Deep learning litigation material preprocessing and classifying method

Abstract

The invention relates to the field of data processing, in particular to a deep learning litigation material preprocessing and classifying method. The method comprises the following steps: acquiring initial litigation material data in a database; converting the initial litigation material data into a high-dimensional vector by using a pre-trained Word2Vec word embedding model to obtain coded litigation material data; performing feature extraction on the coded litigation material data through a CNN convolutional neural network to obtain feature litigation material data; establishing a B i-LSTM litigation material classification and prediction model based on a bidirectional B i-LSTM long-short term memory network, and performing combinatorial optimization on the B i-LSTM litigation material classification and prediction model by using a whale optimization algorithm; and inputting the feature litigation material data into the target B i-LSTM litigation material classification prediction model for classification, obtaining a user query record of encrypted and classified litigation material data in real time, and performing data query recommendation classification according to the user query record. The category to which the litigation material belongs can be automatically judged according to the input litigation material, and efficient and accurate classification is realized.

Description

Technical Field

[0001] The present invention relates to the field of data analysis, and in particular to a deep learning litigation material preprocessing and classification method. Background Art

[0002] Today, as judicial popularization gradually takes root in people's hearts, using the law to protect one's rights has become more and more frequent. Existing litigation materials are usually accompanied by a variety of text evidence, image evidence and video evidence, which makes the litigation materials numerous and complex. Traditional litigation material processing and classification methods can no longer meet the needs of modern judicial practice. Traditional classification methods often rely on manual work, which is not only inefficient, but also easily affected by human factors, resulting in inaccurate and inconsistent classification results. Therefore, how to effectively process a large amount of litigation materials and achieve accurate classification has become a technical problem that needs to be solved urgently in the judicial field at this stage. Summary of the invention

[0003] The purpose of the present invention is to solve the above problems and to design a deep learning litigation material preprocessing and classification method.

[0004] To achieve the above-mentioned purpose, the technical solution of the present invention is that, further, in the above-mentioned method for preprocessing and classifying litigation materials through deep learning, the method for preprocessing and classifying litigation materials comprises the following steps:

[0005] Obtaining litigation material data to be classified from a database, and performing data preprocessing on the litigation material data to be classified to obtain initial litigation material data;

[0006] The initial litigation material data is converted into a high-dimensional vector using a pre-trained Word2Vec word embedding model to obtain encoded litigation material data; features of the encoded litigation material data are extracted using a CNN convolutional neural network to obtain feature litigation material data;

[0007] A Bi-LSTM litigation material classification prediction model is established based on a bidirectional Bi-LSTM long short-term memory network, and the number of input layer neurons, the number of hidden layer neurons and the learning rate of the Bi-LSTM litigation material classification prediction model are combined and optimized using a whale optimization algorithm to obtain a target Bi-LSTM litigation material classification prediction model;

[0008] Inputting the characteristic litigation material data into the target Bi-LSTM litigation material classification prediction model for classification to obtain classified litigation material data;

[0009] Encrypting the classified litigation material data using an AES symmetric encryption algorithm to obtain encrypted classified litigation material data;

[0010] The encrypted classified litigation material data is transmitted to a preset litigation material classification database for storage, user query records of the encrypted classified litigation material data are obtained in real time, and data query recommendation classification is performed based on the user query records.

[0011] Furthermore, in the above litigation material preprocessing and classification method, the acquiring of litigation material data to be classified in the database, and performing data preprocessing on the litigation material data to be classified to obtain initial litigation material data include:

[0012] Obtaining litigation material data to be classified in the database, wherein the litigation material data to be classified includes at least: documentary evidence, physical evidence, audio-visual materials, electronic image data, witness testimony, expert opinions, inspection records, identity certification materials, power of attorney related materials, plaintiff's basic information, defendant's basic information, litigation requests, facts and reasons;

[0013] Performing data noise reduction processing on the litigation material data to be classified to obtain noise-reduced litigation material data;

[0014] Performing data enhancement processing on the noise-reduced litigation material data to obtain enhanced litigation material data;

[0015] The enhanced litigation material data is normalized to obtain initial litigation material data.

