Mediation plan providing method, mediation plan training method, related devices, and storage medium

Abstract

The mediation plan providing method, the training method, the related device and the storage medium, wherein the mediation plan providing method comprises: obtaining input information related to a case to be mediated; inputting the input information into a mediation plan providing model trained to obtain one or more mediation plans corresponding to the case to be mediated. The mediation plan providing method in the embodiments of the present specification can efficiently and directly provide mediation plans, which helps to assist mediation personnel to improve mediation efficiency; in addition, more mediation reference information can be provided when selecting the content of the mediation plan, which helps to improve the success rate of mediation.

Description

Technical Field

[0001] The embodiments of this specification relate to the field of data processing technology, and more particularly to a method for providing a mediation plan, a training method, related devices, and a storage medium. Background Art

[0002] With the rapid development of the commercial economy, the number of various commercial disputes has also increased significantly. The internet economy, in particular, has become closely integrated with people's daily lives, providing fast and accurate online shopping services. However, this has also led to a large number of legal disputes, such as those involving online transactions.

[0003] Mediation is a highly efficient way to resolve disputes. However, there are currently no suitable intelligent tools to assist mediators and judges in formulating mediation solutions, hindering efficient dispute resolution. Summary of the Invention

[0004] In view of this, the embodiments of this specification provide a method for providing a mediation plan, a training method, a related device and a storage medium, which can efficiently generate an accurate mediation plan.

[0005] First, the embodiment of this specification provides a method for providing a mediation plan, including:

[0006] Obtain case information for cases awaiting mediation;

[0007] Obtaining mediation reference information of the case to be mediated based on the case information;

[0008] The case information and the mediation reference information are input into a mediation plan providing model to obtain one or more mediation plans for the case to be mediated.

[0009] Optionally, the mediation reference information includes similar case information; the similar case information is obtained in the following manner:

[0010] Extract the first case from the preset judicial document database;

[0011] Calculating the similarity between the case to be mediated and the first case based on the case information;

[0012] The first case whose similarity exceeds a preset threshold is taken as a similar case of the case to be mediated, and similar case information is obtained.

[0013] Optionally, the judicial document database is obtained by:

[0014] Historical judicial documents are classified according to judicial elements to obtain a judicial document database.

[0015] Optionally, the method further includes:

[0016] The judicial document database is structured.

[0017] Optionally, extracting the first case from a preset judgment document library includes:

[0018] A first case whose judicial elements match the case to be mediated is extracted from the preset judgment document library.

[0019] Optionally, calculating the similarity between the case to be mediated and the first case based on the case information includes:

[0020] Extracting a case text from the case information, and performing vector encoding on the case text of the case to be mediated to obtain a first text semantic feature;

[0021] Extracting the case text of the first case, and performing vector encoding on the case text of the first case to obtain a second text semantic feature;

[0022] Calculate the cosine distance between the first text semantic feature and the second text semantic feature to obtain similarity.

[0023] Optionally, the similar case information includes statistical results of litigation outcomes of similar cases, and the method further includes:

[0024] Compile statistics on the litigation results of multiple similar cases to obtain statistical results;

[0025] The statistical results are used to be input into the mediation plan providing model together with the case information and the similar case information to obtain one or more mediation plans for the case to be mediated.

[0026] Optionally, the mediation reference information also includes characteristic information of the parties, which includes: information representing one or more combinations of the parties' litigation records, execution records, and risk preferences.

[0027] Optionally, the mediation reference information further includes: litigation result prediction information; the litigation result prediction information is obtained by:

[0028] Calculate the legal article prediction information of the applicable legal article of the case to be mediated based on the case information;

[0029] Support for the parties' claims in the case information is predicted based on the legal provision prediction information to obtain the claim result prediction information.

[0030] The embodiments of this specification provide a training method for a mediation plan provision model suitable for providing a mediation plan, which includes:

[0031] Obtaining a sample data set; wherein the sample data set includes a training data set, and the training data set includes: case information of each historical case, mediation reference information obtained based on the case information, and mediation results;

[0032] The sample data set is input into a mediation plan providing model, so that the mediation plan providing model is trained according to the loss between the mediation plan obtained by the model and the mediation result.

[0033] Optionally, the mediation reference information includes similar case information; the similar case information is obtained in the following manner:

[0034] Extract the first case from the preset judicial document database;

[0035] Calculating the similarity between the historical case and the first case based on the case information;

[0036] The first case whose similarity exceeds a preset threshold is regarded as a similar case of the historical case, and similar case information is obtained.

[0037] Optionally, the judicial document database is obtained by:

[0038] Historical judicial documents are classified according to judicial elements to obtain a judicial document database.

[0039] Optionally, the method further includes:

[0040] The judicial document database is structured.

[0041] Optionally, extracting the first case from a preset judgment document library includes:

[0042] A first case whose judicial elements match the historical case is extracted from the preset judgment document library.

[0043] Optionally, calculating the similarity between the historical case and the first case based on the case information includes:

[0044] Extracting case text from the case information, and performing vector encoding on the case text of the historical case to obtain a first text semantic feature;

[0045] Extracting the case text of the first case, and performing vector encoding on the case text of the first case to obtain a second text semantic feature;

[0046] Calculate the cosine distance between the first text semantic feature and the second text semantic feature to obtain similarity.

[0047] Optionally, the similar case information includes statistical results of litigation outcomes of similar cases, and the method further includes:

[0048] Compile statistics on the litigation results of multiple similar cases to obtain statistical results;

[0049] The statistical results are used to be input into the mediation plan providing model together with the case information and the similar case information to obtain one or more mediation plans for the historical cases.

[0050] Optionally, the mediation reference information also includes characteristic information of the parties, which includes: information representing one or more combinations of the parties' litigation records, execution records, and risk preferences.

[0051] Optionally, the mediation reference information further includes: litigation result prediction information; the litigation result prediction information is obtained by:

[0052] Calculate applicable laws and regulations for historical cases based on the case information;

[0053] Support for the parties' claims in the case information is predicted based on the legal provision prediction information to obtain the claim result prediction information.

[0054] This specification also provides an embodiment of a mediation plan providing device, which includes:

[0055] A data acquisition unit, used to obtain case information of the case to be mediated;

[0056] An information obtaining unit, configured to obtain mediation reference information of the case to be mediated based on the case information;

[0057] The input unit is used to input the case information and the mediation reference information into the mediation plan providing model to obtain one or more mediation plans for the case to be mediated.

[0058] The embodiment of this specification also provides a training device for a mediation plan providing model suitable for providing a mediation plan, which includes:

[0059] A data acquisition unit, configured to acquire a sample data set; wherein the sample data set includes a training data set, and the training data set includes: case information about each historical case, mediation reference information obtained based on the case information, and mediation results;

[0060] A training unit inputs the sample data set into a mediation plan providing model, so that the mediation plan providing model is trained based on the loss between the mediation plan obtained by the training unit and the mediation result. This specification also provides an electronic device including a memory and a processor, wherein the memory stores computer instructions executable on the processor, and when the processor executes the computer instructions, it performs the steps of the method described in any of the aforementioned embodiments.

[0061] An embodiment of this specification also provides a computer-readable storage medium having computer instructions stored thereon, which, when executed, execute the steps of the method described in any of the aforementioned embodiments.

[0062] Compared with the prior art, the technical solutions of the embodiments of this specification have the following beneficial effects:

[0063] On the one hand, the method for providing mediation plans in the embodiments of this specification is adopted, by obtaining case information and mediation reference information related to the case to be mediated, inputting them into the mediation plan providing model, and obtaining one or more mediation plans corresponding to the case to be mediated as output; the entire process is automatically completed based on the trained mediation plan providing model, which is efficient and direct; moreover, directly providing them to mediators, judges, etc. in the form of mediation plans will be more efficient in resolving disputes than the indirect method of providing legal clauses and similar cases.

