An AI-based contract ambiguous clause risk assessment method and system

Abstract

This invention relates to the field of intelligent contract review technology, and in particular to an AI-based method and system for risk assessment of ambiguous contract clauses. The method includes the following steps: acquiring the contract text to be assessed; extracting semantic features from the contract text to obtain candidate ambiguous clauses; constructing a historical ambiguous clause sample library, and performing multi-dimensional fuzziness quantification on the candidate ambiguous clauses to obtain fuzziness quantification results, thereby identifying the ambiguous contract clauses; performing risk propagation path analysis on the ambiguous contract clauses to obtain a risk propagation impact index, and constructing a risk propagation chain using the ambiguous contract clauses as nodes; obtaining the chain reaction amplification factor of the ambiguous contract clauses based on the risk propagation chain, and determining the risk assessment result of the ambiguous contract clauses based on the chain reaction amplification factor. This invention effectively identifies ambiguous contract clauses and performs risk assessment, significantly improving the efficiency of contract review.

Description

Technical Field

[0001] This invention relates to the field of intelligent contract review technology, and in particular to an AI-based method and system for risk assessment of ambiguous contract clauses. Background Technology

[0002] In commercial activities, contracts, as the core legal documents for economic activities between enterprises, directly impact the rights and interests of businesses due to the clarity and rigor of their terms. With the expansion of business scale and the increasing complexity of transactions, the workload of contract review has exploded, making manual review insufficient to meet practical needs. Therefore, artificial intelligence-based intelligent contract review solutions have emerged. For example, keyword extraction algorithms are used to classify contract clauses, and then risk information is extracted from the clauses using a pre-set risk keyword list or knowledge graph embedding technology. Finally, a single-dimensional risk scoring model is used to determine the risk level of each clause. Such solutions can identify explicit risk clauses in contracts to a certain extent, such as unilateral disclaimers and clauses concerning unfavorable jurisdiction, achieving preliminary automation of contract review.

[0003] However, existing technologies have several shortcomings: First, risk identification is primarily limited to explicit risk statements within the clauses, lacking the ability to identify potential risks arising from ambiguous or unclear statements, and failing to uncover performance dispute risks hidden in vague expressions. Second, using keyword matching or rule templates for clause analysis makes it difficult to conduct multi-dimensional quantitative assessments of clauses, resulting in low accuracy in identifying ambiguous clauses. Furthermore, risk assessments are typically conducted in isolation for contract clauses, ignoring the interdependencies and risk propagation effects between clauses; ambiguity in one clause may affect the interpretation and performance of other clauses through references, supplements, etc. Finally, risk assessment results are mostly single-dimensional level outputs, lacking tracing of risk propagation paths and quantitative analysis of risk amplification effects, making it difficult to provide legal professionals with sufficient risk explanations and targeted modification suggestions. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides an AI-based method and system for risk assessment of ambiguous contract clauses.

[0005] To achieve the above objectives, in a first aspect, this invention provides an AI-based risk assessment method for ambiguous contract clauses. The method includes the following steps: acquiring the contract text to be assessed; extracting semantic features from the contract text to obtain candidate ambiguous clauses; constructing a historical ambiguous clause sample library and performing multi-dimensional fuzziness quantification on the candidate ambiguous clauses to obtain fuzziness quantification results, thereby identifying the ambiguous contract clauses; performing risk propagation path analysis on the ambiguous contract clauses to obtain a risk propagation impact index, and constructing a risk propagation chain using the ambiguous contract clauses as nodes; obtaining the chain reaction amplification factor of the ambiguous contract clauses based on the risk propagation chain, and determining the risk assessment result of the ambiguous contract clauses based on the chain reaction amplification factor. This invention breaks through the efficiency bottleneck of traditional manual assessment, achieving a comprehensive and in-depth analysis of the risks of ambiguous contract clauses, and improving the accuracy and comprehensiveness of the assessment.

[0006] Optionally, obtaining the contract text to be evaluated and extracting semantic features from the contract text to obtain candidate fuzzy clauses includes: preprocessing the text structure of the contract text to obtain independent clause texts; using the independent clause texts as input, extracting semantic feature vectors from the independent clause texts through a pre-trained language model; obtaining the semantic dispersion of the independent clause texts through the semantic feature vectors, and determining the candidate fuzzy clauses through the semantic dispersion. This invention achieves digital parsing of clause semantics, avoiding subjective biases in manual identification, laying a precise data foundation for subsequent fuzziness quantification and risk assessment, and improving the efficiency and accuracy of clause selection.

[0007] Optionally, obtaining the semantic dispersion of the independent clause text through the semantic feature vector includes: segmenting the independent clause text into a sentence sequence and extracting sentence semantic vectors from the semantic feature vector; determining a global semantic center vector through the sentence semantic vectors and calculating the vector distance between the sentence semantic vectors and the global semantic center vector; and quantifying the semantic dispersion based on the vector distance. This invention transforms abstract semantic ambiguity into a calculable numerical indicator, breaking the limitations of traditional qualitative analysis, making the selection criteria for ambiguous clauses more objective and quantifiable, and enhancing the reliability of the evaluation results.

[0008] Optionally, the step of constructing a historical fuzzy clause sample library and performing multi-dimensional fuzziness quantification on the candidate fuzzy clauses to obtain fuzziness quantification results, thereby determining fuzzy contract clauses, includes: obtaining the text of fuzzy clauses from historical contracts and their fuzziness type tags to construct the historical fuzzy clause sample library; based on the historical fuzzy clause sample library, performing semantic certainty assessment, structural integrity assessment, and logical consistency assessment on the candidate fuzzy clauses to obtain semantic certainty scores, structural integrity indices, and logical contradiction indices as the fuzziness quantification results; constructing an open expression pattern library, and obtaining the pattern matching degree of the candidate fuzzy clauses based on the open expression pattern library; and determining the fuzzy contract clauses through the pattern matching degree and the fuzziness quantification results. This invention achieves a multi-dimensional and comprehensive assessment of clause fuzziness, taking into account both historical experience and the identification of novel fuzzy expressions, effectively avoiding omissions and misjudgments, and improving the comprehensiveness of determining fuzzy contract clauses.