[0016] Furthermore, in the above litigation material preprocessing and classification method, the pre-trained Word2Vec word embedding model is used to convert the initial litigation material data into a high-dimensional vector to obtain encoded litigation material data; and the CNN convolutional neural network is used to extract features of the encoded litigation material data to obtain feature litigation material data, including:

[0017] Obtain historical litigation material data from a database, and input text words in the historical litigation material data into a Word2Vec word embedding model for training;

[0018] Using a pre-trained Word2Vec word embedding model, the initial litigation material data is converted into a high-dimensional vector to obtain encoded litigation material data;

[0019] Constructing a CNN convolutional neural network model, using a sigmoid function as an activation function of the CNN convolutional neural network model, and using the CNN convolutional neural network model to recognize image data and video data;

[0020] The encoded litigation material data is used as input to train the CNN convolutional neural network, and the weight parameters in the neural network are adjusted through the back-propagation algorithm to output characteristic litigation material data.

[0021] Furthermore, in the above litigation material preprocessing and classification method, a Bi-LSTM litigation material classification prediction model is established based on a bidirectional Bi-LSTM long short-term memory network, and the number of input layer neurons, the number of hidden layer neurons and the learning rate of the Bi-LSTM litigation material classification prediction model are combined and optimized using a whale optimization algorithm to obtain a target Bi-LSTM litigation material classification prediction model, including:

[0022] Establish a Bi-LSTM litigation material classification prediction model based on a bidirectional Bi-LSTM long short-term memory network;

[0023] Constructing a WOA whale algorithm, and randomly initializing the WOA whale algorithm using a Sobol sequence initialization method;

[0024] Evaluate the individual fitness in the WOA whale algorithm to obtain the group fitness of the WOA whale algorithm;

[0025] Based on the group fitness, the group positioning parameters of the WOA whale algorithm are obtained; the parameters of the WOA whale algorithm are updated to obtain an optimized I WOA whale optimization algorithm.

[0026] Furthermore, in the above litigation material preprocessing and classification method, the Bi-LSTM litigation material classification prediction model is established based on the bidirectional Bi-LSTM long short-term memory network, and the number of input layer neurons, the number of hidden layer neurons and the learning rate of the Bi-LSTM litigation material classification prediction model are combined and optimized using the whale optimization algorithm to obtain the target Bi-LSTM litigation material classification prediction model, which also includes:

[0027] The I WOA whale optimization algorithm is used to perform combined optimization on the number of input layer neurons, the number of hidden layer neurons and the learning rate of the Bi-LSTM litigation material classification prediction model;

[0028] The I WOA whale optimization algorithm uses the RMSE, a prediction accuracy evaluation index of the Bi-LSTM litigation material classification prediction model, as the fitness value, the hyperparameter as the independent variable, and the fitness value as the objective function;

[0029] Run the I WOA whale optimization algorithm to perform hyperparameter optimization operations, take the optimal fitness value as the goal, use a variety of optimization methods to determine the best hyperparameter combination, and output the hyperparameters corresponding to the optimal fitness value after reaching the termination condition;

[0030] The hyperparameters are introduced into the Bi-LSTM litigation material classification prediction model, and the target Bi-LSTM litigation material classification prediction model is obtained by combining the model weights and learning parameters.

[0031] Furthermore, in the above litigation material preprocessing and classification method, the classified litigation material data is encrypted using the AES symmetric encryption algorithm to obtain the encrypted classified litigation material data, including:

[0032] Acquiring classified litigation material data, wherein the classified litigation material data at least includes criminal litigation material data and civil litigation material data;

[0033] Initialize the AES symmetric encryption algorithm, and generate a round key according to the key length of the AES symmetric encryption algorithm;

[0034] Performing an XOR operation on the initial key and the classified litigation material data to obtain an operation result, and using the operation result as an input of an initial encryption round;

[0035] According to the selected key length and encryption mode, multiple rounds of round functions are repeatedly applied, and the round key is added in the last step of the round function to obtain encrypted classified litigation material data.

[0036] Furthermore, in the above litigation material preprocessing and classification method, the encrypted and classified litigation material data is transmitted to a preset litigation material classification database for storage, user query records of the encrypted and classified litigation material data are obtained in real time, and data query recommendation classification is performed according to the user query records, including:

[0037] Transmitting the encrypted classified litigation material data to a preset litigation material classification database for storage;

[0038] Acquire user query records of the encrypted and classified litigation material data in real time, analyze the query conditions, query frequency and query results in the user query records, and obtain the types of data frequently queried by users;

[0039] Based on the types of data frequently queried by the users, a query cache mechanism is added to compress or archive the encrypted classified litigation material data that is not frequently queried.