[0064] On the other hand, by adopting the method for providing mediation plans in the embodiments of this specification, the mediation reference information may include legal information, litigation support, party characteristics information, etc. By referring to the mediation plan generated by this information, the content of the mediation plan can be based on the law, more acceptable to the parties, and improve the success rate of mediation. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1A It is a schematic diagram of an application scenario example in the embodiments of this specification.

[0066] Figure 1B This is a schematic diagram of another application scenario example in the embodiments of this specification.

[0067] Figure 1C This is a schematic diagram of another application scenario example in the embodiments of this specification.

[0068] Figure 2 It is a flowchart of the method for providing a mediation plan in the embodiment of this specification.

[0069] Figure 3 It is a flowchart of obtaining similar case information in the embodiments of this specification.

[0070] Figure 4It is a flowchart of calculating similarity in the process of obtaining similar case information in the embodiment of this specification.

[0071] Figure 5 It is a structural diagram of the system for obtaining similar case information in the embodiment of this specification.

[0072] Figure 6 It is a flowchart of obtaining litigation result prediction information in the embodiment of this specification.

[0073] Figure 7 It is a structural diagram of the legal issue prediction model in the embodiments of this specification.

[0074] Figure 8 It is a structural diagram of the mediation plan providing model in the embodiment of this specification.

[0075] Figure 9 yes Figure 8 The mediation plan provides a structural diagram of some functional units in the model corresponding to case information processing.

[0076] Figure 10 It is a schematic diagram of the principle of the residual structure in the embodiment of this specification.

[0077] Figure 11 It is a flow chart of a training method for a mediation plan providing model suitable for providing mediation plans in an embodiment of this specification.

[0078] Figure 12 It is a structural diagram of a device for providing a mediation plan in an embodiment of this specification.

[0079] Figure 13 It is a structural diagram of a training device for a mediation plan providing model suitable for providing a mediation plan in an embodiment of this specification.

[0080] Figure 14 It is a schematic diagram of the structure of an electronic device in an embodiment of this specification. DETAILED DESCRIPTION

[0081] The number of legal disputes, particularly online transaction disputes, is increasing dramatically. Many online transaction disputes involve smaller amounts, making them more suitable for mediation. To improve mediation efficiency, mediation has gradually shifted from offline to online methods.

[0082] However, current online mediation basically relies on the mediator’s own experience. Therefore, the goal of online mediation is only to break through the spatial barriers, but does not provide corresponding assistance to the mediator in handling disputes.

[0083] In view of this, in the embodiments of this specification, artificial intelligence technology is applied to the field of legal dispute mediation, so that mediation plans can be output intelligently and automatically to better assist mediators in handling dispute mediation matters quickly, efficiently and accurately.

[0084] It should be noted that the embodiments of this specification can be applied in various scenarios where legal dispute mediation is required, such as assisting mediators in online transaction disputes, or in judicial proceedings to assist judges in efficiently handling dispute mediation matters.

[0085] Reference Figure 1A The schematic diagram of an application scenario example in the embodiment of this specification is shown in FIG. In this application scenario, the mediation plan interface 11 is displayed on the screen of the user terminal 10. Figure 1A In the example, the user terminal is a desktop computer, but in other examples, the user terminal 10 may also be a smart phone, a laptop computer, a tablet computer, a smart bracelet, a smart watch or other smart devices, not limited to the diagram.

[0086] The mediation plan interface is illustrative only. Optionally, the mediation plan interface 11 displays multiple "mediation plans" for selection, such as "Mediation Plan 1," "Mediation Plan 2," and "Mediation Plan 3." A mediation plan can be defined as a solution agreed upon between the parties to a dispute, typically including support for the plaintiff's claim and constraints on enforcement measures.

[0087] For example, Figure 1A The document also displays the plaintiff and defendant's demands. Plaintiff A seeks "triple compensation and a refund of the purchase price," while defendant B seeks "a refund of the purchase price." The proposed mediation plan will fall somewhere between these two, tending to lead A and B to an agreement. For example, Mediation Plan One offers "a refund of the purchase price and triple compensation," Mediation Plan Two offers "a refund of the purchase price and double compensation," and Mediation Plan Three offers "a refund of the purchase price."

[0088] Optionally, the mediation plan interface 11 also provides a success rate for each mediation plan, such as 60%, 20%, and 15% in the figure, and can be sorted for selection. The success rate can be predicted by considering one or a combination of factors such as the case, the law, and the parties involved.

[0089] The mediator (or a judge or other mediation personnel) can select a mediation plan to mediate between the parties after referring to the mediation plan interface 11. The mediation plan with a higher success rate is selected, and the parties may reach a consensus more quickly during the mediation.

[0090] refer to Figure 1BThe diagram shown is a schematic diagram of another application scenario example in the embodiment of this specification.

[0091] The scenario in this example can be a scenario where dispute mediation is conducted offline, such as a court or judicial bureau. In these scenarios, the mediator 12 can use, for example, Figure 1A The user terminal 10 in the embodiment assists in the work. During the communication process with the party 13 , the mediation work is performed with reference to the information of each mediation plan provided by the user terminal 10 .

[0092] Refer again Figure 1C The diagram shown is a schematic diagram of another application scenario example in the embodiment of this specification.

[0093] The application scenario in this example can be a scenario of online dispute mediation. In the communication system provided in the figure, the parties 13 and the mediator 12 communicate online with each other through user terminals 10, 14, and 15 accessing the network 16. Each of the user terminals 10, 14, and 15 must have network communication capabilities; each of the user terminals 10, 14, and 15 can be, for example, a smart phone, a tablet computer, a desktop computer, a laptop computer, etc., equipped with a display screen and a wired or wireless communication module; the network 16 can be a wired or wireless Internet network, etc.

[0094] In this scenario, the mediator 12 can communicate with the parties 13 online, and the user terminal 10 of the mediator 12 can Figure 1A The user terminal 10 displays a mediation plan interface, and the mediator 12 can refer to the mediation plan to perform mediation work.

[0095] Optionally, a service terminal 17 (e.g., a server or server group) running the smart mediation platform can be provided in this scenario. Both the mediator 12 and the parties 13 can communicate with the service terminal 17 through their respective user terminals 10, 14, and 15, log in to the smart mediation platform using their respective identity IDs, and complete the mediation work through online communication. The online communication methods include: online multi-party video, multi-party telephone conference, multi-party online conversation, online text communication, etc.

[0096] Optionally, the smart mediation platform can be used to generate the mediation plans and send them to the user terminal 10 held by the mediator 12, for example Figure 1AOptionally, the service terminal 17 and each user terminal 10, 14, 15 can interact based on a browser / server (B / S) architecture, that is, the data transmitted by the smart mediation platform to each user terminal 10, 14, 15 can be displayed in the web browser (Browser) of the user terminal 10, 14, 15, or through a page provided in a platform program (such as an Alipay applet); alternatively, the service terminal 17 and each user terminal 10, 14, 15 can also interact based on a server / client (C / S) architecture, and the mediation plan interface 11 can be displayed on the screen of the user terminal 10 through client software that communicates with the server software.

[0097] The following uses multiple embodiments to illustrate the process and principle of the mediation plan providing method in the embodiments of this specification.

[0098] refer to Figure 2 The flowchart of the method for providing a mediation plan in the embodiment of this specification is shown, and the execution process includes the following steps:

[0099] S21, obtaining case information of the case to be mediated.

[0100] In a specific implementation, the obtaining refers to receiving from the outside or generating locally.

