found, such as missing tags or skipped numbers, the system will mark the corresponding event as a "structurally abnormal unit" for further processing.

[0100] In the embodiments of this application, content verification is used to evaluate the semantic rationality and consistency of event units. The system can utilize techniques such as semantic similarity calculation, contextual consistency analysis, or entity recognition verification to detect the logical relationship between the content and context of each event unit. For example, when there is an obvious time reversal or causal conflict between two adjacent events, such as "Event 1: Project approval completed; Event 2: Project application submitted," the system can determine that the logical order is abnormal and mark the unit as a "semantically abnormal unit." In addition, content verification can also verify whether the entities mentioned in the events are consistent, such as "company name" and "participant identity," to prevent semantic drift or entity misuse.

[0101] When the system detects an anomalous unit within a semantic event unit, it can trigger local regeneration or local content correction operations to fix the anomalous event without affecting the overall structure. Local regeneration can re-analyze and extract the original text fragment corresponding to the anomalous unit by calling the language model or pre-trained event generation module, thereby generating a new event description that conforms to semantic and format specifications. Local content correction is suitable for cases with minor deviations, such as missing labels, misaligned fields, or ambiguous expressions. The system can fine-tune and complete the original event content based on contextual information to maintain the continuity of event logic and semantic integrity.

[0102] In some embodiments, the system can re-perform consistency verification after performing local corrections to ensure that the corrected event units are consistent with the overall semantic sequence. If the correction result still does not meet the preset quality standard, the system can trigger a manual review mechanism or record relevant information in the error log for subsequent model optimization or rule iteration. By introducing structural verification, content verification, and automatic repair mechanisms, the system achieves closed-loop quality control in the event summarization stage, which not only reduces the risk of error propagation in semantic segmentation but also improves the reliability and maintainability of the text structuring process.

[0103] In the embodiments of this application, after the event summary is classified, structured semantic modeling can be performed on multiple semantic event units to generate module attribute information, thereby providing a clear semantic framework and attribute hierarchy for subsequent data conversion and format verification.

[0104] Figure 3 A flowchart illustrating a method for performing modular analysis according to some exemplary embodiments of this application is shown schematically.

[0105] like Figure 3 As shown, the method for performing modular analysis includes operations S310 to S320.

[0106] In operation S310, a module template is obtained based on a preset module configuration file. The module configuration file is used to define the module template and to define the mapping relationship between the module name and attribute fields.

[0107] In embodiments of this application, module configuration files can be stored in a system knowledge base to define the names, attribute fields, and hierarchical relationships between different types of modules. For example, for financial text, module templates such as "Customer Information Module," "Risk Assessment Module," and "Approval Conclusion Module" can be defined. Each module template contains multiple attribute fields, such as "Customer Name," "Credit Rating," and "Approval Result."

[0108] In operation S320, the corresponding module template is matched based on the semantic identifier information of the semantic event unit, and the attribute fields are filled according to the mapping relationship to generate the module attribute information.

[0109] In embodiments of this application, semantic identification information can be generated during the preceding event summary stage, including event categories, keywords, or contextual semantic tags. The system uses this semantic identification information to find the most matching module template in the configuration file and fills the corresponding fields with the entities and attribute values ​​identified in the event unit. For example, when the event unit is "Risk Management Department completes customer credit assessment," it can automatically match "Risk Assessment Module," filling "Risk Management Department" into the "Execution Department" field and "Customer Credit Assessment" into the "Assessment Content" field. In this way, the system achieves accurate mapping from natural language events to structured attribute data.

[0110] Compared to existing technologies, the modular analysis method provided in this embodiment enables automatic matching of semantic event units with predefined module templates, allowing the system to perform structured decomposition of complex text without manual intervention. Through parameterized definitions in module configuration files, the module structure can be flexibly extended for different business scenarios, avoiding the limitations caused by fixed templates. Simultaneously, the system introduces a mapping relationship verification mechanism when generating module attribute information, ensuring that the attribute filling process conforms to logical consistency and format specifications, thereby improving the stability and universality of the structured results.

