CN-121981101-A - Contract generation auxiliary decision-making method based on multiple agents

CN121981101ACN 121981101 ACN121981101 ACN 121981101ACN-121981101-A

Abstract

The invention relates to the technical field of intelligent contract generation, and discloses a contract generation auxiliary decision-making method based on multiple intelligent agents, which comprises the following steps: extracting initial contract elements from a contract generation request, constructing a shared working space containing a global knowledge graph and a modification influence abstract chain, constructing a collaboration network containing a plurality of special agents, calling coordinated control agents to activate the special agents in sequence to provide a modification scheme and update the shared working space, extracting conflict contexts from the modification influence abstract chain when consistency conflicts are detected, calling related special agents to negotiate and update the modification scheme, and generating a complete contract draft when all contract generation tasks are executed and no conflicts exist. According to the invention, multi-agent cooperation is realized through the shared working space, conflict tracing and intelligent negotiation are realized based on modification of the influence abstract chain, a multi-layer confidence propagation model optimization decision mechanism is introduced, and the quality, efficiency and traceability of contract generation are obviously improved.

Inventors

HU RONG
WANG JUN
GENG ZHE
LI HANG

Assignees

盘古云链(天津)数字科技有限公司

Dates

Publication Date: 20260505
Application Date: 20260407

Claims (9)