[0101] Among them, the case information of the case to be mediated can come from the text information in the paper or electronic documents of the case to be mediated, such as the text information recorded before the court to describe the case of the case to be mediated, wherein the content may include: party information (such as the name and identity of the plaintiff and defendant), party claims (such as the plaintiff's claim), case process description, etc.; optionally, the recording method includes: handwriting on a paper document, inputting characters or voice into an electronic device through an input device (such as a keyboard, microphone), etc.; wherein, the handwritten recording material (which can be recorded by, for example, a mediator, a clerk, etc.) can be formed into an image by scanning or taking a photo, and then recognized into text information in the form of electronic data by, for example, optical character recognition (OCR) text recognition technology; or, the voice input data can be converted into text information in the form of electronic data by voice recognition technology.

[0102] S22: Obtain mediation reference information of the case to be mediated based on the case information.

[0103] In an optional example, the mediation reference information may include: similar case information.

[0104] like Figure 3As shown, the method for obtaining similar case information in the embodiment of this specification can be as follows:

[0105] Step S31: Extract the first case from the preset judgment document library.

[0106] In a specific implementation, the judicial document library is obtained in the following manner: historical judicial documents are classified according to judicial elements to obtain the judicial document library.

[0107] Specifically, the judicial elements include: representation information of text information related to the legal concepts involved in the case trial. Judicial elements are concepts abstracted from the law. Taking loan dispute cases as an example, assuming that the text information such as "whether there is an electronic agreement", "loan agreement", "routine loan", "delivery of funds", and "payment of funds" recorded in the case information are related to the two elements that need to be identified in the actual trial of loan dispute cases: the agreement between the two parties on private lending (i.e., "loan agreement"), and the fact that the lender must pay the money in private lending cases (i.e., "delivery of funds"), then these text information can be represented as judicial elements through coding.

[0108] Optionally, the judicial document library may be structured, for example, by associating and storing judicial document data marked with judicial elements in a structured database.

[0109] Step S32: Calculate the similarity between the case to be mediated and the first case based on the case information.

[0110] like Figure 4 As shown, the calculation method of the similarity in step S32 may include:

[0111] Step S41: extracting the case text from the case information, and performing vector encoding on the case text of the case to be mediated to obtain a first text semantic feature.

[0112] Step S42: extracting the case text of the first case, and performing vector encoding on the case text of the first case to obtain a second text semantic feature;

[0113] Step S43: Calculate the cosine distance between the first text semantic feature and the second text semantic feature to obtain similarity.

[0114] Vector encoding involves converting the case text in the case information of the first case and the case to be mediated into semantic features in vector form, and then calculating the cosine distance between the semantic features to measure the textual semantic similarity between the case texts of each first case and the case to be mediated. Optional vector conversion methods include one-hot encoding and word embedding.

[0115] Alternatively, a bidirectional long short-term memory (BiLSTM) model can be used to measure text similarity. BiLSTM combines a forward-facing long short-term memory (LSTM) model with a backward-facing LSTM. LSTM is well-suited for natural language processing; LSTM can learn to remember and forget certain information through training, making it easier to capture long-range dependencies. However, LSTM models themselves lack the ability to encode information from the back of a sentence to the front, so BiLSTM can effectively compensate for this. However, it should be noted that the above is merely an example of how similarity calculations can be implemented, and is not intended to be limiting.

[0116] It is understandable that if text semantic features are used for the first match, more noise will be introduced; therefore, in the above example, the first case that is similar to the case to be mediated in judicial elements is first screened out, and then similar cases are screened out from the first case based on the similarity in text semantics, which can effectively improve the efficiency of obtaining similar case information.

[0117] Step S33: The first case whose similarity exceeds a preset threshold is taken as a similar case of the case to be mediated, and similar case information is obtained.

[0118] For example, the similarity can be measured by a probability value between [0, 1], and the preset threshold can be defined at, for example, 0.45, 0.5, 0.55, 0.6, 0.65, 0.7, etc. When the preset threshold is reached, the corresponding first cases are considered to be similar cases, and similar case information is obtained; the similarities can also be sorted to select several with the highest similarity as similar cases.

[0119] In a specific embodiment, the similar case information includes statistical results obtained by analyzing the litigation outcomes of various types of cases, such as distribution information. For example, among 100 similar cases awaiting mediation, 10 had verdict A, such as "the plaintiff won the case, refund was issued, and the amount was doubled in compensation," and 20 had verdict B, such as "the plaintiff won the case, refund was issued, and the amount was doubled in compensation," etc. After statistical analysis, relevant statistical results can be obtained. These statistical results are used as input to the mediation plan provision model to obtain one or more mediation plans for the case awaiting mediation. The mediation plan provision model learns from these statistical results to help accurately predict whether the content of the mediation plan for the case awaiting mediation is consistent with actual precedent.

[0120] In a specific implementation, when extracting the first case from the judicial document database based on the judicial element matching, the corresponding judicial element matching degree can also be calculated, which can be used to be stored / output as an interpretation basis for obtaining the judicial documents of similar cases. Optionally, the judicial element matching degree can be expressed as statistical results on the number, word frequency, and other dimensions of the case information of the case to be mediated and the judicial documents of similar cases having the same or similar judicial elements. For example, the case information of the case to be mediated and the output judicial documents of similar cases both contain the statistical results of the number or word frequency of words such as "food poisoning" and "dining at xx food store". It should be noted that the above are only some example implementations of judicial element matching degree, and are not limitations on other implementation methods.

[0121] refer to Figure 5 The structure diagram of the system for obtaining similar cases and similar case information in the embodiment of this specification is shown.

[0122] In a specific implementation, a structured judicial document library 54 can be constructed. For example, a judicial element library 52 can be constructed by manually annotating some judicial document data in an existing judicial document library 51 and extracting judicial elements from the judicial documents. A sample set (which may include a training set and a test set) is generated using the annotated judicial document data and the corresponding judicial element labels. This is used to train a judicial element parsing model 53 implemented using a machine learning model (e.g., a deep neural network). Once the judicial element parsing model 53 is trained, it can be used to input various other judicial documents, which will be classified into judicial labels by the judicial element parsing model 53 to construct a structured judicial document library 54.

[0123] In the process of obtaining similar cases, a pre-built similar case matching model 55 can be used to complete it; when it is necessary to obtain similar cases and similar case information related to the case to be mediated, the case information of the case to be mediated is input into the similar case matching model 55, so as to extract the first case by matching judicial elements in the judgment document library 54, and then determine similar cases by performing text semantic-based similarity matching on the first case and the case to be mediated, and then obtain similar case information, such as judgment document data of similar cases, statistical results of litigation results of similar cases, etc.

[0124] In a specific implementation, the judicial elements can be extracted from the case information of the case to be mediated based on natural language processing (NLP) technology for use in the matching. Natural language processing (NLP) technology includes sentence segmentation, word segmentation, syntactic analysis, entity recognition, entity relationship extraction, pattern recognition, and information extraction methods.

[0125] Optionally, the similarity matching process performed by the similar case matching model 55 can be as follows: Figure 3 The process is shown in the following figure.

[0126] It should be noted that in actual implementation, Figure 5 The process of constructing the structured judicial document library 54 and the process of obtaining similar case information can be executed by different implementers. For example, enterprise A constructs a structured judicial document library, and enterprise B uses the judicial document library to conduct, for example, Figure 3 The process obtains similar case information, etc., and is not restricted to being implemented by the same implementer, nor is it restricted to being executed continuously in the same process.

[0127] in addition, Figure 5 The principle architecture is only an example and can be changed according to specific implementation requirements. For example, the method of obtaining the structured document library can be changed, and it can be completed by using an existing database or manual classification and indexing; or, the algorithms such as judicial element matching and similarity calculation of the similar case matching model 55 can also be changed, and are not limited to the above description. In addition, Figure 3 、 Figure 4 、 Figure 5 The embodiments are not limited to be applied in the method of providing mediation plans.

