WO2020052184A1 - 裁判文书处理方法、装置、计算机设备和存储介质 - Google Patents

裁判文书处理方法、装置、计算机设备和存储介质 Download PDF

Info

Publication number
WO2020052184A1
WO2020052184A1 PCT/CN2019/071516 CN2019071516W WO2020052184A1 WO 2020052184 A1 WO2020052184 A1 WO 2020052184A1 CN 2019071516 W CN2019071516 W CN 2019071516W WO 2020052184 A1 WO2020052184 A1 WO 2020052184A1
Authority
WO
WIPO (PCT)
Prior art keywords
paragraph
judgment
amount
preset
target
Prior art date
Application number
PCT/CN2019/071516
Other languages
English (en)
French (fr)
Inventor
叶素兰
窦文伟
毛皎龙
刘媛源
苏晓明
Original Assignee
平安科技（深圳）有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 平安科技（深圳）有限公司 filed Critical 平安科技（深圳）有限公司
Publication of WO2020052184A1 publication Critical patent/WO2020052184A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F40/00—Handling natural language data
    • G06F40/30—Semantic analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/10—Services
    • G06Q50/18—Legal services

Definitions

  • the present application relates to a method, device, computer equipment, and storage medium for processing referee documents.
  • Judgment documents refer to the legal documents of the results of the adjudication, and are the documents that record the process and results of the trial proceedings of the people's court.
  • the adjudication documents include the plaintiff, lawyer, court of trial, plaintiff's claim, court decision, and case acceptance fee.
  • you can understand the relevant situation of the case For example, based on the judgment documents, you can understand the original court, lawyers and courts involved in the case.
  • the content of the plaintiff request and judgment in the judgment document, and the review process can reflect the professional ability of the lawyer.
  • the ruling documents have a fixed format, information such as the plaintiffs, lawyers, and courts of trial can be automatically extracted through specified rules.
  • the writing style of the arbitral documents differs greatly, and there is no fixed law.
  • the professional competence of lawyers is usually determined based on the results of the analysis by manually locating and analyzing the claims, judgment content, and review process in the judgment documents.
  • this analysis method of lawyer's professional ability requires a lot of manpower and material resources, and the analysis process will take a lot of time, and there is a problem of inefficient processing of referee documents, resulting in a low analysis efficiency of lawyer's professional ability .
  • a method, a device, a computer device, and a storage medium for processing a referee document are provided.
  • a referee document processing method includes:
  • the plaintiff's lawyer's success rate and the court's lawyer's impairment rate are calculated respectively.
  • a referee document processing device includes:
  • Amount item extraction module for extracting a claim amount item from the claim paragraph through a trained entity recognition model, and extracting a judgement amount item from the decision paragraph;
  • Amount value extraction module for extracting the claim corresponding to the claim amount item from the claim paragraph based on the preset amount item expression when the claim amount item and the judgment amount item are extracted. Requesting the value of the amount, and extracting the value of the judgment amount corresponding to the judgment amount item from the judgment paragraph; and
  • a calculation module is configured to calculate the plaintiff's lawyer's success rate and the Yankee's lawyer's impairment rate based on the claim amount and the corresponding claim amount, and the judgment amount and the corresponding judgment amount respectively.
  • a computer device includes a memory and one or more processors.
  • the memory stores computer-readable instructions, and the computer-readable instructions, when executed by the one or more processors, cause the one or more The processors implement the steps of the referee document processing method provided in any one of the embodiments of the present application.
  • One or more non-volatile computer-readable storage media storing computer-readable instructions, which when executed by one or more processors, cause the one or more processors to implement any The steps of the referee document processing method provided in one embodiment.
  • FIG. 1 is an application scenario diagram of a referee document processing method according to one or more embodiments.
  • FIG. 2 is a schematic flowchart of a referee document processing method according to one or more embodiments.
  • FIG. 3 is a schematic flowchart of a referee document processing method in another embodiment.
  • FIG. 4 is a block diagram of a referee word processing device according to one or more embodiments.
  • FIG. 5 is a block diagram of a referee word processing device in another embodiment.
  • FIG. 6 is a block diagram of a computer device according to one or more embodiments.
  • the judgment document processing method provided in this application can be applied to the application environment shown in FIG. 1.
  • the terminal 102 communicates with the server 104 through the network through the network.
  • the server 104 extracts the appeal paragraph and the judgment paragraph from the obtained referee document through the trained paragraph extraction model, and extracts the claim amount item from the appeal paragraph through the trained entity recognition model and the judgment paragraph.
  • the judgment amount is further extracted based on the preset amount expression, and the claim amount corresponding to the claim amount and the judgment amount corresponding to the judgment amount are respectively extracted to calculate the plaintiff according to the extracted claim amount and the judgment amount.
  • the lawyer's winning rate and the lawyer's impairment rate, and the calculated plaintiff's lawyer's winning rate and the court's lawyer's impairment rate are sent to the terminal 102.
  • the terminal 102 may be, but is not limited to, various personal computers, notebook computers, smart phones, tablet computers, and portable wearable devices.
  • the server 104 may be implemented by an independent server or a server cluster composed of multiple servers.
  • a method for processing a referee document is provided.
  • the method is applied to the server in FIG. 1 as an example, and includes the following steps:
  • Judgment documents are legal documents that record the adjudication process and results, and are documents that record the process and results of litigation activities conducted by the people's courts.
  • the judgment documents include the plaintiff, the court, the plaintiff's lawyer, the court's lawyer, the court of trial, the plaintiff's litigation request, the result of the court's judgment and the case acceptance fee.
  • the server when the server receives the referee document processing instruction, the server obtains the corresponding referee document according to the received referee document processing instruction.
  • the server correspondingly queries the pre-stored referee documents locally according to the received referee document processing instructions.
  • the server may specifically receive the referee document processing instruction sent by the terminal, and obtain the corresponding referee document from the terminal according to the received referee document instruction.
  • the referee document processing instruction carries a referee document to be processed.
  • the server parses the received referee document processing instruction and obtains the corresponding referee document.
  • the paragraph extraction model is a model obtained by performing model training according to a pre-obtained training sample set, and is used to correspondingly extract an appeal paragraph and a judgment paragraph from a referee document.
  • the claim paragraph is the paragraph used to describe the plaintiff's claim.
  • a judgment paragraph is a paragraph that describes the outcome of a court decision.
  • the server inputs the obtained referee document into a pre-trained paragraph extraction model, performs prediction through the paragraph extracting model, and obtains corresponding appeal paragraphs and judgment paragraphs respectively, so as to extract corresponding judgement documents from the referee documents.
  • Appeal and judgment paragraphs are the following paragraphs.
  • the server extracts the appeal paragraph and the judgment paragraph from the obtained referee documents respectively by using the pre-trained first paragraph extraction model and the second paragraph extraction model. Specifically, the server inputs the obtained referee document into a pre-trained first paragraph extraction model for prediction, and obtains an appeal paragraph in the referee document. Similarly, the server inputs the obtained judgment document into a pre-trained second paragraph extraction model for prediction, and obtains the judgment paragraph in the judgment document.
  • the first paragraph extraction model is a paragraph extraction model obtained by performing model training according to the target referee documents and corresponding target appeal paragraphs
  • the second paragraph extraction model is a paragraph extraction model obtained by performing model training according to the target referee documents and corresponding target judgment paragraphs.
  • the pre-trained paragraph extraction model is a long-term memory neural network model.
  • the server obtains the referee document, it obtains the extraction paragraph of the appeal paragraph and the extraction paragraph of the judgment paragraph respectively.
  • the server inputs the obtained judgment document and the application paragraph extraction question into a pre-trained paragraph extraction model for prediction, and obtains the application paragraph in the judgment document. Further, the server inputs the judgment document and the obtained judgment paragraph extraction problem into the pre-trained paragraph extraction model for prediction, and obtains the judgment paragraph in the judgment document.
  • the server extracts the appeal paragraph and the judgment paragraph from the referee document through a pre-trained paragraph extraction model, and then filters the extracted appeal paragraph based on a preset appeal expression,
  • the claim section below performs the following steps to extract the claim amount and the value of the claim amount.
  • the server filters the extracted decision paragraphs based on a preset decision expression, and performs the following related steps of extracting a judgment amount item and a judgment amount value on the filtered decision paragraphs.
  • the preset claim expression refers to a preset regular expression used to filter claim paragraphs through regular matching.
  • the preset decision expression refers to a preset regular expression for filtering decision paragraphs through a regular matching manner.
  • the entity recognition model is a model obtained by performing model training according to a pre-obtained training sample set and used to extract an amount item from the obtained paragraph.
  • the claim amount refers to the damage compensation item specified in the plaintiff's claim.
  • the claim amount can specifically refer to the damages items that the plaintiff claimed in the plaintiff's litigation request for the court's compensation, such as mental loss, medical costs, nutrition costs, and child support.
  • the judgment amount refers to the damage compensation items specified in the court judgment result, which are to be compensated by the court to the plaintiff, such as mental loss, medical expenses and child support.
  • the amount of claim and the amount of judgment may be the same or different.
  • the server inputs the claim paragraph extracted from the referee document into a pre-trained entity recognition model, performs prediction through the entity recognition model, and obtains the claim amount item in the claim paragraph.
  • the server inputs the judgment paragraph extracted from the judgment document into a pre-trained entity recognition model for prediction, and obtains the judgment amount item in the judgment paragraph.
  • the entity recognition model that extracts the claim amount item from the claim paragraph may be the same entity recognition model as the entity recognition model that extracts the decision paragraph from the decision paragraph, or may be based on different training sample sets. Train different entity recognition models.
  • the preset amount item expression refers to a preset regular expression used to extract an amount value from a specified paragraph by a regular matching method.
  • the expression of the preset amount item may specifically be a regular expression that extracts an amount value corresponding to the amount item from the specified paragraph, such as "mental loss fee. *? Yuan".
  • the claim amount value refers to the amount of compensation specified in the plaintiff's claim.
  • the value of the claim amount may specifically refer to the amount of compensation paid by the court in the plaintiff's claim, such as 20,000 yuan.
  • the amount of claim corresponds to the value of claim.
  • the value of the judgment amount refers to the amount of compensation specified by the court in the result of the court's award to the plaintiff, such as 10,000 yuan.
  • the server when the claim amount item is extracted from the appeal paragraph and the judgment amount item is extracted from the judgment paragraph, the server performs the preset amount expression corresponding to the appeal paragraph and the extracted claim paragraph. Match to extract the claim amount value corresponding to the claim amount item from the claim paragraph. Similarly, the server matches a preset amount expression corresponding to the decision paragraph with the extracted decision paragraph to extract a judgment amount value corresponding to the decision amount from the decision paragraph.
  • the expression of the preset amount corresponding to the appeal paragraph may be the same as or different from the expression of the preset amount corresponding to the judgment paragraph.
  • the plaintiff's lawyer's success rate refers to the quantified value of the plaintiff's lawyers fighting for benefits / compensation for the plaintiff.
  • Defendant's lawyer's impairment rate refers to the quantified value of the court's lawyer's reduction of loss / compensation for the court.
  • the plaintiff's lawyer's success rate may specifically be the ratio of the total amount of the judgment in the court decision result to the total amount of the claim in the plaintiff's litigation request.
  • the deduction rate of the court's lawyer can be the difference between the total amount of the claim in the plaintiff's lawsuit and the total amount of the judgment in the court's decision, which accounts for the total amount of the claim in the plaintiff's lawsuit.
  • the server determines the corresponding total claim amount according to the claim amount item and the corresponding claim amount value extracted from the claim paragraph, and according to the judgment amount item and the corresponding judgment amount value extracted from the decision paragraph Determine the total amount of the corresponding judgment. Further, the server calculates the plaintiff's lawyer's winning rate and the court's lawyer's impairment rate respectively according to the determined total amount of claims and the total amount of judgment, according to the first preset calculation method.
  • the first preset calculation method is a preset calculation method for instructing the server to calculate the winning rate of the plaintiff's lawyer and the reducing rate of the lawyer according to the determined total amount of the claim and the total amount of the judgment.
  • the above judgment document processing method automatically extracts the corresponding appeal paragraphs and judgment paragraphs from the obtained judgment documents through the paragraph extraction model, and then automatically extracts the claim amount items from the extracted appeal paragraphs through the entity recognition model. , And automatically extract the judgment amount from the judgment paragraph, which improves the extraction efficiency of the judgment amount in the judgment documents, thereby improving the processing efficiency of the judgment documents.
  • the corresponding claim amount and judgment amount are automatically extracted based on the preset amount expression and the extracted claim amount and judgment amount, and the amount is increased.
  • the value extraction efficiency further improves the processing efficiency of judgment documents, thereby improving the analysis efficiency of lawyers' professional capabilities.
  • the method for processing a judgment document further includes: when no claim amount and judgment amount are extracted, extracting the case acceptance fee paragraph from the judgment paragraph based on the preset acceptance fee expression; based on the preset The expression of acceptance fee sharing, extracts the acceptance fee distribution data from the case acceptance fee paragraph; and calculates the plaintiff's lawyer's winning rate and the court's lawyer's impairment rate based on the receiving fee's distribution data.
  • the preset acceptance fee expression is a preset regular expression for extracting the case acceptance fee paragraph from the judgment paragraph by a regular matching method.
  • the case acceptance fee paragraph in the adjudication document is usually a separate paragraph in the judgment paragraph, and usually has specified keywords, such as case acceptance fee, case acceptance fee, litigation fee and case appeal fee, etc. It is ". *? Acceptance fee”.
  • the case acceptance fee paragraph refers to the paragraph used to describe the court's designation of the case acceptance fee.
  • the case processing fee paragraph is specifically used to describe the designated processing fee and the distribution of the processing fee.
  • the preset processing fee sharing expression is a preset regular expression for extracting processing fee allocation data from the case processing fee paragraph by means of regular matching.
  • Preset acceptance fee sharing expressions such as ". *? Fee. *? Yuan. *? Plaintiff. *? Burden” or ". *? Defendant. *? Burden. *? Yuan” etc.
  • Acceptance fee allocation data refers to quantified data on the allocation of case acceptance fees.
  • the processing fee allocation data can be used to characterize the processing fees borne by the plaintiff and the lawyer respectively.
  • the processing fee allocation data may include the processing fee borne by the plaintiff and the processing fee borne by the court.
  • the processing fee sharing data may also include the total case processing fee specified in the court judgment result, and / or other costs involved in the trial of the case, such as security fees and announcement fees.
  • the processing fee sharing data can also include the legal status, name and burden of the processing fee sharing object. It is understandable that the object of apportionment fees can be individuals or companies.
  • the server matches the preset acceptance fee expression with the extracted judgment paragraph to The corresponding case acceptance fee paragraph was extracted from the judgment paragraph. Further, the server matches a preset expression of processing fee sharing with the extracted case processing fee paragraph to extract corresponding processing fee sharing data from the case processing fee paragraph.
  • the server correspondingly determines the total case acceptance fee, the plaintiff bears the case acceptance fee, and the court bears the case acceptance fee according to the extracted acceptance fee sharing data.
  • the server calculates the corresponding plaintiff's lawyer's success rate and lawyer's impairment rate according to the determined total case acceptance fee, the plaintiff's case acceptance fee, and the court's case acceptance fee, respectively, according to the second preset calculation method.
  • the second preset calculation method is a preset calculation method, which is used to instruct the server to calculate the winning rate of the plaintiff's lawyer and the impairment rate of the Yankee's lawyer correspondingly according to the extracted processing fee sharing data.
  • the server when the claim amount is not extracted from the appeal paragraph or the judgment paragraph is not extracted, the server extracts the corresponding case acceptance from the judgment paragraph based on the preset acceptance fee expression. Fee paragraph.
  • the server when the server extracts the processing fee allocation data from the case processing fee paragraph, the server preprocesses the extracted processing fee sharing data, and then calculates the processing fee accordingly based on the preprocessed processing fee sharing data.
  • Pre-processing includes, but is not limited to, deduplication of the amount value in the acceptance fee allocation data, or converting the amount value in the acceptance fee allocation data that does not conform to a preset format into an amount value in a preset standard form.
  • the server extracts the specified keywords in the case acceptance fee paragraph.
  • the server pre-processes the acceptance fee allocation data according to a preset pre-processing method corresponding to the extracted specified keywords.
  • the specified keywords include, but are not limited to, change, increase, half charge, total, total, total, ten thousand yuan and one hundred million yuan.
  • the specified keywords can also be "percent” or “percent”, or the percent symbol "%”.
  • the amount value in the processing fee allocation data is deduplicated; when keywords such as percentage, ten thousand yuan, and / or hundred million yuan are extracted , It indicates that the amount corresponding to the keyword is an amount that does not conform to a preset format, and the server converts the amount to an amount in a preset standard form.
  • case acceptance fee paragraph is "case acceptance fee of 20,000 yuan, halved collection of 10,000 yuan, security fee of 5,000 yuan, a total of 15,000 yuan, 10% by Plaintiff A and 90% by Defendant B" .
  • the specified keywords extracted by the server from the case acceptance fee paragraph are "ten thousand yuan”, “half charge”, “total” and "%”.
  • the server removes the value “20,000 yuan” before the keyword according to the specified keyword “half charge”, and removes the duplicate value “10,000 yuan” and “5,000 yuan” before the keyword according to the specified keyword “total” .
  • the server converts the corresponding amount value to the preset standard form according to the keywords "10,000 yuan” and "%", that is, 10,000 yuan to 10,000 yuan, 10% to 1500 yuan, and 90% to 13,500 yuan. It is worth noting that the pre-processing of the application fee sharing data is not limited to the above examples.
  • extracting the processing fee allocation data from the case processing fee paragraph based on the preset processing fee sharing expression including: sequentially extracting the preset keywords in the case processing fee paragraph in accordance with the semantic order; and according to the preset keywords According to the preset classification conditions, determine the type of acceptance fee allocation for the case acceptance fee paragraph. According to the preset type of acceptance fee allocation expression corresponding to the acceptance fee allocation type, extract the acceptance fee allocation data from the case acceptance fee paragraph.
  • Semantic order refers to the logical order of language.
  • the semantic order may specifically refer to the order in which the words constituting the case acceptance fee paragraph appear, that is, the word order.
  • the preset keywords are keywords that are set in advance, such as "fees", “total”, and “burden”.
  • the processing fee allocation type refers to the type of allocation corresponding to the processing fee allocation data in the case processing fee paragraph.
  • the expenses and apportionment in the case acceptance fee paragraph correspond to at least one combination form, and the composition of the fee and apportionment situation corresponds to the apportionment fee apportionment type.
  • the combined form of expenses and allocation includes but is not limited to the allocation of a single expense, the total allocation of multiple expenses, and the separate allocation of multiple expenses.
  • the server extracts corresponding preset keywords from the case acceptance fee paragraphs in sequence from the front to back according to the semantic order of the case acceptance fee paragraphs.
  • the server determines, according to the extracted preset keywords and the extraction order of the preset keywords, the type of processing fee allocation corresponding to the case processing fee paragraph according to the preset classification conditions.
  • the server determines the preset processing fee sharing expression according to the type of processing fee allocation corresponding to the case processing fee paragraph, and matches the determined preset processing fee sharing expression with the case processing fee paragraph so as to extract from the case processing fee paragraph. Extract the corresponding processing fee allocation data.
  • the server determines the number of the first preset keywords contained in the extracted preset keywords correspondingly.
  • the server determines the processing fee allocation type corresponding to the corresponding case processing fee paragraph as the first allocation type.
  • the server sequentially judges whether there is a second preset among the extracted preset keywords in accordance with the extraction order of the preset keywords. The next preset keyword of the keyword is the first preset keyword. If it does not exist, the server determines the corresponding processing fee allocation type as the second allocation type; if it exists, the server determines the corresponding processing fee allocation type as the third allocation type.
  • the preset keywords include a first preset keyword and a second preset keyword, the first preset keywords such as "fees", "total” and “total”, and the second preset keywords such as "burden” and "Commit” and so on.
  • the case acceptance fee paragraph is "case acceptance fee of 1,000 yuan, which shall be borne by the plaintiff A".
  • the preset keywords extracted from the case acceptance fee paragraph are "fees” and "burden”.
  • the preset The keyword contains a first preset keyword, so the corresponding type of acceptance fee allocation is determined as the first allocation type.
  • the case acceptance fee paragraph is "the case acceptance fee is 1,000 yuan, the security fee is 500 yuan, and the plaintiff A shall bear it"
  • the case acceptance fee paragraph contains 2 first preset keywords and there is no second preset
  • the next preset keyword for the keywords is the first preset keyword, so the type of acceptance fee allocation is determined as the second allocation type.
  • case acceptance fee paragraph is "case acceptance fee of 1,000 yuan, borne by plaintiff A, security fee of 500 yuan, borne by court B"
  • case acceptance fee paragraph contains 2 first preset keywords and exists
  • the next keyword of the second preset keyword is the first preset keyword, so the type of acceptance fee allocation is determined as the third allocation type.
  • the server pre-stores a corresponding preset regular expression.
  • the server matches the preset regular expression corresponding to each type of processing fee allocation type with the case processing fee paragraph, respectively. When the match is successful, it determines the processing fee distribution type corresponding to the preset regular expression that matches successfully, as The type of apportionment fee corresponding to the case acceptance fee paragraph.
  • the corresponding preset acceptance fee allocation expression and the preset regular expression may be the same or different.
