A legal judgment prediction method based on prior knowledge and subtask dependencies

Abstract

The present invention discloses a legal judgment prediction method based on prior knowledge and subtask dependencies. The method is applied to the field of natural language processing technology to solve the problem of low legal judgment prediction accuracy. The method comprises: obtaining factual information in a legal document to be tested, wherein the factual information is used to characterize the case fact description in the legal document to be tested; inputting the factual information into a legal judgment prediction model, and outputting a legal article prediction result, a crime prediction result, and a sentence prediction result corresponding to the factual information through the legal judgment prediction model, wherein the legal judgment prediction model comprises a legal article predictor, a crime predictor, a correction module, and a sentence predictor. In an embodiment of the present invention, the legal article prediction and the crime prediction are corrected by the correction module, thereby significantly improving the prediction accuracy.

Description

Technical Field

[0001] The present invention relates to the technical field of natural language processing, and in particular to a legal judgment prediction method based on prior knowledge and subtask dependencies. Background Art

[0002] The task of automated judgment is to predict the outcome of a court ruling based on the facts contained in legal documents. This approach, on the one hand, provides higher-quality judgments for those without legal background; on the other, it serves as a legal reference for legal professionals. In recent years, there has been extensive research on automated judgment. Initially, this problem was treated as a simple text classification problem, employing traditional methods such as keyword matching. With the development of deep learning, more researchers have begun to leverage deep learning frameworks to extract information from text to assist automated judgment. However, most existing legal judgment prediction methods suffer from low accuracy.

[0003] Therefore, the existing technology still needs to be improved and developed. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a legal judgment prediction method based on prior knowledge and sub-task dependencies in response to the above-mentioned defects of the existing technology, aiming to solve the problem of low prediction accuracy of the legal judgment prediction method in the existing technology.

[0005] The technical solutions adopted by the present invention to solve the problem are as follows:

[0006] In a first aspect, an embodiment of the present invention provides a legal decision prediction method based on prior knowledge and subtask dependencies, wherein the method includes:

[0007] Acquiring factual information from the legal document to be tested, wherein the factual information is used to represent the description of case facts in the legal document to be tested;

[0008] The factual information is input into a legal judgment prediction model, and the legal judgment prediction model outputs a legal article prediction result, a crime prediction result and a sentence prediction result corresponding to the factual information, wherein the legal judgment prediction model includes a legal article predictor, a crime predictor, a correction module and a sentence predictor.

[0009] In a second aspect, a legal judgment prediction device based on prior knowledge and subtask dependencies is provided, wherein the device includes: a fact information acquisition module, configured to acquire fact information in a legal document to be tested, wherein the fact information is used to represent a description of case facts in the legal document to be tested;

[0010] A prediction result output module is used to input the factual information into a legal judgment prediction model, and output a legal article prediction result, a crime prediction result and a sentence prediction result corresponding to the factual information through the legal judgment prediction model, wherein the legal judgment prediction model includes a legal article predictor, a crime predictor, a correction module and a sentence predictor.

[0011] On the third aspect, an intelligent terminal is also provided, comprising a memory and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by one or more processors, wherein the one or more programs include a method for executing a legal judgment prediction method based on prior knowledge and sub-task dependencies as described above.

[0012] In a fourth aspect, a non-temporary computer-readable storage medium is also provided. When the instructions in the storage medium are executed by a processor of an electronic device, the electronic device is enabled to execute a legal judgment prediction method based on prior knowledge and sub-task dependencies as described in any one of the above.

[0013] Beneficial effects of the present invention: The embodiment of the present invention first obtains factual information in the legal document to be tested, wherein the factual information is used to characterize the case fact description in the legal document to be tested; then the factual information is input into the legal judgment prediction model, and the legal judgment prediction model outputs the legal article prediction result, crime prediction result and sentence prediction result corresponding to the factual information, wherein the legal judgment prediction model includes a legal article predictor, a crime predictor, a correction module and a sentence predictor; it can be seen that the embodiment of the present invention corrects the legal article prediction and crime prediction through the correction module, so that the prediction accuracy is greatly improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0015] Figure 1 A flowchart of a legal decision prediction method based on prior knowledge and subtask dependencies provided by an embodiment of the present invention.

[0016] Figure 2 A diagram of a legal decision prediction model provided by an embodiment of the present invention.

[0017] Figure 3 A dependency diagram of legal provisions and crimes provided in an embodiment of the present invention.

[0018] Figure 4 This is a diagram of the correction algorithm provided by an embodiment of the present invention.

