Event relation extraction method and device, electronic equipment and storage medium

Abstract

The invention provides an event relationship extraction method and device, electronic equipment and a storage medium, and relates to the technical field of information processing. The event relation extraction method comprises the following steps: obtaining a to-be-analyzed document and a to-be-analyzed event pair; inputting the to-be-analyzed document and the to-be-analyzed event pair into a relationship acquisition model to obtain a target event relationship, the target event relationship being used for representing an event relationship between the to-be-analyzed event pair, the target event relationship is obtained by predicting event relationships through a plurality of agents in the relationship acquisition model and integrating the event relationships, and the event relationships predicted by the plurality of agents are obtained by simulating debate competition interaction through the plurality of agents. According to the method, the performance of the large model on the event relation extraction task can be effectively improved, and the accuracy of the event relation extraction result is guaranteed.

Description

Technical Field

[0001] The present invention relates to the field of information processing technologies, and in particular, to an event relationship extraction method, apparatus, electronic device, and storage medium. Background Art

[0002] Event relationship extraction refers to extracting relationships such as causality, time sequence, and sub-events between events from a given document. A document contains multiple events. By obtaining the relationships between events, the development context of the events can be effectively understood, enhancing the understanding of the document content. In recent years, large language models (LLMs) have achieved very remarkable performance in various tasks in the field of natural language processing and demonstrated powerful understanding and generation capabilities. However, large language models still have the problem of hallucinations, generating plausible but incorrect results, which limits the application of large language models. Event knowledge can sort out the relationships between document contents and enhance the reasoning ability of large language models. Therefore, in the era of large language models, the event relationship extraction task becomes more important.

[0003] Current event relationship extraction methods based on large language models mainly use a single large language model, regarding it as an agent to gradually extract event relationships. However, existing verification has shown that a single large language model is not only overly confident in its own output results but also very stubborn, which will affect the model performance and the accuracy of the output event relationship results. Summary of the Invention

[0004] The present invention provides an event relationship extraction method, apparatus, electronic device, and storage medium to solve the problem that a single large language model has poor performance in event relationship extraction in the prior art, effectively improve the performance of large language models in the event relationship extraction task, and ensure the accuracy of event relationship extraction results.

[0005] The present invention provides an event relationship extraction method, including the following steps: Obtain a document to be analyzed and an event pair to be analyzed, where the event pair to be analyzed is a pair of events in the document to be analyzed; Input the document to be analyzed and the event pair to be analyzed into a relationship acquisition model to obtain a target event relationship, where the target event relationship is used to represent the event relationship between the event pair to be analyzed. Among them, the target event relationship is obtained by separately predicting event relationships by multiple agents in the relationship acquisition model and integrating them. The multiple agents separately predict event relationships by simulating the interaction of a debate competition, and each agent in the multiple agents is used to represent an instance of a large language model.

[0006] An event relationship extraction method provided by the present invention, wherein the multiple agents include a host agent, a proponent agent, an opponent agent, and an audience agent. The step of inputting the document to be analyzed and the event pair to be analyzed into a relationship acquisition model to obtain a target event relationship includes: According to the document to be analyzed and the event pair to be analyzed, obtain the proponent's debate view corresponding to the proponent agent and the opponent's debate view corresponding to the opponent agent, and the proponent's debate view is different from the opponent's debate view; Based on the proponent's debate view and the opponent's debate view, simulate the interaction process among different participants in a debate competition through the host agent, the proponent agent, the opponent agent, and the audience agent to obtain the event relationships predicted by each of the multiple agents; Integrate the event relationships predicted by each of the multiple agents to obtain the target event relationship.

[0007] An event relationship extraction method provided by the present invention, wherein the step of obtaining the proponent's debate view corresponding to the proponent agent and the opponent's debate view corresponding to the opponent agent according to the document to be analyzed and the event pair to be analyzed includes: Generate a summary of the event pair to be analyzed, and use the summary to predict the event relationship to obtain the proponent's debate view; Extract sentences related to the event to be analyzed from the document to be analyzed to obtain a sentence set, and use the sentence set to predict the event relationship to obtain the opponent's debate view.

[0008] An event relationship extraction method provided by the present invention, wherein the interaction process among different participants in the debate competition includes an opening stage, an argumentation stage, a rebuttal stage, a free discussion stage, and a summary stage. The step of simulating the interaction process among different participants in a debate competition through the host agent, the proponent agent, the opponent agent, and the audience agent based on the proponent's debate view and the opponent's debate view to obtain the event relationships predicted by each of the multiple agents includes: In the opening stage, the host agent introduces the debate rules, process, theme, the roles of the proponent agent and the opponent agent, the document to be analyzed, the event pair to be analyzed, the proponent's debate view corresponding to the proponent agent, and the opponent's debate view corresponding to the opponent agent; In the argumentation stage, the first debater agent of the proponent agent and the first debater agent of the opponent agent speak in turn, respectively stating their corresponding debate views and problem-solving ideas, and the problem-solving ideas are used to represent the strategies adopted for predicting the event relationship between the event pairs to be analyzed; In the refutation stage, the proponent agent and the opponent agent take turns to question each other's debate viewpoints and problem-solving ideas; In the free discussion stage, the proponent agent and the opponent agent respectively clarify their own debate viewpoints; In the summary stage, the first debater agent of the proponent agent and the first debater agent of the opponent agent respectively predict a first event relationship and a second event relationship based on all the information of the interaction process. The audience agent predicts a third event relationship based on all the information of the interaction process, the first event relationship, and the second event relationship. The host agent predicts a fourth event relationship based on the first event relationship, the second event relationship, and the third event relationship.

[0009] According to an event relationship extraction method provided by the present invention, in the refutation stage, the proponent agent and the opponent agent take turns to question each other's debate viewpoints and problem-solving ideas, including: The second debater agent of the proponent agent asks questions about the deficiencies in the debate viewpoints and problem-solving ideas put forward by the first debater agent of the opponent agent; The second debater agent of the opponent agent replies to the questions raised by the second debater agent of the proponent agent. At the same time, the second debater agent of the opponent agent asks questions about the deficiencies in the debate viewpoints and problem-solving ideas of the second debater agent of the proponent agent; The third debater agent of the opponent agent replies to the questions raised by the second debater agent of the proponent agent. At the same time, the third debater agent of the opponent agent asks questions about the deficiencies in the debate viewpoints and problem-solving ideas of the third debater agent of the proponent agent; The proponent agent and the opponent agent conduct Q&A in this way until all the agents in the proponent agent and the opponent agent participate in the Q&A.

