A brand franchise qualification auditing and risk early warning system and method

Abstract

The application discloses a brand franchising qualification auditing and risk early warning system and method, relates to the technical field of brand franchising qualification auditing, and constructs a dependency syntax relation tree by carrying out chapter level, paragraph level, clause level and lexical level segmentation on a contract text and identifying variant clauses; risk offset amount scores are output by tracking the spatial offset distance and expression intensity asymmetry between obligation clauses and corresponding right clauses through semantic links; the sentiment polarity and sentiment intensity scores of each clause are output by a pre-trained franchising contract sentiment model, and an asymmetrical commitment structure quantification is constructed; the implicit risk feature vectors of historical contracts are clustered, systematic fraud tools are identified, and risk warnings are batch pushed. The application realizes the technical leap of contract examination from surface compliance to deep implicit risk mining, and fills the technical blank of sentiment analysis and structural risk detection in the field of intelligent examination of franchising contracts.

Description

Technical Field

[0001] This invention relates to the field of brand franchise qualification verification technology, specifically to a brand franchise qualification verification and risk warning system and method. Background Technology

[0002] Franchise agreements are the core of franchising legal relationships. They are typically drafted unilaterally by the franchisor with standardized terms, limiting franchisees' bargaining power at the time of signing. Some franchise agreements pose potential risks to franchisees through asymmetrical rights and obligations or hidden standard clauses. Therefore, a brand franchise qualification verification and risk warning system and methodology are needed.

[0003] Existing technology, such as the invention application patent disclosed in CN120725377A, discloses a brand franchise management system and method, which relates to the field of franchise management technology. The system includes a franchise pre-screening module, a franchise management module, and a franchise support module. First, the franchise pre-screening module determines the preset franchise store locations and basic raw material transportation plans for franchisees to select target franchise stores. During operation, the franchise management module collects delivery and product data from target franchise stores, analyzes and judges the quality of franchise stores, and adjusts the raw material transportation plan. For high-quality franchise stores, the franchise support module collects regional product consumption habit data, analyzes and generates differentiated product combination recommendation plans, and adjusts the raw material support transportation plan. From the initial site selection to operation management and support, the system uses data calculation and model analysis to achieve scientific and precise brand franchise management, thereby improving the operational efficiency of franchise stores and brand competitiveness.

[0004] Regarding the aforementioned solutions, the inventors of this application have discovered at least the following technical problems: 1. Existing contract review technologies mainly rely on keyword matching, clause compliance comparison, and surface semantic understanding based on large language models. Their analytical depth is limited to judging the semantic compliance of isolated clauses, lacking the ability to deeply explore the implicit relationships between clauses within a contract. Specifically, existing technologies cannot identify the positional offset relationship between obligation clauses and corresponding right clauses in the contract text space, nor can they detect the asymmetrical structure of expression strength formed by the use of absolute expressions in obligation clauses and vague modifications in corresponding right clauses. This results in a large number of implicit risks designed through clause matching relationships not being automatically identified.

[0005] 2. Existing contract review technologies completely lack a sentiment analysis dimension. All AI-based contract analysis solutions treat contract clauses as purely semantic facts, without incorporating quantitative analysis of sentiment polarity and intensity. This results in existing technologies being unable to identify implicit suggestive language, patterns of shirking responsibility, and asymmetrical sentiment structures between obligation and rights clauses in contract texts. Contract review remains at the level of factual judgment and cannot delve into the level of intent judgment, thus making it difficult to detect fraudulent contracts that use neutral language to conceal unfavorable obligations.

[0006] 3. Existing contract review technologies have significant gaps in cross-contract analysis and external verification. Current solutions cannot conduct longitudinal comparisons of different versions of contract templates from the same brand to detect the evolution trend of risk clauses, nor can they cross-verify the results of uncovering hidden risks within the contract with publicly available judicial judgments to enhance the legal credibility of early warning results. As a result, systemic fraud tools cannot be detected in a timely manner, and early warning conclusions lack case law support, making it difficult for them to play a substantial role in dispute prevention and judicial proceedings. Summary of the Invention

[0007] To address the aforementioned technical shortcomings, the present invention aims to provide a brand franchise qualification review and risk warning system and method.

[0008] To solve the above-mentioned technical problems, the present invention adopts the following technical solution: In the first aspect, the present invention provides a brand franchise qualification review and risk warning system, including the following modules: multi-granularity semantic segmentation and dependency parsing module: used to perform multi-granularity semantic segmentation on the target franchise contract text of the target enterprise according to the chapter level, paragraph level, clause level and lexical level, and identify variant clauses with the same semantic substance but different expressions.

[0009] Implicit Obligation and Right Offset Detection Module: This module is used to semantically match and align each obligation clause in the target franchise agreement with its corresponding right clause. It constructs a semantic link from the obligation clause to the right clause through dependency parsing. If it detects that the obligation clause has no corresponding right clause as a semantic endpoint, or that the right clause is placed at a position that exceeds the preset standard distance threshold from the obligation clause, or that the obligation clause uses absolute expression while the corresponding right clause uses fuzzy modification, then the clause is determined to be an implicit risk clause and a risk offset score is output.