[0019] Figure 5 A schematic diagram of a correction process for an implementation method provided in an embodiment of the present invention.

[0020] Figure 6 This is a diagram of the legal judgment experiment results provided by an embodiment of the present invention.

[0021] Figure 7 This is a principle block diagram of a legal decision prediction device based on prior knowledge and subtask dependencies provided by an embodiment of the present invention.

[0022] Figure 8 This is a block diagram of the internal structure of the smart terminal provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0023] The present invention discloses a method for predicting legal decisions based on prior knowledge and subtask dependencies. To clarify the objectives, technical solutions, and effects of the present invention, the present invention is further described below with reference to the accompanying drawings and examples. It should be understood that the specific examples described herein are intended only to illustrate the present invention and are not intended to limit the present invention.

[0024] It will be understood by those skilled in the art that, unless expressly stated otherwise, the singular forms "a", "an", "said" and "the" used herein may also include the plural forms. It should be further understood that the term "comprising" used in the description of the present invention refers to the presence of the features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof. It should be understood that when we refer to an element as being "connected" or "coupled" to another element, it may be directly connected or coupled to the other element, or there may be intermediate elements. In addition, "connected" or "coupled" as used herein may include wireless connections or wireless couplings. The term "and / or" used herein includes all or any units and all combinations of one or more associated listed items.

[0025] It will be understood by those skilled in the art that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by those skilled in the art in the art to which the present invention belongs. It should also be understood that terms such as those defined in common dictionaries should be understood to have meanings consistent with their meanings in the context of the prior art and will not be interpreted in an idealized or overly formal sense unless specifically defined as herein.

[0026] Since most legal judgment prediction devices in the existing technology do not introduce prior knowledge or use the output of the current task as prior knowledge for subsequent tasks, the prediction accuracy of the legal judgment prediction method is low.

[0027] To address the problems of the existing technology, this embodiment provides a legal judgment prediction method based on prior knowledge and subtask dependencies. This method uses a correction module to correct legal article predictions and crime predictions, significantly improving prediction accuracy. In specific implementation, the method first obtains factual information from the legal document to be tested, where the factual information is used to characterize the case fact description in the legal document to be tested. The method then inputs the factual information into a legal judgment prediction model, which then outputs legal article prediction results, crime prediction results, and sentence prediction results corresponding to the factual information. The legal judgment prediction model includes a legal article predictor, a crime predictor, a correction module, and a sentence predictor.

[0028] Exemplary Methods

[0029] This embodiment provides a legal judgment prediction method based on prior knowledge and subtask dependencies, which can be applied to smart terminals for natural language processing. Figure 1 As shown, the method includes:

[0030] Step S100: Obtain factual information in the legal document to be tested, wherein the factual information is used to represent the case fact description in the legal document to be tested;

[0031] Specifically, the legal document to be tested is the legal document for which a legal judgment outcome prediction is required. The factual information is the description of the case facts in the legal document to be tested. The factual information is the description of the factual part, that is, the case details. For example, the factual information can include the relationship between the plaintiff and the defendant, the cause, process, and result of the incident.

[0032] The legal documents to be tested can be downloaded from the legal system dataset or the internet, and then the factual information can be extracted manually or by machine. For example, documents can be downloaded from the China Judgment Documents website, and then AI can be used to extract information such as the relationship between the plaintiff and the defendant, the cause, process, and outcome of the incident.

[0033] After obtaining the factual information, you can execute Figure 1 The following steps are shown: S200, inputting the factual information into the legal judgment prediction model, and outputting the legal article prediction result, crime prediction result and sentence prediction result corresponding to the factual information through the legal judgment prediction model, wherein the legal judgment prediction model includes a legal article predictor, a crime predictor, a correction module and a sentence predictor.

[0034] Specifically, if Figure 2As shown, the legal judgment prediction model is pre-trained and includes a legal provision predictor, a charge predictor, a correction module, and a sentence prediction module. The legal provision predictor, the charge predictor, and the sentence prediction module can all employ models such as CNN models, LSTM models, and pre-trained BERT models. In one implementation, hyperparameters of the predictor are set, such as the learning rate, maximum input length, and representation dimension.

[0035] The factual information can be input into the law article predictor, crime predictor, correction module and sentence predictor in sequence, or the factual information can be input into the law article predictor and crime predictor respectively, and then the two output results of the law article predictor and crime predictor are input into the correction module and crime predictor in sequence, without specific limitation.

