Text information processing method, device, storage medium and processor
Through the multi-task prediction model, the text information in the litigation scenario is predicted and processed, which solves the problem of inefficient case handling in the litigation scenario, and accurately predicts the legal information and litigation prediction information of the case, assists in case trial and mediation.
Patent Information
- Application Number
- CN201911120523.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2019-11-15
- Publication Date
- 2024-11-12
- Estimated Expiration
- 2039-11-15
AI Technical Summary
In litigation scenarios, the litigation results and the legal information used cannot be predicted, resulting in inefficient case handling.
The multi-task prediction model is used to predict the text information in the litigation scenario, and the legal information and litigation prediction information of the case are predicted. The trained model provides joint prediction information and legal information for submitted cases to assist the case. trial.
It improves the efficiency of case handling in litigation scenarios, provides reference results for case trials, helps case mediators formulate mediation strategies, and improves the accuracy and efficiency of judicial assistance.
Smart Images

Figure CN112818671B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of text information processing, and in particular to a text information processing method, device, storage medium and processor. Background Art
[0002] At present, with the enhancement of people's legal awareness, there are more and more cases, especially civil cases, and the Supreme Court also encourages more civil cases to enter the multi-party mediation dispute stage. However, currently, mediation is usually carried out manually according to the actual situation of the case, which takes a long time. There is no solution to predict the litigation results and legal information used in the litigation scene, resulting in low efficiency in case trials.
[0003] Currently, no effective solution has been proposed to the above-mentioned problem of inefficient handling of cases in litigation scenarios due to the inability to predict the litigation results and the legal information used in the litigation scenarios. Summary of the invention
[0004] Embodiments of the present invention provide a method, device, storage medium and processor for processing text information, so as to at least solve the technical problem of inefficient processing of cases in litigation scenarios due to the inability to predict the litigation results of the case and the legal information used in the litigation scenario.
[0005] According to one aspect of an embodiment of the present invention, a method for processing text information is provided. The method comprises: obtaining target text information, wherein the target text information is used to describe case information of a case in a litigation scenario; performing prediction processing on the target text information based on a multi-task prediction model to predict legal information and litigation prediction information of the case, wherein the multi-task prediction model is trained by using text information of case information of case samples used to describe case information in the litigation scenario and legal information samples of the case samples; and outputting the legal information and litigation prediction information of the case.
[0006] According to another aspect of an embodiment of the present invention, a method for processing text information is also provided. The method comprises: displaying target text information on a text display interface, wherein the target text information is used to describe case information of a case in a litigation scenario; and displaying legal information and litigation prediction information of the case on the text display interface, wherein the legal information and litigation prediction information of the case are the result of predicting the target text information based on a multi-task prediction model, and the multi-task prediction model is trained by using text information of case information of case samples in a litigation scenario and legal information samples of the case samples.
[0007] According to another aspect of an embodiment of the present invention, a method for processing case information is also provided. The method includes: obtaining case information of a case, wherein the case information of the case includes at least one of the following: a complaint text, a defense text, a legal and regulatory text, an evidence text, and a judgment text; performing prediction processing on the case information of the case based on a multi-task prediction model, and predicting the legal information and litigation prediction information of the case, wherein the multi-task prediction model is a neural network model, which is used to encode different types of files in the case information of the case; and outputting the legal information and litigation prediction information of the case.
[0008] According to another aspect of an embodiment of the present invention, a text information processing device is also provided. The device includes: a first acquisition unit, used to acquire target text information, wherein the target text information is used to describe case information of a case in a litigation scenario; a first processing unit, used to perform prediction processing on the target text information based on a multi-task prediction model, and predict the legal information of the case and litigation prediction information, wherein the multi-task prediction model is trained by text information of case information of case samples used to describe case information in the litigation scenario and legal information samples of the case samples; a first output unit, used to output the legal information of the case and litigation prediction information.
[0009] According to another aspect of an embodiment of the present invention, a text information processing device is also provided. The device includes: a first display unit, used to display target text information on a text display interface, wherein the target text information is used to describe case information of a case in a litigation scenario; a second display unit, used to display legal information and litigation prediction information of the case on the text display interface, wherein the legal information and litigation prediction information of the case are the results of predictive processing of the target text information based on a multi-task prediction model, and the multi-task prediction model is trained by using text information of case information of case samples in the litigation scenario and legal information samples of the case samples.
[0010] According to another aspect of an embodiment of the present invention, a case information processing device is also provided. The device includes: a second acquisition unit, used to acquire case information of a case, wherein the case information of the case includes at least one of the following: a complaint text, a defense text, a legal and regulatory text, an evidence text, and a judgment text; a second processing unit, used to perform prediction processing on the case information of the case based on a multi-task prediction model, and predict the legal information and litigation prediction information of the case, wherein the multi-task prediction model is a neural network model, which is used to encode different types of files in the case information of the case; a second output unit, used to output the legal information and litigation prediction information of the case.
[0011] According to another aspect of an embodiment of the present invention, a storage medium is also provided. The storage medium includes a stored program, wherein when the program is running, the device where the storage medium is located is controlled to perform the following steps: obtaining target text information, wherein the target text information is used to describe case information of a case in a litigation scenario; performing prediction processing on the target text information based on a multi-task prediction model to predict legal information and litigation prediction information of the case, wherein the multi-task prediction model is trained by text information of case information of case samples used to describe case information in the litigation scenario and legal information samples of the case samples; and outputting the legal information and litigation prediction information of the case.
[0012] According to another aspect of an embodiment of the present invention, a processor is also provided. The processor is used to run a program, wherein the following steps are performed when the program is run: obtaining target text information, wherein the target text information is used to describe case information of a case in a litigation scenario; performing prediction processing on the target text information based on a multi-task prediction model to predict legal information and litigation prediction information of the case, wherein the multi-task prediction model is trained by text information of case information of case samples used to describe case information in the litigation scenario and legal information samples of the case samples; and outputting the legal information and litigation prediction information of the case.
[0013] According to another aspect of an embodiment of the present invention, a mobile terminal is also provided. The mobile terminal includes: a processor; a transmission device connected to the processor, used to obtain target text information, wherein the target text information is used to describe case information of a case in a litigation scenario; and a memory connected to the processor, used to provide the processor with instructions for processing the following processing steps: predicting the target text information based on a multi-task prediction model to predict legal information and litigation prediction information of the case, wherein the multi-task prediction model is trained by text information of case information of case samples used to describe case information in the litigation scenario and legal information samples of the case samples; outputting the legal information and litigation prediction information of the case.
[0014] In an embodiment of the present invention, target text information is obtained, wherein the target text information is used to describe case information of a case in a litigation scenario; the target text information is predicted and processed based on a multi-task prediction model to predict the legal information and litigation prediction information of the case, wherein the multi-task prediction model is trained by text information used to describe case information of case samples in the litigation scenario and legal information samples of the case samples; and the legal information and litigation prediction information of the case are output. That is, through the pre-trained multi-task prediction model, the litigation prediction information and legal information of the submitted case are jointly predicted, thereby achieving the purpose of assisting the trial of the case, realizing the improvement of the efficiency of handling cases in the litigation scenario, and thus solving the technical problem of low efficiency in handling cases in the litigation scenario due to the inability to predict the litigation results of the case and the legal information used in the litigation scenario. BRIEF DESCRIPTION OF THE DRAWINGS
[0015] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of this application. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:
[0016] Figure 1 A hardware structure block diagram of a computer terminal (or mobile device) for implementing a text information processing method is shown;
[0017] Figure 2 is a flowchart of a method for processing text information according to an embodiment of the present invention;
[0018] Figure 3a is a flowchart of another text information processing method according to an embodiment of the present invention;
[0019] Figure 3b is a schematic diagram of interaction in another text information processing method according to an embodiment of the present invention;
[0020] Figure 4 is a flow chart of a method for processing case information according to an embodiment of the present invention;
[0021] Figure 5 is a schematic diagram of a text information processing system according to an embodiment of the present invention;
[0022] Figure 6 is a schematic diagram of an architecture of a multi-task model according to an embodiment of the present invention;
[0023] Figure 7 is a schematic diagram of a network structure of a multi-task model according to an embodiment of the present invention;
[0024] Figure 8is a schematic diagram of a text information processing device according to an embodiment of the present invention;
[0025] Fig. 9 is a schematic diagram of another text information processing device according to an embodiment of the present invention;
[0026] Fig.10 is a schematic diagram of a case information processing device according to an embodiment of the present invention; and
[0027] Fig.11 is a structural block diagram of a mobile terminal according to an embodiment of the present invention. DETAILED DESCRIPTION
[0028] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.
[0029] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.
[0030] First, some nouns or terms that appear in the description of the embodiments of the present application are subject to the following explanations:
[0031] Term Frequency–Inverse Document Frequency (TF-IDF) is a commonly used weighting technique for information retrieval and data mining.
[0032] Continuous Bag-of-Words (CBOW) model: The training input of the CBOW model is the word vector corresponding to the context-related words of a certain feature word, and the output is the word vector of this specific word.
[0033] The Skip-gram neural network model, whose input is the word vector of a specific word, and the output is the context word vector corresponding to the specific word;
[0034] A new deep contextual word representation algorithm (Embeddings from Language Models, ELMo for short) is used to model complex features of words (such as syntax and semantics) and how words change in language context (i.e., modeling polysemous words);
[0035] The self-attention mechanism is applied to various tasks of natural language processing based on deep learning.
[0036] Example 1
[0037] According to an embodiment of the present invention, a method embodiment of a method for processing text information is also provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.
[0038] The method embodiment provided in the first embodiment of the present application can be executed in a mobile terminal, a computer terminal or a similar computing device. Figure 1 FIG. 1 shows a hardware structure block diagram of a computer terminal (or mobile device) for implementing a text information processing method. Figure 1 As shown, the computer terminal 10 (or mobile device 10) may include one or more (102a, 102b, ..., 102n are used to illustrate) processors 112 (the processor 112 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA), a memory 114 for storing data, and a transmission device for communication functions. In addition, it may also include: a display, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of the I / O interface), a network interface, a power supply and / or a camera. It can be understood by those skilled in the art that Figure 1 The structure shown is only for illustration and does not limit the structure of the above electronic device. Figure 1 More or fewer components as shown, or with Figure 1 Different configurations are shown.
