Judicial intervention evaluation system and method based on multi-dimensional index fusion

Abstract

The invention discloses a judicial intervention evaluation system and method based on multi-dimensional index fusion, and particularly relates to the technical field of judicial aid decision making, in particular to a judicial intervention evaluation system and method based on multi-dimensional index fusion. The system mainly comprises a data acquisition module, a feature extraction module, a score calculation module and a result output module. The method comprises the following steps: automatically analyzing judicial documents through a pre-trained large language model, and extracting feature indexes of three dimensions of stockholder compression behaviors, financial anomalies and internal relief feasibility; dynamically determining an index weight by adopting an entropy weight method, and generating a judicial intervention necessity index in combination with a quantization rule; and performing intelligent prediction by using the trained neural network model, and finally outputting a 4-level judicial intervention suggestion. According to the invention, automatic and standardized evaluation of company surplus allocation dispute cases is realized, and judicial referees are effectively assisted.

Description

Technical Field

[0001] This invention relates to the field of judicial auxiliary decision-making technology, and more specifically, to a judicial intervention evaluation system and method based on the fusion of multi-dimensional indicators. Background Technology

[0002] Judicial intervention in disputes over the distribution of company profits is a core challenge in corporate law practice. With the development of my country's market economy and the increasing complexity of corporate governance structures, the number of such cases is on the rise, and judicial practice is gradually shifting from absolute respect for corporate autonomy to conditional intervention. In recent years, interdisciplinary research in artificial intelligence and law has become an important trend. The construction of smart courts is driving judicial decision-making to evolve from traditional experience-based judgment to data-driven analysis, and utilizing technological means to assist judges in evidence analysis, requirement review, and outcome prediction has become a clear development direction.

[0003] However, existing technological solutions have shortcomings. First, current judicial practice relies heavily on judges' subjective judgment, lacking unified and quantifiable review standards. This leads to significant ambiguity in identifying key elements such as "abuse of shareholder rights" and "financial irregularities," resulting in inconsistent judgments and inconsistent rulings in similar cases. Second, existing auxiliary tools are mostly limited to document information management or simple rule matching, lacking the ability to deeply integrate and analyze multi-source heterogeneous data (such as financial data, governance behavior, and procedural facts), and thus failing to systematically assess the necessity and appropriateness of judicial intervention. Finally, minority shareholders face a structural disadvantage in terms of evidence, and existing technologies have failed to effectively construct a tiered burden-of-proof transfer mechanism, resulting in high barriers to rights protection.

[0004] Therefore, this paper proposes a judicial intervention evaluation system and method based on the integration of multi-dimensional indicators to address the above-mentioned problems. The aim is to solve the core issues mentioned above: how to construct a standardized and quantifiable evaluation system to overcome the three major shortcomings of judicial intervention in corporate profit distribution—namely, the ambiguity in the determination of substantive requirements, the insufficient operability of procedural rules, and the imbalance in the allocation of the burden of proof. This will provide judges with objective and consistent decision-making support and effectively reduce the rights protection costs for minority shareholders. Summary of the Invention

[0005] In order to overcome the above-mentioned defects of the prior art, the embodiments of the present invention provide a judicial intervention assessment system and method based on multi-dimensional indicator fusion to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a judicial intervention assessment system based on multi-dimensional indicator fusion, comprising:

[0007] The data acquisition module is used to acquire textual data of legal documents and structured financial data related to disputes over the distribution of company profits.

[0008] The feature extraction module parses the judicial document text data based on a pre-trained large language model and extracts feature indicators from three primary dimensions: shareholder oppression behavior, financial anomalies, and feasibility of internal relief.

[0009] The scoring calculation module dynamically scores each secondary indicator from 0 to 5 points according to preset quantitative rules, and determines the weight of each dimension based on the entropy weight method to calculate the necessity index of judicial intervention.

[0010] The results output module is used to output recommendations on the level of judicial intervention, which include four types: compulsory intervention, limited intervention, procedural review, and priority of corporate autonomy.