[0128] In an optional example, the mediation reference information may include: party characteristics information. This party characteristics information refers to the characteristics of the parties' tendencies in approaching legal issues in the case to be mediated. Optionally, the party characteristics information includes information indicating one or more combinations of the parties' litigation records, enforcement records, and risk preferences. Risk preferences refer to the parties' tolerance for risk, such as investment risk and legal risk.

[0129] Optionally, the party characteristic information may be the result of direct encoding (e.g., through feature extraction) of the party's litigation records, enforcement records, and risk preferences, or the result of encoding statistical information contained in such information. For example, statistical information obtained by analyzing data from the party's litigation records, such as the amount of outstanding debts and the number of lawsuits filed, etc.; risk preferences, such as those that favor low legal risk and a preference for compensation and expedited case resolution to avoid impacting reputation, can all be represented by encoded data, such as in vector form.

[0130] By incorporating the characteristics of the parties into the mediation plan provision model, the model can learn from the parties' profiles of legal disputes when predicting mediation plans, making the predicted mediation plans more accurate and improving the mediation success rate. For example, if the defendant has a history of multiple defaults, the predicted mediation plan may be closer to the plaintiff's claim than if the defendant had no such history.

[0131] In an optional example, the mediation reference information may include: litigation result prediction information. Figure 6 As shown, the claim result prediction information is obtained in the following manner:

[0132] Step S61: Calculating and obtaining legal article prediction information of applicable legal articles of the case to be mediated based on the case information;

[0133] Step S62: predicting support for the party's claim in the case information based on the legal provision prediction information to obtain the claim result prediction information.

[0134] The claim outcome prediction information can be used to represent the legal support for the plaintiff's claim in a pending mediation case. For example, if the plaintiff's claim is not legally supported, the claim outcome prediction information can be used to represent this. For example, the claim outcome prediction information can be expressed as a binary value of 0 or 1, indicating support or disapproval. Alternatively, the claim outcome prediction information can be expressed as a probability value in the range [0, 1], reflecting the probability that the law supports the plaintiff's claim. Alternatively, the claim outcome prediction information can be expressed in a specific encoding format.

[0135] In a specific implementation, a legal issue prediction model can be used to process the input case information of a case to be mediated and output the predicted litigation outcome information. Logically, the legal issue prediction model can predict relevant legal provisions based on the case information of the case to be mediated, using technical means such as similarity matching based on text semantics (which may also include the aforementioned judicial elements). Based on the predicted legal provisions, the model can determine whether the litigation claim is supported to obtain the predicted litigation outcome information.

[0136] In specific implementations, reference can be made to Figure 7 FIG. 7 shows a structural schematic diagram of a legal issue prediction model 70 in an embodiment of the present specification. The legal issue prediction model can be a multi-task learning (MTL) model. The obtaining of the legal provision prediction result is taken as a first task, and the obtaining of the claim prediction result is taken as a second task. In specific implementations, the legal issue prediction model can be implemented by a neural network.

[0137] The legal issue prediction model 70 includes an input layer 71, a hidden layer 72, and an output layer 73.

[0138] In specific implementations, the input information of the legal issue prediction model 70 includes legal provision text information, case information, and case discrete feature information. Optionally, the legal provision text information can include the original text of a legal provision, etc.; the case information is from text information in paper or electronic documents of a case to be mediated, such as text information describing the case in the pre-trial record of the case to be mediated, text information of a complaint, etc.; and the case discrete feature information can be discrete information of the case sorted (such as extracted or counted) by people or machines, such as the number of criminal charge vocabularies, the age range of the defendant, the amount of theft, the weight of drugs, the alcohol concentration in the blood, etc.

[0139] In optional examples, the legal provision text information, the case information, and the case discrete feature information can be converted into vector form in the input layer 71, before being input into the input layer 71, or in the hidden layer 72; that is, a vector conversion unit can be integrated in the input layer 71, or the vector conversion unit can be set outside the legal issue prediction model 70, or a vector conversion unit can be integrated in the hidden layer 72, such as the first layer of the hidden layer, etc. In optional examples, the vector form can include an embedding vector form, by which high-dimensional sparse (or discrete) data can be converted into a low-dimensional vector that retains semantic relationships. For example, for text processing, the corresponding embedding vector can have a word embedding vector, a sentence embedding vector, a paragraph embedding vector, an article embedding vector, etc.

[0140] Optionally, considering the semantic complexity of the case information, a text semantic vector representation model such as a context can be set in the input layer or before the input layer to process the case information to obtain a case information vector, such as an ELMo model pre-trained by related field text, etc. The ELMo model considers the context information, and the word vector obtained under different context semantic conditions is different, which can solve the problem of polysemy, etc.

[0141] Optionally, the discrete feature information of the case can be converted into a sparse vector through discretization when inputting the layer, and then converted into an embedding vector when inputting the hidden layer 72.

[0142] exist Figure 7 In the example, the hidden layer includes multiple encoding layers corresponding to input information processing and output information. These multiple encoding layers include: a legal provision information encoding layer 721, a case text encoding layer 722, a discrete feature encoding layer 723, a legal provision prediction encoding layer 724, and a litigation result encoding layer 725. The legal provision information encoding layer 721, the case text encoding layer 722, and the discrete feature encoding layer 723 can further encode the vectors of legal provision information, case information, and case discrete feature information input from the input layer, respectively, to obtain vectors that more accurately reflect the semantic features of the context.

[0143] For example, corresponding to the processing of legal text information, the legal information encoding layer 721 and the case text encoding layer 722 can have an attention calculation unit, such as using a self-attention algorithm to give higher weight to key information in the legal information, and further encode the vector of the legal information through the attention calculation unit to obtain a text embedding vector that can reflect the context semantics more accurately.

[0144] Corresponding to processing the discrete feature information of the case, the discrete feature encoding layer 723 may have, for example, a factorization machine, a linear regression model, etc., to convert the discrete feature information of the case into a continuous vector, such as an embedding vector; the advantage of using a factorization machine is that when performing vector conversion processing on the discrete feature information of the case, it also associates and combines the various discrete features in the discrete feature information of the case.

[0145] The vectors output by the legal article information encoding layer 721, the case text encoding layer 722, and the discrete feature encoding layer 723 are respectively input into the legal article prediction encoding layer 724 and the litigation result encoding layer 725. Logically speaking, the legal article prediction encoding layer 724 can use the pre-learned relationship between case information, case discrete feature information, and related legal articles to predict the related legal articles based on the input case information and case discrete feature information vectors, encode and generate legal article prediction information (e.g., in vector form) representing the related legal articles, and output it to the output layer. The legal article prediction information can also be input into the litigation result encoding layer 725. The litigation result encoding layer 725 uses the pre-learned relationship between legal article prediction information, case information, case discrete feature information, and litigation result prediction information to encode and generate a vector of litigation result prediction information representing the support of the related legal articles for the litigation request in the case information based on the input case information and case discrete feature information vectors, as well as the legal article prediction information vectors, and outputs it to the output layer.

[0146] The output layer 73 outputs corresponding claim result prediction information according to the vector; optionally, the output layer 73 can also generate legal provision prediction result output according to the legal provision prediction information, to provide information of predicted legal provisions applicable to the case to be mediated.

[0147] In specific implementations, the first task of legal provision prediction, which is actually a multi-label classification task of relevant legal provisions, can adopt a cross-entropy function as a loss function; the second task of claim result prediction, which is a binary classification problem of supporting or not supporting a claim, can also adopt a cross-entropy function as a loss function. For the loss function of multi-task, the comprehensive loss can be calculated by weighted sum or the like.

