Information verification system and processing method and device based on multi-agent cooperation

Abstract

The invention discloses an information verification system and a processing method and device based on multi-agent cooperation, and belongs to the field of information processing. The information verification system comprises a management agent and a plurality of execution agents. The management agent is used for acquiring to-be-verified information; based on the to-be-verified information, scheduling the plurality of execution agents to execute information verification; the plurality of execution agents are used for generating a verification report based on the to-be-verified information; the verification report comprises at least two of the following information: a verification result of the to-be-verified information; statement information, wherein the statement information is a proposition to be verified extracted from the information to be verified; an evidence chain of the statement information, wherein the evidence chain comprises at least one piece of evidence information; a verification result of the statement information; and summarizing information, wherein the summarizing information is summarized for the verification process of the to-be-verified information. And on the basis of multi-agent information verification, reproduction of complete verification logic can be realized.

Description

Technical Field

[0001] This application relates to the field of information processing, and in particular to an information verification system, a processing method and apparatus based on multi-agent cooperation. Background Technology

[0002] With the rapid development of the internet, it has become an important channel for people to publish information and obtain news. Online information is characterized by its fast dissemination speed and wide coverage. However, precisely because users of all identities and stances can express their opinions online, it easily leads to the generation and spread of false information, which can have varying degrees of impact on the public and even society. False information refers to untrue information.

[0003] In related technologies, in order to detect fake information, it is often done by training a corresponding fake information detection model. The fake information detection model is fine-tuned on a dataset of labeled real and fake information to build a classifier.

[0004] However, this method can only determine whether a piece of information to be verified is false. Summary of the Invention

[0005] This application provides an information verification system, a processing method and apparatus based on multi-agent cooperation, and the technical solution is as follows: According to one aspect of this application, an information verification system is provided, the information verification system comprising a management agent and multiple execution agents; The management agent is used to acquire information to be verified; and based on the information to be verified, to schedule the multiple execution agents to perform information verification. The plurality of execution agents are used to generate a verification report based on the information to be verified; the verification report includes at least two of the following information: The verification result of the information to be verified; Statement information, wherein the statement information is a proposition to be verified extracted from the information to be verified; The chain of evidence for the stated information, the chain of evidence including at least one piece of evidence information; The verification result of the stated information; Summary information, which is a summary of the verification process for the information to be verified.

[0006] According to one aspect of this application, an information verification method based on multi-agent cooperation is provided, the method being applied to a system comprising: a management agent and multiple execution agents; The management agent acquires the information to be verified. The management agent, based on the information to be verified, schedules the multiple execution agents to perform information verification; The plurality of execution agents generate a verification report based on the information to be verified; the verification report includes at least two of the following information: The verification result of the information to be verified; Statement information, wherein the statement information is a proposition to be verified extracted from the information to be verified; The chain of evidence for the stated information, the chain of evidence including at least one piece of evidence information; The verification result of the stated information; Summary information, which is a summary of the verification process for the information to be verified.

[0007] According to one aspect of this application, a multi-agent collaborative processing method is provided, characterized in that the method is applied to a system comprising: a management agent, multiple execution agents, a global shared memory, and multiple local team memories corresponding to the multiple execution agents; the method includes: In response to the acquired task to be executed, the management agent initializes the global shared memory; The management agent monitors the task status in the global shared memory and dynamically schedules the target agent from the multiple execution agents based on changes in the task status; The target agent responds to the scheduling by reading task data from the global shared memory, performing processing using the corresponding local team memory, and writing the processing result and the updated task status back to the global shared memory. The multiple execution agents complete the task to be executed through collaborative reading and writing of the globally shared memory.

[0008] According to one aspect of this application, an information verification device is provided, the device being applied to a system comprising: a management agent and multiple execution agents; The management agent module is used to acquire information to be verified; and based on the information to be verified, to schedule the multiple execution agents to perform information verification. The plurality of execution intelligent agent modules are used to generate a verification report based on the information to be verified; the verification report includes at least two of the following information: The verification result of the information to be verified; Statement information, wherein the statement information is a proposition to be verified extracted from the information to be verified; The chain of evidence for the stated information, the chain of evidence including at least one piece of evidence information; The verification result of the stated information; Summary information, which is a summary of the verification process for the information to be verified.

[0009] According to one aspect of this application, a processing apparatus based on multi-agent collaboration is provided, characterized in that the apparatus is applied to a system comprising: a management agent, multiple execution agents, a global shared memory, and multiple local team memories corresponding to the multiple execution agents; the apparatus includes: The management agent module is used to initialize the global shared memory in response to the acquired task to be executed; The management agent module is used to monitor the task status in the global shared memory and dynamically schedule the target agent from the multiple execution agents based on the changes in the task status. The target intelligent agent is used to respond to the scheduling, read task data from the global shared memory, perform processing using the corresponding local team memory, and write back the processing result and the updated task status to the global shared memory. The multiple execution agents complete the task to be executed through collaborative reading and writing of the globally shared memory.

[0010] According to one aspect of this application, a computer device is provided, the computer device including a processor and a memory, the memory storing a computer program, the computer program being loaded and executed by the processor to implement an information verification method based on multi-agent cooperation, and / or a processing method based on multi-agent cooperation.

[0011] According to one aspect of this application, a computer-readable storage medium is provided, wherein a computer program is stored therein, the computer program being loaded and executed by a processor to implement an information verification method based on multi-agent cooperation, and / or a processing method based on multi-agent cooperation.

[0012] According to one aspect of this application, a computer program product is provided, the computer program product comprising a computer program stored in a computer-readable storage medium, wherein a processor reads from the computer-readable storage medium and executes the computer program to implement an information verification method based on multi-agent cooperation, and / or a processing method based on multi-agent cooperation.

[0013] The beneficial effects of the technical solution provided in this application include at least the following: By utilizing multiple agents to complete the information verification process, the system fully leverages the capabilities of these agents to achieve efficient information verification. Furthermore, after multi-agent collaboration for information verification, a complete verification report is generated. This report not only provides the traditional verification result (e.g., true or false information) but also offers the stated information, the evidence chain for that statement, the verification result, and a summary. Compared to the completely black-box approach of traditional methods (where the user cannot know how the verification result was obtained), this results in a more reliable, interpretable, and traceable output. Moreover, the information verification system includes a management agent that schedules multiple agents, eliminating the need for users to manually schedule different agents to perform information verification tasks, thus improving the interactive efficiency of verifying information. Attached Figure Description

[0014] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0015] Figure 1 An architectural diagram of a computer system provided in an exemplary embodiment of this application is shown; Figure 2 This invention provides a schematic diagram of the structure of an information verification system according to an exemplary embodiment of the present application. Figure 3 A flowchart of an exemplary embodiment of the present application for information verification based on multi-agent collaboration is shown. Figure 4 A flowchart of an exemplary embodiment of the present application for information verification based on multi-agent collaboration is shown. Figure 5 This invention provides a schematic diagram of the structure of an information verification system according to an exemplary embodiment of the present application. Figure 6 This invention provides a schematic diagram of the structure of an information verification system according to an exemplary embodiment of the present application. Figure 7 This invention provides a schematic diagram of the structure of an information verification system according to an exemplary embodiment of the present application. Figure 8 This invention provides a structural block diagram of an information verification device based on multi-agent collaboration, according to an exemplary embodiment of this application. Figure 9This invention provides a structural block diagram of an information verification device based on multi-agent collaboration, according to an exemplary embodiment of this application. Figure 10 This application shows a structural block diagram of a server provided in an exemplary embodiment; Figure 11 A structural block diagram of a terminal provided in an exemplary embodiment of this application is shown. Detailed Implementation

[0016] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described in further detail below with reference to the accompanying drawings.

[0017] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0018] The terminology used in this disclosure is for the purpose of describing particular embodiments only and is not intended to be limiting of the disclosure. The singular forms “a,” “the,” and “the” as used in this disclosure and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any and all possible combinations of one or more of the associated listed items.

[0019] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions. For example, the settings and operation information involved in this application were obtained with full authorization.

[0020] It should be understood that although the terms first, second, etc., may be used in this disclosure to describe various information, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, a first parameter may also be referred to as a second parameter without departing from the scope of this disclosure, and similarly, a second parameter may also be referred to as a first parameter. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to determination."

[0021] First, let me introduce the relevant terms used in this application: An intelligent agent (also known as an intelligent entity) is an intelligent entity capable of perceiving its environment and taking actions to achieve specific goals. Generally, intelligent agents are driven by large language models (LLMs) and possess planning, memory, and scheduling capabilities. Planning capability refers to the agent's ability to break down a complex, large goal into a series of executable smaller tasks. Memory capability refers to the agent's ability to remember historical interaction information and learned experiences for continuous optimization of subsequent decisions. Scheduling capability refers to the agent's ability to invoke external tools (such as search engines, calculators, code executors, etc.) to complete specific tasks, compensating for its own limitations.

[0022] Multi-agent framework: This refers to a framework that allows multiple independent agents to work collaboratively to complete larger or more complex tasks that are difficult for a single agent to solve. Within this framework, different agents may play different roles and possess different expertise. The framework is responsible for task allocation and coordinating communication and cooperation among the agents. This can significantly improve the efficiency and quality of completing complex tasks, simulating the working model of a human team.

[0023] Tools: refers to external functions or services that intelligent agents can call, such as API (Application Programming Interface) calls, database queries, file operations, etc.

[0024] Figure 1 A schematic diagram of the architecture of a computer system provided in an exemplary embodiment of this application is shown. The computer system includes: a terminal 110, a management agent 120, and multiple execution agents 130.

[0025] Terminal 110 includes, but is not limited to, mobile phones, tablets, smart voice interaction devices, game consoles, wearable devices, multimedia playback devices, PCs (Personal Computers), in-vehicle terminals, smart home appliances, and other electronic devices. Optionally, a client 111 of the target application can be installed and run on the terminal. For example, the target application is an application that supports interaction with an intelligent agent, such as supporting dialogue with an intelligent agent and instructing the intelligent agent to perform specified tasks (such as information verification, false information detection, report generation, etc.). Alternatively, the target application can also be an application containing intelligent agent functionality, such as reading applications, social applications, consulting applications, various types of browsers, enterprise applications, conferencing applications, instant messaging applications, virtual reality (VR) applications, augmented reality (AR) applications, extended reality (XR) applications, etc., which are not limited in this application embodiment. The form of the target application is not limited in this application embodiment, including but not limited to clients, applications, and mini-programs installed on the terminal, and can also be in web page form.