[0039] It should be noted that the one or more processors 112 and / or other data processing circuits described above may generally be referred to herein as "data processing circuits". The data processing circuits may be embodied in whole or in part as software, hardware, firmware, or any other combination thereof. In addition, the data processing circuit may be a single independent processing module, or may be incorporated in whole or in part into any of the other components in the computer terminal 10 (or mobile device). As described in the embodiments of the present application, the data processing circuit acts as a processor control (e.g., selection of a variable resistor terminal path connected to an interface).
[0040] The memory 114 can be used to store software programs and modules of application software, such as program instructions / data storage devices corresponding to the text information processing method in the embodiment of the present invention. The processor 112 executes various functional applications and data processing by running the software programs and modules stored in the memory 114, that is, the text information processing method of the above-mentioned application program is realized. The memory 114 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some examples, the memory 114 may further include a memory remotely arranged relative to the processor 112, and these remote memories may be connected to the computer terminal 10 via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.
[0041] The transmission device is used to receive or send data via a network. The specific example of the above network may include a wireless network provided by a communication provider of the computer terminal 10. In one example, the transmission device includes a network adapter (Network Interface Controller, NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission device can be a radio frequency (Radio Frequency, RF) module, which is used to communicate with the Internet wirelessly.
[0042] The display may be, for example, a touch screen liquid crystal display (LCD) that enables a user to interact with a user interface of the computer terminal 10 (or mobile device).
[0043] It should be noted that, in some optional embodiments, the above Figure 1 The computer device (or mobile device) shown may include hardware elements (including circuits), software elements (including computer code stored on a computer-readable medium), or a combination of hardware elements and software elements. It should be noted that Figure 1This is merely one example of a particular embodiment and is intended to illustrate the types of components that may be present in the above-described computer device (or mobile device).
[0044] exist Figure 1 In the operating environment shown, this application provides Figure 2 It should be noted that the text information processing method of this embodiment can be Figure 1 The illustrated embodiment is executed by a mobile terminal.
[0045] Figure 2 FIG. 1 is a flowchart of a method for processing text information according to an embodiment of the present invention. Figure 2 As shown, the method comprises the following steps:
[0046] Step S202, obtaining target text information, wherein the target text information is used to describe case information of a case in a litigation scenario.
[0047] In the technical solution provided in step S202 of the present invention, the case in the litigation scenario may be a case that requires dispute mediation, such as a civil case. The case information of this embodiment may include text materials such as the complaint text, the defense text, the legal and regulatory text, the evidence text, and the judgment text required in the process of hearing the case. The target text information of this embodiment is used to describe the case information of the case in the above litigation scenario, and may include case information.
[0048] Step S204: predict the target text information based on the multi-task prediction model to obtain the legal information and litigation prediction information of the case.
[0049] In the technical solution provided in the above step S204 of the present invention, after acquiring the target text information, the target text information is predictively processed based on the multi-task prediction model to predict the legal information and litigation prediction information of the case, wherein the multi-task prediction model is trained by using text information of case information of case samples used to describe the litigation scenario and legal information samples of the case samples to obtain a model for predicting the legal information and litigation prediction information of different cases.
[0050] In this embodiment, after obtaining the target text information, the target text information can be predicted and processed by a pre-trained multi-task prediction model. The multi-task prediction model can be a neural network model, which can be obtained by training with case samples collected in the litigation scene. For example, by pre-collecting text information describing case information of case samples in the litigation scene, litigation prediction information of case samples, and legal information samples of case samples associated with the litigation prediction information of case samples, a multi-task learning algorithm strategy is selected, and the sub-neural network model is trained to obtain a multi-task prediction model, wherein the sub-neural network model can be an initially established detection model. The legal information of the case obtained by predicting and processing the task prediction model of this embodiment is highly correlated with the litigation prediction information, wherein the legal information can be laws and regulations, such as legal clauses, and the litigation prediction information of the case is also the result of the lawsuit, wherein the legal information of the case and the litigation prediction information are highly correlated.
[0051] Optionally, when the sub-neural network model is trained through the text information of the case information of the case sample, the litigation prediction information of the case sample and the legal information of the case sample associated with the litigation prediction information of the case sample, the text information of the case information of the case sample, the litigation prediction information of the case sample and the legal information of the case sample associated with the litigation prediction information of the case sample can be preprocessed according to the distribution consistency algorithm, denoising, sampling and other algorithms, and then feature extraction, feature transformation, feature normalization, feature combination and the like are performed from the preprocessed data to obtain features for training the sub-neural network model, and the features are further processed through optimization algorithms, hypothesis functions, loss functions, decision boundaries, convergence speeds, iteration strategies and the like to obtain labels of the relationship between the text information of the case and the legal information and litigation prediction information of the case, and then the parameters of the sub-neural network model are determined through the labels, and the multi-task prediction model is determined through the parameters. This embodiment can also perform cross-validation, evaluation targets, overfitting, underfitting and other evaluations on the multi-task prediction model finally obtained by training, so as to obtain a model for predicting the legal information and litigation prediction information of different cases.
[0052] Step S206: output the legal information and litigation prediction information of the case.
[0053] In the technical solution provided in the above step S206 of the present invention, after predicting and processing the target text information based on the multi-task prediction model and predicting the legal information and litigation prediction information of the case, the legal information and litigation prediction information of the case are output, and the legal information and litigation prediction information can be output to the display interface as a reference result for the case trial, so that the case mediator can formulate a case mediation strategy based on the predicted legal information and litigation prediction information. This embodiment can also be flexibly applied to other judicial assistance scenarios, for example, the predicted legal information and litigation prediction information are provided to lawyers and parties for reference, thereby assisting the trial of the case and improving the efficiency of case handling.
[0054] Through the above steps S202 to S206, the target text information is obtained, wherein the target text information is used to describe the case information of the case in the litigation scenario; the target text information is predicted and processed based on the multi-task prediction model to predict the legal information and litigation prediction information of the case, wherein the multi-task prediction model is trained by the text information of the case information of the case sample used to describe the litigation scenario and the legal information sample of the case sample; and the legal information and litigation prediction information of the case are output. In other words, through the pre-trained multi-task prediction model, the litigation prediction information and the legal information of the submitted case are jointly predicted, so as to achieve the purpose of assisting the trial of the case, realize the improvement of the efficiency of handling the case in the litigation scenario, and then solve the technical problem of low efficiency in handling the case in the litigation scenario due to the inability to predict the litigation result of the case and the legal information used in the litigation scenario.
[0055] As an optional implementation, in step S204, before processing the target text information based on the multi-task prediction model to obtain the legal information and litigation prediction information of the case, the method also includes: training with the text information of the legal information samples and the case samples to obtain the first task prediction model, wherein the multi-task prediction model includes the first task prediction model, and the first task prediction model is used to predict the legal information of the case; training with the legal information samples, the text information of the case samples and the legal information output by the first task prediction model to obtain the second task prediction model, wherein the multi-task prediction model includes the second task prediction model, and the second task prediction model is used to predict the litigation prediction information of the case.
[0056] In this embodiment, before processing the target text information based on the multi-task prediction model to obtain the legal information and litigation prediction information of the case, it is necessary to train the multi-task prediction model. Since the two tasks of the legal information of the case and the litigation prediction information are highly related, this embodiment can adopt a multi-task learning algorithm strategy to train a first task prediction model for predicting the legal information of the case and a second task prediction model for predicting the litigation prediction information of the case. The first task prediction model and the second task prediction model of this embodiment are also related and can share features to improve the accuracy of the results of the two tasks of the legal information of the case and the litigation prediction information.
[0057] Optionally, this embodiment trains through legal information samples of case samples and text information of case samples to obtain a first task prediction model, wherein the legal information samples may be the formal text of legal clauses, the text information of case samples may be case information of case samples, and may be text materials such as indictments, replies, and judgments of different cases, and the more text materials the better; this embodiment may use the legal information output by the first task prediction model as input information for training the second task prediction model, thereby obtaining a second task prediction model through training through the above-mentioned legal information samples, text information of case samples, and legal information output by the first task prediction model, thereby ensuring that the trained second task prediction model is highly correlated with the first task prediction model.
[0058] As an optional implementation, the method also includes: encoding the legal information sample to obtain a first encoding result; encoding the text information of the case sample to obtain a second encoding result; encoding the discrete features of the text information of the case sample to obtain a third encoding result; training through the legal information sample and the text information of the case sample to obtain the first task prediction model includes: training through the first encoding result, the second encoding result and the third encoding result to obtain the first task prediction model, wherein the first encoding result, the second encoding result and the third encoding result are the results of the input layer of the first task prediction model; training through the legal information sample, the text information of the case sample and the legal information output by the first task prediction model to obtain the second task prediction model includes: training through the first encoding result, the second encoding result, the third encoding result and the legal information output by the first task prediction model to obtain the second task prediction model, wherein the first encoding result, the second encoding result and the third encoding result are the results of the input layer of the second task prediction model.
[0059] In this embodiment, the legal information sample can be encoded to obtain a first encoding result, which is the result in the input layer. The legal information sample can be encoded in the legal information encoding layer of the input layer to obtain the first encoding result; this embodiment can also encode the text information of the case sample to obtain a second encoding result, which is also the result of the input layer. The text information of the case sample can be encoded in the case text information encoding layer of the input layer to obtain the second encoding result; this embodiment can also obtain discrete features of the text information of the case sample, encode the discrete features, and obtain a third encoding result. The third encoding result is also the result of the input layer. The discrete features of the text information can be encoded in the discrete feature input layer of the input layer to obtain the third encoding result.
[0060] Optionally, when this embodiment trains the first task prediction model through text information of legal information samples and case samples, the first encoding result, the second encoding result and the third encoding result are used as the results of the input layer of the first task prediction model to train the first task prediction model.
[0061] The input layer of the first task prediction model and the input layer of the second task prediction model of this embodiment can have the same encoding method. Optionally, when this embodiment is trained through legal information samples, text information of case samples and legal information output by the first task prediction model to obtain the second task prediction model, the first encoding result, the second encoding result and the third encoding result can be used as the results of the input layer of the second task prediction model to train the second task prediction model.
[0062] As an optional implementation, encoding the legal information sample to obtain a first encoding result includes: segmenting the legal information sample to obtain a first word vector sequence; and determining the first word vector sequence as the first encoding result.