[0010] The sentiment and fraud intent analysis module is used to perform in-depth analysis of the sentiment of the target franchise contract terms through a pre-trained franchise contract sentiment model. It identifies implicit suggestive word clusters, liability inference patterns and asymmetric commitment structures in the target franchise contract text, and outputs a sentiment risk score that includes sentiment polarity distribution and sentiment intensity differences.

[0011] Cross-contract pattern clustering and early warning module: It is used to cluster and analyze the hidden risk mining results of all historical franchise contracts of the same brand owner for the target franchise contract. If the same contract template is detected to repeatedly induce the same type of dispute between different franchisees, the contract template is identified as a systemic fraud tool and risk warnings are pushed to all existing and potential franchisees in batches.

[0012] In a second aspect, the present invention provides a method for brand franchise qualification review and risk warning, comprising the following steps: Step 1, multi-granular semantic segmentation and dependency parsing: the target franchise contract text of the target enterprise is segmented into multiple granular semantic segments at the chapter level, paragraph level, clause level and lexical level to identify variant clauses that have the same semantic substance but different expressions.

[0013] Step 2, Implicit Obligation and Right Shift Detection: Semantically match and align each obligation clause in the target franchise agreement with its corresponding right clause. Construct a semantic link from the obligation clause to the right clause through dependency parsing. If an obligation clause is found to have no corresponding right clause as a semantic endpoint, or the right clause is placed at a position that exceeds the preset standard distance threshold from the obligation clause, or the obligation clause uses absolute expression while the corresponding right clause uses fuzzy modification, then the clause is determined to be an implicit risk clause and a risk shift score is output.

[0014] Step 3: Sentiment and Fraudulent Intent Analysis: Through a pre-trained franchise contract sentiment model, an in-depth analysis of the sentiment of the target franchise contract terms is conducted to identify implicit suggestive word clusters, liability inference patterns, and asymmetric commitment structures in the target franchise contract text, and output sentiment risk scores that include sentiment polarity distribution and sentiment intensity differences.

[0015] Step 4: Cross-contract pattern clustering and early warning: Cluster analysis is performed on the hidden risk mining results of all historical franchise contracts of the same brand owner for the target franchise contract. If the same contract template is detected to repeatedly induce the same type of dispute between different franchisees, the contract template is identified as a systemic fraud tool, and risk warnings are pushed to all existing and potential franchisees in batches.

[0016] The beneficial effects of this invention are as follows: 1. This embodiment of the solution is based on an implicit risk detection mechanism derived from the analysis of the alignment relationship between obligation clauses and rights clauses. Existing contract review technologies are limited to semantic compliance judgments of isolated clauses and cannot identify structural risks between clauses. This invention, by constructing a semantic link between obligation clauses and rights clauses, for the first time uses the spatial offset and asymmetric expression intensity of rights clauses relative to obligation clauses as the core dimensions of risk detection. When rights clauses are deliberately placed far after obligation clauses or when obligation clauses use absolute expressions while corresponding rights clauses use vague modifications, risk counting is automatically triggered. This fundamentally changes the technical paradigm of contract risk review from partial clause analysis to overall structural perspective, enabling the systematic and automatic identification of implicit risks long hidden in the alignment relationship of contract clauses.

[0017] 2. This embodiment of the solution introduces sentiment analysis technology into the field of franchise contract review and constructs specialized quantitative models for sentiment polarity skewness coefficient and sentiment intensity difference coefficient. Existing technologies treat contract clauses entirely as semantic facts, failing to identify fraudulent contracts that use neutral language to conceal unfavorable obligations. This invention, through a pre-trained franchise contract sentiment model, independently outputs sentiment polarity and sentiment intensity scores for each contract clause. It then constructs a sentiment polarity skewness coefficient to quantify the degree of sentiment deviation between the obligation clause group and the rights clause group, and constructs a sentiment intensity difference coefficient to quantify the asymmetrical distribution of the two types of clauses in terms of sentiment intensity. This achieves a leap in the analytical dimension of contract text from the level of factual judgment to the level of intent judgment, filling the technological gap of completely lacking sentiment analysis in the field of intelligent contract review.

[0018] 3. This solution constructs a complete technical chain from isolated analysis of a single contract to multi-dimensional systemic fraud identification, and forms a closed-loop iterative mechanism for monitoring, auditing, early warning, verification, and correction. Existing technologies cannot perform longitudinal comparisons of different versions of contract templates from the same brand to detect the version evolution trend of risk clauses, nor can they cross-verify the results of internal contract mining with publicly available judicial judgments to enhance the legal credibility of early warning conclusions. This invention aggregates highly consistent risk characteristic patterns from the historical contracts of the same brand into a standard risk profile of systemic fraud tools through cross-contract pattern clustering. It uses version evolution risk gradient quantification to detect the brand's hidden behavior of systematically strengthening risk clauses through contract version iterations. It achieves cross-verification of hidden risk mining results with real precedents through semantic matching of judicial judgments, upgrading early warning conclusions from algorithmic speculation to credible risk assessment results supported by case law. This provides a complete technical solution for franchise contract risk prevention and control, from single-point detection to systematic defense. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a schematic diagram of the system structure connection of the present invention.

