Contract data hierarchical protection method and device and computer storage medium

Abstract

This application discloses a method, apparatus, and computer storage medium for graded protection of contract data, relating to the field of data security technology. The method includes: extracting at least one pair of contract elements from a contract dataset; determining the target association strength of the contract element pair through a dual-path association extraction mechanism; dynamically adjusting the initial semantic sensitivity of a first contract element based on the target association strength of the contract element pair to obtain the target semantic sensitivity of the first contract element; determining a first graded result for the first contract element based on the target semantic sensitivity of the first contract element; and determining a first security protection strategy for the first contract element based on the first graded result. This application solves the technical problem in the prior art where sensitivity assessment is based solely on static keywords or single attributes, without considering the dynamic transmission effect of semantic associations between contract elements, leading to inaccurate identification of sensitive data and a disconnect between the graded results and actual security risks.

Description

Technical Field

[0001] This application relates to the field of data security technology, and more specifically, to a method, apparatus, and computer storage medium for hierarchical protection of contract data. Background Technology

[0002] Currently, sensitivity assessment of contract data largely relies on static keyword matching or isolated judgments of single attributes, such as directly labeling ID numbers as highly sensitive or classifying contract amounts according to fixed thresholds. However, such methods cannot identify semantic relationships across fields in contract text, especially in scenarios with implicit coupling, such as when precision instruments and high insured amounts appear simultaneously. Traditional methods cannot identify the transmission effect of their combined risks, resulting in the assessment results of sensitive elements only reflecting the static characteristics of isolated attributes, failing to reflect their true risk level in the semantic context. Consequently, the sensitivity data classification results do not match the actual security threat level, making it difficult to support the deployment of refined dynamic protection strategies.

[0003] There is currently no effective solution to the above problems. Summary of the Invention

[0004] This application provides a method, apparatus, and computer storage medium for graded protection of contract data, which at least solves the technical problems in the prior art that sensitivity assessment is based solely on static keywords or single attributes without considering the dynamic transmission effect of semantic relationships between contract elements, resulting in inaccurate identification of sensitive data and a disconnect between the graded results and actual security risks.

[0005] According to one aspect of the embodiments of this application, a method for hierarchical protection of contract data is provided, comprising: extracting at least one pair of contract elements based on a contract dataset, wherein the pair of contract elements consists of a first contract element and a second contract element that have potential semantic association, and the initial semantic sensitivity of the first contract element is lower than that of the second contract element; determining the target association strength of the pair of contract elements through a dual-path association extraction mechanism, wherein the dual-path association extraction mechanism is used to collaboratively identify the association relationship between the first contract element and the second contract element through explicit rule paths and implicit semantic paths; dynamically adjusting the initial semantic sensitivity of the first contract element according to the target association strength of the pair of contract elements to obtain the target semantic sensitivity of the first contract element; determining a first hierarchical result of the first contract element based on the target semantic sensitivity of the first contract element, and determining a first security protection strategy for the first contract element based on the first hierarchical result; determining a second hierarchical result of the second contract element based on the initial semantic sensitivity of the second contract element, and determining a second security protection strategy for the second contract element based on the second hierarchical result.

[0006] Optionally, at least one pair of contract elements is extracted from the contract dataset, including: extracting contract elements from the preprocessed contract dataset to form a set of contract elements; and extracting at least one pair of contract elements from the set of contract elements.

[0007] Optionally, after extracting contract elements from the preprocessed contract dataset and forming a set of contract elements, the method further includes: dividing the values ​​of contract elements into multiple equivalence classes based on the value distribution characteristics of contract elements in the historical contract dataset and the corresponding industry security risk perception, wherein each equivalence class includes a set of values ​​with the same exposure risk level in the business scenario; calculating the frequency of each equivalence class in the contract dataset to obtain frequency sensitivity; obtaining the average semantic score of each equivalence class and the scenario correction coefficient of the performance scenario corresponding to each equivalence class, wherein the average semantic score is used to characterize the security risk level of each equivalence class in the business scenario, and the scenario correction coefficient is used to characterize the sensitivity multiplier adjustment of the same contract element in different scenarios; and substituting the frequency sensitivity, average semantic score, and scenario correction coefficient of the equivalence class to which the contract element belongs into a preset calculation model to obtain the initial semantic sensitivity of the contract element.

[0008] Optionally, the target association strength of the contract element pair is determined through a dual-path association extraction mechanism, including: using a pre-set keyword rule base and sentence template to identify the explicit relationship between the first contract element and the second contract element in the contract element pair, and determining the explicit association strength of the contract element pair based on the explicit relationship; using a pre-trained language model to encode the context fragments of the first contract element and the second contract element in the contract element pair respectively, generating a first context vector and a second context vector; performing an inner product operation on the first context vector and the second context vector to obtain the implicit association strength of the contract element pair; and determining the target association strength of the contract element pair based on the explicit and implicit association strengths of the contract element pair.

[0009] Optionally, the target association strength of the contract element pair is determined based on the explicit association strength and implicit association strength of the contract element pair, including: when the explicit association strength of the contract element pair is detected to be less than the implicit association strength of the contract element pair, the implicit association strength of the contract element pair is taken as the target association strength; when the explicit association strength of the contract element pair is detected to be greater than or equal to the implicit association strength of the contract element pair, the explicit association strength of the contract element pair is taken as the target association strength.

[0010] Optionally, the initial semantic sensitivity of the first contract element is dynamically adjusted based on the target association strength of the contract element pair to obtain the target semantic sensitivity of the first contract element, including: when the target association strength of the contract element pair is detected to be greater than or equal to the first association strength threshold: if the initial semantic sensitivity of the second contract element is greater than or equal to the first preset threshold, then a first value is added to the initial semantic sensitivity of the first contract element to obtain the target semantic sensitivity of the first contract element; if the initial semantic sensitivity of the second contract element is greater than or equal to the second preset threshold and less than the first preset threshold, then a second value is added to the initial semantic sensitivity of the first contract element to obtain the target semantic sensitivity of the first contract element, wherein the first value is greater than the second value.