[0040] Furthermore, in a system for implementing a deep learning litigation material preprocessing and classification method, the system includes the following modules:

[0041] A data acquisition module is used to acquire the litigation material data to be classified in the database, and perform data preprocessing on the litigation material data to be classified to obtain initial litigation material data;

[0042] A data processing module is used to convert the initial litigation material data into a high-dimensional vector using a pre-trained Word2Vec word embedding model to obtain encoded litigation material data; and extract features from the encoded litigation material data using a CNN convolutional neural network to obtain feature litigation material data;

[0043] A model optimization module is used to establish a Bi-LSTM litigation material classification prediction model based on a bidirectional Bi-LSTM long short-term memory network, and to use a whale optimization algorithm to perform combined optimization on the number of input layer neurons, the number of hidden layer neurons, and the learning rate of the Bi-LSTM litigation material classification prediction model to obtain a target Bi-LSTM litigation material classification prediction model;

[0044] A material identification module, used for inputting the characteristic litigation material data into the target Bi-LSTM litigation material classification prediction model for classification, and obtaining classified litigation material data;

[0045] A material encryption module, used to encrypt the classified litigation material data using an AES symmetric encryption algorithm to obtain encrypted classified litigation material data;

[0046] The material storage module is used to transfer the encrypted classified litigation material data to a preset litigation material classification database for storage, obtain user query records of the encrypted classified litigation material data in real time, and perform data query recommendation classification based on the user query records.

[0047] Furthermore, in a system for implementing a deep learning litigation material preprocessing and classification method, the system includes the following submodules:

[0048] Establish a submodule for establishing a Bi-LSTM litigation material classification prediction model based on a bidirectional Bi-LSTM long short-term memory network;

[0049] An initialization submodule is used to construct a WOA whale algorithm and randomly initialize the WOA whale algorithm using a Sobol sequence initialization method;

[0050] An evaluation submodule, used to evaluate the individual fitness in the WOA whale algorithm to obtain the group fitness of the WOA whale algorithm;

[0051] The parameter submodule is used to obtain the group positioning parameters of the WOA whale algorithm based on the group fitness; update the parameters of the WOA whale algorithm to obtain the optimized I WOA whale optimization algorithm.

[0052] Furthermore, in a system for implementing a deep learning litigation material preprocessing and classification method, the system includes the following submodules:

[0053] Establish a submodule for establishing a Bi-LSTM litigation material classification prediction model based on a bidirectional Bi-LSTM long short-term memory network;

[0054] An initialization submodule is used to construct a WOA whale algorithm and randomly initialize the WOA whale algorithm using a Sobol sequence initialization method;

[0055] An evaluation submodule, used to evaluate the individual fitness in the WOA whale algorithm to obtain the group fitness of the WOA whale algorithm;

[0056] The parameter submodule is used to obtain the group positioning parameters of the WOA whale algorithm based on the group fitness; update the parameters of the WOA whale algorithm to obtain the optimized I WOA whale optimization algorithm.

[0057] The beneficial effects are as follows: by acquiring the litigation material data to be classified in the database, data preprocessing is performed on the litigation material data to be classified to obtain initial litigation material data; the initial litigation material data is converted into a high-dimensional vector by using a pre-trained Word2Vec word embedding model to obtain encoded litigation material data; the encoded litigation material data is feature extracted by using a CNN convolutional neural network to obtain feature litigation material data; a Bi-LSTM litigation material classification prediction model is established based on a bidirectional Bi-LSTM long short-term memory network, and the number of input layer neurons, the number of hidden layer neurons and the learning rate of the Bi-LSTM litigation material classification prediction model are combined and optimized by using a whale optimization algorithm to obtain a target Bi-LSTM litigation material classification prediction model; the feature litigation material data is input into the target Bi-LSTM litigation material classification prediction model for classification to obtain classified litigation material data; the classified litigation material data is encrypted by using an AES symmetric encryption algorithm to obtain encrypted classified litigation material data; the encrypted classified litigation material data is transmitted to a preset litigation material classification database for storage, user query records of the encrypted classified litigation material data are obtained in real time, and data query recommendation classification is performed according to the user query records. It can automatically determine the category of the input litigation materials and achieve efficient and accurate classification, which can better capture the semantic information of text data. By using image recognition technology to extract and identify features of image data, it can more accurately determine the image content and thus achieve accurate and efficient litigation material processing and classification. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the following detailed description of the preferred embodiment.The drawings are only for the purpose of illustrating the preferred embodiments and are not to be construed as limiting the invention.