[0021] Figure 2 This is a schematic diagram of the implementation steps of the method of the present invention. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] Examples of embodiments of the present invention Figure 1 As shown, a brand franchise qualification review and risk warning system includes the following modules: multi-granularity semantic segmentation and dependency syntax analysis module, implicit obligation and right offset detection module, sentiment tendency and fraudulent intent analysis module, and cross-contract pattern clustering and early warning module.

[0024] The implicit obligation and right offset detection module is connected to the multi-granularity semantic segmentation and dependency parsing module and the sentiment tendency and fraudulent intent analysis module, respectively. The sentiment tendency and fraudulent intent analysis module is connected to the cross-contract pattern clustering and early warning module.

[0025] Multi-granularity semantic segmentation and dependency parsing module: used to perform multi-granularity semantic segmentation of the target company's franchise contract text at the document level, paragraph level, clause level and lexical level, and identify variant clauses with the same semantic substance but different expressions.

[0026] In a specific embodiment, the target franchise agreement text of the target enterprise is subjected to multi-granularity semantic segmentation at the chapter level, paragraph level, clause level and lexical level. The specific segmentation process is as follows: Chapter-level segmentation identifies the macro-structural boundaries of the contract name, the given conditions clause, the definition clause, the rights and obligations clause, the liability for breach of contract clause and the dispute resolution clause.

[0027] It should be noted that the document-level segmentation first identifies the start and end points of each macro section by matching the chapter, section, and clause heading patterns in the contract text using regular expressions. Specifically, the contract name is identified by matching consecutive heading lines ending with keywords such as "contract," "agreement," or "book" at the beginning of the text. "Wherein" clauses are identified by matching independent paragraph combinations beginning with introductory words such as "wherein," "considering," or "for clarity." "Definition" clauses are identified by matching consecutive clause blocks containing keywords such as "definition," "the following terms," ​​"referring to," or "referring to," usually located after "wherein" clauses. "Rights and obligations" clauses are identified by matching clause groups containing high-frequency modal verbs such as "rights," "obligations," "have the right," "should," or "must," and then aggregating and identifying them based on the semantic classification results of the clause titles or the first sentence of the clauses. The breach of contract clauses are identified by matching clauses containing keywords such as breach, termination, compensation, and remedy. The dispute resolution clauses are identified by matching clauses containing keywords such as dispute, arbitration, litigation, and jurisdiction, which are usually located at the end of the contract. After the title level identification, the system further completes the boundaries based on the hierarchical structure of the clause numbers. When the hierarchical change of the clause number indicates the start of a new chapter, the macro-structure boundary is automatically segmented. For contract texts that do not use a chapter numbering system, the semantic difference between adjacent clauses is calculated through a sliding window based on the contextual mutation points of the clause semantic vector. The position where the difference exceeds the preset threshold is marked as the chapter boundary. Finally, a complete chapter-level structure division result is formed and passed to the paragraph-level segmentation module.

[0028] Paragraph-level segmentation identifies the boundaries of each natural paragraph and marks the paragraph number.

[0029] Clause-level segmentation divides each numbered clause into an independent semantic unit, and segments unnumbered natural paragraphs according to semantic integrity.

[0030] Lexical segmentation performs word segmentation, part-of-speech tagging, and named entity recognition on keywords in the clauses.

[0031] The dependency syntax relation tree is constructed with clauses as nodes and semantic references, logical progressions, conditional constraints, and definition references as edges, forming a directed acyclic graph structure.

[0032] In a specific embodiment, the identification of variant clauses that have the same semantic substance but different expressions is carried out as follows: Variant clause identification is achieved by calculating the cosine similarity of the semantic vectors of different clauses, and clauses with similarity higher than a threshold are normalized to the same semantic category. The formula for calculating the cosine similarity of semantic vectors is as follows: ,in, Terms and conditions and The semantic encoding vector, when A value ≥0.85 is considered a variant clause.

[0033] Implicit Obligation and Right Offset Detection Module: This module is used to semantically match and align each obligation clause in the target franchise agreement with its corresponding right clause. It constructs a semantic link from the obligation clause to the right clause through dependency parsing. If it detects that the obligation clause has no corresponding right clause as a semantic endpoint, or that the right clause is placed at a position that exceeds the preset standard distance threshold from the obligation clause, or that the obligation clause uses absolute expression while the corresponding right clause uses fuzzy modification, then the clause is determined to be an implicit risk clause and a risk offset score is output.

