A file compliance determination method and related apparatus

Abstract

The specification discloses a file compliance judgment method and related devices, wherein the method introduces a multi-modal compliance large model to uniformly analyze and compare external rules files and internal rules files, integrates text, logical structure and image elements into the same analysis framework, and realizes systematic judgment on file compliance. By constructing a structured compliance rule set and a compliance knowledge graph, and combining a gap analysis large model to identify and quantitatively evaluate the gap, the compliance analysis is changed from traditional manual comparison to automatic, structured and quantifiable processing, which significantly improves the comprehensiveness, accuracy and efficiency of compliance review, and reduces the risk caused by human understanding bias.

Description

Technical Field

[0001] This specification relates to the field of computer technology, and in particular to a method and apparatus for determining document compliance. Background Technology

[0002] Current compliance assessments primarily rely on manual comparison or rule matching based on a single text dimension. These methods typically only search and judge the textual content of clauses, failing to consider non-textual elements such as logical structure, signatures, and tables. This results in insufficient accuracy and completeness in analyzing complex compliance documents. Furthermore, the correspondence between external and internal regulations in existing technologies is mostly maintained manually, lacking a unified structured expression and dynamic update mechanism. This makes it difficult to systematically identify different types of compliance issues such as omissions, conflicts, or insufficient constraints. Summary of the Invention

[0003] This specification provides a method and related apparatus for determining document compliance, the technical solution of which is as follows:

[0004] Firstly, this specification provides a method for determining document compliance. The method includes: using a multimodal compliance model to perform multimodal analysis on external and internal regulatory document data, extracting multimodal feature data, including text content, logical structure information, and image features; performing deep semantic analysis of the text content using natural language, extracting key elements based on subject-verb-object parsing logic, and generating a structured compliance rule set, wherein the rule clauses in the compliance rule set include external and internal regulatory clauses, and each rule clause has a corresponding tag; maintaining a compliance knowledge graph based on the compliance rule set, the compliance knowledge graph containing the relationships between the external and internal regulatory clauses; based on the relationships, using a gap analysis model to perform multi-dimensional comparisons of the associated internal and external regulatory clauses to obtain compliance gap results; calculating quantitative evaluation indicators based on the compliance gap results, and generating a compliance analysis report, the compliance analysis report including the quantitative evaluation indicators and a compliance gap list.

[0005] Secondly, this specification provides a document compliance determination device, which includes: a multimodal parsing module, used to perform multimodal parsing on external and internal regulatory document data using a multimodal compliance big data model, and extract multimodal feature data, including text content, logical structure information, and image features; and a deep semantic analysis module, used to perform natural language processing on the text content, extract key elements based on subject-verb-object parsing logic, and generate a structured compliance rule set, wherein the rule clauses in the compliance rule set include external regulatory clauses and internal regulatory clauses, and each rule clause has a corresponding label, wherein the external regulatory clauses are external regulations. The clauses in the document data and the internal regulation clauses are clauses in the internal regulation document data; the knowledge graph maintenance module is used to maintain a compliance knowledge graph based on the compliance rule set, and the compliance knowledge graph contains the relationship between the external regulation clauses and the internal regulation clauses; the multi-dimensional comparison module is used to perform multi-dimensional comparison of the related internal regulation clauses and external regulation clauses based on the relationship and using a gap analysis model to obtain compliance gap results; the quantitative assessment calculation module is used to calculate quantitative assessment indicators based on the compliance gap results and generate a compliance analysis report, and the compliance analysis report contains the quantitative assessment indicators and a compliance gap list.

[0006] Thirdly, this specification provides a computer storage medium storing a plurality of instructions adapted for loading by a processor and executing the above-described method steps.

[0007] Fourthly, this specification provides an electronic device that may include: a processor and a memory; wherein the memory stores a computer program adapted to be loaded by the processor and to execute the above-described method steps.

[0008] Fifthly, this specification provides a computer program product that stores at least one instruction, which is loaded by a processor and executes the above-described method steps.

[0009] The beneficial effects of the technical solutions provided in some embodiments of this specification include at least the following:

[0010] In one or more embodiments of this specification, a multimodal compliance model is introduced to uniformly parse and compare external and internal compliance documents, incorporating text, logical structure, and image elements into the same analytical framework to achieve a systematic determination of document compliance. By constructing a structured compliance rule set and a compliance knowledge graph, and combining it with a gap analysis model for gap identification and quantitative assessment, compliance analysis is transformed from traditional manual comparison to automated, structured, and quantifiable processing, significantly improving the comprehensiveness, accuracy, and efficiency of compliance reviews and reducing the risks associated with human misunderstanding. Attached Figure Description

[0011] To more clearly illustrate the technical solutions in this specification or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this specification. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 This is a schematic diagram of a document compliance determination system provided in this manual.

[0013] Figure 2 This is a flowchart illustrating a document compliance determination method provided in this manual.

[0014] Figure 3 It is based on Figure 2 A flowchart illustrating a specific implementation of step S100 in the document compliance determination method shown in the corresponding embodiment.

[0015] Figure 4 It is based on Figure 2 A flowchart illustrating a specific implementation of step S400 in the document compliance determination method shown in the corresponding embodiment.

[0016] Figure 5 It is based on Figure 4 A flowchart illustrating a specific implementation of step S410 in the document compliance determination method shown in the corresponding embodiment.

[0017] Figure 6 This is a schematic diagram of a document compliance determination device provided in this specification.

[0018] Figure 7 This is a schematic diagram of the structure of an electronic device provided in this specification.

[0019] Figure 8 This is a schematic diagram of the operating system and user space provided in this manual.

[0020] Figure 9 yes Figure 7 Architecture diagram of the Android operating system in China.

[0021] Figure 10 yes Figure 7 Architecture diagram of the iOS operating system. Detailed Implementation

[0022] The technical solutions in this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments in this specification, not all of them. Based on the embodiments in this specification, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this specification. In the description of this specification, it should be understood that the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance. In the description of this specification, it should be noted that unless otherwise expressly specified and limited, "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or devices. Those skilled in the art can understand the specific meaning of the above terms in this specification based on the specific circumstances. Furthermore, in the description of this specification, unless otherwise stated, "a plurality of" means two or more. "AND / OR" describes the relationship between related objects, indicating that there can be three relationships. For example, A AND / OR B can represent: A alone, A and B simultaneously, or B alone. The character " / " generally indicates that the objects before and after it are in an "OR" relationship.

[0023] The present specification will now be described in detail with reference to specific embodiments.

[0024] Please see Figure 1 This is a schematic diagram illustrating a scenario for a document compliance assessment system provided in this specification. Figure 1 As shown, the document compliance determination system may include at least a client cluster and a service platform 100. The client cluster may include at least one client, such as... Figure 1As shown, the cluster specifically includes client 1 corresponding to user 1, client 2 corresponding to user 2, ..., client n corresponding to user n, where n is an integer greater than 0. Each client in the client cluster can be an electronic device with communication capabilities, including but not limited to: wearable devices, handheld devices, personal computers, tablets, in-vehicle devices, smartphones, computing devices, or other processing devices connected to a wireless modem. Electronic devices may have different names in different networks, such as: user equipment, access terminal, user unit, user station, mobile station, mobile station, remote station, remote terminal, mobile device, user terminal, terminal, wireless communication equipment, user agent or user device, cellular phone, cordless phone, personal digital assistant (PDA), and electronic devices in 5G networks or future evolved networks. The service platform 100 can be a standalone server device, such as a rack-mount, blade, tower, or cabinet-type server, or a workstation, mainframe, or other hardware device with strong computing power. Alternatively, it can be a server cluster composed of multiple servers, where each server is symmetrically arranged, with each server functionally and equally important in the transaction chain. Each server can independently provide services, meaning it can provide services without the assistance of other servers. In one or more embodiments of this specification, the service platform 100 can establish a communication connection with at least one client in the client cluster. Based on this communication connection, data interaction can be completed during the document compliance determination process, such as online transaction data interaction. For example, the service platform 100 can recommend content to the client based on the target neural network model obtained by the document compliance determination method of this specification. Alternatively, the service platform 100 can obtain training data from the client, such as first training data.

[0025] It should be noted that the service platform 100 establishes a communication connection with at least one client in the client cluster via a network for interactive communication. This network can be a wireless network or a wired network. Wireless networks include, but are not limited to, cellular networks, wireless LANs, infrared networks, or Bluetooth networks. Wired networks include, but are not limited to, Ethernet, universal serial bus (USB), or controller area networks. In one or more embodiments of the specification, technologies and / or formats including Hyper Text Markup Language (HTML), Extensible Markup Language (XML), etc., are used to represent data exchanged over the network (such as target compressed packets). Furthermore, conventional encryption technologies such as Secure Socket Layer (SSL), Transport Layer Security (TLS), Virtual Private Network (VPN), and Internet Protocol Security (IPsec) can be used to encrypt all or some links. In other embodiments, customized and / or dedicated data communication technologies can be used to replace or supplement the aforementioned data communication technologies.

[0026] The document compliance determination system embodiments provided in this specification and the document compliance determination methods described in one or more embodiments belong to the same concept. The execution entity corresponding to the document compliance determination method involved in one or more embodiments of this specification can be the aforementioned service platform 100; the execution entity corresponding to the document compliance determination method involved in one or more embodiments of this specification can also be the electronic device corresponding to the client, specifically determined based on the actual application environment. The implementation process of the document compliance determination system embodiments can be detailed in the following method embodiments, and will not be repeated here.

[0027] based on Figure 1 The illustrated scenario diagram is followed by a detailed description of the document compliance determination method provided in one or more embodiments of this specification. Please refer to [link to documentation]. Figure 2 This document provides a flowchart illustrating a document compliance determination method according to one or more embodiments. This method can be implemented using a computer program and can run on a document compliance determination device based on the von Neumann architecture. The computer program can be integrated into an application or run as a standalone utility application. The document compliance determination device can be a service platform.

[0028] Specifically, the compliance determination method for this document includes: S100, using a multimodal compliance model to perform multimodal parsing on external and internal regulatory document data, extracting multimodal feature data, including text content, logical structure information, and image features. S200, performing deep semantic analysis of the text content using natural language, extracting key elements based on subject-verb-object parsing logic, and generating a structured compliance rule set. The rule clauses in the compliance rule set include external and internal regulatory clauses, and each rule clause has a corresponding tag. The external regulatory clauses are clauses in the external regulatory document data, and the internal regulatory clauses are clauses in the internal regulatory document data. S300, maintaining a compliance knowledge graph based on the compliance rule set, the compliance knowledge graph containing the relationships between the external and internal regulatory clauses. S400, based on the relationships, using a gap analysis model to perform multi-dimensional comparisons of the associated internal and external regulatory clauses to obtain compliance gap results. S500, based on the compliance gap results, calculate quantitative assessment indicators and generate a compliance analysis report, which includes the quantitative assessment indicators and a compliance gap list.