[0026] Optionally, the user interacts with the management agent 120 and / or multiple execution agents 130 through the client 111, which can be considered as the interaction entry point between the management agent 120 and / or multiple execution agents 130.

[0027] In some embodiments, the management agent 120 is deployed on a server. Optionally, the server can be a standalone physical server, a server cluster or distributed system consisting of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms, but is not limited to these. Optionally, the server can be a server providing background services for the target application in terminal 110.

[0028] In some embodiments, each agent in the plurality of executing agents 130 is deployed on a server. Optionally, the server can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms, but is not limited to these. Optionally, the server can be a server that provides background services for the target application in terminal 110.

[0029] Optionally, the multiple executing agents 130 are n agents, including agent 1, agent 2, ... agent n, where n is a positive integer.

[0030] In some embodiments, the management agent 120 and the various agents in the multiple execution agents 130 are deployed on the same server; or, the management agent 120 and the various agents in the multiple execution agents 130 are deployed on different servers; or, the management agent 120 is deployed on a separate server, and some agents in the multiple execution agents 130 are deployed on the same server, while others are deployed on other servers, such as agents 1 to 3 (not shown in the figure) being deployed on the same server, agents 4 (not shown in the figure) to 6 (not shown in the figure) being deployed on the same server, agents 7 (not shown in the figure) to 9 (not shown in the figure) being deployed on the same server, and so on; or, the management agent 120 and the various execution agents 130 are divided based on the function of the agents, with agents with the same function deployed on the same server, and agents with different functions deployed on different servers, etc. This application does not limit the correspondence between agents and servers.

[0031] It should be noted that the hardware devices corresponding to the intelligent agents are not limited in the embodiments of this application. That is, the management intelligent agent 120 and each intelligent agent in the multiple execution intelligent agents 130 can also be deployed in the terminal. The protection scope of the embodiments of this application is not limited thereto.

[0032] Optionally, each agent in the multiple executing agents 130 can be divided into multiple teams based on the tasks they can perform and the target tasks to be performed. For example, such as... Figure 1As shown, Agent 1 and Agent 2 belong to Agent Team 1. For example, the objective task is information verification, i.e., verifying the authenticity of input information. For information verification, it is necessary to determine the relevant evidence for the information to be verified, and to judge the credibility of this evidence, and then, based on this evidence as a whole, judge the authenticity of the corresponding information to be verified. In this case, the objective task may include a fact extraction team and a fact verification team. The fact extraction team is used to extract statement information from the information to be verified. Statement information is a proposition to be verified extracted from the information to be verified. For example, if the information to be verified is "On year A, month B, day C, a man committed action E in place D, and then another man cleaned up after him; it was an extremely heinous crime," then two statements can be extracted: one is "On year A, month B, day C, a man committed action E in place D," and the other is "another man cleaned up after him." However, "extremely heinous crime" in the information to be verified is an emotional opinion, not a fact / proposition to be verified. The fact-verification team is responsible for verifying the stated information. Specifically, this involves using various methods (such as utilizing tools) to find the corresponding evidence and form a chain of evidence to determine the veracity of the statement. For example, regarding the statement "On year A, month B, day C, a man committed act E in location D," the evidence that could be collected might include: "A child witnessed the man appearing in location D on year A, month B, day C," "A shopping receipt for the tool the man purchased for act E was collected," "Traces of use in performing act E were detected on the tool," and "Biological traces of the man remained on the tool." Based on this evidence, it can be determined that the statement is true, not false.

[0033] Optionally, for cases where the objective task is information verification, a report generation team may also be included. This team summarizes the results of the fact extraction team and the fact verification team, generating a summary of the entire verification process for the information to be verified. The generated verification report may include one or more of the following: the verification result of the information to be verified, such as whether the information to be verified is true or false; statement information, which refers to the propositions to be verified extracted from the information to be verified; the chain of evidence for the statement information, which includes at least one piece of evidence; the verification result of the statement information; and summary information, i.e., a summary of the verification process for the information to be verified.

[0034] Optionally, the computer system also includes a global shared memory 140. The global shared memory 140 is used to store core information related to the target task. For example, for a target task involving information verification, the global shared memory stores the aforementioned statement information, the evidence chain of the statement information, and the verification results of the statement information.

[0035] Optionally, the computer system also includes multiple local team memories. Each local team memory corresponds one-to-one with a multiple executive agent, such as... Figure 1 As shown. Alternatively, one of the multiple local team memories corresponds to one agent team, and an agent team includes at least two agents. The local team memory is used to store intermediate information about an agent or agent team performing a corresponding sub-task. For example, the fact-verifying team mentioned above might retrieve a lot of raw information for a statement, which is stored in the fact-verifying team's local team memory. The fact-verifying team will also extract evidence information based on the raw information stored in the local team memory. Taking the example above, the evidence information "On date A, month B, day C, a child witnessed the man appearing at location D" might be extracted from a news report (raw information).

[0036] Optionally, the computer system further includes at least one tool 150, which is invoked by the intelligent agent to implement functions not supported by the intelligent agent. For example, at least one tool 150 includes a search tool for searching web pages, documents, etc., related to the stated information or the information to be verified. As another example, at least one tool 150 includes a knowledge base for storing processing of historical information to be verified, such as statements of historical information to be verified, chains of evidence, verification results of statements, and verification results of information to be verified.

[0037] In some embodiments, the devices in the computer system described above are connected via wired or wireless networks. For example, terminal 110 is connected to management agent 120 via a wired or wireless network; terminal device 110 is connected to each agent in the plurality of execution agents 130 via a wired or wireless network; management agent 120 is connected to each agent in the plurality of execution agents 130 via a wired or wireless network; management agent 120 and shared team memory 140 are connected via a wired or wireless network; each agent in the plurality of execution agents 130 and shared team memory 140 are connected via a wired or wireless network; each agent in the plurality of execution agents 130 and corresponding local team memory are connected via a wired or wireless network; each agent in the plurality of execution agents 130 and each tool in at least one tool 150 are connected via a wired or wireless network, and so on.

[0038] Figure 2 This illustration shows a schematic diagram of an information verification system provided in an exemplary embodiment of this application. The information verification system includes a management agent 120 and multiple execution agents 130. Management agent 120 is used to obtain information to be verified; based on the information to be verified, multiple execution agents 130 are scheduled to perform information verification.

[0039] Information verification refers to the process of breaking down the information to be verified into one or more propositions, and determining the verification result of the information based on the verification results of one or more propositions. A proposition refers to the statement content in the information to be verified, such as a specific declarative sentence, rather than a vague discussion or emotional expression. For example, if the information to be verified is "On year A, month B, day C, a man committed act E in place D, and then another man cleaned up after him; it was an extremely heinous crime," then two statements can be extracted: one is "On year A, month B, day C, a man committed act E in place D," and the other is "another man cleaned up after him." The phrase "extremely heinous crime" in the information to be verified is an emotional opinion, not a fact / proposition to be verified.

[0040] Multiple execution agents 130 are used to generate verification reports based on the information to be verified.

[0041] The verification report includes at least two of the following information: The verification result of the information to be verified; Statement information, which is a proposition to be verified extracted from the information to be verified; A chain of evidence containing information, wherein the chain of evidence includes at least one piece of evidentiary information; The verification results of the stated information; Summary information refers to a summary of the verification process for the information to be verified.

[0042] For example, the verification result of the information to be verified can be any of the following: highly credible, partially false, purely false, or unverifiable. Highly credible is the first verification result, meaning the information to be verified is true, or the probability of it being true is higher than a first probability threshold, or the probability of it being false is lower than a second probability threshold. Generally, the statements in the information to be verified corresponding to the first verification result have a complete and logically unconflicting chain of evidence. For example, all the statements in the information to be verified are true (or have been verified). Partially false is the second verification result, meaning the information to be verified is partially true, partially false, or questionable, or the probability of it being true is lower than a first probability threshold but higher than a third probability threshold, or the probability of it being false is higher than a second probability threshold but lower than a fourth probability threshold. For example, some statements in the information to be verified are true (or have been verified), while others are false (or have been disproven) or questionable (or lack sufficient evidence). The third verification result is that the information to be verified is entirely false, or the probability that the information to be verified is true is lower than the third probability threshold, or the probability that the information to be verified is false is higher than the fourth probability threshold. For example, the verification result for all statements in the information to be verified is false. The fourth verification result is that the statements in the information to be verified cannot be accurately concluded due to insufficient evidence. Based on the current evidence, they cannot be identified as either true or false. For example, all statements in the information to be verified are questionable.

[0043] Optionally, the information to be verified can be unimodal information, such as one or more pieces of text, one or more images, one or more pieces of video, one or more pieces of audio, one or more hyperlinks, etc.; or, the information to be verified can also be multimodal information, such as multiple pieces of text and one image, one piece of text and one video, multiple pieces of text and one audio, multiple images and one hyperlink, one piece of text and one hyperlink, one piece of text and multiple images and one video, etc. The embodiments of this application do not limit the form and modality of the information to be verified, but the scope of protection of the embodiments of this application is not limited thereto.

[0044] Optionally, the statement information is a proposition to be verified extracted from the information to be verified. One piece of information to be verified includes one or more statements. These statements can be unimodal or multimodal. The modality of the statement information is generally related to the information to be verified, but not necessarily identical. For example, if the information to be verified consists of text and multiple images, the statement information could be a combination of: text only, images only, or a combination of text and images. The text in the statement information could be extracted from the text of the information to be verified, extracted from the images in the information to be verified, or a combination of both. For example, if the text in the information to be verified is "A geological disaster occurred in XX location," and the image is a landslide next to a sign for XX village, the statement information could be "A landslide occurred in XX village of XX location," where "XX location" comes from the text content of the information to be verified, and "A landslide occurred in XX village" is extracted from the text content of the information to be verified.

[0045] Optionally, the chain of evidence for the statement of information includes at least one piece of evidentiary information, which is information related to the statement of information. For example, evidentiary information corroborates the events that occurred in the statement of information in various aspects (such as time, place, event, etc.).

[0046] Optionally, the verification result of the statement information is the result obtained based on the chain of evidence of the statement information. The verification result of the statement information is used to indicate the authenticity of the statement information, similar to the verification result of the information to be verified mentioned above.