[0036] Step S200 includes the following steps:

[0037] S201, encoding the fact information to obtain an encoding result;

[0038] S202: Obtaining an initial legal article prediction result and an initial crime prediction result based on the coding result;

[0039] S203, inputting the initial legal article prediction result and the initial crime prediction result into the correction module, and outputting the legal article prediction result and the crime prediction result through the correction module;

[0040] S204, outputting the legal article prediction result and the crime prediction result;

[0041] S205. Determine a sentence prediction result corresponding to the factual information based on the factual information, the legal article prediction result, and the crime prediction result.

[0042] Specifically, in step S201, encoding is the process of converting information from one form or format to another, also known as code in a computer programming language. Encoding can be performed using an autoencoder, an autoencoder, or the encoding portion of a transformer network. After encoding the factual information, an encoded result is obtained.

[0043] Step S202 includes the following steps: inputting the coding result into the law article predictor, and obtaining an initial law article prediction result through the law article predictor; inputting the coding result into the crime name predictor, and obtaining an initial crime name prediction result through the crime name predictor, wherein the crime name prediction result is used to characterize the prediction of the crime name. In this way, after the coding result is input into the law article predictor and the crime name predictor respectively, the initial law article prediction result and the initial crime name prediction result can be obtained respectively. In the prior art, the case judgment process of human judges follows the order from determining the law article, determining the crime to determining the sentence. In the embodiment of the present invention, the initial law article prediction result and the initial crime name prediction result are executed in parallel, making it more in line with the working logic of artificial intelligence.

[0044] In step S203, the initial law article prediction result and the initial crime name prediction result may not conform to prior knowledge. In order to improve the accuracy of the prediction, the embodiment of the present invention inputs the initial law article prediction result and the initial crime name prediction result into the correction module, and outputs the law article prediction result and the crime name prediction result through the correction module.

[0045] In one implementation, the correction process of the correction module includes the following steps: obtaining a first parameter and initializing it; obtaining a data set, the number of the initial law article prediction results, and the number of the initial crime name prediction results; obtaining prior knowledge based on the data set; wherein the prior knowledge is the distribution inferred from the law articles and crimes; when the first parameter is less than a preset threshold and the initial law article prediction result and the initial crime name prediction result are inconsistent with the prior knowledge, performing correction operations on the initial law article prediction result and the initial crime name prediction result; using the initial law article prediction result as the law article prediction result, and using the initial crime name prediction result as the crime name prediction result.

[0046] Specifically, the first parameter represents the number of times, represented by t, and t is initialized to 0, and the data set, the number of the initial legal provisions prediction results L_max, and the number of the initial crime prediction results C_max are obtained; wherein, L_max=max(L), C_max=max(C). The data set can come from the legal document to be tested, and the data set contains legal provisions and crimes, and the crime is the name of the crime. Based on the data set, prior knowledge is obtained; wherein, the prior knowledge is the distribution inferred from the legal provisions and crimes; accordingly, the prior knowledge obtained based on the data set includes the following steps: obtaining the legal provisions and crimes in the data set; parsing the dependency between the legal provisions and the crimes to obtain prior knowledge.

[0047] In this embodiment, if Figure 3As shown, dependency relationships mean that legal provisions need to reference crimes, and crimes need to reference legal provisions. By manually or machine-extracting legal provisions and crimes from a dataset, the dependency relationships between these provisions and crimes are analyzed to obtain prior knowledge. For example, a legal provision for defrauding an exceptionally large amount of public or private property requires reference to the crimes of fraud and swindling, while the crime of fraud requires reference to the crimes of defrauding an exceptionally large amount of public or private property, defrauding status, honor, or treatment, and molesting women. This allows for prior knowledge of the relationship between defrauding an exceptionally large amount of public or private property and the crime of fraud.

[0048] After obtaining prior knowledge, such as Figure 4-5 As shown, when the first parameter t is less than a preset threshold value (such as 10), the initial law article prediction result and the initial crime prediction result are corrected, which can improve the accuracy of the sentence prediction; when the initial law article prediction result and the initial crime prediction result are inconsistent with the prior knowledge, the initial law article prediction result and the initial crime prediction result are corrected; accordingly, the correction operation on the initial law article prediction result and the initial crime prediction result includes the following steps: when the number of the initial law article prediction results is greater than the number of the initial crime prediction results, the initial crime prediction result is set to 0, and the first parameter is accumulated by a preset value; when the number of the initial law article prediction results is less than the number of the initial crime prediction results, the initial law article prediction result is set to 0, and the first parameter is accumulated by a preset value.

