Big data identification and calculation model construction of civil abuse of right of action behavior based on the constraints of laws and regulations

Litigation is a more legitimate means of conflict resolution than any other, and its effects are most significant and useful, and it is also the ultimate means of resolving conflicts. In litigation, civil litigation is the most commonly used means of solving trivial disputes in people’s daily life, and it can be said that civil litigation is the last safeguard for citizens to protect their rights, and it is the most effective way for the national public power organs to legally help citizens deal with disputes [1-3]. In recent years, due to the democratic idea of the rule of law continues to promote the deep and thorough, the government continues to expand the strength of openness, citizens’ awareness of their rights, legal awareness continues to strengthen, people are no longer satisfied with the “self-sufficient” way of relief, and turn to the state power organs to reach out to the hand for help, in order to protect their own legitimate rights and interests from being infringed upon, which makes litigation this This makes litigation as a public remedy to shine [4-6]. However, things are two-sided, especially in the dialectical mode of thinking, civil litigation can not be spared. Great shine at the same time also appeared some problems, the most prominent of which is the abuse of the right to sue, not only violates the civil litigation fairness and efficiency of the purpose, but also disturbed the original judicial order, and led to a large loss of judicial resources, resulting in an unnecessary consumption of judicial resources [7-9].

After years of reform, China has formed a “court in the middle of the guide, the active participation of the parties” mode of civil litigation, fully respect the parties’ right of disposal, the parties enjoy full and equal rights to participate in litigation, the civil litigation mode of partyism conforms to the trend of development of the times, and has significant and positive effects on the promotion of the construction of the social and democratic system of law and the promotion of judicial justice. Judicial justice has significant and positive significance [10-12]. However, due to the current civil litigation system is too biased to emphasize the protection of rights, the regulation of rights is far from enough, to the exercise of the right to leave a lot of white space, at the same time, effective and open social integrity system has not yet been fully established, litigation value orientation to a certain extent out of balance, the side effect of the inevitable litigation is to trigger a variety of rights of the over-inflation of the litigation, breeding part of the main body of the litigation for the pursuit of their own interests maximization, with the help of litigation Maximize the interests of some litigation subjects in pursuit of their own interests, with the help of litigation beyond the legitimate scope of the right to seek improper rights and interests or infringe on the rights and interests of others of this alienation phenomenon, that is, the abuse of the right to litigation [13-15]. Abuse of the right to litigation has become more and more intense trend, its impact should not be underestimated, first of all, contrary to the civil procedural law, the purpose of the legislation of the litigation process, undermining the fairness of the value of stability and efficiency, and secondly, to the others on the material, spiritual damage caused by the great, and once again resulting in a waste of the limited judicial resources, interfering with the court of law the normal work of the trial, and seriously undermined the judicial authority and fairness [16-18].

Abuse of the right of action has become China’s civil trial judicial practice can not be ignored. Abuse of the right of action by the parties to bring an unlawful lawsuit, the court shall exercise the discretionary power of lexis to be recognized, sanction [19]. However, China has no legal provisions directly defining the abuse of the right of action, the abuse of the punishment and the relief of the victim is also the lack of detailed provisions, so that the judicial practice on the abuse of the phenomenon of the right of action has been a very limited response to the abuse of the right of action, the abuse of the right of the person often do not get the due legal responsibility, low cost of violation of the law objectively undoubtedly further promote the abuse of rights of the wanton [20-22]. It is necessary and urgent to clarify the definition and nature of the abuse of the right of appeal, and to set up a system to prevent and regulate the abuse of the right of appeal in accordance with the actual situation in China.

Civil abuse of rights of action line in the objective aspects of the existence of a variety of forms, before the beginning of its behavioral identification research, this paper first from the civil right of action has the connotation of the civil abuse of rights of action from the behavior of in-depth analysis. Aiming at the lack of civil abuse of rights of action behavior data, uneven problem of behavioral data processing, after verification using SMOTE oversampling method is more suitable. Then, the recognition model based on LightGBM is constructed, and the performance of the model is examined through model training and comparative tests, which provides a practical and feasible method for the identification of civil abusive litigation behavior big data.