  • the server when the type of processing fee allocation corresponding to the case processing fee paragraph is the first type of allocation, the server extracts from the case processing fee paragraph based on a preset processing fee sharing expression corresponding to the first type of distribution. Corresponding processing fee allocation data. Similarly, when the processing fee allocation type corresponding to the case acceptance fee paragraph is the second allocation type, the server extracts the corresponding processing fee from the case acceptance fee paragraph based on the preset processing fee allocation expression corresponding to the second allocation type. Share the data.
  • the processing fee allocation data may include, but is not limited to, the total amount of case processing fees, the processing fee distribution objects, and the amount allocated by each processing fee distribution object.
  • the server divides the case acceptance fee paragraph into more than one unit sentence according to a preset split condition.
  • the acceptance fee allocation type corresponding to each unit sentence is the first allocation type or the second allocation type.
  • the server extracts, from the unit sentence, the acceptance fee allocation data corresponding to the unit sentence based on the preset acceptance fee allocation expression corresponding to the acceptance fee allocation model corresponding to the unit sentence.
  • the server determines the processing fee allocation data corresponding to the processing fee paragraph of the corresponding case according to the processing fee allocation data extracted from each unit sentence.
  • case acceptance fee paragraph is "case acceptance fee of 1,000 yuan, security fee of 500 yuan, a total of 1,500 yuan, borne by the plaintiff A, announcement costs of 600 yuan, borne by the court B", in that order from the case acceptance fee paragraph
  • the server determines that the type of acceptance fee corresponding to the case acceptance fee paragraph is the third type of assessment.
  • the server divides the case acceptance fee paragraph between the sentences where the keywords “burden” and “fees” are preset, and the corresponding two unit sentences are “case acceptance fee 1000 yuan and security fee 500 yuan, A total of 1,500 yuan, which shall be borne by Plaintiff A "and” Announcement Fee 600 yuan, which shall be borne by Defendant B ".
  • the server may abstract the extracted processing fee allocation data into a corresponding fee allocation model.
  • the acceptance fee sharing model is, for example, "cost X, spreader A pays Y, sharer B pays Z,” or "costs total X, sharer A pays Y, and sharer B pays Z".
  • the server divides it into unit sentences of the first assessment type and / or the second assessment type.
  • the corresponding acceptance fee allocation data is extracted from the case acceptance fee paragraphs through different preset acceptance fee allocation expressions, which improves the extraction of the admission fee allocation data. Efficiency, thereby increasing the efficiency of the analysis of lawyers' professional capabilities.
  • the method for processing a referee document further includes: extracting a target tag from the referee document based on a preset tag expression; and when the target tag does not match the preset tag set, execute step S210.
  • the preset label expression is a preset regular expression for extracting a target label from a referee document by a regular matching method.
  • the preset label expression can be specifically used to extract the target label from the decision paragraph in a regular matching manner.
  • Target labels refer to pre-designated document labels extracted from the judgment documents, such as rejection, withdrawal, retrial, trademark disputes, trademark infringement disputes, and second instance trials.
  • the preset label set is a label set composed of more than one preset label.
  • the preset label is a preset document label.
  • the preset labels include, but are not limited to, dismissal, withdrawal, retrial, trademark disputes, trademark infringement disputes, and second instance trials.
  • the server matches the preset tag expression with each paragraph in the referee document to extract the corresponding target tag from the referee document, and matches the extracted target tag with the preset tag set.
  • the server correspondingly determines the calculation method of the plaintiff's lawyer's winning rate and the court's lawyer's impairment rate according to the matching results, and according to the determined calculation method, the plaintiff's lawyer's winning rate and the court's lawyer's impairment rate are correspondingly calculated.
  • the server When the matching result is a matching failure, that is, when the target tag does not match the preset tag set, the server according to the extracted claim amount item and the corresponding claim amount value, and the judgment amount item and the corresponding judgment amount value, Calculate the plaintiff's lawyer's winning rate and the court's lawyer's impairment rate separately.
  • the server when the matching result is a successful match, that is, when the target tag matches the preset tag set, the server correspondingly determines that the plaintiff's lawyer wins the case according to the preset tag in the preset tag set that matches the target tag.
  • the calculation method of the rate and the lawyer's impairment rate, and the plaintiff's lawyer's winning rate and the court's lawyer's impairment rate shall be calculated correspondingly according to the determined calculation method.
  • the preset label set includes an arbitration label, an intellectual property label, and a second-instance label.
  • the server sorts different types of preset tags in the preset tag set according to priority, and matches the target tags with the preset tags in the preset tag set in order according to the priority order.
  • the priority ranking may specifically be that the referee label has priority over the intellectual property label, and the intellectual property label has priority over the second instance label.
  • Ruling labels can include multiple labels, such as dismissal, withdrawal, and retrial.
  • Intellectual property labels can include multiple labels, such as disputes over ownership of trademark rights and disputes over trademark infringement.
  • the server matches the target tag with the ruling tag in the preset tag set.
  • the server determines the plaintiff's lawyer's winning rate and the court's lawyer's impairment rate correspondingly according to the successfully matched ruling tag.
  • the server matches the target tag with the knowledge product tag in the preset tag set.
  • the server uses the successfully matched intellectual property tag Correspondingly determine the plaintiff's lawyer's winning rate and the court's lawyer's impairment rate.
  • the server matches the target tag with the second-trial tag in the preset tag set.
  • the server accepts the case from the extracted case.
  • the data of appropriation of appropriation fee is extracted from the fee paragraph, and the winning rate of the plaintiff's lawyer and the rate of impairment of the court's lawyer are correspondingly calculated according to the extracted apportionment fee apportionment data.
  • the server determines that the plaintiff's lawyer's success rate is 0 and the court's lawyer's impairment rate is 100%.
  • the server determines that the plaintiff's lawyer's success rate is 50% and the court's lawyer's impairment rate is 50%.
  • the server extracts the corresponding target tag from the referee document or judgment paragraph by means of keyword matching. In one embodiment, the server extracts a target tag that matches the intellectual property tag from the judgment document. When no target tag is extracted, the server further extracts a target tag that matches the ruling tag.
  • multiple calculation methods are provided for calculating the plaintiff's lawyer's winning rate and the court's lawyer's impairment rate, and according to the corresponding calculation method of the extracted target tag, the plaintiff's lawyer's winning rate and the court's lawyer's impairment rate are correspondingly calculated, which improves Calculation efficiency, thus improving the analysis efficiency of lawyers' professional ability.
  • the preset label set includes an intellectual property label
  • the method for processing the referee document further includes: when the target label and the intellectual property label match, the judgment paragraph and knowledge Match the preset intellectual property expression corresponding to the property right label; when the judgment paragraph matches the preset intellectual property expression successfully, determine the plaintiff's lawyer's success rate and the court's lawyer's impairment according to the preset determination method corresponding to the preset intellectual property expression. Rate; when the decision paragraph fails to match the preset intellectual property expression, step S210 is performed.
  • An intellectual property label is a label used to characterize a corresponding case as an intellectual property case.
  • Intellectual property labels include, but are not limited to, disputes over ownership of trademark rights and disputes over trademark infringement.
  • the preset intellectual property expression is a preset target expression for extracting a corresponding intellectual property decision result from a judgment paragraph by a regular matching method. Preset intellectual property expressions such as "rejection of litigation request for. *?", “Unsupported litigation request”, “rejection of other litigation requests for. *?", And "stop”.
  • the server correspondingly determines a preset intellectual property expression corresponding to the successfully matched intellectual property tag, and compares the determined preset intellectual property expression with Decision paragraphs are matched.
  • the server correspondingly determines the plaintiff's lawyer's success rate and the court's lawyer's impairment rate according to the preset determination method corresponding to the preset intellectual property expression.
  • the server calculates the plaintiff's lawyer's success rate and the judgment amount according to the extracted claim amount and the corresponding claim amount, and the judgment amount and the corresponding judgment amount. Defendant's lawyer impairment rate.
  • each intellectual property tag corresponds to at least one preset intellectual property expression.
  • the server matches the multiple preset intellectual property expressions with the judgment paragraph respectively, and according to the successfully matched preset intellectual property expressions
  • the preset determination method corresponding to this formula corresponds to the determination of the plaintiff's lawyer's success rate and the court's lawyer's impairment rate.
  • the target label is a trademark ownership dispute
  • the preset intellectual property expressions corresponding to the trademark ownership dispute include "rejection of litigation request *.”, "Unsupported litigation request”, and “rejection. *?” Other lawsuits.
  • the server determines that the plaintiff's lawyer's success rate is 0 and the court's attrition rate is 100% .
  • the preset intellectual property expression that successfully matches the judgment paragraph is "rejection of other lawsuits of. *?"
  • the plaintiff's lawyer's success rate and the court's lawyer's impairment rate are both determined to be 50%.
  • the server determines that the plaintiff's lawyer's success rate is 100% and the court's lawyer's impairment rate is 0.
  • the calculation method of the plaintiff's lawyer's winning rate and the court's lawyer's impairment rate is determined according to the corresponding intellectual property label corresponding to the judgment paragraph, which improves the calculation efficiency and thus improves the professional capacity of lawyers. Analytical efficiency.
  • the training steps of the paragraph extraction model include: obtaining multiple target referee documents; marking the target appeal paragraph and target judgment paragraph in each target referee document separately; obtaining the corresponding claims of the target claim paragraph Paragraph extraction problem and decision paragraph extraction problem corresponding to target judgment paragraph; target judgment documents, appeal paragraph extraction problem and judgment paragraph extraction problem are taken as input features, and corresponding target appeal paragraph and target judgment paragraph are taken as expected output features , Training the long-term memory neural network to obtain a trained paragraph extraction model.
  • the claim paragraph extraction question is used to extract the questioning data of the claim paragraph from the target judgment document, for example, "Which paragraph does the plaintiff's claim correspond to?”
  • the judgment paragraph extraction question is used to extract the question data of the judgment paragraph from the target judgment document, such as "Which paragraph does the court decision result correspond to?”
  • Long-term memory neural networks can be specifically End-To-End Memory Networks, which can be single-layer memory networks or multilayer memory networks.
  • the server obtains multiple target judgment documents, and marks the target application paragraph and target judgment paragraph in each target judgment document separately.
  • the server respectively obtains the question of extracting the request paragraph corresponding to the target appeal paragraph and the question of extracting the decision paragraph corresponding to the target judgment paragraph.
  • the server takes the target referee document and the corresponding application paragraph extraction question as input features, and uses the corresponding target application paragraph as the desired output feature to obtain the first training sample set.
  • the server takes the target judgment document and the corresponding decision paragraph extraction problem as input features, and uses the corresponding target decision paragraph as the desired output feature to obtain a second training sample set.
  • the server alternately trains the long-term memory neural network to obtain a trained paragraph extraction model.
  • Alternate training refers to using the first training sample in the first training sample set and the second training sample in the second training sample set to alternately train the long-term memory neural network.
  • the server performs training according to the first training sample set to obtain a paragraph extraction model for extracting an appeal paragraph from a referee document. Similarly, the server performs training according to the second training sample set to obtain a paragraph extraction model for extracting a judgment paragraph from a referee document.
  • the input features of the first training sample set and the second training sample set are both target referee documents.
  • the server uses the cross-entropy loss function as an objective function to train the long-term memory neural network during model training.
  • the model training is performed based on the training sample set, and the paragraph extraction model obtained through the training is used to extract the appeal paragraph and the judgment paragraph from the referee documents respectively, which improves the extraction efficiency.
  • the training steps of the entity recognition model include: obtaining multiple target referee documents; extracting target appeal paragraphs and target decision paragraphs from each target referee document separately through the trained paragraph extraction model; The target amount items in the target claim paragraph and the target decision paragraph; the target claim paragraph and the target decision paragraph are used as input features, and the corresponding target amount items are used as the desired output features.
  • the initialized entity recognition model is trained to obtain the trained Entity recognition model.
  • the server obtains a plurality of target judgment documents, and extracts corresponding target appeal paragraphs and target judgment paragraphs respectively through a pre-trained paragraph extraction model.
  • the server respectively marks the corresponding target amount, that is, the target claim amount in the target claim paragraph and the target judgment amount in the target judgment paragraph.
  • the server takes the target claim paragraph as an input feature and the corresponding target claim amount item as the desired output feature to obtain a corresponding claim training sample set.
  • the server uses the target decision paragraph as the input feature and the corresponding target decision amount item as the desired output feature to obtain the corresponding decision training sample set.
  • the server alternately trains the initialized entity recognition model according to the request training sample set and the decision training sample set to obtain a trained entity recognition model.
  • the server is trained according to the training set of claims to obtain an entity recognition model for extracting a claim amount item from the claims paragraph. Similarly, the server performs training according to the decision training sample set to obtain an entity recognition model for extracting a decision amount item from the decision paragraph.
  • the entity recognition model may specifically be a bilstm-crf model.
  • training is performed based on the training sample set, and the entity recognition model obtained through training is used to extract a claim amount item from a claim paragraph and a decision amount item from a decision paragraph, thereby improving extraction efficiency.
  • Characters can be Chinese characters, numbers, letters and symbols.
  • a method for processing a referee document is provided.
  • the method specifically includes the following steps:
  • S310 Extract a target label from a referee document based on a preset label expression.
  • the plaintiff's lawyer's success rate and the court's lawyer's impairment rate are calculated according to the amount of claim and the corresponding amount of claim, and the amount of judgment and the corresponding amount of judgment.
  • multiple calculation methods are provided to calculate the plaintiff's lawyer's winning rate and the court's lawyer's impairment rate, so as to facilitate the selection of a suitable calculation method according to the obtained judgment documents, improve the calculation efficiency, and improve the professional capacity of the lawyer. Analytical efficiency.
  • steps in the flowcharts of FIG. 2 and FIG. 3 are sequentially displayed according to the directions of the arrows, these steps are not necessarily performed sequentially in the order indicated by the arrows. Unless explicitly stated in this document, the execution of these steps is not strictly limited, and these steps can be performed in other orders. Moreover, at least a part of the steps in FIG. 2 and FIG. 3 may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily performed at the same time, but may be performed at different times. These sub-steps or The execution order of the phases is not necessarily performed sequentially, but may be performed in turn or alternately with other steps or at least a part of the sub-steps or phases of other steps.
  • a referee word processing device 400 which includes: an acquisition module 401, a paragraph extraction model 402, an amount extraction module 403, an amount extraction module 404, and a calculation module 405. among them:
  • the obtaining module 401 is configured to obtain a referee document.
  • the paragraph extraction model 402 is used to extract the appeal paragraph and the judgment paragraph from the referee document through the trained paragraph extraction model.
  • Amount item extraction module 403 is configured to extract a claim amount item from a claim paragraph through a trained entity recognition model, and extract a judgement amount item from a judgment paragraph.
  • Amount value extraction module 404 for extracting a claim amount item and a judgment amount item, based on a preset amount item expression, extracting a claim amount value corresponding to the claim amount item from the claim paragraph, and from the judgment The judgment amount value corresponding to the judgment amount item is extracted in the paragraph.
  • the calculation module 405 is configured to calculate the plaintiff's lawyer's success rate and the Yankee's lawyer's impairment rate based on the amount of the claim and the corresponding amount of the claim, and the amount of the judgment and the corresponding amount of the judgment.
  • the referee document processing device 400 further includes: an allocation data extraction module 406;
  • the paragraph extraction model 402 is also used to extract the case acceptance fee paragraph from the judgment paragraph based on the preset acceptance fee expression when the claim amount and judgment amount terms are not extracted; the allocation data extraction module 406 is used to Set the processing fee sharing expression to extract the processing fee sharing data from the case processing fee paragraph.
  • the calculation module 405 is also used to calculate the plaintiff's lawyer's winning rate and the court's lawyer's impairment rate based on the processing fee sharing data.
  • the shared data extraction module 406 is further configured to sequentially extract the preset keywords in the case acceptance fee paragraphs in accordance with the semantic order; determine the corresponding acceptance of the case acceptance fee paragraphs according to the preset keywords and according to the preset classification conditions.
  • Type of fee allocation According to the preset expression of the fee allocation corresponding to the type of handling fee allocation, the handling fee allocation data is extracted from the case handling fee paragraph.
  • the referee document processing device 400 further includes: a label extraction module 407; a label extraction module 407 for extracting a target label from the referee document based on a preset label expression; when the target label and the preset label set If they do not match, the calculation module 405 is caused to perform the steps of calculating the plaintiff's lawyer's winning rate and the court's lawyer's impairment rate based on the amount of the claim and the corresponding amount of the claim, and the amount of the judgment and the corresponding amount of the judgment.
  • the preset label set includes an intellectual property label; the label extraction module 407 is further configured to: when the target label matches the intellectual property label, match the judgment paragraph with the preset intellectual property corresponding to the intellectual property label Match the expression; when the judgment paragraph matches the preset intellectual property expression successfully, the calculation module 405 is further configured to determine the plaintiff's lawyer's success rate and the court's lawyer's impairment rate according to the preset determination method corresponding to the preset intellectual property expression; When the judgment paragraph fails to match the preset intellectual property expression, the calculation module 405 executes the calculation of the plaintiff's lawyer's success rate and the victim according to the amount of the claim and the corresponding amount of the claim, and the value of the judgment and the corresponding amount of the judgment Steps for attorney impairment rate.
  • the referee word processing device 400 further includes: a model training module 408;
  • a model training module 408 is used to obtain multiple target referee documents; label the target appeal paragraph and target judgment paragraph in each target referee document separately; obtain the target claim paragraph extraction question corresponding to the target claim paragraph and the target decision paragraph Corresponding decision paragraph extraction problem; the target judgement document, appeal paragraph extraction problem and judgment paragraph extraction problem are used as input features, and the corresponding target appeal paragraph and target judgment paragraph are used as expected output features to train the long-term memory neural network Obtain a trained paragraph extraction model.
  • the model training module 408 is further configured to obtain a plurality of target referee documents; the target claim paragraph and the target decision paragraph are separately extracted from each target referee document through the trained paragraph extraction model; The target amount items in the target claim paragraph and the target decision paragraph; the target claim paragraph and the target decision paragraph are used as input features, and the corresponding target amount items are used as the desired output features.
  • the initialized entity recognition model is trained to obtain the trained Entity recognition model.
  • each module in the referee document processing device can be implemented in whole or in part by software, hardware, and a combination thereof.
  • the above-mentioned modules may be embedded in the hardware form or independent of the processor in the computer device, or may be stored in the memory of the computer device in the form of software, so that the processor calls and performs the operations corresponding to the above modules.
  • a computer device is provided.
  • the computer device may be a server, and its internal structure diagram may be as shown in FIG. 6.
  • the computer device includes a processor, a memory, a network interface, and a database connected through a system bus.
  • the processor of the computer device is used to provide computing and control capabilities.
  • the memory of the computer device includes a non-volatile storage medium and an internal memory.
  • the non-volatile storage medium stores an operating system, computer-readable instructions, and a database.
  • the internal memory provides an environment for operating the operating system and computer-readable instructions in a non-volatile storage medium.
  • the database of the computer equipment is used to store referee documents and preset amount expressions.
  • the network interface of the computer device is used to communicate with an external terminal through a network connection.
  • the computer-readable instructions are executed by a processor to implement a referee document processing method.
  • FIG. 6 is only a block diagram of a part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer equipment to which the scheme of the present application is applied.
  • the specific computer equipment may be Include more or fewer parts than shown in the figure, or combine certain parts, or have a different arrangement of parts.
  • a computer device includes a memory and one or more processors.
  • Computer-readable instructions are stored in the memory, and when the computer-readable instructions are executed by one or more processors, the one or more processors implement any one of the present application.
  • One or more non-volatile computer-readable storage media storing computer-readable instructions, and when the computer-readable instructions are executed by one or more processors, the one or more processors implement one of the embodiments of the present application Provide steps for referee document processing methods.
  • Non-volatile memory may 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.
  • RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous chain Synchlink DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).
  • SRAM static RAM
  • DRAM dynamic RAM
  • SDRAM synchronous DRAM
  • DDRSDRAM dual data rate SDRAM
  • ESDRAM enhanced SDRAM
  • SLDRAM synchronous chain Synchlink DRAM
  • Rambus direct RAM
  • DRAM direct memory bus dynamic RAM
  • RDRAM memory bus dynamic RAM