[0010] According to an event relationship extraction method provided by the present invention, in the free discussion stage, the proponent agent and the opponent agent respectively clarify their own debate viewpoints, including: Combined with the previous interaction information, the proponent agent regenerates the summary of the event pair to be analyzed, and reiterates its own debate viewpoints and problem-solving ideas. The opponent agent re-extracts the set of sentences related to the event pair to be analyzed in the document to be analyzed, and reiterates its own debate viewpoints and problem-solving ideas; The proponent agent and the opponent agent respectively refute each other's debate viewpoints, and the maximum number of rounds of refutation is four.

[0011] According to an event relationship extraction method provided by the present invention, before obtaining the document to be analyzed and the event pair to be analyzed, the method further includes: Setting a plurality of agents in the relationship acquisition model, the plurality of agents being set with different positions, the plurality of agents including a proponent agent, an opponent agent, a moderator agent, and an audience agent; Setting the rules of the debate process in the relationship acquisition model, the debate including an opening stage, a debate stage, and a summary stage, and the debate stage including a statement stage, a rebuttal stage, and a free discussion stage; Setting the debate statement rules in the relationship acquisition model, the debate statement rules being used to instruct the proponent agent and the opponent agent to predict the event relationship, the debate methods to be adopted, and the debate steps to be adopted according to the document to be analyzed and the event pair to be analyzed before the debate.

[0012] The present invention also provides an event relationship extraction device, including the following modules: An information acquisition module, configured to acquire a document to be analyzed and an event pair to be analyzed, where the event pair to be analyzed is a pair of events in the document to be analyzed; A relationship extraction module, configured to input the document to be analyzed and the event pair to be analyzed into a relationship acquisition model to obtain a target event relationship, where the target event relationship is used to characterize the event relationship between the event pair to be analyzed. Among them, the target event relationship is obtained by separately predicting the event relationship by a plurality of agents in the relationship acquisition model and integrating them. The plurality of agents separately predicting the event relationship is obtained by simulating the interaction of a debate by the plurality of agents, and each agent in the plurality of agents is used to represent an instance of a large model.

[0013] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored on the memory and running on the processor. When the processor executes the computer program, the event relationship extraction method as described in any one of the above is implemented.

[0014] The present invention also provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the event relationship extraction method as described in any one of the above is implemented.

[0015] The present invention also provides a computer program product, including a computer program. When the computer program is executed by a processor, the event relationship extraction method as described in any one of the above is implemented.

[0016] The event relationship extraction method, device, electronic device, and storage medium provided by the present invention, for the event relationship extraction task, enhance the understanding ability of the large model for event relationships by organizing multiple agents to conduct a debate competition through a relationship acquisition model based on the large model, thereby improving the accuracy of the large model in event relationship extraction. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0018] Figure 1 It is a flowchart of the event relationship extraction method provided by the present invention.

[0019] Figure 2 It is a flowchart of the method for obtaining the target event relationship provided by the present invention.

[0020] Figure 3 It is a flowchart of the method for obtaining debate viewpoints provided by the present invention.

[0021] Figure 4 It is a flowchart of the method for obtaining the event relationships predicted by each of the multiple agents provided by the present invention.

[0022] Figure 5 It is a flowchart of the method for constructing the relationship acquisition model provided by the present invention.

[0023] Figure 6 It is a schematic diagram of the overall architecture of the event relationship extraction system provided by the present invention.

[0024] Figure 7 It is a schematic diagram of the structure of the event relationship extraction device provided by the present invention.

[0025] Figure 8 It is a schematic diagram of the structure of the electronic device provided by the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0026] To make the objectives, technical solutions, and advantages of the present invention clearer, the following will clearly and completely describe the technical solutions in the present invention in conjunction with the accompanying drawings in the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art without creative efforts based on the embodiments in the present invention belong to the scope of protection of the present invention.

[0027] Event relation extraction refers to extracting relationships such as causality, temporal order, and sub-events between events from a given document. A document contains multiple events. By obtaining the relationships between events, the development context of the events can be effectively understood, enhancing the understanding of the document content. In recent years, large language models (LLMs) have achieved very remarkable performance in various natural language processing tasks and demonstrated powerful understanding and generation capabilities. However, LLMs still have the problem of hallucinations, generating plausible but incorrect results, which limits their applications. Event knowledge can sort out the relationships between document contents and enhance the reasoning ability of LLMs. Therefore, in the era of LLMs, the task of event relation extraction becomes even more important.

[0028] Current research has started to explore the performance of LLMs in event relation extraction tasks, mainly using the chain of thought and self-reflection methods. The event relation extraction method based on the chain of thought allows the LLM to output the intermediate process step by step until the relation category is output. The self-reflection method, based on the chain of thought, enables the LLM to iteratively check its output results until the maximum number of iterations or until the LLM approves the output results.

[0029] Current event relation extraction methods based on LLMs mainly use a single LLM, regarding it as an agent to gradually extract event relations. However, existing verification has shown that a single LLM is not only overly confident in its output results but also very stubborn, which will affect the model performance and the accuracy of the output event relation results.

[0030] In view of this, the embodiments of the present invention provide an event relation extraction method, which obtains a document to be analyzed and an event pair to be analyzed; inputs the document to be analyzed and the event pair to be analyzed into a relation acquisition model to obtain a target event relation, where the target event relation is used to represent the event relation between the event pair to be analyzed. Among them, the target event relation is obtained by separately predicting event relations by multiple agents in the relation acquisition model and integrating them, and the multiple agents separately predict event relations through simulating the interaction in a debate competition. This method can effectively improve the performance of LLMs in event relation extraction tasks and ensure the accuracy of event relation extraction results.

[0031] Next, the technical solutions in the embodiments of the present invention will be described with reference to the accompanying drawings in the embodiments of the present invention.

[0032] Figure 1It is a schematic flowchart of the event relationship extraction method provided by the present invention. The event relationship extraction method can be applied to an electronic device, which can be various types of devices with information processing capabilities during implementation. For example, the electronic device may include a personal computer, a laptop, a handheld computer, or a server, etc.; the electronic device may also be a mobile terminal, for example, the mobile terminal may include a mobile phone, a vehicle-mounted computer, a tablet computer, or a projector, etc. As Figure 1 shown, the method may include the following steps 101 to step 102: Step 101: Obtain a document to be analyzed and an event pair to be analyzed, where the event pair to be analyzed is a pair of events in the document to be analyzed.

[0033] It should be noted that the manner of obtaining the document to be analyzed and the event pair to be analyzed may be obtained through user input or may be obtained through transmission from other devices. The present invention does not limit the method of obtaining the document to be analyzed and the event pair to be analyzed.

[0034] Step 102: Input the document to be analyzed and the event pair to be analyzed into a relationship acquisition model to obtain a target event relationship, where the target event relationship is used to characterize the event relationship between the event pair to be analyzed. Among them, the target event relationship is obtained by separately predicting event relationships by multiple agents in the relationship acquisition model and integrating them. The multiple agents separately predict event relationships by simulating a debate interaction among the multiple agents, and each agent among the multiple agents is used to represent an instance of a large model.