2

Civil abuse of rights of action based on legal constraints

A legal definition of abusive litigation begins with a discussion of the concept of the right to litigation, which also delimits the scope of abusive litigation. In civil procedure law, the right of action refers to the right of the subject of a legal relationship to bring a lawsuit to the court and request it to protect his or her rights and interests. From the perspective of the attributes of the legal relationship related to litigation, the right of action can be divided into the right of action in the broad sense and the right of action in the narrow sense. The broad right of action includes civil right of action, administrative right of action and criminal right of action. The right of action in the narrow sense refers only to the civil right of action, that is, the subject of civil legal relations to request the protection of civil rights and interests of the right of action. The right of action studied in this paper and the related abusive behavior are based on the narrow concept of the right of action.

From the point of view of the connotation of the civil right of action, it mainly includes the following contents.

First, the right of action is characterized by both freedom and normality. In order to protect their own rights and interests, the main body of civil legal relations, by filing a lawsuit to initiate civil litigation procedures. And this right is guaranteed by the Constitution and civil procedure law. When a dispute arises between civil subjects or their rights are unlawfully infringed upon by others, the parties concerned may institute legal proceedings to protect their rights in accordance with the provisions of the law. No organization or individual may unlawfully restrict a party’s right to bring a lawsuit in accordance with the law, and it is up to the free will of the parties to decide whether or not to bring a lawsuit to court and what kind of lawsuit request to make. Of course, the exercise of this right must be limited to the extent permitted by law.

Secondly, the right of action has the attributes of public law. The parties choose to the trial organs, this is a request for national public power to protect their own rights, which on behalf of the state to exercise judicial power between the court has produced a public law level of legal relations. Trial organs shall not refuse to trial, should for the parties to request the rights of relief to provide a smooth channel, at the same time, the trial organs also through the exercise of judicial power, confirm the attribution of civil rights and the corresponding legal responsibility.

Third, the right of action does not belong to one party alone. All parties to a civil legal relationship have the right to sue, can be sued to the trial organ to request the right to relief. The plaintiff who initiates a lawsuit enjoys the right of action, while the defendant may still enjoy the right to reply and file counterclaims in the proceedings.

Fourth, the right of action has both substantive and procedural attributes. Most scholars of procedural law believe that, from the content of the lawsuit, at the same time embodies the nature of these two aspects. The civil subject to file a lawsuit, so that the dispute between themselves and others into the judicial process to solve, which reflects the attributes of the program. And civil litigation in the parties must put forward specific claims and factual reasons, which reflects the entity’s attributes.

Fifth, the right of action is closely related to the civil rights enjoyed by the parties, but the existence of the right of action is not based on the real existence of civil rights as a prerequisite. When the parties to a civil lawsuit sue the court, they often think that they enjoy the relevant civil rights, and their suits also have specific claims and factual reasons. However, whether the party’s claim can be recognized by the trial authority and whether the factual grounds for its claim can be adopted will not constitute an obstacle to the party’s right of action. Therefore, whether the civil rights of the parties can be recognized and protected by the court is not a condition for the exercise of the right of action. When a party sues, as long as there is a clear claim, factual grounds and the subject is qualified.

Abusive behavior, that is, the abuse of the right of action, the concept of which also exists in the legal profession different views. Some researchers advocate that the concept of abusive litigation should be defined in terms of the traditional procedural law theory of the legal system. In the country’s civil judicial practice, the principle of legal substantive characterization is applied.

In litigation, the court enjoys the right of free adjudication according to the norms of civil procedure law, and this right is also subject to legal constraints. The path of legal and regulatory constraints on civil abuse of the right of action can be embodied in three aspects: clear supervision of the main body, mandatory disclosure of information and strengthening of legal responsibility. The litigation subject enjoys the right of action in the civil procedure law protection, the law also provides its rights whether the establishment of the standard and judgment basis. If the exercise of the right is not legally justified, it cannot be established. However, different researchers in-depth analysis of abusive litigation behavior there are also differences. Some people advocate the view of “subjective theory”, that is to say, to determine whether it constitutes abusive litigation, should mainly analyze the subjective aspects of the litigation subject to initiate litigation, the parties in the prosecution of the subjective malice is a necessary condition for the determination of abusive litigation. From this point of view, it is believed that the civil procedure law should provide for the principle of good faith, in order to regulate the parties to exercise the right of action and effectively prevent abusive behavior. Others, on the other hand, advocate the criterion of “objectivity”, i.e., the subjective aspect of the subject of the litigation should not be used as a basis for judgment, but should be examined from the objective aspect of the exercise of the rights of the parties to comply with the requirements stipulated in the law. Failure to do so constitutes abusive conduct.