[0063] In this embodiment, the legal information sample can be encoded in the legal information encoding layer to obtain a first encoding result, and the legal information sample can be segmented to obtain multiple segmented words. This embodiment can use any segmentation tool to segment the legal information sample, and there is no limitation here. This embodiment can convert each word into a word vector, and the word vector can be trained by the corpus after segmentation. The word vector can be trained by the CBOW model, Skip-gram neural network model, ELMo model, bert model, etc. Preferably, the word vector of this embodiment is pre-trained using the ELMo model. Compared with the previous fixed word vectors, the ELMo model takes into account the context information, and the word vectors of different contexts are different, so as to solve the problem of polysemous words. After training the word vector with the corpus after segmentation, this embodiment can encode all the legal terms in the legal information sample into a first word vector sequence, and then determine the first word vector sequence as the first encoding result.
[0064] As an optional implementation, encoding the text information of the case sample to obtain a second encoding result includes: segmenting the text information of the case sample to obtain a second word vector sequence; encoding the second word vector sequence to obtain a second encoding result.
[0065] In this embodiment, the text information of the case sample can be encoded in the case text information encoding layer to obtain a second encoding result, the text information of the case sample can be segmented and converted into a second word vector sequence, and then the second word vector sequence can be encoded. The second word vector can be encoded through the cudnn / gru model and the Self-attention mechanism to obtain a second encoding result, which is the final case text encoding.
[0066] As an optional implementation, encoding the discrete features of the text information of the case sample to obtain a third encoding result includes: encoding the discrete features through a multi-layer perceptron to obtain a third encoding result with continuous semantics.
[0067] This embodiment can encode the discrete features of the text information of the case sample at the discrete feature input layer, can use a multi-layer perceptron to encode and convert the discrete features, and can convert the discrete feature vectors into a continuous semantic space to obtain a third encoding result, which is more suitable for the training of a deep learning model.
[0068] As an optional implementation, training is performed through the first encoding result, the second encoding result and the third encoding result to obtain the first task prediction model, including: fusion encoding the second encoding result and the third encoding result to obtain a fourth encoding result; splicing the fourth encoding result and the first encoding result to obtain a fifth encoding result; training is performed through the fifth encoding result to obtain the first task prediction model, wherein the fifth encoding result is the result of the intermediate layer of the first task prediction model.
[0069] In this embodiment, when the first task prediction model is obtained by training the first coding result, the second coding result and the third coding result, the second coding result and the third coding result can be fused and encoded at the legal article coding layer. For example, the second coding result of the case text information coding layer and the third coding result of the discrete feature input layer can be fused and encoded by the self-attention and highway strategies to obtain the third coding result. After the second coding result and the third coding result are fused and encoded to obtain the fourth coding result, the fourth coding result and the first coding result can be spliced, for example, the fourth coding result and the first coding result of the legal article information coding layer are spliced to obtain the fifth coding result. The fifth coding result is the result of the intermediate layer of the first task prediction model, and the intermediate layer can be the legal article coding layer. That is, in this embodiment, the case text information coding layer and the discrete feature input layer are first fused and encoded by the self-attention and highway strategies, and then spliced with the legal article information coding layer to finally obtain the legal article coding layer.
[0070] As an optional implementation, training through the fifth encoding result to obtain the first task prediction model includes: determining the output layer of the first task prediction model through the fifth encoding result and the objective function to obtain the first task prediction model.
[0071] In this embodiment, when the first task prediction model is obtained by training through the fifth encoding result, the output layer of the first task prediction model can be determined through the fifth encoding result and the objective function, thereby obtaining the first task prediction model. Optionally, the output layer of the first task prediction model of this embodiment can be a legal provision output layer, and the objective function can be a Softmax function. The legal provision information of the legal provision output layer is obtained by predicting the fifth encoding result of the legal provision encoding layer and the Softmax function.
[0072] As an optional implementation method, the second task prediction model is obtained by training with the first encoding result, the second encoding result, the third encoding result and the legal information output by the first task prediction model, including: classifying the first encoding result by a pre-constructed classifier of the legal information sample to obtain a classification result; splicing the classification result, the intermediate classification result used to obtain the classification result, the second encoding result and the third encoding result to obtain a spliced result; fusion encoding the spliced result to obtain a sixth encoding result, wherein the sixth encoding result is the result of the intermediate layer of the second task prediction model; and training with the sixth encoding result and the legal information output by the first task prediction model to obtain the second task prediction model.
[0073] In this embodiment, when the legal information output by the first encoding result, the second encoding result, the third encoding result and the first task prediction model is trained to obtain the second task prediction model, a classifier of the legal information sample can be constructed first, and the classifier can correspond to the legal label encoding layer, can be an attribute label classifier, can be a multi-label classifier, and can be constructed manually. In this embodiment, the first encoding result can be classified by the constructed classifier to obtain a classification result, and the classification result, the intermediate classification result used to obtain the classification result, the second encoding result and the third encoding result can be spliced to obtain a spliced result. Optionally, this embodiment determines the legal article result coding layer according to the intermediate result used to obtain the classification result, and splices the classification result of the legal label coding layer, the intermediate classification result of the legal article result coding layer, the second coding result of the case text information coding layer, and the third coding result of the discrete feature input layer. That is, the legal label coding layer, the legal article result coding layer, the case text information coding layer and the discrete feature input layer are spliced to obtain the splicing result, and then the splicing result is fusion-encoded. The highway strategy can be used to fusion-encode the splicing result to obtain a sixth coding result. The sixth coding result is the result of the intermediate layer of the second task prediction model, and the intermediate layer can be the claim result coding layer.
[0074] After the splicing results are fused and encoded to obtain the sixth encoding result, the sixth encoding result and the legal information output by the first task prediction model can be used for training to obtain the second task prediction model. That is, the predicted legal information can be used as input information for the predicted litigation prediction information. Model training is added to the last layer to obtain the second task prediction model.
[0075] As an optional implementation method, training is performed by using the sixth encoding result and the legal information output by the first task prediction model to obtain the second task prediction model, including: determining the output layer of the second task prediction model by using the sixth encoding result and the objective function; training the output layer of the second task prediction model by using the legal information output by the first task prediction model to obtain the second task prediction model.
[0076] In this embodiment, when the second task prediction model is obtained by training the legal information output by the sixth encoding result and the first task prediction model, the output layer of the second task prediction model can also be determined by the sixth encoding result and the objective function, thereby obtaining the second task prediction model. Optionally, the output layer of the second task prediction model of this embodiment can be the output layer of the litigation prediction information, and the objective function can be the Softmax function, and the litigation prediction information is obtained by the sixth encoding result of the litigation result encoding layer and the Softmax function prediction.
[0077] As an optional implementation method, the output layer of the second task prediction model is trained with the legal information output by the first task prediction model to obtain the second task prediction model, including: obtaining a first loss function corresponding to the legal information output by the first task prediction model; obtaining a second loss function corresponding to the output layer of the second task prediction model; and training the output layer of the second task prediction model with the first loss function and the second loss function to obtain the second task prediction model.
[0078] In this embodiment, the legal information output by the first task prediction model corresponds to the first loss function, and the output layer of the second task prediction model corresponds to the second loss function. The goal of model optimization in this embodiment is to combine the first loss function and the second loss function, train the output layer of the second task prediction model, and obtain the second task prediction model, thereby ensuring that the first task prediction model and the second task prediction model are highly correlated.
[0079] As an optional implementation, the method also includes: extracting statistical features of words and vector features of words from the text information of case samples; training through legal information samples and text information of case samples to obtain a first task prediction model, including: training through legal information samples, statistical features of words and vector features of words to obtain a first task prediction model; training through legal information samples, text information of case samples and legal information output by the first task prediction model to obtain a second task prediction model, including: training through legal information samples, statistical features of words, vector features of words and legal information output by the first task prediction model to obtain a second task prediction model.
[0080] This embodiment can also establish a text preprocessing and pretraining model, which is used to preprocess the target text information, so as to facilitate the training of the multi-task prediction model, for example, preprocessing text materials such as the indictment text, the defense text, and the evidence text, among which the indictment text and the defense text are the most important case analysis materials, followed by the evidence text submitted by the plaintiff and the defendant, which is a very important feature source for the case description and defense text. Optionally, this embodiment extracts the statistical features of words and the vector features of words from the text information of the case sample, wherein the statistical features can be TF-IDF.
[0081] Optionally, the preprocessing steps of this embodiment include: word segmentation of text information, any mainstream word segmentation tool can be used, and a legal dictionary needs to be added; part-of-speech tagging, that is, marking each word segment with a part-of-speech mark, such as verbs, nouns, conjunctions, auxiliary words, etc.; entity recognition, such as entity recognition of names, place names, and institution names, which is not limited here; mining statistical features such as TF-TDF and TF; word vector pre-training, when this embodiment performs word vector pre-training, word vectors can be trained for different cases separately, and text information such as indictments, defense statements, and judgments of different cases can be used, wherein the more text information, the better. Word segmentation is performed on text information of different cases, and word vectors are trained through the corpus after word segmentation. CBOW model, Skip-gram model, ELMo model, bert model, etc. can be used to train word vectors, which are not limited here. Preferably, this embodiment uses the ELMo model to train word vectors to solve problems such as polysemous words.
[0082] After extracting the statistical features of words and the vector features of words from the text information of case samples, when training with legal information samples and the text information of case samples to obtain the first task prediction model, the first task prediction model can be obtained by training with legal information samples, the statistical features of words and the vector features of words. When training with legal information samples, the text information of case samples and the legal information output by the first task prediction model to obtain the second task prediction model, the second task prediction model can be obtained by training with legal information samples, the statistical features of words, the vector features of words and the legal information output by the first task prediction model.
[0083] exist Figure 1 Under the operating environment shown, this application also provides Figure 3a It should be noted that the text information processing method of this embodiment can be Figure 1 The illustrated embodiment is executed by a mobile terminal.
[0084] Figure 3a FIG. 1 is a flowchart of another text information processing method according to an embodiment of the present invention. Figure 3a As shown, the method comprises the following steps:
[0085] Step S302, displaying target text information on a text display interface, wherein the target text information is used to describe case information of a case in a litigation scenario.