[0049] In this embodiment, when the number L_max of the initial legal article prediction results is greater than the number C_max of the initial crime prediction results, the initial crime prediction result is set to 0, and the first parameter is accumulated by a preset value (such as 1); for example: when L_max>C_max, C[c_max_index]=0, t=t+1; when the number L_max of the initial legal article prediction results is less than the number C_max of the initial crime prediction results, the initial crime prediction result is set to 0, and the first parameter is accumulated by a preset value (such as 1); for example: when L_max>C_max, L[l_max_index]=0, t=t+1. When the first parameter is greater than a preset threshold value or the initial legal article prediction result and the initial crime prediction result do not contradict the prior knowledge, the above cycle is no longer performed, and the initial legal article prediction result is used as the legal article prediction result, and the initial crime prediction result is used as the crime prediction result.

[0050] In step S204, the existing technology directly obtains the final sentence prediction result based on the initial law article prediction result and the initial crime prediction result; since the accuracy of the initial law article prediction result and the initial crime prediction result themselves needs to be improved, the obtained sentence prediction result is not accurate enough. In the present invention, the initial law article prediction result and the initial crime prediction result are corrected to obtain the said law article prediction result and the said crime prediction result, which are more accurate, and the said law article prediction result and the said crime prediction result can be directly output, which can serve as a basis for obtaining more accurate sentence prediction results later.

[0051] Step S205 includes the following steps: splicing the factual information, the legal article prediction result and the crime prediction result to obtain spliced ​​text information; inputting the spliced ​​text information into the sentence predictor, and outputting the sentence prediction result corresponding to the factual information through the sentence predictor.

[0052] Specifically, the spliced ​​text information can be obtained by sequentially splicing the factual information, the legal article prediction result, and the crime prediction result; the spliced ​​text information can also be obtained by sequentially splicing the legal article prediction result, the factual information, and the crime prediction result; or the spliced ​​text information can be obtained by sequentially splicing the legal article prediction result, the crime prediction result, and the factual information; there is no specific limitation. Finally, the spliced ​​text information is input into the sentence prediction device, which outputs the sentence prediction result corresponding to the factual information.

[0053] In one implementation, training fact information and labels corresponding to the training fact information are obtained, the training fact information is input into an initial model, and a model result is output through the initial model; a cross-entropy loss function is obtained according to the model result and the label, and the initial model is trained based on the cross-entropy loss function to obtain a legal decision prediction model. The optimal legal decision prediction model is selectively saved using the early stopping method, and training is stopped if the predictor performance does not improve for 10 consecutive times.

[0054] The innovation of this invention is that it corrects the results of legal judgment predictions based on explicit outcome dependencies, significantly improving prediction accuracy. Furthermore, the corrected results are then spliced ​​into the sentence prediction engine, assisting it in making predictions and improving accuracy.

[0055] This paper uses Accuracy, Macro Precision (MP), Macro Recall (MR), and Macro F1 (F1) to measure the performance of all models. Since we want to comprehensively evaluate the performance of all models, we use F1 score as the main indicator. Figure 6Figure 2 shows the experimental results of all models on the CAIL-small and CAIL-big datasets. Our PKJudge framework brings significant and consistent improvements to all classic baseline models.

[0056] The Legal Judgment Prediction Model (PKJudge model) can be applied to the field of legal intelligence. Intuitively, it can be applied to the task of legal judgment prediction. On the one hand, it can effectively provide reliable references for legal practitioners. On the other hand, the public can obtain high-quality legal consulting services at a lower cost. From a broader perspective, this outcome-dependent legal intelligence model concept can be applied to nearly all legal multi-task problems, and even multi-task problems in other fields.

[0057] Exemplary devices

[0058] like Figure 7 As shown in , an embodiment of the present invention provides a legal judgment prediction device based on prior knowledge and subtask dependency, which includes a fact information acquisition module 301 and a prediction result output module 302:

[0059] The fact information acquisition module 301 is used to acquire fact information in the legal document to be tested, wherein the fact information is used to represent the case fact description in the legal document to be tested;

[0060] The prediction result output module 302 is used to input the factual information into the legal judgment prediction model, and output the legal article prediction result, crime prediction result and sentence prediction result corresponding to the factual information through the legal judgment prediction model, wherein the legal judgment prediction model includes a legal article predictor, a crime predictor, a correction module and a sentence predictor.