3

Big Data Identification Model for Abusive Litigation Behavior

3.1

Behavioral data processing

The problem of unbalanced datasets is extremely common in the fields of machine learning and data mining that deal with problems such as classification and clustering. An unbalanced dataset refers to a situation where there is a significant imbalance in the number of samples from different categories in a classification problem. That is, some categories have significantly more samples than others, resulting in an uneven distribution of categories in the dataset. This situation leads to a situation where it is easier for machine learning models to learn the dominant categories and perform poorly on fewer categories. Resampling techniques are one of the effective ways to cope with the problem of unbalanced datasets. Resampling is able to adjust the number of samples from different categories in the dataset to balance the category distribution and solve the problem of poor learning of machine learning models. According to the methods for dealing with imbalanced category distribution, resampling techniques can be divided into three main categories: undersampling, oversampling and mixed sampling.

The main methods of undersampling include reducing the majority class samples or increasing the minority class samples to balance the sample distribution of different classes. Its common methods include random undersampling, undersampling with put-back and prototype selection method.

The core idea of oversampling is to generate new training data by transforming the original data in order to increase the number of sample minority classes, so as to realize the balance of sample distribution. Its common method’s include SMOTE and random oversampling.

Mixed sampling methods are a combination of different sampling strategies designed to effectively deal with unbalanced datasets. Common hybrid sampling methods include random hybrid sampling, resampling hybrid sampling and data-enhanced hybrid sampling.

Information gain is an important feature selection method based on the concept of information entropy, aiming at identifying key features and eliminating irrelevant features so as to improve model performance. The computation of information gain is based on information entropy, which is a measure of the information uncertainty of a data set and reflects the purity of the data set. Under the concept of information entropy, when the number of categories in the dataset is high, the information entropy is higher, indicating a higher uncertainty in the dataset; conversely, when the number of categories in the dataset is low, the information entropy is lower, indicating a lower uncertainty in the dataset. This is because the calculation of information entropy takes into account the cloth of each category in the dataset, thus quantifying the degree of chaos in the dataset. The core idea of information gain is to quantify the change in information entropy before and after the features partition the dataset. High information gain implies that features have a greater influence in dataset division because it can significantly reduce the information entropy after division and improve the purity of the dataset. Therefore, information gain is widely used in feature selection, which helps to select features that have significant influence on the classification problem and improve the prediction performance of the model. The calculation process is as follows:

Setting variable $X = x_{1}, x_{2}, x_{3}, \dots, x_{n}$ , the information entropy of variable $X$ is: (1) $e (x) = - \sum_{i = 1}^{n} ρ_{i} \log_{2} ρ_{i}$

In the formula, $ρ_{i}$ indicates the probability that a sample belongs to category $x$ , the more values of $X$ , the more information it carries, the smaller the information entropy, indicating that the higher the purity of the data set, and vice versa, the lower.

Classification prediction, set the category $C = C_{1}, C_{2}, C_{3}, \dots, C_{n}$ , the formula for the information entropy of the categorized data set is: (2) $e (X) = - \sum_{i = 1}^{n} ρ (C_{i}) \log_{2} ρ (C_{i})$

If the classified sample has feature $T$ , set feature $T = T_{1}, T_{2}, T_{3}, \dots, T_{n}$ , at this point, the information entropy formula for feature $T$ is: (3) $\begin{matrix} (C | T) = ρ_{1} e (C | T) = T_{1} + ρ_{2} e (C | T = T_{2}) + \dots + ρ_{n} e (C | T = T_{n}) \\ = \sum_{i = 1}^{n} ρ_{i} e (C | T = T_{i}) \end{matrix}$

Where $ρ_{i}$ represents the probability of occurrence of each value, a weighted average can be calculated. At this point, the information gain from feature $T$ of the sample dataset is the difference between the information entropy of the sample and that of the sample when the sample has the feature: (4) $I G (T) = e (C) - e (C | T)$

3.2

Recognition model based on LightGBM algorithm

The particularly rapid development of information and communication technologies has led to rapid growth in data size and increasing data complexity, constantly challenging the processing and high-speed computing capabilities of existing data platform architectures, and the emergence of big data. Big data generally refers to a collection of data that cannot be sensed, captured, processed, and served in a manageable period of time using traditional equipment, software, hardware, etc. It is necessary to update the processing model in order to enable the data processing platform to have stronger insight, better decision-making capabilities and process management optimization capabilities, and to carry out reasonable management of massive, high-growth, and diversified information assets.