[0111] In the embodiments of this application, after completing the modular analysis of each semantic event unit and generating module attribute information, the system can further perform association analysis on the module attribute information based on the temporal order or participating entity relationship of the semantic event units to ensure semantic consistency and logical coordination between different event units.

[0112] In some embodiments, association analysis can be achieved by constructing a time-series model or semantic association graph between events. The system uses timestamps, entity identifiers, and event types as key nodes to sort and compare multiple event units. For example, when the text involves three events: "customer submits funding application," "risk control department completes review," and "funds successfully disbursed," the system can automatically determine that the three are consecutive events on the same business chain based on the consistency of event time and the entity "customer," thereby establishing a temporal association relationship between the events.

[0113] In another embodiment, the system can identify attribute conflicts between different semantic event units based on the aforementioned association analysis results. Attribute conflicts can manifest as inconsistent attribute values ​​for the same entity in different events, logically contradictory events occurring within the same time period, or information duplication due to event overlap. For example, when "Event 1: Customer credit rating is A" and "Event 2: Risk assessment result is B" both point to the same customer, the system can identify this inconsistency through entity matching and attribute comparison, and mark it as an attribute conflict. For time-related conflicts, the system can also detect abnormal event sequences, such as "Approval completed" appearing before "Application submitted," automatically determining it as a logical conflict. Through the detection process, the system can discover potential contradictions in structured data, ensuring the logical consistency of the event chain and the authenticity of the data.

[0114] When the system detects an attribute conflict, it can automatically trigger an adjustment mechanism to correct the corresponding module attribute information, eliminate the conflict, and restore semantic consistency. The adjustment methods may include strategies such as local recalculation, attribute inheritance, or weighted fusion. For example, when there are multiple values for the same attribute in different events, the system can determine the optimal value based on the time of occurrence, data source reliability, or confidence score and update it to the main module attribute; in the case of logical conflicts, the system can trace back to the original text or upstream event unit, re-extract relevant attributes, and perform consistency verification. This automatic adjustment mechanism can achieve self-error correction and optimization of data without affecting the overall structure, thereby enhancing the stability and credibility of the structured results.

[0115] Figure 4 FIG. schematically shows a schematic diagram of a method for structured conversion based on module attribute information according to some exemplary embodiments of the present application.

[0116] As Figure 4 shown, the method for structured conversion based on module attribute information may include operations S410 to S420.

[0117] In operation S410, parse the intermediate data file containing the module attribute information to extract the key-value pairs corresponding to the module attribute information.

[0118] In operation S420, based on a preset mapping configuration file, convert the key-value pairs into structured data.

[0119] In an embodiment of the present application, after generating the module attribute information, the intermediate data file containing this information can be parsed to extract the key-value pairs corresponding to the module attributes. Among them, the intermediate data file can be a semi-structured file output in the event analysis stage. The system can read the file content line by line through a parser, identify the module tags, attribute names, and their corresponding attribute values, and convert them into a set of key-value pairs according to the preset field extraction rules. This parsing process not only retains the hierarchical relationship between modules but also can extract meta-information related to attributes, such as data source, confidence score, and the event number to which it belongs. For example, when the intermediate file contains content such as "<module name>Risk Assessment< / module name><attribute name>Credit Rating< / attribute name><attribute value>A< / attribute value>", the system can automatically extract the key-value pair {"Credit Rating": "A"} and map it to the corresponding module node to provide the basic data for subsequent structured conversion.

[0120] In some embodiments, to ensure the completeness and accuracy of parsing, the system can employ a multi-layered verification mechanism to perform structural verification on intermediate data files. Specifically, it can check label closure, field completeness, and hierarchical nesting relationships. When missing labels, duplicate fields, or invalid nodes are detected, an error correction or regeneration mechanism can be triggered. Furthermore, a multi-file parallel parsing mode can be supported, simultaneously parsing and extracting key-value pairs from intermediate data files from different event modules, and merging them into a unified set of key-value pairs using a unique identifier after parsing.