[0009] Optionally, the step of constructing an open expression pattern library and obtaining the pattern matching degree of the candidate ambiguous clauses based on the open expression pattern library includes: constructing the open expression pattern library through various ambiguous expression forms, including degree ambiguity patterns, time ambiguity patterns, scope ambiguity patterns, and conditional ambiguity patterns; based on the open expression pattern library, performing semantic similarity matching between the candidate ambiguous clauses and the ambiguous expression forms to obtain a similarity matching score; and weighted fusion of the similarity matching scores to obtain the pattern matching degree. This invention achieves systematic identification of various ambiguous expressions, accurately captures implicit ambiguity features in clauses, provides a core basis for the accurate determination of ambiguous clauses in contracts, and improves the accuracy of ambiguous clause identification.

[0010] Optionally, the step of performing risk propagation path analysis on the ambiguous contract clauses to obtain a risk propagation impact index, and constructing a risk propagation chain using the ambiguous contract clauses as nodes, includes: constructing a clause dependency relationship network through the element dependency relationships of the ambiguous contract clauses; obtaining the risk propagation impact index of the ambiguous contract clauses based on the clause dependency relationship network; identifying key risk hub clauses based on the risk propagation impact index, and using the key risk hub clauses as the starting point for traversal to construct the risk propagation chain. This invention breaks through the limitations of traditional assessments that focus on a single clause, achieving a full understanding of the contract risk propagation path and providing a clear direction for risk control of ambiguous contract clauses.

[0011] Optionally, obtaining the risk propagation impact index of the ambiguous contract clause based on the clause dependency relationship network includes: obtaining the initial influence value of the ambiguous contract clause in the clause dependency relationship network based on the fuzziness quantification result; calculating the cumulative influence value of the ambiguous contract clause based on the initial influence value and the edge weights of the clause dependency relationship network; and normalizing the cumulative influence value to obtain the risk propagation impact index. This invention accurately identifies key risk hubs through quantitative indices, making the assessment of risk propagation paths more data-driven and improving the accuracy of risk prevention and control.

[0012] Optionally, the step of obtaining the chain reaction amplification factor of the ambiguous contract clause based on the risk propagation chain, and determining the risk assessment result of the ambiguous contract clause based on the chain reaction amplification factor, includes: obtaining the original risk benchmark value of the ambiguous contract clause; obtaining the risk attenuation coefficient and node coupling coefficient based on the risk propagation chain, and calculating the chain reaction amplification factor in combination with the risk propagation impact index; and using the product of the original risk benchmark value and the chain reaction amplification factor as the independent risk assessment value of the ambiguous contract clause to obtain the risk assessment result. This invention quantifies the potential risk amplification capability of clauses, enabling dynamic and accurate assessment of the risks of ambiguous contract clauses, and avoiding underestimation of the potential risks of clauses.

[0013] Optionally, obtaining the risk attenuation coefficient and node coupling coefficient based on the risk propagation chain includes: obtaining the propagation path length from the starting node to the current node in the risk propagation chain, and calculating the risk attenuation coefficient based on the propagation path length; and calculating the node coupling coefficient based on the strength of the dependency relationship between adjacent nodes in the risk propagation chain. This invention makes the calculation of the chain reaction amplification factor more closely resemble actual risk transmission scenarios, improving the scientific rigor and practicality of risk assessment results.

[0014] Secondly, this invention provides an AI-based risk assessment system for ambiguous contract clauses. The system executes the AI-based risk assessment method for ambiguous contract clauses provided by this invention. The system includes an input device, an output device, a processor, and a memory, which are interconnected. The memory stores a computer program, which includes program instructions, and the processor is configured to invoke the program instructions. This invention, through the collaboration of high-performance hardware, automates the assessment process, improving the overall efficiency and scalability of risk assessment for ambiguous contract clauses. Attached Figure Description

[0015] Figure 1 This is a flowchart of an AI-based risk assessment method for ambiguous contract clauses according to an embodiment of the present invention. Figure 2 This is a framework diagram of an AI-based risk assessment system for ambiguous contract clauses, according to an embodiment of the present invention. Detailed Implementation

[0016] Specific embodiments of the present invention will now be described in detail. It should be noted that the embodiments described herein are for illustrative purposes only and are not intended to limit the invention. In the following description, numerous specific details are set forth in order to provide a thorough understanding of the invention. However, it will be apparent to those skilled in the art that these specific details are not necessary to practice the invention. In other instances, well-known circuits, software, or methods have not been specifically described to avoid obscuring the invention.

[0017] Throughout this specification, references to "an embodiment," "an embodiment," "an example," or "an example" mean that a particular feature, structure, or characteristic described in connection with that embodiment or example is included in at least one embodiment of the invention. Therefore, the phrases "in an embodiment," "in an embodiment," "an example," or "an example" appearing in various places throughout the specification do not necessarily refer to the same embodiment or example. Furthermore, specific features, structures, or characteristics can be combined in one or more embodiments or examples in any suitable combination and / or sub-combination. Moreover, those skilled in the art will understand that the illustrations provided herein are for illustrative purposes and are not necessarily drawn to scale.

[0018] Please see Figure 1 One embodiment of the present invention provides an AI-based method for risk assessment of ambiguous contract clauses, the method comprising the following steps: S1. Obtain the contract text to be evaluated, and extract semantic features from the contract text to obtain candidate fuzzy clauses.

[0019] Specifically, S1 includes the following steps: S11. Perform text structure preprocessing on the contract text to be evaluated to obtain independent clause texts.

[0020] First, obtain the contract text to be evaluated. The contract text format includes, but is not limited to, PDF, Word, TXT, and scanned copies. If it is a scanned copy, it needs to be converted into an editable text format using optical character recognition technology.

[0021] Subsequently, the text was cleaned in a structured manner to remove irrelevant information such as headers, footers, watermarks, and annotations from the contract text to be evaluated, to standardize the punctuation format, and to convert full-width characters to half-width characters.

[0022] Next, clause boundary detection is performed based on explicit structural markers in the contract text to be evaluated. Explicit structural markers include, but are not limited to: numbering (such as "1", "Article 1", "a"), line breaks, and specific keywords (such as "Party A" and "Party B" on separate lines). The system uses regular expressions to match the above markers and segments the continuous text into initial paragraphs.

[0023] Then, each initial paragraph is checked for context. If a paragraph begins with introductory words such as "whereas" or "it is hereby proven" and subsequent paragraphs have clear numbers, it is merged into the subsequent clause as a preamble. If a paragraph is numbered consecutively but the content is too short (e.g., less than 10 characters), it is determined to be a title or a blank line.

[0024] In this embodiment, the contract text to be evaluated is processed as described above to obtain independent clause texts. Each independent clause text corresponds to a complete contract clause. This preprocessing structure ensures the quality of subsequent input data and avoids the problem of misaligned clause boundaries caused by format confusion.