[0011] Optionally, the shareholder oppression behavior dimension includes three secondary indicators: disguised profit distribution, capital solidification, and concealment or transfer of assets. The disguised profit distribution indicator is scored based on the multiple relationship between executive compensation and the industry average, and the capital solidification indicator is scored based on the deviation between the proportion of retained profits to net assets and the industry benchmark.

[0012] The indicator of concealing or transferring assets is scored based on the degree of deviation between the related transaction price and the market price.

[0013] Optionally, the financial anomaly dimension includes two secondary indicators: the existence of distributable profits and the concealment or transfer of profits. The existence of distributable profits indicator is scored based on the validity of the audit report and the results of cross-validation, while the concealment or transfer of profits indicator is scored based on the proportion of inflated costs or the proportion of off-book funds.

[0014] Optionally, the internal remedy feasibility dimension includes two secondary indicators: obstruction of the right to know and control decision-making mechanism. The obstruction of the right to know indicator is scored based on the number of times shareholders’ access requests are rejected and the duration of litigation. The control decision-making mechanism indicator is scored based on the pass rate of shareholder meeting proposals or the concentration of voting rights of the actual controller.

[0015] Optionally, the scoring calculation module is further configured to: dynamically adjust the scoring threshold according to industry type and enterprise size, then support users to input proof of reasonable business purpose to trigger the exemption mechanism, and finally update the market price benchmark and industry average based on the moving average method or industry annual report data.

[0016] Optionally, the system further includes a model training module, which uses a feedforward neural network to train on the structured features of historical cases to predict the level of judicial intervention. The neural network includes an input layer, at least one hidden layer, and an output layer. The hidden layer uses the ReLU activation function and introduces a Dropout mechanism. The output layer corresponds to the probability distribution of four types of intervention.

[0017] Optionally, the model training module uses a cross-validation strategy to evaluate model performance and performs feature importance analysis based on Shapley values ​​to enhance model interpretability.

[0018] Optionally, the system also provides a visual interface for inputting three primary dimension ratings and displaying the intervention recommendation level and corresponding explanations in real time.

[0019] A judicial intervention assessment method based on multi-dimensional indicator fusion includes the following steps:

[0020] S1. By analyzing judicial documents using a large language model, feature indicators are extracted from three dimensions: shareholder oppression behavior, financial anomalies, and the feasibility of internal relief.

[0021] S2. Each secondary indicator is dynamically scored from 0 to 5 points according to preset rules;

[0022] S3. Determine the weights of each dimension based on the entropy weight method and calculate the necessity index of judicial intervention;

[0023] S4. Based on this index, four levels of judicial intervention recommendations are generated.

[0024] Optionally, the method further includes: dynamically adjusting the scoring threshold according to industry type and enterprise size, training a neural network model based on historical cases to predict the intervention level of new cases, and receiving user input through a visual interface to display the prediction results and feature importance analysis.

[0025] The technical effects and advantages of this invention are as follows:

[0026] Compared to existing technologies, this invention constructs a quantifiable evaluation system that integrates three primary dimensions (shareholder oppression, financial anomalies, and feasibility of internal remedies) and seven secondary indicators, transforming abstract legal principles into concrete, calculable scoring indicators. The system utilizes structured data extracted from court documents and automatically generates a score of 0-5 for each indicator based on preset quantitative rules (such as executive compensation exceeding the industry average multiple, retained profit ratio deviation, and related-party transaction price deviation rate). This innovation realizes a paradigm shift in the elements of judicial intervention from subjective qualitative judgment to objective quantitative analysis. Its effect is to reduce inconsistencies in judgment standards caused by differences in judges' personal experience. Its advantage lies in providing unified and clear review standards and operational guidelines for judicial intervention, enhancing the consistency and predictability of judgments.