[0034] In a specific embodiment, the semantic link from the obligation clause to the right clause is constructed through dependency parsing. The specific analysis process is as follows: the identification of the obligation clause is based on the matching of a mandatory modal verb lexicon, which includes ought, must, guarantee, and promise.

[0035] The identification of rights clauses is based on matching with a lexicon of authorizing modal verbs, which includes the right to, may, and enjoy.

[0036] It should be noted that, firstly, the lexical matching should employ a composite strategy combining exact matching and lemmatization. This involves not only matching the original forms of words like "should," "must," "guarantee," and "promise," but also mapping variant words like "should," "must," "ensure," and "promise" to their corresponding lexical forms through lemmatization algorithms to achieve generalized matching. Simultaneously, a negative word detection mechanism should be introduced to suppress the activation state of a modal verb when negative words like "not," "not," or "not" appear before it, thus avoiding misjudgment. Secondly, the semantic category determination of clauses should be combined with dependency parsing results for syntactic role verification. Only when a mandatory modal verb functions as an adverbial modifier of a core predicate or as the active verb in the syntactic dependency tree can it be considered an activation signal for an obligation clause. If the modal verb appears in a descriptive statement of the term's meaning in a definition clause or in a "given" clause as merely a description of background facts without specific rights or obligations, then false detections should be excluded based on the syntactic structure at the depth of the modal verb and whether its subject is a clearly defined party to the contract. Third, for implicit obligation clauses that lack mandatory modal verbs but should be identified as obligation clauses based on the contract context, the system needs to further analyze whether the clause contains a combination of frequency adverbs and action verbs, or whether it indirectly expresses the obligation through a causal relationship structure of conditional adverbial clauses and result clauses. If it is determined to be one of these situations, it will also be included in the set of obligation clauses. Fourth, the identification of rights clauses, in addition to matching authorizing modal verbs, also needs to detect whether the rights clause is subject to preconditions. If there are restrictive modifiers such as written consent from the brand owner or within the scope specified by the brand owner, the enforceability of the rights clause needs to be assessed based on the dependency hierarchy of the restrictive condition in the syntactic structure and the degree to which it limits the scope of the authorized rights. This assessment is used to assist in judging the validity of the rights clause in subsequent offset detection. Fifth, the system outputs a corresponding identification confidence score for each identified obligation clause and rights clause. Clauses with scores lower than the preset acceptance threshold will be submitted for manual review to ensure the quality of input data for downstream semantic link construction and offset detection.

[0037] For each obligation clause, its semantic encoding vector is extracted. Within the entire contract text, an attention mechanism is used to calculate the matching degree between this vector and the semantic encoding vector of each right clause. The right clause with the highest matching degree is selected as the corresponding clause. The matching degree calculation formula is as follows: ,in, Indicates the first Obligations clauses Indicates the first Rights and obligations For the first The semantic encoding vector of the obligations clause, For the first The semantic encoding vector of the rights clause, For encoding dimensions, The total number of rights clauses; All rights and obligations The semantic link is constructed with the obligation clause as the starting node, the corresponding right clause as the ending node, and the matching degree as the edge weight, forming a directed edge.

[0038] If the global highest matching degree of a certain obligation clause is lower than the preset matching threshold of 0.6, it is marked as an isolated obligation clause without a corresponding rights clause.

[0039] In a specific embodiment, the process of determining the clause as a hidden risk clause and outputting a risk offset score is as follows: The determination of no corresponding rights clause is based on the global highest matching degree being lower than a preset matching threshold of 0.6, triggering a first type of risk count; the distance offset is calculated using the character offset between the rights clause and the obligation clause in the contract text as the unit of measurement. If the offset exceeds the standard distance threshold dynamically calculated based on the total contract length, a second type of risk count is triggered. The distance offset factor calculation formula is: ,in, For character offset, The total number of characters in the contract. .

[0040] The strength of expression is detected by matching a pre-trained absolute expression lexicon with a fuzzy modification lexicon. If an obligation clause matches an absolute expression lexicon and the corresponding right clause matches a fuzzy modification lexicon, a third type of risk count is triggered. The formula for calculating the expression strength asymmetry factor is as follows: ,in, The number of absolute statements in the obligation clauses. The number of times the vague modifiers in the corresponding rights clauses are hit. This refers to the total number of words in each clause.