1. A multi-agent based contract generation aid decision making method, the method comprising: S1, extracting initial contract elements from a contract generation request, and constructing a shared working space, wherein the shared working space comprises a contract generation task which is being executed, a global knowledge graph for storing entities in the contract and relations among the entities, and a modification influence abstract chain for recording the history modification operation of each agent on the contract and the influence range of each agent; S2, constructing a collaboration network containing a plurality of special intelligent agents, and generating task dependency relations among the special intelligent agents based on the initial contract elements; S3, calling a coordination control intelligent agent to activate corresponding special intelligent agents in sequence according to the task dependency relationship, reading the executing contract generation task from the shared working space by the activated special intelligent agent, modifying the contract generation task to generate a corresponding modification influence abstract, and updating the modified contract generation task and the modification influence abstract to the shared working space; S4, when the coordination control intelligent agent detects that the current modification has consistency conflict with the global knowledge graph, extracting a history abstract related to the conflict from the modification influence abstract chain, constructing a conflict context, calling an associated special intelligent agent, negotiating based on the conflict context, generating a new modification and updating to the shared working space; and S5, generating a complete contract draft based on the contract generation task when all the contract generation tasks are executed and no conflict point exists.
2. The multi-agent based contract generation aid decision making method according to claim 1, wherein: In the step S1, the modification effect abstract chain adopts a chain data structure with version pointing, wherein each modification effect abstract comprises a current version identifier, a preamble version identifier, a modification operator and a list of affected entities and relationships between entities in a global knowledge graph; In the step S3, when the activated special agent updates the modified contract generating task to the shared working space, further generating a new modification influence abstract according to the global knowledge-graph entity related to the modification, setting the associated current version identifier as an increased new value, and pointing the preamble version identifier to the latest modified version; In the step S4, when a conflict is detected, the coordination control agent identifies an entity in the global knowledge graph related to the conflict, and extracts all modification influence digests from the modification influence digest chain from the current version to the version in which the entity is introduced for the first time by using the entity as a key, thereby forming the conflict context.
3. The multi-agent based contract generation aid decision making method according to claim 1, wherein: In the step S1, the global knowledge graph is constructed based on a multi-layer confidence propagation model, wherein an entity is at least associated with an initial confidence degree given by a special agent of an initial source of the entity, and a change history list for recording modification or verification of the entity by a subsequent special agent or a coordination control agent; In the step S3, when the special agent modifies the current contract generation task, if the modification involves confirmation or change of the existing entity relationship, the identification of the special agent associated with the current operation and the confidence adjustment value thereof are synchronously added in the change history list of the corresponding entity when the global knowledge graph is updated; And in the step S4, when the coordination control intelligent agent detects the conflict, the method further comprises the steps of calculating the current comprehensive confidence coefficient of the entity relation related to the conflict, triggering negotiation if the comprehensive confidence coefficient is lower than a preset confidence coefficient threshold value, automatically adopting a history version with the highest confidence coefficient if the comprehensive confidence coefficient is higher than the preset confidence coefficient threshold value, and writing the adoption behavior into the modification influence abstract chain.
4. The multi-agent based contract generation aid decision making method according to claim 1, wherein: Wherein the step of constructing a collaboration network comprising a plurality of specialized agents comprises: Dynamically screening special agents matched with a current contract generation task from a preset agent resource pool according to the industry field, legal type and risk dimension related to the initial contract element, wherein the special agents comprise an industry expert agent, a legal expert agent, a risk assessment agent and a language optimization agent; Loading a corresponding domain knowledge base and a corresponding special prompt template for each special intelligent agent, and initializing each special intelligent agent based on the initial contract elements; Registering each initialized special agent to a blank relation network, and distributing a special communication identifier and a message queue for each special agent to form the collaboration network.
5. The multi-agent based contract generation aid decision making method according to claim 4, wherein: after the collaboration network is formed, the method further comprises the step of generating task dependency relations among the special intelligent agents based on the initial contract elements: Carrying out dependency analysis on the initial contract elements, and identifying mandatory preconditions and optional optimization sequences existing in the initial contract elements; constructing a basic dependency skeleton based on the mandatory preconditions, wherein the basic dependency skeleton comprises a directed edge set conforming to business logic and legal standards; performing topological ordering on the optional optimization sequence with the aim of minimizing inter-agent communication overhead to generate a dynamic execution sequence, and attaching the dynamic execution sequence to the scheduling configuration of the cooperative network in a metadata form; and writing the basic dependency skeleton and the dynamic execution sequence into a task queue manager of the coordination control agent as the task dependency relationship.
6. The multi-agent based contract generation aid decision making method according to claim 1, wherein: the step of calling the coordination control agent to activate the corresponding special agent in turn according to the task dependency relationship comprises the following steps: the coordination control intelligent agent reads a special intelligent agent currently provided with a to-be-executed mark from the task dependency relationship and a pre-condition list of the special intelligent agent; Inquiring the global knowledge graph, and for the special intelligent agent meeting all the preconditions in the corresponding precondition list, calculating the matching degree between the special intelligent agent and the entity which is not processed in the current contract generation task based on the professional field of the special intelligent agent; The coordination control agent sequentially sends activation instructions to the special agents according to the sequence from high to low of the matching degree, and the activated special agents are marked as 'executing' in the shared working space.
7. The multi-agent based contract generation aid decision making method according to claim 1, wherein: Wherein the activated dedicated agent reads the executing contract generation task from the shared workspace and modifies the contract generation task, the step of generating a corresponding modification influence summary comprising: The activated special agent analyzes the current contract generation task, identifies the to-be-processed clauses belonging to the professional field of the special agent, and further locks the corresponding node positions and the corresponding global identifiers of the to-be-processed clauses in the global knowledge graph; The activated special agent performs semantic analysis and logic reasoning on the to-be-processed clause based on the domain knowledge base of the special agent to generate a modification scheme, performs local consistency pre-verification on the modification scheme and the current state of the related entity in the global knowledge graph, and judges whether potential conflict exists or not; If the local consistency pre-verification passes, executing the modification scheme, and synchronously generating a modification influence abstract, wherein the abstract comprises global identifiers of modified clauses, text contents before and after modification, a list of influenced entities and relationships among the entities and an influence propagation range list, namely, other clause identifier lists possibly needing synchronous adjustment due to the modification; If the local consistency pre-verification is not passed, according to the pre-judged conflict type, a corresponding strategy is called from a preset correction strategy library to adjust the modification scheme, and the pre-verification is repeated until the pre-verification is passed or the maximum try number is reached; If the local consistency pre-verification reaches the maximum number of attempts and fails, the modification is terminated and the exception is reported.
8. The multi-agent based contract generation aid decision making method according to claim 1, wherein: wherein the step of the coordination control agent detecting a consistency conflict of the current modification with the global knowledge-graph comprises: Performing logic consistency check on the global knowledge graph entity related to the current modification and the relation between the entities, and judging that logic consistency conflict exists if the fact that circular dependence or type mismatch exists is recognized; carrying out semantic consistency verification on terms related to current modification, calculating semantic similarity between a modified text and a history definition based on the history definition of a related entity in the global knowledge graph, and judging that semantic consistency conflict exists if the similarity is lower than a preset similarity threshold; And carrying out compliance consistency verification on the clauses related to the current modification, reading a compliance attribute mark preset by legal expert intelligent agents in the global knowledge graph, and judging that compliance conflict exists if the modification results in forced deletion or change of the attribute mark.
9. The multi-agent based contract generation aid decision making method according to claim 1, wherein: Wherein the step of coordinating control agents invoking associated dedicated agents, negotiating based on the conflicting context, generating new modifications and updating to the shared workspace comprises: The coordination control intelligent agent invokes a matched negotiation protocol from a preset negotiation protocol library according to the conflict type, wherein the negotiation protocol comprises a legal priority protocol, an industry convention priority protocol and a weighted negotiation protocol; transmitting the conflict context and the selected negotiation protocol to the associated special intelligent agents, and receiving the correction scheme and scheme basis returned by each special intelligent agent; and determining a final scheme according to preset arbitration rules in the selected negotiation protocol, and adding the record of the negotiation to the modification influence abstract chain.