[0121] After extracting the key-value pairs corresponding to the module attribute information, these pairs can be converted into structured data based on a preset mapping configuration file. The mapping configuration file defines the correspondence between attribute fields and fields in the target data model, as well as the conversion rules, such as data type constraints, field hierarchy, and unit conversions. After reading the mapping configuration file, the system can perform field mapping and format conversion operations to populate the original key-value pairs into the target data structure.

[0122] To enhance data standardization, the system can reference industry or domain-standard data models during the mapping and transformation process. Through standardized mapping, the structured data generated by the system can be directly interfaced with other business systems or regulatory interfaces, achieving cross-system data sharing and interoperability.

[0123] Through the above methods, the system can efficiently extract module attribute information from intermediate data files and accurately convert it into structured data, realizing the mapping from semantic attributes to standardized data models, thereby improving the automation and consistency of data structuring.

[0124] In the embodiments of this application, after generating structured data, the system can perform format validation on the structured data to ensure that the output results meet the preset structural specifications and data integrity requirements. Format validation can be performed according to the system's built-in structured data validation rules, which define field types, hierarchical relationships, required fields, value ranges, and logical constraints between fields. During the validation process, the system can check the generated structured data item by item to confirm that the type of each field is consistent with the template definition, the hierarchical nesting relationship is correct, and all required fields are effectively filled.

[0125] When the system detects that structured data has failed validation, it can locate the corresponding target module attribute information based on the validation report. By establishing a mapping index between structured data and module attributes, the system can accurately trace the source attribute field that caused the anomaly. For example, when the "Risk Rating" field is found to have an incorrect type or a missing value, the system can locate the "Risk Assessment Module" that generated the field and its corresponding attribute information based on the index. After locating the error, the system can further analyze the cause of the error, such as inconsistent mapping rules, missing original attributes, or template filling errors, and take different repair strategies according to the anomaly type.

[0126] In some embodiments, after locating abnormal module attribute information, the system can trigger an automatic repair mechanism to regenerate or restructure the attribute information until it passes format verification. Regeneration can be achieved by calling a language model or attribute extraction algorithm to re-extract the correct value of the attribute from the original text or intermediate file; while structural adjustment is applicable to cases where field names, hierarchical positions, or data types are incorrect. The system can automatically correct the field structure and reorganize the hierarchical relationships according to the mapping configuration file. After the repair is completed, the system re-executes format verification. If the verification result still fails, iterative correction can continue until the verification rules are met. Through this cyclical verification and repair mechanism, the system ensures that the final generated target structured data meets the expected format requirements and data quality standards.

[0127] Furthermore, to enhance the flexibility and scalability of verification, the system also allows users to customize verification rules or select industry-standard rule sets. For example, in financial application scenarios, users can load field verification templates that comply with regulatory reporting standards, enabling the system to automatically check whether field naming conventions, date formats, and encoding methods meet regulatory requirements during format verification.

[0128] Through the above implementation methods, the system introduces format verification and automatic repair steps after structure conversion, forming a quality control system from structure generation to verification closed loop, thereby improving the stability and reliability of structured data output and ensuring that the final generated target structured data meets business specifications, format standards and cross-system compatibility requirements.

[0129] In the embodiments of this application, after identifying the semantic event unit, the system can match the corresponding module template based on the semantic identifier information of the semantic event unit, and fill the attribute fields according to the mapping relationship to generate module attribute information. Specifically, the system first performs semantic recognition and classification on the semantic event unit, extracting event type or keyword information, such as "meeting", "product launch", "strategic cooperation", "risk assessment", etc. The system compares the extracted semantic features with the module templates defined in the module configuration file, determines the most suitable module structure based on the matching result, and automatically loads the corresponding template. For example, when the keyword "product launch" is identified in the semantic event unit, the system will match the "product launch module template", and extract and fill the attributes of the text content according to the attribute fields defined in the template (such as product name, release time, release channel, etc.), and finally generate module attribute information corresponding to the event semantics.