[0029] In the embodiments of this specification, external and internal regulatory document data, including various information formats such as text, images, tables, seals, and signatures, are uniformly parsed and structured. Based on this, a compliance knowledge graph is constructed, and the correspondence between internal and external regulatory clauses is modeled semantically and structurally. A gap analysis model is then used to perform multi-dimensional comparative analysis of the two, automatically identifying and classifying compliance gaps. Finally, a compliance analysis report containing quantitative assessment indicators and a list of compliance gaps is generated. The embodiments of this specification can cover multimodal compliance elements, improve the accuracy, completeness, and interpretability of compliance judgments, and provide quantitative basis for compliance management and risk control.

[0030] It should be noted that the external regulatory documents refer to laws, regulations, regulatory standards, policy documents, or regulatory notices issued by regulatory agencies, while the internal regulatory documents refer to management systems, operating procedures, or business specifications developed internally by the organization. These documents can be electronic documents, scanned documents, or image files. In other words, external regulatory documents are standard documents, while internal regulatory documents are documents subject to compliance. This specification uses external regulatory documents as a standard to determine whether internal regulatory documents have compliance issues.

[0031] In S100, a multimodal compliance model is used to perform multimodal analysis on external and internal regulation document data to extract multimodal feature data. The multimodal compliance model unifies the analysis of different information formats in the external and internal regulation document data, and the extracted multimodal feature data includes at least text content, logical structure information, and image features. The image features are used to represent visual entities in the documents, and the logical structure information is used to represent the correspondence between visual entities and text content, thus providing basic data support for subsequent semantic analysis and compliance determination.

[0032] Specifically, in some embodiments, the specific implementation of step S100 can be found in [reference needed]. Figure 3 . Figure 3 It is based on Figure 2 The detailed description of step S100 in the document compliance determination method shown in the corresponding embodiment is as follows: In the document compliance determination method, the multimodal compliance big model includes a visual encoder and a language big model. The multimodal feature data includes image features, text content, and logical structure information. Step S100 may include the following steps: S110, using the visual encoder to extract entity features from the external and internal compliance document data to obtain image features of visual entities, including seals, signatures, and tables. S120, using the language big model to perform text recognition and semantic understanding on the external and internal compliance document data to obtain text content, and fusing and aligning the image features of the visual entities with the text content to determine the logical structure information between the visual entities and the text content.

[0033] In this embodiment, the multimodal compliance big model is composed of a visual encoder and a language big model working together. The visual encoder is used to identify and extract features from visual entities in the file data; the language big model is used to identify and semantically understand textual information in the file data, and to achieve fusion and alignment between textual information and visual entities.

[0034] This embodiment utilizes a multimodal parsing process within a multimodal compliance big data model to ultimately generate multimodal feature data encompassing image features, text content, and logical structure information. This multimodal feature data serves as the foundational input for subsequent deep semantic analysis of natural language, generation of compliance rule sets, maintenance of compliance knowledge graphs, and compliance gap analysis. By introducing a multimodal compliance big data model that collaborates with a visual encoder and a language big data model, this embodiment is able to comprehensively parse complex document structures and multimodal compliance elements, improving the accuracy and reliability of document compliance determination.

[0035] In step S110, a visual encoder is used to extract entity features from external and internal regulatory document data to obtain image features of visual entities. These external and internal regulatory document data typically contain visual entities such as seals, signatures, and tables. These visual entities are used to represent formal compliance elements or structured data elements in the compliance determination process. By processing the document data with the visual encoder, the aforementioned visual entities can be identified from the document page, and their appearance, spatial location, and structural features can be characterized to generate corresponding image feature data. These image features are used for subsequent fusion analysis with text content to determine the correspondence between visual entities and specific clauses.

[0036] Specifically, in some embodiments, the specific implementation of step S110 can be found in the following embodiments. This embodiment is based on... Figure 3 The detailed description of step S110 in the document compliance determination method shown in the corresponding embodiment is as follows: In this method, the visual encoder includes an image input layer, a slicing layer, and multiple stages of sliding window transformation blocks. Step S110 may include the following steps: inputting the outer specification file data and the inner specification file data into the image input layer in the form of images to obtain corresponding three-dimensional outer specification tensors and three-dimensional inner specification tensors. Dividing the three-dimensional outer specification tensors and three-dimensional inner specification tensors into multiple non-overlapping image blocks through the slicing layer to obtain corresponding image block sequence vectors. Extracting entity features from the image blocks in the image block sequence vector one by one according to the stage sequence through the multiple stages of sliding window transformation blocks to obtain image features of visual entities.

[0037] In this embodiment, the visual encoder adopts a hierarchical architecture to better capture long-distance dependencies in document images. Its structure is well-suited for processing document images containing complex layouts, tables, and stamps of varying sizes, effectively preserving the spatial location information of visual entities. The visual encoder includes an image input layer, a slicing layer, and multiple stages of sliding window transform blocks for extracting entity features from external and internal specification document data. Specifically, the image input layer receives preprocessed page image data from the external or internal specification document; the slicing layer segments the input image data; and the multiple stages of sliding window transform blocks process the segmented image blocks stage by stage to extract image features of visual entities.

[0038] This embodiment employs a visual encoder structure that includes a slice segmentation layer and a multi-stage sliding window transform block to achieve refined feature extraction of local entities in large-format document images. It can effectively handle visual entities of different scales and distribution locations, improve the recognition accuracy of key elements such as seals, signatures, and tables, and avoid information loss caused by differences in image size or layout. This provides a stable and reliable visual feature foundation for subsequent multimodal fusion and compliance analysis.

[0039] In some specific implementations, the outer and inner specification file data can be input to the image input layer as images. The images can be scanned document page images or page images generated from electronic documents. Through the image input layer, each page image is converted into a corresponding three-dimensional outer specification tensor or three-dimensional inner specification tensor, where the three-dimensional tensor is used to represent the pixel information of the image in the spatial and channel dimensions. The image input layer enables unified image input of the outer and inner specification file data, providing a standardized data format for subsequent segmentation and feature extraction. After obtaining the three-dimensional outer and inner specification tensors, the slicing layer is used to segment the outer and inner specification file data. Specifically, the slicing layer divides the three-dimensional outer and inner specification tensors into multiple non-overlapping image blocks according to preset segmentation rules, thereby forming a corresponding image block sequence vector. Each image block is used to represent local region information in the original document image. By segmenting a full-page document image into multiple image blocks, the data scale of a single processing step can be reduced while preserving the detailed features of visual entities within local regions, providing a foundation for subsequent entity feature extraction. After image slicing, multiple stages of sliding window transformation blocks process the image blocks in the image block sequence vector one by one in sequence to extract image features of visual entities. Specifically, each stage of the sliding window transformation block includes a multi-layer moving window-based multi-head self-attention mechanism and a multilayer perceptron (MLP). Each stage of the sliding window transformation block scans and transforms the image blocks within different receptive fields, thereby gradually extracting feature information related to visual entities such as seals, signatures, and tables, forming a deep visual feature map containing visual texture features and spatial layout features of seals, signatures, and tables. Through multi-stage processing, the visual encoder can capture the morphological and structural features of visual entities at different scales. After processing by the multiple stages of sliding window transformation blocks, image features representing visual entities are output. These image features are used for subsequent fusion and alignment with text content to determine the logical structural relationship between visual entities and corresponding text clauses. Through the above processing flow, image features containing visual entities such as seals, signatures, and forms are finally obtained. These image features, as part of the multimodal feature data, participate in subsequent multimodal fusion, compliance rule generation, compliance gap analysis, and compliance analysis report generation processes. By employing a visual encoder structure including an image input layer, a slice segmentation layer, and a multi-stage sliding window transform block, this embodiment can stably and accurately extract visual entity features in complex document environments, improving the completeness and reliability of multimodal parsing.

[0040] In S120, a large language model is used to perform text recognition and semantic understanding on external and internal regulatory document data to obtain text content. After recognizing the text content in the documents, the large language model performs contextual semantic analysis to extract clause statements, system descriptions, and their semantic relationships, thereby obtaining structured text content information. This text content is used to subsequently generate compliance rule clauses and participate in compliance gap analysis.

[0041] After extracting the image features of visual entities and understanding the semantics of the text content, the image features of the visual entities and the text content are fused and aligned to determine the logical structural information between them. Specifically, by performing alignment analysis on the image features of visual entities and the text content in a unified semantic space, the position of the clause, chapter, or semantic unit corresponding to the visual entity is determined, thereby clarifying the association between the seal, signature, or form and the specific text content. Through the above fusion and alignment process, logical structural information is generated to characterize the internal structural relationships of the document. This logical structural information describes the logical correspondence between the text content and the visual entities within the document.

[0042] Specifically, in some embodiments, the specific implementation of step S120 can be found in the following embodiments. This embodiment is based on... Figure 3 The detailed description of step S120 in the document compliance determination method shown in the corresponding embodiment is as follows: In the document compliance determination method, the language large model includes an embedding layer, a concatenation input layer, a decoding layer, a multimodal alignment layer, and an output layer. Step S120 may include the following steps: The external and internal specification document data are pre-recognized by OCR and then input into the embedding layer to obtain a text embedding vector sequence. The image features of the visual entities are compressed and dimensionally transformed by the multimodal alignment layer to obtain a visual embedding vector sequence. The text embedding vector sequence and the visual embedding vector sequence are concatenated into the concatenation input layer to obtain a combined sequence. The decoding layer fuses the combined sequence using a multi-head self-attention mechanism and a feedforward neural network to obtain a hidden state vector. The output layer performs layer normalization and linear transformation on the hidden state vector to obtain structured text content and logical structure information.

[0043] In this embodiment, the large language model adopts a Transformer-Decoder-only architecture similar to LLaMA or GPT series structures, including an embedding layer, a concatenation input layer, a decoding layer, a multimodal alignment layer, and an output layer. This is used for text recognition, semantic understanding, and multimodal feature fusion processing of external and internal specification file data. Simultaneously, during multimodal alignment and concatenation, a Q-Former or linear projection layer is used to map the image feature dimensions output by the visual encoder to the text embedding space of the large language model, achieving visual and text fusion alignment. Specifically, the embedding layer converts text information into vector representations usable for semantic computation; the multimodal alignment layer processes the image features of visual entities, aligning them with the text embedding vectors within a unified semantic space; the concatenation input layer concatenates the text embedding vector sequence and the visual embedding vector sequence; the decoding layer performs semantic modeling on the fused multimodal information; and the output layer generates structured text content and logical structure information.