Machine learning methods are widely used in the field of recognition due to their high intelligence and high efficiency. However, traditional algorithms such as support vector machine, K-nearest neighbor and other algorithms have simple decision-making mechanism and insufficient learning ability on complex datasets. In order to improve the recognition accuracy of abusive litigation behavior, this paper proposes a recognition method based on integrated learning on the basis of data balancing processing and feature selection, which provides an accurate, efficient, and stable solution for the recognition of abusive litigation behavior.

LightGBM algorithm is an improved version of Gradient Boosting Decision Tree (GBDT) algorithm, which has a wide range of applications in the field of electric power data mining, however, it still suffers from the problems of easy overfitting under large sample conditions and slow training speed.

LightGBM is an integrated learning algorithm with decision tree as the base learner, and iteratively trains each decision tree in a gradient boosting manner until all the decision trees are trained, and the detailed principle of the algorithm is as follows.

Decision tree is a typical machine learning algorithm, a classification decision tree has only one root node, while there can be more than one child and leaf nodes. The leaf nodes of the decision tree map the final classification results and the child nodes correspond to the classification features. The root node midpoint contains all the training samples and the child nodes contain the samples after the features are divided. The key step in the decision tree learning process is to perform feature division, the specific division principle is as follows:

Let the percentage of category $k$ in training set $D$ be $p_{k} (k = 1, 2, ..., n)$ , and define the information entropy as (5) $E n d (D) = - \sum_{i = 1}^{n} p_{k} \log_{2} p_{k}$

Taking discrete features as an example, if feature $u$ has a total of $M$ possible values $\{u_{1}, u_{2}, ..., u_{M}\}$ , when the training set $D$ is divided over feature $u$ it produces $M$ branch nodes. Assume that all the samples in feature $u$ with value $u_{m}$ are contained in the $m$ th branch node: (6) $G a i n (D, u) = E n t (D) - \sum_{m - 1}^{M} \frac{D^{m}}{D} E n t (D_{m})$

The main rationale for the decision tree algorithm in the feature partitioning process is to maximize the information gain after partitioning.

Gradient Boosting Decision Tree is an integrated learning framework with decision trees as the base classifiers, GBDT utilizes the fast gradient descent method to approximate the solution of each decision tree, and in each iteration, the newly built decision tree is made to reduce the loss function along the direction where the loss function is reduced the fastest, a negative gradient direction to reduce the loss function to obtain higher prediction accuracy.

Assume the data set is ${\{y_{i}, x_{i}\}}_{1}^{N}$ , where $x = \{x_{1}, x_{2}, \dots, x_{n}\}$ is the input and $y = \{y_{1}, y_{2}, \dots, y_{n}\}$ is the output, and define the evaluation function $F (x)$ to represent the mapping relationship between input $x$ to output $y$ . Construct the loss function $L (y, F (x))$ , then the final learning objective of the algorithm is to find the function $F^{*} (x)$ that minimizes the expected value of the loss function $L (y, F (x))$ for all $\{y, x\}$ joint distributions of input one output mapping relations. $F^{*} (x)$ is defined as: (7) $F^{*} (x) = \arg \min_{F} E_{y, x} L (y, F (x)) = \arg \min_{F} E_{x} [E_{y} (L (y, F (x))) | x]$ where $E_{x}, E_{y, x}$ represents the mathematical expectation of $x, {y, x}$ , respectively.

The GBDT algorithm models the final strong classifier by linearly combining the $m$ weak classifiers to obtain the final strong classifier: (8) $F_{M} (x) = \sum_{m = 0}^{M} β_{m} h (x; a_{m})$ where $β_{m}$ represents the linear combination coefficients of the $m$ nd decision tree and $h (x; a_{m})$ represents the decision tree model.