[0130] In some embodiments, when the system fails to match a suitable module template in the preset module configuration file, it can automatically trigger a module extension mechanism to dynamically add or modify templates. The system can generate new module templates or supplement existing ones based on the contextual semantic features and identified attributes of semantic event units. For example, when the system detects that the input text contains the previously undefined event type "strategic alliance signing," it can automatically create a "strategic cooperation module template" based on the semantic extraction results and add attribute fields such as "partner name," "signing date," and "scope of cooperation." Simultaneously, the system updates the mapping relationships in the module configuration file to ensure the new template remains compatible with the overall system structure, thereby guaranteeing that subsequent events can directly reuse the template for attribute filling.

[0131] In another embodiment, in addition to automatically triggering the module extension mechanism, the system also allows users to manually add or modify module templates through configuration files, enabling flexible customization for specific business scenarios. Specifically, a module configuration file editing interface can be provided, allowing users to directly add module templates or adjust the attribute fields, mapping relationships, and hierarchical structure of existing templates according to business needs.

[0132] When modifying existing modules, users can also update template content based on adjustments to business processes. For example, if the "Approval Result Module" originally contained the attribute fields "Approval Status" and "Approver," users can add "Approval Time" and "Approval Remarks" fields through the configuration file, thus ensuring the module structure remains consistent with the latest business logic. When loading the updated configuration file, the system automatically re-parses the template structure and synchronizes the mapping relationships, ensuring that the reference relationships and data structures between the modified module template and other modules are compatible.

[0133] After obtaining the updated module template and mapping relationships, the system can regenerate module attribute information using these relationships. By reloading the updated configuration file, the system performs attribute extraction, field binding, and hierarchical mapping operations on unmatched event units. For example, after dynamic template expansion, the system can re-analyze previously unrecognized "supply chain cooperation" events, extracting their main attribute fields (such as "cooperating companies," "cooperating products," and "cooperation period") and generating module attribute information that conforms to the new template structure. This process can be completed automatically without interrupting the overall analysis workflow, ensuring the system's adaptability and scalability in complex semantic environments.

[0134] Furthermore, to prevent structural conflicts or mapping confusion caused by module template expansion, the system can perform template consistency checks after updating the mapping relationships. Specifically, the system can verify whether the field naming, hierarchical relationships, and data types of the new template are compatible with existing templates, and detect whether there are duplicate field names or logical conflicts. Once the verification is successful, the system will officially register the new template in the module library and update the mapping relationships to the global index, enabling the system to automatically apply the updated module structure in subsequent event processing.

[0135] According to embodiments of this application, intelligent matching and dynamic expansion from semantic event units to module templates are achieved. The system can not only extract attributes within a preset template range, but also automatically generate and update templates based on new semantics, thus forming a closed-loop mechanism of semantic recognition, template matching, template expansion, and attribute generation. This mechanism enhances the intelligence and adaptability of the modular analysis process, enabling the system to continuously adapt to the structured analysis needs of multi-domain and multi-type text scenarios.

[0136] According to embodiments of this application, by sequentially processing unstructured text through event summarization and classification, modular analysis, structured transformation, and format verification, a fully automated structured processing workflow from semantic understanding to standardized data output is achieved. A step-by-step processing design approach breaks down the complex text structuring task into multiple independent and verifiable steps, reducing the "illusion" risk when large language models directly generate results. Through the event summarization and classification step, the system can automatically identify semantic boundaries in the text, breaking down long content into multiple independent semantic event units, thereby improving the accuracy of information extraction and contextual consistency. In the modular analysis stage, embodiments of this application introduce module configuration files and mapping relationship mechanisms, enabling the system to have flexible structure definition capabilities and supporting the dynamic addition or modification of module templates based on new semantics, thus forming a continuously evolving structured knowledge system. By introducing a semantic association analysis mechanism, consistency verification of module attribute information is performed based on the temporal sequence of events and the relationships between participating entities. This automatically identifies logical contradictions and attribute conflicts between different events and completes self-repair through local regeneration or attribute adjustment. This design enables the system to have cross-event-level semantic reasoning capabilities, thereby ensuring the coherence and authenticity of structured results at the global semantic level.