[0041] It should be noted that, firstly, the construction of the absolute expression lexicon needs to cover three semantic levels. The first level consists of unconditional limiting words, including words such as any, all, everything, and regardless of whether they are used to eliminate conditional restrictions. The second level consists of degree-enhancing words, including words such as absolute, complete, thorough, and sufficient to enhance the indisputability of obligations. The third level consists of time-constant words, including words such as always, forever, at any time, and long-term to eliminate time boundary constraints. The lexicon is incrementally expanded by reverse mining clauses deemed by the court to be unbalanced in rights and obligations from publicly available franchise contract dispute judgments, and the generalization coverage of the lexicon is achieved by using a synonym embedding method. Second, the construction of the fuzzy modifier lexicon needs to cover two semantic directions. The first direction is conditional modifiers, including modifiers such as "when conditions permit," "after consent," "according to the actual situation," and "within a reasonable range," which are used to attach preconditions to the exercise of rights. The second direction is degree-weakening modifiers, including modifiers such as "appropriate," "as much as possible," "as far as possible," "relevant," and "corresponding," which are used to reduce the certainty of rights protection. The lexicon is expanded in a supervisory manner by analyzing the differences in expression between standard format contract texts and fair contract texts confirmed by court judgments. Third, the execution of matching rules needs to adopt a composite strategy combining exact matching and semantic similarity matching. For standard words already included in the lexicon, exact character matching is used; for expressions without variants, semantic vector cosine similarity calculation is used. When the similarity between an expression and any word in the lexicon exceeds a preset threshold, it is considered a match. Fourth, the determination of asymmetry requires further calculation of the ratio between the absolute expression density of the obligation clause and the fuzzy modification density of the right clause, under the condition that the clauses match successfully. The third type of risk counting is triggered only when the ratio exceeds the preset asymmetry threshold. In the case where there are multiple absolute expressions in the obligation clause and only a single fuzzy modification in the corresponding right clause, the system outputs a continuous value of the expression intensity asymmetry factor based on the exponential decay relationship between the product of the number of absolute expressions and the number of fuzzy modification relative to the total number of words in the clause. The closer the value is to 1, the higher the degree of asymmetry, thereby realizing a refined quantitative assessment of the expression intensity asymmetry.

[0042] The risk offset score is calculated by combining the above three types of risk factors: ,in, The indicator variable for isolated obligation clauses is 1 if it exists and 0 otherwise. The score is normalized to the 0-1 interval. A value of ≥0.7 is considered a high-risk implicit offset clause.

[0043] The sentiment and fraud intent analysis module is used to perform in-depth analysis of the sentiment of the target franchise contract terms through a pre-trained franchise contract sentiment model. It identifies implicit suggestive word clusters, liability inference patterns and asymmetric commitment structures in the target franchise contract text, and outputs a sentiment risk score that includes sentiment polarity distribution and sentiment intensity differences.

[0044] In a specific embodiment, the identification process for implicit suggestive word clusters, liability inference patterns, and asymmetric commitment structures in the target franchise agreement text is as follows: The franchise agreement sentiment model uses a pre-trained language model as a base and undergoes supervised fine-tuning training on domain corpora containing franchise agreement texts and franchise dispute judgments, outputting the sentiment polarity scores of each obligation clause. And the emotional polarity score of each rights clause And the emotional intensity score of each obligation clause. And the emotional intensity score of each rights clause .

[0045] The identification of suggestive tone clusters is achieved through a pattern matching algorithm. The frequency of co-occurrence of suggestive words with excessive promises and words promising high returns in the target word list is detected. When the frequency exceeds the threshold of 0.3, it is marked as a risk of suggestive tone.

[0046] The identification of the liability presumption pattern is achieved by analyzing the explicitness or implicitness of the subject of the sentence and the voice of the verb. If the responsible party in the clause uses the passive voice, omits the subject, or uses a generic pronoun without explicit reference, it is determined to be a liability presumption pattern.

[0047] The identification of asymmetric commitment structures involves comparing the sentiment polarity distribution of obligation clauses and rights clauses, introducing a sentiment polarity skewness coefficient: ,in, and These refer to the total number of obligation clauses and the total number of right clauses in the contract. This represents the overall mean of the emotional polarity across all clauses.

[0048] when Furthermore, if the number of neutral clauses with an emotional intensity score below 0.2 in the obligation clauses exceeds 40%, it is judged as an asymmetric commitment structure.

[0049] In a specific embodiment, the output includes an emotional risk score based on the distribution of emotional polarity and the difference in emotional intensity. The specific output process is as follows: the distribution of emotional polarity is presented in the form of a histogram or box plot, showing the distribution of polarity scores between the obligation clause group and the rights clause group, and marking the positions of abnormal deviation clauses.

[0050] The difference in emotional intensity is based on the difference between the mean emotional intensity of the obligation clause group and the rights clause group. The formula for calculating the emotional intensity difference coefficient is as follows: , When the difference exceeds the preset difference threshold of 0.25, the emotional risk score increases linearly; the emotional risk score adopts a logarithmic-exponential fusion formula: ,in, =1.5, when An emotional risk alert is triggered when the value is ≥0.65, and a paragraph with high-risk clauses is displayed.

[0051] Cross-contract pattern clustering and early warning module: It is used to cluster and analyze the hidden risk mining results of all historical franchise contracts of the same brand owner for the target franchise contract. If the same contract template is detected to repeatedly induce the same type of dispute between different franchisees, the contract template is identified as a systemic fraud tool and risk warnings are pushed to all existing and potential franchisees in batches.