[0011] Optionally, dynamically adjusting the initial semantic sensitivity of the first contract element based on the target association strength of the contract element pair to obtain the target semantic sensitivity of the first contract element further includes: when the target association strength of the contract element pair is detected to be less than the first association strength threshold and greater than or equal to the second association strength threshold: if the initial semantic sensitivity of the second contract element is greater than or equal to the first preset threshold, then a second value is added to the initial semantic sensitivity of the first contract element to obtain the target semantic sensitivity of the first contract element; if the initial semantic sensitivity of the second contract element is greater than or equal to the second preset threshold and less than the first preset threshold, then the initial semantic sensitivity of the first contract element is used as the target semantic sensitivity of the first contract element.

[0012] According to another aspect of the embodiments of this application, a contract data hierarchical protection device is also provided, comprising: a first processing unit, configured to extract at least one contract element pair based on a contract dataset, wherein the contract element pair consists of a first contract element and a second contract element that have potential semantic association, and the initial semantic sensitivity of the first contract element is lower than the initial semantic sensitivity of the second contract element; a second processing unit, configured to determine the target association strength of the contract element pair through a dual-path association extraction mechanism, wherein the dual-path association extraction mechanism is used to collaboratively identify the association relationship between the first contract element and the second contract element through explicit rule paths and implicit semantic paths; a third processing unit, configured to dynamically adjust the initial semantic sensitivity of the first contract element according to the target association strength of the contract element pair to obtain the target semantic sensitivity of the first contract element; a fourth processing unit, configured to determine a first hierarchical result of the first contract element according to the target semantic sensitivity of the first contract element, and determine a first security protection strategy for the first contract element according to the first hierarchical result; and a fifth processing unit, configured to determine a second hierarchical result of the second contract element according to the initial semantic sensitivity of the second contract element, and determine a second security protection strategy for the second contract element according to the second hierarchical result.

[0013] According to another aspect of the embodiments of this application, a computer-readable storage medium is also provided, which stores a computer program, wherein when the computer program is executed, the device where the computer-readable storage medium is located performs the above-described contract data hierarchical protection method.

[0014] According to another aspect of the embodiments of this application, an electronic device is also provided, including one or more processors and a memory, the memory being used to store one or more programs, wherein when the one or more programs are executed by one or more processors, the one or more processors cause the one or more processors to perform the above-described contract data hierarchical protection method.

[0015] According to another aspect of the embodiments of this application, a computer program product is also provided, including a computer program or instructions that, when executed by a processor, implement the above-described contract data hierarchical protection method.

[0016] In this application, the contract data hierarchical protection method first extracts at least one pair of contract elements from a contract dataset. Each pair consists of a first contract element and a second contract element with potential semantic association, where the initial semantic sensitivity of the first contract element is lower than that of the second contract element. A dual-path association extraction mechanism is used to determine the target association strength of the contract element pair. This mechanism collaboratively identifies the association between the first and second contract elements through explicit rule paths and implicit semantic paths. Based on the target association strength of the contract element pair, the initial semantic sensitivity of the first contract element is dynamically adjusted to obtain its target semantic sensitivity. A first hierarchical result is determined for the first contract element based on its target semantic sensitivity, and a first security protection strategy is determined based on the first hierarchical result. Finally, a second hierarchical result is determined for the second contract element based on its initial semantic sensitivity, and a second security protection strategy is determined based on the second hierarchical result.

[0017] In this embodiment, a dual-path association extraction mechanism is adopted. The explicit rule path identifies the clear semantic association patterns between contract elements, while the implicit semantic path captures element pairs that are not explicitly expressed but are highly semantically related. This achieves the purpose of quantifying the association strength between contract element pairs and dynamically transmitting sensitivity accordingly. This achieves the technical effect of accurately increasing the sensitivity of low-sensitivity elements under the influence of strongly associated high-sensitivity elements. In turn, it solves the technical problem in the prior art that sensitivity assessment is based only on static keywords or single attributes and does not consider the dynamic transmission effect of semantic association between contract elements, resulting in inaccurate identification of sensitive data and a disconnect between the classification results and actual security risks. Attached Figure Description

[0018] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0019] Figure 1 This is a flowchart of an optional hierarchical protection method for contract data according to an embodiment of this application;

[0020] Figure 2 This is a flowchart of an optional technical solution according to an embodiment of this application;

[0021] Figure 3 This is a schematic diagram illustrating an optional initial semantic sensitivity calculation according to an embodiment of this application;

[0022] Figure 4 This is an optional semantic association dynamically adjusted relationship diagram according to an embodiment of this application;

[0023] Figure 5 This is a flowchart of an optional specific instance method according to an embodiment of this application;

[0024] Figure 6 This is a schematic diagram of an optional contract data hierarchical protection device according to an embodiment of this application. Detailed Implementation

[0025] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0026] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0027] According to an embodiment of this application, a method embodiment for hierarchical protection of contract data is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0028] According to the embodiments of this application, a contract data hierarchical protection system (hereinafter referred to as the system) can be used as the execution subject of the contract data hierarchical protection method of this application. The contract data hierarchical protection system can be a software system or an embedded system combining software and hardware. Of course, the execution subject of the method in the embodiments of this application can also be other forms of execution subject, such as devices or equipment. Those skilled in the art should know that this application does not particularly limit the specific form of the execution subject of the method.

