A method and system for whole-process case handling assistance driven by a large language model

Abstract

The application discloses a method and system for whole-process case handling assistance driven by a large language model, and comprises the following steps: S1, judging whether received case file materials belong to an existing case; if yes, entering S2; if no, entering S3; S2, automatically identifying and filing the case file materials into a holographic archive of a corresponding target case, wherein a target case type template already exists in the holographic archive, and entering S4; S3, creating a new case as a target case, automatically identifying and filing the case file materials into a holographic archive of the target case, and matching a corresponding case type template from a case type template library as a target case type template according to a case type of the target case, and entering S4; S4, extracting behavior facts from the case file materials by using an industry special large model, and filling entities into the target case type template by using the behavior facts; S5, adapting a condition of a situation identification rule according to the behavior facts, and giving a situation identification qualitative conclusion; and S6, analyzing requirements of the target case type supported by the situation identification qualitative conclusion, and giving a case conclusion.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of case handling, in particular to a method and system for whole-process auxiliary case handling driven by a large language model. BACKGROUND

[0002] The applicant previously designed to fill in the case template (knowledge graph principle construction) with the case file materials to form a case graph, and the case graph is used for reference by the case handling personnel, so that the case handling personnel can quickly understand the case situation. On this basis, the large language model is used to identify the behavior facts from the case file materials, and the case conclusion such as whether the evidence involved is established can be analyzed based on the scenario identification, which is more convenient for the case handling personnel to understand the case situation and improve the work efficiency of the case handling personnel. SUMMARY

[0003] The present application provides a method and system for whole-process auxiliary case handling driven by a large language model.

[0004] The present application solves the above technical problems by the following technical scheme:

[0005] The present application provides a method for whole-process auxiliary case handling driven by a large language model, which is characterized by comprising the following steps:

[0006] S1, receiving a case file material, judging whether the case file material belongs to an existing case, if yes, entering step S2, if not, entering step S3;

[0007] S2, automatically identifying the case file material into the holographic archive of the corresponding target case, the holographic archive already exists the matched target case template, entering step S4;

[0008] S3, creating a new case as a target case, automatically identifying the case file material into the holographic archive of the target case, and matching the corresponding case template from the case template library according to the case type of the target case as the target case template, entering step S4;

[0009] S4, extracting the behavior facts in the case file material according to the target case template and using the industry special large model, and filling the entities in the target case template with the extracted behavior facts, the target case template is constructed by using the knowledge graph principle;

[0010] The industry special large model is a large language model obtained by fine-tuning and reinforcement learning based on a model base and using industry-specific task design and related data sets;

[0011] S5, adapting the scenario identification rule condition according to the extracted behavior facts, and giving a scenario identification qualitative conclusion;

[0012] S6, according to the case identification qualitative conclusion analysis case identification qualitative conclusion supported by the target case of the elements of the case, based on the supported target case of the elements of the case gives the case conclusion;

[0013] S7, the target case atlas constituted by the entity filling, the case identification qualitative conclusion and the case conclusion are stored in the holographic archives of the target case.

[0014] The application also provides a system for whole-process auxiliary case handling driven by a large language model, which is characterized in that it comprises a case file judgment module, a first filing module, a second filing module, an extraction and filling module, a case identification module, a case analysis module and a storage module.

[0015] The case file judgment module is used to receive a case file material, judge whether the case file material belongs to an existing case, and call the first filing module if yes, or call the second filing module if no.

[0016] The first filing module is used to automatically identify the case file material into the holographic archives of the corresponding target case, and the holographic archives already exist the matched target case template, and call the extraction and filling module.

[0017] The second filing module is used to create a new case as a target case, automatically identify the case file material into the holographic archives of the target case, match the corresponding case template from the case template library as the target case template according to the case type of the target case, and call the extraction and filling module.

[0018] The extraction and filling module is used to extract the behavior facts in the case file material according to the target case template and using an industry-specific large model, and perform entity filling on the target case template using the extracted behavior facts, and the target case template is constructed using the knowledge graph principle.