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • General Physics & Mathematics (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Business, Economics & Management (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Tourism & Hospitality (AREA)
  • Evolutionary Biology (AREA)
  • Computational Linguistics (AREA)
  • Audiology, Speech & Language Pathology (AREA)
  • Computing Systems (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Strategic Management (AREA)
  • Primary Health Care (AREA)
  • Molecular Biology (AREA)
  • General Business, Economics & Management (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Marketing (AREA)
  • Human Resources & Organizations (AREA)
  • Economics (AREA)
  • Technology Law (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)
  • Financial Or Insurance-Related Operations Such As Payment And Settlement (AREA)

Abstract

一种裁判文书处理方法，包括：获取裁判文书；通过已训练的段落提取模型从裁判文书中提取诉请段落和判决段落；通过已训练的实体识别模型从所述诉请段落中提取诉请金额项，以及从所述判决段落中提取判决金额项；当提取到所述诉请金额项和所述判决金额项时，基于预设金额项表达式，从所述诉请段落中提取与所述诉请金额项对应的诉请金额值，以及从所述判决段落中提取与所述判决金额项对应的判决金额值；根据所述诉请金额项和相应的所述诉请金额值，以及所述判决金额项和相应的所述判决金额值，分别计算原告律师胜诉率和被告律师减损率。