[0025] S12. Using the independent clause text as input, extract the semantic feature vector of the independent clause text through a pre-trained language model.

[0026] In this embodiment, each individual clause text obtained from preprocessing is input into a pre-trained language model for the contract domain. The pre-trained language model adopts the BERT model (Bidirectional Encoder Representations from Transformers) based on the Transformer architecture and has been unsupervised pre-trained on a large-scale Chinese contract corpus (including various business contracts, legal documents, judgments, etc.).

[0027] Specifically, during input processing, the independent clause text is first segmented: using a segmenter matched with the pre-trained language model, the text is divided into a sequence of tokens, and [CLS] and [SEP] tags are added to the beginning and end of the sequence, respectively. The token sequence is converted into an input ID, an attention mask, and a paragraph type ID; then, the above three sets of tensors are input into the Transformer encoder of the pre-trained language model. The encoder gradually fuses the contextual semantics through a multi-head self-attention mechanism and a feedforward neural network; finally, the high-dimensional vector corresponding to the [CLS] tag in the final output of the pre-trained language model is extracted as the semantic feature vector of the independent clause text.

[0028] It should be noted that the semantic feature vector integrates the semantic information of the entire clause and can represent the core legal meaning of the clause.

[0029] In an optional embodiment, the semantic feature vectors of all independent clause texts constitute a feature matrix, which transforms the natural language text into a computable numerical vector, laying the data foundation for subsequent semantic discreteness calculation and fuzzy quantization.

[0030] S13. Obtain the semantic dispersion of the independent clause text through the semantic feature vector, and determine the candidate fuzzy clause through the semantic dispersion.

[0031] In this embodiment, for independent clause text, a sentence boundary detection model based on bidirectional long short-term memory network and conditional random field (or simple punctuation rules: period, semicolon, question mark, exclamation mark, etc.) is used to segment it into sentence sequences; each sentence sequence is input into the aforementioned pre-trained language model, and the [CLS] vector of the sentence is extracted from the semantic feature vector as the sentence semantic vector.

[0032] Furthermore, the global semantic center vector is defined as the mean of the semantic vectors of all sentences in the corresponding independent clause text, satisfying the following relationship: in, The global semantic center vector. The total number of sentence sequences. For the index of the sentence sequence, For sentence semantic vectors, An index for the text of the independent clauses.

[0033] Simultaneously, the vector distance (cosine distance or Euclidean distance) between the semantic vector of each sentence and the global semantic center vector is calculated. In this embodiment, cosine distance is preferred because it is not sensitive to the absolute length of the vector and is more suitable for measuring the consistency of semantic direction. The vector distance satisfies the following relationship: in, For vector distance, An index for the text of the independent clauses. For the index of the sentence sequence, For sentence semantic vectors, The global semantic center vector. Let L2 be the L2 norm of the vector.

[0034] In this embodiment, semantic discreteness is quantized based on vector distance. Semantic discreteness is defined as the mean of the vector distances of all sentence sequences, satisfying the following relationship: in, For the first The semantic dispersion of individual clause texts The total number of sentence sequences. For the index of the sentence sequence, This represents the vector distance.

[0035] It should be noted that semantic dispersion reflects the degree of concentration of semantics within the text of independent clauses: the smaller the value, the more consistent the semantics of each sentence and the clear the wording of the clause; the larger the value, the more divergent the semantics between sentences and the existence of internal ambiguity or mixed wording in the clause.

[0036] Furthermore, the system presets a semantic dispersion threshold. If the semantic dispersion is greater than the threshold, the clause is marked as a candidate fuzzy clause and enters the subsequent multi-dimensional fuzziness quantification stage; otherwise, it is judged as a clear clause and no in-depth risk assessment is conducted.

[0037] Specifically, a dynamic threshold strategy is adopted, which determines the semantic dispersion threshold based on the distribution characteristics of the semantic dispersion of all independent clauses in the contract text to be evaluated, satisfying the following relationship: in, The semantic dispersion threshold, The arithmetic mean of semantic dispersion. This is an adjustment coefficient (taken as 1.0 to 1.5). denoted as the standard deviation of semantic dispersion.

[0038] S2. Construct a sample library of historical fuzzy clauses, and perform multi-dimensional fuzziness quantification on the candidate fuzzy clauses to obtain fuzziness quantification results, so as to determine the fuzzy clauses in the contract.

[0039] Specifically, S2 includes the following steps: S21. Obtain the text of the fuzzy clauses in historical contracts and their fuzzy type tags to construct the historical fuzzy clause sample library.

[0040] In this embodiment, historical contracts that have been reviewed and evaluated by professional legal personnel or legal experts are obtained from the company's historical contract database or a publicly available contract case library. Each contract is accompanied by a detailed review report, which clearly indicates the type of ambiguity in the clauses (such as semantic ambiguity, structural omissions, logical contradictions) and the corresponding ambiguity level rating.

[0041] Specifically, each evaluated historical contract is segmented into clauses, and the original text of clauses marked as "fuzzy" is extracted and their fuzziness type labels are recorded. For example, the fuzziness type label for "Party A shall make payment within a reasonable period" in a procurement contract is "time-fuzzy"; the fuzziness type label for "both parties shall assume appropriate confidentiality obligations" in a confidentiality agreement is "scope-fuzzy". The system combines the extracted fuzzy clause text with its label into key-value pairs and stores them in the database.

[0042] Furthermore, metadata is attached to each sample clause: contract type (such as sales contract, lease contract, employment contract), clause position (which clause), and ambiguity score (0-10 points). To ensure the diversity and representativeness of the sample library, the system extracts samples from evaluated contracts in different industries, contract types, and legal fields to construct a historical ambiguity clause sample library.

[0043] Once completed, the historical ambiguous clause sample library supports functions such as searching by tag, filtering by contract type, and sorting by degree of ambiguity. It also serves as a benchmark for subsequent multi-dimensional ambiguity quantification, enabling the system to quickly determine the degree of ambiguity of new clauses by analogy with historical cases.

[0044] S22. Based on the historical fuzzy clause sample library, the candidate fuzzy clauses are subjected to semantic deterministic evaluation, structural integrity evaluation and logical consistency evaluation to obtain semantic deterministic score, structural integrity index and logical contradiction index as the fuzziness quantification result.

[0045] In this embodiment, semantic certainty assessment aims to quantify the clarity of key legal concepts expressed in the clauses.