[0044] This embodiment introduces a multimodal alignment layer and a fusion decoding mechanism into the large language model, enabling deep fusion of visual features and textual semantics within a unified representation space. This accurately depicts the logical relationship between text content and corresponding visual entities, generating structured text content and logical structure information. Consequently, it enhances the understanding of document clause hierarchy, relationships, and dependencies of formal elements, providing higher-quality semantic input for rule extraction and compliance judgment.

[0045] In some specific implementations, the external and internal specification file data can be pre-recognized using OCR before entering the language model to obtain text information from the files. Subsequently, the OCR-pre-recognized text information is input into the embedding layer to generate a corresponding text embedding vector sequence. This text embedding vector sequence represents the distribution of text content in the semantic space, providing a foundation for subsequent multimodal fusion processing. After visual entity image feature extraction, the image features of the visual entities are processed through the multimodal alignment layer. Specifically, visual features can be compressed and dimensionally transformed using learnable query vectors or fully connected layers to generate a visual embedding vector sequence, ensuring consistency between the visual embedding vector sequence and the text embedding vector sequence in vector dimension and semantic space, thus meeting the input requirements for multimodal fusion processing. After obtaining the text embedding vector sequence and the visual embedding vector sequence, they are input into a concatenation layer for concatenation, forming a combined sequence. Subsequently, the decoding layer uses a multi-head self-attention mechanism and a feedforward neural network to fuse the combined sequence, understanding the correspondence between the seal and the text, and the row and column logic of the table. Specifically, the decoding layer models the correlation between different modal features in the combined sequence through a multi-head self-attention mechanism, and further processes the fused features through a feedforward neural network to obtain a hidden state vector. This allows text content and visual entity information to interact and model within a unified semantic space, thereby enhancing the understanding of the overall semantic and structural relationships of the document. After obtaining the hidden state vector, the output layer normalizes and linearly transforms it to generate structured text content and logical structure information, such as JSON-formatted key-value pairs describing the presence of a certain seal and its associated clause. The structured text content is used for subsequent deep semantic analysis of natural language and generation of compliance rule sets, while the logical structure information describes the correspondence between visual entities and text content and participates as an important component of multimodal feature data in subsequent compliance gap analysis and compliance analysis report generation. Through the above-described structural design and processing flow of the large language model, this embodiment enables deep semantic understanding and structured expression of document content in multimodal compliance scenarios, improving the usability and reliability of multimodal parsing results in the compliance determination process.

[0046] In S200, the text content is subjected to deep semantic analysis of natural language, and key elements are extracted based on subject-verb-object parsing logic to generate a structured set of compliance rules.

[0047] Specifically, syntactic and semantic analysis can be performed on the clauses in the text content to identify the subject as the regulatory object, the predicate as the compliance action, and the object as the recipient of the compliance action. Based on preset regulatory object determination rules, each identified regulatory object is judged one by one. When the regulatory object meets the preset rules, the corresponding subject, predicate, and object are combined to form a rule clause. The rule clauses include external regulation clauses and internal regulation clauses, where external regulation clauses are derived from external regulation document data, and internal regulation clauses are derived from internal regulation document data. In some embodiments, each rule clause can be configured with a corresponding tag, which is used to characterize the business area, risk level, or gap level to which the rule clause belongs, thereby forming a structured compliance rule set.

[0048] Specifically, in some embodiments, the specific implementation of step S200 can be found in the following embodiments. This embodiment is based on... Figure 2 The detailed description of step S200 in the document compliance determination method shown in the corresponding embodiment includes the following steps: performing syntactic analysis on the clauses in the text content to extract the subject as the regulatory object, the predicate as the compliance action, and the object as the action recipient. Determining whether each regulatory object conforms to preset rules. If the regulatory object conforms to the preset rules, combining the corresponding subject, predicate, and object to form a rule clause. Matching corresponding tags to each rule clause to generate the compliance rule set.

[0049] In this embodiment, deep semantic analysis of the clause statements is performed based on subject-verb-object parsing logic to transform natural language clauses into structured rule clauses, thus clearly expressing the regulatory objects, compliance actions, and the recipients of those actions. This embodiment effectively eliminates ambiguity in natural language, improves the machine understandability of compliance rules, and provides a unified and standardized data foundation for subsequent compliance knowledge graph construction, gap analysis, and risk classification through rule tagging.

[0050] In some specific implementations, the clauses in the text content undergo syntactic analysis. Syntactic analysis breaks down the grammatical components of the clauses, extracting the subject, predicate, and object. The subject represents the regulatory object targeted by the clause, the predicate represents the compliance action stipulated in the clause, and the object represents the recipient or object of the compliance action, transforming the clause from unstructured natural language expression into a structured set of semantic elements. After extracting the subject, predicate, and object, the regulatory object corresponding to each subject is individually assessed to determine whether it conforms to preset rules. These preset rules define the types of regulatory objects that need to be included in the compliance assessment, such as financial institutions, payment institutions, or other legally regulated entities. When a regulatory object conforms to the preset rules, it indicates that the corresponding clause is a compliance obligation clause and should proceed to the subsequent rule clause generation process. This avoids including descriptive or non-binding content that does not fall within the scope of compliance obligations into the compliance rule set, thereby improving the accuracy of the compliance rule set. When a regulatory object conforms to the preset rules, the corresponding subject, predicate, and object are combined to form a rule clause. The rule clauses are used to represent a specific compliance obligation or requirement, and are categorized as external or internal regulation clauses based on their source document type. This transforms the natural language statements into structured rule clauses that can be calculated and compared. Rule clauses originating from external regulation documents are identified as external regulation clauses, and those originating from internal regulation documents are identified as internal regulation clauses. After generating the rule clauses, corresponding tags are matched to each clause, forming a structured compliance rule set. These tags represent the business area, risk level, or gap level to which the rule clause belongs. By configuring tags for the rule clauses, the compliance rule set possesses multi-dimensional attribute description capabilities, facilitating subsequent construction of a compliance knowledge graph, compliance gap analysis, and calculation of quantitative assessment indicators.

[0051] In S300, a compliance knowledge graph is maintained based on the aforementioned set of compliance rules.

[0052] Specifically, each rule clause in the compliance rule set can be instantiated as an entity node in the compliance knowledge graph. Based on the semantic similarity, referencing relationships, and hierarchical relationships between the rule clauses, associations are established between these entity nodes. These associations represent the corresponding states between external and internal regulatory clauses. In this way, the compliance knowledge graph contains the associations between external and internal regulatory clauses, providing a semantic and structural foundation for subsequent compliance gap analysis.

[0053] Specifically, in some embodiments, the specific implementation of step S300 can be found in the following embodiments. This embodiment is based on... Figure 2The detailed description of step S300 in the document compliance determination method shown in the corresponding embodiment includes the following steps: instantiating each rule clause in the compliance rule set as an entity node in a graph; establishing edges between entity nodes based on the reference relationships, hierarchical relationships, and association relationships between the rule clauses, wherein the edges include at least one of constraint relationships, implementation relationships, and conflict relationships; dynamically adding entity nodes in response to the receipt of external compliance document data, and updating the existing relationship network between entity nodes based on semantic reasoning.

[0054] In this embodiment, the relationship between external and internal regulations is explicitly modeled by instantiating the structured compliance rule set into a compliance knowledge graph. This embodiment can intuitively reflect the constraint, implementation, or conflict relationships between different clauses, and supports dynamic expansion and semantic reasoning updates when external regulations are updated, thereby enhancing the maintainability and scalability of the compliance system and avoiding the problem that static rule bases are difficult to adapt to regulatory changes.

[0055] In some specific implementations, each rule clause in the compliance rule set is instantiated as an entity node in a compliance knowledge graph, transforming the compliance rule set from a rule list into a graph structure usable for relational modeling and semantic reasoning. Each entity node corresponds to a specific rule clause, representing its semantic content and attribute information. The attribute information may include the rule clause's source type, business domain, risk level, and gap level label. After the rule clauses are instantiated, edges are established between entity nodes based on the reference relationships, hierarchical relationships, and associations between internal and external regulations. This allows the compliance knowledge graph to intuitively reflect the corresponding structure and logical state between external and internal regulations. These edges represent the logical relationships between rule clauses, including at least one or more of constraint relationships, implementation relationships, and conflict relationships. Constraint relationships represent the normative binding effect of external regulations on internal regulations; implementation relationships represent the execution or acceptance of external regulations by internal regulations; and conflict relationships represent inconsistencies between internal and external regulations. When new external regulatory documents are received, the rule clauses parsed from the external regulatory documents are instantiated as new entity nodes and added to the compliance knowledge graph. Subsequently, semantic reasoning is used to analyze and update the relationships between the newly added entity nodes and existing entity nodes, thereby adjusting the relationship network between entity nodes. This ensures that the compliance knowledge graph can continuously evolve with changes in external regulatory documents, maintaining its adaptability to the latest regulatory requirements and providing an accurate and real-time relationship foundation for subsequent compliance gap analysis.

[0056] In S400, based on the relationships in the compliance knowledge graph, a gap analysis model is used to compare the related internal and external regulatory clauses in multiple dimensions to obtain compliance gap results.

[0057] Specifically, a gap analysis model can be used to comprehensively consider the consistency between internal and external regulatory clauses across multiple dimensions, including semantic content, formal elements, and logical structure, and to conduct a comparative analysis of the two. Through this multi-dimensional comparison, it can be identified whether the internal regulatory clauses meet the requirements of the external regulatory clauses, and thereby determine whether a compliance gap exists. The compliance gap results are used to characterize the degree of compliance of the internal regulatory clauses with respect to the external regulatory clauses.

[0058] Specifically, in some embodiments, the specific implementation of step S400 can be found in [reference needed]. Figure 4 . Figure 4 It is based on Figure 2 The detailed description of step S400 in the document compliance determination method shown in the corresponding embodiment includes the following steps: S410, based on the correlation, using a gap analysis model to perform a multi-dimensional comparison of the related internal and external regulatory clauses to identify the compliance gaps between them. S420, classifying each compliance gap to determine the corresponding gap type.

[0059] In this embodiment, internal and external regulatory clauses are compared in a targeted manner based on the relationships in the compliance knowledge graph, avoiding invalid analysis between irrelevant clauses. A large-scale gap analysis model is used for multi-dimensional comparison and gap classification, enabling compliance issues to be accurately located and differentiated by nature. This provides a clear basis for subsequent risk assessment and rectification decisions, thereby improving the interpretability and practical value of the compliance analysis results.