Description

裁判文书处理方法、装置、计算机设备和存储介质
本申请要求于2018年09月10日提交中国专利局，申请号为2018110519287，申请名称为“裁判文书处理方法、装置、计算机设备和存储介质”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本申请涉及一种裁判文书处理方法、装置、计算机设备和存储介质。
背景技术
裁判文书是指裁判结果的法律文书，是记载人民法院审理诉讼活动过程和结果的凭证。裁判文书中包括原被告、律师、庭审法院、原告诉讼请求、法院判决结果和案件受理费等。通过分析裁判文书可以了解案件的相关情况，比如基于裁判文书可以了解案件涉及的原被告、律师和庭审法院等信息。同时，裁判文书中的原告诉请和判决内容，以及审查过程能够反映律师的专业能力等。由于裁判文书有固定格式，可以通过指定规则自动提取原被告、律师和庭审法院等信息。然而，由于案件类型不同、庭审法院不同和书记员不同等造成裁判文书的书写风格差异较大，没有固定规律。
目前，通常是通过人工定位并分析裁判文书中的诉请、判决内容和审查过程等，根据分析结果确定律师的专业能力。然而，发明人意识到，该种律师专业能力的分析方式需要耗费大量的人力物力，且分析过程会耗费大量的时间，存在裁判文书的处理效率低的问题，从而导致律师专业能力的分析效率低。
发明内容
根据本申请公开的各种实施例，提供一种裁判文书处理方法、装置、计算机设备和存储介质。
一种裁判文书处理方法包括：
获取裁判文书；
通过已训练的段落提取模型从所述裁判文书中提取诉请段落和判决段落；
通过已训练的实体识别模型从所述诉请段落中提取诉请金额项，以及从所述判决段落中提取判决金额项；
当提取到所述诉请金额项和所述判决金额项时，基于预设金额项表达式，从所述诉请段落中提取与所述诉请金额项对应的诉请金额值，以及从所述判决段落中提取与所述判决金额项对应的判决金额值；及
根据所述诉请金额项和相应的所述诉请金额值，以及所述判决金额项和相应的所述判决金额值，分别计算原告律师胜诉率和被告律师减损率。
一种裁判文书处理装置包括：
获取模块，用于获取裁判文书；
段落提取模型，用于通过已训练的段落提取模型从所述裁判文书中提取诉请段落和判决段落；
金额项提取模块，用于通过已训练的实体识别模型从所述诉请段落中提取诉请金额项，以及从所述判决段落中提取判决金额项；
金额值提取模块，用于当提取到所述诉请金额项和所述判决金额项时，基于预设金额项表达式，从所述诉请段落中提取与所述诉请金额项对应的诉请金额值，以及从所述判决段落中提取与所述判决金额项对应的判决金额值；及
计算模块，用于根据所述诉请金额项和相应的所述诉请金额值，以及所述判决金额项和相应的所述判决金额值，分别计算原告律师胜诉率和被告律师减损率。
一种计算机设备，包括存储器和一个或多个处理器，所述存储器中储存有计算机可读指令，所述计算机可读指令被所述一个或多个处理器执行时，使得所述一个或多个处理器实现本申请任意一个实施例中提供的裁判文书处理方法的步骤。
一个或多个存储有计算机可读指令的非易失性计算机可读存储介质，所述计算机可读指令被一个或多个处理器执行时，使得所述一个或多个处理器实现本申请任意一个实施例中提供的裁判文书处理方法的步骤。
本申请的一个或多个实施例的细节在下面的附图和描述中提出。本申请的其它特征和优点将从说明书、附图以及权利要求书变得明显。计算机可读指令计算机可读指令计算机可读指令计算机可读指令
附图说明
为了更清楚地说明本申请实施例中的技术方案，下面将对实施例中所需要使用的附图作简单地介绍，显而易见地，下面描述中的附图仅仅是本申请的一些实施例，对于本领域普通技术人员来讲，在不付出创造性劳动的前提下，还可以根据这些附图获得其它的附图。
图1为根据一个或多个实施例中裁判文书处理方法的应用场景图。
图2为根据一个或多个实施例中裁判文书处理方法的流程示意图。
图3为另一个实施例中裁判文书处理方法的流程示意图。
图4为根据一个或多个实施例中裁判文书处理装置的框图。
图5为另一个实施例中裁判文书处理装置的框图。
图6为根据一个或多个实施例中计算机设备的框图。
具体实施方式
为了使本申请的技术方案及优点更加清楚明白，以下结合附图及实施例，对本申请进 行进一步详细说明。应当理解，此处描述的具体实施例仅仅用以解释本申请，并不用于限定本申请。
本申请提供的裁判文书处理方法，可以应用于如图1所示的应用环境中。终端102通过网络与服务器104通过网络进行通信。服务器104通过已训练的段落提取模型从所获取到的裁判文书中提取诉请段落和判决段落，并通过已训练的实体识别模型从诉请段落中提取诉请金额项，以及从判决段落中提取判决金额项，进而基于预设金额项表达式分别提取诉请金额项对应的诉请金额值和判决金额项对应的判决金额值，以根据所提取的诉请金额值和判决金额值对应计算原告律师胜诉率和被告律师减损率，并将所计算的原告律师胜诉率和被告律师减损率发送至终端102。终端102可以但不限于是各种个人计算机、笔记本电脑、智能手机、平板电脑和便携式可穿戴设备，服务器104可以用独立的服务器或者是多个服务器组成的服务器集群来实现。
在其中一个实施例中，如图2所示，提供了一种裁判文书处理方法，以该方法应用于图1中的服务器为例进行说明，包括以下步骤：
S202，获取裁判文书。
裁判文书是记录裁判过程和裁判结果的法律文书，是记载人民法院审理诉讼活动过程和结果的凭证。裁判文书包括原告、被告、原告律师、被告律师、庭审法院、原告诉讼请求、法院判决结果和案件受理费等。
具体地，服务器接收到裁判文书处理指令时，根据所接收到的裁判文书处理指令获取相应的裁判文书。服务器根据所接收到的裁判文书处理指令在本地对应查询预存储的裁判文书。服务器具体可以接收终端发送的裁判文书处理指令，根据所接收到的裁判文书指令从终端获取相应的裁判文书。
在其中一个实施例中，裁判文书处理指令中携带有待处理的裁判文书。服务器解析所接收到的裁判文书处理指令，获得相应的裁判文书。
S204，通过已训练的段落提取模型从裁判文书中提取诉请段落和判决段落。
段落提取模型是根据预先获取的训练样本集进行模型训练获得的、用于从裁判文书中对应提取诉请段落和判决段落的模型。诉请段落是指用于描述原告诉讼请求的段落。判决段落是指用于描述法院判决结果的段落。
具体地，服务器将所获取到的裁判文书输入预先训练好的段落提取模型中，通过该段落提取模型进行预测，分别获得相应的诉请段落和判决段落，以从该裁判文书中分别提取相应的诉请段落和判决段落。
在其中一个实施例中，服务器通过预先训练好的第一段落提取模型和第二段落提取模型，分别从所获取到的裁判文书中提取诉请段落和判决段落。具体地，服务器将所获取到的裁判文书输入预先训练好的第一段落提取模型进行预测，获得该裁判文书中的诉请段落。类似地，服务器将所获取到的该裁判文书输入到预先训练好的第二段落提取模型进行预测，获得该裁判文书中的判决段落。第一段落提取模型是根据目标裁判文书和相应的目 标诉请段落进行模型训练获得的段落提取模型，第二段落提取模型是根据目标裁判文书和相应的目标判决段落进行模型训练获得的段落提取模型。
在其中一个实施例中，预先训练好的段落提取模型为长久记忆神经网络模型。服务器获取到裁判文书时，分别获取诉请段落提取问题和判决段落提取问题。服务器将所获取到的裁判文书和诉请段落提取问题输入到预先训练好的段落提取模型进行预测，获得该裁判文书中的诉请段落。进一步地，服务器将该裁判文书和所获取到的判决段落提取问题输入到该预先训练好的段落提取模型进行预测，获得该裁判文书中的判决段落。
在其中一个实施例中，服务器通过预先训练好的段落提取模型从裁判文书中提取出诉请段落和判决段落后，基于预设诉请表达式对所提取出的诉请段落进行筛选，对筛选出的诉请段落执行下述提取诉请金额项和诉请金额值的相关步骤。类似地，服务器基于预设判决表达式对所提取出的判决段落进行筛选，对筛选出的判决段落执行下述提取判决金额项和判决金额值的相关步骤。预设诉请表达式是指预先设定的用于通过正则匹配方式筛选诉请段落的正则表达式。预设判决表达式是指预先设定的用于通过正则匹配方式筛选判决段落的正则表达式。预设判决表达式比如“(？：准许(？：原告|上诉人)？.*撤回.*(？：起诉|上诉)|按撤诉处理|准予撤诉)”，或者(？：发回.*重审|移送.*审理|指令.*？再审)等。值得说明的是，预设判决表达式以及相应的表达式形式不仅仅局限于示例，预设判决表达的形式可以比示例更简单或更复杂。通过预设诉请表达式对所提取出的诉请段落和判决段落进行筛选，提高了诉请段落和判决段落提取的准确性。
S206，通过已训练的实体识别模型从诉请段落中提取诉请金额项，以及从判决段落中提取判决金额项。
实体识别模型是根据预先获取的训练样本集进行模型训练获得的、用于从所获取到的段落中提取金额项的模型。诉请金额项是指原告诉讼请求中指定的损害赔偿项目。诉请金额项具体可以是指原告在原告诉讼请求中所提出的要求被告赔偿的损害赔偿项目，比如精神损失费、医疗费、营养费和子女抚养费等。判决金额项是指法院判决结果中指定的由被告向原告赔偿的损害赔偿项目，比如精神损失费、医疗费和子女抚养费等。诉请金额项和判决金额项可以相同，也可以存在差异。
具体地，服务器将从裁判文书中提取的诉请段落输入预先训练好的实体识别模型中，通过该实体识别模型进行预测，获得该诉请段落中的诉请金额项。类似地，服务器将从裁判文书中提取的判决段落输入预先训练好的实体识别模型进行预测，获得该判决段落中的判决金额项。
在其中一个实施例中，从诉请段落中提取诉请金额项的实体识别模型，与从判决段落中提取判决段落的实体识别模型可以同一个实体识别模型，也可以是基于不同的训练样本集分别训练获得的不同的实体识别模型。
S208，当提取到诉请金额项和判决金额项时，基于预设金额项表达式，从诉请段落中提取与诉请金额项对应的诉请金额值，以及从判决段落中提取与判决金额项对应的判决金 额值。
预设金额项表达式是指预先设定的用于通过正则匹配方式从指定段落中提取的金额值的正则表达式。预设金额项表达式具体可以是从指定段落中提取与金额项对应的金额值的正则表达式，比如“精神损失费.*？元”。诉请金额值是指原告诉讼请求中指定的赔偿金额数值。诉请金额值具体可以是指原告在原告诉讼请求中提出的由被告赔偿的赔偿金额数值，比如2万元。诉请金额项与诉请金额值相对应。判决金额值是指法院判决结果中指定的由被告向原告赔偿的赔偿金额数值，比如1万元。
具体地，当从诉请段落中提取到诉请金额项、且从判决段落中提取到判决金额项时，服务器将诉请段落对应的预设金额项表达式与所提取到的诉请段落进行匹配，以从诉请段落中提取出与诉请金额项对应的诉请金额值。类似地，服务器将判决段落对应的预设金额项表达式与所提取的判决段落进行匹配，以从判决段落中提取出与判决金额项对应的判决金额值。诉请段落对应的预设金额项表达式，与判决段落对应的预设金额项表达式可以相同也可以不同。
S210，根据诉请金额项和相应的诉请金额值，以及判决金额项和相应的判决金额值，分别计算原告律师胜诉率和被告律师减损率。
原告律师胜诉率是指原告律师为原告争取利益/赔偿的量化值。被告律师减损率是指指被告律师为被告减小损失/赔偿的量化值。在本实施例中，原告律师胜诉率具体可以是法院判决结果中的判决总金额，占原告诉讼请求中的诉请总金额的比率。被告律师减损率具体可以是原告诉讼请求中的诉请总金额与法院判决结果中的判决总金额之间的差值，占原告诉讼请求中的诉请总金额的比率。可以理解的是，原告律师胜诉率越大表明原告律师的专业能力越强，同样的，被告律师减损率越大表明被告律师的专业能力越强。原告律师胜诉率与被告律师减损率的总和为1。
具体地，服务器根据从诉请段落中提取到的诉请金额项和相应的诉请金额值确定相应的诉请总金额，并根据从判决段落中提取到的判决金额项和相应的判决金额值确定相应的判决总金额。进一步地，服务器根据所确定的诉请总金额和判决总金额，按照第一预设计算方式分别计算原告律师胜诉率和被告律师减损率。第一预设计算方式是预先设定的计算方式，用于指示服务器如何根据所确定的诉请总金额和判决总金额，对应计算原告律师胜诉率和被告律师减损率。第一预设计算方式具体可以是根据诉请总金额和判决总金额计算原告律师胜诉率和被告律师减损率的计算表达式，比如原告律师胜诉率＝判决总金额/诉请总金额，被告律师减损率＝1-判决总金额/诉请总金额。
上述裁判文书处理方法，通过段落提取模型从所获取到的裁判文书中自动提取相应的诉请段落和判决段落，进而通过实体识别模型分别从所提取到的诉请段落中自动提取诉请金额项，并从判决段落中自动提取判决金额项，提高了裁判文书中金额项的提取效率，从而提高了裁判文书的处理效率。当提取到诉请金额项和判决金额项时，基于预设金额项表达式，以及所提取到的诉请金额项和判决金额项自动提取相应的诉请金额值和判决金额 值，提高了金额值的提取效率，进一步提高了判断文书的处理效率，从而提高了律师专业能力的分析效率。
在其中一个实施例中，上述裁判文书处理方法还包括：当没有提取到诉请金额项和判决金额项时，基于预设受理费表达式，从判决段落中提取案件受理费段落；基于预设受理费分摊表达式，从案件受理费段落提取受理费分摊数据；根据受理费分摊数据分别计算原告律师胜诉率和被告律师减损率。
预设受理费表达式是预先设定的用于通过正则匹配方式从判决段落中提取案件受理费段落的正则表达式。由于裁判文书中的案件受理费段落通常为判决段落中的独立段落，且通常具有指定关键词，比如案件受理费、本案受理费、诉讼费和案件上诉费等，预设受理费表达式比如可以是“.*？受理费”。案件受理费段落是指用于描述法院针对受理案件指定受理费的段落。案件受理费段落具体用于描述指定的受理费和受理费分摊情况。
预设受理费分摊表达式是预先设定的用于通过正则匹配方式从案件受理费段落中提取受理费分摊数据的正则表达式。预设受理费分摊表达式比如“.*？费.*？元.*？原告.*？负担”，或者“.*？被告.*？负担.*？元”等。受理费分摊数据是指案件受理费分摊情况的量化数据。受理费分摊数据可用于表征原告和被告各自承担的受理费。受理费分摊数据具体可以包括原告承担的受理费和被告承担的受理费。受理费分摊数据还可以包括法院判决结果中指定的案件受理费总额，和/或案件审理过程中涉及的其他费用，如保全费和公告费等。受理费分摊数据还可以包括受理费分摊对象的法律地位、名称和负担金额。可以理解的是受理费分摊对象可以是个人，也可以是公司。