[0046] First, a pre-trained legal element named entity recognition model is used to extract various legal element words from candidate fuzzy clauses, including but not limited to time elements (such as time limit, date, working day), quantity elements (such as amount, quantity, proportion), behavioral elements (such as delivery, payment, guarantee), and conditional elements (such as unless, if, if).

[0047] It should be noted that the legal element named entity recognition model is based on a bidirectional long short-term memory network and a conditional random field architecture. It has been fully trained on a corpus of contracts labeled with legal elements and can accurately identify the element boundaries and types in candidate fuzzy clauses.

[0048] Next, after extracting the element words, for each element word, it is checked whether there are fuzzy modifiers before and after it. The predefined set of fuzzy modifiers includes: reasonable, appropriate, serious, significant, remarkable, as far as possible, timely, as soon as possible, approximately, around, certain, several, etc. If such modifiers exist, the element is marked as a fuzzy element; otherwise, it is marked as a deterministic element.

[0049] Subsequently, in order to improve the accuracy of the assessment, the system further compares the current clause with the same type of clauses in the historical fuzzy clause sample library. Based on the legal attribute category of the clause (such as breach of contract clauses, price clauses, etc.), the system searches for similar clauses in the sample library and counts the average density of fuzzy elements and common fuzzy patterns in similar clauses.

[0050] Finally, the semantic determinism score is obtained by calculating the deviation between the fuzzy element density of the current candidate fuzzy clause and the average density of the historical fuzzy clause sample library, which satisfies the following relationship: in, For semantic certainty score, For the number of deterministic elements, For the number of fuzzy elements, For the fuzzy element density of the current clause, The average fuzzy density of similar clauses in the sample library. It is a very small positive number.

[0051] In this embodiment, structural integrity assessment is used to detect whether candidate ambiguous clauses are missing core constituent elements that are legally required.

[0052] Build a template library of essential elements for different types of clauses. For example, the essential elements of a breach of contract clause include: the circumstances of the breach, the method of assuming liability for breach of contract, the method of calculating liquidated damages, and grounds for exemption from liability; the essential elements of a price clause include: amount, currency, payment method, payment period, and invoice information; the essential elements of a confidentiality clause include: the scope of confidential information, the confidentiality period, confidentiality obligations, and exceptions.

[0053] When conducting a structural integrity assessment, firstly, the aforementioned legal element named entity recognition model is used to identify elements in candidate ambiguous clauses, obtaining the actual set of elements contained in the candidate ambiguous clauses; then, based on the legal attribute category of the clause, the corresponding element template is retrieved from the essential element template library to obtain the essential element set; subsequently, the structural integrity index is obtained through the actual element set and the essential element set, which is the ratio of the number of essential elements in the detected actual element set to the total number in the essential element set, satisfying the following relationship: in, As a structural integrity indicator, For the collection of actual elements, A collection of essential elements. This indicates taking the intersection. This indicates the number of elements to be retrieved from the set.

[0054] It should be noted that, if This indicates that all essential elements are present; if This indicates that there are missing elements; the more missing elements, the lower the index value.

[0055] In this embodiment, logical consistency assessment is used to detect logical inconsistencies within candidate ambiguous clauses and between the clauses and other parts of the contract.

[0056] Logical relation triples are extracted from the clause text, including conditional relations (such as "if...then..."), causal relations (such as "because...leads to..."), and temporal relations (such as "after..."). The extraction method adopts a relation extraction model based on dependency parsing, which can identify conditional conjunctions, causal conjunctions, and temporal conjunctions and the clauses they govern.

[0057] For each extracted pair of logical relation triples, the system checks for logical conflicts. Common types of internal logical conflicts include: the same action being given contradictory time constraints (such as "Party A shall make payment within 5 days" and "The payment deadline shall be notified separately by Party A"), and two mutually exclusive consequences being stipulated under the same condition (such as "If Party B breaches the contract, Party A has the right to terminate the contract" and "If Party B breaches the contract, both parties shall continue to perform").

[0058] If any internal logical contradiction is detected, the system records the contradiction weight as 0.5. For logical consistency between clauses, the system constructs a logical relationship graph of the entire contract: each clause is a node, and the references, supplements, exclusions, and conflicts between clauses are edges. The system traverses all node pairs to detect whether there are explicit or implicit contradictions. For example, if the main contract clause stipulates that "disputes shall be under the jurisdiction of the court where Party A is located," while the supplementary agreement stipulates that "disputes shall be under the jurisdiction of the court where Party B is located," then there is a jurisdictional conflict.

[0059] In this embodiment, the logical contradiction index satisfies the following relationship: in, The logical contradiction index. The number of contradictory relationships detected. This represents the total number of logical relationships extracted.

[0060] The aforementioned semantic certainty score, structural integrity index, and logical contradiction index together constitute the fuzziness quantification result, which serves as an important basis for subsequently determining ambiguous clauses in contracts.

[0061] S23. Construct an open expression pattern library, and obtain the pattern matching degree of the candidate fuzzy clauses based on the open expression pattern library.

[0062] In this embodiment, the open expression pattern library is a set of semantic frameworks for identifying ambiguous expression patterns in contract texts. Its construction does not rely on a fixed list of keywords, but is based on semantic roles and syntactic structures.

[0063] Specifically, through numerous cases of ambiguous contract clauses annotated by legal experts, four core types of ambiguous expressions have been identified: **Degree Ambiguity:** This type lacks specific quantitative standards when describing quantity, intensity, or quality. A typical structure includes "Subject + Should / Need + Degree Ambiguity Word + Behavior / State," where degree ambiguity words include reasonable, appropriate, significant, substantial, serious, as far as possible, and sufficient. **Time Ambiguity:** This type lacks clear definitions when describing points in time, time periods, or frequencies. A typical structure includes "Behavior + Time Ambiguity Word," where time ambiguity words include timely, as soon as possible, periodically, irregularly, reasonable period, and appropriate time. **Scope Ambiguity:** This type lacks clear limitations when describing the scope of application and the boundaries of responsibility. A typical structure includes "Includes / Coveres + Scope Ambiguity Word + List Items," where scope ambiguity words include related, everything, any, and all. **Condition Ambiguity:** This type lacks clear and verifiable standards when describing triggering conditions. A typical structure includes "If / Such + Condition Ambiguity Word + Then," where condition ambiguity words include change of circumstances, force majeure, when necessary, mutual agreement, and deemed appropriate.