[0059] Figure 1 This is a schematic diagram of a first embodiment of a deep learning litigation material preprocessing and classification method in an embodiment of the present invention;

[0060] Figure 2 It is a schematic diagram of a second embodiment of a method for preprocessing and classifying litigation materials through deep learning in an embodiment of the present invention;

[0061] Figure 3 Schematic diagram of a third embodiment of a deep learning litigation material preprocessing and classification method in an embodiment of the present invention. DETAILED DESCRIPTION

[0062] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0063] Those skilled in the art will appreciate that, unless otherwise stated, the singular forms "a", "an", "the" used herein may also include plural forms. It should be further understood that the term "comprising" used in the specification of the present invention refers to the presence of the features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof.

[0064] The present invention will be described in detail below in conjunction with the accompanying drawings. Figure 1 As shown, a deep learning litigation material preprocessing and classification method, the litigation material preprocessing and classification method includes the following steps:

[0065] Step 101, obtaining litigation material data to be classified in a database, performing data preprocessing on the litigation material data to be classified, and obtaining initial litigation material data;

[0066] Specifically, the present embodiment also includes obtaining litigation material data to be classified in the database, and the litigation material data to be classified includes at least: documentary evidence, physical evidence, audio-visual materials, electronic image data, witness testimony, appraisal opinions, inspection records, identity certification materials, power of attorney related materials, plaintiff's basic information, defendant's basic information, litigation requests, facts and reasons; performing data denoising processing on the litigation material data to be classified to obtain denoised litigation material data; performing data enhancement processing on the denoised litigation material data to obtain enhanced litigation material data; and performing normalization processing on the enhanced litigation material data to obtain initial litigation material data.

[0067] Step 102: using a pre-trained Word2Vec word embedding model to convert the initial litigation material data into a high-dimensional vector to obtain encoded litigation material data; performing feature extraction on the encoded litigation material data through a CNN convolutional neural network to obtain feature litigation material data;

[0068] Specifically, the present embodiment also includes obtaining historical litigation material data in a database, inputting text vocabulary in the historical litigation material data into a Word2Vec word embedding model for training; using a pre-trained Word2Vec word embedding model to convert the initial litigation material data into a high-dimensional vector to obtain encoded litigation material data; constructing a CNN convolutional neural network model, using the Sigmoid function as the activation function of the CNN convolutional neural network model, and using the CNN convolutional neural network model to recognize image data and video data; using the encoded litigation material data as input, training the CNN convolutional neural network, adjusting the weight parameters in the neural network through a back-propagation algorithm, and outputting characteristic litigation material data.

[0069] Step 103: Establish a Bi-LSTM litigation material classification prediction model based on a bidirectional Bi-LSTM long short-term memory network, and use the whale optimization algorithm to perform combined optimization on the number of input layer neurons, the number of hidden layer neurons, and the learning rate of the Bi-LSTM litigation material classification prediction model to obtain a target Bi-LSTM litigation material classification prediction model;

[0070] Specifically, this embodiment also includes establishing a Bi-LSTM litigation material classification prediction model based on a bidirectional Bi-LSTM long short-term memory network; constructing a WOA whale algorithm, and randomly initializing the WOA whale algorithm using a Sobol sequence initialization method; evaluating the individual fitness in the WOA whale algorithm to obtain the group fitness of the WOA whale algorithm; obtaining the group positioning parameters of the WOA whale algorithm based on the group fitness; updating the parameters of the WOA whale algorithm to obtain an optimized I WOA whale optimization algorithm.

[0071] Specifically, this embodiment also includes using the IWOA whale optimization algorithm to perform combined optimization on the number of input layer neurons, the number of hidden layer neurons and the learning rate of the Bi-LSTM litigation material classification prediction model; the IWOA whale optimization algorithm uses the prediction accuracy assessment indicator RMSE of the Bi-LSTM litigation material classification prediction model as the fitness value, the hyperparameter as the independent variable, and the fitness value as the objective function; the IWOA whale optimization algorithm is run to perform hyperparameter optimization operations, with the optimal fitness value as the target, and a variety of optimization methods are used to determine the best hyperparameter combination. After the termination condition is reached, the hyperparameter corresponding to the optimal fitness value is output; the hyperparameter is introduced into the Bi-LSTM litigation material classification prediction model, and the target Bi-LSTM litigation material classification prediction model is obtained by combining the model weights and learning parameters.