[0148] In some examples, when training the legal problem prediction model 70, an alternating training manner can be adopted, i.e., the losses respectively output by the loss functions of the first task and the second task are alternately optimized to train the legal problem prediction model, or a joint training manner is adopted, i.e., a joint loss obtained by combining (e.g., calculating a weighted sum or the like) the losses of the first task and the second task is optimized to train the adjustment legal problem prediction model 70.

[0149] S23, inputting the case information and the mediation reference information into a mediation plan providing model to obtain one or more mediation plans of the case to be mediated.

[0150] In specific implementations, each mediation plan content is taken as a label, and the contents of various mediation results in the collected historical cases are taken as labels. The mediation plan model is a multi-classification label model that classifies the labels of mediation plan contents according to the input information.

[0151] Reference Figure 8 The structure of the mediation plan providing model 80 is shown in the schematic diagram. The mediation plan providing model 80 can be implemented based on a deep neural network, including an input layer 81, a hidden layer 82, and an output layer 83.

[0152] The input layer 81 is configured to receive case information of a case to be mediated and mediation reference information. Optionally, the mediation reference information includes at least one of case information, party feature information, and claim result prediction information.

[0153] Although the mediation plan providing model 80 is shown as a deep neural network in the embodiment, the mediation plan providing model 80 can also be implemented by other machine learning algorithms, such as a support vector machine (SVM) algorithm, a decision tree algorithm, a random forest algorithm, a gradient boosting algorithm, a k-nearest neighbor algorithm, a naive Bayes algorithm, a linear regression algorithm, a logistic regression algorithm, or the like. Figure 8The examples illustrate mediation reference information as including predicted litigation outcome information, similar case information, and party characteristics. However, it is understood that this information can be modified in practice based on the accuracy requirements of the actual mediation plan, and is not intended to limit implementation. For example, various pieces of information in the mediation reference information can be modified as needed, such as by adding or deleting information.

[0154] Optionally, the mediation plan providing model 80 includes: at least one vector conversion unit, which is used to convert the input information into the form of an embedded vector for transmission to the back layer. In the case where there are multiple input information, there can also be multiple vector conversion units. For example, Figure 8 In the example, each of the case information, litigation result prediction information, similar case information, and party characteristic information is correspondingly configured with a vector conversion unit, namely, vector conversion units 821, 822, 823, and 824.

[0155] Optionally, case information, litigation result prediction information, similar case information, and party feature information can be converted into a continuous vector with reduced dimensionality, such as an embedding vector, by vector conversion units 821, 822, 823, and 824.

[0156] For example, the case information may be text information, such as text information from pre-court records describing the case to be mediated. Figure 9 In the example, the vector conversion unit 821 includes: a text semantic vector representation model 8211, which is used to convert case information into a context-related first text semantic vector.

[0157] In a specific implementation, the text semantic vector representation model 8211 may include a text pre-training model, that is, a model whose parameters have been pre-trained using a text dataset, such as an ELMo model trained on a large amount of legal text, a BERT (Bidirectional Encoder Representations from Transformers) model, etc. Among them, the ELMo model is based on the BiLM (bidirectional language model). The ELMo model can learn the complexity of vocabulary usage, such as grammar and semantics; it can also learn the polysemy of vocabulary in different contexts; compared with fixed word vectors, ELMo takes contextual information into account, and the word vectors in different contexts are different, which can solve problems such as polysemy and is more suitable for feature extraction of semantically complex case information. The BERT model learns feature representations for words by running a self-supervised learning method based on a large amount of corpus.

[0158] For example, in the ELMo model, the text information of the case to be mediated in a text form can be converted into an embedding vector after being input into the ELMo model after preprocessing (such as word segmentation, etc.), as the first text semantic vector.

[0159] Optionally, in Figure 9 In the mediation plan providing model 80, the attention calculation unit 84 is further adapted to calculate the attention weight of each word corresponding to the first text semantic vector to obtain a second text semantic vector. The attention refers to the attention mechanism in deep learning, which refers to the feature that the attention weight of a person to the elements (such as each word in a sentence or each block in an image) in an object is unevenly distributed, and the key information and non-key information in the information to be processed are distinguished (for example, by weight).

[0160] In a specific implementation, the attention mechanism includes additive attention, multiplicative (dot product) attention, self-attention, and key-value attention, etc. The attention calculation unit 84 can be implemented by an attention model based on one or more of the above attention mechanisms.

[0161] For example, in the ELMo model, the vector output by the ELMo model is input into, for example, a self-attention calculation unit for processing, which can maximize the preservation of the feature semantic information of the text, and obtain an embedding vector that can reflect more accurate context semantics.

[0162] Optionally, the text pre-training model and the attention calculation unit can also be integrated together. For example, if the text pre-training model is a BERT model, the multi-head self-attention (Multi-Head Attention) calculation unit is integrated in the encoder (Encoder) and decoder (Decoder) therein.

[0163] Optionally, for the input of the lawsuit result prediction information, the case information, or the party characteristic information, etc., there can be many discrete features, so a regression model can be used as the corresponding vector conversion unit 822, 823, 824 to convert into a continuous vector, such as a factorization machine (FM), a linear regression model, etc. Among them, considering that the lawsuit result prediction information, the case information, and the party characteristic information can be high-dimensional sparse discrete features, which can cause too many connection parameters of neurons of the neural network, so the factorization machine is more suitable for solving this problem; the processing of the factorization machine is equivalent to low-dimensional embedding of high-dimensional sparse discrete features, on the one hand, it can convert discrete features into continuous dense vectors, which is convenient for subsequent neural network layer calculation; on the other hand, the factorization machine can automatically combine features of discrete features.

[0164] In specific implementation, each of the vector conversion units 821, 822, 823, and 824 can be implemented in the hidden layer 82, for converting the input information into a vector (such as an embedding vector) and transmitting to the back layer.

[0165] In specific implementation, the hidden layer 82 includes a nonlinear transformation layer 825, which is suitable for performing nonlinear transformation on at least part of the features from the input layer 81 direction and outputting to the output layer 83 direction. The nonlinear transformation layer 825 is located at the back layer of each of the vector conversion units 821, 822, 823, and 824, for receiving the vectors output by each of the vector conversion units 821, 822, 823, and 824.

[0166] Optionally, the nonlinear transformation layer 825 can be implemented based on a highway neural network (Highway Network) or a residual neural network (ResNet). Since the depth of the deep neural network needs to be increased to process complex tasks, there are problems such as gradient disappearance during training, and the highway neural network and the residual neural network that appears later have the ability to transmit the original input across layers, thereby solving the training difficulty problem of the deeper deep neural network. Among them, the network depth that can be implemented by the residual neural network is greater than that of the highway neural network, so when implementing the nonlinear transformation layer 825, it can be selected according to the actual neural network depth requirement.

[0167] The forward propagation of the highway network can be simplified as the following expression:

[0168] y = H(x, W H )·T(x, W T )+x·(1-T(x, W T )); (1)

[0169] Where H and T are nonlinear transformation functions; when T = 0, y = x; when T = 1, y = H(x, W H ); From the second half of the formula, we can see that the highway structure uses the function T to control the part of the original feature x that passes directly without undergoing nonlinear transformation.

[0170] refer to Figure 10 The residual structure (residual block) is improved based on the above highway structure and is expressed as follows using formula (2):

[0171] y=F(x,W i )+x; (2)

[0172] Among them, F is the residual function, x represents the input, y represents the expected output; x passes through the weight layer (the weight parameter is W i ) and the partial output F(x,W after the ReLu function i ) is summed with the original input x to obtain y.

[0173] The residual structure of ResNet contains two types of mapping, one is identity mapping (Identity Mapping) and the other is residual mapping (Residual Mapping). The identity mapping refers to itself, that is, the x in the formula, and the residual mapping refers to the "difference", that is, yx, which is the F(x) part in the formula.