[0086] In the technical solution provided in step S302 of the present invention, the case in the litigation scenario may be a case that requires dispute mediation. The case information of this embodiment may include text materials such as the complaint text, the defense text, the legal and regulatory text, the evidence text, and the judgment text required in the process of hearing the case. The target text information is displayed on the text display interface, and the target text information is used to describe the case information of the case in the above litigation scenario, and may include case information.
[0087] Step S304, displaying the legal information and litigation prediction information of the case on the text display interface, wherein the legal information and litigation prediction information of the case are the results of predictive processing of the target text information based on the multi-task prediction model.
[0088] In the technical solution provided in the above step S304 of the present invention, after the target text information is displayed on the text display interface, the legal information and litigation prediction information of the case are displayed on the text display interface, wherein the multi-task prediction model is trained by using text information of case information of case samples used to describe the litigation scenario and legal information samples of the case samples.
[0089] In this embodiment, after the target text information is displayed on the text display interface, the target text information can be predicted and processed by a pre-trained multi-task prediction model, which can be a neural network model and can be obtained by training case samples collected in litigation scenarios. For example, by pre-collecting text information describing case information of case samples in litigation scenarios, litigation prediction information of case samples, and legal information samples of case samples associated with litigation prediction information of case samples, a multi-task learning algorithm strategy is selected, and a sub-neural network model is trained to obtain a multi-task prediction model, wherein the sub-neural network model can be an initially established detection model. The legal information of the case obtained by predicting and processing the task prediction model of this embodiment is highly correlated with the litigation prediction information.
[0090] Specific as Figure 3b As shown, Figure 3bThis is a schematic diagram of the interaction in another text information processing method according to an embodiment of the present invention. The user can drag text materials such as the complaint text, the defense text, the legal and regulatory text, the evidence text and the judgment text required in the case trial process into the "Add" text box, and click the "Information Processing" button to perform litigation prediction on the target text information, and finally generate legal information and litigation prediction information related to the case, wherein the litigation prediction information may include the probability of litigation success, risk information, potential problems, the success rate of appeal after losing the case, or related information on the appeal matter.
[0091] After the target text information is predicted and processed based on the multi-task prediction model, and the legal information and litigation prediction information of the case are predicted, the legal information and litigation prediction information of the case are displayed on the text display interface, so that the predicted legal information and litigation prediction information are used as reference results for the case trial, so that the case mediator can formulate a case mediation strategy based on the predicted legal information and litigation prediction information. This embodiment can also be flexibly applied to other judicial assistance scenarios, for example, the predicted legal information and litigation prediction information are provided to lawyers and parties for reference, thereby assisting the trial of the case and improving the efficiency of case handling.
[0092] exist Figure 1 Under the operating environment shown, this application also provides Figure 4 It should be noted that the case information processing method of this embodiment can be Figure 1 The illustrated embodiment is executed by a mobile terminal.
[0093] Figure 4 FIG. 1 is a flowchart of a method for processing case information according to an embodiment of the present invention. Figure 4 As shown, the method comprises the following steps:
[0094] Step S402, obtaining case information of the case.
[0095] In the technical solution provided in the above step S402 of the present invention, the case information of the case includes at least one of the following: the text of the complaint, the text of the defense, the text of laws and regulations, the text of evidence and the text of the judgment.
[0096] In this embodiment, the case information may include text materials such as the indictment text, the defense text, the laws and regulations text, the evidence text and the judgment text required in the process of hearing the case.
[0097] Step S404, predicting and processing the case information of the case based on the multi-task prediction model, predicting the legal information and litigation prediction information of the case, wherein the multi-task prediction model is a neural network model, which is used to encode different types of files in the case information of the case.
[0098] In the technical solution provided in the above step S404 of the present invention, after obtaining the case information of the case, the case information of the case is predicted and processed based on the multi-task prediction model to predict the legal information and litigation prediction information of the case, wherein the multi-task prediction model is a neural network model, which is used to encode different types of files in the case information of the case.
[0099] In this embodiment, after obtaining the case information of the case, the case information of the case can be predicted and processed by a pre-trained multi-task prediction model. The multi-task prediction model can be a neural network model, which can encode different types of files in the case information of the case to obtain the legal information and litigation prediction information of the case.
[0100] Optionally, when the multi-task prediction model encodes different types of files in the case information of a case, it can segment the text in the file at the legal information encoding layer of the multi-task prediction model to obtain a word vector sequence, segment the text in the file at the case text information encoding layer, convert it into a word vector sequence, and then encode the word vector sequence through cudnn / gru+Self-attention to obtain the case text encoding, and use a multi-layer perceptron network to encode and convert the text in the file at the discrete feature input layer to obtain the legal information and litigation prediction information of the case.
[0101] Optionally, this embodiment pre-collects text information describing case information of case samples in litigation scenarios, litigation prediction information of case samples, and legal information samples of case samples associated with litigation prediction information of case samples, selects a multi-task learning algorithm strategy, trains the sub-neural network model, and obtains a multi-task prediction model, wherein the sub-neural network model may be an initially established detection model. The legal information of the case obtained by predicting and processing the task prediction model of this embodiment is highly correlated with the litigation prediction information, wherein the legal information may be laws and regulations, such as legal clauses, and the litigation prediction information of the case is also the result of the claim, wherein the legal information of the case is highly correlated with the litigation prediction information.
[0102] Step S406: output the legal information and litigation prediction information of the case.
[0103] In the technical solution provided in the above step S406 of the present invention, after predicting and processing the case information of the case based on the multi-task prediction model and predicting the legal information and litigation prediction information of the case, the legal information and litigation prediction information of the case are output. The legal information and litigation prediction information of the case can be output to a display interface, thereby serving as a reference result for the trial of the case.
[0104] As an optional implementation, in step S404, before predicting and processing the case information based on the multi-task prediction model to obtain the legal information and litigation prediction information of the case, the method also includes: training with legal information samples of case samples and case information of case samples to obtain a first task prediction model, wherein the multi-task prediction model includes a first task prediction model, and the first task prediction model is used to predict the legal information of the case; training with legal information samples, case information of case samples and legal information output by the first task prediction model to obtain a second task prediction model, wherein the multi-task prediction model includes a second task prediction model, and the second task prediction model is used to predict the litigation prediction information of the case.
[0105] In this embodiment, before processing case information based on the multi-task prediction model and predicting the legal information and litigation prediction information of the case, it is necessary to train the multi-task prediction model. Since the two tasks of the legal information of the case and the litigation prediction information are highly related, this embodiment can adopt a multi-task learning algorithm strategy to train a first task prediction model for predicting the legal information of the case and a second task prediction model for predicting the litigation prediction information of the case. The first task prediction model and the second task prediction model of this embodiment are also related and can share features to improve the accuracy of the results of the two tasks of the legal information of the case and the litigation prediction information.
[0106] Optionally, this embodiment obtains a first task prediction model by training with legal information samples of case samples and case information of case samples; this embodiment can use the legal information output by the first task prediction model as input information for training the second task prediction model, thereby obtaining a second task prediction model by training with the above-mentioned legal information samples, case information of case samples and legal information output by the first task prediction model, thereby ensuring that the trained second task prediction model is highly correlated with the first task prediction model.
[0107] As an optional implementation, the method also includes: encoding the legal information sample to obtain a first encoding result; encoding the case information of the case sample to obtain a second encoding result; encoding the discrete features of the case information of the case sample to obtain a third encoding result; training with the legal information sample and the case information of the case sample to obtain the first task prediction model includes: training with the first encoding result, the second encoding result and the third encoding result to obtain the first task prediction model, wherein the first encoding result, the second encoding result and the third encoding result are the results of the input layer of the first task prediction model; training with the legal information sample, the case information of the case sample and the legal information output by the first task prediction model to obtain the second task prediction model includes: training with the first encoding result, the second encoding result, the third encoding result and the legal information output by the first task prediction model to obtain the second task prediction model, wherein the first encoding result, the second encoding result and the third encoding result are the results of the input layer of the second task prediction model.
[0108] In this embodiment, the legal information sample can be encoded to obtain a first encoding result, which is the result in the input layer. The legal information sample can be encoded in the legal information encoding layer of the input layer to obtain the first encoding result; this embodiment can also encode the case information of the case sample to obtain a second encoding result, which is also the result of the input layer. The case information of the case sample can be encoded in the case text information encoding layer of the input layer to obtain the second encoding result; this embodiment can also obtain discrete features of the text information of the case sample, encode the discrete features, and obtain a third encoding result. The third encoding result is also the result of the input layer. The discrete features of the case information can be encoded in the discrete feature input layer of the input layer to obtain the third encoding result.
[0109] Optionally, when this embodiment is trained through the case information of legal information samples and case samples to obtain the first task prediction model, the first encoding result, the second encoding result and the third encoding result are used as the results of the input layer of the first task prediction model to train the first task prediction model.
[0110] The input layer of the first task prediction model and the input layer of the second task prediction model of this embodiment can have the same encoding method. Optionally, when this embodiment is trained through legal information samples, text information of case samples and legal information output by the first task prediction model to obtain the second task prediction model, the first encoding result, the second encoding result and the third encoding result can be used as the results of the input layer of the second task prediction model to train the second task prediction model.
[0111] This embodiment uses a pre-trained multi-task prediction model to jointly predict litigation prediction information and legal information for submitted cases, thereby achieving the purpose of assisting the trial of cases, improving the efficiency of handling cases in litigation scenarios, and further solving the technical problem of inefficient handling of cases in litigation scenarios due to the inability to predict the litigation results of cases and the legal information used in litigation scenarios.
[0112] Example 2
[0113] The technical solution of the present invention is illustrated below in conjunction with preferred implementation modes.
[0114] This embodiment can be applied to the research of integrated convenient service technology and equipment for the entire litigation process. It is a cross-network online mediation technology that supports multi-party access, and aims to use artificial intelligence technology to improve the mediation efficiency of people's mediators and judges.
[0115] This embodiment proposes a solution for jointly predicting litigation prediction information and legal provisions based on the case information submitted by the parties. The predicted litigation prediction information and predicted legal provisions can be used as reference results for mediators to facilitate their formulation of mediation strategies. It can also be flexibly applied to other types of judicial scenarios, for example, as a reference for lawyers and parties.
[0116] Figure 5 Schematic diagram of a text information processing system according to an embodiment of the present invention. Figure 5 As shown, the text information processing system 50 may include: a text preprocessing and pretraining module 51 and a neural network multi-task prediction module 52.