[0029] Figure 1 This is an optional hierarchical protection method for contract data according to an embodiment of this application, such as... Figure 1 As shown, the method includes the following steps:

[0030] Step S101: Extract at least one pair of contract elements based on the contract dataset, wherein the pair of contract elements consists of a first contract element and a second contract element that have potential semantic association, and the initial semantic sensitivity of the first contract element is lower than that of the second contract element.

[0031] Optionally, contract elements refer to the smallest data unit in the contract text that carries specific business semantics, such as the shipper's name, insured value, cargo name, flight number, etc.

[0032] Optionally, a contract element pair refers to an ordered pair of two contract elements in the same contract clause or adjacent context, which are formed by potential semantic and logical dependencies or combined risk relationships.

[0033] Optionally, latent semantic association represents that although two contract elements are not explicitly connected through explicit grammatical structures (such as inclusion or being), they have implicit combined risks in the business context (such as the simultaneous occurrence of precision instruments and high insurance value, constituting high-value transportation risk).

[0034] Optionally, after the contract dataset is segmented and entity recognized, the system traverses the contract elements of each contract text and pairs any two different types of contract elements that are adjacent or in the same sentence to form a contract element pair.

[0035] Step S102: Determine the target association strength of the contract element pair through the dual-path association extraction mechanism. The dual-path association extraction mechanism is used to collaboratively identify the association relationship between the first contract element and the second contract element through explicit rule path and implicit semantic path.

[0036] Optionally, the dual-path association extraction mechanism includes two parallel channels: an explicit rule path and an implicit semantic path, used to complementarily determine whether two contract elements are related.

[0037] The explicit rule path performs pattern matching based on predefined contract clause sentence templates to identify explicit associations with clear grammatical structure and definite semantics. The implicit semantic path uses a pre-trained language model to semantically encode the contract context, calculates the similarity between two contract elements in the semantic space, and captures implicit associations that are highly semantically related but do not use trigger words.

[0038] Optionally, the target association strength is the final confidence value obtained by fusing the association scores output by the explicit and implicit paths respectively, reflecting the reliability and strength of the semantic association between the two contract elements.

[0039] Optionally, for each pair of contract elements, the system executes two paths in parallel:

[0040] Explicit path: Load the civil aviation contract keyword rule base (such as attributes like "contains", "requires submission", etc.), and match whether the sentence containing the pair of elements conforms to the preset sentence pattern. If the match is successful, assign an explicit association strength;

[0041] Implicit path: Input the context text of the two elements into the fine-tuned BERT model, generate context vectors respectively, calculate the normalized cosine similarity, and if it is greater than or equal to the preset value, it is determined to be a strong implicit association.

[0042] Fusion output: Taking the maximum value between the explicit and implicit path scores as the target association strength of the contract element pair helps to prioritize high-confidence evidence.

[0043] Optionally, the fine-tuned BERT model used in the application is a specialized language model based on the general BERT architecture, which is pre-trained for domain-adaptive learning and task-oriented fine-tuning for the domain characteristics of civil aviation contract texts. Its model structure consists of the following three core components:

[0044] 1. Input embedding layer.

[0045] The input embedding layer consists of the sum of three vectors:

[0046] Word embedding: Maps each word or subword in the input text to a fixed-dimensional vector. It uses word segmentation algorithms to process professional terms in the civil aviation field (such as insured value, dangerous goods, and unified social credit code), which helps prevent compound words from being incorrectly split.

[0047] Segment embedding: Used to distinguish different semantic segments in a contract (such as shipper information and performance terms), and is uniformly set to 0 in single-sentence input scenarios;

[0048] Position embedding: Assigning an absolute position code to each word or subword in the sequence, enabling the model to perceive word order structure and preserve the semantic dependencies of fixed word order in contract terms such as "Goods name: Name X, Insured amount: Amount X".

[0049] 2. Multi-layer Transformer encoder stack.

[0050] This model uses a 12-layer Transformer encoder, with each layer consisting of two sub-modules connected in series:

[0051] One of the sub-modules is a multi-head self-attention mechanism: it uses 12 parallel attention heads, each of which calculates the attention weights of the query, key, and value vectors to capture semantic dependencies of different granularities between words or sub-words, thereby achieving bidirectional context modeling.

[0052] Another submodule is the feedforward neural network: each attention layer is followed by a two-layer fully connected network, which uses the GELU activation function for non-linear transformation to enhance the model's expressive power.

[0053] In addition, each layer includes residual connections and layer normalization to mitigate gradient vanishing and accelerate convergence.

[0054] 3. Output presentation layer.

[0055] The hidden state of each word or sub-word is output as the final semantic representation. In this model, an average pooling strategy is used to average the vectors of all words or sub-words in the context segment containing the target contract element, generating a context-aware sentence vector for that contract element.

[0056] Step S103: Based on the target association strength of the contract element pair, dynamically adjust the initial semantic sensitivity of the first contract element to obtain the target semantic sensitivity of the first contract element.

[0057] Step S104: Determine the first classification result of the first contract element based on the target semantic sensitivity of the first contract element, and determine the first security protection strategy of the first contract element based on the first classification result.

[0058] Optionally, the first classification result is based on the target semantic sensitivity of the first contract element, and the sensitivity level label (such as high sensitivity, medium sensitivity, etc.) is divided according to a preset level range to guide the matching of security protection strategies.

[0059] Optionally, the first security protection strategy is access control, encryption, or desensitization measures corresponding to the first classification result.

[0060] In this embodiment, the system presets five levels of classification intervals (such as extremely high sensitivity, high sensitivity, medium sensitivity, etc.), maps the target semantic sensitivity of the first contract element to the corresponding level, and matches the corresponding security protection strategy according to the classification structure.