[0027] Compared to existing technologies, this invention replaces the traditional method relying on subjective expert assignment by introducing an objective weight allocation mechanism based on entropy weighting. This algorithm automatically determines the weight percentage of each indicator in the final decision by calculating the information entropy of various indicator data in historical case samples. This innovation allows the model weights to directly reflect the distinguishability and importance of each indicator in practice, ensuring that the core logic of the evaluation model originates from judicial practice itself, rather than being pre-set by humans. The advantage lies in improving the scientific rigor and objectivity of the model weights, avoiding subjective bias, and making the final Judicial Intervention Necessity Index (PJI Index) more accurately reflect the true situation of the case.

[0028] Compared to existing technologies, this invention enhances the adaptability and fairness of the model by integrating dynamic threshold adjustment and exemption mechanisms. The system does not mechanically apply fixed standards, but dynamically adjusts the scoring benchmark thresholds based on industry type (e.g., asset-heavy vs. asset-light industries) and company size (large, medium, and small). Simultaneously, it allows companies to submit reasonable proof of business purpose (e.g., investment plans) to apply for exemption from scoring. This innovation enables the quantitative standards to flexibly adapt to complex and ever-changing business practices, effectively preventing misjudgments of special cases and reducing the substantial injustice caused by a "one-size-fits-all" approach. Its advantage lies in pursuing consistency in adjudication while also considering justice in individual cases, making the model more practical and widely accepted.

[0029] Compared to existing technologies, this invention combines the deep semantic parsing capabilities of a large language model with the efficient decision-making capabilities of a lightweight neural network to construct an efficient and interpretable intelligent decision-making pipeline. The system first uses a large language model to accurately extract structured features from complex judicial documents. These features are then input into a locally deployed neural network for high-speed reasoning, ultimately outputting intervention level recommendations supplemented by interpretability analyses such as SHAP values. This innovation ensures both a powerful understanding of complex textual information and achieves low-latency, high-concurrency real-time auxiliary decision-making, significantly improving the automation and efficiency of judicial document processing. Its advantage lies in not only providing conclusions but also revealing the decision-making basis through feature importance ranking, enhancing judges' trust in the system's output and conforming to the principle of judicial transparency. Attached Figure Description

[0030] Figure 1 This is a system framework diagram of the present invention. Detailed Implementation

[0031] The following will refer to the appendices in the embodiments of the present invention. Figure 1The technical solutions in the embodiments of the present invention are clearly and completely described herein. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0032] Example 1:

[0033] A judicial intervention assessment system based on the fusion of multi-dimensional indicators includes:

[0034] The data acquisition module is used to obtain the text data of judgment documents in cases of disputes over the distribution of company profits from the judicial document disclosure platform, and to extract structured financial data from the enterprise information disclosure system. The data acquisition module connects to external databases through application programming interfaces to achieve batch data capture and cleaning, and to establish a local case database.

[0035] The feature extraction module uses a pre-trained large language model based on the Transformer architecture to perform semantic parsing of judicial documents. It constructs legal element extraction templates through prompting engineering technology, automatically identifies and outputs feature index values ​​corresponding to three primary dimensions: shareholder oppression behavior, financial abnormalities, and feasibility of internal relief. This module adopts a zero-shot learning mechanism to adapt to the processing of unlabeled cases.

[0036] The scoring calculation module dynamically scores each secondary indicator from 0 to 5 points according to preset quantitative rules. The quantitative rules are established based on industry benchmark data and legal norms. The scoring calculation module uses the entropy weight method to calculate the weight coefficients based on the frequency and distinguishability of each indicator in historical cases, and generates the necessity index of judicial intervention through weighted fusion.

[0037] The results output module maps the numerical range of the judicial intervention necessity index to four levels of judicial intervention recommendations, including compulsory intervention, limited intervention, procedural review, and priority for corporate autonomy. This module provides a structured data output interface and a visualization display interface.