[0117] The text preprocessing and pretraining module 51 is used to preprocess the text information such as the indictment, the defense, and the evidence, so as to facilitate the use of the subsequent neural network multi-task prediction module 52. Among them, the indictment and the defense in the text information are the most important case analysis materials, followed by the evidence submitted by the plaintiff and the defendant in the text information. The text describing and defending the case is a very important source of features. This embodiment mainly extracts two types of features for the above text information, including statistical features of words, such as TF-IDF, and can also include vector features.
[0118] Optionally, the preprocessing steps of this embodiment include: word segmentation of text information, any mainstream word segmentation tool can be used, and a legal dictionary needs to be added; part-of-speech tagging, that is, marking each word with a part-of-speech tag, such as verbs, nouns, conjunctions, auxiliary words, etc.; entity recognition, such as entity recognition of names of people, places, and institutions, which is not limited here; and statistical features such as TF-TDF and TF can also be mined.
[0119] When pre-training word vectors, this embodiment can train word vectors for different cases, and can use text information such as indictments, defense statements, and judgments of different cases, where the more text information, the better. The text information of different cases is segmented, and the word vectors are trained through the corpus after segmentation. The CBOW model, Skip-gram model, ELMo model, bert model, etc. can be used, and no restrictions are made here. Optionally, this embodiment uses the ELMo model.
[0120] The neural network multi-task prediction module of this embodiment is introduced below.
[0121] In this embodiment, the legal terms and the claim results are two highly related tasks. Therefore, a multi-task learning algorithm strategy is selected to enable the two tasks to share features and improve the results of the two tasks.
[0122] Figure 6 is a schematic diagram of an architecture of a multi-task model according to an embodiment of the present invention, Figure 7 Schematic diagram of a network structure of a multi-task model according to an embodiment of the present invention. Figure 6 As shown, the input layer includes the law information encoding layer, the case text information encoding layer, and the discrete feature input layer; the middle layer includes the law encoding layer and the litigation result encoding layer; and the output layer includes the law prediction results and the litigation prediction results.
[0123] The legal information coding layer of this embodiment corresponds to Figure 7 The law information encoding layer (law infoembedding) in Figure 7 : is a schematic diagram of the network structure of a multi-task model according to an embodiment of the present invention. In the legal provision information encoding layer, the formal text of the legal provisions is segmented, and then each word is converted into a pre-trained word vector. All legal provisions can be encoded into a word vector sequence. The word vector of this embodiment can be pre-trained using the ELMo model. The advantage is that compared with the previous fixed word vectors, the ELMo model takes into account the context information. The word vectors of different contexts are different, which can solve problems such as polysemous words.
[0124] Case text information encoding layer, corresponding to Figure 7 The case text information encoding layer (arti-level embedding) in the embodiment can first segment the case text and convert it into a word vector sequence. Then the word vector sequence is encoded through cudnn / gru+self-attention to finally obtain the case text encoding.
[0125] Discrete feature input layer, corresponding to Figure 7In the discrete feature input layer (discrete feature), this embodiment uses a multi-layer perceptron network to encode and transform the discrete features of the case, which can convert the discrete feature vector into a continuous semantic space, making it more suitable for the training of the deep learning model.
[0126] Legal article coding layer, corresponding to Figure 7 The input of this part includes three parts of the input layer. Optionally, this embodiment uses the self-attention and highway strategies to fuse the case text information encoding layer and the discrete feature input layer to obtain the encoding result, and then concatenates the encoding result with the law information encoding layer to finally obtain the law encoding layer.
[0127] The application result coding layer corresponds to Figure 7 The first step is to construct an attribute label classifier (attr classifier), which is a multi-label classifier, where the label refers to the legal label, which can be constructed manually. Optionally, the attribute label classifier is constructed through the attention mechanism. The legal label encoding layer is obtained; the second step is to splice the legal label encoding layer (attr embedding), the legal result encoding layer (law result embedding, obtained from the intermediate result of the legal classifier), the discrete feature input layer (discrete embedding), and the case text information encoding layer, and then fuse the encoding through the highway strategy to obtain the claim result encoding layer (accu embedding).
[0128] The output layer, the legal article output layer is predicted by the legal article encoding layer (law embedding) + Softmax function, and the claim result is predicted by the claim result encoding layer (accu embedding) + Softmax function. The two tasks, the legal article prediction result and the claim prediction result, have the same encoding method at the input layer, and the legal article prediction result is used as the input information of the claim result, and the model training is added to the last layer. The optimization goal of the model in this embodiment is to combine the loss functions of the two subtasks to obtain the final loss function, and then ensure the high correlation of the model results through the above two methods.
[0129] This embodiment uses a pre-trained multi-task prediction model to jointly predict litigation prediction information and legal information for submitted cases, wherein the litigation prediction information and the legal information are highly correlated, thereby achieving the purpose of assisting the trial of the case and improving the efficiency of handling cases in litigation scenarios, thereby solving the technical problem of inefficient handling of cases in litigation scenarios due to the inability to predict the litigation results of the case and the legal information used in the litigation scenario.
[0130] Example 3
[0131] According to an embodiment of the present invention, there is also provided a method for implementing the above Figure 2 The text information processing method and the text information processing device are shown.
[0132] Figure 8 is a schematic diagram of a text information processing device according to an embodiment of the present invention. Figure 8 As shown, the text information processing device 80 includes: a first acquisition unit 81, a first processing unit 82 and a first output unit 83.
[0133] The first acquisition unit 81 is used to acquire target text information, wherein the target text information is used to describe case information of a case in a litigation scenario.
[0134] The first processing unit 82 is used to perform prediction processing on the target text information based on the multi-task prediction model to predict the legal information and litigation prediction information of the case, wherein the multi-task prediction model is trained by using text information of case information of case samples used to describe the litigation scenario and legal information samples of the case samples.
[0135] The first output unit 83 is used to output the legal information and litigation prediction information of the case.
[0136] It should be noted that the first acquisition unit 81, the first processing unit 82 and the first output unit 83 correspond to steps S202 to S206 of Example 1, and the three modules and the corresponding steps implement the same examples and application scenarios, but are not limited to the contents disclosed in the above-mentioned Example 1. It should be noted that the above-mentioned modules, as part of the device, can be run in the computer terminal 10 provided in Example 1.
[0137] According to an embodiment of the present invention, another method for implementing the above Figure 3a The text information processing method and the text information processing device are shown. Fig. 9 FIG. 2 is a schematic diagram of another text information processing device according to an embodiment of the present invention. Fig. 9As shown, the text information processing 90 includes: a first display unit 91 and a second display unit 92.
[0138] The first display unit 91 is used to display target text information on a text display interface, wherein the target text information is used to describe case information of a case in a litigation scenario.
[0139] The second display unit 92 is used to display the legal information and litigation prediction information of the case in a text display interface, wherein the legal information and litigation prediction information of the case are the result of predictive processing of the target text information based on a multi-task prediction model, and the multi-task prediction model is trained by using text information of case information of case samples in litigation scenarios and legal information samples of the case samples.
[0140] It should be noted that the first display unit 91 and the second display unit 92 correspond to step S302 and step S304 of embodiment 1, and the examples and application scenarios implemented by the two modules and the corresponding steps are the same, but are not limited to the contents disclosed in the above embodiment 1. It should be noted that the above modules, as part of the device, can be run in the computer terminal 10 provided in embodiment 1.
[0141] According to an embodiment of the present invention, another method for implementing the above Figure 4 A case information processing device and a case information processing method are shown. Fig.10 Schematic diagram of a case information processing device according to an embodiment of the present invention. Fig.10 As shown, the case information processing device 100 includes: a second acquisition unit 101, a second processing unit 102 and a second output unit 103.
[0142] The second acquisition unit 101 is used to acquire case information of a case, wherein the case information of a case includes at least one of the following: a text of an indictment, a text of a defense, a text of laws and regulations, a text of evidence, and a text of a judgment.
[0143] The second processing unit 102 is used to predict the case information of the case based on the multi-task prediction model to predict the legal information and litigation prediction information of the case, wherein the multi-task prediction model is a neural network model, which is used to encode different types of files in the case information of the case.
[0144] The second output unit 103 is used to output the legal information and litigation prediction information of the case.
[0145] This embodiment uses a pre-trained multi-task prediction model to jointly predict litigation prediction information and legal information for submitted cases, thereby achieving the purpose of assisting the trial of cases, improving the efficiency of handling cases in litigation scenarios, and further solving the technical problem of inefficient handling of cases in litigation scenarios due to the inability to predict the litigation results of cases and the legal information used in litigation scenarios.
[0146] Example 4
[0147] An embodiment of the present invention may provide a mobile terminal, which may be any computer terminal device in a computer terminal group.
[0148] Optionally, in this embodiment, the mobile terminal may be located in at least one network device among a plurality of network devices of the computer network.
[0149] In this embodiment, the above-mentioned mobile terminal can execute the program code of the following steps in the text information processing method of the application: obtaining target text information, wherein the target text information is used to describe case information of a case in a litigation scenario; performing predictive processing on the target text information based on a multi-task prediction model to predict legal information and litigation prediction information of the case, wherein the multi-task prediction model is trained by text information used to describe case information of case samples in the litigation scenario and legal information samples of the case samples; and outputting the legal information and litigation prediction information of the case.
[0150] Optionally, Fig.11 is a structural block diagram of a mobile terminal according to an embodiment of the present invention. Fig.11 As shown, the mobile terminal A may include: one or more (only one is shown in the figure) processors 112 , a memory 114 , and a transmission device 116 .
[0151] Among them, a transmission device is connected to the processor and is used to obtain target text information, wherein the target text information is used to describe case information of a case in a litigation scenario; and a memory is connected to the processor and is used to provide the processor with instructions for processing the following processing steps: predicting the target text information based on a multi-task prediction model to predict legal information and litigation prediction information of the case, wherein the multi-task prediction model is trained by text information of case information of case samples used to describe case information in a litigation scenario and legal information samples of the case samples; and outputting the legal information and litigation prediction information of the case.
[0152] Among them, the memory can be used to store software programs and modules, such as program instructions / modules corresponding to the text information processing method and device in the embodiment of the present invention. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory, that is, realizing the above-mentioned text information processing method. The memory may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory may further include a memory remotely arranged relative to the processor, and these remote memories can be connected to the mobile terminal A via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.