[0061] For example, when the semantic sensitivity of the target falls into the high-sensitivity range, the corresponding first-level result is high-sensitivity. The system then automatically assigns SM4 encryption (national cryptographic standard) + departmental-level access control + operation log auditing as the security protection strategy according to the preset strategy mapping table. The preset strategy mapping table is shown in Table 1.

[0062] Table 1

[0063]

[0064] Step S105: Determine the second classification result of the second contract element based on the initial semantic sensitivity of the second contract element, and determine the second security protection strategy of the second contract element based on the second classification result.

[0065] Figure 2 This is a flowchart of an optional technical solution according to an embodiment of this application. For example... Figure 2 As shown, this solution proceeds sequentially through data acquisition and preprocessing, attribute frequency-sensitive quantification, attribute semantic-sensitive quantification, dynamic adjustment of semantic relevance, and comprehensive sensitivity.

[0066] In one alternative implementation, at least one pair of contract elements is extracted from the contract dataset, including: extracting contract elements from the preprocessed contract dataset to form a set of contract elements; and extracting at least one pair of contract elements from the set of contract elements.

[0067] Optionally, the preprocessed contract dataset refers to the structured data set formed after the original civil aviation contract text has been cleaned (removing contract numbers, page numbers, and format specifiers), segmented (using a dictionary that incorporates industry terms, such as insurance amount, dangerous goods, and handler, to prevent incorrect segmentation), and identified entities (in this embodiment, a named entity recognition model finely tuned based on 5,000 industry corpora is used).

[0068] Optionally, a contract element refers to the smallest semantic entity in the contract text that carries independent business semantics. It has a clear type label, such as unified social credit code, insured amount, goods name, transport flight number, and liquidated damages ratio. Each contract element has been labeled as an identity identifier, commercial value, or performance scenario in the preprocessing stage.

[0069] Optionally, for each preprocessed contract text, a fine-tuned named entity recognition model is invoked to automatically scan and extract all contract elements that conform to the semantics of civil aviation business, forming a set of all contract elements within the contract text.

[0070] Optionally, by transforming unstructured contract text into a set of computable and associative standardized semantic units, a high-quality, highly consistent, and unambiguous basic data source can be provided for the construction of subsequent contract element pairs, reducing misjudgments of associations caused by text noise or terminology recognition errors.

[0071] Optionally, a contract element pair refers to a binary pair consisting of any two different contract elements in the set of contract elements. The conditions for its formation are: the two exist in the same contract paragraph or sentence and have a potential semantic connection intention, regardless of whether their connection strength has been determined, only semantic co-occurrence is required.

[0072] Optionally, for each contract text's set of contract elements, a sliding window or intra-sentence pairing strategy is used to traverse all possible combinations of contract elements. Pairs of elements of the same type are excluded to focus on cross-class associations, while element pairs from different categories that co-occur in the same semantic unit are retained. For example, in the example above, two valid contract element pairs are extracted from the set {Company A, Code X, Amount X}: (Code X, Amount X) and (Company A, Amount X), forming a potential association pair between the identity category and the commercial value category, serving as input units for subsequent dynamic adjustments.

[0073] In an optional embodiment, after extracting contract elements from the preprocessed contract dataset to form a set of contract elements, the method further includes: dividing the values ​​of contract elements into multiple equivalence classes based on the value distribution characteristics of contract elements in the historical contract dataset and the corresponding industry security risk perception, wherein each equivalence class includes a set of values ​​with the same exposure risk level in the business scenario; calculating the frequency of each equivalence class in the contract dataset to obtain frequency sensitivity; obtaining the average semantic score of each equivalence class and the scenario correction coefficient of the performance scenario corresponding to each equivalence class, wherein the average semantic score is used to characterize the security risk level of each equivalence class in the business scenario, and the scenario correction coefficient is used to characterize the sensitivity multiplier adjustment of the same contract element in different scenarios; and substituting the frequency sensitivity, average semantic score, and scenario correction coefficient of the equivalence class to which the contract element belongs into a preset calculation model to obtain the initial semantic sensitivity of the contract element.

[0074] Optionally, an equivalence class refers to a set of groups formed after merging the values ​​of a certain contract element according to business semantics. All values ​​within a group have similar leakage consequences and exposure risk levels in civil aviation security practices, rather than being a simple numerical equidistant division. The exposure risk level refers to the comprehensive degree of economic losses, compliance penalties, or operational safety impacts that may occur in the civil aviation business environment when a certain value is illegally accessed or leaked.

[0075] Specifically, for each contract element, based on the value distribution of that element in historical civil aviation contract data and combined with industry experts' understanding of risk levels, the values ​​are divided into several semantically equivalent intervals. Although the values ​​within each equivalence class are different, they are classified into the same risk level because their business risks are similar.

[0076] Optionally, continuous or discrete original values ​​can be transformed into risk units with business consistency to reduce sensitivity fluctuations caused by small numerical differences, making the assessment closer to the real industry risk perception and improving the robustness of the model.

[0077] Optionally, frequency sensitivity refers to the inverse mapping value of the frequency of a certain equivalence class in the historical contract dataset, reflecting the scarcity and locatability of this value in the industry. That is, the less frequently it appears, the easier it is to lock a specific contract or entity after leakage, and the higher the risk.

[0078] Optionally, the frequency sensitivity is calculated as follows: First, count the number of times each equivalence class appears in the contract dataset to obtain the relative frequency of each equivalence class, and then calculate the sensitivity using a normalization formula. For example, the frequency sensitivity of the k-th equivalence class is calculated as shown in formula (1).

[0079] (1)

[0080] in, This represents the maximum frequency of a given attribute across all equivalence classes. Indicates the minimum frequency of occurrence. This represents the frequency of the k-th equivalence class. When... The larger the value, the smaller the difference between the frequency of the contract element's value and the maximum frequency, meaning that the value is relatively scarce and the sensitivity is higher.