[0060] In S410, the first step is to identify the clause pairs requiring compliance gap analysis based on the relationships maintained in the compliance knowledge graph. This ensures that the compliance gap analysis is conducted between clause pairs with clear semantic connections, thereby avoiding invalid comparisons between irrelevant clauses. The relationships represent the semantic and structural correspondence between external and internal regulatory clauses. By reading the established relationships in the compliance knowledge graph, one or more internal regulatory clauses associated with each external regulatory clause are selected as input comparison objects for the gap analysis model. After determining the clause comparison objects, the gap analysis model is used to perform a multi-dimensional comparison between the associated internal and external regulatory clauses to identify whether a compliance gap exists. Specifically, the gap analysis model can comprehensively consider the consistency between internal and external regulatory clauses in multiple dimensions, such as semantic content, logical structure, and formal elements, and analyze and judge them. When the gap analysis model determines that an internal regulatory clause does not meet the requirements of an external regulatory clause in at least one of the above dimensions, a compliance gap is identified. The compliance gap results are used to characterize the degree of compliance of internal regulations with external regulations, and serve as the basis for subsequent gap classification and quantitative assessment.

[0061] Specifically, in some embodiments, the specific implementation of step S410 can be found in [reference needed]. Figure 5 . Figure 5 It is based on Figure 4 The detailed description of step S410 in the document compliance determination method shown in the corresponding embodiment includes the following steps: S412, using a gap analysis model to perform a multi-dimensional comparison of the relevant internal and external regulatory clauses to obtain a multi-dimensional comparison result; S414, based on the multi-dimensional comparison result, identifying the compliance gap between the two.

[0062] In this embodiment, the compliance gap identification process is broken down into two stages: multi-dimensional comparison result generation and gap determination. This makes the gap analysis logic clearer and more controllable. The multi-dimensional comparison results provide traceable analytical basis for gap identification, helping to reduce the uncertainty caused by black-box judgments, thereby improving the accuracy and stability of compliance gap identification and enhancing the credibility of the analysis results in audit and compliance management scenarios.

[0063] In S412, the gap analysis model, upon receiving a pair of internal and external regulatory clauses determined by the compliance knowledge graph, performs a multi-dimensional comparative analysis. This multi-dimensional comparison includes, but is not limited to, comparing the consistency of the semantic content of the clauses, the completeness of the structural elements involved in the clauses, and the conformity of the formal elements associated with the clauses. By comparing and analyzing the internal and external regulatory clauses across multiple dimensions, multi-dimensional comparison results are generated. These results record the consistency or difference status of the internal and external regulatory clauses in each comparison dimension in a structured form, intuitively reflecting the differences between the internal and external regulatory clauses across different dimensions and providing a basis for identifying compliance gaps.

[0064] Specifically, in some embodiments, the specific implementation of step S412 can be found in the following embodiments. This embodiment is based on... Figure 5 The detailed description of step S412 in the document compliance determination method shown in the corresponding embodiment is as follows: In this method, the gap analysis model includes calling a semantic similarity model, a signature verification model, and a table verification model. The multidimensional comparison results include semantic difference results, signature verification results, and table verification results. Step S412 may include the following steps: calling the semantic similarity model to determine the semantic similarity between external and internal regulations, obtaining semantic difference results; calling the signature verification model to verify the signatures in the internal and external regulation document data, obtaining signature verification results; and responding to table recognition, calling the table verification model to verify the consistency of the numerical range or logical structure information required by the internal and external regulations, obtaining table verification results.

[0065] In this embodiment, the gap analysis model includes a semantic similarity model, a signature verification model, and a table verification model. These models are used to perform multi-dimensional comparisons of related internal and external regulatory clauses and generate multi-dimensional comparison results. Specifically, the semantic similarity model compares the semantic content of external and internal regulatory clauses; the signature verification model checks the consistency and completeness of signature elements in external and internal regulatory document data; and the table verification model checks the consistency of numerical ranges or logical structures in table data within documents. This implementation, by introducing dedicated models such as semantic similarity, signature verification, and table verification into the gap analysis model, achieves multi-layered coverage of compliance issues. It can simultaneously handle semantic consistency, formal legality, and structured data compliance issues, avoiding omissions caused by single-dimensional analysis, making compliance gap identification more comprehensive, and thus improving the completeness and refinement of the overall compliance judgment results.

[0066] In some specific implementations, the semantic similarity model analyzes and judges the semantic consistency between external and internal regulatory clauses based on the semantic representation of the clause text, obtaining a semantic difference result to characterize the degree of semantic difference between the two. This semantic difference result reflects whether the internal regulatory clauses meet the compliance requirements stipulated by the external regulatory clauses in terms of semantic content. The signature verification model determines whether the document contains necessary signature elements and analyzes the completeness, consistency, and correspondence between the signatures and relevant clauses, generating a signature verification result. This signature verification result characterizes whether the document meets the requirements of the external regulatory clauses at the formal compliance level. The table verification model analyzes the consistency between internal regulatory data and external regulatory requirements in terms of numerical range or logical structure information, generating a table verification result. This table verification result characterizes whether the internal regulatory document complies with the constraints of the external regulatory clauses at the structured data level. Through the collaborative processing of the semantic similarity model, signature verification model, and table verification model, a multi-dimensional comparison result including semantic difference results, signature verification results, and table verification results is generated. The multidimensional comparison results record the comparison of related internal and external regulatory clauses in different dimensions in a structured form, and serve as the basic input for subsequent compliance gap identification, gap type determination and quantitative assessment indicator calculation.

[0067] It should be noted that in some embodiments, the semantic similarity model employs a cross-encoder based on the BERT architecture. Compared to the dual-tower model, the cross-encoder can more accurately capture subtle semantic differences between two long and complex sentences. This semantic similarity model includes a clause input layer, a BERT encoding layer, and a classification layer. The process of calling the semantic similarity model to determine the semantic similarity between external and internal rule clauses, and obtaining semantic difference results, specifically includes: inputting the external and internal rule clauses into the text input layer to form clause pairs; the BERT encoding layer performing full interaction on the clause pairs to capture deep semantic interaction information and marking the corresponding context vectors in the clause pairs; and the classification layer classifying the clause pairs based on the context vectors to obtain semantic difference results.

[0068] Specifically, the clause input layer is used to uniformly format the texts of external and internal regulations, and combine them into clause pairs. These clause pairs serve as the basic input units for subsequent semantic analysis, representing the clause combination relationships that require semantic similarity determination. By forming clause pairs at the clause input layer, external and internal regulations are processed within the same semantic analysis context, providing a foundation for deep semantic interaction analysis. The BERT encoding layer performs full interaction modeling on the external and internal regulations texts in the clause pairs, capturing deep semantic interaction information at the word, phrase, and sentence levels through context association, and marking the corresponding context vectors in the clause pairs. These context vectors represent the semantic correspondence and difference features between external and internal regulations, serving as an important basis for subsequent semantic difference determination. Based on the context vectors, the classification layer processes the clause pairs through a fully connected layer and then an activation function to determine the semantic consistency or semantic difference between external and internal regulations, and outputs the semantic difference results. The activation function can be Softmax or Sigmoid; this specification does not limit the choice. The semantic difference results are used to characterize the degree of semantic compliance between internal regulatory clauses and external regulatory clauses. Therefore, they can be used for subsequent compliance gap identification, gap type determination, and quantitative assessment indicator calculation. This embodiment uses a BERT-based semantic similarity model to perform full interactive semantic modeling on clause pairs, which can effectively capture deep semantic differences and avoid misjudgments caused by relying solely on keyword matching. This solution significantly improves the accuracy of semantic similarity judgment, making the semantic difference results more consistent with the true semantic requirements of regulatory clauses, thereby improving the accuracy of compliance gap identification at the text level.

[0069] In some embodiments, the signature verification model employs a Siamese convolutional neural network with a ResNet-50 backbone, comprising a signature input layer, a feature extraction layer, and a distance metric layer. The process of calling the signature verification model to verify signatures in internal and external document data to obtain verification results specifically includes: extracting signatures from the internal and external document data and inputting them into the signature input layer to obtain signature pairs; the feature extraction layer using two ResNet networks with shared weights to extract features from the internal and external signatures in the signature pairs, obtaining corresponding high-dimensional feature vectors; and the distance metric layer calculating the similarity of the high-dimensional feature vectors to obtain the signature verification result. This structure is suitable for using metric learning to determine whether two signature images belong to the same entity or whether they have been tampered with.

[0070] Specifically, the signature input layer performs standardized formatting on the signatures in the internal and external specification documents, and combines the corresponding internal and external signatures to form signature pairs. These signature pairs serve as the basic input unit for subsequent signature consistency checks, representing the signature combination relationship requiring similarity assessment. The feature extraction layer uses two ResNet networks with shared weights to perform image feature analysis on the internal and external signatures, extracting high-dimensional feature vectors that characterize the signature's morphological, structural, and texture features. By employing shared weights, the internal and external signatures are represented in the same feature space, ensuring feature comparability. The distance metric layer calculates the similarity of the high-dimensional feature vectors to determine the degree of consistency between the internal and external signatures in terms of appearance and structural features, and generates signature verification results accordingly. The similarity calculation can be based on the Euclidean distance or cosine similarity between two feature vectors. The signature verification result can be a similarity score. If the similarity score is lower than a threshold, it is determined to be inconsistent, thus characterizing whether the document meets the formal compliance requirements of external regulations at the signature element level. Therefore, it can be used for subsequent compliance gap identification, gap type determination, and quantitative evaluation index calculation. This embodiment uses a shared-weight feature extraction network to model the features of internal and external regulatory signatures and calculates similarity based on feature distance, achieving automated verification of signature consistency. This solution can effectively identify forged, missing, or mismatched signatures, enhance the ability to review document formal compliance, reduce manual verification workload, and improve the reliability of compliance judgments.

[0071] In some embodiments, the table verification model employs a graph neural network combined with Table-BERT. The table is modeled as a graph structure, with cells as nodes and row-column relationships as edges, suitable for handling consistency of numerical ranges and logical structure. It includes a graph construction layer, a graph convolutional layer, and a consistency prediction layer. In response to table recognition, the table verification model is invoked to verify the consistency of the numerical range or logical structure information required by the internal specification data and the external specification, obtaining the table verification result. Specifically, this includes: extracting the table from the internal specification file data and the table requirements from the external specification file data and inputting them into the graph construction layer; converting table cells into node features; and converting row and column adjacency relationships into edge features. The graph convolutional layer aggregates neighbor node information, updates the representation of each cell, captures the corresponding logic of the table header values, and obtains the target node feature matrix. The consistency prediction layer compares the target node feature matrix with the numerical range features required by the external specification to obtain the table verification result.