[0064] In this embodiment, for each of the above types of patterns, the system constructs a semantic framework: defining the framework's name, category, core predicate slots, argument slots, and typical examples. It then constructs an open representation pattern library based on the semantic framework and stores it in a structured format (such as JSON). Each pattern entry includes: pattern ID, category label, semantic framework description (including predicate semantic class and argument role constraints), a set of positive examples (at least 5 typical fuzzy representation instances), and a set of negative examples (corresponding clear representation instances).

[0065] It should be noted that the open expression pattern library supports the active learning process to periodically identify uncovered vague expression patterns from newly reviewed contracts. After confirmation by legal experts, these patterns are automatically added to the pattern library, enabling the library to continuously expand and evolve.

[0066] In this embodiment, for each candidate fuzzy clause, the system first segments it into several short sentence or phrase units, each unit being no more than 30 words in length to maintain semantic integrity. Then, for each short sentence unit, the system sequentially matches it with four types of fuzzy patterns in the open representation pattern library to obtain a similarity matching score.

[0067] Similarity matching employs a frame alignment method based on semantic role annotation; specifically, it includes the following steps: First, dependency parsing and semantic role labeling are performed on short sentence units to extract the core predicates and their arguments (including agent, patient, time, place, manner, cause, etc.) in the sentence, forming a semantic framework instance of the sentence.

[0068] For example, for the short sentence unit "Party B shall deliver the goods within a reasonable period", the core predicate extracted is "delivery", the agent argument is "Party B", the patient argument is "goods", and the time argument is "within a reasonable period".

[0069] Then, the extracted semantic frame instances are aligned with each pattern frame in the pattern library. The alignment process includes: predicate similarity for predicate alignment, determining whether the core predicate of the sentence matches the predefined predicate semantic class in the pattern frame (e.g., deliver, provide, transfer); and argument alignment, checking whether the argument types appearing in the sentence cover the key argument slots in the pattern frame, obtaining the number of argument slots. The frame alignment score satisfies the following relationship: in, Score for frame alignment. The number of matching argument slots, This represents the total number of argument slots in the pattern framework. This represents the predicate similarity.

[0070] It should be noted that if the predicate semantic classes match perfectly, the predicate similarity is 1; if they belong to a hypernym relationship, the predicate similarity is 0.7; otherwise, the predicate similarity is 0. If so, the short phrase unit is determined to match the fuzzy pattern.

[0071] Furthermore, for each candidate fuzzy clause, the system aggregates all the patterns matched by the short phrase units and statistically analyzes the matching results by pattern category. The similarity matching score of the candidate fuzzy clause for each fuzzy pattern is defined as the maximum value of the frame alignment score of all short phrase units in that pattern.

[0072] For example, if a clause matches both the degree ambiguity pattern (frame alignment score of 0.88 for short sentence units) and the time ambiguity pattern (frame alignment score of 0.76 for short sentence units), then the degree ambiguity similarity match score is... similarity matching score with temporal ambiguity .

[0073] It should be noted that if a certain pattern does not match any short phrase units, then the similarity matching score for that pattern is 0.

[0074] In this embodiment, the similarity matching scores of candidate fuzzy clauses for four types of fuzzy patterns are obtained. (Degree ambiguous) (Time is unclear) (The scope is ambiguous) After the conditions are ambiguous, the system performs a weighted fusion calculation to determine the overall pattern matching degree of the clause.

[0075] Because different types of ambiguity affect contract risk to varying degrees, the system sets weighting coefficients based on legal expert experience and historical data analysis. Degree ambiguity and time ambiguity directly relate to the clarity of the core rights and obligations of the contract and have the greatest impact on performance disputes, thus receiving higher weights; scope ambiguity affects the boundaries of liability and the scope of application, with a secondary impact; condition ambiguity usually involves triggering conditions under specific circumstances, and its impact is relatively small.

[0076] Specifically, in this embodiment, the weights are set as follows: , , , The pattern matching degree satisfies the following relationship: in, For pattern matching degree, For weights with varying degrees of ambiguity, For similarity matching with fuzzy degree of similarity, For weights with temporal fuzziness, For temporally ambiguous similarity matching scores, For weights with fuzzy scope, For similarity matching scores with fuzzy scope, For weights with fuzzy conditions, The score is given for similarity matches with fuzzy conditions.

[0077] In an optional embodiment, a pattern diversity adjustment factor is introduced; if the number of matched fuzzy pattern categories is less than two, then... Multiply by this factor. This is because a single type of vague statement is often easier to clarify through partial modifications than a clause that mixes multiple vague types, and its overall risk level is relatively lower.

[0078] S24. Determine the ambiguous clauses of the contract using the pattern matching degree and the fuzziness quantification result.

[0079] First, the system normalizes the four indicators and unifies them into numerical values ​​describing the tendency towards ambiguity. Higher semantic certainty scores and structural integrity scores indicate clearer clauses, therefore they need to be converted into ambiguity tendency values: Ambiguity Tendency Values , The logical contradiction index and pattern matching degree are inherently more ambiguous the higher they are, and no transformation is needed.

[0080] Then, the system calculates the comprehensive fuzzy index of the candidate fuzzy clauses, and uses a weighted summation method to satisfy the following relationship: in, To integrate the fuzzy index, These are the weighting coefficients. For semantic certainty score, As a structural integrity indicator, The logical contradiction index. This represents the pattern matching degree.

[0081] It should be noted that the weighting coefficients satisfy... Based on the experience of legal experts, semantic ambiguity has the most fundamental impact on contract interpretation; therefore, this embodiment sets... Structural deficiencies and logical contradictions respectively affect the enforceability and self-consistency of clauses. , Pattern matching degree is used as an auxiliary verification metric, and is set as follows: .

[0082] Furthermore, the system presets two thresholds: a low fuzziness threshold. and high fuzziness threshold .like The clause was deemed clear and will not be included in the subsequent risk propagation analysis process; if If it is determined to be a general ambiguous clause, it is marked as to be observed and selectively included in the analysis; if It was determined to be a highly ambiguous clause and will be included in subsequent risk propagation path analysis and risk assessment.

[0083] In this embodiment, the system stores the text of the contract clauses (generally vague clauses and highly vague clauses) and their corresponding comprehensive fuzziness index and scores of each dimension together as input data for the subsequent risk propagation path analysis module.

[0084] S3. Perform risk propagation path analysis on the ambiguous clauses of the contract to obtain the risk propagation impact index, and construct a risk propagation chain using the ambiguous clauses of the contract as nodes.

[0085] Specifically, S3 includes the following steps: S31. Construct a clause dependency relationship network through the element dependency relationships of the ambiguous clauses in the contract.