[0038] The shareholder oppression behavior dimension includes three secondary indicators: disguised profit distribution, capital solidification, and concealment or transfer of assets. The disguised profit distribution indicator is quantitatively assessed by calculating the ratio of the company's senior executives' compensation to the industry average compensation. The industry average compensation data comes from the industry-specific wage guidance data released by the National Bureau of Statistics. The system adopts a dynamic update mechanism to regularly synchronize the latest statistical results. The specific scoring rules are as follows: 5 points are assigned when the ratio exceeds 3 times, 4 points are assigned when it is 2 to 3 times, 3 points are assigned when it is 1.5 to 2 times, 2 points are assigned when it is 1 to 1.5 times, and 1 point is assigned when it is less than 1 time.

[0039] Where, PJI=∑(w j*S j );

[0040] PJI: Judicial Intervention Necessity Index, a continuous value based on comprehensive evaluation.

[0041] w j The weight coefficient of the j-th indicator satisfies ∑w j =1, and is objectively determined by the entropy weight method.

[0042] S j : The standardized score of the j-th indicator, ranging from 0 to 5.

[0043] This formula is the core algorithm of this invention. By weighted and fused multi-dimensional quantitative indicators, it transforms the discrete judicial requirements evaluation into a unified and comparable continuous value, providing a precise basis for subsequent classification recommendations. It achieves dimensionality reduction and fusion from multi-dimensional features to a single decision variable.

[0044] The capital consolidation indicator is evaluated by calculating the proportion of retained earnings to net assets and comparing it with the industry benchmark. The industry benchmark is determined by the industry median method. A deviation of more than 20 percentage points is assigned 5 points, 10 to 20 percentage points is assigned 4 points, 5 to 10 percentage points is assigned 3 points, and within 5 percentage points is assigned 2 points. At the same time, a time dimension factor is introduced, and the score is adjusted upward for companies that have not held a shareholders' meeting for three consecutive years.

[0045] The indicators for concealing or transferring assets are assessed by comparing the deviation between the prices of related-party transactions and the fair market price. The fair market price is calculated using the moving average method to average the price of similar transactions over 12 months. A price deviation of more than 30% is assigned 5 points, 20% to 30% is assigned 4 points, 10% to 20% is assigned 3 points, and less than 10% is assigned 1 point. The criteria for identifying related-party transactions are implemented in accordance with Accounting Standard No. 36 for Enterprises.

[0046] The financial anomaly dimension includes two secondary indicators: the existence of distributable profits and the concealment or transfer of profits. The existence of distributable profits indicator uses the audit report as the core basis for identification. The system constructs an audit report credibility assessment submodule, which identifies the type of audit opinion through natural language processing technology. Cases with unqualified opinions and consistent data are assigned 1 point, while cases with qualified opinions or disclaimers of opinion are cross-validated with bank statements and tax declaration data. If there are contradictions, 5 points are assigned.

[0047] The indicators for concealed or transferred profits are assessed by calculating the proportion of inflated costs or off-book funds. The proportion of inflated costs is based on the industry average cost structure. A deviation of more than 20% is assigned 5 points, 10% to 20% is assigned 4 points, 5% to 10% is assigned 3 points, and less than 5% is assigned 2 points. The proportion of off-book funds is calculated by comparing financial statements with bank account cash flow data. A proportion of more than 15% is assigned 5 points, 10% to 15% is assigned 4 points, 5% to 10% is assigned 3 points, and less than 5% is assigned 1 point.

[0048] The feasibility dimension of internal remedies includes two secondary indicators: obstruction of the right to know and control decision-making mechanism. The obstruction of the right to know indicator is assessed by statistically analyzing the number of times shareholders’ access requests are rejected and the duration of related litigation. More than 2 rejections and litigation periods exceeding 18 months are assigned 5 points, 1 rejection and litigation period of 12 to 18 months are assigned 4 points, 1 rejection and litigation period of 6 to 12 months are assigned 3 points, and no rejection record is assigned 1 point. The criteria for determining access requests are based on Article 8 of the Judicial Interpretation IV of the Company Law regarding improper purposes.

[0049] The control decision-making mechanism indicators are evaluated by calculating the approval rate of shareholder meeting proposals and the concentration of voting rights of the actual controller. A proposal approval rate of over 90% is assigned 5 points, 60% to 90% is assigned 4 points, and less than 60% is assigned 2 points. The concentration of voting rights is calculated using the Herfindahl-Hirschman Index, and the score is increased by one level when the index exceeds 0.5.