[0019] The industry-specific large model is a large language model obtained by fine-tuning and reinforcement learning based on a model base and using industry-specific task design and related data sets.

[0020] The case identification module is used to adapt the case identification rule conditions according to the extracted behavior facts, and give the case identification qualitative conclusion.

[0021] The case analysis module is used to analyze the elements of the target case supported by the case identification qualitative conclusion according to the case identification qualitative conclusion, and give the case conclusion based on the supported elements of the target case.

[0022] The storage module is used to store the target case atlas constituted by the entity filling, the case identification qualitative conclusion and the case conclusion into the holographic archives of the target case.

[0023] The positive progress effect of the present application is that the present application makes a plot determination on the basis of the behavior facts proposed by the large language model from the case file materials, and based on the plot determination, it can further analyze the conclusion of the case, such as whether the incriminating evidence is established, which is more convenient for the case handling personnel to understand the case situation and improves the work efficiency of the case handling personnel. BRIEF DESCRIPTION OF DRAWINGS

[0024] Figure 1 The flowchart of the method for whole-process case handling assistance driven by a large language model according to the preferred embodiment of the present application.

[0025] Figure 2 The structural block diagram of the system for whole-process case handling assistance driven by a large language model according to the preferred embodiment of the present application. DETAILED DESCRIPTION

[0026] To make the objectives, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0027] As shown in Figure 1 The present embodiment provides a method for whole-process case handling assistance driven by a large language model, which comprises the following steps:

[0028] Step 101: receiving a case file material, determining whether the case file material belongs to an existing case, if yes, entering step 102, if not, entering step 103.

[0029] Step 102: automatically identifying the case file material into the holographic archive of the corresponding target case, the holographic archive already exists the matched target case template, entering step 104.

[0030] Step 103: creating a new case as a target case, automatically identifying the case file material into the holographic archive of the target case, and matching the corresponding case template from the case template library as the target case template according to the case type of the target case, entering step 104.

[0031] In the present scheme, if the case material is the new case material of the existing case, the case material is directly identified into the holographic archive of the corresponding target case in the existing case; if the case material does not belong to the existing case, a target case needs to be newly created, and the case material is identified into the holographic archive of the newly created target case.

[0032] Step 104, extracting the behavior facts in the case file materials according to the target case type template and using the industry-specific large model, and filling the target case type template with the extracted behavior facts.

[0033] The target case type template is constructed using the knowledge graph principle. The industry-specific large model is a large language model based on a model base and fine-tuned and reinforced learning using industry-specific task design and related data sets.

[0034] The case file materials are document materials, and step 104 includes the following steps:

[0035] S41, determine whether the document material is a formatted document or a non-formatted document. If it is a formatted document, go to step S42. If it is a non-formatted document, go to step S43.

[0036] S42, first extract the standard content in the formatted document using the pre-set regular expression and the pre-trained BERT model, extract the entity information from it and fill it into the corresponding node in the target case type template, and then extract the non-standard content in the formatted document using the industry-specific large model, extract the entity information from it and fill it into the corresponding node in the target case type template, and extract the non-standard content in the formatted document using the industry-specific large model according to the node breadth-first strategy in the target case type template.

[0037] S43, directly extract the standard content and non-standard content in the non-formatted document using the industry-specific large model, extract the entity information from it and fill it into the corresponding node in the target case type template, and extract the non-standard content in the non-formatted document using the industry-specific large model according to the node breadth-first strategy in the target case type template.

[0038] For specific implementation of steps S42 and S43, please refer to the invention patent with application number 2024106805417 and title "A method and system for improving holographic archive extraction efficiency and quality".