[0153] The processor can also call the information and application programs stored in the memory through the transmission device to perform the following steps: before processing the target text information based on the multi-task prediction model to obtain the legal information and litigation prediction information of the case, training is performed through the legal information samples and the text information of the case samples to obtain the first task prediction model, wherein the multi-task prediction model includes the first task prediction model, and the first task prediction model is used to predict the legal information of the case; training is performed through the legal information samples, the text information of the case samples and the legal information output by the first task prediction model to obtain the second task prediction model, wherein the multi-task prediction model includes the second task prediction model, and the second task prediction model is used to predict the litigation prediction information of the case.
[0154] Optionally, the processor also executes the program code of the following steps: encoding the legal information sample to obtain a first encoding result; encoding the text information of the case sample to obtain a second encoding result; encoding the discrete features of the text information of the case sample to obtain a third encoding result; training with the first encoding result, the second encoding result and the third encoding result to obtain a first task prediction model, wherein the first encoding result, the second encoding result and the third encoding result are the results of the input layer of the first task prediction model; training with the first encoding result, the second encoding result, the third encoding result and the legal information output by the first task prediction model to obtain a second task prediction model, wherein the first encoding result, the second encoding result and the third encoding result are the results of the input layer of the second task prediction model.
[0155] Optionally, the processor further executes the program code of the following steps: segmenting the legal information sample to obtain a first word vector sequence; and determining the first word vector sequence as a first encoding result.
[0156] Optionally, the processor further executes the program code of the following steps: segmenting the text information of the case sample to obtain a second word vector sequence; encoding the second word vector sequence to obtain a second encoding result.
[0157] Optionally, the processor further executes the program code of the following steps: encoding the discrete features through a multi-layer perceptron to obtain a third encoding result with continuous semantics.
[0158] Optionally, the processor also executes the program code of the following steps: fusing the second encoding result and the third encoding result to obtain a fourth encoding result; concatenating the fourth encoding result and the first encoding result to obtain a fifth encoding result; training through the fifth encoding result to obtain the first task prediction model, wherein the fifth encoding result is the result of the intermediate layer of the first task prediction model.
[0159] Optionally, the processor further executes the program code of the following steps: determining the output layer of the first task prediction model through the fifth encoding result and the objective function to obtain the first task prediction model.
[0160] Optionally, the processor also executes the program code of the following steps: classifying the first coding result by a pre-constructed classifier of the legal information sample to obtain a classification result; splicing the classification result, the intermediate classification result used to obtain the classification result, the second coding result and the third coding result to obtain a spliced result; fusion encoding the spliced result to obtain a sixth coding result, wherein the sixth coding result is the result of the intermediate layer of the second task prediction model; training by the sixth coding result and the legal information output by the first task prediction model to obtain the second task prediction model.
[0161] Optionally, the processor also executes the program code of the following steps: determining the output layer of the second task prediction model through the sixth encoding result and the objective function; training the output layer of the second task prediction model through the legal information output by the first task prediction model to obtain the second task prediction model.
[0162] Optionally, the processor also executes the program code of the following steps: obtaining a first loss function corresponding to the legal information output by the first task prediction model; obtaining a second loss function corresponding to the output layer of the second task prediction model; training the output layer of the second task prediction model by the first loss function and the second loss function to obtain the second task prediction model.
[0163] Optionally, the processor also executes the program code of the following steps: extracting statistical features of words and vector features of words from the text information of case samples; training through legal information samples, statistical features of words and vector features of words to obtain a first task prediction model; training through legal information samples, text information of case samples and legal information output by the first task prediction model to obtain a second task prediction model, including: training through legal information samples, statistical features of words, vector features of words and legal information output by the first task prediction model to obtain a second task prediction model.
[0164] Optionally, the above-mentioned mobile terminal can execute the program code of the following steps in the text information processing method of the application: displaying target text information on a text display interface, wherein the target text information is used to describe case information of a case in a litigation scenario; displaying legal information and litigation prediction information of the case on a text display interface, wherein the legal information and litigation prediction information of the case are the results of predictive processing of the target text information based on a multi-task prediction model, and the multi-task prediction model is trained by using text information of case information of case samples used to describe the litigation scenario and legal information samples of the case samples.
[0165] Optionally, the mobile terminal can execute the program code of the following steps in the case information processing method of the application: obtaining case information of the case, wherein the case information of the case includes at least one of the following: complaint text, defense text, legal and regulatory text, evidence text and judgment text; predicting the case information of the case based on a multi-task prediction model to obtain legal information and litigation prediction information of the case, wherein the multi-task prediction model is a neural network model, which is used to encode different types of files in the case information of the case; and outputting the legal information and litigation prediction information of the case.
[0166] The processor can call the information and application programs stored in the memory through the transmission device to execute the following steps: before predicting and processing the case information based on the multi-task prediction model to predict the legal information and litigation prediction information of the case, the method also includes: training with legal information samples of case samples and case information of case samples to obtain a first task prediction model, wherein the multi-task prediction model includes a first task prediction model, and the first task prediction model is used to predict the legal information of the case; training with legal information samples, case information of case samples and legal information output by the first task prediction model to obtain a second task prediction model, wherein the multi-task prediction model includes a second task prediction model, and the second task prediction model is used to predict the litigation prediction information of the case.
[0167] Optionally, the processor also executes the program code of the following steps: encoding the legal information sample to obtain a first encoding result; encoding the case information of the case sample to obtain a second encoding result; encoding the discrete features of the case information of the case sample to obtain a third encoding result; training with the first encoding result, the second encoding result and the third encoding result to obtain a first task prediction model, wherein the first encoding result, the second encoding result and the third encoding result are the results of the input layer of the first task prediction model; training with the first encoding result, the second encoding result, the third encoding result and the legal information output by the first task prediction model to obtain a second task prediction model, wherein the first encoding result, the second encoding result and the third encoding result are the results of the input layer of the second task prediction model.
[0168] According to an embodiment of the present invention, a text information processing method is provided, which obtains target text information, wherein the target text information is used to describe case information of a case in a litigation scenario; performs prediction processing on the target text information based on a multi-task prediction model to predict the legal information and litigation prediction information of the case, wherein the multi-task prediction model is trained by text information describing case information of case samples in the litigation scenario and legal information samples of the case samples; and outputs the legal information and litigation prediction information of the case. That is, through the pre-trained multi-task prediction model, the litigation prediction information and legal information of the submitted case are jointly predicted, thereby achieving the purpose of assisting the trial of the case, realizing the improvement of the efficiency of processing cases in the litigation scenario, and further solving the technical problem of low efficiency in processing cases in the litigation scenario due to the inability to predict the litigation results of the case and the legal information used in the litigation scenario.
[0169] It can be understood by those skilled in the art that Fig.11 The structure shown is for illustration only, and the mobile terminal may also be a smart phone (such as an Android phone, an iOS phone, etc.), a tablet computer, a handheld computer, a mobile Internet device (Mobile Internet Devices, MID), a PAD, and other terminal devices. Fig.11 It does not limit the structure of the above mobile terminal. For example, mobile terminal A may also include Fig.11 More or fewer components (such as network interfaces, display devices, etc.) shown in, or having Fig.11 Different configurations are shown.
[0170] A person of ordinary skill in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by instructing the hardware related to the terminal device through a program, and the program can be stored in a computer-readable storage medium, and the storage medium may include: a flash drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, etc.
[0171] Example 5
[0172] The embodiment of the present invention further provides a storage medium. Optionally, in this embodiment, the storage medium can be used to store the program code executed by the text information processing method provided in the above embodiment 1.
[0173] Optionally, in this embodiment, the above storage medium may be located in any computer terminal in a computer terminal group in a computer network, or in any mobile terminal in a mobile terminal group.
[0174] Optionally, in this embodiment, the storage medium is configured to store program code for executing the following steps: obtaining target text information, wherein the target text information is used to describe case information of a case in a litigation scenario; performing predictive processing on the target text information based on a multi-task prediction model to predict legal information and litigation prediction information of the case, wherein the multi-task prediction model is trained by training text information of case information of case samples used to describe case information in a litigation scenario and legal information samples of the case samples; and outputting the legal information and litigation prediction information of the case.
[0175] Optionally, the storage medium is also configured to store program code for executing the following steps: before processing the target text information based on the multi-task prediction model to obtain the legal information and litigation prediction information of the case, the method also includes: training with legal information samples and text information of case samples to obtain a first task prediction model, wherein the multi-task prediction model includes a first task prediction model, and the first task prediction model is used to predict the legal information of the case; training with legal information samples, text information of case samples and legal information output by the first task prediction model to obtain a second task prediction model, wherein the multi-task prediction model includes a second task prediction model, and the second task prediction model is used to predict the litigation prediction information of the case.
[0176] Optionally, the storage medium is also configured to store program codes for performing the following steps: encoding a legal information sample to obtain a first encoding result; encoding the text information of a case sample to obtain a second encoding result; encoding discrete features of the text information of the case sample to obtain a third encoding result; training with the first encoding result, the second encoding result, and the third encoding result to obtain a first task prediction model, wherein the first encoding result, the second encoding result, and the third encoding result are results of an input layer of the first task prediction model; and training with the first encoding result, the second encoding result, the third encoding result, and the legal information output by the first task prediction model to obtain a second task prediction model, wherein the first encoding result, the second encoding result, and the third encoding result are results of an input layer of the second task prediction model.
[0177] Optionally, the storage medium is also configured to store program codes for executing the following steps: segmenting the legal information sample to obtain a first word vector sequence; and determining the first word vector sequence as a first encoding result.
[0178] Optionally, the storage medium is also configured to store program codes for executing the following steps: segmenting the text information of the case sample to obtain a second word vector sequence; encoding the second word vector sequence to obtain a second encoding result.
[0179] Optionally, the storage medium is further configured to store a program code for executing the following steps: encoding discrete features through a multi-layer perceptron to obtain a third encoding result with continuous semantics.
[0180] Optionally, the storage medium is also configured to store program code for executing the following steps: fusion encoding the second encoding result and the third encoding result to obtain a fourth encoding result; splicing the fourth encoding result and the first encoding result to obtain a fifth encoding result; training through the fifth encoding result to obtain a first task prediction model, wherein the fifth encoding result is the result of the intermediate layer of the first task prediction model.