具体地，当没有从诉请段落中提取到诉请金额项、且没有从判决段落中提取到判决金额项时，服务器将预设受理费表达式与所提取到的判决段落进行匹配，以从判决段落中提取出相应的案件受理费段落。进一步地，服务器将预设受理费分摊表达式与所提取到的案件受理费段落进行匹配，以从案件受理费段落中提取相应的受理费分摊数据。服务器根据所提取到的受理费分摊数据对应确定案件受理费总额、原告承担案件受理费和被告承担案件受理费。服务器根据所确定的案件受理费总额、原告承担案件受理费和被告承担案件受理费，按照第二预设计算方式分别计算相应的原告律师胜诉率和被告律师减损率。
第二预设计算方式是预先设定的计算方式，用于指示服务器如何根据所提取的受理费分摊数据，对应计算原告律师胜诉率和被告律师减损率。第二预设计算方式具体可以是根据受理费分摊数据对应计算原告律师胜诉率和被告律师减损率的计算表达式，比如原告律师胜诉率＝被告承担案件受理费/案件受理费总额，被告律师减损率＝原告承担案件受理费/案件受理费总额。
在其中一个实施例中，当没有从诉请段落中提取到诉请金额，或者没有从判决段落中提取到判决段落时，服务器基于预设受理费表达式，从判决段落中提取相应的案件受理费段落。
在其中一个实施例中，服务器从案件受理费段落中提取到受理费分摊数据时，对所提 取到的受理费分摊数据进行预处理，再根据预处理后的受理费分摊数据对应计算案件受理费总额、原告承担案件受理费和被告承担案件受理费。预处理包括但不限于对受理费分摊数据中的金额值进行去重处理，或者将受理费分摊数据中不符合预设格式的金额值转换为预设标准形式的金额值。具体地，服务器在案件受理费段落中提取指定关键词，当提取到指定关键词时，服务器按照所提取到的指定关键词所对应的预设预处理方式对受理费分摊数据进行预处理。指定关键词包括但不限于变更、增加、减半收取、总计、共计、合计、万元和亿元等。指定关键词也可以是“百分”或“百分之”，或者百分符号“％”。比如当提取到变更、增加、减半收取和/或总计等关键词时，对受理费分摊数据中的金额值进行去重处理；当提取到百分、万元和/或亿元等关键词时，表明该关键词对应的金额值为不符合预设格式的金额值，服务器将该金额值转换为预设标准形式的金额值。
举例说明，假设提取到的案件受理费段落为“案件受理费2万元，减半收取1万元，保全费5000元，共计15000元，由原告A负担10％，由被告B负担90％”。服务器从该案件受理费段落中提取到的指定关键词为“万元”、“减半收取”、“共计”和“％”。服务器根据指定关键词“减半收取”去除该关键词之前的金额值“2万元”，根据指定关键词“共计”去除该关键词之前的重复金额值“1万元”和“5000元”。同时，服务器根据关键词“万元”和“％”分别将各自对应的金额值转换为预设标准形式，即将1万元转换为10000元，将10％转换为1500元，将90％转换为13500元。值得说明的是的，受理费分摊数据的预处理并不仅仅局限于上述举例说明。
上述实施例中，在没有提取到诉请金额项和判决金额项时，提供了另一种计算原告律师胜诉率和被告律师减损率的计算方式，以便于从不同维度分析律师专业能力，提高了律师专业能力的分析效率。
在其中一个实施例中，基于预设受理费分摊表达式，从案件受理费段落提取受理费分摊数据，包括：按照语义顺序依次提取案件受理费段落中的预设关键词；根据预设关键词按照预设分类条件确定案件受理费段落对应的受理费分摊类型；根据受理费分摊类型对应的预设受理费分摊表达式，从案件受理费段落中提取受理费分摊数据。
语义顺序是指语言逻辑顺序。语义顺序具体可以是指构成案件受理费段落的各个词出现的先后顺序，即词序。预设关键词是预先设定的关键词，比如“费”、“共计”和“负担”等。受理费分摊类型是指案件受理费段落中的受理费分摊数据所对应的分摊类型。案件受理费段落中的费用和分摊情况对应有至少一种的组合形式，费用和分摊情况的组成形式与受理费分摊类型相对应。费用和分摊情况的组合形式包括但不限于单一费用的分摊、多种费用的合计分摊和多种费用的分别分摊等。
具体地，服务器按照案件受理费段落的语义顺序，从前往后依次从该案件受理费段落中提取相应的预设关键词。服务器根据所提取到的预设关键词以及预设关键词的提取顺序，按照预设分类条件确定案件受理费段落对应的受理费分摊类型。服务器根据案件受理费段落所对应的受理费分摊类型，对应确定预设受理费分摊表达式，将所确定的预设受理 费分摊表达式与案件受理费段落进行匹配，以从案件受理费段落中提取相应的受理费分摊数据。
在其中一个实施例中，服务器提取到预设关键词后，对应确定所提取到的预设关键词中所包含的第一预设关键词的数量。当所提取到的预设关键词中包含一个第一预设关键词时，服务器将相应案件受理费段落对应的受理费分摊类型确定为第一分摊类型。当所提取到的预设关键词中包含多于一个的第一预设关键词时，服务器按照预设关键词的提取顺序，依次判断所提取到的预设关键词中，是否存在第二预设关键词的下一个预设关键词为第一预设关键词。若不存在，服务器将相应的受理费分摊类型确定为第二分摊类型；若存在，服务器将相应的受理费分摊类型确定为第三分摊类型。预设关键词包括第一预设关键词和第二预设关键词，第一预设关键词比如“费”、“合计”和“共计”等，第二预设关键词比如“负担”和“承担”等。
举例说明，首先，假设案件受理费段落为“案件受理费1000元，由原告A负担”，从该案件受理费段落中提取到的预设关键词为“费”和“负担”，该预设关键词中包含一个第一预设关键词，故将相应的受理费分摊类型确定为第一分摊类型。其次，假设案件受理费段落为“案件受理费1000元，保全费500元，由原告A负担”，由于该案件受理费段落中包含2个第一预设关键词、且不存在第二预设关键词的下一个预设关键词为第一预设关键词，故将受理费分摊类型确定为第二分摊类型。再次，假设案件受理费段落为“案件受理费1000元，由原告A负担，保全费500元，由被告B负担”，由于该案件受理费段落中包含2个第一预设关键词、且存在第二预设关键词的下一个关键词为第一预设关键词，故将受理费分摊类型确定为第三分摊类型。
在其中一个实施例中，对于每类受理费分摊类型，服务器预存储有相应的预设正则表达式。服务器将每类受理费分摊类型所对应的预设正则表达式，分别与案件受理费段落进行匹配，当匹配成功时，将匹配成功的预设正则表达式所对应的受理费分摊类型，确定为该案件受理费段落所对应的受理费分摊类型。对于每类受理费分摊类型，相应的预设受理费分摊表达式和预设正则表达式可以相同也可以不同。
在其中一个实施例中，当案件受理费段落对应的受理费分摊类型为第一分摊类型时，服务器基于该第一分摊类型所对应的预设受理费分摊表达式，从案件受理费段落中提取相应的受理费分摊数据。类似地，当案件受理费段落对应的受理费分摊类型为第二分摊类型时，服务器基于该第二分摊类型所对应的预设受理费分摊表达式，从案件受理费段落中提取相应的受理费分摊数据。受理费分摊数据具体可以包括但不限于是案件受理费总额、受理费分摊对象和每个受理费分摊对象所分摊的金额。
在其中一个实施例中，当案件受理费段落对应的受理费分摊类型为第三分摊类型时，服务器按照预设分割条件将该案件受理费段落划分为多于一个的单元句。每个单元句所对应的受理费分摊类型为第一分摊类型或第二分摊类型。对于每个单元句，服务器基于单元句对应的受理费分摊模型所对应的预设受理费分摊表达式，从该单元句中提取该单元句所 对应的受理费分摊数据。服务器根据从每个单元句中分别提取到的受理费分摊数据，对应确定相应案件受理费段落所对应的受理费分摊数据。
举例说明，假设从案件受理费段落为“案件受理费1000元、保全费500元，共计1500元，由原告A负担，公告费600元，由被告B负担”，从该案件受理费段落中依次提取到的预设关键词分别为：费、费、共计、负担、费和负担时，服务器判定该案件受理费段落对应的受理费分摊类型为第三分摊类型。由于依次提取到的预设关键词中，第二预设关键词“负担”后存在第一预设关键词“费”，即依次提取到的预设关键词存在“负担”到“费”的变化，服务器对案件受理费段落在预设关键词“负担”和“费”各自所在的句子之间进行分割，对应获得的两个单元句分别为“案件受理费1000元、保全费500元，共计1500元，由原告A负担”和“公告费600元，由被告B负担”。
在其中一个实施例中，对于第一分摊类型和第二分摊类型的案件受理费段落或单元句，服务器可将所提取到的受理费分摊数据抽象成相应的费用分摊模型。受理费分摊模型比如“费用X元，分摊人A负担Y元，分摊人B负担Z元”，或者“费用共计X元，分摊人A负担Y元，分摊人B负担Z元”等。对于第三分摊类型的案件受理费段落，服务器将其分割为第一分摊类型和/或第二分摊类型的单元句。
上述实施例中，针对不同受理费分摊类型的案件受理费段落，通过不同的预设受理费分摊表达式从该案件受理费段落中提取相应的受理费分摊数据，提高了受理费分摊数据的提取效率，从而提高了律师专业能力的分析效率。
在其中一个实施例中，步骤S210之前，上述裁判文书处理方法还包括：基于预设标签表达式，从裁判文书中提取目标标签；当目标标签与预设标签集合不匹配时，执行步骤S210。
预设标签表达式是预先设定的用于通过正则匹配方式从裁判文书中提取目标标签的正则表达式。预设标签表达式具体可用于通过正则匹配方式从判决段落中提取目标标签。目标标签是指预先指定的从裁判文书中提取的文书标签，比如驳回、撤回、发回重审、商标权权属纠纷、侵害商标权纠纷和二审等。预设标签集合是由多于一个的预设标签组成的标签集合。预设标签是预先设定的文书标签。预设标签包括但不限于是驳回、撤回、发回重审、商标权权属纠纷、侵害商标权纠纷和二审。
具体地，服务器将预设标签表达式与裁判文书中的各个段落分别进行匹配，以从裁判文书中提取相应的目标标签，并将所提取到的目标标签与预设标签集合进行匹配。服务器根据匹配结果对应确定原告律师胜诉率和被告律师减损率的计算方式，根据所确定的计算方式对应计算原告律师胜诉率和被告律师减损率。当匹配结果为匹配失败时，即当目标标签与预设标签集合不匹配时，服务器根据所提取到的诉请金额项和相应的诉请金额值，以及判决金额项和相应的判决金额值，分别计算原告律师胜诉率和被告律师减损率。
在其中一个实施例中，当匹配结果为匹配成功时，即当目标标签与预设标签集合相匹配时，服务器根据预设标签集合中与目标标签相匹配的预设标签，对应确定原告律师胜诉 率和被告律师减损率的计算方式，并根据所确定的计算方式对应计算原告律师胜诉率和被告律师减损率。
在其中一个实施例中，预设标签集合中包括裁定标签、知识产权标签和二审标签。服务器对预设标签集合中不同类型的预设标签按照优先级进行排序，并按照优先级排序将目标标签依次与预设标签集合中的预设标签进行匹配。优先级排序具体可以是裁判标签的优先级优于知识产权标签，知识产权标签的优先级优于二审标签。裁定标签可以包括多个标签，比如驳回、撤回和发回重审。知识产权标签可以包括多个标签，比如商标权权属纠纷、侵害商标权纠纷。
进一步地，首先，服务器将目标标签与预设标签集合中的裁定标签进行匹配，当匹配成功时，服务器根据匹配成功的裁定标签对应确定原告律师胜诉率和被告律师减损率。其次，当匹配失败时，服务器将目标标签与预设标签集合中的知识产品标签进行匹配，当预设标签集合中存在与目标标签相匹配的知识产权标签时，服务器根据匹配成功的知识产权标签对应确定原告律师胜诉率和被告律师减损率。再次，当预设标签集合中不存在与目标标签相匹配的知识产权标签时，服务器将目标标签与预设标签集合中的二审标签进行匹配，当匹配成功时，服务器从所提取到的案件受理费段落中提取受理费分摊数据，并根据所提取到的受理费分摊数据对应计算原告律师胜诉率和被告律师减损率。
举例说明，当目标标签为驳回时，服务器则判定原告律师胜诉率为0、被告律师减损率100％。当目标标签为撤回时，服务器则判定原告律师胜诉率为50％、被告律师减损率50％。
在其中一个实施例中，服务器通过关键词匹配的方式从裁判文书或判决段落中提取相应的目标标签。在一个实施例中，服务器从裁判文书中提取与知识产权标签相匹配的目标标签，当没有提取到目标标签时，服务器进一步提取与裁定标签相匹配的目标标签。
上述实施例中，提供了多种计算原告律师胜诉率和被告律师减损率的计算方式，并根据所提取的目标标签所对应的计算方式，对应计算原告律师胜诉率和被告律师减损率，提高了计算效率，从而提高了律师专业能力的分析效率。
在其中一个实施例中，预设标签集合中包括知识产权标签；从裁判文书中提取目标标签之后，上述裁判文书处理方法还包括：当目标标签与知识产权标签相匹配时，将判决段落与知识产权标签所对应的预设知识产权表达式进行匹配；当判决段落与预设知识产权表达式匹配成功时，根据预设知识产权表达式对应的预设确定方式确定原告律师胜诉率和被告律师减损率；当判决段落与预设知识产权表达式匹配失败时，执行步骤S210。
知识产权标签是用于表征相应案件为知识产权案件的标签。知识产权标签包括但不限于商标权权属纠纷和侵害商标权纠纷。预设知识产权表达式是预先设定的用于通过正则匹配方式从判决段落中提取与知识产权判决结果相对应的目标表达式。预设知识产权表达式比如“驳回.*？的诉讼请求”、“诉讼请求不予支持”、“驳回.*？的其他诉讼请求”和“停止”等。
具体地，当目标标签与预设标签集合中的知识产权标签相匹配时，服务器对应确定匹配成功的知识产权标签所对应的预设知识产权表达式，将所确定的预设知识产权表达式与判决段落进行匹配。当判决段落与预设知识产权表达式匹配成功时，服务器根据该预设知识产权表达式所对应的预设确定方式，对应确定原告律师胜诉率和被告律师减损率。当判决段落与预设知识产权表达式匹配失败时，服务器根据所提取到的诉请金额项和相应的诉请金额值，以及判决金额项和相应的判决金额值，分别计算原告律师胜诉率和被告律师减损率。
在其中一个实施例中，知识产权标签有多个，每个知识产权标签对应有至少一个预设知识产权表达式。当与目标标签匹配成功的知识产权标签对应有多个预设知识产权表达式时，服务器将该多个预设知识产权表达式分别与判决段落进行匹配，并根据匹配成功的预设知识产权表达式所对应的预设确定方式，对应确定原告律师胜诉率和被告律师减损率。