[0039] Step S42:

[0040] S421, for the target case type template, extract the standard content in the formatted document using the pre-set regular expression and the pre-trained BERT model, and extract the entity information from it and fill it into the corresponding node in the target case type template;

[0041] S422, determine whether each node in the i-th layer of the target case type template has been filled with entity information, if yes, go to step S423, if no, go to step S424, where 1≤i≤N, N is the number of layers of the target case type template, and N is a positive integer;

[0042] S423, i = i + 1, repeat step S422;

[0043] S424, using an industry-specific large model to extract non-standard content in the formatted document, and extracting entity information from the non-standard content to fill in the corresponding node of the layer in the target case template that has not been filled in and has not been marked as not needing extraction and filling operation;

[0044] Using an industry-specific large model to extract the process according to the node breadth-first strategy in the target case template layer by layer:

[0045] Traverse the layer, and for a node in the layer that has not been filled in and has not been marked as not needing extraction and filling operation, if entity content is extracted from the formatted document, it is filled into the node, and if the node is the tail node of a triple, reverse extraction verification is needed, if consistent, it means that the entity content of the node is correct, then the next layer node to which the node belongs needs to be extracted and filled in, if inconsistent, it means that the entity content of the node is incorrect, and all nodes to which the node belongs do not need to be extracted and filled in, and all nodes to which the node belongs are marked as not needing extraction and filling operation, if the entity content of the node is not extracted from the formatted document, all nodes to which the node belongs do not need to be extracted and filled in, and all nodes to which the node belongs are marked as not needing extraction and filling operation;

[0046] S425, determine whether all nodes in the i-th layer of the target case template have been traversed, if yes, go to step S426, if no, repeat step S424;

[0047] S426, i = i + 1, determine whether i ≤ N, if yes, repeat step S424, if no, end the process.

[0048] Among them, the behavior fact written according to a certain standard constitutes standard content, and the behavior fact not written according to a certain standard constitutes non-standard content.

[0049] Moreover, before the quality of the BERT model meets the preset requirements, the standard content in the formatted document is extracted using a pre-set regular expression, the data extracted by the regular expression is accumulated as a data sample set to train the BERT model, and after the quality of the BERT model meets the preset requirements after sample training, the trained BERT model is used to extract the standard content in the formatted document.

[0050] In this step, for the standard content in the formatted document, regular expressions and BERT models are used to extract and fill in the target case template, with higher extraction accuracy and efficiency. For non-standard content in formatted documents, standard content in non-formatted documents, and non-standard content, industry-specific large models are used to extract and fill in the target case template.

[0051] Step 105, according to the extracted behavior facts, adapt the conditions of the case recognition rules, and give the qualitative conclusion of the case recognition.

[0052] Step 106, according to the qualitative conclusion of the case recognition, analyze the requirements of the target case supported by the qualitative conclusion of the case recognition, and give the case conclusion based on the requirements of the supported target case.

[0053] Step 107, store the target case graph constituted by the filled entities, the qualitative conclusion of the case recognition, and the case conclusion into the holographic archive of the target case.

[0054] In step 106, analyze which requirements of the target case are supported by the qualitative conclusion of the case recognition, and determine whether the target case is established based on the supported requirements. If it is completely established, give the case conclusion that the target case is established. If it is not completely established, analyze the possible case based on the supported requirements, take the possible case as a new target case, and take the case template corresponding to the new target case as a new target case template. Execute steps 104-105, and then display the target case graph and the qualitative conclusion of the case recognition to the case handling personnel to determine whether the target case is established or the new target case is established. After the case handling personnel confirm, store the confirmed established case and the corresponding target case graph and qualitative conclusion of the case recognition into the holographic archive of the target case.