[0050] The scoring calculation module is also configured to: dynamically adjust the scoring threshold based on industry type and enterprise size, with different capital fixation threshold ranges set for asset-heavy and asset-light industries; and classify enterprise size into three levels—large, medium, and small—based on registered capital and number of employees, each with different proportional fluctuation ranges. It supports users inputting proof of reasonable business purpose to trigger an exemption mechanism, with supporting materials including investment project plans, board resolutions, or third-party evaluation reports. The system extracts key information and verifies reasonableness through document parsing technology. It updates market price benchmarks and industry averages based on moving averages or industry annual report data, establishing a dynamic industry database and automatically synchronizing the latest published industry statistics quarterly for threshold recalibration.

[0051] The system also includes a model training module, which uses a feedforward neural network to train on the structured features of historical cases to predict the level of judicial intervention. The neural network includes an input layer that receives scores from three first-level dimensions, and a hidden layer consisting of two fully connected layers. The first layer maps the 3-dimensional input to a 64-dimensional feature space, and the second layer compresses it to a 32-dimensional feature space. The hidden layers use the ReLU activation function to introduce non-linear transformation capability, and a Dropout mechanism with a ratio of 0.2 is introduced after each layer to prevent overfitting. The output layer maps the 32-dimensional features to a 4-dimensional output space through a linear transformation, corresponding to the probability distribution of the four intervention types. Finally, the probability values ​​are normalized using the Softmax function. The training process uses the cross-entropy loss function and the Adam optimizer, and a 5-fold cross-validation strategy is used to evaluate the model's generalization performance.

[0052] The model training module uses a cross-validation strategy to evaluate model performance. The dataset is randomly divided into 5 mutually exclusive subsets. Four subsets are used as the training set and one subset is used as the validation set in turn. The training process is repeated 5 times and the average accuracy is calculated. Feature importance analysis is performed based on Shapley values. Feature weight ranking is generated by calculating the marginal contribution of each feature to the model output. The influence of each indicator on the prediction results is displayed in a visual form to enhance the interpretability of the model.

[0053] The system also provides a visual interface for inputting three primary dimension scores and displaying the intervention recommendation level and corresponding explanations in real time. The interface accepts user input in the form of a web form, with the input range limited to integer values ​​between 0 and 15. The backend service standardizes the input data and then inputs it into a trained neural network model. After obtaining the prediction results, it matches the corresponding legal basis and judgment principles from a preset interpretation dictionary. Finally, it returns the data to the front end in structured JSON format for display. The displayed content includes intervention level labels, confidence percentage, key indicator impact analysis, and a summary of recommended measures.

[0054] A judicial intervention assessment method based on multi-dimensional indicator fusion includes the following steps: parsing judicial documents through a large language model, constructing a legal element extraction template using prompting engineering technology, and automatically identifying and extracting feature indicators from three dimensions: shareholder oppression behavior, financial anomalies, and the feasibility of internal relief.

[0055] Each secondary indicator is dynamically scored from 0 to 5 points according to preset rules. The scoring rules are established based on industry benchmark data and legal requirements. The entropy weight method is used to calculate the weight coefficient based on the frequency and distinguishability of each indicator in historical cases. The necessity index of judicial intervention is calculated based on a weighted fusion method. The index value range is mapped to four levels of judicial intervention recommendations. The user input of three-dimensional score values ​​is received through a visual interface. After standardization, the scores are input into a neural network model for prediction, and the intervention level recommendations and explanations are output.

[0056] Among them, E j =-(1 / ln(m))*∑(p ij *ln(p ij )), w j =(1-E j ) / ∑(1-E j );

[0057] E j : The information entropy value of the j-th indicator.

[0058] m: The total number of sample cases involved in the calculation.

[0059] p ij The percentage of the score of the i-th sample on the j-th indicator relative to the total score of that indicator.

[0060] w j : The final weight of the j-th indicator.