[0061] Based on the above embodiment, the present invention also provides an intelligent terminal, whose principle block diagram can be shown as follows: Figure 8 As shown. The intelligent terminal includes a processor, a memory, a network interface, a display screen, and a temperature sensor connected through a system bus. The processor of the intelligent terminal is used to provide computing and control capabilities. The memory of the intelligent terminal includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the intelligent terminal is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a legal judgment prediction method based on prior knowledge and sub-task dependencies is implemented. The display screen of the intelligent terminal can be a liquid crystal display screen or an electronic ink display screen, and the temperature sensor of the intelligent terminal is pre-set inside the intelligent terminal to detect the operating temperature of the internal device.

[0062] Those skilled in the art will understand that Figure 8 The schematic diagram is merely a block diagram of a portion of the structure related to the solution of the present invention and does not constitute a limitation on the smart terminal to which the solution of the present invention is applied. A specific smart terminal may include more or fewer components than shown in the diagram, or combine certain components, or have a different arrangement of components.

[0063] In one embodiment, a smart terminal is provided, comprising a memory and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by one or more processors. The one or more programs include instructions for performing the following operations:

[0064] Acquiring factual information from the legal document to be tested, wherein the factual information is used to represent the description of case facts in the legal document to be tested;

[0065] The factual information is input into a legal judgment prediction model, and the legal judgment prediction model outputs a legal article prediction result, a crime prediction result and a sentence prediction result corresponding to the factual information, wherein the legal judgment prediction model includes a legal article predictor, a crime predictor, a correction module and a sentence predictor.

[0066] Those skilled in the art will appreciate that all or part of the processes in the above-described embodiments can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the above-described embodiments. Among them, any reference to memory, storage, database or other media used in the embodiments provided by the present invention can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct RAM bus dynamic RAM (DRDRAM), and RAMbus dynamic RAM (RDRAM).

[0067] In summary, the present invention discloses a legal judgment prediction method based on prior knowledge and subtask dependency. The method is applied to the field of natural language processing technology to solve the problem of low accuracy in legal judgment prediction. The method includes: obtaining factual information in the legal document to be tested, wherein the factual information is used to characterize the case fact description in the legal document to be tested; inputting the factual information into a legal judgment prediction model, and outputting a legal article prediction result, a crime prediction result and a sentence prediction result corresponding to the factual information through the legal judgment prediction model, wherein the legal judgment prediction model includes a legal article predictor, a crime predictor, a correction module and a sentence predictor. The embodiment of the present invention corrects the legal article prediction and the crime prediction through the correction module, so that the prediction accuracy is greatly improved.

[0068] Based on the above embodiments, the present invention discloses a legal decision prediction method based on prior knowledge and subtask dependencies. It should be understood that the application of the present invention is not limited to the above examples. For ordinary technicians in this field, it can be improved or transformed according to the above description. All these improvements and transformations should fall within the scope of protection of the claims attached to the present invention.

Claims

1. A legal judgment prediction method based on prior knowledge and subtask dependencies, characterized by: The method comprises: Acquiring factual information from the legal document to be tested, wherein the factual information is used to represent the description of case facts in the legal document to be tested; Inputting the factual information into a legal judgment prediction model, and outputting a legal article prediction result, a crime prediction result, and a sentence prediction result corresponding to the factual information through the legal judgment prediction model, wherein the legal judgment prediction model includes a legal article predictor, a crime predictor, a correction module, and a sentence predictor; The inputting of the factual information into the legal judgment prediction model, and the outputting of the legal article prediction result, the crime prediction result, and the sentence prediction result corresponding to the factual information by the legal judgment prediction model include: Encoding the fact information to obtain an encoding result; According to the coding results, an initial legal article prediction result and an initial crime prediction result are obtained; Inputting the initial legal article prediction result and the initial crime prediction result into the correction module, and outputting the legal article prediction result and the crime prediction result through the correction module; Outputting the legal article prediction result and the crime prediction result; Determining a sentence prediction result corresponding to the factual information based on the factual information, the legal article prediction result, and the crime prediction result; The initial legal article prediction result and the initial crime prediction result obtained based on the coding result include: Inputting the encoding result into the legal article predictor, and obtaining an initial legal article prediction result through the legal article predictor; Inputting the encoding result into the crime prediction device, obtaining an initial crime prediction result through the crime prediction device, wherein the crime prediction result is used to represent the prediction of the crime name; The correction process of the correction module includes: Get the first parameter and initialize it; Obtaining a data set, the number of the initial legal article prediction results, and the number of the initial crime prediction results; Obtaining prior knowledge based on the data set; wherein the prior knowledge is a distribution inferred from legal provisions and crimes; When the first parameter is less than a preset threshold and the initial legal article prediction result and the initial crime prediction result are inconsistent with the prior knowledge, performing a correction operation on the initial legal article prediction result and the initial crime prediction result; The initial legal article prediction result is used as the legal article prediction result, and the initial crime prediction result is used as the crime prediction result; The obtaining of prior knowledge according to the data set includes: Obtain the legal provisions and crimes in the dataset; Analyze the dependency relationship between the legal provisions and the crimes to obtain prior knowledge; The correction operation on the initial legal article prediction result and the initial crime prediction result includes: When the number of the initial legal article prediction results is greater than the number of the initial crime prediction results, the initial crime prediction results are set to 0, and the first parameter is accumulated by a preset value; When the number of the initial legal article prediction results is less than the number of the initial crime prediction results, the initial legal article prediction results are set to 0, and the first parameter is accumulated with a preset value.