[0055] For example, receiving an A1 case material, judging that the A1 case material belongs to an existing case A, then automatically identifying the A1 case material into the holographic archive of the A case, the holographic archive of the A case already exists a matched target case type template such as a theft crime template, extracting the behavior facts in the A1 case material according to the theft crime template and using the industry special large model, filling the entities of the theft crime template with the extracted behavior facts, obtaining the theft crime graph of the A case. According to the extracted behavior facts, adapt the condition of the situation identification rule (such as the confession situation, the reduced punishment situation of theft crime, the aggravated punishment situation of theft crime, etc.), give the situation identification qualitative conclusion. Analyze the situation identification qualitative conclusion to support which elements of the theft crime, based on the supported elements to determine whether the theft crime is established, if it is completely established, give the case conclusion that the theft crime of the A case is established, if it does not completely meet the elements of the theft crime, based on the supported elements to analyze the possible case type such as robbery, match the robbery template, extract the behavior facts in the A1 case material according to the theft crime template and use the industry special large model, fill the entities of the robbery template with the extracted behavior facts, obtain the robbery graph of the A case. According to the extracted behavior facts, adapt the condition of the situation identification rule (such as the confession situation, the reduced punishment situation of theft crime, the aggravated punishment situation of theft crime, etc.), give the situation identification qualitative conclusion, and then display the target case graph and the situation identification qualitative conclusion of the two times to the case handling personnel to determine whether the A case is a theft crime or a robbery.

[0056] As shown in Figure 2 The embodiment also provides a system for whole-process auxiliary case handling driven by a large language model, which comprises a case judgment module 1, a first filing module 2, a second filing module 3, an extraction and filling module 4, a situation identification module 5, a case analysis module 6, a storage module 7 and a display module 8.

[0057] The case judgment module 1 is used for receiving a case material, judging whether the case material belongs to an existing case, calling the first filing module 2 if yes, and calling the second filing module 3 if no.

[0058] The first filing module 2 is used for automatically identifying the case material into the holographic archive of the corresponding target case, the holographic archive already exists a matched target case type template, and calling the extraction and filling module 4.

[0059] The second filing module 3 is used for creating a new case as a target case, automatically identifying the case material into the holographic archive of the target case, matching the corresponding case type template from the case type template library as the target case type template according to the case type of the target case, and calling the extraction and filling module 4.

[0060] The extraction and filling module 4 is used to extract behavior facts in the case file materials according to the target case type template and using the industry-specific large model, fill the target case type template with the extracted behavior facts, and the target case type template is constructed using the knowledge graph principle; the industry-specific large model is a large language model obtained by fine-tuning and reinforcement learning based on a model base and using industry-specific task design and related data sets.

[0061] The case file materials are document materials, and the extraction and filling module 4 includes a type judgment unit, a first extraction and filling unit, and a second extraction and filling unit.

[0062] The type judgment unit is used to judge whether the document material is a formatted document or a non-formatted document, and the first extraction and filling unit is called when it is a formatted document, and the second extraction and filling unit is called when it is a non-formatted document.

[0063] The first extraction and filling unit is used to first extract the standard content in the formatted document using a pre-set regular expression and a pre-trained BERT model, extract entity information from it and fill it into the corresponding node in the target case type template, and then extract the non-standard content in the formatted document using the industry-specific large model, extract entity information from it and fill it into the corresponding node in the target case type template, and extract the non-standard content in the formatted document using the industry-specific large model according to the node breadth-first strategy in the target case type template.

[0064] The second extraction and filling unit is used to directly extract the standard content and non-standard content in the non-formatted document using the industry-specific large model, extract entity information from it and fill it into the corresponding node in the target case type template, and extract the non-standard content in the non-formatted document using the industry-specific large model according to the node breadth-first strategy in the target case type template.

[0065] The behavior facts constitute the standard content according to certain standards, and the behavior facts constitute the non-standard content without writing according to certain standards.

[0066] Moreover, the pre-set regular expression is used to extract the standard content in the formatted document before the quality of the BERT model meets the preset requirements, and the data extracted by the regular expression is accumulated as a data sample set to train the BERT model, and the trained BERT model is used to extract the standard content in the formatted document after the quality of the BERT model meets the preset requirements after sample training.

[0067] The scenario identification module 5 is used to adapt the scenario identification rule conditions according to the extracted behavior facts, and give a scenario identification qualitative conclusion.

[0068] The case analysis module 6 is used to analyze the elements of the target case supported by the case determination qualitative conclusion, and give a case conclusion based on the supported elements of the target case.