[0081] Optionally, the average semantic score is calculated by a team of industry experts who subjectively score the security risks of each equivalence class in the context of civil aviation business, reflecting the "degree of danger" of the contract element value in business perception.

[0082] Specifically, let the number of people in the review panel be m, and the t-th expert... The score for the k-th equivalence class of contract element i The value ranges from [0,1]. The average semantic score of the equivalence class is obtained by calculating the average of the expert scores. .

[0083] (2)

[0084] Optionally, scene correction coefficient The multiplier factor is assigned based on the specific performance scenario corresponding to the equivalence class. The exposure risk differs for the same value under different scenarios. In this embodiment, under emergency / high-risk scenarios... In ordinary scenarios In low-risk scenarios .

[0085] Optionally, expert experience and scenario context can be introduced to compensate for the shortcomings of purely data-driven approaches, so that sensitivity assessments can reflect both objective distribution patterns and industry knowledge and dynamic risk environments, thereby improving the business relevance of the assessment.

[0086] Optionally, the preset calculation model is a weighted fusion formula used to integrate frequency sensitivity, average semantic score and scene correction coefficient to output the initial semantic sensitivity of contract elements, as shown in formula (3):

[0087] (3)

[0088] in, As the weight of objective frequency sensitivity, The weight of subjective semantic sensitivity satisfies , A higher value indicates higher sensitivity.

[0089] Figure 3 This is a schematic diagram illustrating an optional initial semantic sensitivity calculation according to an embodiment of this application. Figure 3 As shown, the initial semantic sensitivity is calculated by frequency sensitivity, averaging of multiple scores, and correction based on the performance scenario.

[0090] In one optional embodiment, the target association strength of the contract element pair is determined through a dual-path association extraction mechanism, including: using a preset keyword rule base and sentence template to identify the explicit relationship between the first contract element and the second contract element in the contract element pair, and determining the explicit association strength of the contract element pair based on the explicit relationship; using a pre-trained language model to encode the context fragments of the first contract element and the second contract element in the contract element pair respectively, generating a first context vector and a second context vector; performing an inner product operation on the first context vector and the second context vector to obtain the implicit association strength of the contract element pair; and determining the target association strength of the contract element pair based on the explicit association strength and the implicit association strength of the contract element pair.

[0091] Optionally, the preset keyword rule base and sentence template refer to a set of structured rules artificially constructed based on the highly stylized and repetitive language characteristics of civil aviation contract texts. It includes trigger words (such as contain, attribute is, need to be submitted, restricted to) and their corresponding grammatical rule templates, which are used to identify the semantic dependencies explicitly expressed in the contract.

[0092] Optionally, an explicit relationship refers to a semantic connection established directly between two contract elements through a clear grammatical structure or fixed expression. The strength of an explicit relationship is a numerical confidence level assigned based on the type of explicit relationship, reflecting the certainty of the relationship as a strong semantic link in the contract text; a higher value indicates a more explicit relationship.

[0093] Specifically, the system loads a preset rule base and performs pattern matching on the sentences containing the contract elements. If a match is successful, an explicit association strength is assigned according to the explicit relationship diagram in Table 2. If no match is found, the explicit association strength is 0.

[0094] Table 2

[0095]

[0096] Optionally, a contextual fragment refers to a local contextual text consisting of 10 words or punctuation marks before and after the target contract element, centered on it, used to capture the semantic environment of the contract element.

[0097] Optionally, the context vector refers to the dense vector representation of the context fragment in the deep semantic space after being encoded by the BERT model, which contains the deep semantic features of the contract element in the context.

[0098] Specifically, the context fragments of the first contract element and the second contract element are extracted and input into the fine-tuned BERT model to obtain the average pooling vector of their last hidden layer, which serves as the first context vector and the second context vector.

[0099] Optionally, the inner product operation refers to the dot product operation on two normalized vectors, and the result is the cosine similarity between the two in the semantic space, with a value range of [-1, 1]. The closer the value is to 1, the more similar the semantics are.

[0100] Optionally, the implicit association strength is obtained by mapping the inner product result to the non-negative association score, which reflects the potential correlation between the two contract elements in the semantic space. After threshold truncation, it is used to determine whether there is an implicit semantic dependency.

[0101] Specifically, regarding the elements of the contract Construct context fragments centered on the two contract elements respectively. The hidden layer vector sequence is obtained after encoding. and First, average pooling is performed to obtain the fragment-level representation, as shown in formula (4):

[0102] (4)

[0103] in, Let d represent a d-dimensional real space, where p and q represent indexes.

[0104] Then proceed Normalization yields the unit vector representation, as shown in formula (5):

[0105] (5)

[0106] Finally, the semantic similarity between the two context segments is calculated based on the inner product as the implicit association strength, as shown in Equation (6):

[0107] (6)

[0108] Among them, similarity It is determined to be a strong implicit association. It is determined to be an implicit association.

[0109] In one optional embodiment, determining the target association strength of a contract element pair based on its explicit association strength and implicit association strength includes: when the explicit association strength of the contract element pair is detected to be less than its implicit association strength, using the implicit association strength of the contract element pair as the target association strength; and when the explicit association strength of the contract element pair is detected to be greater than or equal to its implicit association strength, using the explicit association strength of the contract element pair as the target association strength.

[0110] Optionally, if there are both explicit and implicit associations among contract elements, the association strength is fused, and the maximum value is taken as the target association strength. The calculation formula is as shown in formula (7):

[0111] (7)

[0112] Figure 4 This is an optional semantically dynamically adjusted relationship diagram according to an embodiment of this application. For example... Figure 4 As shown, the dynamic adjustment of semantic association includes selecting candidate pairs; extracting explicit relationships; determining whether there is an explicit association, and if so, recording the strength of the explicit association; if not, recording the strength of the explicit association as 0; extracting implicit associations and calculating their strength; fusing association strengths and transmitting sensitivity; and outputting the final sensitivity.