[0137] Corresponding to the above-described text processing method, embodiments of this application also provide a text processing apparatus.

[0138] Figure 5 A schematic block diagram of a text processing apparatus according to an embodiment of this application is shown.

[0139] like Figure 5 As shown, the text processing device 500 of this embodiment includes a data acquisition module 510, an event summary classification module 520, a modular analysis module 530, and a structured conversion and format verification module 540.

[0140] The data acquisition module 510 can be used to acquire unstructured text to be processed. In one embodiment, the data acquisition module 510 can be used to perform the operation S210 described above, which will not be repeated here.

[0141] The event summary classification module 520 can be used to perform event summary classification on the unstructured text to obtain multiple semantic event units. In one embodiment, the event summary classification module 520 can be used to perform the operation S220 described above, which will not be repeated here.

[0142] The modular analysis module 530 can be used to perform modular analysis on the semantic event unit to obtain the corresponding module attribute information. In one embodiment, the modular analysis module 530 can be used to perform the operation S230 described above, which will not be repeated here.

[0143] The structure conversion and format verification module 540 can be used to perform structure conversion and format verification based on the module attribute information to obtain the target structured data corresponding to the unstructured text. In one embodiment, the structure conversion and format verification module 540 can be used to perform the operation S240 described above, which will not be repeated here.

[0144] According to an embodiment of this application, the event summary classification module 520 can also be used to guide the large model to extract a preliminary event summary of the unstructured text according to the sequence number identification rule based on a preset prompt template; and to divide the preliminary event summary into boundaries based on contextual logical relationships or semantic similarity to obtain the semantic event unit.

[0145] According to embodiments of this application, the event summary classification module 520 can also be used to perform structural and content checks on the semantic event units; and in response to detecting abnormal units in the semantic event units, to perform local regeneration or local content correction on the abnormal units.

[0146] According to embodiments of this application, the modular analysis module 530 can also be used to obtain a module template based on a preset module configuration file, wherein the module configuration file is used to define the module template and to define the mapping relationship between module names and attribute fields; and to match the corresponding module template based on the semantic identifier information of the semantic event unit, and fill the attribute fields according to the mapping relationship to generate the module attribute information.

[0147] According to an embodiment of this application, the modular analysis module 530 can also be used to perform association analysis on the module attribute information based on the temporal order or participating entity relationship of the semantic event units; identify attribute conflicts between different semantic event units according to the association analysis results; and adjust the corresponding module attribute information in response to the identification of attribute conflicts.

[0148] According to an embodiment of this application, the modular analysis module 530 can also be used to match the corresponding module template based on the event type or keyword of the semantic event unit; in response to no matching module template, obtain the added or modified module template and the updated mapping relationship; and generate the module attribute information using the updated mapping relationship.

[0149] According to an embodiment of this application, the structure conversion and format verification module 540 can also be used to parse an intermediate data file containing the module attribute information to extract key-value pairs corresponding to the module attribute information; and to convert the key-value pairs into structured data based on a preset mapping configuration file.

[0150] According to an embodiment of this application, the structure conversion and format verification module 540 can also be used to perform format verification on the structured data based on preset structured data verification rules; in response to the format verification failing, locate the target module attribute information; and regenerate or restructure the target module attribute information until the format verification passes, so as to obtain the target structured data.

[0151] According to embodiments of this application, any multiple modules among the data acquisition module 510, event summary classification module 520, modular analysis module 530, and structure conversion and format verification module 540 can be merged into one module, or any one of these modules can be split into multiple modules. Alternatively, at least some of the functions of one or more of these modules can be combined with at least some of the functions of other modules and implemented in one module. According to embodiments of this application, at least one of the data acquisition module 510, event summary classification module 520, modular analysis module 530, and structure conversion and format verification module 540 can be at least partially implemented as hardware circuitry, such as a field-programmable gate array (FPGA), a programmable logic array (PLA), a system-on-a-chip, a system-on-a-substrate, a system-on-package, an application-specific integrated circuit (ASIC), or any other reasonable means of integrating or packaging circuitry, or implemented in software, hardware, or firmware, or in any suitable combination of any of these three implementation methods. Alternatively, at least one of the data acquisition module 510, the event summary classification module 520, the modular analysis module 530, and the structure conversion and format verification module 540 may be implemented at least partially as a computer program module, which can perform corresponding functions when the computer program module is run.