[0072] Step 104: input the characteristic litigation material data into the target Bi-LSTM litigation material classification prediction model for classification, and obtain classified litigation material data;

[0073] Specifically, the present embodiment also includes criminal litigation material data and civil litigation material data. The criminal litigation material data includes conventional criminal litigation cases: cases involving criminal offenses, such as robbery, theft, intentional homicide, etc. In such cases, the defendant will face corresponding judicial procedures, such as arrest, trial, and judgment. Major criminal cases: Such cases usually involve vicious criminal acts, which are of great social harm and danger, such as terrorist attacks, large-scale drug trafficking, etc. Due to the importance of the case, the trial of major criminal cases usually goes through a more rigorous and complicated procedure. Special criminal cases: including military criminal cases, confidentiality infringement cases, foreign-related criminal cases, etc. The characteristic of such cases is that the legal provisions and procedures related to other fields need to be taken into account when handling them. Juvenile criminal cases: cases involving juvenile criminal acts. Factors such as the protection of the legal rights and interests of minors, education and correction need to be considered during the trial process.

[0074] Data on civil litigation materials include disputes over property rights: These include disputes between citizens and between citizens and legal persons arising from property rights, mostly referring to disputes arising from the possession, use, income and disposal of property. Contractual behavior disputes: These involve disputes between citizens arising from contractual behaviors such as sale, lease, loan, gift, pawn, etc., as well as disputes arising from inheritance. Debt and compensation disputes: These include debt disputes arising from unjust enrichment, gratuitous management, etc., as well as compensation disputes arising from property damage. Personal rights disputes: These mainly refer to disputes arising from infringement of citizens' right to health, right to name, right to reputation, right to honor and right to portrait. Intellectual property disputes: These involve disputes arising from infringement of citizens' invention rights (patent rights) and copyrights. Marriage and family disputes: These mainly include divorce and disputes over property division and child support arising from divorce, as well as disputes over support, upbringing and maintenance between family members. Economic and other disputes: These include disputes arising from economic contracts, enterprise labor employment, enterprise contracting, land contracting, neighboring rights, etc., as well as other civil litigation cases that should be accepted by the people's courts as stipulated by law or the judicial interpretation documents of the Supreme People's Court.

[0075] Step 105: Encrypt the classified litigation material data using the AES symmetric encryption algorithm to obtain encrypted classified litigation material data;

[0076] Specifically, the present embodiment also includes obtaining classified litigation material data, which at least includes criminal litigation material data and civil litigation material data; initializing the AES symmetric encryption algorithm, and generating round keys according to the key length of the AES symmetric encryption algorithm; performing an XOR operation on the initial key and the classified litigation material data to obtain an operation result, and using the operation result as the input of the initial encryption round; repeatedly applying multiple rounds of round functions according to the selected key length and encryption mode, adding the round key in the last step of the round function, and obtaining encrypted classified litigation material data.

[0077] Step 106: The encrypted and classified litigation material data is transmitted to a preset litigation material classification database for storage, user query records of the encrypted and classified litigation material data are obtained in real time, and data query recommendation classification is performed based on the user query records.

[0078] Specifically, the present embodiment also includes transmitting the encrypted classified litigation material data to a preset litigation material classification database for storage; obtaining user query records of the encrypted classified litigation material data in real time, analyzing the query conditions, query frequency and query results in the user query records, and obtaining the types of data that users frequently query; based on the types of data that users frequently query, adding a query cache mechanism, and compressing or archiving the encrypted classified litigation material data that is not frequently queried.