[0069] The storage module 7 is used to store the target case graph constituted by the filled entity, the case determination qualitative conclusion and the case conclusion into the holographic archives of the target case.

[0070] The case analysis module 6 is used to analyze which elements of the target case are supported by the case determination qualitative conclusion, and determine whether the target case is established based on the supported elements. If the target case is completely established, a case conclusion that the target case is established is given. If the target case is not completely established, possible cases are analyzed based on the supported elements. The possible cases are taken as a new target case, and the case template corresponding to the new target case is taken as a new target case template. The extraction and filling module 4 and the case determination module 5 are called again, and the display module 8 is called again to display the target case graph and the case determination qualitative conclusion of the previous and the latter two times to the case handling personnel, so as to determine whether the target case is established or the new target case is established.

[0071] Although the specific embodiments of the present application are described above, those skilled in the art should understand that these are only illustrative, and the protection scope of the present application is defined by the appended claims. Those skilled in the art can make various changes or modifications to these embodiments without departing from the principles and essence of the present application, and these changes and modifications all fall within the protection scope of the present application.

Claims

1. A method for whole-process case handling assistance driven by a large language model, characterized in that, It comprises the following steps: S1, receiving a case file material, judging whether the case file material belongs to an existing case, if yes, entering step S2, if not, entering step S3; S2, automatically identifying the case file material into the holographic archive of the corresponding target case, the holographic archive already exists the matched target case template, entering step S4; S3, creating a new case as a target case, automatically identifying the case file material into the holographic archive of the target case, matching the corresponding case template from the case template library as the target case template according to the case type of the target case, entering step S4; S4, extracting the behavior facts in the case file material according to the target case template and using the industry special large model, filling the entities in the target case template with the extracted behavior facts, the target case template is constructed using the knowledge graph principle; The industry special large model is a large language model obtained by fine-tuning and reinforcement learning based on a model base and using industry-specific task design and related data sets; S5, adapting the scenario recognition rule condition according to the extracted behavior facts, and giving a scenario recognition qualitative conclusion; S6, analyzing the requirements of the target case according to the scenario recognition qualitative conclusion, and giving a case conclusion based on the requirements of the target case; S7, storing the target case graph constituted by the entity filling, the scenario recognition qualitative conclusion and the case conclusion into the holographic archive of the target case; The case file material is a document material, and step S4 comprises the following steps: S41, judging whether the document material is a formatted document or a non-formatted document, if it is a formatted document, entering step S42, if it is a non-formatted document, entering step S43; S42, first extracting the standard content in the formatted document using the pre-set regular expression and the pre-trained BERT model, extracting the entity information from the standard content and filling it into the corresponding node of the target case template, then extracting the non-standard content in the formatted document using the industry special large model, extracting the entity information from the non-standard content and filling it into the corresponding node of the target case template, and extracting the non-standard content in the formatted document using the industry special large model according to the node breadth-first strategy in the target case template; S43, directly extracting the standard content and non-standard content in the non-formatted document using the industry special large model, extracting the entity information from the standard content and non-standard content and filling it into the corresponding node of the target case template, and extracting the non-standard content in the formatted document using the industry special large model according to the node breadth-first strategy in the target case template; Wherein, the behavior facts written according to certain standards constitute standard content, and the behavior facts not written according to certain standards constitute non-standard content; Step S42: S421, for the target case template, extracting the standard content in the formatted document using the pre-set regular expression and the pre-trained BERT model, and extracting the entity information from the standard content and filling it into the corresponding node of the target case template; S422, judge whether each node of the i layer in the target case cause template has been filled with entity information, if yes, go to step S423, if not, go to step S424, wherein 1≤i≤N, N is the number of layers of the target case cause template, and N is a positive integer; S423, i=i+1, repeat step S422; S424, using an industry-specific large model to extract non-standard content in the formatted document, and extracting entity information from the non-standard content to fill in the corresponding node in the layer of the target case cause template which has not been filled and has not been marked as not needing extraction and filling operation; During the extraction process using the industry-specific large model, the nodes are extracted layer by layer according to the node breadth-first strategy in the target case cause template: Traverse the layer, wherein for a certain node in the layer which has not been filled and has not been marked as not needing extraction and filling operation, if entity content is extracted from the formatted document, fill it into the node, if the node is the tail node of a triple, reverse extraction verification is needed, if consistent, it means that the entity content of the node is correct, then the next layer node to which the node belongs needs to be extracted and filled, if inconsistent, it means that the entity content of the node is incorrect, all nodes to which the node belongs do not need to be extracted and filled, and all nodes to which the node belongs are marked as not needing extraction and filling operation, if the entity content of the node is not extracted from the formatted document, all nodes to which the node belongs do not need to be extracted and filled, and all nodes to which the node belongs are marked as not needing extraction and filling operation; S425, judge whether each node of the i layer in the target case cause template has been traversed, if yes, go to step S426, if not, repeat step S424; S426, i=i+1, judge whether i≤N, if yes, repeat step S424, if not, end the process.