[0035] It should be noted that the relationship acquisition model includes multiple agents. By simulating a debate interaction among the multiple agents, event relationships predicted by the multiple agents are obtained, and then the multiple predicted event relationships are integrated to obtain the target event relationship, that is, the event relationship between the event pair to be analyzed. The event relationship extraction method based on multiple agents of a large model can effectively improve the performance of the large model in the event relationship extraction task and enhance the reasoning ability of the large model.

[0036] It can be understood that the present invention, by giving an article, an event and , during the process of a debate among different agents, the final event relationship is obtained through analysis and discussion. For the event relationship extraction task, a relationship acquisition model based on a large model is constructed, and a method of multiple-agent debate is adopted, that is, a debate competition is organized by large model agents to enhance the large model's understanding ability of event relationships and improve the accuracy of the large model in event relationship extraction.

[0037] Exemplarily, the present invention obtains a given document , where represents the document the th word, where and respectively represent the and th event trigger words, and . is a predefined set of event relationship types. An event trigger word refers to a word in a document that can trigger an event. In event-related tasks, event trigger words are generally used to represent events, so the word "event" will be directly used in the following descriptions. Assume is an event prediction system based on large model multi-agent, i.e., a relationship acquisition model. Given events and , the event prediction system needs to predict the relationship between events and and based on the document content

[0038] Figure 2 is a schematic flow diagram of the method for obtaining the target event relationship provided by the present invention. As Figure 2 shown, the multiple agents include a host agent, a proponent agent, an opponent agent, and an audience agent. The inputting the document to be analyzed and the pair of events to be analyzed into the relationship acquisition model to obtain the target event relationship may include: Step 201: According to the document to be analyzed and the pair of events to be analyzed, obtain the proponent's debate view corresponding to the proponent agent and the opponent's debate view corresponding to the opponent agent, and the proponent's debate view is different from the opponent's debate view.

[0039] It should be noted that there are many ways to obtain the proponent's debate view corresponding to the proponent agent and the opponent's debate view corresponding to the opponent agent. For example, it can be obtained by analyzing the pair of events to be analyzed and then combining with the document to be analyzed, or by analyzing the compressed document to be analyzed and then combining with the pair of events to be analyzed. The present invention does not limit the way of obtaining the proponent's debate view corresponding to the proponent agent and the opponent's debate view corresponding to the opponent agent according to the document to be analyzed and the pair of events to be analyzed.

[0040] Among them, it can be clear that since the proponent's debate view is different from the opponent's debate view, by establishing the proponent's debate view and the opponent's debate view with different positions, the accuracy of the obtained target event relationship can be further improved.

[0041] Step 202: Based on the pro - debate view and the con - debate view, simulate the interaction process among different participants in the debate competition through the host agent, pro - side agent, con - side agent, and audience agent, and obtain the event relationships predicted by each of the multiple agents.

[0042] It should be noted that based on the pro - debate view and the con - debate view, simulating the interaction process among different participants in the debate competition through the host agent, pro - side agent, con - side agent, and audience agent can be carried out in accordance with the process of a real court debate or in accordance with the process of an ordinary debate competition, etc. The present invention does not limit the way of simulating the interaction process among different participants in the debate competition through the host agent, pro - side agent, con - side agent, and audience agent based on the pro - debate view and the con - debate view.

[0043] Among them, in the debate process, the pro - side agent and the con - side agent can, based on their respective debate views, carry out multiple rounds of interaction around the event relationship, and gradually improve and generate their respective answers through discussion. The debate process aims to enable different types of debate competition participants to give their own event relationship results respectively through the discussion between the pro - side agent and the con - side agent.

[0044] Step 203: Integrate the event relationships predicted by each of the multiple agents to obtain the target event relationship.

[0045] It should be noted that after obtaining the event relationships predicted by each of the multiple agents, the event relationship extraction results of the agents are integrated. The goal of this part is to fuse the prediction results of different types of agents to generate a more comprehensive and accurate final prediction result. After the interaction process ends, each type of agent, including the pro - side agent, con - side agent, host agent, and audience agent, outputs its own event relationship prediction result. Next, taking these four prediction results as a reference, use a large - model to generate the final event relationship.

[0046] It can be understood that for the problem of agent role diversity, the present invention classifies agents into four types: host agent, pro - side agent, con - side agent, and audience agent, and explores the understanding of event relationships by different participants and its impact on the final result. In addition, the present invention generates the pro - debate view corresponding to the pro - side agent and the con - debate view corresponding to the con - side agent, then simulates the interaction process among different participants in the debate competition through multiple agents to obtain the event relationships predicted by each of the multiple agents, and finally integrates the event relationships predicted by each of the multiple agents to obtain the target event relationship. That is, for the event relationship extraction task, a multi - agent debate method based on a large - model is constructed, which enhances the ability of the large - model in event relationship extraction and improves the accuracy of the event relationship extraction result.

[0047] Figure 3 It is a schematic flowchart of the method for obtaining debate viewpoints provided by the present invention. As Figure 3 shown, the obtaining of the positive debate viewpoints corresponding to the positive agent and the negative debate viewpoints corresponding to the negative agent according to the document to be analyzed and the pair of events to be analyzed may include: Step 301: Generate a summary for the pair of events to be analyzed, and use the summary to predict the event relationship to obtain the positive debate viewpoints; Step 302: Extract sentences related to the events to be analyzed from the document to be analyzed to obtain a sentence set, and use the sentence set to predict the event relationship to obtain the negative debate viewpoints.

[0048] It should be noted that a key challenge in event relationship extraction is long-distance dependence, that is, in a document, two events may be distributed in different sentences, and how to model the dependence relationship between events is a difficult problem. The present invention adopts the method of compressing the document to reduce the distance between events.

[0049] The present invention proposes two document compression strategies of generating first and then extracting, and selecting first and then extracting as the debate viewpoints of the positive and negative sides. Among them, generating first and then extracting means generating a summary for the pair of events to be analyzed and using the summary to predict the event relationship. Selecting first and then extracting means extracting a sentence set related to the pair of events to be analyzed from the document to be analyzed and using the sentence set to predict the event relationship.

[0050] It can be understood that during the agent debate process, the positive agent and the negative agent hold different viewpoints respectively. That is, for the multi-agent interaction problem, the present invention considers the adversarial and cooperative relationships between agents, thereby effectively improving the agents' understanding of the content of the document to be analyzed.