[0086] In this embodiment, the determined ambiguous terms of the contract are obtained. Each clause corresponds to its position number in the original contract. The system performs element dependency analysis on each clause: extracting the core legal elements (subject, action, condition, time, amount, etc.) in the clause, and identifying the internal dependencies between elements, such as "payment obligation" depending on the "delivery and acceptance" condition.

[0087] Building upon this, the system further analyzes the cross-clause dependencies between clauses. It iterates through all clause combinations. Detection Does it exist for Explicit or implicit references: Explicit references include direct references to clause numbers (e.g., "see clause X"); implicit references include the use of the same terminology (e.g., ...). Define "deliverables" Using this term), logical sequence (such as...) The provision states that "in case of a dispute", The provision states that "disputes shall be resolved through arbitration".

[0088] Furthermore, based on the reference type, the system defines directed edges. Assign weights Explicit references have a weight of 0.9, implicit references have a weight of 0.7, and logical continuity has a weight of 0.5. Conflict relationships are also detected (e.g., inconsistencies in the same provision), with conflicting edges having a weight of 1.0. Thus, a clause dependency network is constructed using clauses as nodes and directed dependencies as edges. This network provides structured graph data for risk propagation analysis, satisfying the following relationships: in, For the network of clause dependencies, For a set of fuzzy clause nodes, It is a directed edge set. This is the weight matrix for the corresponding edge.

[0089] S32. Based on the aforementioned clause dependency network, obtain the risk propagation impact index of the ambiguous clauses in the contract.

[0090] After constructing the network of contractual dependencies, the system needs to quantify the ability of each node (the ambiguous contractual clause) to propagate risk within the network, i.e., the risk propagation impact index. The calculation of this index involves three progressively advancing steps: First, assign an initial influence value to each node based on the fuzzy quantification results; second, calculate the cumulative influence value of each node based on the network structure and edge weights; finally, normalize the cumulative value to obtain a risk propagation impact index with uniform dimensions.

[0091] The initial influence value reflects the node's potential ability to spread risk outward as a risk source. It should also reflect the ambiguity of the clause itself and its structural importance in the network.

[0092] In this embodiment, the self-fuzziness intensity is derived from the fuzziness quantization result. A maximum value strategy is used to characterize the degree of fuzziness defects most prominent in the clause. The self-fuzziness intensity satisfies the following relationship: in, For its own blur intensity, This indicates taking the maximum value. For semantic certainty score, As a structural integrity indicator, This is the logical contradiction index.

[0093] It should be noted that the self-ambiguity intensity is taken as the maximum value of the three dimensions, indicating the most serious ambiguity problem of the clause in terms of semantics, structure or logic, so as to highlight the structural defects.

[0094] Furthermore, the structural importance of nodes in the network is calculated. In a dependency network, the sum of outgoing edge weights of a node reflects its influence on other terms. The larger the sum of outgoing edge weights, the more downstream nodes directly associated with that term, or the stronger the association, and the higher its potential as a source of risk. Structural importance is defined as the ratio of the sum of the outgoing edge weights of a node to the sum of the largest outgoing edge weights in the network.

[0095] It should be noted that if a node has no outgoing edges, its structural importance is zero.

[0096] Finally, the initial influence value is obtained by fusing its own fuzziness intensity and structural importance, satisfying the following relationship: in, This is the initial influence value. For balance coefficient, For its own blur intensity, For structural importance.

[0097] It should be noted that taking That is, the contribution of its own ambiguity intensity is slightly higher than its structural importance. The reason is that even if a clause is located at a network hub, if it is completely clear (…), it will still contribute more to the overall structure. ), nor should it be considered a high-risk source; conversely, a vague but isolated clause ( It still has some initial influence, but it will be appropriately reduced.

[0098] In this embodiment, risk propagates along directed edges from upstream to downstream nodes in the terms dependency network. The system employs a random walk iterative algorithm with a restart term to simulate this process. Let the nodes... In the The cumulative influence value after the next iteration is initial state In each iteration, the nodes The influence of a node is determined by its initial influence (restart item) and the influence transmitted from upstream nodes. The update rules satisfy the following relationship: in, For nodes In the Cumulative influence value after each iteration For the propagation coefficient, This is the initial influence value. For nodes The upstream node, For all nodes that exist, point to them. The set of upstream nodes of a directed edge. For nodes In the Cumulative influence value after each iteration For nodes To the node edge weights, For nodes downstream node index, For the set of downstream nodes of a node, For nodes To the node Edge weights.

[0099] It should be noted that the propagation coefficient is taken as... This means that in each iteration, 85% of a node's influence comes from upstream propagation, and 15% comes from its own initial value (restart term).

[0100] The above updates are performed iteratively until the sum of the absolute values ​​of the changes in two consecutive iterations for all nodes is less than the convergence threshold. The maximum number of iterations, 100, may be reached. Once converged, the node can be obtained. The cumulative value of influence.

[0101] In this embodiment, since the number and network structure of ambiguous clauses differ in different contracts, the absolute value range of the cumulative influence value is not uniform, making direct cross-contract comparison impossible. Therefore, the cumulative influence value of all ambiguous clauses is subjected to minimum-maximum normalization to obtain a risk propagation impact index with uniform dimensions.

[0102] After normalization, clauses with a risk propagation impact index close to 1 indicate that they have the strongest risk propagation influence in the network and are the core nodes of risk diffusion; clauses with an index close to 0 indicate that their risk impact is weak or isolated. This index serves as the core basis for subsequently identifying key risk hub clauses and constructing risk propagation chains.

[0103] S33. Based on the risk propagation impact index, key risk hub clauses are identified, and the key risk hub clauses are used as the starting point for traversal to construct the risk propagation chain.

[0104] After obtaining the risk propagation impact index for each ambiguous clause in a contract, the system first sets a pivot threshold. In this embodiment This threshold can be dynamically adjusted based on the contract risk sensitivity (it can be lowered to 0.6 when the risk tolerance is low), and ambiguous clauses in contracts that exceed this threshold will be marked as key risk hub clauses.

[0105] Key risk hub clauses have a dual nature: on the one hand, they have prominent inherent ambiguity (reflected in the high degree of self-ambiguity in the initial influence value); on the other hand, they occupy a core position in the dependency network, possessing a large outbound weight and / or a high inbound acceptance, enabling them to widely propagate risks to other clauses.

[0106] Then, the system uses each key risk hub clause as the starting point for traversal and performs a depth-first traversal in the clause dependency network. The traversal strictly follows the direction of the directed edges, starting from the starting point, visiting its direct successor nodes in turn, and then recursively visiting the successors of the successor nodes, until it is impossible to extend further downstream.