[0174] In response to the phenomenon that the accuracy of the training set decreases as the network deepens, ResNet provides two options, namely identity mapping and residual mapping. If the network has reached the target state, continuing to deepen the network will approach 0, leaving only the identity mapping. In this way, the network is theoretically always in the target state, and the network performance will not decrease with increasing depth.

[0175] In actual applications, common deep residual network models include 50 layers, 101 layers, 152 layers, etc.; however, in the scenario of the multi-classification model of the mediation plan in the embodiments of this specification, if the actual network depth does not need to reach the number of layers of the above-mentioned deep residual network model, a highway neural network can be preferred.

[0176] Optionally, the attention calculation unit in the previous example can be integrated into the nonlinear transformation layer. For example, if the nonlinear transformation layer is a highway neural network, the attention calculation unit can be implemented in the highway neural network.

[0177] Optionally, the mediation plan providing model also outputs the mediation success rate corresponding to each mediation plan, that is, Figure 1AFor example, the mediation success rate can be obtained based on the predicted value of the mediation plan provided by the mediation plan providing model. For example, the output layer of the mediation plan providing model includes a Sigmod function for outputting a predicted value of the label corresponding to each mediation plan content, and the success rate of the mediation plan can be obtained based on the predicted value.

[0178] The predicted value output by the Sigmod function for each label is between 0 and 1. This predicted value can be converted into a corresponding success rate, for example, an 80% success rate for mediation plan one, a 50% success rate for mediation plan two, and a 20% success rate for mediation plan three. Alternatively, the success rate can be calculated using the predicted value as one of the reference factors. Of course, other functions, such as the softmax function, are also possible, and the present invention is not limited to this.

[0179] Reference Figure 11 The embodiment of this specification also provides a flow chart of a training method for a mediation plan providing model suitable for providing a mediation plan. The training process may specifically include the following steps:

[0180] S111: Obtain a sample data set.

[0181] The sample dataset includes a training dataset, which includes case information for each historical case, mediation reference information derived from the case information, and mediation results. In some examples, the sample dataset may also include a test dataset. The test dataset and training dataset are simply two parts of the same dataset divided for different purposes.

[0182] Optionally, the mediation reference information includes similar case information; the similar case information is obtained in the following manner: extracting a first case from a preset judicial document library; calculating the similarity between the historical case and the first case based on the case information; taking the first case whose similarity exceeds a preset threshold as a similar case to the historical case, and obtaining similar case information.

[0183] Optionally, the judgment document library is obtained by: classifying historical judgment documents according to judicial elements to obtain the judgment document library. Optionally, the method further includes: performing structured processing on the judgment document library.

[0184] Optionally, extracting the first case from the preset judgment document library includes: extracting the first case whose judicial elements match the historical case from the preset judgment document library.

[0185] Optionally, the calculating the similarity between the historical case and the first case based on the case information includes: extracting the case text from the case information, and performing vector encoding on the case text of the historical case to obtain a first text semantic feature; extracting the case text of the first case, and performing vector encoding on the case text of the first case to obtain a second text semantic feature; calculating the cosine distance between the first text semantic feature and the second text semantic feature to obtain the similarity.

[0186] Optionally, the similar case information includes statistical results of litigation outcomes of similar cases, and the statistical results are used to input into the mediation plan providing model to obtain one or more mediation plans for the historical cases.

[0187] It should be noted that the specific principles for obtaining similar case information can be referred to the previous Figures 3-5 The embodiments are not repeated here.

[0188] Optionally, the mediation reference information also includes characteristic information of the parties, which includes: information representing one or more combinations of the parties' litigation records, execution records, and risk preferences.

[0189] Optionally, the mediation reference information also includes: litigation result prediction information; the litigation result prediction information is obtained in the following manner: calculating the applicable legal provisions of historical cases based on the case information; predicting the support for the parties' litigation in the case information based on the legal provision prediction information to obtain the litigation result prediction information.

[0190] It should be noted that the specific principles for obtaining the prediction information of the litigation results can be referred to the previous Figures 6-7 The embodiments are not repeated here.

[0191] S112: Inputting the sample data set into a mediation plan providing model, so that the mediation plan providing model is trained according to the loss between the mediation plan obtained by the model and the mediation result.

[0192] Optionally, the mediation plan model is a multi-classification label model, which is used to classify the input information to obtain labels about the content of the mediation plan.

[0193] Optionally, the mediation plan providing model is implemented based on a neural network and includes an input layer, a hidden layer, and an output layer; wherein the hidden layer includes a nonlinear transformation layer adapted to perform a nonlinear transformation on at least a portion of the features from the input layer before outputting the features toward the output layer. Optionally, the nonlinear transformation layer is implemented based on a highway neural network or a residual neural network.

[0194] Optionally, the mediation plan providing model includes: at least one vector conversion unit, used to convert the input information into the form of an embedded vector for transmission to the back layer.

[0195] Optionally, the vector is a continuous dense vector.

[0196] Optionally, the mediation reference information includes: case information of historical cases, and the case information is in text form; the vector conversion unit includes: a text semantic vector representation model, which is used to convert the case information into a context-related first text semantic vector.

[0197] Optionally, the text semantic vector representation model includes: a text pre-trained model pre-trained with legal text data.

[0198] Optionally, the mediation plan providing model further includes: an attention calculation unit, adapted to calculate the attention weights of each word corresponding to the first text semantic vector to obtain a second text semantic vector.

[0199] Optionally, the at least one vector conversion unit includes: a regression model, used to convert at least one of the similar case information, party characteristic information, and litigation prediction result information in the mediation reference information into a vector form; wherein the regression model can associate and combine discrete features in the processed information.

[0200] In actual implementation, the mediation plan providing model can be implemented through a deep neural network, and part or all of the various sub-networks in the hidden layer of the mediation plan providing model, such as nonlinear conversion units (such as highway neural networks or residual neural networks, etc.), various vector conversion units (such as factorization machines, text semantic vector representation models), and attention calculation units, etc., can be trained when the mediation plan providing model is trained through the sample data set.

[0201] Since the predicted mediation plan in the embodiment of this specification can be a multi-classification label task model, the cross-entropy loss function can be selected as the loss function of the mediation plan providing model, which is used to measure the difference between the true probability distribution and the predicted value distribution as loss; the true probability distribution comes from the mediation results of the input historical cases, and the predicted value distribution comes from the predicted mediation plans output by the mediation plan providing model.

[0202] The mediation plan providing model is adjusted by inputting a sample data set, and the parameters of the hidden layer of the mediation plan providing model are adjusted for the purpose of optimizing the loss until the output value of the loss function converges to a preset threshold, and the training is determined to be completed.

[0203] In practice, to efficiently minimize the loss function and complete training as quickly as possible, some training methods that can automatically optimize the learning rate can be used. For example, batch gradient descent (BGD), stochastic gradient descent (SGD), momentum, AdaGrad (an adaptive learning rate algorithm), and adaptive moment estimation (Adam) can be used.

[0204] Optionally, the mediation plan providing model further outputs a mediation success rate corresponding to each mediation plan. Optionally, the mediation success rate is obtained based on the predicted value of the obtained mediation plan, that is, for example, the predicted value of the content label of each mediation plan output by the Sigmund function of the output layer.

[0205] In some embodiments of this specification, Figure 12 The schematic diagram of the structure of the mediation plan providing device 120 is shown. The specific implementation of the mediation plan providing device 120 can refer to the previous mediation plan providing method, which will not be repeated here.

[0206] The mediation plan providing device 120 may include:

[0207] The data acquisition unit 121 is used to obtain case information of the case to be mediated;

[0208] An information obtaining unit 122 is configured to obtain mediation reference information of the case to be mediated based on the case information;

[0209] The input unit 123 is configured to input the case information and the mediation reference information into the mediation plan providing model 124 to obtain one or more mediation plans for the case to be mediated.