[0152] Figure 6 A block diagram schematically illustrates an electronic device suitable for implementing a text processing method according to an embodiment of this application.

[0153] like Figure 6 As shown, an electronic device 600 according to an embodiment of this application includes a processor 601, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 602 or a program loaded from a storage portion 608 into a random access memory (RAM) 603. The processor 601 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or an associated chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 601 may also include onboard memory for caching purposes. The processor 601 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of this application.

[0154] RAM 603 stores various programs and data required for the operation of electronic device 600. Processor 601, ROM 602, and RAM 603 are interconnected via bus 604. Processor 601 executes various operations of the method flow according to embodiments of this application by executing programs in ROM 602 and / or RAM 603. It should be noted that the programs may also be stored in one or more memories other than ROM 602 and RAM 603. Processor 601 may also execute various operations of the method flow according to embodiments of this application by executing programs stored in said one or more memories.

[0155] According to embodiments of this application, the electronic device 600 may further include an input / output (I / O) interface 605, which is also connected to a bus 604. The electronic device 600 may also include one or more of the following components connected to the input / output (I / O) interface 605: an input section 606 including a keyboard, mouse, etc.; an output section 607 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 608 including a hard disk, etc.; and a communication section 609 including a network interface card such as a LAN card, modem, etc. The communication section 609 performs communication processing via a network such as the Internet. A drive 610 is also connected to the input / output (I / O) interface 605 as needed. A removable medium 611, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 610 as needed so that computer programs read from it can be installed into the storage section 608 as needed.

[0156] This application also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not assembled into the device / apparatus / system. The computer-readable storage medium carries one or more programs, which, when executed, implement the method according to the embodiments of this application.

[0157] According to embodiments of this application, the computer-readable storage medium can be a non-volatile computer-readable storage medium, such as including but not limited to: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this application, the computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. For example, according to embodiments of this application, the computer-readable storage medium may include ROM 602 and / or RAM 603 and / or one or more memories other than ROM 602 and RAM 603 described above.

[0158] Embodiments of this application also include a computer program product comprising a computer program containing program code for performing the methods shown in the flowchart. When the computer program product is run on a computer system, the program code is used to enable the computer system to implement the text processing method provided in the embodiments of this application.

[0159] When the computer program is executed by the processor 601, it performs the functions defined in the system / apparatus of this application embodiment. According to the embodiments of this application, the systems, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0160] In one embodiment, the computer program may rely on a tangible storage medium such as an optical storage device or a magnetic storage device. In another embodiment, the computer program may also be transmitted and distributed in the form of signals over a network medium, and downloaded and installed via the communication section 609, and / or installed from the removable medium 611. The program code contained in the computer program can be transmitted using any suitable network medium, including but not limited to: wireless, wired, etc., or any suitable combination thereof.

[0161] In such an embodiment, the computer program can be downloaded and installed from a network via the communication section 609, and / or installed from the removable medium 611. When the computer program is executed by the processor 601, it performs the functions defined in the system of this application embodiment. According to the embodiments of this application, the systems, devices, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0162] According to embodiments of this application, program code for executing the computer programs provided in the embodiments of this application can be written in any combination of one or more programming languages. Specifically, these computational programs can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages ​​include, but are not limited to, languages ​​such as Java, C++, Python, "C", or similar programming languages. The program code can be executed entirely on the user's computing device, partially on the user's device, partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0163] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0164] Those skilled in the art will understand that the features described in the various embodiments of this application can be combined and / or combined in various ways, even if such combinations or combinations are not explicitly described in this application. In particular, the features described in the various embodiments of this application can be combined and / or combined in various ways without departing from the spirit and teachings of this application. All such combinations and / or combinations fall within the scope of this application.