In the gradient boosting process, GBDT uses the segmented greedy algorithm to solve the parameters of each decision tree step by step with the most rapid descent method until the last decision tree is solved. Let the initial value of $F (x)$ be $F_{0} (x)$ and the value of $F_{m} (x)$ during the $m$ th iteration: (9) $F_{m} (x) = F_{m - 1} (x) + β_{m} h (x; a_{m})$

Let $- g_{m} (x_{i})$ represent the value of the gradient in the negative direction for the $m$ nd iteration, $i = 1, 2, \dots, N$ , then there is: (10) $- g_{m} (x_{i}) = - {[\frac{\partial L (y_{i}, F (x_{i}))}{\partial F (x_{i})}]}_{F (x) = F_{n - 1} (x)}$ (11) $(β_{m}, a_{m}) = \arg \min_{β, a} \sum_{i = 1}^{N} L (y_{i}, F_{m - 1} (x_{i}) + β h (x_{i}; a))$

$a_{m}, β_{m}$ respectively according to Eq: (12) $a_{m} = \arg \min_{β, α} \sum_{i = 1}^{N} {[- g_{m} (x_{i}) - β h (x_{i}; a)]}^{2}$ (13) $β_{m} = \arg \min_{β} \sum_{i = 1}^{N} L (y_{i}, F_{m - 1} (x_{i}) + β h (x_{i}; a_{m}))$

The LightGBM algorithm is a further improvement on the GBDT algorithm, and the main improvement strategies are the incorporation of the histogram algorithm, the leaf-wise leaf-growth strategy with depth constraints, and the support for parallel computing.

The Lightgbm algorithm supports both feature parallelism and data parallelism, and optimizes these two parallel methods, solving the problem that only slicing data can be done in data parallel processing without being able to save all the data locally. In addition, the Lightgbm algorithm uses a decentralized statute to operate on the data in parallel, and can solve the task of merging histograms on different machines, so the Lightgbm algorithm not only reduces the amount of data processing, but also has a greater advantage in improving the speed of data processing, which is more applicable to big data analysis and mining.

Samples are unusually difficult to accurately classify due to spatial distribution problems, and such samples are called difficult samples, in order to further increase the classification accuracy of Lightgbm classifier for such samples, the Lightgbm algorithm is optimized by using the focal point loss function to improve the model’s learning ability for difficult samples in the dataset.

The original log-loss function of Lightgbm for the $m$ st decision tree before improvement: (14) $L_{o r i g i n a l} = - \sum_{j = 1}^{N} L (y_{i}, F_{m - 1} (x_{i}; A_{m - 1})) = - \sum_{j = 1}^{N} y_{i} \log (p_{j})$ where $N$ represents the number of categories, $p_{j}$ represents the probability that a sample $i$ is predicted to be a category $j$ , $F_{m} - 1 (x_{i}; A_{m - 1})$ is the predicted value of the model consisting of the current $m - 1$ decision tree for the input sample $x_{i}$ with parameter $A_{m} - 1$ , and $A_{m - 1}$ represents the set of parameters for the previous $m - 1$ trees, including $\{a_{1}, a_{2}, ..., a_{m - 1}\}, L (y_{i}, F_{m - 1} (x_{i}; A_{m - 1}))$ represents the error function between the true value $y_{i}$ and the predicted value.

Focus loss function: (15) $L_{f o c a l_{l} o s s} = - \sum_{j = 1}^{N} {(1 - p_{j})}^{γ} y_{i} \log (p_{j})$

$p_{j} \approx 1, (1 - p_{j}) \approx 0$ when a sample has a higher probability of being correctly categorized, at which point the weight of the sample with the higher probability of being correctly categorized is reduced and will receive less attention in the next iteration. Parameter $γ$ serves to regulate the rate at which the weights of easily categorized samples are reduced, and increasing $γ$ enhances the impact of the regulator. The focus loss function makes the classifier pay more attention to the contribution of misclassified samples during the training process, so that its probability of correct classification in the next iteration process can be improved, thus improving the classifier training accuracy.