[0052] In one specific embodiment, the clustering analysis of the implicit risk mining results of all historical franchise contracts of the same brand as the target franchise contract is performed as follows: All historical franchise contracts of the same brand as the target franchise contract are retrieved from the system database, and the implicit risk feature vector of each contract is extracted. , , , and ,in, For the contract sample index number, , The total number of contract samples. Encoding for asymmetric commitment structure types, The entropy value represents the location distribution of high-risk clauses. This refers to the number of risk clauses.

[0053] A density-based clustering algorithm is used to perform unsupervised clustering of risk feature vectors, with cosine distance as the distance metric. ,in, For the index number of another contract sample, the neighborhood radius Set the minimum number of samples to 15% of the feature space diameter. Introducing a clustering risk density coefficient: ,in, For clustering, The number of samples within the cluster. The cluster center is the mean vector of all samples within the cluster. is the standard deviation of the distance within the cluster.

[0054] If the number of contracts within a cluster is ≥5 and If the value is ≥0.75, the contract template corresponding to that cluster is identified as a systemic fraud tool, and the cluster center is determined accordingly. As a standard risk profile.

[0055] In one specific embodiment, the process of sending risk warnings to all existing and potential franchisees in batches is as follows: The risk warning includes the identification of contract templates identified as systemic fraud tools, excerpts and positioning of high-risk clauses, and risk deviation scores. With Emotional Risk Score And risk avoidance advice for franchisees; The push channels include message center push notifications, SMS notifications, and emails for existing franchisees, and risk warning pop-ups that are forcibly displayed through the system interface when potential franchisees inquire about brand information.

[0056] All push notifications are generated into immutable logs and stored in an audit database.

[0057] Examples of embodiments of the present invention Figure 2 As shown, a brand franchise qualification review and risk warning method includes the following steps: Step 1, multi-granular semantic segmentation and dependency parsing: The target franchise contract text of the target enterprise is segmented into multiple granular semantic segments at the chapter level, paragraph level, clause level and lexical level to identify variant clauses with the same semantic substance but different expressions.

[0058] Step 2, Implicit Obligation and Right Shift Detection: Semantically match and align each obligation clause in the target franchise agreement with its corresponding right clause. Construct a semantic link from the obligation clause to the right clause through dependency parsing. If an obligation clause is found to have no corresponding right clause as a semantic endpoint, or the right clause is placed at a position that exceeds the preset standard distance threshold from the obligation clause, or the obligation clause uses absolute expression while the corresponding right clause uses fuzzy modification, then the clause is determined to be an implicit risk clause and a risk shift score is output.

[0059] Step 3: Sentiment and Fraudulent Intent Analysis: Through a pre-trained franchise contract sentiment model, an in-depth analysis of the sentiment of the target franchise contract terms is conducted to identify implicit suggestive word clusters, liability inference patterns, and asymmetric commitment structures in the target franchise contract text, and output sentiment risk scores that include sentiment polarity distribution and sentiment intensity differences.

[0060] Step 4: Cross-contract pattern clustering and early warning: Cluster analysis is performed on the hidden risk mining results of all historical franchise contracts of the same brand owner for the target franchise contract. If the same contract template is detected to repeatedly induce the same type of dispute between different franchisees, the contract template is identified as a systemic fraud tool, and risk warnings are pushed to all existing and potential franchisees in batches.

[0061] The examples described in this invention are not limited to the specific embodiments listed above. The examples are merely illustrative to facilitate understanding of the invention and do not constitute a limitation on the scope of protection of this invention. Any modifications, equivalent substitutions, etc., made within the spirit and principles of this invention should be included within the scope of protection.

[0062] The above description is merely an example and illustration of the concept of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described or use similar methods to replace them, as long as they do not deviate from the concept of the invention or exceed the scope defined in this specification, they should all fall within the protection scope of the present invention.

Claims

1. A brand franchise qualification review and risk warning system, characterized in that, Includes the following modules: Multi-granularity semantic segmentation and dependency parsing module: used to perform multi-granularity semantic segmentation of the target enterprise's franchise contract text at the chapter level, paragraph level, clause level and lexical level, and identify variant clauses with the same semantic substance but different expressions; Implicit Obligation and Right Offset Detection Module: This module is used to semantically match and align each obligation clause in the target franchise agreement with its corresponding right clause. It constructs a semantic link from the obligation clause to the right clause through dependency parsing. If it detects that the obligation clause has no corresponding right clause as a semantic endpoint, or that the right clause is placed at a position that exceeds the preset standard distance threshold from the obligation clause, or that the obligation clause uses absolute expression while the corresponding right clause uses fuzzy modification, then the clause is determined to be an implicit risk clause and a risk offset score is output. The sentiment and fraud intent analysis module is used to perform in-depth analysis of the sentiment of the target franchise contract terms through a pre-trained franchise contract sentiment model. It identifies implicit suggestive word clusters, liability inference patterns and asymmetric commitment structures in the target franchise contract text, and outputs a sentiment risk score that includes sentiment polarity distribution and sentiment intensity differences. Cross-contract pattern clustering and early warning module: It is used to cluster and analyze the hidden risk mining results of all historical franchise contracts of the same brand owner for the target franchise contract. If the same contract template is detected to repeatedly induce the same type of dispute between different franchisees, the contract template is identified as a systemic fraud tool and risk warnings are pushed to all existing and potential franchisees in batches.