[0181] Optionally, the storage medium is also configured to store program code for executing the following steps: determining the output layer of the first task prediction model through the fifth encoding result and the objective function to obtain the first task prediction model.
[0182] Optionally, the storage medium is also configured to store program code for executing the following steps: classifying the first coding result by a pre-constructed classifier of the legal information sample to obtain a classification result; splicing the classification result, the intermediate classification result used to obtain the classification result, the second coding result and the third coding result to obtain a spliced result; fusion encoding the spliced result to obtain a sixth coding result, wherein the sixth coding result is the result of the intermediate layer of the second task prediction model; training by the sixth coding result and the legal information output by the first task prediction model to obtain the second task prediction model.
[0183] Optionally, the storage medium is also configured to store program code for executing the following steps: determining the output layer of the second task prediction model through the sixth encoding result and the objective function; training the output layer of the second task prediction model through the legal information output by the first task prediction model to obtain the second task prediction model.
[0184] Optionally, the storage medium is also configured to store program code for executing the following steps: obtaining a first loss function corresponding to the legal information output by the first task prediction model; obtaining a second loss function corresponding to the output layer of the second task prediction model; training the output layer of the second task prediction model by the first loss function and the second loss function to obtain the second task prediction model.
[0185] Optionally, the storage medium is also configured to store program code for executing the following steps: extracting statistical features of words and vector features of words from text information of case samples; training through legal information samples, statistical features of words and vector features of words to obtain a first task prediction model; training through legal information samples, text information of case samples and legal information output by the first task prediction model to obtain a second task prediction model, including: training through legal information samples, statistical features of words, vector features of words and legal information output by the first task prediction model to obtain a second task prediction model.
[0186] Optionally, in this embodiment, the storage medium is also configured to store program code for executing the following steps: displaying target text information on a text display interface, wherein the target text information is used to describe case information of a case in a litigation scenario; displaying legal information and litigation prediction information of the case on a text display interface, wherein the legal information and litigation prediction information of the case are the result of predictive processing of the target text information based on a multi-task prediction model, and the multi-task prediction model is trained by using text information of case information of case samples used to describe the litigation scenario and legal information samples of the case samples.
[0187] Optionally, in this embodiment, the storage medium is also configured to store program codes for executing the following steps: obtaining case information of a case, wherein the case information of the case includes at least one of the following: a complaint text, a defense text, a legal and regulatory text, an evidence text, and a judgment text; performing predictive processing on the case information of the case based on a multi-task prediction model to predict legal information and litigation prediction information of the case, wherein the multi-task prediction model is a neural network model for encoding different types of files in the case information of the case; and outputting the legal information and litigation prediction information of the case.
[0188] Optionally, the storage medium is also configured to store program code for executing the following steps: before predicting and processing case information based on the multi-task prediction model to obtain legal information and litigation prediction information of the case, training is performed through legal information samples of case samples and case information of case samples to obtain a first task prediction model, wherein the multi-task prediction model includes a first task prediction model, and the first task prediction model is used to predict the legal information of the case; training is performed through legal information samples, case information of case samples, and legal information output by the first task prediction model to obtain a second task prediction model, wherein the multi-task prediction model includes a second task prediction model, and the second task prediction model is used to predict the litigation prediction information of the case.
[0189] Optionally, the storage medium is also configured to store program codes for executing the following steps: encoding a legal information sample to obtain a first encoding result; encoding the case information of a case sample to obtain a second encoding result; encoding discrete features of the case information of the case sample to obtain a third encoding result; training with the first encoding result, the second encoding result and the third encoding result to obtain a first task prediction model, wherein the first encoding result, the second encoding result and the third encoding result are the results of the input layer of the first task prediction model; training with the legal information sample, the case information of the case sample and the legal information output by the first task prediction model to obtain a second task prediction model, including: training with the first encoding result, the second encoding result, the third encoding result and the legal information output by the first task prediction model to obtain a second task prediction model, wherein the first encoding result, the second encoding result and the third encoding result are the results of the input layer of the second task prediction model.
[0190] Example 6
[0191] The embodiment of the present invention further provides a processor. Optionally, in this embodiment, the processor runs the program of the text information processing method provided in the embodiment 1 above.
[0192] Optionally, in this embodiment, the processor is configured to run the program code of the following steps: obtaining target text information, wherein the target text information is used to describe case information of a case in a litigation scenario; performing predictive processing on the target text information based on a multi-task prediction model to predict legal information and litigation prediction information of the case, wherein the multi-task prediction model is trained by training text information of case information of case samples used to describe case information in the litigation scenario and legal information samples of the case samples; and outputting the legal information and litigation prediction information of the case.
[0193] Optionally, in this embodiment, the processor is also configured to run the program code of the following steps: displaying target text information on a text display interface, wherein the target text information is used to describe case information of a case in a litigation scenario; displaying legal information and litigation prediction information of the case on a text display interface, wherein the legal information and litigation prediction information of the case are the results of predictive processing of the target text information based on a multi-task prediction model, and the multi-task prediction model is trained by using text information of case information of case samples used to describe the litigation scenario and legal information samples of the case samples.
[0194] Optionally, in this embodiment, the processor is also configured to run the program code of the following steps: obtaining case information of the case, wherein the case information of the case includes at least one of the following: the text of the complaint, the text of the defense, the text of laws and regulations, the text of evidence and the text of the judgment; performing predictive processing on the case information of the case based on the multi-task prediction model to predict the legal information and litigation prediction information of the case, wherein the multi-task prediction model is a neural network model for encoding different types of files in the case information of the case; and outputting the legal information and litigation prediction information of the case.
[0195] The serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages or disadvantages of the embodiments.
[0196] In the above embodiments of the present invention, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.
[0197] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.
[0198] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.
[0199] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.
[0200] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, magnetic disk or optical disk and other media that can store program codes.
[0201] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.
Claims
1. A method for processing text information, It is characterized in that include: Acquire target text information, wherein the target text information is used to describe case information of a case in a litigation scenario; Predicting the target text information based on a multi-task prediction model to obtain legal information and litigation prediction information of the case, wherein the multi-task prediction model is trained by using text information describing case information of case samples in the litigation scenario and legal information samples of the case samples; Output legal information and litigation prediction information of the case; The method also includes: encoding the legal information sample to obtain a first encoding result; encoding the text information of the case sample to obtain a second encoding result; encoding the discrete features of the text information of the case sample to obtain a third encoding result; wherein the first encoding result, the second encoding result and the third encoding result are the results of the input layer of the first task prediction model, wherein the multi-task prediction model includes the first task prediction model, and the first task prediction model is used to predict the legal information of the case; and the first encoding result, the second encoding result and the third encoding result are the results of the input layer of the second task prediction model, wherein the multi-task prediction model includes the second task prediction model, and the second task prediction model is used to predict the litigation prediction information of the case.
2. The method according to claim 1, It is characterized in that Before processing the target text information based on the multi-task prediction model to obtain the legal information and litigation prediction information of the case, the method further includes: The first task prediction model is obtained by training the legal information sample and the text information of the case sample; The second task prediction model is obtained by training with the legal information sample, the text information of the case sample and the legal information output by the first task prediction model.
3. The method according to claim 2, It is characterized in that The first task prediction model is obtained by training with the text information of the legal information sample and the case sample, including: the first task prediction model is obtained by training with the first encoding result, the second encoding result and the third encoding result; The second task prediction model is obtained by training with the legal information sample, the text information of the case sample and the legal information output by the first task prediction model, including: the second task prediction model is obtained by training with the first encoding result, the second encoding result, the third encoding result and the legal information output by the first task prediction model.
4. The method according to claim 3, It is characterized in that The legal information sample is encoded to obtain a first encoding result including: Segmenting the legal information sample to obtain a first word vector sequence; Determine the first word vector sequence as the first encoding result.
5. The method according to claim 3, It is characterized in that The text information of the case sample is encoded to obtain a second encoding result including: Segmenting the text information of the case sample to obtain a second word vector sequence; Encode the second word vector sequence to obtain the second encoding result.
6. The method according to claim 3, It is characterized in that The discrete features of the text information of the case sample are encoded to obtain a third encoding result including: The discrete features are encoded through a multi-layer perceptron to obtain a third encoding result with continuous semantics.
7. The method according to claim 3, It is characterized in that The first task prediction model is obtained by training the first encoding result, the second encoding result, and the third encoding result, including: Performing fusion encoding on the second encoding result and the third encoding result to obtain a fourth encoding result; Concatenate the fourth encoding result and the first encoding result to obtain a fifth encoding result; The first task prediction model is obtained by training through the fifth encoding result, wherein the fifth encoding result is the result of the intermediate layer of the first task prediction model.
8. The method according to claim 7, It is characterized in that The first task prediction model is obtained by training through the fifth encoding result, including: The output layer of the first task prediction model is determined by the fifth encoding result and the objective function to obtain the first task prediction model.
9. The method according to claim 3, It is characterized in that The second task prediction model is obtained by training the first encoding result, the second encoding result, the third encoding result and the legal information output by the first task prediction model, including: Classifying the first coding result by using a pre-constructed classifier of the legal information sample to obtain a classification result; splicing the classification result, the intermediate classification result used to obtain the classification result, the second encoding result and the third encoding result to obtain a splicing result; Performing fusion encoding on the splicing result to obtain a sixth encoding result, wherein the sixth encoding result is a result of an intermediate layer of the second task prediction model; The second task prediction model is obtained by training with the sixth encoding result and the legal information output by the first task prediction model.
10. The method according to claim 9, It is characterized in that The second task prediction model is obtained by training the sixth encoding result and the legal information output by the first task prediction model, including: Determine the output layer of the second task prediction model according to the sixth encoding result and the objective function; The output layer of the second task prediction model is trained using the legal information output by the first task prediction model to obtain the second task prediction model.
11. The method according to claim 10, It is characterized in that The output layer of the second task prediction model is trained by using the legal information output by the first task prediction model, so as to obtain the second task prediction model, comprising: Obtaining a first loss function corresponding to the legal information output by the first task prediction model; Obtaining a second loss function corresponding to the output layer of the second task prediction model; The output layer of the second task prediction model is trained by using the first loss function and the second loss function to obtain the second task prediction model.