2. The legal judgment prediction method based on prior knowledge and subtask dependencies according to claim 1 is characterized in that: Determining the sentence prediction result corresponding to the factual information based on the factual information, the legal article prediction result, and the crime prediction result includes: Splicing the factual information, the legal article prediction result, and the crime prediction result to obtain spliced ​​text information; The spliced ​​text information is input into the sentence predictor, and the sentence predictor outputs a sentence prediction result corresponding to the factual information.

3. A legal judgment prediction device based on prior knowledge and subtask dependencies, characterized in that: The device comprises: A fact information acquisition module is used to acquire fact information in the legal document to be tested, wherein the fact information is used to represent the case fact description in the legal document to be tested; a prediction result output module, configured to input the factual information into a legal judgment prediction model, and output, through the legal judgment prediction model, a legal article prediction result, a crime prediction result, and a sentence prediction result corresponding to the factual information, wherein the legal judgment prediction model includes a legal article predictor, a crime predictor, a correction module, and a sentence predictor; The inputting of the factual information into the legal judgment prediction model, and the outputting of the legal article prediction result, the crime prediction result, and the sentence prediction result corresponding to the factual information by the legal judgment prediction model include: Encoding the fact information to obtain an encoding result; According to the coding results, an initial legal article prediction result and an initial crime prediction result are obtained; Inputting the initial legal article prediction result and the initial crime prediction result into the correction module, and outputting the legal article prediction result and the crime prediction result through the correction module; Outputting the legal article prediction result and the crime prediction result; Determining a sentence prediction result corresponding to the factual information based on the factual information, the legal article prediction result, and the crime prediction result; The correction process of the correction module includes: Get the first parameter and initialize it; Obtaining a data set, the number of the initial legal article prediction results, and the number of the initial crime prediction results; Obtaining prior knowledge based on the data set; wherein the prior knowledge is a distribution inferred from legal provisions and crimes; When the first parameter is less than a preset threshold and the initial legal article prediction result and the initial crime prediction result are inconsistent with the prior knowledge, performing a correction operation on the initial legal article prediction result and the initial crime prediction result; The initial legal article prediction result is used as the legal article prediction result, and the initial crime prediction result is used as the crime prediction result; The initial legal article prediction result and the initial crime prediction result obtained based on the coding result include: Inputting the encoding result into the legal article predictor, and obtaining an initial legal article prediction result through the legal article predictor; Inputting the encoding result into the crime prediction device, obtaining an initial crime prediction result through the crime prediction device, wherein the crime prediction result is used to represent the prediction of the crime name; The obtaining of prior knowledge according to the data set includes: Obtain the legal provisions and crimes in the dataset; Analyze the dependency relationship between the legal provisions and the crimes to obtain prior knowledge; The correction operation on the initial legal article prediction result and the initial crime prediction result includes: When the number of the initial legal article prediction results is greater than the number of the initial crime prediction results, the initial crime prediction results are set to 0, and the first parameter is accumulated by a preset value; When the number of the initial legal article prediction results is less than the number of the initial crime prediction results, the initial legal article prediction results are set to 0, and the first parameter is accumulated with a preset value.

4. An intelligent terminal, characterized in that: The device comprises a memory and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by one or more processors, and the one or more programs include being used to execute the method according to any one of claims 1 to 2.

5. A non-transitory computer-readable storage medium, characterized in that When the instructions in the storage medium are executed by a processor of an electronic device, the electronic device is enabled to execute the method according to any one of claims 1 to 2.

Images