[0047] For example, regarding the statement "On date C, month B, year A, a man committed act E at location D," a complete and logically consistent chain of evidence typically includes the following: "A child witnessed the man appearing at location D on date C, month B, year A," "A shopping receipt was collected showing the man purchasing tools for act E," "Traces of use for act E were detected on the tools," and "Biological traces of the man remained on the tools." Here, "A child witnessed the man appearing at location D on date C, month B, year A" serves as eyewitness testimony, indicating the man was likely present. "The shopping receipt was collected showing the man purchasing tools for act E" serves as proof of purchase, indicating the man had the means to commit the act. "Traces of use for act E were detected on the tools" and "Biological traces of the man remained on the tools" are concrete evidence that act E was committed, ruling out the possibility that the man did not commit act E. By combining the above evidence, the verification result of the statement "On date C, month B, year A, a man committed act E at location D" can be obtained. If the verification result is confirmed, then the statement can be considered true. If the chain of evidence only includes the two pieces of evidence: "A child witnessed the man appearing in location D on date A, month B, day C" and "A shopping receipt was collected showing the man purchasing tools for action E," then the chain of evidence is incomplete. This is because the above evidence only establishes the precondition that the man performed action E in location D on date A, month B, day C, but does not prove that he actually performed action E. In this case, based on this chain of evidence, the verification result of the statement might be deemed insufficient evidence. If the chain of evidence includes "A shopping receipt was collected showing the man purchasing tools for action E" and "An elderly person stated on date A, month B, day C, day C, that the man appeared in location E," then there is a conflict regarding the location of the man's appearance. Since the evidence concerning the location of the man's appearance only provides a simple and approximate date, if the publicly known information indicates that location D and location E are far apart and difficult to reach within a day, then this indicates a contradiction between the chain of evidence and the statement. If the evidence is authentic and reliable, the verification result of the statement can be deemed falsified, meaning the statement is false information.

[0048] Optionally, the summary information is a summary of the verification process for the information to be verified. For example, the summary information is used to indicate which statements in the information to be verified are supported by what evidence, which statements are identified as false information by what evidence, and so on.

[0049] Optionally, each of the multiple execution agents 130 can be simply referred to as an agent, and these execution agents are used to execute the target task or a subtask of the target task. For example, a target task can be broken down into multiple subtasks, and each subtask in the multiple subtasks is executed by one or more execution agents.

[0050] Optionally, each of the multiple execution agents 130 is scheduled by a management agent. For example, the management agent determines whether to schedule a second agent to execute a second subtask based on the execution status of a first agent for a first subtask. The second subtask depends on the first subtask; for example, the second subtask processes the output of the first subtask, meaning the input of the second subtask is the output of the first subtask. Alternatively, the management agent determines whether to call back a fourth agent to re-execute a fourth subtask based on the execution result of a third agent for a third subtask. For example, the fourth subtask is a preceding subtask of the third subtask, and the execution result of the third subtask is related to the execution result of the fourth subtask. For instance, if the execution result of the fourth subtask is of low quality, the execution result of the third subtask will also be of low quality. However, the execution result of the fourth subtask is difficult to quantify, while the execution result of the third subtask is easily quantifiable. In this case, the quality of the execution result of the third subtask can be used to determine whether the fourth subtask needs to be re-executed. For example, the third subtask is the verification task of the stated information, and the fourth subtask is the evidence chain collection task of the stated information. If the evidence chain of the stated information is insufficient, it is easy to obtain the verification result of doubtful information with insufficient evidence. At this time, the fourth subtask can be called back to collect more evidence to verify the stated information.

[0051] Optionally, all of the multiple execution agents 130 are scheduled to perform information verification tasks for the information to be verified, or some of the multiple execution agents 130 are scheduled to perform information verification tasks for the information to be verified.

[0052] In summary, the information verification system provided in this application embodiment demonstrates a direction for information verification based on multi-agent collaboration. By utilizing multiple executive agents to complete the information verification process, it fully leverages the capabilities of the agents to achieve efficient information verification. Furthermore, after information verification using multi-agent collaboration, a complete verification report is generated. This report not only provides the verification result (such as true or false information) for the information to be verified in traditional methods, but also provides the statement information, the evidence chain for the statement information, the verification result of the statement information, and summary information. Compared to the completely black-box approach of traditional methods (i.e., the user cannot know how the verification result was obtained), the final output has higher reliability, interpretability, and traceability. Moreover, this information verification system also includes a management agent that schedules multiple executive agents, eliminating the need for the user to manually schedule different agents to perform information verification tasks, thus improving the interactive efficiency of information verification for a given piece of information.

[0053] The following section will further describe the process of verifying information by at least one entity.

[0054] 1. Information verification process of multiple agents In some embodiments, the plurality of executive agents include a fact extraction agent, a fact verification agent, and a report generation agent.

[0055] A fact extraction agent is used to obtain at least one statement based on the information to be verified.

[0056] For example, a fact extraction agent extracts at least one statement from the information to be verified. This can be achieved through tasks such as Named Entity Recognition (NER), Relation Extraction (RE), and Event Extraction (EE). Entity Recognition involves identifying named entities with specific meanings from the information to be verified and classifying them into predefined categories (such as names of people, places, organizations, times, products, diseases, etc.). Relation Extraction identifies semantic relationships between entity pairs from the information to be verified and categorizes these relationships into predefined types (such as "founder-company," "located-place," "production-product"). Its goal is to reveal the logical connections between entities and form a structured knowledge network. Event Extraction identifies the occurrence of specific events from the information to be verified and extracts the core elements of the events (such as trigger words, time, place, participants, and results). It should be understood that some of the information extracted through the above methods may be opinions or attitudes toward the event. In this case, it is necessary to filter the extracted information to obtain the final statement.

[0057] For example, for each piece of extracted information, the following vocabulary is identified: sentiment words, evaluative adjectives, or subjective judgments. These words often indicate that the corresponding information is an opinion rather than a fact. Sentiment words include "excellent," "terrible," "surprising," etc. Evaluative adjectives include "efficient," "reliable," etc. Subjective judgments include "I think," "experts say...maybe," etc. Optionally, modal words and time indicators that have not yet occurred can also be excluded. Modal words include "expected," "possible," "hopeful," and time indicators that have not yet occurred include "future," "next year," etc. These words indicate that the corresponding information is a predicted but not yet occurred event.

[0058] In some embodiments, a fact extraction agent is used to obtain at least one statement and the weights of each statement in the at least one statement based on the information to be verified, wherein the weights of each statement are used to indicate the importance of each statement in the information to be verified.

[0059] Optionally, the weight of each statement is calculated by the thinking model of the fact extraction agent.

[0060] Optionally, the weight of each statement is related to at least one of the semantic function, position of occurrence, etc., represented by each statement in the information to be verified.

[0061] Optionally, the semantic function of the stated information refers to the semantic function that the stated information undertakes in the information to be verified. For example, the semantic function includes at least one of the following: core argument, core evidence, supporting evidence, and background information. A core argument refers to the stated information being the core conclusion of the information to be verified, such as the completion of the development of the XXX vaccine, which can reduce the severity rate of XX disease. Core evidence refers to the stated information being key data or facts in the information to be verified that directly support the core argument, such as clinical trials showing that the severity rate of those vaccinated with the XXX vaccine decreased by 90%. Supporting evidence refers to the stated information being supplementary information in the information to be verified that indirectly supports the argument, such as the XXX vaccine covering more than 100 countries. Background information refers to the stated information being facts that provide context in the information to be verified, such as the development history of the XXX vaccine. Generally speaking, the weight of the stated information whose semantic function is the core argument > the weight of the stated information whose semantic function is the core evidence > the weight of the stated information whose semantic function is the supporting evidence > the weight of the stated information whose semantic function is the background information.

[0062] Optionally, the location of the statement information is used to indicate its position within the information to be verified. This location may include the title, the first sentence of the first paragraph, the first sentence of other paragraphs, or other positions. For example, different positions correspond to different weights: a statement appearing in the title has a weight of 1.0, a statement appearing in the first sentence of the first paragraph has a weight of 0.8, a statement appearing in the first sentence of another paragraph has a weight of 0.6, and a statement appearing in other positions has a weight of 0.3. It should be understood that the above weight settings are for illustrative purposes only and can be flexibly set in actual scenarios; this application does not limit these settings.

[0063] Optionally, the final weight of the stated information = a × weight corresponding to the semantic function + b × weight corresponding to the position of occurrence. Where a and b are coefficients preset by the developer.

[0064] It should be understood that the weight can also be related to other factors, such as the irreplaceability of the statement information in the information to be verified, the role of the statement information in the logical chain of the information to be verified, the authority of the entity involved in the statement information, etc., which will not be elaborated here in the embodiments of this application.

[0065] It should be noted that while the above examples use text representing declarative information and information to be verified, in some embodiments, the declarative information and information to be verified can also be multimodal information. In this case, the fact extraction agent can be jointly scheduled with other multimodal processing agents. For example, the multiple execution agents may include at least one of the following: an image analysis agent, a video analysis agent, and an audio analysis agent. The management agent, based on the information to be verified, schedules at least one of the image analysis agent, video analysis agent, and audio analysis agent to extract multimodal features from the information to be verified; and schedules the fact extraction agent based on the multimodal features and the information to be verified, thereby obtaining declarative information that incorporates multimodal features.

[0066] By determining the weight of the stated information, the importance of the stated information within the information to be verified is highlighted. This allows for a more reasonable focus on the more important stated information during the final verification process based on the stated information, leading to a more scientific and accurate assessment of the overall authenticity of the information to be verified. Consequently, the final verification results are more targeted, systematic, and practical.

[0067] A fact-verifying agent is used to invoke at least one tool for each statement in at least one statement to obtain the evidence chain of the statement and / or the verification result of the statement.

[0068] Optionally, the fact-verifying agent invokes at least one tool for each statement in at least one statement to obtain multiple pieces of evidence for the statement; based on the multiple pieces of evidence for the statement, it obtains the evidence chain for the statement; and based on the evidence chain for the statement, it obtains the verification result for the statement.

[0069] For example, the evidence chain of the statement information includes multiple pieces of evidence, or includes at least one of multiple pieces of evidence. For instance, the evidence chain of the statement information is obtained by filtering multiple pieces of evidence. The filtering process of multiple pieces of evidence is used for at least one of the following: removing evidence information from different sources but with the same content; merging evidence information from different sources but with a content similarity higher than a preset threshold.