2. The brand franchise qualification review and risk warning system according to claim 1, characterized in that, The target franchise agreement text of the target enterprise is segmented semantically at the document level, paragraph level, clause level, and lexical level. The specific segmentation process is as follows: Chapter-level segmentation identifies the macro-structural boundaries of contract names, disclaimers, definition clauses, rights and obligations clauses, liability for breach of contract clauses, and dispute resolution clauses; Paragraph-level segmentation identifies the boundaries of each natural paragraph and marks the paragraph number; Clause-level segmentation divides each numbered clause into an independent semantic unit, and segments unnumbered natural paragraphs according to semantic integrity. Lexical segmentation performs word segmentation, part-of-speech tagging, and named entity recognition on keywords in the clauses. The dependency syntax relation tree is constructed with clauses as nodes and semantic references, logical progressions, conditional constraints, and definition references as edges, forming a directed acyclic graph structure.

3. The brand franchise qualification review and risk warning system according to claim 2, characterized in that, The identification process for variant clauses that have the same semantic substance but different expressions is as follows: Variant clause identification normalizes clauses with similarity scores above a threshold into the same semantic category by calculating the cosine similarity of their semantic vectors. The formula for calculating the cosine similarity of semantic vectors is as follows: ,in, Terms and conditions and The semantic encoding vector, when A value ≥0.85 is considered a variant clause.

4. The brand franchise qualification review and risk warning system according to claim 3, characterized in that, The semantic link from obligation clauses to rights clauses is constructed through dependency parsing. The specific analysis process is as follows: The identification of obligations is based on matching with a lexicon of mandatory modal verbs, which include ought, must, guarantee, and promise; The identification of rights clauses is based on matching with a lexicon of authorizing modal verbs, which includes the verbs of right, may, and enjoy. For each obligation clause, its semantic encoding vector is extracted. Within the entire contract text, an attention mechanism is used to calculate the matching degree between this vector and the semantic encoding vector of each right clause. The right clause with the highest matching degree is selected as the corresponding clause. The matching degree calculation formula is as follows: ,in, Indicates the first Obligations clauses Indicates the first Rights and obligations For the first The semantic encoding vector of the obligations clause, For the first The semantic encoding vector of the rights clause, For encoding dimensions, The total number of rights clauses; All rights and obligations The semantic link is constructed with the obligation clause as the starting node, the corresponding right clause as the ending node, and the matching degree as the edge weight, forming a directed edge; If the global highest matching degree of a certain obligation clause is lower than the preset matching threshold of 0.6, it is marked as an isolated obligation clause without a corresponding rights clause.

5. The brand franchise qualification review and risk warning system according to claim 4, characterized in that, The process of identifying this clause as a hidden risk clause and outputting a risk offset score is as follows: The determination of the absence of corresponding rights clauses is based on the global highest matching degree being lower than the preset matching threshold of 0.6, triggering the first type of risk count. The distance offset is calculated using the character offset between the rights clauses and obligations clauses in the contract text as the unit of measurement. If the offset exceeds the standard distance threshold dynamically calculated based on the total length of the contract, the second type of risk count is triggered. The distance offset factor calculation formula is as follows: ,in, For character offset, The total number of characters in the contract. ; The strength of expression is detected by matching a pre-trained absolute expression lexicon with a fuzzy modification lexicon. If an obligation clause matches an absolute expression lexicon and the corresponding right clause matches a fuzzy modification lexicon, a third type of risk count is triggered. The formula for calculating the expression strength asymmetry factor is as follows: ,in, The number of absolute statements in the obligation clauses. The number of times the vague modifiers in the corresponding rights clauses are hit. For the total number of words in each clause; The risk offset score is calculated by combining the above three types of risk factors: ,in, The indicator variable for isolated obligation clauses is 1 if it exists and 0 otherwise. The score is normalized to the 0-1 interval. A value of ≥0.7 is considered a high-risk implicit offset clause.