The framework of the final recognition method is shown in Fig. 1, which is divided into three core steps, namely, the balanced processing of cyber-attack data proposed in Chapter 3, the selection of the optimal feature sub-selection for recognition, and the construction of the recognition model. It achieves the optimization work of the recognition model of civil abuse of rights of action, and SMOTE reduces the recognition false alarm rate. The data dimensionality reduction is achieved by using the Joint Maximum Mutual Information (JMIM) algorithm, which removes the redundant features and improves the recognition efficiency and accuracy. Based on the focal loss function optimization LightGBM classifier (FLGB) is trained in identifying the optimal feature subset to obtain the recognition model, which is combined with data balancing processing and feature subset selection to form the recognition method, and the final recognition model is named CKS-JMIM-FLGB.

4

Empirical research on big data identification model of abusive litigation behavior

The problem of abusive litigation behavior recognition is a typical binary classification problem, so in terms of evaluation criteria, four common metrics such as recall, accuracy, AUC, and F1 are mainly compared. For the binary classification problem, the performance measures mainly rely on the confusion matrix. ACC is the proportion of correctly categorized samples reflecting the accuracy rate.AUC is intuitively the area under the ROC curve, which is loosely interpreted as the likelihood that the probability value of a sample predicted to be positive is greater than the probability value predicted to be negative, and is a measure of the effectiveness of the model’s true problem classification. The F1Score is a compromise between check accuracy and recall that measures the classification status of unbalanced data. The data used in this paper is the S region, and the original data includes 83,600 civil lawsuit records in the time period of September to December 2020. The data set after data processing is divided into two parts: the training set and the test set, the training data is 70% of the randomly selected data and the test data is the remaining 30%.

4.1

Impact study of data processing

The structure of the data processing is shown in Table 1, and this section will explore the extent to which the imbalance of the data due to the right to sue behavior affects the model, comparing the results of the data before and after the balancing process is performed to analyze whether there is any change in the classification performance of its experimental model. The simulation set was processed using SMOTE oversampling as well as random undersampling techniques. The datasets of the three processing methods are substituted into the LightGBM model for training, and then the results of the three evaluation indexes of the three processing methods are compared to analyze the necessity of data balancing processing.

Table 1.

The structure of the process

Processing method	1	2	Tot
Unprocessing	0.685	0.315	1
SMOTE	0.510	0.490	1
Random owe sampling	0.510	0.490	1

The model evaluation comparison is shown in Figure 2, where the oversampling and undersampling processing is done using the imblearn package, and then the model is trained and evaluated by the sklearn package. The results show that the dataset has the highest values of the three indicators after SMOTE oversampling, with each indicator being 0.9715, 0.9981, and 0.9721, respectively, which demonstrates the better effect of this method in unbalanced data processing. After random undersampling processing data, the three index values are much lower than after SMOTE oversampling method processing, even lower than when it is not processed, so it shows that the undersampling method processing data, although it makes the data balanced, but in the process of processing to remove a lot of sample values in most of the categories, in this experiment removed nearly 75% of the sample size, if the data imbalance is higher. If the imbalance of the data is higher, then this method will remove more sample size, which will lead to the sample information contained in the obtained sample is more incomplete, so the indicators show the results are not very satisfactory, which also proves that the SMOTE oversampling method is more suitable.

4.2

Comparative analysis of recognition models

This experimental test was conducted to validate the effectiveness of the constructed model by generating a 223-dimensional feature matrix after preprocessing, feature engineering, and feature processing of the collected data. Positive and negative samples from the data were randomly sampled in the ratio of 7:3, respectively, to form the training and test sets.

In the algorithm comparison, the commonly used algorithms for machine learning are compared, including SVM, random forest, XGBoost, and deep forest (gcForest). The comparison of the feature engineering effect of different models is shown in Table 2, here mainly through the two types of comprehensive indicators AUC and ACC values are compared to further measure the recognition rate of the abuse of the right to sue behavior samples. From the table, it can be seen that without feature engineering, the AUC, ACC and Recall of each algorithm are relatively low, and after adding data processing, there are more key attributes of abusive litigation behavior embedded in the data, so it can be seen that the AUC, ACC, and Recall have been improved, and the accuracy rate is more than 90%, which further proves that the feature engineering processing in this paper has a great enhancement.

Table 2.