[0051] Figure 4 It is a schematic flowchart of the method for obtaining the event relationships predicted by multiple agents respectively provided by the present invention. As Figure 4 shown, the interaction process between different participants in the debate competition includes an opening stage, a statement stage, a rebuttal stage, a free discussion stage, and a summary stage. Based on the positive debate viewpoints and the negative debate viewpoints, through the host agent, the positive agent, the negative agent, and the audience agent to simulate the interaction process between different participants in the debate competition, the obtaining of the event relationships predicted by multiple agents respectively may include: Step 401: In the opening stage, through the host agent, introduce the debate rules, process, theme, the roles of the positive agent and the negative agent, the document to be analyzed, the pair of events to be analyzed, the positive debate viewpoints corresponding to the positive agent, and the negative debate viewpoints corresponding to the negative agent.

[0052] It should be noted that in the opening stage, the debate is initiated by the host agent, which introduces the debate rules, procedures, and the roles of both sides of the debate (the affirmative side and the negative side). The host agent is also responsible for clarifying the theme of this debate, that is, the type of event relationship to be predicted and the relevant background information of the document to be analyzed input.

[0053] Step 402: In the statement stage, the first debater agent of the affirmative side agent and the first debater agent of the negative side agent speak in turn, respectively stating their corresponding debate viewpoints and problem-solving ideas, where the problem-solving ideas are used to represent the strategies adopted for predicting the event relationship between the events to be analyzed.

[0054] It should be noted that in the statement stage, the first debater agent of the affirmative side and the first debater agent of the negative side speak in turn, respectively stating their viewpoints, the core evidence supporting their event relationship predictions, and problem-solving ideas. The first debater agent of the affirmative side puts forward viewpoints based on the "generate first and then extract" strategy, and the first debater agent of the negative side makes corresponding statements based on the "select first and then extract" strategy.

[0055] Step 403: In the rebuttal stage, the affirmative side agent and the negative side agent take turns to question each other's debate viewpoints and problem-solving ideas.

[0056] It should be noted that in the rebuttal stage, the agents of both sides take turns to question each other's viewpoints and evidence.

[0057] Furthermore, in the rebuttal stage, the affirmative side agent and the negative side agent taking turns to question each other's debate viewpoints and problem-solving ideas may include: the second debater agent of the affirmative side agent asks questions about the deficiencies in the debate viewpoints and problem-solving ideas put forward by the first debater agent of the negative side agent; the second debater agent of the negative side agent replies to the questions raised by the second debater agent of the affirmative side agent, and at the same time, the second debater agent of the negative side agent asks questions about the deficiencies in the debate viewpoints and problem-solving ideas of the second debater agent of the affirmative side agent; the third debater agent of the negative side agent replies to the questions raised by the second debater agent of the affirmative side agent, and at the same time, the third debater agent of the negative side agent asks questions about the deficiencies in the debate viewpoints and problem-solving ideas of the third debater agent of the affirmative side agent; the affirmative side agent and the negative side agent conduct Q&A in this way until all agents in the affirmative side agent and the negative side agent participate in the Q&A.

[0058] Exemplarily, the second debater agent of the affirmative side raises questions about the possible deficiencies of the negative side's strategy, and the second debater agent of the negative side replies to the questions of the second debater agent of the affirmative side. At the same time, the second debater agent of the negative side raises questions about the possible deficiencies of the affirmative side's view. The affirmative side and the negative side conduct Q&A in this way until all agents on both sides participate in the Q&A.

[0059] Step 404: In the free discussion stage, the affirmative agents and the negative agents respectively clarify their respective debate views.

[0060] It should be noted that in the free discussion stage, when entering the open discussion stage, the affirmative agents and the negative agents respectively clarify their views.

[0061] Furthermore, in the free discussion stage, the affirmative agents and the negative agents respectively clarifying their respective debate views may include: combining the previous interaction information, the affirmative agents regenerate the summary of the event pair to be analyzed, and reiterate their own debate views and problem-solving ideas. The negative agents re-extract the set of sentences related to the event pair to be analyzed in the document to be analyzed, and reiterate their own debate views and problem-solving ideas. The affirmative agents and the negative agents respectively refute the debate views of the other party, and the maximum number of rounds of refutation is four.

[0062] Exemplarily, combining the previous information, the affirmative agents regenerate the summary of the event pair and reiterate their own views and problem-solving ideas. Similarly, the negative agents re-extract the set of sentences useful for predicting the event relationship in the article and reiterate their own views and problem-solving ideas. The affirmative agents and the negative agents can respectively refute the views of the other party. The maximum number of rounds of discussion in this stage is 4.

[0063] Step 405: In the summary stage, the first debater agent of the affirmative agents and the first debater agent of the negative agents respectively predict the first event relationship and the second event relationship based on all the information of the interaction process. The audience agent predicts the third event relationship based on all the information of the interaction process, the first event relationship and the second event relationship. The host agent predicts the fourth event relationship based on the first event relationship, the second event relationship and the third event relationship.

[0064] It should be noted that in the summary stage, at the end of the debate, the first debater agent of the affirmative side and the first debater agent of the negative side respectively summarize their final views and prediction results. Subsequently, the audience agent, as an impartial party, comprehensively summarizes the debate process. It combines the arguments of the affirmative and negative sides, as well as the information it observes during the debate, to generate a comprehensive event relationship result. In addition, the host agent also participates in the summary stage. As the supervisor and rule maker of the entire debate process, the host agent comprehensively considers the performances of the affirmative and negative sides, as well as the analysis results of the audience agent in the summary.

[0065] The debate process includes multiple stages to simulate the interactions among different participants in a real debate competition, ensuring the accuracy and comprehensiveness of the event relationship result.

[0066] In some embodiments, before obtaining the document to be analyzed and the event pair to be analyzed, a relationship acquisition model can also be constructed.

[0067] Figure 5 It is a schematic flowchart of the method for constructing the relationship acquisition model provided by the present invention. That is, constructing the agents participating in the debate and setting the rules of the agent debate competition. This part aims to set the agents and rules in the agent debate competition, mainly including three sub-parts: agent role setting, agent debate competition rules, and debate speech preparation. As Figure 5 shown, before obtaining the document to be analyzed and the event pair to be analyzed, the method may further include: Step 501: Set multiple agents in the relationship acquisition model, where the multiple agents have different positions, and the multiple agents include an affirmative agent, a negative agent, a host agent, and an audience agent.

[0068] It should be noted that first, the agent role setting needs to be carried out. An agent debate competition requires agents with different positions to participate, mainly including three types of agents: debate agents, namely the affirmative agent and the negative agent, the host agent, and the audience agent.