[0113] In one optional embodiment, the initial semantic sensitivity of the first contract element is dynamically adjusted based on the target association strength of the contract element pair to obtain the target semantic sensitivity of the first contract element. This includes: when the target association strength of the contract element pair is detected to be greater than or equal to a first association strength threshold: if the initial semantic sensitivity of the second contract element is greater than or equal to a first preset threshold, then a first value is added to the initial semantic sensitivity of the first contract element to obtain the target semantic sensitivity of the first contract element; if the initial semantic sensitivity of the second contract element is greater than or equal to a second preset threshold and less than the first preset threshold, then a second value is added to the initial semantic sensitivity of the first contract element to obtain the target semantic sensitivity of the first contract element, wherein the first value is greater than the second value.

[0114] Optionally, the first association strength threshold refers to a preset association strength threshold, which is used to determine whether there is a strong association between two contract elements.

[0115] In this embodiment, the threshold is set to 0.9. When the target association strength calculated in the system reading step is greater than or equal to 0.9, the sensitivity adjustment process is entered; otherwise, this step is skipped, and the initial semantic sensitivity of the first contract element remains unchanged.

[0116] Specifically, when the target association strength is greater than or equal to 0.9: if the initial semantic sensitivity of the second contract element is greater than or equal to the first preset threshold, then a first value is added to the initial semantic sensitivity of the first contract element. For example, when the initial semantic sensitivity of the first contract element is 0.75, it is increased by 0.1, resulting in a target semantic sensitivity of 0.85. If the initial semantic sensitivity of the second contract element is greater than or equal to the second preset threshold and less than the first preset threshold, then a second value is added to the initial semantic sensitivity of the first contract element. For example, when the initial semantic sensitivity of the first contract element is 0.4, it is increased by 0.05, resulting in a target semantic sensitivity of 0.45.

[0117] Optionally, in this system, the target semantic sensitivity of any contract element shall not exceed 0.95. If the adjusted target semantic sensitivity exceeds the range, it shall be forcibly assigned a value of 0.95 and marked as extremely sensitive, so as to reserve a standard boundary for subsequent classification.

[0118] In an optional embodiment, the initial semantic sensitivity of the first contract element is dynamically adjusted based on the target association strength of the contract element pair to obtain the target semantic sensitivity of the first contract element. The method further includes: when the target association strength of the contract element pair is detected to be less than a first association strength threshold and greater than or equal to a second association strength threshold: if the initial semantic sensitivity of the second contract element is greater than or equal to a first preset threshold, a second value is added to the initial semantic sensitivity of the first contract element to obtain the target semantic sensitivity of the first contract element; if the initial semantic sensitivity of the second contract element is greater than or equal to the second preset threshold and less than the first preset threshold, the initial semantic sensitivity of the first contract element is used as the target semantic sensitivity of the first contract element.

[0119] Optionally, the second association strength threshold is used to define the lower limit of the association.

[0120] In this embodiment, the second association strength threshold is set to 0.7. The system reads the target association strength value to determine whether it meets the requirement that the target association strength is greater than or equal to 0.7 and less than 0.9. If it meets the requirement, this process is initiated; otherwise, this step is skipped, and the no-adjustment or strong-association adjustment logic is executed respectively.

[0121] Specifically, when the target association strength is greater than or equal to 0.7 and less than 0.9, if the initial semantic sensitivity of the second contract element is greater than or equal to the first preset threshold, then a second value is added to the initial semantic sensitivity of the first contract element. For example, if the initial semantic sensitivity of the first contract element is 0.35, it is increased by 0.05 to obtain a target semantic sensitivity of 0.40.

[0122] If the initial semantic sensitivity of the second contract element is greater than or equal to the second preset threshold and less than the first preset threshold, then the initial semantic sensitivity of the first contract element will not be adjusted. For example, if the initial semantic sensitivity of the first contract element is 0.4, it will be directly used as the target semantic sensitivity, that is, the target semantic sensitivity will remain at 0.4.

[0123] Optionally, to avoid sensitivity overflow or over-adjustment, the system sets constraints: if the target semantic sensitivity is greater than or equal to 0.95, it is marked as extremely sensitive and the target semantic sensitivity is forcibly assigned a value of 1; if the initial sensitivity is less than or equal to 0.3, the reduction operation is not performed.

[0124] Figure 5 This is a flowchart of an optional specific example method according to an embodiment of this application. For example... Figure 5As shown, the system first collects and preprocesses civil aviation contract data, then calculates attribute frequency sensitivity and attribute semantic sensitivity, dynamically adjusts semantic relevance, including implicit and explicit association extraction. Finally, it integrates sensitivity grading and customized security protection strategies, and verifies the effectiveness of sensitivity grading.

[0125] Figure 6 This is a schematic diagram of an optional contract data hierarchical protection device according to an embodiment of this application. According to another embodiment of this application, a contract data hierarchical protection device is also provided, including: a first processing unit 601, a second processing unit 602, a third processing unit 603, a fourth processing unit 604, and a fifth processing unit 605.

[0126] The system comprises the following components: a first processing unit 601, used to extract at least one pair of contract elements based on a contract dataset, wherein the pair consists of a first contract element and a second contract element with potential semantic association, and the initial semantic sensitivity of the first contract element is lower than that of the second contract element; a second processing unit 602, used to determine the target association strength of the pair of contract elements through a dual-path association extraction mechanism, wherein the dual-path association extraction mechanism is used to collaboratively identify the association relationship between the first contract element and the second contract element through explicit rule paths and implicit semantic paths; a third processing unit 603, used to dynamically adjust the initial semantic sensitivity of the first contract element according to the target association strength of the pair of contract elements to obtain the target semantic sensitivity of the first contract element; a fourth processing unit 604, used to determine the first classification result of the first contract element based on the target semantic sensitivity of the first contract element, and to determine the first security protection strategy of the first contract element based on the first classification result; and a fifth processing unit 605, used to determine the second classification result of the second contract element based on the initial semantic sensitivity of the second contract element, and to determine the second security protection strategy of the second contract element based on the second classification result.