[0070] Optionally, the chain of evidence may include at least one discrete piece of evidence, or the chain of evidence may be obtained by organizing at least one piece of evidence around a pre-defined logic (such as the five elements of subject-event-time-place-causation).

[0071] Optionally, the confidence level corresponding to each piece of evidence in the chain of evidence is calculated. The verification result of the statement is then obtained based on the confidence level corresponding to the chain of evidence. For example, the confidence level corresponding to the chain of evidence is calculated based on the similarity between at least one piece of evidence and the statement; or, the confidence level corresponding to the chain of evidence is calculated based on the confidence levels corresponding to at least one piece of evidence. For instance, first, the total confidence level of each piece of evidence is calculated; if there is contradictory evidence among at least one piece of evidence, the corresponding confidence level is subtracted from the total confidence level to obtain the final confidence level.

[0072] In some embodiments, the fact-verifying agent is configured to, for the i-th statement in at least one set of statements, invoke at least one tool to query and obtain verification information for the i-th statement, wherein the verification information includes at least one piece of evidence information, the source of at least one piece of evidence information, and the confidence level of at least one piece of evidence information; based on the verification information for the i-th statement, obtain the evidence chain for the i-th statement and / or the verification result for the i-th statement; let i = i + 1, and repeat the execution from the step of invoking at least one tool to query and obtain the verification information for the i-th statement until the verification of at least one statement is completed, where i is a positive integer.

[0073] Optionally, the confidence level of at least one piece of evidentiary information is determined based on at least one of the following factors: source; time of publication of the evidentiary information; quality of the evidentiary information.

[0074] For example, different sources correspond to different levels of confidence. A source may be at least one of the following: official media; authoritative institutions; social media; or ordinary users. Generally speaking, official media has a higher level of confidence than authoritative institutions, authoritative institutions have a higher level of confidence than social media, and social media has a higher level of confidence than ordinary users.

[0075] For example, the closer the release time of the evidence information is to the present time, the higher its confidence level.

[0076] For example, the quality of evidentiary information refers to the degree of detail in the evidentiary information. For instance, the confidence level of evidentiary information that details the time, place, clothing of the participants, appearance of the participants, and the process of the event is higher than that of evidentiary information that only describes the process of the event.

[0077] In some embodiments, a fact-verifying agent is configured to invoke at least one tool to query verification information of the i-th statement in at least one statement, respectively, under multiple query conditions. The multiple query conditions are restrictions on querying the verification information, and the query conditions include at least one of the following aspects: time range; evidence set; search engine; modality of verification information.

[0078] In other words, the fact-verifying agent supports multiple rounds of searching and verifying evidence for a single statement. Different rounds can correspond to different query conditions.

[0079] Optionally, the time range refers to the search range of evidence information in the time dimension. For example, in the first round of query, the time range can be unlimited; in the second round of query, the time range is limited to within one month; in the third round of query, the time range is limited to within one week, and so on.

[0080] Optionally, an evidence set, similar to a search engine, is used to limit the scope of the search for evidentiary information. For example, an evidence set could include one of the following types: official media; archives. Restrictions on the search engine could allow the use of only pre-agreed search engines for evidentiary information in a given round of queries. Restrictions on both the evidence set and the search engine can prevent unreliable information sources from leading to excessively low confidence levels in the evidentiary information.

[0081] Optionally, the modality of the verification information includes at least one of the following: text; image; video; audio; hyperlink.

[0082] Optionally, the query under multiple query conditions can be performed during a single verification process for the i-th statement. For example, the management agent schedules the fact verification agent to verify at least one statement or the i-th statement. The fact verification agent calls at least one tool to query the verification information of the i-th statement under multiple query conditions for the i-th statement. Based on the verification information of the i-th statement, the verification result of the i-th statement is obtained.

[0083] Alternatively, queries under multiple query conditions can be performed during multiple verification processes for the i-th statement. For example, the management agent schedules a fact-verifying agent to verify at least one statement, or the i-th statement. The fact-verifying agent, for the i-th statement among the at least one statement, invokes at least one tool to query the first verification information of the i-th statement under the first query condition of multiple query conditions; based on the first verification information of the i-th statement, it obtains the verification result of the i-th statement. If the confidence level of the verification result of the i-th statement is lower than the confidence threshold, and / or the verification result of the i-th statement is insufficient evidence, then the management agent schedules the fact-verifying agent again to verify the i-th statement, or the fact-verifying agent actively re-executes the verification of the i-th statement. In this case, a different verification condition can be used, such as a second verification condition. For example, the fact-verifying agent, for the i-th statement in at least one set of statements, invokes at least one tool to retrieve second verification information for the i-th statement under a second query condition with multiple query conditions; based on the second verification information of the i-th statement, it obtains a second verification result for the i-th statement. It then continues to determine whether further verification is needed, and so on, until the obtained verification result is true or false information, and / or the verification count reaches the verification limit. The verification limit is set in advance by the developer or user.

[0084] In some embodiments, at least one tool includes at least one of the following: Search tools are used to search for at least one of the following: web pages, documents, and evidence information related to the stated information; Web scraping tools are used to scrape web pages and / or document content to obtain evidentiary information; A knowledge base is used to record content related to historical information that needs to be verified.

[0085] The report-generating agent is used to generate a verification report based on at least one of the following: statement information, the chain of evidence for the statement information, and the verification result of the statement information.

[0086] Optionally, each of the aforementioned execution agents can be a single execution agent or a team of execution agents. That is, the aforementioned fact extraction agent can be a fact extraction team, the aforementioned fact verification agent can be a fact verification team, and the aforementioned report generation agent can be a report generation team. The fact extraction team can include multiple execution agents, the fact verification team can include multiple execution agents, and the report generation team can include multiple execution agents. Different execution agents are used to execute sub-tasks within their respective teams. This application embodiment does not limit this aspect.

[0087] In summary, the system provided in this application implements information verification through multiple execution agents with different functions, thereby improving the reliability of information verification through a more reasonable division of labor. Furthermore, compared to training a single execution agent capable of completing the entire information verification process, training execution agents capable of completing each sub-task in the information verification process is less difficult, and the trained execution agents are more specialized, enabling them to better complete the entire information verification process and improving the efficiency and robustness of information verification.

[0088] In the presence of multiple agents, it is also necessary to consider the data generated during the collaboration of these agents. Therefore, embodiments of this application design an information verification system that includes global shared memory and local team memory.

[0089] 2. Data Management In some embodiments, the information verification system further includes a global shared memory and multiple local team memories. Optionally, the multiple local team memories correspond one-to-one with the multiple executing agents. Alternatively, the multiple local team memories have a many-to-one relationship with the multiple executing agents. Or, the multiple local team memories have a one-to-many relationship with the multiple executing agents.

[0090] A fact extraction agent is used to write at least one statement into the global shared memory.

[0091] A fact-verifying agent is configured to read at least one statement from a global shared memory; for each statement in the at least one statement, invoke at least one tool to obtain at least one piece of evidence corresponding to each statement; write at least one piece of evidence into the fact-verifying agent's local team memory; based on the at least one piece of evidence corresponding to each statement stored in the fact-verifying agent's local team memory, determine the evidence chain corresponding to each statement and / or the verification result of each statement; and write the evidence chain corresponding to each statement and / or the verification result of each statement into the global shared memory.

[0092] The report-generating agent generates a verification report based on information stored in the globally shared memory.

[0093] In other words, the global shared memory is used to store core information related to the target task. For example, for the target task of information verification, the information that the global shared memory supports storing includes at least one of the following: statement information; the evidence chain of the statement information; the verification result of the statement information; the verification result of the information to be verified; summary information; verification report; the source of the evidence information in the evidence chain; the confidence level of the evidence information; and the verification status of the statement information. The verification status of the statement information includes verified and unverified.

[0094] Optionally, local team memory is used to store information from the process of handling sub-tasks or target tasks, such as evidence information, original content obtained when searching for evidence, etc.

[0095] Optionally, the report-generating agent stores the generated verification report in a global shared memory. The management agent reads the verification report from the global shared memory and sends it to the client. The client can then display the verification report to show the user the verification results for the information to be verified.

[0096] Optionally, the local team memory is also used to store core information of the target task generated during execution. After the executing agent finishes its corresponding subtask, it reads the core information of the target task from the local team memory and stores it in the global shared memory. For example, for a fact extraction agent, the intermediate data and the final result (i.e., at least one statement) generated during the processing of the information to be verified are initially stored in the local team memory until the fact extraction subtask ends, at least one statement from the local team memory is then written to the global shared memory. The same applies to the fact verification agent; the evidence chain of the statement, the verification result of the statement, etc., are first saved in the local team memory when generated. They are only written to the global shared memory after the verification of a statement is completed, or after the verification of all statements in at least one statement is completed.

[0097] The setup of global shared memory and local team memory allows the global shared memory to record only the state of core tasks (such as the final verification results of each statement), maintaining a high signal-to-noise ratio. Local team memory, on the other hand, handles intermediate processes and "dirty work" within their respective subtasks, ensuring focus within the team and preventing irrelevant information from interfering with other teams' decisions. This avoids the problem of chaotic memory management in multi-agent collaboration. For example, in traditional methods, all information generated during system operation—including successful attempts, failed paths, intermediate considerations, and redundant search results—is often mixed in the same memory pool. This results in effective information being overwhelmed by noise, leading to extremely low information density, which not only affects decision-making efficiency but may also cause incorrect judgments due to interference from irrelevant information.

[0098] Optionally, a management agent is used to schedule individual execution agents among multiple execution agents based on information stored in a globally shared memory.

[0099] Optionally, the global shared memory is also used to store the execution status of each executing agent. For example, after the fact extraction agent completes the writing of at least one statement, it adds status information such as "fact extraction subtask completed" to the global shared memory. After the management agent reads this status information from the global shared memory, it determines that the task of the fact extraction agent has been completed and can schedule the fact verification agent to execute the next task. Alternatively, if the fact extraction agent writes at least one statement to the global shared memory, and the management agent detects that the time elapsed since the last statement written by the fact extraction agent is greater than a time elapsed threshold, it confirms that the fact extraction agent has completed the fact extraction subtask and automatically calls the fact verification agent.

[0100] For example, a management agent is used for at least one of the following: Once the fact extraction agent has written at least one statement to the global shared memory, schedule the fact verification agent. When the number of evidence information written into the evidence chain by the fact-verifying agent in the global shared memory reaches a first quantity, the scheduling report generation agent is activated. When the number of verification results written to the global shared memory by the fact-verifying agent reaches the second threshold, the scheduling report generation agent is activated.