[0069] Step 502: Set the rules of the debate competition process in the relationship acquisition model. The debate competition includes an opening stage, a debate stage, and a summary stage. The debate stage includes a speech stage, a rebuttal stage, and a free discussion stage; It should be noted that secondly, the rules of the intelligent agent debate competition need to be set. The intelligent agent debate competition mainly includes three steps: the opening stage, the debate stage, and the summary stage. The opening stage refers to the host intelligent agent introducing the rules of the intelligent agent debate competition, the debate topic, and the positions of different debate intelligent agents, namely the positive intelligent agent and the negative intelligent agent, and announcing the start of the debate. The debate stage refers to the debate intelligent agents, namely the positive intelligent agent and the negative intelligent agent, respectively elaborating on their own positions and viewpoints, and communicating and discussing with the debate intelligent agents of other positions, namely the positive intelligent agent and the negative intelligent agent, to deepen the understanding of the relationship between events. The summary stage refers to the debate intelligent agents, namely the positive intelligent agent and the negative intelligent agent, the host intelligent agent, and the audience intelligent agent respectively summarizing their own viewpoints and outputting their respective event relationships.

[0070] Step 503: Set the debate statement rules in the relationship acquisition model. The debate statement rules are used to instruct the positive intelligent agent and the negative intelligent agent to predict the event relationship, the debate methods to be adopted, and the debate steps to be adopted based on the document to be analyzed and the event to be analyzed before the debate competition.

[0071] It should be noted that finally, the preparation of the debate statements is carried out. Before the host intelligent agent announces the start of the intelligent agent debate competition, the debate intelligent agents of different positions, namely the positive intelligent agent and the negative intelligent agent, need to prepare their debate statements. Specifically, given an article and the events therein and , the debate intelligent agents of different positions, namely the positive intelligent agent and the negative intelligent agent, need to give their preliminary results, which mainly include three aspects: the event relationship result, the method adopted, and the specific step analysis.

[0072] Furthermore, in step 501, the objective of the present invention is to adopt the method of the intelligent agent debate competition, and different types of intelligent agents deepen their understanding of the event relationship through debate. The intelligent agents participating in the intelligent agent debate competition are mainly divided into three types: the debate intelligent agents, namely the positive intelligent agent and the negative intelligent agent, the host intelligent agent, and the audience intelligent agent.

[0073] (1) The host intelligent agent. This intelligent agent is responsible for explaining and implementing the competition rules, ensuring the fairness of the competition and the smooth progress of the debate process. In addition, the host intelligent agent also needs to guide the discussion and act as a referee when necessary. At the end of the intelligent agent debate competition, the host intelligent agent needs to give its own understanding of the problem.

[0074] (2) Debate agents. The debate agents consist of the pro-side agents and the con-side agents, and are the main participants in the debate. By presenting their own views and discussing with the other side, they promote a deeper understanding of the debate topic among all participants. The pro-side includes three debate agents: the first pro debater, the second pro debater, and the third pro debater. The con-side includes three debate agents: the first con debater, the second con debater, and the third con debater. Among them, the responsibility of the first pro (con) debater is to present the views of their own side at the beginning of the debate and, at the summary stage, give the final result of the event relationship of their own side based on the discussion. The responsibilities of the second and third pro (con) debaters are to answer the questions raised by the other side in the debate result and to raise questions against the views of the other side.

[0075] (3) Audience agents. As observers of the debate, these agents deepen their understanding of the topic by listening to the debate content. From the perspective of bystanders, audience agents can evaluate and understand the debate topic more objectively.

[0076] In some embodiments, the present invention also constructs a memory mechanism for the agents. By introducing a dynamic memory unit for the agents, this mechanism enables them to record key information, track the progress of the debate during the debate process, and dynamically adjust their statements and positions according to new inputs. The memory mechanism mainly includes the following three core modules: (1) Information storage module. This module is used to store all the information received by the agents in the debate, including the statements, questions of the opponents, and their own speech records.

[0077] (2) Information retrieval and update module. The agents can retrieve the stored information in real time to support their quick response to the views of the opponents or to put forward new arguments. At the same time, when new important information appears in the debate, this module can dynamically update the stored content to ensure the integrity and timeliness of the memory content.

[0078] (3) Memory optimization module. To avoid memory overload, this module uses a large model to streamline the memory information and optimize the memory capacity of the agents. It preferentially retains the information crucial to the current debate topic and eliminates irrelevant or low-correlation information.

[0079] It can be understood that the memory mechanism of the agents can support the agents to efficiently store, manage, and utilize information during the debate process to enhance their reasoning and decision-making abilities.

[0080] Next, the exemplary application of the embodiments of the present invention in a practical application scenario will be described.

[0081] Figure 6It is a schematic diagram of the overall architecture of the event relationship extraction system provided by the present invention. The main technical problems to be solved by the event relationship extraction method provided by the present invention are: how to effectively integrate the large model multi-agent method with the event relationship extraction task, and how to design an effective multi-agent interaction method. The event relationship extraction system based on the large model multi-agent proposed by the present invention is mainly divided into four parts: the first part is to construct the agents participating in the debate and set the rules of the agent debate competition; the second part is to construct the memory mechanism of the agents; the third part is for the agents to extract the relationship between events through mutual interaction; the fourth part is to integrate the event relationship extraction results of the agents. As Figure 6 shown, the event relationship extraction system includes the following functional units: The article is the document to be analyzed obtained. The preprocessing unit. Each agent is an instance of a large model. The event relationship extraction method based on the large model multi-agent obtains the relationship between events through the way of agent debate competition. The agent debate competition includes three types of agents: the host agent, the debate agents, namely the positive agent and the negative agent, and the audience agent. The positive agent and the negative agent each hold a debate view and need to generate corresponding problem-solving ideas according to the view.

[0082] The memory unit. After the agent debate competition starts, the content output by each of the host agent, the debate agents, namely the positive agent and the negative agent, and the audience agent in each round will be stored in the information storage module. When starting a new round of dialogue, the agent will obtain historical information from the information storage module to enhance the effectiveness and accuracy of the debate. To avoid memory overload, the memory optimization module uses the large model to streamline the memory information and optimize the memory capacity of the agent. Priority is given to retaining the information crucial to the current debate topic and eliminating irrelevant or low-correlation information.

[0083] The debate unit. Different types of agents obtain the relationship between events through the debate competition. During the entire agent debate competition process, the host agent introduces the rules of the agent debate competition and controls the process of the entire agent debate competition. The positive agent and the negative agent respectively elaborate on their own views and problem-solving ideas, raise questions about the views and problem-solving ideas of the other party, and answer the questions raised by the other party. The audience agent listens to the entire debate process. At the end of the agent debate competition, the host agent, the debate agents, namely the positive agent and the negative agent, and the audience agent respectively give their own answers.

[0084] The prediction integration unit. Using the large model to generate the final event relationship result from the four prediction results.

[0085] Based on the above architecture, the present invention provides an event relationship extraction method based on the large model multi-agent, and the specific implementation steps are as follows: (1)Obtain the article and event pairs ; Obtain the viewpoints of the debate agents, namely the pro-agent and the con-agent: generate first and then extract, and select first and then extract.