12. The method according to any one of claims 2 to 11, It is characterized in that The method further comprises: extracting statistical features of words and vector features of words from the text information of the case sample; The first task prediction model is obtained by training the legal information sample and the text information of the case sample, including: the first task prediction model is obtained by training the legal information sample, the statistical features of the word and the vector features of the word; The second task prediction model is obtained by training the legal information samples, the text information of the case samples and the legal information output by the first task prediction model, including: the second task prediction model is obtained by training the legal information samples, the statistical features of the words, the vector features of the words and the legal information output by the first task prediction model.
13. A method for processing text information, It is characterized in that include: Displaying target text information on a text display interface, wherein the target text information is used to describe case information of a case in a litigation scenario; Displaying the legal information and litigation prediction information of the case on the text display interface, wherein the legal information and litigation prediction information of the case are the result of predicting the target text information based on a multi-task prediction model, and the multi-task prediction model is trained by using text information of case information of case samples in the litigation scenario and legal information samples of the case samples; The method also includes: encoding the legal information sample to obtain a first encoding result; encoding the text information of the case sample to obtain a second encoding result; encoding the discrete features of the text information of the case sample to obtain a third encoding result; wherein the first encoding result, the second encoding result and the third encoding result are the results of the input layer of the first task prediction model, wherein the multi-task prediction model includes the first task prediction model, and the first task prediction model is used to predict the legal information of the case; and the first encoding result, the second encoding result and the third encoding result are the results of the input layer of the second task prediction model, wherein the multi-task prediction model includes the second task prediction model, and the second task prediction model is used to predict the litigation prediction information of the case.
14. A method for processing case information, It is characterized in that include: Obtaining case information of the case, wherein the case information of the case includes at least one of the following: the text of the indictment, the text of the defense, the text of the law and regulations, the text of the evidence, and the text of the judgment; Predicting the case information of the case based on a multi-task prediction model to obtain legal information and litigation prediction information of the case, wherein the multi-task prediction model is a neural network model, which is used to encode different types of files in the case information of the case; Output legal information and litigation prediction information of the case; The method also includes: encoding the legal information sample to obtain a first encoding result; encoding the case information of the case sample to obtain a second encoding result; encoding the discrete features of the case information of the case sample to obtain a third encoding result, wherein the first encoding result, the second encoding result and the third encoding result are the results of the input layer of the first task prediction model, wherein the multi-task prediction model includes the first task prediction model, and the first task prediction model is used to predict the legal information of the case; and the first encoding result, the second encoding result and the third encoding result are the results of the input layer of the second task prediction model, wherein the multi-task prediction model includes the second task prediction model, and the second task prediction model is used to predict the litigation prediction information of the case.
15. The method according to claim 14, It is characterized in that Before predicting the case information based on the multi-task prediction model to obtain the legal information and litigation prediction information of the case, the method further includes: The first task prediction model is obtained by training the legal information samples of the case samples and the case information of the case samples; The second task prediction model is obtained by training through the legal information samples, the case information of the case samples and the legal information output by the first task prediction model.
16. The method according to claim 15, It is characterized in that The first task prediction model is obtained by training the legal information sample and the case information of the case sample, including: training the first encoding result, the second encoding result and the third encoding result to obtain the first task prediction model; The second task prediction model is obtained by training with the legal information sample, the case information of the case sample and the legal information output by the first task prediction model, including: the second task prediction model is obtained by training with the first encoding result, the second encoding result, the third encoding result and the legal information output by the first task prediction model.
17. A text information processing device, It is characterized in that include: A first acquisition unit is used to acquire target text information, wherein the target text information is used to describe case information of a case in a litigation scenario; A first processing unit is used to perform prediction processing on the target text information based on a multi-task prediction model to predict the legal information and litigation prediction information of the case, wherein the multi-task prediction model is trained by using text information of case information of case samples in the litigation scenario and legal information samples of the case samples; A first output unit, used to output the legal information and litigation prediction information of the case; Wherein, the text information processing device is also used to: encode the legal information sample to obtain a first encoding result; encode the text information of the case sample to obtain a second encoding result; encode the discrete features of the text information of the case sample to obtain a third encoding result; wherein the first encoding result, the second encoding result and the third encoding result are the results of the input layer of the first task prediction model, wherein the multi-task prediction model includes the first task prediction model, and the first task prediction model is used to predict the legal information of the case; and the first encoding result, the second encoding result and the third encoding result are the results of the input layer of the second task prediction model, wherein the multi-task prediction model includes the second task prediction model, and the second task prediction model is used to predict the litigation prediction information of the case.
18. A text information processing device, It is characterized in that include: A first display unit, configured to display target text information on a text display interface, wherein the target text information is used to describe case information of a case in a litigation scenario; a second display unit, configured to display the legal information and litigation prediction information of the case on the text display interface, wherein the legal information and litigation prediction information of the case are the result of prediction processing of the target text information based on a multi-task prediction model, and the multi-task prediction model is trained by using text information of case information of case samples in the litigation scenario and legal information samples of the case samples; Wherein, the text information processing device is also used to: encode the legal information sample to obtain a first encoding result; encode the text information of the case sample to obtain a second encoding result; encode the discrete features of the text information of the case sample to obtain a third encoding result; wherein the first encoding result, the second encoding result and the third encoding result are the results of the input layer of the first task prediction model, wherein the multi-task prediction model includes the first task prediction model, and the first task prediction model is used to predict the legal information of the case; and the first encoding result, the second encoding result and the third encoding result are the results of the input layer of the second task prediction model, wherein the multi-task prediction model includes the second task prediction model, and the second task prediction model is used to predict the litigation prediction information of the case.
19. A case information processing device, It is characterized in that include: A second acquisition unit is used to acquire case information of the case, wherein the case information of the case includes at least one of the following: a text of an indictment, a text of a defense, a text of laws and regulations, a text of evidence, and a text of a judgment; a second processing unit, configured to perform prediction processing on the case information of the case based on a multi-task prediction model, and predict the legal information and litigation prediction information of the case, wherein the multi-task prediction model is a neural network model, and is configured to perform encoding processing on different types of files in the case information of the case; A second output unit, used to output the legal information and litigation prediction information of the case; Wherein, the case information processing device is also used to: encode the legal information sample to obtain a first encoding result; encode the text information of the case sample to obtain a second encoding result; encode the discrete features of the text information of the case sample to obtain a third encoding result; wherein the first encoding result, the second encoding result and the third encoding result are the results of the input layer of the first task prediction model, wherein the multi-task prediction model includes the first task prediction model, and the first task prediction model is used to predict the legal information of the case; and the first encoding result, the second encoding result and the third encoding result are the results of the input layer of the second task prediction model, wherein the multi-task prediction model includes the second task prediction model, and the second task prediction model is used to predict the litigation prediction information of the case.
20. A storage medium, It is characterized in that The storage medium includes a stored program, wherein when the program is running, the device where the storage medium is located is controlled to perform the following steps: Acquire target text information, wherein the target text information is used to describe case information of a case in a litigation scenario; Predicting the target text information based on a multi-task prediction model to obtain legal information and litigation prediction information of the case, wherein the multi-task prediction model is trained by using text information describing case information of case samples in the litigation scenario and legal information samples of the case samples; Output legal information and litigation prediction information of the case; Encode the legal information sample to obtain a first encoding result; encode the text information of the case sample to obtain a second encoding result; encode the discrete features of the text information of the case sample to obtain a third encoding result; wherein the first encoding result, the second encoding result and the third encoding result are the results of the input layer of the first task prediction model, wherein the multi-task prediction model includes the first task prediction model, and the first task prediction model is used to predict the legal information of the case; and the first encoding result, the second encoding result and the third encoding result are the results of the input layer of the second task prediction model, wherein the multi-task prediction model includes the second task prediction model, and the second task prediction model is used to predict the litigation prediction information of the case.
21. A processor, It is characterized in that The processor is used to run a program, wherein the following steps are performed when the program is run: Acquire target text information, wherein the target text information is used to describe case information of a case in a litigation scenario; Predicting the target text information based on a multi-task prediction model to obtain legal information and litigation prediction information of the case, wherein the multi-task prediction model is trained by using text information describing case information of case samples in the litigation scenario and legal information samples of the case samples; Output legal information and litigation prediction information of the case; Encode the legal information sample to obtain a first encoding result; encode the text information of the case sample to obtain a second encoding result; encode the discrete features of the text information of the case sample to obtain a third encoding result; wherein the first encoding result, the second encoding result and the third encoding result are the results of the input layer of the first task prediction model, wherein the multi-task prediction model includes the first task prediction model, and the first task prediction model is used to predict the legal information of the case; and the first encoding result, the second encoding result and the third encoding result are the results of the input layer of the second task prediction model, wherein the multi-task prediction model includes the second task prediction model, and the second task prediction model is used to predict the litigation prediction information of the case.
22. A mobile terminal, It is characterized in that include: processor; a transmission device connected to the processor, and configured to obtain target text information, wherein the target text information is used to describe case information of a case in a litigation scenario; and A memory connected to the processor, for providing the processor with instructions for processing the following processing steps: predicting the target text information based on a multi-task prediction model to predict the legal information and litigation prediction information of the case, wherein the multi-task prediction model is trained by text information of case information of case samples used to describe the litigation scenario and legal information samples of the case samples; outputting the legal information and litigation prediction information of the case; encoding the legal information sample to obtain a first encoding result; encoding the text information of the case sample to obtain a second encoding result; encoding the discrete features of the text information of the case sample to obtain a third encoding result; wherein the first encoding result, the second encoding result and the third encoding result are the results of the input layer of the first task prediction model, wherein the multi-task prediction model includes the first task prediction model, and the first task prediction model is used to predict the legal information of the case; and the first encoding result, the second encoding result and the third encoding result are the results of the input layer of the second task prediction model, wherein the multi-task prediction model includes the second task prediction model, and the second task prediction model is used to predict the litigation prediction information of the case.
Citation Information
Patent Citations
An apparatus for assisting judicial case decision based on machine learning
CN109241285A
User legal provision prediction method based on a filtering door mechanism
CN109829055A
Intelligent law assessment method and system
CN110222145A
Litigation result prediction method and device, computer equipment and storage medium
CN110377632A
Law service system and the method for lawsuit orderusing an internet
KR1020020078157A
Cited By
Machine-learned language models which generate intermediate textual analysis in service of contextual text generation
US12481834B2
Machine-Learned Language Models Which Generate Intermediate Textual Analysis in Service of Contextual Text Generation
US20240256786A1