[0101] For example, the first quantity and the second quantity are equal to the number of statements, or the first quantity and the second quantity are less than the number of statements. For instance, if there are at least one statement and there are n statements, the first quantity can be equal to n, or the first quantity can be less than n; the second quantity can also be equal to n, or the second quantity can be less than n. n is a positive integer.

[0102] The management agent supports the dynamic scheduling of multiple execution agents based on the task status in the globally shared memory, enabling dynamic and non-linear intelligent task scheduling in accordance with real-time task progress. It can decide whether to continue verification, request supplementary evidence, transfer to the next team, or terminate the process early, thus possessing high flexibility and intelligence in dealing with unknown and complex situations.

[0103] Optionally, a management agent initializes the global shared memory in response to the acquisition of the information to be verified. Alternatively, a management agent initializes the global shared memory in response to the input of the information to be verified.

[0104] Optionally, a management agent is used to update a knowledge base based on information in a globally shared memory. The knowledge base records content related to historical information to be verified.

[0105] For example, the information stored in the global shared memory is structured information, which includes at least one of the following: statement information; evidence chain of the statement information; verification result of the statement information; verification result of the information to be verified; summary information; verification report; source of evidence information in the evidence chain; confidence level of evidence information; verification status of statement information.

[0106] Figure 3 A flowchart illustrating an exemplary embodiment of this application provides an information verification method based on multi-agent cooperation. This method is executed by an information verification system, which may be as described above. Figure 2 The information verification system shown in the figure. The method includes: Step 210: The management agent obtains the information to be verified.

[0107] The information to be verified is the information entered by the user through the client. The information to be verified can be unimodal or multimodal.

[0108] Step 220: The management agent schedules multiple execution agents to perform information verification based on the information to be verified.

[0109] Information verification refers to the process of breaking down the information to be verified into one or more propositions, and determining the verification result of the information based on the verification results of one or more propositions. A proposition refers to the statement content in the information to be verified, such as a specific declarative sentence, rather than a vague discussion or emotional expression. For example, if the information to be verified is "On year A, month B, day C, a man committed act E in place D, and then another man cleaned up after him; it was an extremely heinous crime," then two statements can be extracted: one is "On year A, month B, day C, a man committed act E in place D," and the other is "another man cleaned up after him." The phrase "extremely heinous crime" in the information to be verified is an emotional opinion, not a fact / proposition to be verified.

[0110] Step 230: Multiple execution agents generate a verification report based on the information to be verified.

[0111] The verification report includes at least two of the following information: The verification result of the information to be verified; Statement information, which is a proposition to be verified extracted from the information to be verified; A chain of evidence containing information, wherein the chain of evidence includes at least one piece of evidentiary information; The verification results of the stated information; Summary information refers to a summary of the verification process for the information to be verified.

[0112] For example, the verification result of the information to be verified can be any of the following: highly credible, partially false, purely false, or unverifiable. Highly credible is the first verification result, meaning the information to be verified is true, or the probability of it being true is higher than a first probability threshold, or the probability of it being false is lower than a second probability threshold. Generally, the statements in the information to be verified corresponding to the first verification result have a complete and logically unconflicting chain of evidence. For example, all the statements in the information to be verified are true (or have been verified). Partially false is the second verification result, meaning the information to be verified is partially true, partially false, or questionable, or the probability of it being true is lower than a first probability threshold but higher than a third probability threshold, or the probability of it being false is higher than a second probability threshold but lower than a fourth probability threshold. For example, some statements in the information to be verified are true (or have been verified), while others are false (or have been disproven) or questionable (or lack sufficient evidence). The third verification result is that the information to be verified is entirely false, or the probability that the information to be verified is true is lower than the third probability threshold, or the probability that the information to be verified is false is higher than the fourth probability threshold. For example, the verification result for all statements in the information to be verified is false. The fourth verification result is that the statements in the information to be verified cannot be accurately concluded due to insufficient evidence. Based on the current evidence, they cannot be identified as either true or false. For example, all statements in the information to be verified are questionable.

[0113] Optionally, the information to be verified can be unimodal information, such as one or more pieces of text, one or more images, one or more pieces of video, one or more pieces of audio, one or more hyperlinks, etc.; or, the information to be verified can also be multimodal information, such as multiple pieces of text and one image, one piece of text and one video, multiple pieces of text and one audio, multiple images and one hyperlink, one piece of text and one hyperlink, one piece of text and multiple images and one video, etc. The embodiments of this application do not limit the form and modality of the information to be verified, but the scope of protection of the embodiments of this application is not limited thereto.

[0114] Optionally, the statement information is a proposition to be verified extracted from the information to be verified. One piece of information to be verified includes one or more statements. These statements can be unimodal or multimodal. The modality of the statement information is generally related to the information to be verified, but not necessarily identical. For example, if the information to be verified consists of text and multiple images, the statement information could be a combination of: text only, images only, or a combination of text and images. The text in the statement information could be extracted from the text of the information to be verified, extracted from the images in the information to be verified, or a combination of both. For example, if the text in the information to be verified is "A geological disaster occurred in XX location," and the image is a landslide next to a sign for XX village, the statement information could be "A landslide occurred in XX village of XX location," where "XX location" comes from the text content of the information to be verified, and "A landslide occurred in XX village" is extracted from the text content of the information to be verified.

[0115] Optionally, the chain of evidence for the statement of information includes at least one piece of evidentiary information, which is information related to the statement of information. For example, evidentiary information corroborates the events that occurred in the statement of information in various aspects (such as time, place, event, etc.).

[0116] Optionally, the verification result of the statement information is the result obtained based on the chain of evidence of the statement information. The verification result of the statement information is used to indicate the authenticity of the statement information, similar to the verification result of the information to be verified mentioned above.

[0117] For example, regarding the statement "On date C, month B, year A, a man committed act E at location D," a complete and logically consistent chain of evidence typically includes the following: "A child witnessed the man appearing at location D on date C, month B, year A," "A shopping receipt was collected showing the man purchasing tools for act E," "Traces of use for act E were detected on the tools," and "Biological traces of the man remained on the tools." Here, "A child witnessed the man appearing at location D on date C, month B, year A" serves as eyewitness testimony, indicating the man was likely present. "The shopping receipt was collected showing the man purchasing tools for act E" serves as proof of purchase, indicating the man had the means to commit the act. "Traces of use for act E were detected on the tools" and "Biological traces of the man remained on the tools" are concrete evidence that act E was committed, ruling out the possibility that the man did not commit act E. By combining the above evidence, the verification result of the statement "On date C, month B, year A, a man committed act E at location D" can be obtained. If the verification result is confirmed, then the statement can be considered true. If the chain of evidence only includes the two pieces of evidence: "A child witnessed the man appearing in location D on date A, month B, day C" and "A shopping receipt was collected showing the man purchasing tools for action E," then the chain of evidence is incomplete. This is because the above evidence only establishes the precondition that the man performed action E in location D on date A, month B, day C, but does not prove that he actually performed action E. In this case, based on this chain of evidence, the verification result of the statement might be deemed insufficient evidence. If the chain of evidence includes "A shopping receipt was collected showing the man purchasing tools for action E" and "An elderly person stated on date A, month B, day C, day C, that the man appeared in location E," then there is a conflict regarding the location of the man's appearance. Since the evidence concerning the location of the man's appearance only provides a simple and approximate date, if the publicly known information indicates that location D and location E are far apart and difficult to reach within a day, then this indicates a contradiction between the chain of evidence and the statement. If the evidence is authentic and reliable, the verification result of the statement can be deemed falsified, meaning the statement is false information.

[0118] Optionally, the summary information is a summary of the verification process for the information to be verified. For example, the summary information is used to indicate which statements in the information to be verified are supported by what evidence, which statements are identified as false information by what evidence, and so on.

[0119] Optionally, each of the multiple execution agents 130 can be simply referred to as an agent, and these execution agents are used to execute the target task or a subtask of the target task. For example, a target task can be broken down into multiple subtasks, and each subtask in the multiple subtasks is executed by one or more execution agents.

[0120] Optionally, each of the multiple execution agents 130 is scheduled by a management agent. For example, the management agent determines whether to schedule a second agent to execute a second subtask based on the execution status of a first agent for a first subtask. The second subtask depends on the first subtask; for example, the second subtask processes the output of the first subtask, meaning the input of the second subtask is the output of the first subtask. Alternatively, the management agent determines whether to call back a fourth agent to re-execute a fourth subtask based on the execution result of a third agent for a third subtask. For example, the fourth subtask is a preceding subtask of the third subtask, and the execution result of the third subtask is related to the execution result of the fourth subtask. For instance, if the execution result of the fourth subtask is of low quality, the execution result of the third subtask will also be of low quality. However, the execution result of the fourth subtask is difficult to quantify, while the execution result of the third subtask is easily quantifiable. In this case, the quality of the execution result of the third subtask can be used to determine whether the fourth subtask needs to be re-executed. For example, the third subtask is the verification task of the stated information, and the fourth subtask is the evidence chain collection task of the stated information. If the evidence chain of the stated information is insufficient, it is easy to obtain the verification result of doubtful information with insufficient evidence. At this time, the fourth subtask can be called back to collect more evidence to verify the stated information.

[0121] Optionally, all of the multiple execution agents 130 are scheduled to perform information verification tasks for the information to be verified, or some of the multiple execution agents 130 are scheduled to perform information verification tasks for the information to be verified.

[0122] In summary, the method provided in this application utilizes multiple intelligent agents to complete the information verification process, fully leveraging the capabilities of these agents to achieve efficient information verification. Furthermore, after multi-agent collaboration for information verification, a complete verification report is generated. This report not only provides the verification result (e.g., true or false information) for the information to be verified as in traditional methods, but also provides the statement information, the evidence chain for the statement information, the verification result of the statement information, and summary information. Compared to the completely black-box approach of traditional methods (i.e., the user cannot know how the verification result was obtained), the final output has higher reliability, interpretability, and traceability. Moreover, this information verification system also includes a management agent that schedules multiple intelligent agents, eliminating the need for the user to manually schedule different agents to perform information verification tasks, thus improving the interactive efficiency of information verification for a given piece of information.

[0123] The following section will further describe the process of verifying information by at least one entity.