Description

Contract generation auxiliary decision-making method based on multiple agents Technical Field The invention relates to the technical field of intelligent contract generation, in particular to a contract generation auxiliary decision-making method based on multiple intelligent agents. Background Traditional contract drafting mainly depends on manual modes, standards are different and risks are difficult to control, most of existing automatic contract generation schemes are generated based on template filling, a rule engine or natural language, the text generation logic of the schemes is single, and complex requirements of multiple dimensions such as legal compliance, business suitability, risk controllability, language normalization and the like in the contract drafting process are difficult to be considered. In recent years, multi-agent systems are applied in the field of complex task processing due to the characteristics of distribution, collaboration, autonomy and the like, in a contract generation scene, researchers try to construct a collaboration system consisting of a plurality of specialized agents, however, in the existing scheme, the agents lack a uniform context sharing mechanism, task scheduling is difficult to adapt to dynamic changes of contract contents, conflict processing still depends on preset rules or manual intervention, and a systematic and intelligent solution is lacking. In the process of contract generation, the modification opinions of different professional perspectives often conflict with each other, when the prior art processes such conflict, the traceability of the conflict source and a negotiation mechanism based on historical basis are lacked, and in addition, when the inherent contradiction or compliance risk occurs in the contract text, the source of the problem is difficult to trace. The knowledge graph is used as a structural knowledge representation mode and is widely applied to the fields of information retrieval, question answering systems and the like, however, the existing scheme mostly uses the knowledge graph as a static knowledge base, and particularly in a multi-agent collaboration scene, the problems of reliability difference of different source information, influence of modification operation on the knowledge graph, decision support based on a historical version and the like are solved, and no mature technical solution exists at present. Disclosure of Invention The invention aims to solve the problems in the prior art and provides a multi-agent-based contract generation auxiliary decision-making method. In order to achieve the purpose, the invention adopts the following technical scheme that the contract generation auxiliary decision-making method based on multiple intelligent agents comprises the following steps: S1, extracting initial contract elements from a contract generation request, and constructing a shared working space, wherein the shared working space comprises a contract generation task which is being executed, a global knowledge graph for storing entities in the contract and relations among the entities, and a modification influence abstract chain for recording the history modification operation of each agent on the contract and the influence range of each agent; S2, constructing a collaboration network containing a plurality of special intelligent agents, and generating task dependency relations among the special intelligent agents based on the initial contract elements; S3, calling a coordination control intelligent agent to activate corresponding special intelligent agents in sequence according to the task dependency relationship, reading the executing contract generation task from the shared working space by the activated special intelligent agent, modifying the contract generation task to generate a corresponding modification influence abstract, and updating the modified contract generation task and the modification influence abstract to the shared working space; S4, when the coordination control intelligent agent detects that the current modification has consistency conflict with the global knowledge graph, extracting a history abstract related to the conflict from the modification influence abstract chain, constructing a conflict context, calling an associated special intelligent agent, negotiating based on the conflict context, generating a new modification and updating to the shared working space; and S5, generating a complete contract draft based on the contract generation task when all the contract generation tasks are executed and no conflict point exists. The technical scheme provided by the invention has the beneficial effects that at least: The invention provides unified context environment and history traceability for a plurality of special intelligent agents by constructing the shared working space containing the global knowledge graph and modifying the influence abstract chain, and simultaneously carries out task arrangement by dyn