[0107] During the traversal, the system records the complete path from the starting point to the current node, forming a risk propagation tree with the starting point as the root node. To avoid infinite traversal caused by circular dependencies, the system maintains a set of visited nodes, and each node appears only once in the same propagation tree. If a visited node is encountered, the traversal along that branch is stopped.

[0108] When multiple key risk hub clauses share the same downstream node, the system retains the path starting from the point with the highest risk propagation impact index, and the remaining paths are cut off at that node.

[0109] Ultimately, the system outputs a set of risk propagation chains. Each risk propagation chain originates from a key risk hub clause and lists the subsequently affected clause nodes in the direction of directed edges, clearly demonstrating the diffusion path and hierarchical relationship of ambiguous risks in the contract.

[0110] S4. Obtain the chain reaction amplification factor of the ambiguous contract clause based on the risk propagation chain, and determine the risk assessment result of the ambiguous contract clause based on the chain reaction amplification factor.

[0111] Specifically, S4 includes the following steps: S41. Obtain the original risk benchmark value of the ambiguous clause in the contract.

[0112] In this embodiment, a geometric mean is used to avoid the over-dominance of a single dimension while reflecting the synergistic effect between dimensions, thereby obtaining the original risk benchmark value, which satisfies the following relationship: in, The original risk benchmark value, For semantic certainty score, As a structural integrity indicator, This is the logical contradiction index.

[0113] It should be noted that the physical meaning of geometric mean is that the overall risk only increases significantly when all three dimensions have a certain degree of defect.

[0114] S42. Based on the risk propagation chain, obtain the risk attenuation coefficient and node coupling coefficient, and calculate the chain reaction amplification factor in combination with the risk propagation impact index.

[0115] In this embodiment, as risk propagates downstream along the directed edges in the propagation chain, its influence gradually decreases with increasing propagation distance. The process involves obtaining the risk propagation chain from the starting node to the current node. propagation path length (In terms of the number of edges, the starting point itself) The risk decay coefficient adopts an exponential decay model, and the decay rate is determined by the decay factor. Control, satisfying the following relationship: in, This is the risk attenuation coefficient. The natural base, As the attenuation factor, This represents the propagation path length.

[0116] It should be noted that taking ;when hour, This indicates that the risk at the starting point itself has not diminished; with As the value increases, the attenuation coefficient decreases exponentially.

[0117] In this embodiment, the node coupling coefficient is used to quantify the risk transmission efficiency between adjacent nodes in the risk propagation chain. The system obtains the position information of the edge from the upstream node to the current node in the propagation chain, including the out-degree of the upstream node in the propagation chain. (i.e., the number of downstream nodes that the upstream node directly points to in the propagation chain) and the depth of the current node in the propagation chain. (Number of edges counting from the starting point). The node coupling coefficients satisfy the following relationship: in, The node coupling coefficient is... The out-degree of the upstream node in the propagation chain. The natural base, This is the depth attenuation coefficient. This represents the depth of the current node in the propagation chain.

[0118] It should be noted that the depth decay coefficient is 0.2; the physical meaning of this formula is that the risk of an upstream node will be evenly distributed to all its direct downstream nodes, and the risk decays exponentially with increasing depth as it propagates along the propagation chain.

[0119] In this embodiment, a weighted superposition method is used to calculate the nodes. The chain reaction amplification factor integrates the risk propagation impact index at the starting point with the attenuation coefficient and coupling coefficient of each node along the path. The path contribution factor is defined as the sum of the products of the attenuation coefficient and coupling coefficient of all nodes along the path, superimposed on the starting point's influence. The chain reaction amplification factor satisfies the following relationship: in, It is a chain reaction amplification factor. The risk transmission impact index is the starting point. Indicates the distance from the starting point to the node. The set of all nodes on the path, For nodes The risk attenuation coefficient, For nodes The node coupling coefficient.

[0120] It should be noted that the starting point itself , The physical meaning of this formula is that during the propagation of the initial risk, the contribution of each node along the path due to its positional attenuation and coupling efficiency is accumulated in a superimposed manner, and finally, together with the influence of the initial risk, it determines the amplification effect. This reflects the characteristic that the risk propagates along the chain by being superimposed at each level rather than simply multiplied.

[0121] S43. The product of the original risk benchmark value and the chain reaction amplification factor is used as the independent risk assessment value of the ambiguous clause in the contract to obtain the risk assessment result.

[0122] In this embodiment, after obtaining the original risk baseline value and chain reaction amplification factor for each ambiguous clause in the contract, the system calculates the independent risk assessment value for that clause. This value combines the inherent risk of the clause itself (original baseline) and its amplification effect (chain reaction) in the risk propagation network, satisfying the following relationship: in, This is an independent risk assessment value. The original risk benchmark value, It is the chain reaction amplification factor.

[0123] Furthermore, the system categorizes ambiguous clauses in the contract into risk levels based on independent risk assessment values. Two risk thresholds are set: a low-risk threshold... and high risk threshold .when When a clause is deemed low-risk, it only needs attention but does not require immediate modification; when When a clause is deemed a medium-risk clause, it is recommended that it be included in the review focus; when If a clause is deemed high-risk, it must be prioritized for modification and become a core point of negotiation.

[0124] In this embodiment, the system outputs the calculated independent risk assessment value and its corresponding risk level as the final risk assessment result of the ambiguous clause in the contract. The assessment result takes into account both the degree of defect of the clause itself and fully reflects its risk propagation influence in the overall contract, thus achieving a comprehensive and quantitative risk assessment of the ambiguous clause.

[0125] Please see Figure 2 In an optional embodiment, the present invention provides an AI-based risk assessment system for ambiguous contract clauses. The system includes an input device, an output device, a processor, and a memory, all interconnected. The memory stores a computer program comprising program instructions. The processor is configured to invoke the program instructions to execute specific steps as described in the embodiments of the AI-based risk assessment method for ambiguous contract clauses provided by the present invention. The AI-based risk assessment system for ambiguous contract clauses provided by the present invention is structurally complete, objective, and stable, enhancing the overall applicability and practical application capabilities of the present invention.