[0124] 1. Information verification process of multiple agents In some embodiments, the plurality of executive agents include a fact extraction agent, a fact verification agent, and a report generation agent.

[0125] The fact extraction agent obtains at least one statement based on the information to be verified.

[0126] The fact-verifying agent invokes at least one tool for each statement in at least one statement to obtain the evidence chain of the statement and / or the verification result of the statement.

[0127] In some embodiments, the fact-verifying agent, for the i-th statement in at least one statement, invokes at least one tool to query and obtain the verification information of the i-th statement. The verification information includes at least one piece of evidence information, the source of at least one piece of evidence information, and the confidence level of at least one piece of evidence information. Based on the verification information of the i-th statement, the agent obtains the evidence chain of the i-th statement and / or the verification result of the i-th statement. Let i = i + 1, and repeat the step from invoking at least one tool to query and obtain the verification information of the i-th statement until the verification of at least one statement is completed, where i is a positive integer.

[0128] In some embodiments, at least one tool includes at least one of the following: Search tools are used to search for at least one of the following: web pages, documents, and evidence information related to the stated information; Web scraping tools are used to scrape web pages and / or document content to obtain evidentiary information; A knowledge base is used to record content related to historical information that needs to be verified.

[0129] The report-generating agent generates a verification report based on at least one of the stated information, the evidence chain of the stated information, and the verification result of the stated information.

[0130] For details, please refer to "1. Information verification process of multiple agents" in the above system embodiment. This application embodiment does not limit this.

[0131] In the presence of multiple agents, it is also necessary to consider the data generated during the collaboration of these agents. Therefore, embodiments of this application design an information verification system that includes global shared memory and local team memory.

[0132] 2. Data Management In some embodiments, the information verification system further includes a global shared memory and multiple local team memories. Optionally, the multiple local team memories correspond one-to-one with the multiple executing agents. Alternatively, the multiple local team memories have a many-to-one relationship with the multiple executing agents. Or, the multiple local team memories have a one-to-many relationship with the multiple executing agents.

[0133] The fact extraction agent writes at least one statement into the global shared memory.

[0134] The fact-verifying agent reads at least one statement from the global shared memory; for each statement in the at least one statement, it invokes at least one tool to obtain at least one piece of evidence corresponding to each statement; it writes at least one piece of evidence into the fact-verifying agent's local team memory; based on the at least one piece of evidence corresponding to each statement stored in the fact-verifying agent's local team memory, it determines the evidence chain corresponding to each statement and / or the verification result of each statement; and it writes the evidence chain corresponding to each statement and / or the verification result of each statement into the global shared memory.

[0135] The report-generating agent generates a verification report based on information stored in the globally shared memory.

[0136] Optionally, the management agent schedules each of the multiple execution agents based on information stored in the global shared memory.

[0137] For example, the management agent performs at least one of the following steps: Once the fact extraction agent has written at least one statement to the global shared memory, schedule the fact verification agent. When the number of evidence information written into the evidence chain by the fact-verifying agent in the global shared memory reaches a first quantity, the scheduling report generation agent is activated. When the number of verification results written to the global shared memory by the fact-verifying agent reaches the second threshold, the scheduling report generation agent is activated.

[0138] Optionally, the management agent initializes the global shared memory in response to the acquisition of the information to be verified. Alternatively, the management agent initializes the global shared memory in response to the input of the information to be verified.

[0139] Optionally, the management agent updates the knowledge base based on information in a globally shared memory. The knowledge base is used to record content related to historical information to be verified.

[0140] For example, the information stored in the global shared memory is structured information, which includes at least one of the following: statement information; evidence chain of the statement information; verification result of the statement information; verification result of the information to be verified; summary information; verification report; source of evidence information in the evidence chain; confidence level of evidence information; verification status of statement information.

[0141] For details, please refer to "2. Data Management" in the above system embodiment. This application embodiment does not limit this aspect.

[0142] Figure 4 This application illustrates an exemplary embodiment of an information verification method based on multi-agent collaboration. The method is applied to a system comprising: a management agent, multiple executive agents, a global shared memory, and multiple local team memories corresponding to the multiple executive agents. The method includes: Step 310: The management agent responds to the acquired task to be executed by initializing the global shared memory.

[0143] Optionally, the management agent writes the acquired tasks to be executed into the global shared memory.

[0144] Optionally, the global shared memory is used to store core information related to the task to be executed. For example, for a task to be executed involving information verification, the information stored in the global shared memory includes at least one of the following: statement information; the chain of evidence for the statement information; the verification result of the statement information; the verification result of the information to be verified; summary information; verification report; the source of the evidence information in the chain of evidence; the confidence level of the evidence information; and the verification status of the statement information. The verification status of the statement information includes verified and unverified.

[0145] In some embodiments, the management agent, in response to the acquired task content, breaks the task content into multiple subtasks.

[0146] Optionally, in response to the acquired task to be executed, the management agent breaks down the task into multiple subtasks and determines the execution order of each subtask.

[0147] Step 320: The management agent monitors the task status in the global shared memory and dynamically schedules the target agent from multiple execution agents based on changes in the task status.

[0148] For example, the task status includes at least one of the following: not executed; executing; completed.

[0149] Optionally, when multiple subtasks exist, the task state includes the task state of each subtask, such as the subtask's task state including at least one of the following: subtask not executed, subtask in execution, or subtask completed. Optionally, when multiple subtasks exist, the management agent monitors the task state of each subtask in the globally shared memory.

[0150] In some embodiments, step 320 may be implemented as: the management agent monitoring the task state in the global shared memory; the management agent performing at least one of the following steps: When the task state in the global shared memory is the target state, the target agent is dynamically scheduled from multiple executing agents, and there is a correspondence between the target state and the target agent. If the task status in the global shared memory is that the first subtask has been completed and the score of the processing result of the first subtask is lower than the first score threshold, the target agent is rescheduled from multiple executing agents and used to execute the first subtask. When the task status in the global shared memory is the execution of the second subtask, and the feedback information of the first agent, including the cooperation request, is read, the target agent is scheduled from multiple executing agents. The first agent is the executing agent that executes the second subtask, and the cooperation request is used to request the scheduling of other agents to assist in completing the second subtask. If the processing status of the third subtask indicated by the task status in the global shared memory meets the rescheduling conditions, a target agent is selected and scheduled from multiple executing agents. The rescheduling conditions include at least one of the following: the third subtask fails to process; the processing latency of the third subtask exceeds a first latency threshold; the processing progress of the third subtask is lower than a first progress threshold; or the real-time load of the executing agent executing the third subtask is higher than a first load threshold.

[0151] For example, the task to be executed can be broken down into multiple subtasks. The management agent monitors the task status of each subtask and schedules the target agent to execute the next subtask when the task status of the preceding subtask is updated to be completed.

[0152] For example, the management agent determines whether to schedule the second agent to execute the second subtask based on the execution status of the first agent on the first subtask. The second subtask depends on the first subtask; for example, the second subtask processes the output of the first subtask, meaning the input of the second subtask is the output of the first subtask. As another example, the management agent determines whether to call back the fourth agent to re-execute the fourth subtask based on the execution result of the third agent on the third subtask. For example, the fourth subtask is a preceding subtask of the third subtask, and the execution result of the third subtask is related to the execution result of the fourth subtask. For instance, if the execution result of the fourth subtask is of low quality, the execution result of the third subtask will also be of low quality. However, the execution result of the fourth subtask is difficult to quantify, while the execution result of the third subtask is easily quantifiable. In this case, the quality of the execution result of the third subtask can be used to determine whether the fourth subtask needs to be re-executed. For example, the third subtask is the verification task of the stated information, and the fourth subtask is the evidence chain collection task of the stated information. If the evidence chain of the stated information is insufficient, it is easy to obtain the verification result of doubtful information with insufficient evidence. At this time, the fourth subtask can be called back to collect more evidence to verify the stated information.

[0153] For example, when the management agent's task status in the global shared memory is "the first subtask has been completed," it reads the processing result of the first subtask from the global shared memory and scores it to obtain a score for the processing result of the first subtask. If the score of the processing result of the first subtask is lower than a first score threshold, a target agent is rescheduled from multiple executing agents to execute the first subtask. The target agent can be the executing agent that previously executed the first subtask, i.e., the executing agent whose score for the processing result of the first subtask was lower than the first score threshold; or, the target agent can be a new agent reselected by the management agent.

[0154] For example, if the first agent encounters an anomaly while executing the second sub-task, such as the task data read from the global shared memory being insufficient to execute the second sub-task, or the task data read from the global shared memory lacking some key data, or the quality of the task data read from the global shared memory being below a quality threshold, etc., the first agent can proactively send feedback information, carrying a collaboration request in the feedback information. If the management agent's task status in the global shared memory is "second sub-task in progress," and it receives the feedback information from the first agent including a collaboration request, it schedules a target agent from among multiple executing agents. The first agent is the executing agent for the second sub-task, and the collaboration request is used to request the scheduling of other agents to assist in completing the second sub-task. The collaboration request may carry the name or identifier of the target agent, or it may carry desired capabilities, indicating the capabilities expected of the target agent. The target agent may be selected by the management agent based on the collaboration request.

[0155] For example, if the processing status of the third subtask indicated by the task status in the global shared memory of the management agent meets the rescheduling conditions, a target agent is selected and scheduled from multiple execution agents; the rescheduling conditions include at least one of the following: the third subtask processing fails; the processing latency of the third subtask exceeds a first latency threshold; the processing progress of the third subtask is lower than a first progress threshold; the real-time load of the execution agent executing the third subtask is higher than a first load threshold.

[0156] Optionally, if the executing agent fails to execute the third subtask, it reports the failure information to the managing agent. For example, if the third subtask is a search task, the failure might occur due to a tool call failure or the inability to find relevant content in the database.

[0157] For example, the real-time load of each executing agent is used to indicate at least one of the following information during the execution of the agent: the number of tasks with strict time constraints (such as real-time response and immediate processing requirements) that the executing agent needs to process per unit time, resource consumption, and related dynamic indicators (such as task queue length and processing latency).

[0158] For example, if the management agent detects that the real-time load of a target agent B among multiple target agents is too high (e.g., higher than the first load) during the execution of multiple sub-tasks, then based on the task type and / or task description of the sub-task executed by the target agent B, the management agent determines from at least two agents an agent C that indicates the replacement of the target agent B, identifies the agent C as the target agent, and schedules the agent C to execute the sub-task.