[0210] Optionally, the mediation reference information includes similar case information; the similar case information is obtained in the following manner:

[0211] Extract the first case from the preset judicial document database;

[0212] Calculating the similarity between the case to be mediated and the first case based on the case information;

[0213] The first case whose similarity exceeds a preset threshold is taken as a similar case of the case to be mediated, and similar case information is obtained.

[0214] Optionally, the judicial document database is obtained by:

[0215] Historical judicial documents are classified according to judicial elements to obtain a judicial document database.

[0216] Optionally, the method further includes:

[0217] The judicial document database is structured.

[0218] Optionally, extracting the first case from a preset judgment document library includes:

[0219] A first case whose judicial elements match the case to be mediated is extracted from the preset judgment document library.

[0220] Optionally, calculating the similarity between the case to be mediated and the first case based on the case information includes:

[0221] Extracting a case text from the case information, and performing vector encoding on the case text of the case to be mediated to obtain a first text semantic feature;

[0222] Extracting the case text of the first case, and performing vector encoding on the case text of the first case to obtain a second text semantic feature;

[0223] Calculate the cosine distance between the first text semantic feature and the second text semantic feature to obtain similarity.

[0224] Optionally, the similar case information includes statistical results of litigation outcomes of similar cases, and the method further includes:

[0225] Compile statistics on the litigation results of multiple similar cases to obtain statistical results;

[0226] The statistical results are used to be input into the mediation plan providing model together with the case information and the similar case information to obtain one or more mediation plans for the case to be mediated.

[0227] Optionally, the mediation reference information also includes characteristic information of the parties, which includes: information representing one or more combinations of the parties' litigation records, execution records, and risk preferences.

[0228] Optionally, the mediation reference information further includes: litigation result prediction information; the litigation result prediction information is obtained by:

[0229] Calculate the legal article prediction information of the applicable legal article of the case to be mediated based on the case information;

[0230] Support for the parties' claims in the case information is predicted based on the legal provision prediction information to obtain the claim result prediction information.

[0231] In some embodiments of this specification, Figure 13 The structure diagram of the training device 130 of the mediation plan providing model suitable for providing mediation plans is shown. The specific implementation of the training device 130 can refer to the training method of the mediation plan providing model, and will not be repeated here.

[0232] The training device 130 includes:

[0233] The data acquisition unit 131 is configured to acquire a sample data set; wherein the sample data set includes a training data set, and the training data set includes: case information about each historical case, mediation reference information obtained based on the case information, and mediation results;

[0234] The training unit 132 inputs the sample data set into the mediation plan providing model 133 so that the mediation plan providing model is trained according to the loss between the mediation plan obtained by the training unit 132 and the mediation result.

[0235] Optionally, the mediation reference information includes similar case information; the similar case information is obtained in the following manner:

[0236] Extract the first case from the preset judicial document database;

[0237] Calculating the similarity between the historical case and the first case based on the case information;

[0238] The first case whose similarity exceeds a preset threshold is regarded as a similar case of the historical case, and similar case information is obtained.

[0239] Optionally, the judicial document database is obtained by:

[0240] Historical judicial documents are classified according to judicial elements to obtain a judicial document database.

[0241] Optionally, the method further includes:

[0242] The judicial document database is structured.

[0243] Optionally, extracting the first case from a preset judgment document library includes:

[0244] A first case whose judicial elements match the historical case is extracted from the preset judgment document library.

[0245] Optionally, calculating the similarity between the historical case and the first case based on the case information includes:

[0246] Extracting case text from the case information, and performing vector encoding on the case text of the historical case to obtain a first text semantic feature;

[0247] Extracting the case text of the first case, and performing vector encoding on the case text of the first case to obtain a second text semantic feature;

[0248] Calculate the cosine distance between the first text semantic feature and the second text semantic feature to obtain similarity.

[0249] Optionally, the similar case information includes statistical results of litigation outcomes of similar cases, and the method further includes:

[0250] Compile statistics on the litigation results of multiple similar cases to obtain statistical results;

[0251] The statistical results are used to be input into the mediation plan providing model together with the case information and the similar case information to obtain one or more mediation plans for the historical cases.

[0252] Optionally, the mediation reference information also includes characteristic information of the parties, which includes: information representing one or more combinations of the parties' litigation records, execution records, and risk preferences.

[0253] Optionally, the mediation reference information further includes: litigation result prediction information; the litigation result prediction information is obtained by:

[0254] Calculate applicable laws and regulations for historical cases based on the case information;

[0255] Support for the parties' claims in the case information is predicted based on the legal provision prediction information to obtain the claim result prediction information.

[0256] It should be noted that in Figure 12 and Figure 13 Each unit in the embodiment may be implemented by software, a combination of software and hardware, or a hardware circuit.

[0257] The embodiment of this specification also provides an electronic device 140, such as Figure 14 The structural diagram of the electronic device 140 shown in the figure, the electronic device 140 may include a memory 141 and a processor 142, the memory 141 stores computer instructions that can be run on the processor, and when the processor runs the computer instructions, it executes the steps of the mediation plan providing method described in any of the aforementioned embodiments or the training method of the mediation plan providing model suitable for mediation plan provision described in any of the aforementioned embodiments.

[0258] In a specific implementation, the processor 142 may be a CPU (Central Processing Unit), an ASIC (Application Specific Integrated Circuit), or one or more integrated circuits configured to implement an embodiment of the present invention. The memory may include high-speed RAM memory, and may also include non-volatile memory (non-volatile memory), such as at least one disk storage. The processor 142 and the memory 141 may communicate via a bus, etc., and the chips may also communicate with each other via corresponding communication interfaces.

[0259] In a specific implementation, the electronic device 140 may further include: a communicator 143; the communicator may include, for example, one or more of a wired network card, a wireless network card, and a 2G / 3G / 4G / 5G module, etc., and can interact with the outside world for information.

[0260] In some examples, the electronic device 140 may be loaded with a trained mediation plan model and run the mediation plan providing method to obtain one or more mediation plans corresponding to the case to be mediated.

[0261] For example, the electronic device 140 may be used in Figure 1A 、 Figure 1B and Figure 1C In the scenario, the service terminal, which is connected to the user terminal in communication, provides one or more mediation plan data to the user terminal 10 of the mediator. The service terminal can be a server / server group or other electronic device 140; Figure 1C In this scenario, the electronic device 140 can be used to implement the "smart mediation platform." The electronic device 140 can obtain input information about the case to be mediated by reading it from a local memory or other communicable storage medium, transmitting it from a user terminal of a mediator or other party, or manually inputting it.

[0262] For another example, the electronic device 140 may also be implemented as Figure 1A 、 Figure 1B ,or Figure 1C The user terminal 10 on the mediator's side has the trained mediation plan model loaded in the local memory, and obtains one or more mediation plans corresponding to the case to be mediated by running the mediation plan providing method locally, and displays the mediation plan interface through the display screen of the user terminal 10.

[0263] An embodiment of this specification also provides a computer-readable storage medium having computer instructions stored thereon, which, when executed, execute the steps of the mediation plan providing method described in any of the aforementioned embodiments or the training method of a mediation plan providing model suitable for providing a mediation plan described in any of the aforementioned embodiments.

[0264] In a specific implementation, the computer-readable storage medium may include, for example, any suitable type of memory unit, memory device, memory item, memory medium, storage device, storage item, storage medium and / or storage unit, such as memory, removable or non-removable media, erasable or non-erasable media, writable or rewritable media, digital or analog media, hard disk, floppy disk, compact disk read-only memory (CD-ROM), compact disk recordable (CD-R), compact disk rewritable (CD-RW), optical disk, magnetic media, magneto-optical media, removable memory cards or disks, various types of digital versatile discs (DVDs), magnetic tapes, cassettes, etc.