[0086] (2)Based on the view of generating first and then extracting, the pro-agent generates the corresponding problem-solving ideas. Specifically, the pro-agent generates an abstract of the event pair and uses the generated abstract to predict the relationship between events. Based on the view of selecting first and then extracting, the con-agent generates the corresponding problem-solving ideas. Specifically, the con-agent extracts a set of sentences useful for predicting event relationships from the article and uses the screened set of sentences to predict the relationship between events.

[0087] (3)In the opening stage, the host agent reads out the rules, procedures of the agent debate and the roles of both sides of the debate (the pro side and the con side). The host agent reads out the article , event pairs and the debate viewpoints of the pro-agent and the con-agent.

[0088] (4)Entering the statement stage, the first pro-debate agent and the first con-debate agent speak in turn, respectively stating their viewpoints and corresponding problem-solving ideas. The first pro-debate agent puts forward problem-solving ideas based on the view of "generate first and then extract", and the first con-debate agent makes corresponding statements based on the view of "select first and then extract".

[0089] (5)Entering the rebuttal stage, the second pro-debate agent raises questions about the possible deficiencies in the con-agent's viewpoints and problem-solving ideas. The second con-debate agent replies to the questions of the second pro-debate agent, and at the same time, the second con-debate agent raises questions about the possible deficiencies in the pro-agent's viewpoints. The pro side and the con side conduct Q&A in this way until all agents on both sides have participated in the Q&A.

[0090] (6)Entering the free discussion stage, the pro-agent and the con-agent clarify their own viewpoints respectively. Combining the previous information, the pro-agent regenerates an abstract of the event pair and reiterates its own viewpoints and problem-solving ideas. Similarly, the con-agent re-extracts a set of sentences useful for predicting event relationships from the article and reiterates its own viewpoints and problem-solving ideas. The pro-agent and the con-agent can respectively refute the viewpoints of the other side. This stage has at most 4 rounds of discussions.

[0091] (7)Entering the summary stage, the first pro-debate agent gives the event relationship of the event pair based on the information output by itself in the statement stage, rebuttal stage and free discussion stage. The first con-debate agent gives the event relationship of the event pair The event relationship. The audience agent predicts the event pair based on the information output by the proponent and opponent agents during the statement stage, rebuttal stage, and free discussion stage. The possible event relationship information. The host agent combines the information output by all agents during the opening stage, statement stage, rebuttal stage, and free discussion stage to predict the event pair The possible event relationship information.

[0092] (8) In the previous step, the first proponent agent, the first opponent agent, the audience agent, and the host agent respectively gave the event pair The possible event relationship results. Combining these four prediction results, the final event relationship result is generated using a large model.

[0093] The event relationship extraction method provided by the present invention was experimentally tested on two public datasets, MAVEN-ERE and EventStoryLine. The experimental results show that compared with methods such as chain of thought and self-reflection, the method based on multi-agent interaction of large models can significantly improve the performance of event relationship extraction and enhance the accuracy of the event relationship extraction results. Further analysis reveals that during the debate process, the performance of different types of agents has improved compared to before the debate competition.

[0094] Based on the foregoing embodiments, an embodiment of the present invention provides an event relationship extraction device. Each module included in the device, as well as each unit included in each module, can be implemented by a processor; of course, it can also be implemented by specific logic circuits; during implementation, the processor can be a central processing unit (CPU), a microprocessor (MPU), a digital signal processor (DSP), or a field programmable gate array (FPGA), etc.

[0095] The event relationship extraction device provided by the present invention is described below. The event relationship extraction device described below can be mutually referred to the event relationship extraction method described above.

[0096] Figure 7 It is a schematic structural diagram of the event relationship extraction device provided by the present invention. As Figure 7 shown, the device 700 includes an information acquisition module 701 and a relationship extraction module 702, where: The information acquisition module 701 is used to acquire the document to be analyzed and the event pair to be analyzed, and the event pair to be analyzed is a pair of events in the document to be analyzed; A relation extraction module 702 is configured to input the document to be analyzed and the event pair to be analyzed into a relation acquisition model to obtain a target event relation, where the target event relation is used to characterize the event relation between the event pairs to be analyzed. The target event relation is obtained by separately predicting event relations by multiple agents in the relation acquisition model and integrating them. The multiple agents separately predict event relations by simulating the interaction in a debate competition. Each agent in the multiple agents is used to represent an instance of a large model.

[0097] In some embodiments, the multiple agents include a host agent, a proponent agent, an opponent agent, and an audience agent. The relation extraction module 702 includes a view acquisition unit, a view debate unit, and an information integration unit. Among them, The view acquisition unit is configured to obtain a proponent debate view corresponding to the proponent agent and an opponent debate view corresponding to the opponent agent according to the document to be analyzed and the event pair to be analyzed. The proponent debate view and the opponent debate view are different. The view debate unit is configured to, based on the proponent debate view and the opponent debate view, simulate the interaction process among different participants in a debate competition through the host agent, the proponent agent, the opponent agent, and the audience agent to obtain the event relations predicted by the multiple agents respectively. The information integration unit is configured to integrate the event relations predicted by the multiple agents respectively to obtain the target event relation.

[0098] In some embodiments, the view acquisition unit is specifically configured to: generate a summary of the event pair to be analyzed and use the summary to predict the event relation to obtain the proponent debate view; extract sentences related to the event to be analyzed from the document to be analyzed to obtain a sentence set, and use the sentence set to predict the event relation to obtain the opponent debate view.

[0099] In some embodiments, the interaction process among different participants in the debate competition includes an opening stage, an argument stage, a rebuttal stage, a free discussion stage, and a summary stage. The view debate unit includes an opening component, an argument component, a rebuttal component, a discussion component, and a summary component. Among them, The opening component is configured to, in the opening stage, introduce the debate competition rules, process, theme, roles of the proponent agent and the opponent agent, the document to be analyzed, the event pair to be analyzed, the proponent debate view corresponding to the proponent agent, and the opponent debate view corresponding to the opponent agent through the host agent. The statement component is used for the first debater agent of the proponent agent and the first debater agent of the opponent agent to speak in turn during the statement stage, respectively stating their corresponding debate viewpoints and problem-solving ideas, where the problem-solving ideas are used to represent the strategies adopted for predicting the event relationships between the to-be-analyzed event pairs; The refutation component is used for the proponent agent and the opponent agent to take turns to question each other's debate viewpoints and problem-solving ideas during the refutation stage; The discussion component is used for the proponent agent and the opponent agent to clarify their respective debate viewpoints during the free discussion stage; The summary component is used for the first debater agent of the proponent agent and the first debater agent of the opponent agent to respectively predict the first event relationship and the second event relationship based on all the information of the interaction process. The audience agent predicts the third event relationship based on all the information of the interaction process, the first event relationship, and the second event relationship. The host agent predicts the fourth event relationship based on the first event relationship, the second event relationship, and the third event relationship.