2. The method for whole-process case handling assistance driven by a large language model according to claim 1, characterized in that, In step S6, analyze which elements of the target case cause are supported by the scenario determination qualitative conclusion, determine whether the target case cause is established based on the supported elements, if completely established, give a case conclusion that the target case cause is established, if not completely established, give a case conclusion that the target case cause is not completely established.

3. The method for whole-process case handling assistance driven by a large language model according to claim 2, characterized in that, If not completely established, analyze possible causes based on the supported elements, take the possible causes as a new target case cause, take the case cause template corresponding to the new target case cause as a new target case cause template, execute steps S4-S5, and then display the target case graph and the scenario determination qualitative conclusion of the two times to the case handling personnel, so that the case handling personnel determine whether the target case cause is established or the new target case cause is established.

4. The method for whole-process case handling assistance driven by a large language model according to claim 1, characterized in that, Before the quality of the BERT model meets the preset requirements, the standard content in the formatted document is extracted using a pre-set regular expression, the data extracted by the regular expression is accumulated as a data sample set to train the BERT model, and after the quality of the BERT model meets the preset requirements after sample training, the trained BERT model is used to extract the standard content in the formatted document.

5. A system for whole-process case handling assistance driven by a large language model, characterized in that, It includes a case file judgment module, a first filing module, a second filing module, an extraction and filling module, a scenario determination module, a case analysis module, and a storage module; The case judgment module is configured to receive a case material, judge whether the case material belongs to an existing case, call a first entry module if the case material belongs to the existing case, and call a second entry module if the case material does not belong to the existing case; The first entry module is configured to automatically identify the case material and enter the case material into a holographic archive of a corresponding target case, the holographic archive already containing a matched target case template, and call an extraction and filling module; The second entry module is configured to create a new case as the target case, automatically identify the case material and enter the case material into a holographic archive of the target case, match a corresponding case template from a case template library as the target case template according to a case type of the target case, and call the extraction and filling module; The extraction and filling module is configured to extract behavior facts in the case material according to the target case template and using an industry-specific large model, and fill entities in the target case template according to the extracted behavior facts, the target case template being constructed using a knowledge graph principle; The industry-specific large model is a large language model obtained by fine-tuning and reinforcement learning based on a model base and using industry-specific task design and related data sets; The scenario identification module is configured to adapt scenario identification rule conditions according to the extracted behavior facts, and give a scenario identification qualitative conclusion; The case analysis module is configured to analyze requirements of a target case template supported by the scenario identification qualitative conclusion according to the scenario identification qualitative conclusion, and give a case conclusion based on the requirements of the target case template supported by the scenario identification qualitative conclusion; The storage module is configured to store the target case template, the scenario identification qualitative conclusion, and the case conclusion after entity filling into the holographic archive of the target case; The case material is a document material, and the extraction and filling module includes a type judgment unit, a first extraction and filling unit, and a second extraction and filling unit; The type judgment unit is configured to judge whether the document material is a formatted document or an unformatted document, call the first extraction and filling unit if the document material is the formatted document, and call the second extraction and filling unit if the document material is the unformatted document; The first extraction and filling unit is configured to first extract standard content in the formatted document using a pre-set regular expression and a pre-trained BERT model, fill entity information extracted from the standard content into corresponding nodes in the target case template, extract non-standard content in the formatted document using the industry-specific large model, extract entity information from the