[0159] Step 330: In response to the scheduling, the target agent reads task data from the global shared memory, performs processing using the corresponding local team memory, and writes the processing results and the updated task state back to the global shared memory.

[0160] In this process, multiple intelligent agents collaborate by reading and writing to a globally shared memory to complete the task to be executed.

[0161] Optionally, the capabilities of each executing agent include at least one of the following: data analysis capability; data collection capability; data cleaning capability; multimodal perception and understanding capability; dynamic learning capability; context awareness capability; intent understanding capability; causal reasoning capability; multi-objective decision-making capability; information extraction capability; information verification capability; fact extraction capability; fact verification capability; and report generation capability.

[0162] Optionally, this multi-agent collaborative processing method can be applied to any of the following fields: information verification; content generation (code / text / images, etc.); data processing; customer service; medical diagnosis; education and tutoring; autonomous driving; financial transactions; image recognition and generation; industrial automation; legal document processing; smart home control; and scientific research assistance.

[0163] For example, the task to be performed can be one or more of the following tasks in combination: verification of real and false information; code review in software engineering; code bug fixing; auxiliary analysis of legal cases; auxiliary analysis of medical reports; and compliance verification of articles. It should be understood that the multi-agent collaborative processing method in this application embodiment can also be applied to other fields and tasks not shown. This application embodiment uses information verification as the task to be performed for explanation, but it is not intended to limit the scope of the application.

[0164] For example, the task to be performed is information verification. Multiple execution agents cooperate by reading and writing to the globally shared memory to complete the verification of the information to be verified and generate a verification report.

[0165] The verification report includes at least two of the following information: The verification result of the information to be verified; Statement information, which is a proposition to be verified extracted from the information to be verified; A chain of evidence containing information, wherein the chain of evidence includes at least one piece of evidentiary information; The verification results of the stated information; Summary information refers to a summary of the verification process for the information to be verified.

[0166] For example, the verification result of the information to be verified can be any of the following: highly credible, partially false, purely false, or unverifiable. Highly credible is the first verification result, meaning the information to be verified is true, or the probability of it being true is higher than a first probability threshold, or the probability of it being false is lower than a second probability threshold. Generally, the statements in the information to be verified corresponding to the first verification result have a complete and logically unconflicting chain of evidence. For example, all the statements in the information to be verified are true (or have been verified). Partially false is the second verification result, meaning the information to be verified is partially true, partially false, or questionable, or the probability of it being true is lower than a first probability threshold but higher than a third probability threshold, or the probability of it being false is higher than a second probability threshold but lower than a fourth probability threshold. For example, some statements in the information to be verified are true (or have been verified), while others are false (or have been disproven) or questionable (or lack sufficient evidence). The third verification result is that the information to be verified is entirely false, or the probability that the information to be verified is true is lower than the third probability threshold, or the probability that the information to be verified is false is higher than the fourth probability threshold. For example, the verification result for all statements in the information to be verified is false. The fourth verification result is that the statements in the information to be verified cannot be accurately concluded due to insufficient evidence. Based on the current evidence, they cannot be identified as either true or false. For example, all statements in the information to be verified are questionable.

[0167] Optionally, the information to be verified can be unimodal information, such as one or more pieces of text, one or more images, one or more pieces of video, one or more pieces of audio, one or more hyperlinks, etc.; or, the information to be verified can also be multimodal information, such as multiple pieces of text and one image, one piece of text and one video, multiple pieces of text and one audio, multiple images and one hyperlink, one piece of text and one hyperlink, one piece of text and multiple images and one video, etc. The embodiments of this application do not limit the form and modality of the information to be verified, but the scope of protection of the embodiments of this application is not limited thereto.

[0168] Optionally, the statement information is a proposition to be verified extracted from the information to be verified. One piece of information to be verified includes one or more statements. These statements can be unimodal or multimodal. The modality of the statement information is generally related to the information to be verified, but not necessarily identical. For example, if the information to be verified consists of text and multiple images, the statement information could be a combination of: text only, images only, or a combination of text and images. The text in the statement information could be extracted from the text of the information to be verified, extracted from the images in the information to be verified, or a combination of both. For example, if the text in the information to be verified is "A geological disaster occurred in XX location," and the image is a landslide next to a sign for XX village, the statement information could be "A landslide occurred in XX village of XX location," where "XX location" comes from the text content of the information to be verified, and "A landslide occurred in XX village" is extracted from the text content of the information to be verified.

[0169] Optionally, the chain of evidence for the statement of information includes at least one piece of evidentiary information, which is information related to the statement of information. For example, evidentiary information corroborates the events that occurred in the statement of information in various aspects (such as time, place, event, etc.).

[0170] Optionally, the verification result of the statement information is the result obtained based on the chain of evidence of the statement information. The verification result of the statement information is used to indicate the authenticity of the statement information, similar to the verification result of the information to be verified mentioned above.

[0171] For example, regarding the statement "On date C, month B, year A, a man committed act E at location D," a complete and logically consistent chain of evidence typically includes the following: "A child witnessed the man appearing at location D on date C, month B, year A," "A shopping receipt was collected showing the man purchasing tools for act E," "Traces of use for act E were detected on the tools," and "Biological traces of the man remained on the tools." Here, "A child witnessed the man appearing at location D on date C, month B, year A" serves as eyewitness testimony, indicating the man was likely present. "The shopping receipt was collected showing the man purchasing tools for act E" serves as proof of purchase, indicating the man had the means to commit the act. "Traces of use for act E were detected on the tools" and "Biological traces of the man remained on the tools" are concrete evidence that act E was committed, ruling out the possibility that the man did not commit act E. By combining the above evidence, the verification result of the statement "On date C, month B, year A, a man committed act E at location D" can be obtained. If the verification result is confirmed, then the statement can be considered true. If the chain of evidence only includes the two pieces of evidence: "A child witnessed the man appearing in location D on date A, month B, day C" and "A shopping receipt was collected showing the man purchasing tools for action E," then the chain of evidence is incomplete. This is because the above evidence only establishes the precondition that the man performed action E in location D on date A, month B, day C, but does not prove that he actually performed action E. In this case, based on this chain of evidence, the verification result of the statement might be deemed insufficient evidence. If the chain of evidence includes "A shopping receipt was collected showing the man purchasing tools for action E" and "An elderly person stated on date A, month B, day C, day C, that the man appeared in location E," then there is a conflict regarding the location of the man's appearance. Since the evidence concerning the location of the man's appearance only provides a simple and approximate date, if the publicly known information indicates that location D and location E are far apart and difficult to reach within a day, then this indicates a contradiction between the chain of evidence and the statement. If the evidence is authentic and reliable, the verification result of the statement can be deemed falsified, meaning the statement is false information.

[0172] Optionally, the summary information is a summary of the verification process for the information to be verified. For example, the summary information is used to indicate which statements in the information to be verified are supported by what evidence, which statements are identified as false information by what evidence, and so on.

[0173] With the rise of Large Language Models (LLMs), automated fake information detection technology has undergone rapid evolution. Current mainstream technical solutions can be broadly categorized as follows: 1. Supervised Learning Models: This is the traditional mainstream approach, based on pre-trained language models such as BERT and RoBERTa, and constructs classifiers by fine-tuning them on a large number of labeled real / fake news datasets. The advantage of this type of method is its strong ability to fit data patterns within a specific domain. To address the data imbalance problem, subsequent research, such as the GSFND framework proposed by Tong et al., attempts to use LLM to generate additional fake news samples to expand the training set and combines reinforcement learning to dynamically adjust the sampling rate.

[0174] 2. Retrieval-Augmented Generation (RAG) Based Approaches: To address the issue of outdated or insufficient knowledge within the model, the RAG framework is widely used. These methods assist LLM in making judgments by retrieving relevant evidence from external knowledge bases (such as Wikipedia and news archives). For example, one approach uses the RAG framework to handle undiscovered false information, improving the accuracy and interpretability of judgments by providing external evidence. Recent research has further refined the application of RAG, such as RAEmoLLM, which innovatively uses sentiment information as a key retrieval method to find cases with similar sentiment expressions to aid in judgment.

[0175] 3. Multi-step Reasoning & Workflow: In order to simulate the complex human analytical process, researchers began to design workflows that include multiple steps.

[0176] Iterative retrieval: Some methods dynamically collect and evaluate evidence through multiple rounds of "retrieval-reasoning-re-retrieval" cycles to address the problem of insufficient information from a single retrieval.

[0177] Structured intelligent agent workflow: Some methods encapsulate the LLM into an intelligent agent that performs verification tasks step by step according to a pre-defined, expert-designed workflow (such as checking inflammatory language, verifying URL sources, and calling search engines).

[0178] This application constructs a multi-agent misinformation detection framework that simulates a human expert team collaboration model. This framework anthropomorphizes the complex misinformation identification task, achieving a highly interpretable, flexible, and self-evolving information detection process.

[0179] The key technical points are as follows: Team-based collaborative architecture: The information detection process is divided into three specialized teams: a fact statement extraction team, a fact verification team, and a summary and analysis team. Each team consists of one or more dedicated agents with clearly defined roles, working together to complete tasks.

[0180] 1. Layered Memory Mechanism: Introducing a two-tiered structure of shared memory (Global Status) and team-specific memory (Team Memory). Shared memory is used to synchronize the status of core tasks across teams (such as the verification progress of each statement), while team-specific memory ensures the focus and independence of work within each team, avoiding information interference.

[0181] 2. LLM-driven dynamic scheduling: Unlike workflows based on fixed rules, this application uses a top-level supervisor agent (i.e., management agent) as the "brain". Based on the understanding and reasoning ability of the large language model, it autonomously decides on the allocation, flow and termination of tasks, thereby achieving greater flexibility and adaptability to complex situations.

[0182] 3. Self-evolving Retrieval-Augmented Generation (RAG) Knowledge Base: This innovative approach proposes constructing a scalable knowledge base for both fake news and real news. This RAG knowledge base is automatically generated and accumulated by the framework during historical task processing. It includes verified factual statements, chains of evidence, writing styles, and other prior knowledge, providing a reference for detection in new tasks and enabling continuous system learning and performance optimization.

[0183] The related technologies have the following significant drawbacks: 1. Insufficient interpretability or lack of process transparency: Solutions based on traditional discriminative models (such as BERT) are like a "black box," only able to provide a final "yes / no" conclusion, without offering any basis for judgment or detailed verification process, making them difficult for users to accept. Although newer methods attempt to enable LLM to generate explanations, these explanations are often "post-hoc attributions" of the model itself, rather than actual, traceable verification steps.