[0265] Computer instructions may include any suitable type of code, e.g., source code, compiled code, interpreted code, executable code, static code, dynamic code, encrypted code, etc., implemented using any suitable high-level, low-level, object-oriented, visual, compiled and / or interpreted programming language.

[0266] For the specific implementation, working principle, specific function and effect of each device, system, equipment or system in the embodiments of the present invention, please refer to the specific introduction in the corresponding method embodiment.

[0267] Although the embodiments of this specification are disclosed above, the present invention is not limited thereto. Any person skilled in the art can make various changes and modifications without departing from the spirit and scope of the embodiments of this specification. Therefore, the scope of protection of the present invention shall be based on the scope defined by the claims.

Claims

1. A method for providing a mediation plan, wherein: include: Obtain case information for cases awaiting mediation; Obtaining mediation reference information for the case to be mediated based on the case information, wherein the mediation reference information includes at least one of: similar case information, party characteristic information, and claim prediction result information; the similar case information is used to represent statistical results of litigation outcomes of at least one similar case; the party characteristic information is characteristic information of the parties to the case to be mediated in their tendencies to approach legal issues; and the claim prediction result information is used to represent the legal support for the claims of the parties to the case to be mediated; The case information and the mediation reference information are input into a mediation plan providing model to obtain one or more mediation plans for the case to be mediated and the success rate of the one or more mediation plans, wherein the mediation plan providing model is used to classify the input information and obtain labels about the content of the mediation plan.

2. According to the method for providing a mediation plan according to claim 1, the similar case information is obtained by: Extract the first case from the preset judicial document database; Calculating the similarity between the case to be mediated and the first case based on the case information; The first case whose similarity exceeds a preset threshold is taken as a similar case of the case to be mediated, and similar case information is obtained.

3. The method for providing a mediation plan according to claim 2, wherein: The judicial documents database is obtained in the following manner: Historical judicial documents are classified according to judicial elements to obtain a judicial document database.

4. The method for providing a mediation plan according to claim 3, wherein: The method further comprises: The judicial document database is structured.

5. The method for providing a mediation plan according to claim 3, wherein: The extracting of the first case from the preset judgment document database includes: A first case whose judicial elements match the case to be mediated is extracted from the preset judgment document library.

6. The method for providing a mediation plan according to claim 5, wherein: Calculating the similarity between the case to be mediated and the first case based on the case information includes: Extracting a case text from the case information, and performing vector encoding on the case text of the case to be mediated to obtain a first text semantic feature; Extracting the case text of the first case, and performing vector encoding on the case text of the first case to obtain a second text semantic feature; Calculate the cosine distance between the first text semantic feature and the second text semantic feature to obtain similarity.

7. The method for providing a mediation plan according to claim 1, wherein the characteristic information of the parties includes: Information representing one or more combinations of a party’s litigation record, execution record, and risk preference.

8. The method for providing a mediation plan according to claim 1, wherein the litigation result prediction information is obtained by: Calculate the legal article prediction information of the applicable legal article of the case to be mediated based on the case information; Support for the parties' claims in the case information is predicted based on the legal provision prediction information to obtain the claim result prediction information.

9. A training method for a mediation plan provision model suitable for providing mediation plans, wherein: include: Obtaining a sample data set; wherein the sample data set includes a training data set, the training data set including: case information of each historical case, mediation reference information obtained based on the case information, and mediation results; the mediation reference information including: at least one of similar case information, party characteristic information, and claim prediction result information; the similar case information is used to represent statistical results of litigation results of at least one similar case; the party characteristic information is characteristic information of the parties in the historical case in their tendencies in dealing with legal issues; and the claim prediction result information is used to represent the legal support of the parties in the historical case; The sample data set is input into a mediation plan providing model so that the mediation plan providing model is trained based on the loss between the mediation plan obtained and the mediation result, wherein the trained mediation plan providing model is used to output one or more mediation plans for the case to be mediated and the success rate of the one or more mediation plans based on the case information and mediation reference information of the case to be mediated, and the mediation plan providing model is used to classify the input information to obtain a label about the content of the mediation plan.

10. The training method according to claim 9, wherein the similar case information is obtained by: Extract the first case from the preset judicial document database; Calculating the similarity between the historical case and the first case based on the case information; The first case whose similarity exceeds a preset threshold is regarded as a similar case of the historical case, and similar case information is obtained.

11. The training method according to claim 10, wherein: The judicial documents database is obtained in the following manner: Historical judicial documents are classified according to judicial elements to obtain a judicial document database.

12. The training method according to claim 11, wherein: The method further comprises: The judicial document database is structured.

13. The training method according to claim 11, wherein: The extracting of the first case from the preset judgment document database includes: A first case whose judicial elements match the historical case is extracted from the preset judgment document library.

14. The training method according to claim 13, wherein: Calculating the similarity between the historical case and the first case based on the case information includes: Extracting case text from the case information, and performing vector encoding on the case text of the historical case to obtain a first text semantic feature; Extracting the case text of the first case, and performing vector encoding on the case text of the first case to obtain a second text semantic feature; Calculate the cosine distance between the first text semantic feature and the second text semantic feature to obtain similarity.

15. The training method according to claim 9, wherein the characteristic information of the party concerned includes: Information representing one or more combinations of a party’s litigation record, execution record, and risk preference.

16. The training method according to claim 9, wherein the claim result prediction information is obtained by: Calculate the applicable laws of historical cases based on the case information; Support for the parties' claims in the case information is predicted based on the legal provision prediction information to obtain the claim result prediction information.

17. A device for providing a mediation plan, wherein: include: A data acquisition unit, used to obtain case information of the case to be mediated; an information acquisition unit, configured to obtain mediation reference information for the case to be mediated based on the case information, wherein the mediation reference information includes at least one of similar case information, party characteristic information, and claim prediction result information, wherein the similar case information is used to represent statistical results of litigation outcomes of at least one similar case, the party characteristic information is characteristic information of the parties to the case to be mediated in their tendencies to approach legal issues, and the claim prediction result information is used to represent the legal support for the claims of the parties to the case to be mediated; An input unit is used to input the case information and the mediation reference information into a mediation plan providing model to obtain one or more mediation plans for the case to be mediated and the success rate of the one or more mediation plans. The mediation plan providing model is used to classify the input information and obtain a label about the content of the mediation plan.

18. A training device for a mediation plan provision model suitable for providing mediation plans, wherein: include: a data acquisition unit configured to obtain a sample data set; wherein the sample data set includes a training data set, the training data set including: case information regarding each historical case, mediation reference information obtained based on the case information, and mediation results; the mediation reference information including: at least one of similar case information, party characteristic information, and claim prediction result information; the similar case information is used to represent statistical results of litigation results of at least one similar case; the party characteristic information is characteristic information of the parties in the historical case in their tendencies in approaching legal issues; and the claim prediction result information is used to represent the legal support of the parties in the historical case; A training unit inputs the sample data set into a mediation plan providing model so that the mediation plan providing model is trained according to the loss between the mediation plan obtained by it and the mediation result, wherein the trained mediation plan providing model is used to output one or more mediation plans for the case to be mediated and the success rate of the one or more mediation plans based on the case information and mediation reference information of the case to be mediated, and the mediation plan providing model is used to classify according to the input information to obtain a label about the content of the mediation plan.

19. An electronic device comprising a memory and a processor, wherein the memory stores computer instructions that can be executed on the processor, wherein: When the processor runs the computer instructions, it executes the method for providing a mediation plan according to any one of claims 1 to 8 or the training method according to any one of claims 9 to 16.

20. A computer-readable storage medium having computer instructions stored thereon, wherein: When the computer instructions are executed, the method for providing a mediation plan according to any one of claims 1 to 8 or the training method according to any one of claims 9 to 16 is executed.

Images