[0127] Optionally, the first processing unit 601 includes: a first sub-processing unit for extracting contract elements from the preprocessed contract dataset to form a contract element set; and a second sub-processing unit for extracting at least one pair of contract elements from the contract element set.

[0128] Optionally, the contract data hierarchical protection device further includes: an equivalence class division unit, used to divide the values ​​of contract elements into multiple equivalence classes based on the value distribution characteristics of contract elements in historical contract datasets and the corresponding industry security risk perceptions, wherein each equivalence class includes a set of values ​​with the same exposure risk level in a business scenario; a frequency sensitivity calculation unit, used to calculate the frequency of each equivalence class in the contract dataset to obtain frequency sensitivity; an acquisition unit, used to acquire the average semantic score of each equivalence class and the scenario correction coefficient of the performance scenario corresponding to each equivalence class, wherein the average semantic score is used to characterize the security risk level of each equivalence class in a business scenario, and the scenario correction coefficient is used to characterize the sensitivity multiplier adjustment of the same contract element in different scenarios; and an initial semantic sensitivity calculation unit, used to substitute the frequency sensitivity, average semantic score, and scenario correction coefficient of the equivalence class to which the contract element belongs into a preset calculation model to obtain the initial semantic sensitivity of the contract element.

[0129] Optionally, the second processing unit 602 includes: a first determining subunit, used to identify the explicit relationship between the first contract element and the second contract element in a contract element pair using a preset keyword rule base and sentence template, and determine the explicit association strength of the contract element pair based on the explicit relationship; an encoding subunit, used to encode the context fragments of the first contract element and the second contract element in the contract element pair respectively using a pre-trained language model to generate a first context vector and a second context vector; a calculation subunit, used to perform an inner product operation on the first context vector and the second context vector to obtain the implicit association strength of the contract element pair; and a second determining subunit, used to determine the target association strength of the contract element pair based on the explicit association strength and the implicit association strength of the contract element pair.

[0130] Optionally, the second determining subunit includes: a first detection module, configured to take the implicit association strength of the contract element pair as the target association strength when the explicit association strength of the contract element pair is detected to be less than the implicit association strength of the contract element pair; and a second detection module, configured to take the explicit association strength of the contract element pair as the target association strength when the explicit association strength of the contract element pair is detected to be greater than or equal to the implicit association strength of the contract element pair.

[0131] Optionally, the third processing unit 603 includes: a first processing subunit, configured to: when the target association strength of the contract element pair is detected to be greater than or equal to a first association strength threshold; if the initial semantic sensitivity of the second contract element is greater than or equal to a first preset threshold, then add a first value to the initial semantic sensitivity of the first contract element to obtain the target semantic sensitivity of the first contract element; if the initial semantic sensitivity of the second contract element is greater than or equal to a second preset threshold and less than the first preset threshold, then add a second value to the initial semantic sensitivity of the first contract element to obtain the target semantic sensitivity of the first contract element, wherein the first value is greater than the second value.

[0132] Optionally, the third processing unit 603 further includes: a second processing subunit, configured to: when the target association strength of the contract element pair is detected to be less than a first association strength threshold and greater than or equal to a second association strength threshold: if the initial semantic sensitivity of the second contract element is greater than or equal to a first preset threshold, then add a second value to the initial semantic sensitivity of the first contract element to obtain the target semantic sensitivity of the first contract element; if the initial semantic sensitivity of the second contract element is greater than or equal to the second preset threshold and less than the first preset threshold, then use the initial semantic sensitivity of the first contract element as the target semantic sensitivity of the first contract element.

[0133] According to another aspect of the embodiments of this application, a computer-readable storage medium is also provided, which stores a computer program, wherein when the computer program is executed, the device where the computer-readable storage medium is located performs the above-described contract data hierarchical protection method.

[0134] According to another aspect of the embodiments of this application, an electronic device is also provided, including one or more processors and a memory, the memory being used to store one or more programs, wherein when the one or more programs are executed by one or more processors, the one or more processors cause the one or more processors to perform the above-described contract data hierarchical protection method.

[0135] According to another aspect of the embodiments of this application, a computer program product is also provided, including a computer program or instructions that, when executed by a processor, implement the above-described contract data hierarchical protection method.

[0136] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0137] In the above embodiments of this application, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0138] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units can be a logical functional division, and in actual implementation, there may be other division methods. For instance, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling, direct coupling, or communication connection may be through some interfaces; the indirect coupling or communication connection between units or modules may be electrical or other forms.

[0139] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0140] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0141] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, read-only memory (ROM), random access memory (RAM), portable hard drive, magnetic disk, or optical disk.

[0142] The above description is only a preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A method for hierarchical protection of contract data, characterized in that, include: At least one pair of contract elements is extracted from the contract dataset, wherein the pair of contract elements consists of a first contract element and a second contract element that have potential semantic association, and the initial semantic sensitivity of the first contract element is lower than that of the second contract element. The target association strength of the contract element pair is determined by a dual-path association extraction mechanism, wherein the dual-path association extraction mechanism is used to collaboratively identify the association relationship between the first contract element and the second contract element through explicit rule paths and implicit semantic paths. Based on the target association strength of the contract element pairs, the initial semantic sensitivity of the first contract element is dynamically adjusted to obtain the target semantic sensitivity of the first contract element. The first classification result of the first contract element is determined based on the target semantic sensitivity of the first contract element, and the first security protection strategy of the first contract element is determined based on the first classification result. The second classification result of the second contract element is determined based on the initial semantic sensitivity of the second contract element, and the second security protection strategy of the second contract element is determined based on the second classification result.