Characteristics of different models

Model	AUC		ACC		Recall
Model	Raw data	After processing	Raw data	After processing	Raw data	After processing
XGBoost	0.524	0.605	0.515	0.962	0.058	0.301
gcForest	0.531	0.638	0.538	0.963	0.064	0.289
Random Forest	0.502	0.560	0.474	0.903	0.004	0.123
SVM	0.504	0.579	0.498	0.958	0.007	0.159
Ours	0.515	0.661	0.477	0.971	0.034	0.240

The ROC curve of the better performing deep forest model is further compared with that of this paper’s model, and the comparison is shown in Fig. 3, which shows that the ROC curve of this paper’s model is closer to the upper left corner than that of the deep forest model, indicating that this paper’s model has a better recognition effect.

In order to verify the stability of the model, ten-fold cross-validation is carried out, and the results of ten ten-fold cross-validation are shown in Table 3. It can be seen that the ten times the mean of the indicators are also good performance, and the standard deviation are relatively low, can reflect the stability of this paper based on the LightGBM algorithm model of the abuse of the right to sue behavior identification model.

Table 3.

Ten times ten times

Rotation	AUC	ACC	Recall
1	0.897	0.881	0.832
2	0.893	0.864	0.901
3	0.918	0.974	0.941
4	0.955	0.978	0.935
5	0.934	0.979	0.942
6	0.913	0.953	0.913
7	0.922	0.962	0.932
8	0.915	0.971	0.931
9	0.929	0.968	0.922
10	0.925	0.988	0.924
Average	0.9201	0.988	0.9173
Std	0.016	0.009	0.013

5

Conclusion

Abuse of the right to sue behavior exists in many aspects of the harm, in order to solve the problem of abuse of the right to sue behavior, this paper focuses on the abuse of the right to sue behavior identification to carry out research, according to the results of this paper can be obtained as follows several conclusions: 1)

for the problem of abusive litigation behavior data, explore the degree of impact on the model due to the imbalance of litigation behavior data, through the comparison of the results of different processing methods, the data after SMOTE oversampling of the three indexes are the highest value, ACC, AUC and F1 indexes were 0.9715, 0.9981, 0.9721. The model in the imbalance of the data after the treatment of the results of the model is better, and the results of the model in the unbalanced data processing. It also proves that SMOTE oversampling method is more suitable.

2)

This paper compares different algorithm models based on real data. After the unbalanced data processing, the AUC, ACC and Recall of different models are improved, which further proves that the data processing has a great enhancement on the overall classification effect of the model. The average AUC, ACC, and Recall in the ten ten-fold cross-validation results are 0.9201, 0.9518, and 0.9173, respectively, which further verifies that the abusive litigation behavior recognition model in this paper has good stability.

Suggestions for abusive litigation behavior: 1)

The establishment of independent pre-trial procedures, civil pre-trial procedures, only after the plaintiff side of the lawsuit, to the trial body before the trial with independence of the intermediate procedures.

2)

The establishment of abusive tort damages liability system, the subject of civil litigation rights, if the abuse of their rights, may cause wrongful infringement of others, in the nature of a tort, should be sanctioned by law.

3)

Improve the administrative penalty mechanism for abusive behavior, in judicial practice, most of the abusive behavior meets the requirements of tort, so it is necessary for the injured party to claim damages by way of a tort claim.

4)

Trial organs should take effective measures to curb the phenomenon of abusive litigation, the trial organs in the process of case adjudication, should guide the parties to establish a reasonable litigation concept, the implementation of the principle of good faith in accordance with the law.

Lingua:: Inglese

Frequenza di pubblicazione:: 1 volte all'anno
Argomenti della rivista:: Scienze biologiche, Scienze della vita, altro, Matematica, Matematica applicata, Matematica generale, Fisica, Fisica, altro

Feed RSS della rivista

Big data identification and calculation model construction of civil abuse of right of action behavior based on the constraints of laws and regulations

Bihua Ou

Baomin Wang

Xiaoying Zhao

Pubblicato online: 23 set 2025

Ricevuto: 14 gen 2025

Accettato: 23 apr 2025

DOI: https://doi.org/10.2478/amns-2025-0996

Parole chiaveCivil abusive litigation behavior, Unbalanced data processing, LightGBM, Behavior recognition model

© 2025 Bihua Ou et al., published by Sciendo

This work is licensed under the Creative Commons Attribution 4.0 International License.

Parole chiave
Civil abusive litigation behavior, Unbalanced data processing, LightGBM, Behavior recognition model