[0072] Specifically, the graph construction layer is used to convert the table structure into a graph structure representation. It maps each cell in the table to a node in the graph and converts the numerical, text, or identifier information in the cells into node features; simultaneously, it converts the row and column relationships between cells in the table into edge features in the graph, thereby constructing graph data representing the overall structure of the table, enabling the row and column relationships and logical connections between cells to be expressed in a structured manner. The graph convolutional layer is used to aggregate the neighbor node information of each node in the graph. Through comprehensive analysis of the node features of adjacent cells, it updates the node representation corresponding to each cell. This processing method captures the correspondence between the table header and data units, as well as the logical structure information between different rows and columns, resulting in an updated node feature matrix, i.e., the target node feature matrix. This target node feature matrix is ​​used to comprehensively represent the overall characteristics of the internal specification table at the structural and content levels. The consistency prediction layer is used to compare and analyze the target node feature matrix with the numerical range features or logical structure features specified in the external specification document data to determine whether the data content and structure in the internal specification table meet the external specification requirements. Specifically, this comparison analysis is usually implemented through binary classification or regression using an MLP (Multi-Level Processing). Based on the comparison results, table verification results are generated. These results may include numerical out-of-bounds flags, logical structure missing flags, etc., and together with semantic difference results and signature verification results, they constitute the basic data for compliance gap analysis, used for subsequent compliance gap identification, gap type determination, and quantitative assessment indicator calculation. This embodiment models the table structure as a graph structure and introduces graph convolution processing to achieve a holistic understanding of the numerical relationships and logical structure in the table. It can accurately capture the correspondence between table headers and data, and perform consistency verification between internal regulations and external regulations, avoiding the inefficiency and omissions of traditional cell-by-cell comparison, thereby significantly improving the accuracy and automation level of structured data compliance verification.

[0073] In S414, after obtaining the multidimensional comparison results, the system identifies whether there is a compliance gap between the internal and external regulatory clauses based on these results. Specifically, when the multidimensional comparison results indicate that an internal regulatory clause fails to meet the requirements of an external regulatory clause in any comparison dimension, a compliance gap is determined to exist between the internal and external regulatory clauses, resulting in a compliance gap identification result. This result characterizes the degree of compliance of the internal regulatory clause with respect to the external regulatory clause and provides data support for subsequent gap type determination, weight setting, and compliance analysis report generation.

[0074] In S420, after identifying compliance gaps, each gap is categorized to determine its corresponding gap type. Specifically, based on the analysis results of the gap analysis model and the relationship information in the compliance knowledge graph, the identified compliance gaps are judged using rules to determine their types. These gap types are used to distinguish compliance issues of different natures and provide a basis for subsequent gap weighting and quantitative assessment. This ensures that the compliance gap results not only reflect the existence of non-compliance but also further clarify the nature and severity of the non-compliance. The final output of the compliance gap results includes the identified compliance gaps and their corresponding gap types, used for subsequent quantitative assessment indicator calculations and compliance analysis report generation.

[0075] Specifically, in some embodiments, the specific implementation of step S420 can be found in the following embodiments. This embodiment is based on... Figure 4 The detailed description of step S420 in the document compliance determination method shown in the corresponding embodiment includes the following steps: If the compliance gap is that no internal regulation clause associated with the external regulation clause can be found in the compliance knowledge graph, then the gap type is determined to be an omission. If the compliance gap is that the content of the internal regulation clause conflicts with the requirements of the external regulation clause, then the gap type is determined to be an error. If the compliance gap is that the binding strength of the internal regulation clause is lower than the binding strength of the external regulation clause, then the gap type is determined to be a deviation.

[0076] In this embodiment, compliance gaps are clearly categorized into three types: omissions, errors, and deviations, allowing for differentiated handling of compliance issues of different natures. This classification mechanism helps accurately reflect the degree of inadequacy of the internal regulatory system relative to external regulatory requirements and provides a direct basis for setting gap weights and assessing risk levels, thereby supporting the development of more targeted compliance rectification strategies and improving the decision-making efficiency of compliance management.

[0077] In some specific implementations, the relationship between external and internal regulatory clauses is retrieved based on the compliance knowledge graph. When no related internal regulatory clause can be found in the compliance knowledge graph for a particular external regulatory clause, it is determined that the external regulatory clause is not covered in the internal regulatory system. In this case, the compliance gap is identified as a missing internal regulatory clause, and the gap type is classified as omission. That is, the omission type is used to characterize compliance issues in the internal regulatory system that do not explicitly address the requirements of external regulations. When there are internal regulatory clauses related to external regulatory clauses in the compliance knowledge graph, the consistency of content between the internal and external regulatory clauses is further judged based on the multi-dimensional comparison results output by the gap analysis model. If the multi-dimensional comparison results show that the content of the internal regulatory clause is inconsistent with the explicit requirements of the external regulatory clause in terms of semantic meaning, logical constraints, or stipulation direction, and this inconsistency constitutes a contradictory or mutually exclusive relationship, the compliance gap is identified as a content conflict gap, and the gap type is classified as error. That is, the error type is used to characterize compliance issues where the internal regulatory clause violates or deviates from the requirements of the external regulatory clause at the content level. When internal and external regulations are consistent in content but differ in binding strength, this difference is further assessed. If the multidimensional comparison results of the gap analysis model indicate that the internal regulations impose lower binding conditions, implementation standards, or restrictions on the regulated entity than the external regulations, the compliance gap is identified as a constraint deficiency gap, and this gap type is classified as a deviation. That is, the deviation type characterizes how internal regulations respond to external regulations but fail to meet the required level of compliance stringency.

[0078] In S500, quantitative assessment indicators are calculated based on the compliance gap results, and a compliance analysis report is generated.

[0079] Specifically, based on the compliance gap results, different types of compliance gaps can be weighted to obtain quantitative assessment indicators that characterize the overall compliance status. These quantitative assessment indicators can reflect the organization's compliance level in specific business areas or at the overall level. In some embodiments, the quantitative assessment indicators can be integrated with a compliance gap list to generate a compliance analysis report. The compliance analysis report includes at least the quantitative assessment indicators and the corresponding compliance gap list, providing compliance managers with structured and quantifiable compliance analysis results.

[0080] Specifically, in some embodiments, the specific implementation of step S500 can be found in the following embodiments. This embodiment is based on... Figure 2The detailed description of step S500 in the document compliance determination method shown in the corresponding embodiment includes the following steps: Assigning corresponding weight coefficients to each compliance gap according to the gap type, wherein the weight coefficient for errors is greater than the weight coefficient for omissions, and the weight coefficient for omissions is greater than the weight coefficient for deviations. Accumulating the weighted compliance gap values ​​by combining the gap level labels corresponding to each rule clause to obtain a total gap value. Determining the quantitative evaluation index based on the difference in the total gap values. Encapsulating the quantitative evaluation index, the compliance gap results, and the compliance knowledge graph to generate a compliance analysis report.

[0081] In this embodiment, by introducing gap level labels and weighting coefficients, compliance gaps of different types and importance are quantified, enabling a numerical assessment of compliance risks. This solution can transform complex compliance issues into comparable quantitative indicators and present them centrally through compliance analysis reports, thereby providing intuitive and objective data support for continuous compliance monitoring, trend analysis, and management decision-making.

[0082] In some specific implementations, rule clauses are assigned gap level labels when generating structured compliance rule sets. These gap level labels characterize the importance or risk level of the corresponding rule clause within the overall compliance system. During quantitative assessment, each compliance gap is assigned a corresponding weight coefficient based on its identified gap type. Specifically, higher weight coefficients are assigned to error-type compliance gaps; second-highest weight coefficients are assigned to omission-type compliance gaps; and relatively lower weight coefficients are assigned to deviation-type compliance gaps. This differentiated weighting ensures that different types of compliance gaps have varying degrees of impact in the quantitative assessment. The weight coefficients of each compliance gap are combined with the gap level labels of their corresponding rule clauses to perform weighted processing on each compliance gap. Specifically, based on the gap level labels corresponding to the rule clauses, the weighted compliance gaps are quantified, and the gap values ​​of each compliance gap are summed to obtain a total gap value reflecting the overall compliance status. This total gap value comprehensively characterizes the overall compliance deviation of the internal regulatory system from external regulatory requirements. Based on the difference in the total gap value, quantitative assessment indicators are determined. The quantitative assessment indicators are used to reflect the degree of compliance between internal regulations and external regulations in numerical form. By comparing the total gap values ​​of different documents or at different time points, a quantitative assessment of changes in compliance levels can be achieved. The quantitative assessment indicators, the compliance gap results, and the compliance knowledge graph are encapsulated to generate a compliance analysis report. The compliance analysis report is used to centrally present the analysis results of document compliance determination, including at least the quantitative assessment indicators, a compliance gap list, and information on the correlation between external and internal regulations. Through the compliance analysis report, the compliance analysis results can be output in a structured form, facilitating subsequent compliance review, rectification decisions, and risk management.

[0083] The following will combine Figure 6 This manual provides a detailed introduction to the document compliance assessment device provided. It should be noted that... Figure 6 The document compliance determination device shown is used to execute this specification. Figures 1-5 The methods of the embodiments shown are illustrated only in connection with this specification for ease of explanation. For specific technical details not disclosed, please refer to this specification. Figures 1-5 The example shown.

[0084] Please see Figure 6This diagram illustrates the structure of the document compliance determination device 600 described in this specification. The document compliance determination device 600 can be implemented as all or part of a user terminal through software, hardware, or a combination of both. According to some embodiments, the document compliance determination device 600 includes a multimodal parsing module 610, a deep semantic analysis module 620, a knowledge graph maintenance module 630, a multi-dimensional comparison module 640, and a quantitative evaluation calculation module 650.

[0085] The multimodal parsing module 610 utilizes a multimodal compliance big data model to perform multimodal parsing on external and internal regulation document data, extracting multimodal feature data, including text content, logical structure information, and image features. The deep semantic analysis module 620 performs natural language processing on the text content, extracting key elements based on subject-verb-object parsing logic to generate a structured compliance rule set. The rule clauses in the compliance rule set include external regulation clauses and internal regulation clauses, each with a corresponding tag. The external regulation clauses are the clauses in the external regulation document data and the internal regulation clauses. The compliance knowledge graph is a set of internal regulations document data. The knowledge graph maintenance module 630 maintains a compliance knowledge graph based on the compliance rule set, which includes the relationships between the external regulations clauses and the internal regulations clauses. The multi-dimensional comparison module 640 performs a multi-dimensional comparison of the related internal and external regulations clauses using a gap analysis model based on these relationships to obtain compliance gap results. The quantitative assessment calculation module 650 calculates quantitative assessment indicators based on the compliance gap results and generates a compliance analysis report, which includes the quantitative assessment indicators and a compliance gap list.