[0079] The beneficial effect lies in that, by acquiring the litigation material data to be classified in the database, data preprocessing is performed on the litigation material data to be classified to obtain initial litigation material data; the initial litigation material data is converted into a high-dimensional vector by using the pre-trained Word2Vec word embedding model to obtain encoded litigation material data; the encoded litigation material data is feature extracted by using the CNN convolutional neural network to obtain feature litigation material data; a Bi-LSTM litigation material classification prediction model is established based on a bidirectional Bi-LSTM long short-term memory network, and the number of input layer neurons, the number of hidden layer neurons and the learning rate of the Bi-LSTM litigation material classification prediction model are combined and optimized by using the whale optimization algorithm to obtain a target Bi-LSTM litigation material classification prediction model; the feature litigation material data is input into the target Bi-LSTM litigation material classification prediction model for classification to obtain classified litigation material data; the classified litigation material data is encrypted by using the AES symmetric encryption algorithm to obtain encrypted classified litigation material data; the encrypted classified litigation material data is transmitted to a preset litigation material classification database for storage, user query records of the encrypted classified litigation material data are obtained in real time, and data query recommendation classification is performed according to the user query records. It can automatically determine the category of the input litigation materials and achieve efficient and accurate classification, which can better capture the semantic information of text data. By using image recognition technology to extract and identify features of image data, it can more accurately determine the image content and thus achieve accurate and efficient litigation material processing and classification.

[0080] In this embodiment, please refer to Figure 2 In a second embodiment of a deep learning litigation material preprocessing and classification method in an embodiment of the present invention, the initial litigation material data is converted into a high-dimensional vector using a pre-trained Word2Vec word embedding model to obtain encoded litigation material data; and feature extraction of the encoded litigation material data is performed through a CNN convolutional neural network to obtain feature litigation material data, including the following steps:

[0081] Step 201, obtaining historical litigation material data in a database, and inputting text words in the historical litigation material data into a Word2Vec word embedding model for training;

[0082] Step 202: using a pre-trained Word2Vec word embedding model to convert the initial litigation material data into a high-dimensional vector to obtain encoded litigation material data;

[0083] Step 203: construct a CNN convolutional neural network model, use the Sigma ID function as the activation function of the CNN convolutional neural network model, and use the CNN convolutional neural network model to recognize image data and video data;

[0084] Step 204: Use the encoded litigation material data as input, train the CNN convolutional neural network, adjust the weight parameters in the neural network through the back propagation algorithm, and output the characteristic litigation material data.

[0085] In this embodiment, please refer to Figure 3 In a third embodiment of a deep learning litigation material preprocessing and classification method according to an embodiment of the present invention, a Bi-LSTM litigation material classification prediction model is established based on a bidirectional Bi-LSTM long short-term memory network, and the number of input layer neurons, the number of hidden layer neurons and the learning rate of the Bi-LSTM litigation material classification prediction model are combined and optimized using a whale optimization algorithm to obtain a target Bi-LSTM litigation material classification prediction model, including the following steps:

[0086] Step 301: Use the I WOA whale optimization algorithm to perform combined optimization on the number of input layer neurons, the number of hidden layer neurons, and the learning rate of the Bi-LSTM litigation material classification prediction model;

[0087] Step 302, the IWOA whale optimization algorithm uses the RMSE, a prediction accuracy evaluation indicator of the Bi-LSTM litigation material classification prediction model, as a fitness value, a hyperparameter as an independent variable, and a fitness value as an objective function;

[0088] Step 303: Run the IWOA whale optimization algorithm to perform hyperparameter optimization operations, take the optimal fitness value as the target, use multiple optimization methods to determine the best hyperparameter combination, and output the hyperparameters corresponding to the optimal fitness value after the termination condition is reached;

[0089] Step 304: Introduce hyperparameters into the Bi-LSTM litigation material classification prediction model, combine model weights and learning parameters, and obtain the target Bi-LSTM litigation material classification prediction model.

[0090] The above shows and describes the basic principles, main features and advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited by the above embodiments. The above embodiments and descriptions are only preferred examples of the present invention and are not intended to limit the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A deep learning litigation material preprocessing and classification method, characterized in that: The litigation material preprocessing and classification method comprises the following steps: Obtaining litigation material data to be classified in a database, and performing data preprocessing on the litigation material data to be classified to obtain initial litigation material data; The initial litigation material data is converted into a high-dimensional vector using a pre-trained Word2Vec word embedding model to obtain encoded litigation material data; features of the encoded litigation material data are extracted using a CNN convolutional neural network to obtain feature litigation material data; A Bi-LSTM litigation material classification prediction model is established based on a bidirectional Bi-LSTM long short-term memory network, and the number of input layer neurons, the number of hidden layer neurons and the learning rate of the Bi-LSTM litigation material classification prediction model are combined and optimized using a whale optimization algorithm to obtain a target Bi-LSTM litigation material classification prediction model; Inputting the characteristic litigation material data into the target Bi-LSTM litigation material classification prediction model for classification to obtain classified litigation material data; Encrypting the classified litigation material data using an AES symmetric encryption algorithm to obtain encrypted classified litigation material data; The encrypted classified litigation material data is transmitted to a preset litigation material classification database for storage, user query records of the encrypted classified litigation material data are obtained in real time, and data query recommendation classification is performed based on the user query records.