2. The method according to claim 1, characterized in that, At least one pair of contract elements was extracted from the contract dataset, including: Extract contract elements from the preprocessed contract dataset to form a contract element set; Extract at least one pair of contract elements from the set of contract elements.

3. The method according to claim 2, characterized in that, After extracting contract elements from the preprocessed contract dataset and forming a set of contract elements, the method further includes: Based on the value distribution characteristics of the contract elements in the historical contract dataset and the corresponding industry security risk perception, the values ​​of the contract elements are divided into multiple equivalence classes, wherein each equivalence class includes a set of values ​​with the same exposure risk level in the business scenario. Calculate the frequency of each equivalence class in the contract dataset to obtain the frequency sensitivity. Obtain the average semantic score of each equivalence class and the scenario correction coefficient of the performance scenario corresponding to each equivalence class, wherein the average semantic score is used to characterize the security risk level of each equivalence class in the business scenario, and the scenario correction coefficient is used to characterize the sensitivity multiplier adjustment of the same contract element in different scenarios; The frequency sensitivity, average semantic score, and scenario correction coefficient of the equivalence class to which the contract element belongs are substituted into a preset calculation model to obtain the initial semantic sensitivity of the contract element.

4. The method according to claim 1, characterized in that, The target association strength of the contract element pairs is determined through a dual-path association extraction mechanism, including: Using a pre-defined keyword rule base and sentence templates, the explicit relationship between the first contract element and the second contract element in the contract element pair is identified, and the explicit association strength of the contract element pair is determined based on the explicit relationship. A pre-trained language model is used to encode the context fragments of the first contract element and the second contract element in the contract element pair, respectively, to generate a first context vector and a second context vector; Perform an inner product operation on the first context vector and the second context vector to obtain the implicit association strength of the contract element pair; The target association strength of the contract element pair is determined based on the explicit and implicit association strengths of the contract element pair.

5. The method according to claim 4, characterized in that, Based on the explicit and implicit association strengths of the contract element pairs, the target association strength of the contract element pairs is determined, including: When it is detected that the explicit association strength of the contract element pair is less than the implicit association strength of the contract element pair, the implicit association strength of the contract element pair is taken as the target association strength. When the explicit association strength of the contract element pair is detected to be greater than or equal to the implicit association strength of the contract element pair, the explicit association strength of the contract element pair is taken as the target association strength.

6. The method according to claim 1, characterized in that, Based on the target association strength of the contract element pairs, the initial semantic sensitivity of the first contract element is dynamically adjusted to obtain the target semantic sensitivity of the first contract element, including: When the target association strength of the contract element pair is detected to be greater than or equal to the first association strength threshold: If the initial semantic sensitivity of the second contract element is greater than or equal to the first preset threshold, then the first value is added to the initial semantic sensitivity of the first contract element to obtain the target semantic sensitivity of the first contract element. If the initial semantic sensitivity of the second contract element is greater than or equal to the second preset threshold and less than the first preset threshold, then a second value is added to the initial semantic sensitivity of the first contract element to obtain the target semantic sensitivity of the first contract element, wherein the first value is greater than the second value.

7. The method according to claim 6, characterized in that, Based on the target association strength of the contract element pairs, the initial semantic sensitivity of the first contract element is dynamically adjusted to obtain the target semantic sensitivity of the first contract element, which further includes: When the target association strength of the contract element pair is detected to be less than the first association strength threshold and greater than or equal to the second association strength threshold: If the initial semantic sensitivity of the second contract element is greater than or equal to the first preset threshold, then a second value is added to the initial semantic sensitivity of the first contract element to obtain the target semantic sensitivity of the first contract element. If the initial semantic sensitivity of the second contract element is greater than or equal to the second preset threshold and less than the first preset threshold, then the initial semantic sensitivity of the first contract element is taken as the target semantic sensitivity of the first contract element.

8. A contract data hierarchical protection device, characterized in that, include: The first processing unit is configured to extract at least one pair of contract elements based on the contract dataset, wherein the pair of contract elements consists of a first contract element and a second contract element that have potential semantic association, and the initial semantic sensitivity of the first contract element is lower than that of the second contract element. The second processing unit is used to determine the target association strength of the contract element pair through a dual-path association extraction mechanism, wherein the dual-path association extraction mechanism is used to collaboratively identify the association relationship between the first contract element and the second contract element through explicit rule paths and implicit semantic paths. The third processing unit is used to dynamically adjust the initial semantic sensitivity of the first contract element according to the target association strength of the contract element pair, so as to obtain the target semantic sensitivity of the first contract element. The fourth processing unit is used to determine the first classification result of the first contract element based on the target semantic sensitivity of the first contract element, and to determine the first security protection strategy of the first contract element based on the first classification result. The fifth processing unit is used to determine the second classification result of the second contract element based on the initial semantic sensitivity of the second contract element, and to determine the second security protection strategy of the second contract element based on the second classification result.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, wherein when the computer program is executed, the device in which the computer-readable storage medium is located performs the contract data hierarchical protection method according to any one of claims 1 to 7.

10. An electronic device, characterized in that, It includes one or more processors and a memory, the memory being used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors cause the one or more processors to perform the contract data hierarchical protection method according to any one of claims 1 to 7.

11. A computer program product, characterized in that, It includes a computer program or instructions that, when executed by a processor, implement the contract data hierarchical protection method according to any one of claims 1 to 7.

Images