[0086] Optionally, the multimodal compliance big model includes a visual encoder and a language big model, and the multimodal feature data includes image features, text content, and logical structure information; the multimodal parsing module 610 specifically includes: a visual encoding submodule, used to extract entity features from the external and internal compliance document data using the visual encoder to obtain image features of visual entities, including seals, signatures, and tables; and a text recognition submodule, used to perform text recognition and semantic understanding on the external and internal compliance document data using the language big model to obtain text content, and to fuse and align the image features of the visual entities with the text content to determine the logical structure information between the visual entities and the text content.

[0087] Optionally, the visual encoder includes an image input layer, a slice segmentation layer, and multiple stages of sliding window transformation blocks; the visual encoding submodule specifically includes: an image input unit, used to input the outer guide file data and the inner guide file data into the image input layer in the form of images to obtain the corresponding three-dimensional outer guide tensor and three-dimensional inner guide tensor; a slice segmentation unit, used to segment the three-dimensional outer guide tensor and the three-dimensional inner guide tensor into multiple non-overlapping image blocks through the slice segmentation layer to obtain the corresponding image block sequence vector; and a sliding window transformation unit, used to extract entity features from the image blocks in the image block sequence vector one by one according to the stage sequence through multiple stages of sliding window transformation blocks to obtain the image features of visual entities.

[0088] Optionally, the large language model includes an embedding layer, a concatenation input layer, a decoding layer, a multimodal alignment layer, and an output layer; the text recognition submodule specifically includes: a text embedding unit, used to input the external and internal specification file data into the embedding layer after OCR pre-recognition to obtain a text embedding vector sequence; a multimodal alignment unit, used to compress and transform the image features of the visual entities through the multimodal alignment layer to obtain a visual embedding vector sequence; a concatenation input unit, used to input the text embedding vector sequence and the visual embedding vector sequence into the concatenation input layer for concatenation to obtain a combined sequence; a fusion processing unit, used by the decoding layer to perform fusion processing on the combined sequence through a multi-head self-attention mechanism and a feedforward neural network to obtain a hidden layer state vector; and a layer normalization unit, used by the output layer to perform layer normalization and linear transformation on the hidden layer state vector to obtain structured text content and logical structure information.

[0089] Optionally, the deep semantic analysis module 620 specifically includes: a syntactic analysis submodule, used to perform syntactic analysis on the clause statements in the text content, extracting the subject as the regulatory object, the predicate as the compliance action, and the object as the action recipient; a rule determination submodule, used to determine whether each regulatory object conforms to a preset rule; a grammatical combination submodule, used to combine the corresponding subject, predicate, and object to form a rule clause if the regulatory object conforms to the preset rule; and a tag matching submodule, used to match corresponding tags for each rule clause to generate the compliance rule set.

[0090] Optionally, the knowledge graph maintenance module 630 specifically includes: a node instantiation submodule, used to set each rule clause in the compliance rule set as an entity node in the graph; a node edge establishment submodule, used to establish edges between entity nodes according to the reference relationship, hierarchical relationship and the association relationship between each rule clause, wherein the edges include at least one of constraint relationship, implementation relationship and conflict relationship; and a node dynamic addition submodule, used to dynamically add entity nodes in response to the receipt of external regulation document data, and update the existing relationship network between entity nodes according to semantic reasoning.

[0091] Optionally, the multi-dimensional comparison module 640 specifically includes: a compliance gap identification submodule, used to perform multi-dimensional comparison of the associated internal and external regulatory clauses based on the relationship and using a gap analysis model to identify the compliance gaps between them; and a compliance gap classification submodule, used to classify each compliance gap and determine the corresponding gap type.

[0092] Optionally, the compliance gap identification submodule specifically includes: a multi-dimensional comparison unit, used to perform multi-dimensional comparison of related internal and external regulatory clauses using a gap analysis model to obtain multi-dimensional comparison results; and a gap identification unit, used to identify the compliance gap between the two based on the multi-dimensional comparison results.

[0093] Optionally, the gap analysis model includes a semantic similarity model, a signature verification model, and a table verification model; the multidimensional comparison results include semantic difference results, signature verification results, and table verification results; the multidimensional comparison unit specifically includes: a semantic similarity subunit, used to call the semantic similarity model to determine the semantic similarity between external and internal regulations, and obtain semantic difference results; a signature verification subunit, used to call the signature verification model to verify the signatures in the internal and external regulations, and obtain signature verification results; and a table verification subunit, used to respond to the identification of tables, call the table verification model to verify the consistency of the numerical range or logical structure information required by the internal and external regulations, and obtain table verification results.

[0094] Optionally, the compliance gap classification submodule specifically includes: an omission determination unit, used to determine the gap type as omission if the compliance gap is that no internal regulation clause associated with the external regulation clause can be found in the compliance knowledge graph; an error determination unit, used to determine the gap type as error if the compliance gap is that the content of the internal regulation clause conflicts with the requirements of the external regulation clause; and a deviation determination unit, used to determine the gap type as deviation if the compliance gap is that the binding strength of the internal regulation clause is lower than the binding strength of the external regulation clause.

[0095] Optionally, the quantitative assessment calculation module 650 specifically includes: a weighting submodule, used to assign corresponding weight coefficients to each compliance gap according to the gap type, wherein the weight coefficient for errors is greater than the weight coefficient for omissions, and the weight coefficient for omissions is greater than the weight coefficient for deviations; a gap accumulation submodule, used to accumulate the gap values ​​of the weighted compliance gaps by combining the gap level labels corresponding to each rule clause, to obtain a total gap value; an indicator determination submodule, used to determine the quantitative assessment indicator based on the difference of the total gap value; and a report encapsulation submodule, used to encapsulate the quantitative assessment indicator, the compliance gap results, and the compliance knowledge graph to generate a compliance analysis report.

[0096] It should be noted that the document compliance determination device provided in the above embodiments is only illustrated by the division of the above functional modules when executing the document compliance determination method. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. In addition, the document compliance determination device and the document compliance determination method embodiments provided in the above embodiments belong to the same concept, and the implementation process is detailed in the method embodiments, which will not be repeated here.

[0097] The serial numbers in this specification are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0098] This specification introduces a multimodal compliance model to uniformly analyze and compare external and internal compliance documents, incorporating text, logical structure, and image elements into a single analytical framework to achieve a systematic assessment of document compliance. By constructing a structured compliance rule set and a compliance knowledge graph, and combining this with a gap analysis model for gap identification and quantitative evaluation, compliance analysis is transformed from traditional manual comparison to automated, structured, and quantifiable processing. This significantly improves the comprehensiveness, accuracy, and efficiency of compliance reviews, while reducing the risks associated with human error.

[0099] This specification also provides a computer storage medium capable of storing multiple instructions adapted to be loaded and executed by a processor as described above. Figures 1-5 The document compliance determination method described in the illustrated embodiment can be found in the following documentation for its specific execution process. Figures 1-5 The specific details of the illustrated embodiments will not be elaborated here.

[0100] This specification also provides a computer program product that stores at least one instruction, said at least one instruction being loaded and executed by the processor as described above. Figures 1-5 The document compliance determination method described in the illustrated embodiment can be found in the following documentation for its specific execution process. Figures 1-5 The specific details of the illustrated embodiments will not be elaborated here.

[0101] Please refer to Figure 7This diagram illustrates a structural block diagram of an electronic device provided in an exemplary embodiment of this specification. The electronic device in this specification may include one or more components such as a processor 110, a memory 120, an input device 130, an output device 140, and a bus 150. The processor 110, memory 120, input device 130, and output device 140 are connected via the bus 150. The processor 110 may include one or more processing cores. The processor 110 connects various parts within the electronic device using various interfaces and lines, and performs various functions and processes data by running or executing instructions, programs, code sets, or instruction sets stored in the memory 120, and by calling data stored in the memory 120. Optionally, the processor 110 may be implemented using at least one hardware form of digital signal processing (DSP), field-programmable gate array (FPGA), or programmable logic array (PLA). The processor 110 may integrate one or a combination of central processing unit (CPU), graphics processing unit (GPU), and modem. The CPU primarily handles the operating system, user interface, and applications; the GPU is responsible for rendering and drawing the displayed content; and the modem handles wireless communication. It is understood that the modem may not be integrated into the processor 110 and may be implemented using a separate communication chip. The memory 120 may include random access memory (RAM) or read-only memory (ROM). Optionally, the memory 120 may include non-transitory computer-readable storage medium. The memory 120 can be used to store instructions, programs, code, code sets, or instruction sets. The memory 120 may include a program storage area and a data storage area. The program storage area may store instructions for implementing the operating system, instructions for implementing at least one function (such as touch functionality, sound playback functionality, image playback functionality, etc.), instructions for implementing the various method embodiments described below, etc. The operating system may be the Android system, including systems deeply developed based on the Android system, the iOS system developed by Apple, including systems deeply developed based on the iOS system, or other systems. The data storage area can also store data created during the use of electronic devices, such as phonebook entries, audio and video data, chat logs, etc.

[0102] See Figure 8 As shown, the memory 120 can be divided into operating system space and user space. The operating system runs in the operating system space, while native and third-party applications run in the user space. To ensure that different third-party applications can achieve good running performance, the operating system allocates corresponding system resources for each third-party application. However, different application scenarios within the same third-party application have different requirements for system resources. For example, in local resource loading scenarios, third-party applications have higher requirements for disk read speed; in animation rendering scenarios, third-party applications have higher requirements for GPU performance. Since the operating system and third-party applications are independent of each other, the operating system often cannot promptly perceive the current application scenario of a third-party application, resulting in the operating system's inability to perform targeted system resource adaptation based on the specific application scenario. To enable the operating system to distinguish the specific application scenario of a third-party application, data communication between the third-party application and the operating system needs to be established, allowing the operating system to obtain the current scenario information of the third-party application at any time, and then perform targeted system resource adaptation based on the current scenario. Taking the Android system as an example, the programs and data stored in the memory 120 are as follows... Figure 9As shown, the memory 120 can store the Linux kernel layer 320, the system runtime library layer 340, the application framework layer 360, and the application layer 380. The Linux kernel layer 320, system runtime library layer 340, and application framework layer 360 belong to the operating system space, while the application layer 380 belongs to the user space. The Linux kernel layer 320 provides low-level drivers for various hardware components of the electronic device, such as display drivers, audio drivers, camera drivers, Bluetooth drivers, Wi-Fi drivers, and power management. The system runtime library layer 340 provides support for key features of the Android system through several C / C++ libraries. For example, the SQLite library provides database support, the OpenGL / ES library provides 3D graphics support, and the Webkit library provides browser kernel support. The system runtime library layer 340 also provides the Android runtime library, which mainly provides core libraries that allow developers to write Android applications using the Java language. The application framework layer 360 provides various APIs that may be used when building applications. Developers can also use these APIs to build their own applications, such as activity management, window management, view management, notification management, content provider, package management, call management, resource management, and location management. At least one application runs in the application layer 380. These applications can be native applications that come with the operating system, such as contacts, SMS, clock, and camera apps; or third-party applications developed by third-party developers, such as games, instant messaging, and photo editing apps. Taking iOS as an example, the programs and data stored in memory 120 are as follows... Figure 10As shown, the iOS system comprises: Core OS layer 420, Core Services layer 440, Media layer 460, and Cocoa Touch Layer 480. Core OS layer 420 includes the operating system kernel, drivers, and low-level program frameworks. These frameworks provide hardware-level functionality for use by the program frameworks in Core Services layer 440. Core Services layer 440 provides applications with the system services and / or program frameworks they require, such as Foundation framework, account framework, advertising framework, data storage framework, network connectivity framework, geolocation framework, motion framework, etc. Media layer 460 provides applications with audiovisual interfaces, such as interfaces related to graphics and images, audio technology, video technology, and AirPlay (wireless audio / video transmission). Cocoa Touch Layer 480 provides various commonly used interface-related frameworks for application development and is responsible for user touch interaction on electronic devices. Examples include local notification services, remote push services, advertising frameworks, game tool frameworks, message user interface (UI) frameworks, user interface UIKit frameworks, map frameworks, and so on.