举例说明，假设目标标签为商标权权属纠纷，商标权权属纠纷对应的预设知识产权表达式有“驳回.*？的诉讼请求”、“诉讼请求不予支持”和“驳回.*？的其他诉讼请求”。当与判决段落匹配成功的预设知识产权表达式为“驳回.*？的诉讼请求”或“诉讼请求不予支持”时，服务器则确定原告律师胜诉率为0、被告律师减损率为100％。当与判决段落匹配成功的预设知识产权表达式为“驳回.*？的其他诉讼请求”时，则确定原告律师胜诉率和被告律师减损率均为50％。
假设目标标签为侵害商标权纠纷，侵害商标权纠纷对应的预设知识产权表达式有“.*？停止.*？”。当判决段落与该预设知识产权表达式匹配成功时，即判决段落中包含有关键词“停止”时，服务器则确定原告律师胜诉率为100％、被告律师减损率为0。
上述实施例中，对于涉及到知识产权的案件，根据判决段落所对应的知识产权标签对应确定计算原告律师胜诉率和被告律师减损率的计算方式，提高了计算效率，从而提高了律师专业能力的分析效率。
在其中一个实施例中，段落提取模型的训练步骤包括：获取多个目标裁判文书；分别标注出每个目标裁判文书中的目标诉请段落和目标判决段落；获取目标诉请段落对应的诉请段落提取问题，以及目标判决段落对应的判决段落提取问题；将目标裁判文书、诉请段落提取问题和判决段落提取问题作为输入特征，将相应的目标诉请段落和目标判决段落作为期望的输出特征，对长久记忆神经网络进行训练获得已训练的段落提取模型。
诉请段落提取问题是用于从目标裁判文书中提取诉请段落的提问数据，比如“原告诉讼请求对应于哪些段落？”。判决段落提取问题是用于从目标裁判文书中提取判决段落的提问数据，比如“法院判决结果对应于哪些段落？”。长久记忆神经网络具体可以是End-To-End Memory Networks，具体可以是单层记忆网络，也可以是多层记忆网络。
具体地，服务器获取多个目标裁判文书，并分别标注出每个目标裁判文书中的目标诉请段落和目标判决段落。服务器分别获取目标诉请段落对应的诉请段落提取问题，以及目标判决段落对应的判决段落提取问题。服务器将目标裁判文书和相应的诉请段落提取问题 作为输入特征，将相应的目标诉请段落作为期望的输出特征，获得第一训练样本集。类似地，服务器将目标裁判文书和相应的判决段落提取问题作为输入特征，将相应的目标判决段落作为期望的输出特征，获得第二训练样本集。进一步地，服务器基于所获取到的第一训练样本集和第二训练样本集，对长久记忆神经网络进行交替训练获得已训练的段落提取模型。交替训练是指分别利用第一训练样本集中的第一训练样本和第二训练样本集中的第二训练样本，交替对长久记忆神经网络进行训练。
在其中一个实施例中，服务器根据第一训练样本集进行训练，获得用于从裁判文书中提取诉请段落的段落提取模型。类似地，服务器根据第二训练样本集进行训练，获得用于从裁判文书中提取判决段落的段落提取模型。在一个实施例中，第一训练样本集和第二训练样本集中的输入特征均为目标裁判文书。
在其中一个实施例中，服务器在进行模型训练时，将交叉熵损失函数作为目标函数对长久记忆神经网络进行训练。
上述实施例中，基于训练样本集进行模型训练，以通过训练获得的段落提取模型从裁判文书中分别提取诉请段落和判决段落，提高了提取效率。
在其中一个实施例中，实体识别模型的训练步骤包括：获取多个目标裁判文书；通过已训练的段落提取模型分别从每个目标裁判文书中提取目标诉请段落和目标判决段落；分别标注出目标诉请段落和目标判决段落中的目标金额项；将目标诉请段落和目标判决段落作为输入特征，将相应的目标金额项作为期望的输出特征，对初始化的实体识别模型进行训练获得已训练的实体识别模型。
具体地，服务器获取多个目标裁判文书，并通过预先训练好的的段落提取模型分别提取相应的目标诉请段落和目标判决段落。对于所提取到的目标诉请段落和目标判决段落，服务器分别标注出相应的目标金额项，即标注出目标诉请段落中的目标诉请金额项和目标判决段落中的目标判决金额项。服务器将目标诉请段落作为输入特征，相应的目标诉请金额项作为期望的输出特征，获得相应的诉请训练样本集。类似地，服务器将目标判决段落作为输入特征，相应的目标判决金额项作为期望的输出特征，获得相应的判决训练样本集。进一步地，服务器根据诉请训练样本集和判决训练样本集对初始化的实体识别模型进行交替训练获得已训练的实体识别模型。
在其中一个实施例中，服务器根据诉请训练样本集进行训练，获得用于从诉请段落中提取诉请金额项的实体识别模型。类似地，服务器根据判决训练样本集进行训练，获得用于从判决段落中提取判决金额项的实体识别模型。
在其中一个实施例中，实体识别模型具体可以是bilstm-crf模型。
上述实施例中，基于训练样本集进行训练，以通过训练获得的实体识别模型从诉请段落提取诉请金额项和判决段落提取判决金额项，提高了提取效率。
指的说明的是，上述各个实施例中的“.*？”代表相应位置处存在一个或多个字符。字符具体可以是汉字、数字、字母和符号等。
如图3所示，在其中一个实施例中，提供了一种裁判文书处理方法，该方法具体包括以下步骤：
S302，获取裁判文书。
S304，通过已训练的段落提取模型从裁判文书中提取诉请段落和判决段落。
S306，通过已训练的实体识别模型从诉请段落中提取诉请金额项，以及从判决段落中提取判决金额项。
S308，当提取到诉请金额项和判决金额项时，基于预设金额项表达式，从诉请段落中提取与诉请金额项对应的诉请金额值，以及从判决段落中提取与判决金额项对应的判决金额值。
S310，基于预设标签表达式，从裁判文书中提取目标标签。
S312，当目标标签与预设标签集合不匹配时，根据诉请金额项和相应的诉请金额值，以及判决金额项和相应的判决金额值，分别计算原告律师胜诉率和被告律师减损率。
S314，当目标标签与知识产权标签相匹配时，将判决段落与知识产权标签所对应的预设知识产权表达式进行匹配。
S316，当判决段落与预设知识产权表达式匹配成功时，根据预设知识产权表达式对应的预设确定方式确定原告律师胜诉率和被告律师减损率。
S318，当判决段落与预设知识产权表达式匹配失败时，根据诉请金额项和相应的诉请金额值，以及判决金额项和相应的判决金额值，分别计算原告律师胜诉率和被告律师减损率。
S320，当没有提取到诉请金额项和判决金额项时，基于预设受理费表达式，从判决段落中提取案件受理费段落。
S322，按照语义顺序依次提取案件受理费段落中的预设关键词。
S324，根据预设关键词按照预设分类条件确定案件受理费段落对应的受理费分摊类型。
S326，根据受理费分摊类型对应的预设受理费分摊表达式，从案件受理费段落中提取受理费分摊数据。
S328，根据受理费分摊数据分别计算原告律师胜诉率和被告律师减损率。
上述实施例中，提供了多种计算原告律师胜诉率和被告律师减损率的计算方式，以便于根据所获取到的裁判文书选择相适应的计算方式，提高了计算效率，从而提高了律师专业能力的分析效率。
应该理解的是，虽然图2和图3的流程图中的各个步骤按照箭头的指示依次显示，但是这些步骤并不是必然按照箭头指示的顺序依次执行。除非本文中有明确的说明，这些步骤的执行并没有严格的顺序限制，这些步骤可以以其它的顺序执行。而且，图2和图3中的至少一部分步骤可以包括多个子步骤或者多个阶段，这些子步骤或者阶段并不必然是在同一时刻执行完成，而是可以在不同的时刻执行，这些子步骤或者阶段的执行顺序也不 必然是依次进行，而是可以与其它步骤或者其它步骤的子步骤或者阶段的至少一部分轮流或者交替地执行。
在其中一个实施例中，如图4所示，提供了一种裁判文书处理装置400，包括：获取模块401、段落提取模型402、金额项提取模块403、金额值提取模块404和计算模块405，其中：
获取模块401，用于获取裁判文书。
段落提取模型402，用于通过已训练的段落提取模型从裁判文书中提取诉请段落和判决段落。
金额项提取模块403，用于通过已训练的实体识别模型从诉请段落中提取诉请金额项，以及从判决段落中提取判决金额项。
金额值提取模块404，用于当提取到诉请金额项和判决金额项时，基于预设金额项表达式，从诉请段落中提取与诉请金额项对应的诉请金额值，以及从判决段落中提取与判决金额项对应的判决金额值。
计算模块405，用于根据诉请金额项和相应的诉请金额值，以及判决金额项和相应的判决金额值，分别计算原告律师胜诉率和被告律师减损率。
如图5所示，在其中一个实施例中，裁判文书处理装置400还包括：分摊数据提取模块406；
段落提取模型402，还用于当没有提取到诉请金额项和判决金额项时，基于预设受理费表达式，从判决段落中提取案件受理费段落；分摊数据提取模块406，用于基于预设受理费分摊表达式，从案件受理费段落提取受理费分摊数据；计算模块405，还用于根据受理费分摊数据分别计算原告律师胜诉率和被告律师减损率。
在其中一个实施例中，分摊数据提取模块406，还用于按照语义顺序依次提取案件受理费段落中的预设关键词；根据预设关键词按照预设分类条件确定案件受理费段落对应的受理费分摊类型；根据受理费分摊类型对应的预设受理费分摊表达式，从案件受理费段落中提取受理费分摊数据。
在其中一个实施例中，裁判文书处理装置400还包括：标签提取模块407；标签提取模块407，用于基于预设标签表达式，从裁判文书中提取目标标签；当目标标签与预设标签集合不匹配时，使得计算模块405执行根据诉请金额项和相应的诉请金额值，以及判决金额项和相应的判决金额值，分别计算原告律师胜诉率和被告律师减损率的步骤。
在其中一个实施例中，预设标签集合中包括知识产权标签；标签提取模块407，还用于当目标标签与知识产权标签相匹配时，将判决段落与知识产权标签所对应的预设知识产权表达式进行匹配；当判决段落与预设知识产权表达式匹配成功时，使得计算模块405还用于根据预设知识产权表达式对应的预设确定方式确定原告律师胜诉率和被告律师减损率；当判决段落与预设知识产权表达式匹配失败时，计算模块405执行根据诉请金额项和 相应的诉请金额值，以及判决金额项和相应的判决金额值，分别计算原告律师胜诉率和被告律师减损率的步骤。
在其中一个实施例中，裁判文书处理装置400还包括：模型训练模块408；
模型训练模块408，用于获取多个目标裁判文书；分别标注出每个目标裁判文书中的目标诉请段落和目标判决段落；获取目标诉请段落对应的诉请段落提取问题，以及目标判决段落对应的判决段落提取问题；将目标裁判文书、诉请段落提取问题和判决段落提取问题作为输入特征，将相应的目标诉请段落和目标判决段落作为期望的输出特征，对长久记忆神经网络进行训练获得已训练的段落提取模型。
在其中一个实施例中，模型训练模块408，还用于获取多个目标裁判文书；通过已训练的段落提取模型分别从每个目标裁判文书中提取目标诉请段落和目标判决段落；分别标注出目标诉请段落和目标判决段落中的目标金额项；将目标诉请段落和目标判决段落作为输入特征，将相应的目标金额项作为期望的输出特征，对初始化的实体识别模型进行训练获得已训练的实体识别模型。
关于裁判文书处理装置的具体限定可以参见上文中对于裁判文书处理方法的限定，在此不再赘述。上述裁判文书处理装置中的各个模块可全部或部分通过软件、硬件及其组合来实现。上述各模块可以硬件形式内嵌于或独立于计算机设备中的处理器中，也可以以软件形式存储于计算机设备中的存储器中，以便于处理器调用执行以上各个模块对应的操作。
在其中一个实施例中，提供了一种计算机设备，该计算机设备可以是服务器，其内部结构图可以如图6所示。该计算机设备包括通过系统总线连接的处理器、存储器、网络接口和数据库。该计算机设备的处理器用于提供计算和控制能力。该计算机设备的存储器包括非易失性存储介质、内存储器。该非易失性存储介质存储有操作系统、计算机可读指令和数据库。该内存储器为非易失性存储介质中的操作系统和计算机可读指令的运行提供环境。该计算机设备的数据库用于存储裁判文书和预设金额项表达式。该计算机设备的网络接口用于与外部的终端通过网络连接通信。该计算机可读指令被处理器执行时以实现一种裁判文书处理方法。
本领域技术人员可以理解，图6中示出的结构，仅仅是与本申请方案相关的部分结构的框图，并不构成对本申请方案所应用于其上的计算机设备的限定，具体的计算机设备可以包括比图中所示更多或更少的部件，或者组合某些部件，或者具有不同的部件布置。
一种计算机设备，包括存储器和一个或多个处理器，存储器中储存有计算机可读指令，计算机可读指令被一个或多个处理器执行时，使得一个或多个处理器实现本申请任意一个实施例中提供的裁判文书处理方法的步骤。
一个或多个存储有计算机可读指令的非易失性计算机可读存储介质，计算机可读指令被一个或多个处理器执行时，使得一个或多个处理器实现本申请任意一个实施例中提供的 裁判文书处理方法的步骤。
本领域普通技术人员可以理解实现上述实施例方法中的全部或部分流程，是可以通过计算机可读指令来指令相关的硬件来完成，所述的计算机可读指令可存储于一非易失性计算机可读取存储介质中，该计算机可读指令在执行时，可包括如上述各方法的实施例的流程。本申请所提供的各实施例中所使用的对存储器、存储、数据库或其它介质的任何引用，均可包括非易失性和/或易失性存储器。非易失性存储器可包括只读存储器(ROM)、可编程ROM(PROM)、电可编程ROM(EPROM)、电可擦除可编程ROM(EEPROM)或闪存。易失性存储器可包括随机存取存储器(RAM)或者外部高速缓冲存储器。作为说明而非局限，RAM以多种形式可得，诸如静态RAM(SRAM)、动态RAM(DRAM)、同步DRAM(SDRAM)、双数据率SDRAM(DDRSDRAM)、增强型SDRAM(ESDRAM)、同步链路(Synchlink)DRAM(SLDRAM)、存储器总线(Rambus)直接RAM(RDRAM)、直接存储器总线动态RAM(DRDRAM)、以及存储器总线动态RAM(RDRAM)等。
以上实施例的各技术特征可以进行任意的组合，为使描述简洁，未对上述实施例中的各个技术特征所有可能的组合都进行描述，然而，只要这些技术特征的组合不存在矛盾，都应当认为是本说明书记载的范围。
以上所述实施例仅表达了本申请的几种实施方式，其描述较为具体和详细，但并不能因此而理解为对发明专利范围的限制。应当指出的是，对于本领域的普通技术人员来说，在不脱离本申请构思的前提下，还可以做出若干变形和改进，这些都属于本申请的保护范围。因此，本申请专利的保护范围应以所附权利要求为准。