[0100] In some embodiments, the refutation component is specifically used for: the second debater agent of the proponent agent to raise questions about the deficiencies in the debate viewpoints and problem-solving ideas put forward by the first debater agent of the opponent agent; The second debater agent of the opponent agent replies to the questions raised by the second debater agent of the proponent agent, and at the same time, the second debater agent of the opponent agent raises questions about the deficiencies in the debate viewpoints and problem-solving ideas of the second debater agent of the proponent agent; The third debater agent of the opponent agent replies to the questions raised by the second debater agent of the proponent agent, and at the same time, the third debater agent of the opponent agent raises questions about the deficiencies in the debate viewpoints and problem-solving ideas of the third debater agent of the proponent agent; The proponent agent and the opponent agent conduct Q&A in this way until all the agents in the proponent agent and the opponent agent participate in the Q&A.

[0101] In some embodiments, the discussion component is specifically used for: combining the previous interaction information, the proponent agent regenerates the summary of the to-be-analyzed event pairs, and reiterates its own debate viewpoints and problem-solving ideas. The opponent agent re-extracts the set of sentences related to the to-be-analyzed event pairs in the to-be-analyzed document, and reiterates its own debate viewpoints and problem-solving ideas; The proponent agent and the opponent agent respectively refute each other's debate viewpoints, and the maximum number of rounds of refutation is four.

[0102] In some embodiments, the device further includes a model setting module, which is specifically configured to: set multiple agents in the relationship acquisition model, where different stances are set for the multiple agents, and the multiple agents include a proponent agent, an opponent agent, a moderator agent, and an audience agent; Set the rules of the debate process in the relationship acquisition model, where the debate includes an opening stage, a debate stage, and a summary stage, and the debate stage includes a statement stage, a rebuttal stage, and a free discussion stage; Set the debate statement rules in the relationship acquisition model, where the debate statement rules are used to instruct the proponent agent and the opponent agent to determine the predicted event relationship, the debate methods to be adopted, and the debate steps to be taken based on the document to be analyzed and the event to be analyzed before the debate.

[0103] In the embodiments of the present invention, the performance of the large model in the event relationship extraction task can be effectively improved, and the accuracy of the event relationship extraction result can be guaranteed.

[0104] Figure 8 It is a schematic physical structure diagram of the electronic device provided by the present invention. As Figure 8 shown, the electronic device may include: a processor 810, a communication interface 820, a memory 830, and a communication bus 840. Among them, the processor 810, the communication interface 820, and the memory 830 communicate with each other through the communication bus 840. The processor 810 can call the logical instructions in the memory 830 to execute the event relationship extraction method, which includes: obtaining a document to be analyzed and an event pair to be analyzed, where the event pair to be analyzed is a pair of events in the document to be analyzed; inputting the document to be analyzed and the event pair to be analyzed into the relationship acquisition model to obtain a target event relationship, where the target event relationship is used to represent the event relationship between the event pair to be analyzed. Among them, the target event relationship is obtained by separately predicting the event relationships by multiple agents in the relationship acquisition model and integrating them. The multiple agents separately predict the event relationships by simulating the interaction of a debate, and each agent in the multiple agents is used to represent an instance of a large model.

[0105] In addition, when the logical instructions in the above-mentioned memory 830 are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs.

[0106] On the other hand, the present invention also provides a computer program product. The computer program product includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the event relationship extraction method provided by the above-mentioned various methods. The method includes: obtaining a document to be analyzed and an event pair to be analyzed, where the event pair to be analyzed is a pair of events in the document to be analyzed; inputting the document to be analyzed and the event pair to be analyzed into a relationship acquisition model to obtain a target event relationship, where the target event relationship is used to characterize the event relationship between the event pair to be analyzed. Among them, the target event relationship is obtained by separately predicting event relationships by multiple agents in the relationship acquisition model and integrating them. The multiple agents separately predict event relationships by simulating debate competition interactions among the multiple agents, and each agent among the multiple agents is used to represent an instance of a large model.

[0107] The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the processes or functions described in the embodiments of the present invention are generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions may be transmitted from one website, computer, server, or data center to another website, computer, server, or data center by wire (such as coaxial cable, optical fiber, digital subscriber line (DSL)) or wirelessly (such as infrared, wireless, microwave, etc.). The computer-readable storage medium may be any available medium that can be stored by a computer or a data storage device such as a server or a data center that includes one or more integrated available media. The available medium may be a magnetic medium (such as a floppy disk, a hard disk, a magnetic tape), an optical medium (such as a DVD), or a semiconductor medium (such as a solid state disk (SSD)), etc.

[0108] In another aspect, the present invention also provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements an event relationship extraction method provided by the above-mentioned various methods. The method includes: obtaining a document to be analyzed and an event pair to be analyzed, where the event pair to be analyzed is a pair of events in the document to be analyzed; inputting the document to be analyzed and the event pair to be analyzed into a relationship acquisition model to obtain a target event relationship, where the target event relationship is used to characterize the event relationship between the event pair to be analyzed. Among them, the target event relationship is obtained by separately predicting event relationships by multiple agents in the relationship acquisition model and integrating them. The multiple agents separately predict event relationships by simulating debate competition interactions among the multiple agents, and each agent in the multiple agents is used to represent an instance of a large model.

[0109] The above computer-readable storage medium may adopt any combination of one or more computer-readable media. The computer-readable media may be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the above. More specific examples (non-exhaustive list) of the computer-readable storage medium include: an electrical connection having one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM), or a flash memory, an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this document, the computer-readable storage medium may be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0110] The computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, which carries the computer-readable program code. Such a propagated data signal may take various forms, including - but not limited to - an electromagnetic signal, an optical signal, or any suitable combination of the above. The computer-readable signal medium may also be any computer-readable medium other than the computer-readable storage medium, which can send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0111] The program code contained on the computer-readable medium may be transmitted by any appropriate medium, including - but not limited to - wireless, wire, optical fiber, radio frequency (RF), etc., or any suitable combination of the above.

[0112] Computer program code for performing the operations of this specification may be written in one or more programming languages or combinations thereof. The programming languages include object-oriented programming languages such as Java, Smalltalk, C++, and also include conventional procedural programming languages such as the "C" language or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, executed as a stand-alone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., through the Internet using an Internet service provider).

[0113] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated. The components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art can understand and implement it without creative effort.