6. The brand franchise qualification review and risk warning system according to claim 5, characterized in that, The identification process for implicit suggestive phrases, liability presumption patterns, and asymmetric commitment structures in the target franchise agreement text is as follows: The franchise agreement sentiment model uses a pre-trained language model as its foundation and undergoes supervised fine-tuning training on a domain corpus containing franchise agreement texts and franchise dispute judgments. It outputs the sentiment polarity scores for each obligation clause. And the emotional polarity score of each rights clause And the emotional intensity score of each obligation clause. And the emotional intensity score of each rights clause ; The identification of suggestive tone word clusters is achieved through a pattern matching algorithm. The frequency of co-occurrence of suggestive words with excessive promises and words with promises of high returns in the target word list is detected. When the frequency exceeds the threshold of 0.3, it is marked as a risk of suggestive tone. The identification of the liability presumption pattern is achieved by analyzing the explicitness or implicitness of the subject of the sentence and the voice of the verb. If the responsible party in the clause uses the passive voice, omits the subject, or uses a generic pronoun without explicit reference, it is determined to be a liability presumption pattern. The identification of asymmetric commitment structures involves comparing the sentiment polarity distribution of obligation clauses and rights clauses, introducing a sentiment polarity skewness coefficient: ,in, and These refer to the total number of obligation clauses and the total number of right clauses in the contract. The overall mean of the emotional polarity of all clauses; when Furthermore, if the number of neutral clauses with an emotional intensity score below 0.2 in the obligation clauses exceeds 40%, it is judged as an asymmetric commitment structure.

7. The brand franchise qualification review and risk warning system according to claim 6, characterized in that, The output includes an emotional risk score that reflects the difference in emotional polarity distribution and emotional intensity. The specific output process is as follows: The distribution of emotional polarity is presented in the form of histograms or box plots, showing the polarity score distribution of the obligation clause group and the rights clause group, and marking the locations of clauses that deviate abnormally. The difference in emotional intensity is based on the difference between the mean emotional intensity of the obligation clause group and the rights clause group. The formula for calculating the emotional intensity difference coefficient is as follows: , When the difference exceeds the preset difference threshold of 0.25, the emotional risk score increases linearly; the emotional risk score adopts a logarithmic-exponential fusion formula: ,in, =1.5, when An emotional risk alert is triggered when the value is ≥0.65, and a paragraph with high-risk clauses is displayed.

8. The brand franchise qualification review and risk warning system according to claim 7, characterized in that, The cluster analysis of the implicit risks identified in all historical franchise agreements of the same brand under the target franchise agreement is performed as follows: Retrieve all historical franchise agreements from the same brand as the target franchise agreement from the system database, and extract the implicit risk feature vector for each agreement. , , , and ,in, For the contract sample index number, , The total number of contract samples. Encoding for asymmetric commitment structure types, The entropy value represents the location distribution of high-risk clauses. Number of risk clauses; A density-based clustering algorithm is used to perform unsupervised clustering of risk feature vectors, with cosine distance as the distance metric. ,in, For another contract sample, the index number and neighborhood radius are... Set the minimum number of samples to 15% of the feature space diameter. Introducing a clustering risk density coefficient: ,in, For clustering, The number of samples within the cluster. The cluster center is the mean vector of all samples within the cluster. The standard deviation of the intra-cluster distance; If the number of contracts within a cluster is ≥5 and If the value is ≥0.75, the contract template corresponding to that cluster is identified as a systemic fraud tool, and the cluster center is determined accordingly. As a standard risk profile.

9. The brand franchise qualification review and risk warning system according to claim 8, characterized in that, The process of sending risk warnings to all existing and potential franchisees in batches is as follows: The risk warning includes identification of contract templates deemed systemic fraud tools, excerpts and definitions of high-risk clauses, and risk deviation scores. With Emotional Risk Score And risk avoidance advice for franchisees; The push channels include message center push notifications, SMS notifications and emails for existing franchisees, and risk warning pop-ups that are forcibly displayed through the system interface when potential franchisees inquire about brand information; All push notifications are generated into immutable logs and stored in an audit database.

10. A method for performing brand franchise qualification review and risk warning system according to any one of claims 1-9, characterized in that, Includes the following steps: Step 1: Multi-granularity semantic segmentation and dependency parsing: Perform multi-granularity semantic segmentation on the target company's franchise agreement text at the document level, paragraph level, clause level, and lexical level to identify variant clauses with the same semantic substance but different expressions. Step 2, Implicit Obligation and Right Shift Detection: Semantically match and align each obligation clause in the target franchise agreement with its corresponding right clause. Construct a semantic link from the obligation clause to the right clause through dependency parsing. If an obligation clause is found to have no corresponding right clause as a semantic endpoint, or the right clause is placed at a position that exceeds the preset standard distance threshold from the obligation clause, or the obligation clause uses absolute expression while the corresponding right clause uses fuzzy modification, then the clause is determined to be an implicit risk clause and a risk shift score is output. Step 3: Sentiment and Fraudulent Intent Analysis: Through a pre-trained franchise contract sentiment model, an in-depth analysis of the sentiment of the target franchise contract terms is conducted to identify implicit suggestive word clusters, liability inference patterns, and asymmetric commitment structures in the target franchise contract text, and output a sentiment risk score that includes sentiment polarity distribution and sentiment intensity differences. Step 4: Cross-contract pattern clustering and early warning: Cluster analysis is performed on the hidden risk mining results of all historical franchise contracts of the same brand owner for the target franchise contract. If the same contract template is detected to repeatedly induce the same type of dispute between different franchisees, the contract template is identified as a systemic fraud tool, and risk warnings are pushed to all existing and potential franchisees in batches.

Images