[0103] exist Figure 10The framework shown includes, but is not limited to, the basic framework in the core service layer 440 and the UIKit framework in the touchable layer 480, which are relevant to most applications. The basic framework provides many fundamental object classes and data types, offering essential system services to all applications, and is independent of the UI. The UIKit framework provides a basic UI class library for creating touch-based user interfaces. iOS applications can use the UIKit framework to provide their UI, thus providing the application's infrastructure for building user interfaces, drawing, handling user interaction events, responding to gestures, etc. The methods and principles for implementing data communication between third-party applications and the operating system in the iOS system can be referenced from the Android system, and will not be elaborated upon here. The input device 130 is used to receive input instructions or data, and includes, but is not limited to, a keyboard, mouse, camera, microphone, or touch device. The output device 140 is used to output instructions or data, and includes, but is not limited to, display devices and speakers. In one example, input device 130 and output device 140 may be combined. Input device 130 and output device 140 are touch displays used to receive touch operations from the user using a finger, stylus, or any suitable object on or near the display, and to display the user interface of various applications. Touch displays are typically located on the front panel of the electronic device. Touch displays can be designed as full-screen, curved screen, or irregularly shaped screen. Touch displays can also be designed as a combination of full-screen and curved screen, or a combination of irregularly shaped and curved screen; this specification does not limit this. Furthermore, those skilled in the art will understand that the structure of the electronic device shown in the above figures does not constitute a limitation on the electronic device. The electronic device may include more or fewer components than shown, or combine certain components, or have different component arrangements. For example, the electronic device may also include radio frequency circuits, input units, sensors, audio circuits, wireless fidelity (WiFi) modules, power supplies, Bluetooth modules, etc., which will not be elaborated further here. In this specification, the entity executing each step can be the electronic device described above. Optionally, the entity executing each step can be the operating system of the electronic device. The operating system can be Android, iOS, or other operating systems; this manual does not limit this.The electronic device described in this manual may also be equipped with a display device. This display device can be any device capable of displaying information, such as a cathode ray tube display (CR), a light-emitting diode display (LED), an e-ink screen, a liquid crystal display (LCD), or a plasma display panel (PDP). Users can use the display device on the electronic device to view displayed text, images, videos, and other information. The electronic device may be a smartphone, tablet, gaming device, AR (Augmented Reality) device, automobile, data storage device, audio playback device, video playback device, laptop, desktop computing device, or wearable device such as a smartwatch, smart glasses, smart helmet, smart bracelet, smart necklace, or smart clothing.

[0104] exist Figure 7 In the illustrated electronic device, which can be a terminal, the processor 110 can be used to call the network optimization application stored in the memory 120 and specifically perform the following operations: Using a multimodal compliance model, perform multimodal parsing on external and internal regulatory document data to extract multimodal feature data, including text content, logical structure information, and image features; perform deep semantic analysis of the text content using natural language, extract key elements based on subject-verb-object parsing logic, and generate a structured compliance rule set, wherein the rule clauses in the compliance rule set include external and internal regulatory clauses, and each rule clause has a corresponding label; the external regulatory clauses are clauses in the external regulatory document data, and the internal regulatory clauses are clauses in the internal regulatory document data; maintain a compliance knowledge graph based on the compliance rule set, the compliance knowledge graph containing the relationships between the external and internal regulatory clauses; based on the relationships, use a gap analysis model to perform multi-dimensional comparisons of the associated internal and external regulatory clauses to obtain compliance gap results; calculate quantitative evaluation indicators based on the compliance gap results, and generate a compliance analysis report, the compliance analysis report including the quantitative evaluation indicators and a compliance gap list.

[0105] In one embodiment, the multimodal compliance big model includes a visual encoder and a language big model, and the multimodal feature data includes image features, text content, and logical structure information. When the processor 110 performs multimodal parsing of the external and internal compliance document data using the multimodal compliance big model to extract multimodal feature data, it specifically performs the following operations: using the visual encoder to extract entity features from the external and internal compliance document data to obtain image features of visual entities, including seals, signatures, and tables; using the language big model to perform text recognition and semantic understanding on the external and internal compliance document data to obtain text content, and fusing and aligning the image features of the visual entities with the text content to determine the logical structure information between the visual entities and the text content.

[0106] In one embodiment, the visual encoder includes an image input layer, a slice segmentation layer, and a sliding window transformation block with multiple stages. When the processor 110 performs entity feature extraction on the outer guide file data and the inner guide file data using the visual encoder to obtain image features of visual entities, it specifically performs the following operations: inputting the outer guide file data and the inner guide file data into the image input layer in the form of images to obtain corresponding three-dimensional outer guide tensors and three-dimensional inner guide tensors; segmenting the three-dimensional outer guide tensors and three-dimensional inner guide tensors into multiple non-overlapping image blocks through the slice segmentation layer to obtain corresponding image block sequence vectors; and extracting entity features from the image blocks in the image block sequence vector one by one according to the stage sequence through the sliding window transformation block with multiple stages to obtain image features of visual entities.

[0107] In one embodiment, the language big model includes an embedding layer, a concatenation input layer, a decoding layer, a multimodal alignment layer, and an output layer. When the processor 110 performs text recognition and semantic understanding on the external and internal specification file data using the language big model to obtain text content, and fuses and aligns the image features of the visual entities with the text content to determine the logical structure information between the visual entities and the text content, the processor 110 specifically performs the following operations: The external and internal specification file data are pre-recognized by OCR and then input into the embedding layer to obtain a text embedding vector sequence; the image features of the visual entities are compressed and dimensionally transformed by the multimodal alignment layer to obtain a visual embedding vector sequence; the text embedding vector sequence and the visual embedding vector sequence are concatenated into the concatenation input layer to obtain a combined sequence; the decoding layer fuses the combined sequence using a multi-head self-attention mechanism and a feedforward neural network to obtain a hidden state vector; the output layer performs layer normalization and linear transformation on the hidden state vector to obtain structured text content and logical structure information.

[0108] In one embodiment, when the processor 110 performs deep semantic analysis of the text content using natural language, extracts key elements based on subject-verb-object parsing logic, and generates a structured compliance rule set, it specifically performs the following operations: performs syntactic analysis on the clause statements in the text content, extracting the subject as the regulatory object, the predicate as the compliance action, and the object as the action recipient; determines whether each regulatory object conforms to a preset rule; if the regulatory object conforms to the preset rule, combines the corresponding subject, predicate, and object to form a rule clause; matches corresponding tags to each rule clause to generate the compliance rule set.

[0109] In one embodiment, when the processor 110 performs the maintenance of the compliance knowledge graph based on the compliance rule set, it specifically performs the following operations: instantiating each rule clause in the compliance rule set into an entity node in the graph; establishing edges between entity nodes according to the reference relationship, hierarchical relationship and the association relationship between each rule clause, wherein the edges include at least one of constraint relationship, implementation relationship and conflict relationship; dynamically adding entity nodes in response to receiving external regulation document data, and updating the existing relationship network between entity nodes according to semantic reasoning.

[0110] In one embodiment, when the processor 110 performs the following operations to obtain compliance gap results by comparing the related internal and external regulatory clauses in multiple dimensions using a gap analysis model based on the relationship: comparing the related internal and external regulatory clauses in multiple dimensions based on the relationship and identifying the compliance gaps between them; classifying each compliance gap and determining the corresponding gap type.

[0111] In one embodiment, when the processor 110 performs the multi-dimensional comparison of related internal and external regulatory clauses using the gap analysis big model to identify the compliance gaps between them, it specifically performs the following operations: using the gap analysis big model to perform a multi-dimensional comparison of related internal and external regulatory clauses to obtain multi-dimensional comparison results; and identifying the compliance gaps between them based on the multi-dimensional comparison results.

[0112] In one embodiment, the gap analysis model includes a semantic similarity model, a signature verification model, and a table verification model; the multidimensional comparison results include semantic difference results, signature verification results, and table verification results; when the processor 110 performs multidimensional comparison of related internal and external regulations using the gap analysis model to obtain multidimensional comparison results, it specifically performs the following operations: calling the semantic similarity model to determine the semantic similarity between the external and internal regulations, and obtaining semantic difference results; calling the signature verification model to verify the signatures in the internal and external regulations, and obtaining signature verification results; and responding to the identification of tables, calling the table verification model to verify the consistency of the numerical range or logical structure information required by the internal and external regulations, and obtaining table verification results.

[0113] In one embodiment, when the processor 110 performs the classification of each compliance gap and determines the corresponding gap type, it specifically performs the following operations: if the compliance gap is that no internal regulation clause associated with the external regulation clause can be found in the compliance knowledge graph, then the gap type is determined to be an omission; if the compliance gap is that the content of the internal regulation clause conflicts with the requirements of the external regulation clause, then the gap type is determined to be an error; if the compliance gap is that the binding strength of the internal regulation clause is lower than the binding strength of the external regulation clause, then the gap type is determined to be a deviation.

[0114] In one embodiment, the label includes a gap level label. When the processor 110 executes the step of calculating a quantitative assessment indicator based on the compliance gap results and generating a compliance analysis report, it specifically performs the following operations: assigning a corresponding weight coefficient to each compliance gap according to the gap type, wherein the weight coefficient for errors is greater than the weight coefficient for omissions, and the weight coefficient for omissions is greater than the weight coefficient for deviations; combining the gap level labels corresponding to each rule clause, summing the weighted compliance gap values ​​to obtain a total gap value; determining the quantitative assessment indicator based on the difference in the total gap value; and encapsulating the quantitative assessment indicator, the compliance gap results, and the compliance knowledge graph to generate a compliance analysis report.