[0114] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. The computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0115] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments or perform equivalent replacements for some of the technical features. These modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for extracting event relations, characterized in that: include: Obtaining a document to be analyzed and an event pair to be analyzed, wherein the event pair to be analyzed is a pair of events in the document to be analyzed; The document to be analyzed and the event pair to be analyzed are input into the relationship acquisition model to obtain a target event relationship, and the target event relationship is used to characterize the event relationship between the event pair to be analyzed, wherein the target event relationship is obtained by respectively predicting and integrating event relationships by multiple agents in the relationship acquisition model, and the event relationships predicted by the multiple agents are obtained by interacting in a simulated debate competition among the multiple agents, and each of the multiple agents is used to characterize an instance of a large model.

2. The event relationship extraction method according to claim 1, characterized in that: The plurality of agents include a host agent, a positive agent, a negative agent, and an audience agent. The step of inputting the document to be analyzed and the event to be analyzed into a relationship acquisition model to obtain a target event relationship includes: According to the pair of the document to be analyzed and the event to be analyzed, obtaining the affirmative debate viewpoint corresponding to the affirmative agent and the negative debate viewpoint corresponding to the negative agent, wherein the affirmative debate viewpoint is different from the negative debate viewpoint; Based on the affirmative debate viewpoints and the negative debate viewpoints, the event relationships predicted by the multiple agents are obtained through the interaction process between different participants in the simulated debate by the host agent, the affirmative agent, the negative agent and the audience agent; The event relations predicted by the multiple agents are integrated to obtain the target event relation.

3. The event relationship extraction method according to claim 2, characterized in that: The obtaining, according to the pair of the document to be analyzed and the event to be analyzed, the affirmative debate viewpoint corresponding to the affirmative intelligent agent and the negative debate viewpoint corresponding to the negative intelligent agent, comprises: Generate a summary for the event pair to be analyzed, and use the summary to predict event relationships to obtain the debate viewpoint of the affirmative side; Sentences related to the event to be analyzed are extracted from the document to be analyzed to obtain a sentence set, and the sentence set is used to predict event relations to obtain the opposing debate point of view.

4. The event relationship extraction method according to claim 2, characterized in that: The interaction process between different participants in the debate competition includes an opening stage, a statement stage, a rebuttal stage, a free discussion stage and a summary stage. The interaction process between different participants in the debate competition is simulated by the host agent, the affirmative agent, the negative agent and the audience agent based on the affirmative debate viewpoint and the negative debate viewpoint, and the event relationship predicted by each of the multiple agents is obtained, including: In the opening stage, the host agent introduces the debate rules, process, theme, roles of the affirmative agent and the negative agent, documents to be analyzed, event pairs to be analyzed, the affirmative debate viewpoints corresponding to the affirmative agent, and the negative debate viewpoints corresponding to the negative agent; In the speech stage, the first debater of the affirmative intelligent agent and the first debater of the negative intelligent agent speak in turn, respectively stating their respective debate viewpoints and problem-solving ideas, wherein the problem-solving ideas are used to represent the strategy adopted to predict the event relationship between the event pairs to be analyzed; In the rebuttal stage, the affirmative agent and the negative agent take turns to question the other party's debate viewpoints and problem-solving ideas; In the free discussion phase, the affirmative agent and the negative agent clarify their respective debate viewpoints; In the summary stage, the first debate agent of the affirmative agent and the first debate agent of the negative agent respectively predict the first event relationship and the second event relationship based on all the information of the interaction process, the audience agent predicts the third event relationship based on all the information of the interaction process, the first event relationship and the second event relationship, and the host agent predicts the fourth event relationship based on the first event relationship, the second event relationship and the third event relationship.

5. The event relationship extraction method according to claim 4, characterized in that: In the rebuttal stage, the affirmative agent and the negative agent take turns to question the other party's debate viewpoints and problem-solving ideas, including: The second debater of the affirmative intelligent agent raises questions about the shortcomings of the debate viewpoints and problem-solving ideas put forward by the first debater of the negative intelligent agent; The second debater of the opposing intelligent body responds to the questions raised by the second debater of the affirmative intelligent body, and at the same time, the second debater of the opposing intelligent body raises questions about the shortcomings of the debate viewpoints and problem-solving ideas of the second debater of the affirmative intelligent body; The third debater of the opposing intelligent body responds to the questions raised by the second debater of the affirmative intelligent body, and at the same time, the third debater of the opposing intelligent body raises questions about the shortcomings of the debate viewpoints and problem-solving ideas of the third debater of the affirmative intelligent body; The affirmative agent and the negative agent conduct question and answer in this manner until all agents in the affirmative agent and the negative agent participate in the question and answer.

6. The event relationship extraction method according to claim 4, characterized in that: During the free discussion phase, the affirmative agent and the negative agent clarify their respective debate viewpoints, including: Combined with the previous interactive information, the affirmative agent regenerates the summary of the event pair to be analyzed, and reiterates its own debate viewpoint and problem-solving ideas; the negative agent re-extracts the sentence set related to the event pair to be analyzed in the document to be analyzed, and reiterates its own debate viewpoint and problem-solving ideas; The affirmative agent and the negative agent refute the other party's debate viewpoints respectively, and the number of rounds of refutation is at most four rounds.

7. The event relationship extraction method according to claim 1, characterized in that: Before obtaining the pair of the document to be analyzed and the event to be analyzed, the method further includes: Setting a plurality of agents in the relationship acquisition model, wherein the plurality of agents are set with different positions, and the plurality of agents include a positive agent, a negative agent, a host agent, and an audience agent; Setting rules for the debate process in the relationship acquisition model, wherein the debate includes an opening phase, a debate phase, and a summary phase, and the debate phase includes a statement phase, a rebuttal phase, and a free discussion phase; The debate speech rules in the relationship acquisition model are set, and the debate speech rules are used to instruct the affirmative agent and the negative agent to predict event relationships, adopt debate methods and adopt debate steps based on the documents to be analyzed and the events to be analyzed before the debate.

8. An event relationship extraction device, characterized in that: include: An information acquisition module, used to acquire a document to be analyzed and an event pair to be analyzed, wherein the event pair to be analyzed is a pair of events in the document to be analyzed; A relationship extraction module is used to input the document to be analyzed and the event pair to be analyzed into a relationship acquisition model to obtain a target event relationship, wherein the target event relationship is used to characterize the event relationship between the event pair to be analyzed, wherein the target event relationship is obtained by respectively predicting and integrating event relationships by multiple agents in the relationship acquisition model, and the event relationships predicted by the multiple agents are obtained by interacting in a simulated debate competition among the multiple agents, and each of the multiple agents is used to characterize an instance of a large model.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the computer program, the event relationship extraction method as described in any one of claims 1 to 7 is implemented.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the event relationship extraction method as described in any one of claims 1 to 7 is implemented.

Images