2. A deep learning litigation material preprocessing and classification method as claimed in claim 1, characterized in that: The step of obtaining the litigation material data to be classified in the database and performing data preprocessing on the litigation material data to be classified to obtain initial litigation material data includes: Obtaining litigation material data to be classified in the database, wherein the litigation material data to be classified includes at least: documentary evidence, physical evidence, audio-visual materials, electronic image data, witness testimony, expert opinions, inspection records, identity certification materials, power of attorney related materials, plaintiff's basic information, defendant's basic information, litigation requests, facts and reasons; Performing data noise reduction processing on the litigation material data to be classified to obtain noise-reduced litigation material data; Performing data enhancement processing on the noise-reduced litigation material data to obtain enhanced litigation material data; The enhanced litigation material data is normalized to obtain initial litigation material data.

3. A deep learning litigation material preprocessing and classification method as claimed in claim 1, characterized in that: The pre-trained Word2Vec word embedding model is used to convert the initial litigation material data into a high-dimensional vector to obtain encoded litigation material data; and the CNN convolutional neural network is used to extract features of the encoded litigation material data to obtain feature litigation material data, including: Obtain historical litigation material data from a database, and input text words in the historical litigation material data into a Word2Vec word embedding model for training; Using a pre-trained Word2Vec word embedding model, the initial litigation material data is converted into a high-dimensional vector to obtain encoded litigation material data; Constructing a CNN convolutional neural network model, using a Sigmoid function as an activation function of the CNN convolutional neural network model, and using the CNN convolutional neural network model to recognize image data and video data; The encoded litigation material data is used as input to train the CNN convolutional neural network, and the weight parameters in the neural network are adjusted through the back-propagation algorithm to output characteristic litigation material data.

4. A deep learning litigation material preprocessing and classification method as claimed in claim 1, characterized in that: The Bi-LSTM litigation material classification prediction model is established based on the bidirectional Bi-LSTM long short-term memory network, and the number of input layer neurons, the number of hidden layer neurons and the learning rate of the Bi-LSTM litigation material classification prediction model are combined and optimized using the whale optimization algorithm to obtain the target Bi-LSTM litigation material classification prediction model, including: A Bi-LSTM litigation material classification prediction model is established based on a bidirectional Bi-LSTM long short-term memory network; Constructing a WOA whale algorithm, and randomly initializing the WOA whale algorithm using a Sobol sequence initialization method; Evaluate the individual fitness in the WOA whale algorithm to obtain the group fitness of the WOA whale algorithm; Based on the group fitness, the group positioning parameters of the WOA whale algorithm are obtained; the parameters of the WOA whale algorithm are updated to obtain an optimized I WOA whale optimization algorithm.

5. A deep learning litigation material preprocessing and classification method as claimed in claim 1, characterized in that: The method further includes establishing a Bi-LSTM litigation material classification prediction model based on a bidirectional Bi-LSTM long short-term memory network, and using a whale optimization algorithm to perform combined optimization on the number of input layer neurons, the number of hidden layer neurons, and the learning rate of the Bi-LSTM litigation material classification prediction model to obtain a target Bi-LSTM litigation material classification prediction model. The I WOA whale optimization algorithm is used to perform combined optimization on the number of input layer neurons, the number of hidden layer neurons and the learning rate of the Bi-LSTM litigation material classification prediction model; The I WOA whale optimization algorithm uses the prediction accuracy evaluation index RMSE of the Bi-LSTM litigation material classification prediction model as the fitness value, the hyperparameter as the independent variable, and the fitness value as the objective function; Run the I WOA whale optimization algorithm to perform hyperparameter optimization operations, take the optimal fitness value as the goal, use a variety of optimization methods to determine the best hyperparameter combination, and output the hyperparameters corresponding to the optimal fitness value after reaching the termination condition; The hyperparameters are introduced into the Bi-LSTM litigation material classification prediction model, and the target Bi-LSTM litigation material classification prediction model is obtained by combining the model weights and learning parameters.