Claims (20)

  1. 一种裁判文书处理方法，包括：
    获取裁判文书；
    通过已训练的段落提取模型从所述裁判文书中提取诉请段落和判决段落；
    通过已训练的实体识别模型从所述诉请段落中提取诉请金额项，以及从所述判决段落中提取判决金额项；
    当提取到所述诉请金额项和所述判决金额项时，基于预设金额项表达式，从所述诉请段落中提取与所述诉请金额项对应的诉请金额值，以及从所述判决段落中提取与所述判决金额项对应的判决金额值；及
    根据所述诉请金额项和相应的所述诉请金额值，以及所述判决金额项和相应的所述判决金额值，分别计算原告律师胜诉率和被告律师减损率。
  2. 根据权利要求1所述的方法，其特征在于，还包括：
    当没有提取到所述诉请金额项和所述判决金额项时，基于预设受理费表达式，从所述判决段落中提取案件受理费段落；
    基于预设受理费分摊表达式，从所述案件受理费段落提取受理费分摊数据；及
    根据所述受理费分摊数据分别计算原告律师胜诉率和被告律师减损率。
  3. 根据权利要求2所述的方法，其特征在于，所述基于预设受理费分摊表达式，从所述案件受理费段落提取受理费分摊数据，包括：
    按照语义顺序依次提取所述案件受理费段落中的预设关键词；
    根据所述预设关键词按照预设分类条件确定所述案件受理费段落对应的受理费分摊类型；及
    根据所述受理费分摊类型对应的预设受理费分摊表达式，从所述案件受理费段落中提取受理费分摊数据。
  4. 根据权利要求1所述的方法，其特征在于，所述根据所述诉请金额项和相应的所述诉请金额值，以及所述判决金额项和相应的所述判决金额值，分别计算原告律师胜诉率和被告律师减损率之前，所述方法还包括：
    基于预设标签表达式，从所述裁判文书中提取目标标签；及
    当所述目标标签与预设标签集合不匹配时，执行所述根据所述诉请金额项和相应的所述诉请金额值，以及所述判决金额项和相应的所述判决金额值，分别计算原告律师胜诉率和被告律师减损率的步骤。
  5. 根据权利要求4所述的方法，其特征在于，所述预设标签集合中包括知识产权标签；所述从所述裁判文书中提取目标标签之后，所述方法还包括：
    当所述目标标签与所述知识产权标签相匹配时，将所述判决段落与所述知识产权标签所对应的预设知识产权表达式进行匹配；
    当所述判决段落与所述预设知识产权表达式匹配成功时，根据所述预设知识产权表达 式对应的预设确定方式确定原告律师胜诉率和被告律师减损率；及
    当所述判决段落与所述预设知识产权表达式匹配失败时，执行所述根据所述诉请金额项和相应的所述诉请金额值，以及所述判决金额项和相应的所述判决金额值，分别计算原告律师胜诉率和被告律师减损率的步骤。
  6. 根据权利要求1至5任意一项所述的方法，其特征在于，所述段落提取模型的训练步骤包括：
    获取多个目标裁判文书；
    分别标注出每个所述目标裁判文书中的目标诉请段落和目标判决段落；
    获取所述目标诉请段落对应的诉请段落提取问题，以及所述目标判决段落对应的判决段落提取问题；及
    将所述目标裁判文书、所述诉请段落提取问题和所述判决段落提取问题作为输入特征，将相应的所述目标诉请段落和所述目标判决段落作为期望的输出特征，对长久记忆神经网络进行训练获得已训练的段落提取模型。
  7. 根据权利要求1至5任意一项所述的方法，其特征在于，所述实体识别模型的训练步骤包括：
    获取多个目标裁判文书；
    通过已训练的段落提取模型分别从每个所述目标裁判文书中提取目标诉请段落和目标判决段落；
    分别标注出所述目标诉请段落和所述目标判决段落中的目标金额项；及
    将所述目标诉请段落和所述目标判决段落作为输入特征，将相应的所述目标金额项作为期望的输出特征，对初始化的实体识别模型进行训练获得已训练的实体识别模型。
  8. 一种裁判文书处理装置，包括：
    获取模块，用于获取裁判文书；
    段落提取模型，用于通过已训练的段落提取模型从所述裁判文书中提取诉请段落和判决段落；
    金额项提取模块，用于通过已训练的实体识别模型从所述诉请段落中提取诉请金额项，以及从所述判决段落中提取判决金额项；
    金额值提取模块，用于当提取到所述诉请金额项和所述判决金额项时，基于预设金额项表达式，从所述诉请段落中提取与所述诉请金额项对应的诉请金额值，以及从所述判决段落中提取与所述判决金额项对应的判决金额值；及
    计算模块，用于根据所述诉请金额项和相应的所述诉请金额值，以及所述判决金额项和相应的所述判决金额值，分别计算原告律师胜诉率和被告律师减损率。
  9. 一种计算机设备，包括存储器和一个或多个处理器，所述存储器中储存有计算机可读指令，所述计算机可读指令被所述一个或多个处理器执行时，使得所述一个或多个处理器执行以下步骤：
    获取裁判文书；
    通过已训练的段落提取模型从所述裁判文书中提取诉请段落和判决段落；
    通过已训练的实体识别模型从所述诉请段落中提取诉请金额项，以及从所述判决段落中提取判决金额项；
    当提取到所述诉请金额项和所述判决金额项时，基于预设金额项表达式，从所述诉请段落中提取与所述诉请金额项对应的诉请金额值，以及从所述判决段落中提取与所述判决金额项对应的判决金额值；及
    根据所述诉请金额项和相应的所述诉请金额值，以及所述判决金额项和相应的所述判决金额值，分别计算原告律师胜诉率和被告律师减损率。
  10. 根据权利要求9所述的计算机设备，其特征在于，所述处理器执行所述计算机可读指令时还执行以下步骤：
    当没有提取到所述诉请金额项和所述判决金额项时，基于预设受理费表达式，从所述判决段落中提取案件受理费段落；
    基于预设受理费分摊表达式，从所述案件受理费段落提取受理费分摊数据；及
    根据所述受理费分摊数据分别计算原告律师胜诉率和被告律师减损率。
  11. 根据权利要求10所述的计算机设备，其特征在于，所述基于预设受理费分摊表达式，从所述案件受理费段落提取受理费分摊数据，包括：
    按照语义顺序依次提取所述案件受理费段落中的预设关键词；
    根据所述预设关键词按照预设分类条件确定所述案件受理费段落对应的受理费分摊类型；及
    根据所述受理费分摊类型对应的预设受理费分摊表达式，从所述案件受理费段落中提取受理费分摊数据。
  12. 根据权利要求9所述的计算机设备，其特征在于，所述计算机可读指令被所述处理器执行时，使得所述处理器在执行所述根据所述诉请金额项和相应的所述诉请金额值，以及所述判决金额项和相应的所述判决金额值，分别计算原告律师胜诉率和被告律师减损率之前，还执行以下步骤：
    基于预设标签表达式，从所述裁判文书中提取目标标签；及
    当所述目标标签与预设标签集合不匹配时，执行所述根据所述诉请金额项和相应的所述诉请金额值，以及所述判决金额项和相应的所述判决金额值，分别计算原告律师胜诉率和被告律师减损率的步骤。
  13. 根据权利要求12所述的计算机设备，其特征在于，所述计算机可读指令被所述处理器执行时，使得所述处理器在执行所述预设标签集合中包括知识产权标签；所述从所述裁判文书中提取目标标签之后，还执行以下步骤：
    当所述目标标签与所述知识产权标签相匹配时，将所述判决段落与所述知识产权标签所对应的预设知识产权表达式进行匹配；
    当所述判决段落与所述预设知识产权表达式匹配成功时，根据所述预设知识产权表达式对应的预设确定方式确定原告律师胜诉率和被告律师减损率；及
    当所述判决段落与所述预设知识产权表达式匹配失败时，执行所述根据所述诉请金额项和相应的所述诉请金额值，以及所述判决金额项和相应的所述判决金额值，分别计算原告律师胜诉率和被告律师减损率的步骤。
  14. 根据权利要求9至13任意一项所述的计算机设备，其特征在于，所述处理器执行所述计算机可读指令时还执行所述段落提取模型的训练步骤，包括：
    获取多个目标裁判文书；
    分别标注出每个所述目标裁判文书中的目标诉请段落和目标判决段落；
    获取所述目标诉请段落对应的诉请段落提取问题，以及所述目标判决段落对应的判决段落提取问题；及
    将所述目标裁判文书、所述诉请段落提取问题和所述判决段落提取问题作为输入特征，将相应的所述目标诉请段落和所述目标判决段落作为期望的输出特征，对长久记忆神经网络进行训练获得已训练的段落提取模型。
  15. 一个或多个存储有计算机可读指令的非易失性计算机可读存储介质，所述计算机可读指令被一个或多个处理器执行时，使得所述一个或多个处理器执行以下步骤：
    获取裁判文书；
    通过已训练的段落提取模型从所述裁判文书中提取诉请段落和判决段落；
    通过已训练的实体识别模型从所述诉请段落中提取诉请金额项，以及从所述判决段落中提取判决金额项；
    当提取到所述诉请金额项和所述判决金额项时，基于预设金额项表达式，从所述诉请段落中提取与所述诉请金额项对应的诉请金额值，以及从所述判决段落中提取与所述判决金额项对应的判决金额值；及
    根据所述诉请金额项和相应的所述诉请金额值，以及所述判决金额项和相应的所述判决金额值，分别计算原告律师胜诉率和被告律师减损率。
  16. 根据权利要求15所述的存储介质，其特征在于，所述计算机可读指令被所述处理器执行时还执行以下步骤：
    当没有提取到所述诉请金额项和所述判决金额项时，基于预设受理费表达式，从所述判决段落中提取案件受理费段落；
    基于预设受理费分摊表达式，从所述案件受理费段落提取受理费分摊数据；及
    根据所述受理费分摊数据分别计算原告律师胜诉率和被告律师减损率。
  17. 根据权利要求16所述的存储介质，其特征在于，所述基于预设受理费分摊表达式，从所述案件受理费段落提取受理费分摊数据，包括：
    按照语义顺序依次提取所述案件受理费段落中的预设关键词；
    根据所述预设关键词按照预设分类条件确定所述案件受理费段落对应的受理费分摊 类型；及
    根据所述受理费分摊类型对应的预设受理费分摊表达式，从所述案件受理费段落中提取受理费分摊数据。
  18. 根据权利要求15所述的存储介质，其特征在于，所述计算机可读指令被所述处理器执行时，使得所述处理器在执行所述根据所述诉请金额项和相应的所述诉请金额值，以及所述判决金额项和相应的所述判决金额值，分别计算原告律师胜诉率和被告律师减损率之前，还执行以下步骤：
    基于预设标签表达式，从所述裁判文书中提取目标标签；及
    当所述目标标签与预设标签集合不匹配时，执行所述根据所述诉请金额项和相应的所述诉请金额值，以及所述判决金额项和相应的所述判决金额值，分别计算原告律师胜诉率和被告律师减损率的步骤。
  19. 根据权利要求18所述的存储介质，其特征在于，所述计算机可读指令被所述处理器执行时，使得所述处理器在执行所述预设标签集合中包括知识产权标签；所述从所述裁判文书中提取目标标签之后，还执行以下步骤：
    当所述目标标签与所述知识产权标签相匹配时，将所述判决段落与所述知识产权标签所对应的预设知识产权表达式进行匹配；
    当所述判决段落与所述预设知识产权表达式匹配成功时，根据所述预设知识产权表达式对应的预设确定方式确定原告律师胜诉率和被告律师减损率；及
    当所述判决段落与所述预设知识产权表达式匹配失败时，执行所述根据所述诉请金额项和相应的所述诉请金额值，以及所述判决金额项和相应的所述判决金额值，分别计算原告律师胜诉率和被告律师减损率的步骤。
  20. 根据权利要求15至19任意一项所述的存储介质，其特征在于，所述计算机可读指令被所述处理器执行时还执行所述段落提取模型的训练步骤，包括：
    获取多个目标裁判文书；
    分别标注出每个所述目标裁判文书中的目标诉请段落和目标判决段落；
    获取所述目标诉请段落对应的诉请段落提取问题，以及所述目标判决段落对应的判决段落提取问题；及
    将所述目标裁判文书、所述诉请段落提取问题和所述判决段落提取问题作为输入特征，将相应的所述目标诉请段落和所述目标判决段落作为期望的输出特征，对长久记忆神经网络进行训练获得已训练的段落提取模型。
PCT/CN2019/071516 2018-09-10 2019-01-14 裁判文书处理方法、装置、计算机设备和存储介质 WO2020052184A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201811051928.7 2018-09-10
CN201811051928.7A CN109446511B (zh) 2018-09-10 2018-09-10 裁判文书处理方法、装置、计算机设备和存储介质