[0165] The embodiments of this application have been described above. However, these embodiments are merely illustrative and not intended to limit the scope of this application. Although various embodiments have been described above, this does not mean that the measures in the various embodiments cannot be used advantageously in combination. Without departing from the scope of this application, those skilled in the art can make various substitutions and modifications, all of which should fall within the scope of this application.

Claims

1. A text processing method characterized by, The method comprises: acquiring unstructured text to be processed; performing event summary classification on the unstructured text to obtain a plurality of semantic event units; performing modular analysis on the semantic event units to obtain corresponding module attribute information; and performing structured conversion and format verification based on the module attribute information to obtain target structured data corresponding to the unstructured text.

2. The method of claim 1, wherein, The event summary classification on the unstructured text comprises: based on a preset prompt template, guiding a large model to extract a preliminary event summary of the unstructured text according to a serial number identification rule; and based on a context logical relationship or semantic similarity, performing boundary division on the preliminary event summary to obtain the semantic event units.

3. The method of claim 2, wherein, The method further comprises: performing structure inspection and content inspection on the semantic event units; and in response to detecting an abnormal unit in the semantic event units, locally regenerating or locally correcting the abnormal unit.

4. The method of claim 1, wherein, The modular analysis on the semantic event units to obtain corresponding module attribute information comprises: based on a preset module configuration file, obtaining a module template, the module configuration file being used to define the module template and being used to define a mapping relationship between a module name and an attribute field; and based on semantic identification information of the semantic event units, matching a corresponding module template, and filling an attribute field according to the mapping relationship to generate the module attribute information.

5. The method of claim 4, wherein, The method further comprises: based on a time sequence or a participation entity relationship of the semantic event units, performing correlation analysis on the module attribute information; according to a correlation analysis result, identifying attribute conflicts between different semantic event units; and in response to identifying an attribute conflict, adjusting corresponding module attribute information. The structured conversion based on the module attribute information comprises:

6. The method of claim 1, wherein, parsing an intermediate data file containing the module attribute information to extract key-value pairs corresponding to the module attribute information; and based on a preset mapping configuration file, converting the key-value pairs into structured data. The format verification comprises: based on a preset structured data verification rule, performing format verification on the structured data; 7. The method of claim 6, wherein, in response to the format verification failing, locating target module attribute information; and regenerating or structurally adjusting the target module attribute information until the format verification passes to obtain the target structured data. The matching of a corresponding module template based on semantic identification information of the semantic event units and the filling of an attribute field according to the mapping relationship to generate the module attribute information comprises: based on an event type or a keyword of the semantic event units, matching a corresponding module template; in response to not matching a module template, acquiring an added or modified module template and an updated mapping relationship; and 8. The method of claim 4 or 5, wherein, generating the module attribute information by using the updated mapping relationship. The apparatus comprises: a data acquisition module configured to acquire unstructured text to be processed; an event summary classification module configured to perform event summary classification on the unstructured text to obtain a plurality of semantic event units; and a modular analysis module configured to perform modular analysis on the semantic event units to obtain corresponding module attribute information.

9. A text processing apparatus characterized by comprising: ​ ​ ​ The modular analysis module is configured to perform modular analysis on the semantic event unit to obtain corresponding module attribute information. The structured conversion and format verification module is configured to perform structured conversion and format verification based on the module attribute information to obtain target structured data corresponding to the unstructured text. 10.An electronic device comprising: one or more processors; memory for storing one or more computer programs, characterized in that the one or more processors execute the one or more computer programs to implement the steps of the method according to any one of claims 1-8.

11. A computer readable storage medium having stored thereon a computer program or instructions, characterized in that, The computer program or instructions, when executed by a processor, implement the steps of the method according to any one of claims 1-8.

12. A computer program product comprising computer programs or instructions, characterized in that, The computer program or instructions, when executed by a processor, implement the steps of the method according to any one of claims 1-8. The computer program or instructions, when executed by a processor, implement the steps of the method according to any one of claims 1-8.

Images