[0061] ln: Natural logarithm function.

[0062] This algorithm objectively determines the importance of each indicator in decision-making by calculating the information entropy of the indicator data. Entropy value E j The smaller the value, the greater the variability of the indicator data, and the more information it provides. The weight w... j The higher the value, the better. This avoids subjective assignment bias and ensures the scientific and objective nature of weight allocation.

[0063] The method also includes: dynamically adjusting the scoring threshold according to industry type and enterprise size, establishing an industry dynamic database to regularly update benchmark values; training a neural network model based on historical cases, using a feedforward network structure with Dropout mechanism to prevent overfitting, and using cross-entropy loss function and Adam optimizer for parameter optimization; calculating feature importance through Shapley value analysis, generating a visual report to show the degree of influence of each indicator on the prediction results; and providing an application programming interface to support third-party system integration, with the interface adopting a RESTful architecture design and supporting JSON format data exchange.

[0064] Among them, T adj =Tbase *(1+α*F size );

[0065] T adj : Adjusted dynamic threshold (such as the capital solidification ratio threshold).

[0066] T base : Basic industry benchmark thresholds.

[0067] α: Adjustment coefficient, a constant set based on experience, used to control the adjustment range.

[0068] F size Enterprise size factor, which is divided into three categories: large (0.1), medium (0), and small (-0.1) based on registered capital and number of employees, and assigned a value accordingly.

[0069] This rule allows the model's scoring benchmark to be flexible enough to adapt to companies of different sizes, preventing systematic bias towards companies of a specific size caused by a "one-size-fits-all" approach, and enhancing the model's applicability and fairness.

[0070] The workflow of this invention is as follows: First, the system uses a data acquisition module to batch capture the text data of judgment documents in cases of disputes over the distribution of company profits from the judicial document disclosure platform, and simultaneously extracts structured financial data from the enterprise information disclosure system, which is then cleaned and stored in the local case database.

[0071] Subsequently, the feature extraction module performs deep semantic analysis of the judgment documents based on a pre-trained large language model. It automatically identifies and extracts the feature values ​​of 7 secondary indicators under 3 primary dimensions: shareholder oppression behavior, financial abnormalities, and feasibility of internal relief through a pre-designed prompt engineering template.

[0072] Next, the scoring calculation module dynamically scores each secondary indicator from 0 to 5 points according to the preset quantitative rules. The scoring threshold is calibrated based on the benchmark value updated regularly by the industry dynamic database and supports automatic floating adjustment according to the size of the enterprise and the industry type.

[0073] Meanwhile, the system uses the entropy weight method to calculate the weight coefficients of each dimension, and generates a judicial intervention necessity index through weighted fusion. This index is input into a feedforward neural network model trained with historical cases. The model contains two hidden layers and uses the Dropout mechanism to prevent overfitting. The output generates a probability distribution of four levels: mandatory intervention, limited intervention, procedural review, and priority of corporate autonomy through the Softmax function.

[0074] Finally, the results output module presents the prediction results and feature importance analysis results through a visualization interface. Users can input three-dimensional score values ​​to obtain intervention suggestions and legal basis explanations in real time. The system also supports data interaction with external judicial systems through application programming interfaces.

[0075] Finally, the following points should be noted: First, in the description of this application, it should be noted that, unless otherwise specified and limited, the terms "installation", "connection", and "linkage" should be interpreted broadly, and can be mechanical or electrical connections, or internal connections between two components, or direct connections. "Up", "down", "left", "right", etc. are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may change.

[0076] Secondly: The accompanying drawings of the embodiments disclosed in this invention only involve the structures involved in the embodiments disclosed in this invention. Other structures can refer to the general design. In the absence of conflict, the same embodiment and different embodiments of this invention can be combined with each other.