6. A deep learning litigation material preprocessing and classification method as claimed in claim 1, characterized in that: The classified litigation material data is encrypted using the AES symmetric encryption algorithm to obtain the encrypted classified litigation material data, including: Acquiring classified litigation material data, wherein the classified litigation material data at least includes criminal litigation material data and civil litigation material data; Initialize the AES symmetric encryption algorithm, and generate a round key according to the key length of the AES symmetric encryption algorithm; Performing an XOR operation on the initial key and the classified litigation material data to obtain an operation result, and using the operation result as an input of an initial encryption round; According to the selected key length and encryption mode, multiple rounds of round functions are repeatedly applied, and the round key is added in the last step of the round function to obtain encrypted classified litigation material data.

7. A deep learning litigation material preprocessing and classification method as claimed in claim 1, characterized in that: The step of transmitting the encrypted classified litigation material data to a preset litigation material classification database for storage, obtaining user query records of the encrypted classified litigation material data in real time, and performing data query recommendation classification according to the user query records includes: Transmitting the encrypted classified litigation material data to a preset litigation material classification database for storage; Acquire user query records of the encrypted and classified litigation material data in real time, analyze the query conditions, query frequency and query results in the user query records, and obtain the types of data frequently queried by users; Based on the types of data frequently queried by the users, a query cache mechanism is added to compress or archive the encrypted classified litigation material data that is not frequently queried.

8. A system for implementing a deep learning litigation material preprocessing and classification method as claimed in claim 1, characterized in that: The system includes the following modules: A data acquisition module is used to acquire the litigation material data to be classified in the database, and perform data preprocessing on the litigation material data to be classified to obtain initial litigation material data; A data processing module is used to convert the initial litigation material data into a high-dimensional vector using a pre-trained Word2Vec word embedding model to obtain encoded litigation material data; and extract features from the encoded litigation material data using a CNN convolutional neural network to obtain feature litigation material data; A model optimization module is used to establish a Bi-LSTM litigation material classification prediction model based on a bidirectional Bi-LSTM long short-term memory network, and to use a whale optimization algorithm to perform combined optimization on the number of input layer neurons, the number of hidden layer neurons, and the learning rate of the Bi-LSTM litigation material classification prediction model to obtain a target Bi-LSTM litigation material classification prediction model; A material identification module, used for inputting the characteristic litigation material data into the target Bi-LSTM litigation material classification prediction model for classification, to obtain classified litigation material data; A material encryption module, used to encrypt the classified litigation material data using an AES symmetric encryption algorithm to obtain encrypted classified litigation material data; The material storage module is used to transfer the encrypted classified litigation material data to a preset litigation material classification database for storage, obtain user query records of the encrypted classified litigation material data in real time, and perform data query recommendation classification based on the user query records.

9. A system for implementing a deep learning litigation material preprocessing and classification method as claimed in claim 1, characterized in that: The system includes the following submodules: An acquisition submodule is used to acquire historical litigation material data in a database, and input text words in the historical litigation material data into a Word2Vec word embedding model for training; A conversion submodule, used to convert the initial litigation material data into a high-dimensional vector using a pre-trained Word2Vec word embedding model to obtain encoded litigation material data; A construction submodule is used to construct a CNN convolutional neural network model, use a Sigmoid function as an activation function of the CNN convolutional neural network model, and use the CNN convolutional neural network model to recognize image data and video data; A submodule is obtained, which is used to take the encoded litigation material data as input, train the CNN convolutional neural network, adjust the weight parameters in the neural network through the back propagation algorithm, and output the characteristic litigation material data.

10. A system for implementing a deep learning litigation material preprocessing and classification method as claimed in claim 1, characterized in that: The system includes the following submodules: Establish a submodule for establishing a Bi-LSTM litigation material classification prediction model based on a bidirectional Bi-LSTM long short-term memory network; An initialization submodule is used to construct a WOA whale algorithm and randomly initialize the WOA whale algorithm using a Sobol sequence initialization method; An evaluation submodule, used to evaluate the individual fitness in the WOA whale algorithm to obtain the group fitness of the WOA whale algorithm; The parameter submodule is used to obtain the group positioning parameters of the WOA whale algorithm based on the group fitness; update the parameters of the WOA whale algorithm to obtain the optimized I WOA whale optimization algorithm.

Images