non-standard content into the corresponding nodes in the target case template, and extract the non-standard content in the formatted document using the industry-specific large model according to a node breadth-first strategy in the target case template layer by layer; The second extraction and filling unit is configured to directly extract standard content and non-standard content in the unformatted document using the industry-specific large model, fill entity information extracted from the standard content and the non-standard content into corresponding nodes in the target case template, and extract the standard content and the non-standard content in the unformatted document using the industry-specific large model according to the node breadth-first strategy in the target case template layer by layer; The behavior facts constitute the standard content according to certain standards, and the behavior facts constitute the non-standard content without writing according to certain standards; The first extraction and filling unit includes: S421, for the target case cause template, using the pre-set regular expression and the pre-trained BERT model to extract the standard content in the formatted document, and extracting entity information to fill in the corresponding nodes in the target case cause template; S422, judging whether each node of the i-th layer in the target case cause template has been filled with entity information, if yes, entering S423, if not, entering S424, wherein 1≤i≤N, N is the number of layers of the target case cause template, and N is a positive integer; S423, i=i+1, repeating S422; S424, using an industry-specific large model to extract non-standard content in the formatted document, and extracting entity information to fill in the corresponding nodes in the target case cause template which have not been filled and have not been marked as no need for extraction and filling operation; During the extraction process using the industry-specific large model, the nodes are extracted layer by layer according to the node breadth-first strategy in the target case cause template: Traverse the layer, wherein for a node in the layer which has not been filled and has not been marked as no need for extraction and filling operation, when the entity content is extracted from the formatted document, fill it into the node, if the node is the tail node of a triple, reverse extraction verification is needed, if consistent, it means that the entity content of the node is correct, then the next layer of nodes to which the node belongs need to be extracted and filled, if inconsistent, it means that the entity content of the node is incorrect, and all nodes to which the node belongs do not need to be extracted and filled, and all nodes to which the node belongs are marked as no need for extraction and filling operation, when the entity content of the node is not extracted from the formatted document, all nodes to which the node belongs do not need to be extracted and filled, and all nodes to which the node belongs are marked as no need for extraction and filling operation; S425, judging whether each node of the i-th layer in the target case cause template has been traversed, if yes, entering S426, if not, repeating S424; S426, i=i+1, judging whether i≤N, if yes, repeating S424, if not, ending the process.

6. The system for whole-process case handling assistance driven by a large language model according to claim 5, wherein, The case analysis module is used to analyze which elements support the target case cause of the scenario determination and qualitative conclusion, determine whether the target case cause is established based on the supported elements, give a case conclusion that the target case cause is established when it is completely established, and give a case conclusion that the target case cause is not completely established when it is not completely established.

7. The system for whole-process case handling assistance driven by a large language model according to claim 6, wherein, The system further comprises a display module, which, when the target case cause is not completely established, analyzes possible case causes based on the supported elements, takes the possible case causes as a new target case cause, takes the case cause template corresponding to the new target case cause as a new target case cause template, calls the extraction and filling module and the scenario determination module, and then calls the display module to display the target case graph and the scenario determination and qualitative conclusion before and after the two times to the case handling personnel, so as to determine whether the target case cause is established or the new target case cause is established.

8. The system for whole-process case handling assistance driven by a large language model of claim 5, wherein, Before the quality of the BERT model reaches the preset requirement, the standard content in the formatted document is extracted by using a preset regular expression, and the data extracted by the regular expression is accumulated as a data sample set to train the BERT model; after the quality of the BERT model reaches the preset requirement after sample training, the standard content in the formatted document is extracted by using the trained BERT model.

Images