[0115] This specification introduces a multimodal compliance model to uniformly analyze and compare external and internal compliance documents, incorporating text, logical structure, and image elements into a single analytical framework to achieve a systematic assessment of document compliance. By constructing a structured compliance rule set and a compliance knowledge graph, and combining this with a gap analysis model for gap identification and quantitative evaluation, compliance analysis is transformed from traditional manual comparison to automated, structured, and quantifiable processing. This significantly improves the comprehensiveness, accuracy, and efficiency of compliance reviews, while reducing the risks associated with human error.

[0116] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory, or random access memory, etc.

[0117] It should be noted that the information (including but not limited to user device information, user personal information, etc.), data (including but not limited to data used for analysis, stored data, displayed data, etc.), and signals involved in the embodiments of this specification are all authorized by the user or fully authorized by all parties, and the collection, use, and processing of related data must comply with the relevant laws, regulations, and standards of the relevant countries and regions. For example, the object characteristics, interactive behavior characteristics, and user information involved in this specification were all obtained under full authorization.

[0118] The above-disclosed embodiments are merely preferred embodiments of this specification and should not be construed as limiting the scope of this specification. Therefore, any equivalent variations made in accordance with the claims of this specification shall still fall within the scope of this specification.

Claims

1. A method for determining document compliance, characterized in that, The method includes: A multimodal compliance big data model is used to perform multimodal analysis on external and internal compliance document data to extract multimodal feature data, which includes text content, logical structure information and image features. Natural language deep semantic analysis is performed on the text content, key elements are extracted based on subject-verb-object parsing logic, and a structured compliance rule set is generated. The rule clauses in the compliance rule set include external rule clauses and internal rule clauses, and each rule clause has a corresponding tag. The external rule clauses are clauses in the external rule document data, and the internal rule clauses are clauses in the internal rule document data. A compliance knowledge graph is maintained based on the aforementioned compliance rule set, and the compliance knowledge graph contains the relationships between the external regulatory clauses and the internal regulatory clauses; Based on the aforementioned relationship, a large gap analysis model is used to compare the related internal and external regulatory clauses from multiple dimensions to obtain compliance gap results. Based on the compliance gap results, quantitative assessment indicators are calculated, and a compliance analysis report is generated. The compliance analysis report includes the quantitative assessment indicators and a list of compliance gaps. The gap analysis model includes a semantic similarity model, a signature verification model, and a table verification model. The specific steps for obtaining compliance gap results by using a gap analysis model to compare the related internal and external regulatory clauses from multiple dimensions based on the aforementioned relationship include: The semantic similarity model is invoked to determine the semantic similarity between external and internal regulations, and the semantic difference results are obtained. The signature verification model is invoked to verify the signatures in internal and external regulatory document data, and the signature verification results are obtained. In response to the recognition of the table, the table verification model is invoked to verify the consistency between the internal specification data and the numerical range or logical structure information required by the external specification, and the table verification result is obtained.

2. The method according to claim 1, characterized in that, The multimodal compliance big model includes a visual encoder and a language big model, and the multimodal feature data includes image features, text content, and logical structure information; The process of using a multimodal compliance big data model to perform multimodal analysis on the external and internal compliance document data and extract multimodal feature data specifically includes: A visual encoder is used to extract entity features from the external and internal specification document data to obtain image features of visual entities, including seals, signatures, and tables. The language big data model is used to perform text recognition and semantic understanding on the external and internal data to obtain the text content. The image features of the visual entities are then fused and aligned with the text content to determine the logical structure information between the visual entities and the text content.

3. The method according to claim 2, characterized in that, The visual encoder includes an image input layer, a slice segmentation layer, and a sliding window transform block with multiple stages. The step of using a visual encoder to extract entity features from the outer and inner specification file data to obtain image features of visual entities specifically includes: The outer guide file data and the inner guide file data are input into the image input layer in the form of images to obtain the corresponding three-dimensional outer guide tensor and three-dimensional inner guide tensor. The slicing layer divides the three-dimensional outer gauge tensor and the three-dimensional inner gauge tensor into multiple non-overlapping image blocks to obtain the corresponding image block sequence vector. By using multiple stages of sliding window transformation blocks to extract entity features from the image blocks in the image block sequence vector one by one in the stage order, the image features of the visual entities are obtained.

4. The method according to claim 2, characterized in that, The large language model includes an embedding layer, a concatenated input layer, a decoding layer, a multimodal alignment layer, and an output layer. The process of using a large language model to perform text recognition and semantic understanding on the external and internal rule file data to obtain text content, and fusing and aligning the image features of the visual entities with the text content to determine the logical structure information between the visual entities and the text content, specifically includes: The external and internal specification file data are pre-recognized by OCR and then input into the embedding layer to obtain a text embedding vector sequence. The image features of the visual entity are compressed and dimensionally transformed by the multimodal alignment layer to obtain a visual embedding vector sequence; The text embedding vector sequence and the visual embedding vector sequence are input to the concatenation layer and then concatenated to obtain a combined sequence. The decoding layer fuses the combined sequence through a multi-head self-attention mechanism and a feedforward neural network to obtain a hidden layer state vector; The output layer performs layer normalization and linear transformation on the hidden layer state vector to obtain structured text content and logical structure information.

5. The method according to claim 1, characterized in that, The process of performing deep semantic analysis of the text content using natural language, extracting key elements based on subject-verb-object parsing logic, and generating a structured set of compliance rules specifically includes: Syntactic analysis is performed on the clauses in the text to extract the subject as the regulatory object, the predicate as the compliance action, and the object as the action recipient; Determine whether each of the monitored objects conforms to the preset rules; If the regulated object meets the preset rules, the corresponding subject, predicate and object will be combined to form rule clauses; Match the corresponding tags to each rule clause to generate the compliance rule set.

6. The method according to claim 1, characterized in that, The maintenance of the compliance knowledge graph based on the compliance rule set specifically includes: Instantiate each rule clause in the compliance rule set as an entity node in the graph; Based on the reference relationships, hierarchical relationships, and association relationships among the various rule clauses, edges are established between entity nodes, and the edges include at least one of the constraint relationships, implementation relationships, and conflict relationships. In response to receiving external specification document data, entity nodes are dynamically added, and the existing relationship network between entity nodes is updated based on semantic reasoning.

7. The method according to claim 1, characterized in that, Based on the aforementioned relationship, a large-scale gap analysis model is used to compare the related internal and external regulatory clauses from multiple dimensions to obtain compliance gap results, specifically including: Based on the aforementioned relationship, a gap analysis model is used to compare the related internal and external regulatory clauses from multiple dimensions to identify the compliance gaps between them. Classify the compliance gaps and determine the corresponding gap types.

8. The method according to claim 7, characterized in that, The method utilizes a gap analysis model to perform multi-dimensional comparisons of related internal and external regulatory clauses to identify compliance gaps between them, specifically including: The gap analysis model is used to compare the relevant internal and external regulatory clauses from multiple dimensions, and the results of the multidimensional comparison are obtained. Based on the multidimensional comparison results, compliance gaps between the two are identified.

9. The method according to claim 8, characterized in that, The multidimensional comparison results include semantic difference results, signature verification results, and table verification results.

10. The method according to claim 7, characterized in that, The classification of compliance gaps and determination of corresponding gap types specifically includes: If the compliance gap is that no internal regulation clause associated with the external regulation clause can be found in the compliance knowledge graph, then the gap type is determined to be an omission. If the compliance gap is a conflict between the content of internal regulations and the requirements of external regulations, then the gap type will be judged as an error. If the compliance gap is that the binding strength of the internal regulations is lower than that of the external regulations, then the gap type is determined to be a deviation.

11. The method according to claim 1, characterized in that, The labels include gap level labels, the compliance gap results include gap types, and the calculation of quantitative assessment indicators and generation of a compliance analysis report based on the compliance gap results specifically includes: Each compliance gap is assigned a corresponding weight coefficient according to the gap type, wherein the weight coefficient for errors is greater than the weight coefficient for omissions, and the weight coefficient for omissions is greater than the weight coefficient for deviations. By combining the gap level labels corresponding to each rule clause, the weighted compliance gaps are summed to obtain the total gap value. The quantitative evaluation index is determined based on the difference in the total gap value; The quantitative assessment indicators, the compliance gap results, and the compliance knowledge graph are encapsulated to generate a compliance analysis report.

12. A document compliance determination device, characterized in that, The device includes: The multimodal parsing module is used to perform multimodal parsing on external and internal compliance document data using the multimodal compliance big model, and extract multimodal feature data, which includes text content, logical structure information and image features; The deep semantic analysis module is used to perform natural language processing on the text content, extract key elements based on subject-verb-object parsing logic, and generate a structured compliance rule set. The rule clauses in the compliance rule set include external rule clauses and internal rule clauses, and each rule clause has a corresponding tag. The external rule clauses are clauses in the external rule document data, and the internal rule clauses are clauses in the internal rule document data. The knowledge graph maintenance module is used to maintain a compliance knowledge graph based on the compliance rule set, wherein the compliance knowledge graph contains the relationship between the external regulation clauses and the internal regulation clauses; The multi-dimensional comparison module is used to compare the related internal and external regulatory clauses in multiple dimensions based on the aforementioned relationship and using a gap analysis model to obtain compliance gap results. The quantitative assessment calculation module is used to calculate quantitative assessment indicators based on the compliance gap results and generate a compliance analysis report, which includes the quantitative assessment indicators and a compliance gap list. The gap analysis model includes a semantic similarity model, a signature verification model, and a table verification model. The multi-dimensional comparison module specifically includes: The semantic similarity subunit is used to call the semantic similarity model to determine the semantic similarity between external and internal rules and obtain semantic difference results. The signature verification subunit is used to call the signature verification model to verify the signatures in the internal and external document data and obtain the signature verification results. The table verification subunit is used to respond to the recognition of the table, call the table verification model, verify the consistency between the internal specification data and the numerical range or logical structure information required by the external specification, and obtain the table verification result.

13. A computer storage medium, characterized in that, The computer storage medium stores a plurality of instructions adapted for loading by a processor and executing the method steps as claimed in any one of claims 1 to 11.

14. A computer program product, characterized in that, The computer program product stores at least one instruction, which is loaded by a processor and executed as a method step as claimed in any one of claims 1 to 11.

15. An electronic device, characterized in that, include: A processor and a memory; wherein the memory stores a computer program adapted to be loaded by the processor and executed the method steps as claimed in any one of claims 1 to 11.

Images