Publications (1)

Publication Number Publication Date
WO2020052184A1 true WO2020052184A1 (zh) 2020-03-19

Family

ID=65530979

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/071516 WO2020052184A1 (zh) 2018-09-10 2019-01-14 裁判文书处理方法、装置、计算机设备和存储介质

Country Status (2)

Country Link
CN (1) CN109446511B (zh)
WO (1) WO2020052184A1 (zh)

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111598742A (zh) * 2020-05-14 2020-08-28 鼎富智能科技有限公司 一种从判决书获取当事人量刑要素的方法及装置
CN111753537A (zh) * 2020-06-12 2020-10-09 鼎富智能科技有限公司 离婚纠纷裁判文书标签提取方法及装置
CN111784505A (zh) * 2020-06-30 2020-10-16 鼎富智能科技有限公司 一种借贷纠纷判决书提取方法及装置
CN112989830A (zh) * 2021-03-08 2021-06-18 武汉大学 一种基于多元特征和机器学习的命名实体识别方法
CN113569538A (zh) * 2020-04-29 2021-10-29 北京国双科技有限公司 文书的生成方法、装置、存储介质及电子设备
CN116484010A (zh) * 2023-03-15 2023-07-25 北京擎盾信息科技有限公司 知识图谱构建方法、装置、存储介质及电子装置

Families Citing this family (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110825872B (zh) * 2019-09-11 2023-05-23 成都数之联科技股份有限公司 一种提取和分类诉讼请求信息的方法及系统
CN110781299B (zh) * 2019-09-18 2024-03-19 平安科技（深圳）有限公司 资产信息识别方法、装置、计算机设备及存储介质
CN110765889B (zh) * 2019-09-29 2024-06-25 平安直通咨询有限公司上海分公司 法律文书的特征提取方法、相关装置及存储介质
CN111126064A (zh) * 2019-11-14 2020-05-08 四川隧唐科技股份有限公司 金额识别方法、装置、计算机设备和可读存储介质
CN111177332B (zh) * 2019-11-27 2023-11-24 中证信用增进股份有限公司 自动提取裁判文书涉案标的和裁判结果的方法及装置
TWI798513B (zh) 2019-12-20 2023-04-11 國立清華大學 自然語言語料用於機器學習決策模型的訓練方法
CN111858938B (zh) * 2020-07-23 2024-05-24 鼎富智能科技有限公司 一种裁判文书标签的提取方法及装置

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106815203A (zh) * 2015-12-01 2017-06-09 北京国双科技有限公司 一种裁判文书中的金额解析方法及装置
CN108197163A (zh) * 2017-12-14 2018-06-22 上海银江智慧智能化技术有限公司 一种基于裁判文书的结构化处理方法
CN108287818A (zh) * 2018-01-03 2018-07-17 小草数语（北京）科技有限公司 裁判文书中金额的提取方法、装置和电子设备
CN108334500A (zh) * 2018-03-05 2018-07-27 上海思贤信息技术股份有限公司 一种基于机器学习算法的裁判文书标注方法及装置

Family Cites Families (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105447750B (zh) * 2015-11-17 2022-06-03 小米科技有限责任公司 信息识别方法、装置、终端及服务器
CN106815201B (zh) * 2015-12-01 2021-06-08 北京国双科技有限公司 一种自动判定裁判文书判决结果的方法及装置
CN106815266B (zh) * 2015-12-01 2020-06-16 北京国双科技有限公司 裁判文书检索方法和装置
KR101838948B1 (ko) * 2016-04-29 2018-03-15 주식회사 헬프미 법률문서 자동 작성 방법 및 장치
CN107632968B (zh) * 2017-05-22 2021-01-05 南京大学 一种面向裁判文书的证据链关系模型的构建方法
CN108197099A (zh) * 2017-12-01 2018-06-22 厦门快商通信息技术有限公司 一种文本信息提取方法及计算机可读存储介质

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106815203A (zh) * 2015-12-01 2017-06-09 北京国双科技有限公司 一种裁判文书中的金额解析方法及装置
CN108197163A (zh) * 2017-12-14 2018-06-22 上海银江智慧智能化技术有限公司 一种基于裁判文书的结构化处理方法
CN108287818A (zh) * 2018-01-03 2018-07-17 小草数语（北京）科技有限公司 裁判文书中金额的提取方法、装置和电子设备
CN108334500A (zh) * 2018-03-05 2018-07-27 上海思贤信息技术股份有限公司 一种基于机器学习算法的裁判文书标注方法及装置

Cited By (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113569538A (zh) * 2020-04-29 2021-10-29 北京国双科技有限公司 文书的生成方法、装置、存储介质及电子设备
CN111598742A (zh) * 2020-05-14 2020-08-28 鼎富智能科技有限公司 一种从判决书获取当事人量刑要素的方法及装置
CN111753537A (zh) * 2020-06-12 2020-10-09 鼎富智能科技有限公司 离婚纠纷裁判文书标签提取方法及装置
CN111784505A (zh) * 2020-06-30 2020-10-16 鼎富智能科技有限公司 一种借贷纠纷判决书提取方法及装置
CN112989830A (zh) * 2021-03-08 2021-06-18 武汉大学 一种基于多元特征和机器学习的命名实体识别方法
CN112989830B (zh) * 2021-03-08 2023-08-18 武汉大学 一种基于多元特征和机器学习的命名实体识别方法
CN116484010A (zh) * 2023-03-15 2023-07-25 北京擎盾信息科技有限公司 知识图谱构建方法、装置、存储介质及电子装置
CN116484010B (zh) * 2023-03-15 2024-01-16 北京擎盾信息科技有限公司 知识图谱构建方法、装置、存储介质及电子装置

Also Published As

Publication number Publication date
CN109446511A (zh) 2019-03-08
CN109446511B (zh) 2022-07-08

Similar Documents

Publication Publication Date Title
WO2020052184A1 (zh) 裁判文书处理方法、装置、计算机设备和存储介质
WO2017092555A1 (zh) 一种裁判文书中的金额解析方法及装置
CN111291570B (zh) 一种实现司法文书中要素识别的方法及装置
CN110781299A (zh) 资产信息识别方法、装置、计算机设备及存储介质
CN106815201B (zh) 一种自动判定裁判文书判决结果的方法及装置
CN110610005A (zh) 基于深度学习的盗窃罪辅助量刑方法
CN110738039A (zh) 一种案件辅助信息的提示方法、装置、存储介质和服务器
CN110765760B (zh) 一种法律案件分配方法、装置、存储介质和服务器
CN110335180A (zh) 案件立案材料智能审查装置
CN106446070A (zh) 一种基于专利群的信息处理装置及方法
US20210133212A1 (en) Data archiving method and computing device implementing same
CN111930805A (zh) 一种信息挖掘方法及计算机设备
Fagan Big Data Legal Scholarship: Toward a Research Program and Practitionaer's Guide
CN108108346A (zh) 文档的主题特征词抽取方法及装置
CN113254651B (zh) 一种裁判文书的分析方法、装置、计算机设备及存储介质
CN116186284A (zh) 知识图谱的构建方法、应用方法及其对应装置
CN116757207A (zh) 基于人工智能的icd自动编码方法及相关设备
CN111611340A (zh) 信息抽取方法、装置、计算机设备和存储介质
CN113886538B (zh) 医保报销信息查询方法、装置、电子设备和存储介质
CN110956140B (zh) 图像信息提取方法、装置、设备以及存储介质
US11593564B2 (en) Systems and methods for extracting patent document templates from a patent corpus
Hong et al. Improving colloquial case legal judgment prediction via abstractive text summarization
Junior et al. A study of the influence of textual features in learning medical prior authorization
CN112530585A (zh) 基于医疗机构的数据处理方法、装置、计算机设备及介质
CN111291153A (zh) 一种信息搜索的方法和装置

Legal Events

Date Code Title Description
NENP Non-entry into the national phase

Ref country code: DE

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 15.06.2021)

122 Ep: pct application non-entry in european phase

Ref document number: 19859689

Country of ref document: EP

Kind code of ref document: A1