[0126] In summary, the present invention provides an AI-based risk assessment method and system for ambiguous contract clauses. First, semantic features of the contract text are extracted using a pre-trained language model, and semantic dispersion is calculated to screen candidate ambiguous clauses. Then, a multi-dimensional quantitative evaluation of the candidate ambiguous clauses is conducted using a historical ambiguous clause sample library and an open expression pattern library to determine the ambiguous contract clauses. Next, a clause dependency relationship network is constructed to calculate the risk propagation impact index and build a risk propagation chain. Finally, a chain reaction amplification factor is obtained based on the risk propagation chain, and the product of the original risk benchmark value and the amplification factor is used as the independent risk assessment value. The method of this invention is easy to understand, computationally simple, and requires relatively little workload, providing a theoretical foundation and technical support for the further development of the field of intelligent contract review technology.

[0127] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be covered within the scope of the claims and specification of the present invention.

Claims

1. A risk assessment method for ambiguous contract clauses based on AI, characterized in that, Includes the following steps: Obtain the contract text to be evaluated, and extract semantic features from the contract text to obtain candidate fuzzy clauses; A historical fuzzy clause sample library is constructed, and the candidate fuzzy clauses are subjected to multi-dimensional fuzziness quantification to obtain fuzziness quantification results, so as to determine the fuzzy clauses in the contract. A risk propagation path analysis was performed on the ambiguous clauses in the contract to obtain a risk propagation impact index, and a risk propagation chain was constructed using the ambiguous clauses in the contract as nodes. Based on the risk propagation chain, obtain the chain reaction amplification factor of the ambiguous contract clause, and determine the risk assessment result of the ambiguous contract clause according to the chain reaction amplification factor.

2. The AI-based risk assessment method for ambiguous contract clauses according to claim 1, characterized in that, The process of obtaining the contract text to be evaluated and extracting semantic features from the contract text to obtain candidate fuzzy clauses includes: The text structure of the contract to be evaluated is preprocessed to obtain individual clause texts; The independent clause text is taken as input, and the semantic feature vector of the independent clause text is extracted through a pre-trained language model; The semantic discreteness of the independent clause text is obtained through the semantic feature vector, and the candidate fuzzy clause is determined through the semantic discreteness.

3. The AI-based risk assessment method for ambiguous contract clauses according to claim 2, characterized in that, The step of obtaining the semantic discreteness of the independent clause text through the semantic feature vector includes: The independent clause text is segmented into a sentence sequence, and the sentence semantic vector of the sentence sequence is extracted from the semantic feature vector; The global semantic center vector is determined using the sentence semantic vector, and the vector distance between the sentence semantic vector and the global semantic center vector is calculated. The semantic discreteness is quantized based on the vector distance.

4. The AI-based risk assessment method for ambiguous contract clauses according to claim 1, characterized in that, The process of constructing a historical sample library of ambiguous clauses and performing multi-dimensional fuzziness quantification on the candidate ambiguous clauses to obtain fuzziness quantification results, in order to determine ambiguous clauses in the contract, includes: Obtain the text of ambiguous clauses from historical contracts and their ambiguous type tags to construct the historical ambiguous clause sample library; Based on the historical fuzzy clause sample library, the candidate fuzzy clauses are subjected to semantic certainty assessment, structural integrity assessment, and logical consistency assessment to obtain semantic certainty score, structural integrity index, and logical contradiction index as the fuzziness quantification results. Construct an open expression pattern library, and obtain the pattern matching degree of the candidate fuzzy clauses based on the open expression pattern library; The ambiguous clauses in the contract are determined by the pattern matching degree and the fuzziness quantification result.

5. The AI-based risk assessment method for ambiguous contract clauses according to claim 4, characterized in that, The construction of an open expression pattern library, and the acquisition of the pattern matching degree of the candidate fuzzy clauses based on the open expression pattern library, includes: The open representation pattern library is constructed by means of various fuzzy representation forms, including degree fuzzy mode, time fuzzy mode, range fuzzy mode and condition fuzzy mode; Based on the open expression pattern library, the candidate fuzzy clauses and the fuzzy expression forms are semantically similar to obtain a similarity matching score; The pattern matching degree is obtained by weighted fusion of the similarity matching scores.

6. The AI-based risk assessment method for ambiguous contract clauses according to claim 1, characterized in that, The process of analyzing the risk propagation path of the ambiguous contract clauses to obtain a risk propagation impact index, and constructing a risk propagation chain using the ambiguous contract clauses as nodes, includes: A clause dependency network is constructed by the element dependency relationships of the aforementioned ambiguous clauses in the contract; Based on the aforementioned clause dependency network, the risk propagation impact index of the ambiguous clauses in the contract is obtained; Key risk hub clauses are identified based on the risk propagation impact index, and these key risk hub clauses are used as the starting point for traversal to construct the risk propagation chain.

7. The AI-based risk assessment method for ambiguous contract clauses according to claim 6, characterized in that, The process of obtaining the risk propagation impact index of the ambiguous contract clauses based on the aforementioned clause dependency network includes: In the aforementioned clause dependency network, the initial influence value of the ambiguous clause in the contract is obtained based on the fuzziness quantification result; Based on the initial influence value, the cumulative influence value of the ambiguous contract clause is calculated by combining the edge weights of the clause dependency relationship network. The risk propagation impact index is obtained by normalizing the cumulative impact value.

8. The AI-based risk assessment method for ambiguous contract clauses according to claim 1, characterized in that, The process of obtaining the chain reaction amplification factor of the ambiguous contract clause based on the risk propagation chain, and determining the risk assessment result of the ambiguous contract clause based on the chain reaction amplification factor, includes: Obtain the original risk benchmark value for the ambiguous clauses in the contract; The risk attenuation coefficient and node coupling coefficient are obtained based on the risk propagation chain, and the chain reaction amplification factor is calculated by combining the risk propagation impact index. The product of the original risk benchmark value and the chain reaction amplification factor is used as the independent risk assessment value of the ambiguous clause in the contract to obtain the risk assessment result.

9. The AI-based risk assessment method for ambiguous contract clauses according to claim 8, characterized in that, The process of obtaining the risk attenuation coefficient and node coupling coefficient based on the risk propagation chain includes: Obtain the propagation path length from the starting node to the current node in the risk propagation chain, and calculate the risk attenuation coefficient based on the propagation path length; The node coupling coefficient is calculated based on the strength of the dependency relationship between adjacent nodes in the risk propagation chain.

10. An AI-based risk assessment system for ambiguous contract clauses, characterized in that, The system includes an input device, an output device, a processor, and a memory, which are interconnected. The memory stores a computer program, which includes program instructions. The processor is configured to invoke the program instructions to execute the AI-based risk assessment method for ambiguous contract terms as described in any one of claims 1-9.

Images