[0077] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A judicial intervention evaluation system based on multi-dimensional indicator fusion, characterized in that, include: The data acquisition module is used to acquire textual data of legal documents and structured financial data related to disputes over the distribution of company profits; The feature extraction module parses the judicial document text data based on a pre-trained large language model and extracts feature indicators from three primary dimensions: shareholder oppression behavior, financial anomalies, and feasibility of internal relief. The scoring calculation module dynamically scores each secondary indicator from 0 to 5 points according to preset quantitative rules, and determines the weight of each dimension based on the entropy weight method to calculate the necessity index of judicial intervention. The results output module is used to output recommendations on the level of judicial intervention, which include four types: compulsory intervention, limited intervention, procedural review, and priority of corporate autonomy.

2. The judicial intervention evaluation system based on multi-dimensional indicator fusion as described in claim 1, characterized in that, The shareholder oppression behavior dimension includes three secondary indicators: disguised profit distribution, capital solidification, and concealment or transfer of assets. The disguised profit distribution indicator is scored based on the multiple relationship between executive compensation and the industry average, and the capital solidification indicator is scored based on the deviation of the proportion of retained profits to net assets from the industry benchmark. The indicator of concealing or transferring assets is scored based on the degree of deviation between the related transaction price and the market price.

3. The judicial intervention evaluation system based on multi-dimensional indicator fusion as described in claim 1, characterized in that, The financial anomaly dimension includes two secondary indicators: the existence of distributable profits and the concealment or transfer of profits. The existence of distributable profits indicator is scored based on the validity of the audit report and the results of cross-validation, while the concealment or transfer of profits indicator is scored based on the proportion of inflated costs or the proportion of off-book funds.

4. The judicial intervention evaluation system based on multi-dimensional indicator fusion as described in claim 1, characterized in that, The internal remedy feasibility dimension includes two secondary indicators: obstruction of the right to know and control decision-making mechanism. The obstruction of the right to know indicator is scored based on the number of times shareholders’ access requests are rejected and the time spent in litigation. The control decision-making mechanism indicator is scored based on the pass rate of shareholder meeting proposals or the concentration of voting rights of the actual controller.

5. A judicial intervention evaluation system based on multi-dimensional indicator fusion as described in claim 1, characterized in that, The scoring calculation module is also configured to: dynamically adjust the scoring threshold according to industry type and enterprise size, then support users to input proof of reasonable business purpose to trigger the exemption mechanism, and finally update the market price benchmark and industry average based on the moving average method or industry annual report data.

6. The judicial intervention evaluation system based on multi-dimensional indicator fusion as described in claim 1, characterized in that, The system also includes a model training module, which uses a feedforward neural network to train on the structured features of historical cases to predict the level of judicial intervention. The neural network includes an input layer, at least one hidden layer and an output layer. The hidden layer uses the ReLU activation function and introduces the Dropout mechanism. The output layer corresponds to the probability distribution of four types of intervention.

7. A judicial intervention assessment system based on multi-dimensional indicator fusion as described in claim 6, characterized in that, The model training module uses cross-validation to evaluate model performance and performs feature importance analysis based on Shapley values ​​to enhance model interpretability.

8. A judicial intervention assessment system based on multi-dimensional indicator fusion as described in claim 1, characterized in that, The system also provides a visual interface for inputting three primary dimension ratings and displaying the intervention recommendation level and corresponding explanations in real time.

9. A judicial intervention assessment method based on multi-dimensional indicator fusion, characterized in that, Includes the following steps: S1. By analyzing judicial documents using a large language model, feature indicators are extracted from three dimensions: shareholder oppression behavior, financial anomalies, and the feasibility of internal relief. S2. Each secondary indicator is dynamically scored from 0 to 5 points according to preset rules; S3. Determine the weights of each dimension based on the entropy weight method and calculate the necessity index of judicial intervention; S4. Based on this index, four levels of judicial intervention recommendations are generated.

10. The judicial intervention assessment method based on multi-dimensional indicator fusion as described in claim 9, characterized in that, The method also includes: dynamically adjusting the scoring threshold according to industry type and enterprise size, training a neural network model based on historical cases to predict the intervention level of new cases, and receiving user input through a visual interface to display the prediction results and feature importance analysis.

Images