[0184] This solution addresses the issue by simulating the investigation process of human experts, recording each step (statement extraction, evidence retrieval, and state update) in a shared Global Memory. The final analysis report is based on this transparent and traceable process log, achieving complete transparency of the process and high interpretability of the results.

[0185] 2. Lack of proactive and in-depth fact-verification processes: Early LLM schemes mainly relied on their internally fixed knowledge for "experience-based judgments," which were prone to "illusions" regarding recent events or specialized domain knowledge. Although the RAG framework solved some problems by introducing external retrieval, they often remained at the level of acquiring superficial evidence and lacked the ability to deeply address the quality of evidence and conflicts.

[0186] This solution addresses the following: The "fact verification team" in this embodiment is empowered to invoke external tools (such as search engines and databases) and proactively conduct multiple rounds of fact verification. More importantly, it is designed to handle complex evidentiary situations, such as identifying and flagging contradictory evidence, rather than simply accepting it. This ensures that the final judgment is based on carefully evaluated facts, rather than unverified search results.

[0187] 3. Rigid collaboration processes and lack of dynamic adaptability: In order to improve intelligence, research in related technologies has begun to adopt multi-step workflows or multi-agent systems, but their collaboration modes are often fixed and linear. For example, FactAgent relies on fixed workflows preset by experts. When encountering complex scenarios not covered by rules, these systems have difficulty coping flexibly and have poor robustness.

[0188] This solution addresses the following: This application introduces an LLM-driven Top-level Supervisor and Team Supervisor, which can dynamically and non-linearly intelligently schedule tasks based on real-time task progress in Global Memory. It can decide whether to continue verification, request supplementary evidence, transfer to the next team, or terminate the process early, thus possessing high flexibility and intelligence in dealing with unknown and complex situations.

[0189] 4. Disorganized memory management mechanism, low information density, and susceptibility to interference: A commonly overlooked problem in some current agent and workflow methods is the disorganized memory management. All information generated during system operation—including successful attempts, failed paths, intermediate considerations, and redundant retrieval results—is often mixed in the same memory pool. This results in valuable information being overwhelmed by noise, leading to extremely low information density. This not only affects decision-making efficiency but may also cause incorrect judgments due to interference from irrelevant information.

[0190] This solution addresses this issue by designing a hierarchical memory mechanism. We introduce Global Memory and Local Memory. Global Memory records only the state of core tasks (such as the final verification results of each statement), maintaining a high signal-to-noise ratio; while Local Memory is used to handle intermediate processes and "dirty work" in their respective subtasks, ensuring focus within the team and preventing irrelevant information from interfering with the decisions of other teams.

[0191] 5. Lack of a closed-loop mechanism for continuous learning and knowledge accumulation: Most solutions in related technologies are "one-off" tests, and their valuable process knowledge is lost after the task is completed. Even if some solutions use knowledge bases, there is a lack of an automated closed loop that feeds the results of one's own work back into the knowledge base.

[0192] This solution addresses the following: This application employs a bootstrap approach, building a self-evolving RAG knowledge base based on the framework itself. It automatically extracts, processes, and stores the structured data (such as false statements, valid evidence chains, conclusions, source URLs, etc.) accumulated in Global Memory from each successfully detected case into the knowledge base, forming a continuously growing prior knowledge asset. This allows the system to "get smarter with use," achieving faster response and higher accuracy in future tasks.

[0193] This application provides a simple, easy-to-use, and highly transparent information detection service.

[0194] 1. User input: Users submit content to be tested to the system, such as a piece of text, an article link, or an image.

[0195] 2. Automated Analysis: Upon receiving a request, the system automatically initiates its internal multi-agent collaborative analysis process. No user intervention is required.

[0196] 3. Output Analysis Report: After the analysis is complete, the system returns a detailed, structured analysis report to the user. This report includes: Overall conclusion: Comprehensive assessment of the tested content (e.g., highly credible, partially false, purely false information, unverifiable).

[0197] Core Statement Breakdown: A list of key factual statements (claims) extracted from the original text.

[0198] Verify details one by one: For each statement, show its verification status (confirmed / disproven / insufficient evidence), a summary of key evidence, and a link to the source.

[0199] Comprehensive Analysis: A summary document generated by the "Summary Analysis Team" that explains the complete logical chain leading to the final conclusion.

[0200] The technical implementation details of this application are as follows: Figure 5 As shown, its core is a non-linear, multi-agent collaborative workflow intelligently scheduled by a top-level supervisor. The entire process revolves around a dynamically updated global memory, rather than fixed linear steps. Its core workflow is as follows.

[0201] Core component description: Management Agent 10, acting as the top-level supervisor of the information verification system, is the system's "commander-in-chief" and a high-level agent. It does not directly execute tasks but instead autonomously decides which specialized team to activate next by observing the current state of Global Memory and combining it with preset system instructions and goals.

[0202] Specialized teams (Extraction / Verification / Writer Team): Each team is responsible for a specific sub-task. Each team may contain one or more agents that perform specific operations (such as extraction, verification, and writing). For example, the teams in this embodiment include a fact extraction team 20, a fact verification team 30, and a report generation team 40.

[0203] Execution Agents: The smallest working unit responsible for specific tasks. Each agent possesses its own local memory, used for thinking and temporarily storing information while executing tasks, and autonomously invoking tools and performing actions through the ReAct loop. For example, the fact extraction team 20 includes sub-management agents 21, extraction agents 22, and update agents 23. The fact verification team 30 includes sub-management agents 31, verification agents 32, and update agents 33. The report generation team 40 includes sub-management agents 41, writing agents 42, and update agents 43.

[0204] Global Memory: A task-level central state database, serving as a shared information dashboard for all teams. It records each claim extracted from the original text and its associated attributes in structured data format, such as ID, status, weight, evidence, URLs, and confidence. Collaboration among all teams is accomplished by reading and writing to Global Memory.

[0205] For example, the processing procedure of this system is as follows: Figure 6 As shown.

[0206] Step 1: Task reception and initialization.

[0207] Users submit content to be tested via an interface. The top-level supervisor receives the task and initializes a global shared memory (Global Status) to monitor the progress of the entire task.

[0208] Step 2: Claim Extraction.

[0209] Top-level oversight assigns tasks to the fact-telling extraction team.

[0210] The team consists of agents such as sub-supervisors, extractors, and updaters.

[0211] The Extractor Agent is responsible for reading the original text, identifying and extracting all verifiable objective factual statements (Claims), and assigning a weight to each specific statement in the content judgment.

[0212] The updater agent writes the extracted Claims list into the global shared memory (Global Status) and marks the initial status of each Claim as "pending verification".

[0213] Once completed, the team reports to the top-level supervisor, who then decides to proceed to the next step.

[0214] Step 3: Claim Verification.

[0215] The top-level supervisor passes on the task and shared memory access permissions containing the claims to be verified to the fact-verification team.

[0216] This team is the core of this framework, and its internal structure is as follows: Figure 7 As shown.

[0217] The sub-supervisor retrieves "claims to be verified" from the shared memory and assigns them to the verifier agent.

[0218] For a single claim, the verifier agent autonomously selects and uses a set of tools for multi-round (Attempt) verification via Tool Calling. The toolset may include: a search tool (search tool 34): calling a search engine to find relevant information; a scraping tool (scraping tool 35): scraping web page content to obtain detailed evidence; and experience (knowledge base 36): accessing the internal RAG disinformation knowledge base to query similar cases or styles.

[0219] Each piece of evidence collected by the verifier agent and its credibility are recorded.

[0220] After the verifier completes the verification of a claim, the updater agent updates the status of the claim in the shared memory (e.g., confirmed, falsified, insufficient evidence) and the associated chain of evidence.

[0221] This process is repeated until all claims are validated or the maximum number of attempts is reached.

[0222] Step 4: Summary Analysis and Report Generation.

[0223] Once all claims have been validated, the top-level supervisor will hand over the task to the summary and analysis team.

[0224] The team's writer agent reads the final state and the entire chain of evidence for all claims from the shared memory. The writer agent can also utilize appropriate toolsets for analysis and report writing. These toolsets may include: search tools 44, data scraping tools 45, and data processing tools 46.

[0225] Based on this information, the writer agent performs logical reasoning and comprehensive analysis to write a complete analysis report, including overall conclusions and the analysis process.

[0226] The Updater Agent writes the final report to the specified location.

[0227] Step 5: Task Completion and Knowledge Accumulation. After the top-level supervisor confirms the report generation, returns the results to the user, and terminates the task. Simultaneously, the system can trigger a background process to process and store valuable information such as verified false statements and newly discovered chains of evidence from this task into the RAG misinformation knowledge base, enabling the system's self-evolution.

[0228] The system provided in this application embodiment significantly improves interpretability: It outputs a detailed report containing evidence chains and analytical logic, rather than a simple "true / false" label, completely solving the "black box" problem of traditional information detection models and making the results credible and traceable. The detection process is highly flexible and intelligent: Through LLM-driven top-level supervisors for dynamic task scheduling, this application can intelligently handle various complex and unstructured false information content, exhibiting stronger robustness and adaptability than systems with fixed rules. It ensures the accuracy of factual judgments: By empowering the Agent to actively verify information using external tools, it ensures that conclusions are based on real-time, objective external evidence, rather than relying solely on outdated knowledge within the model, effectively avoiding the "illusion" problem of LLM. It achieves self-learning and evolution: The unique RAG knowledge base accumulation mechanism enables the system to continuously learn from historical tasks, constantly accumulating prior knowledge such as false information characteristics, factual evidence, and writing style, thus performing more efficiently and accurately in future detection tasks. It is highly modular and scalable: The team-based architecture design makes each module functionally cohesive, easy to optimize and upgrade independently. It allows for easy addition of new agent roles (such as image source tracing agents) or new tools to meet more diverse detection needs in the future. On real-world public datasets LIAR and LIAR2, it achieves state-of-the-art (SOTA) performance, with macro-F1 scores of 85.3% and 89% respectively, surpassing other methods by more than 6%.

[0229] Figure 8 This illustration shows a structural block diagram of an information verification device based on multi-agent collaboration provided in an exemplary embodiment of this application. The device is applied to a system comprising: a management agent and multiple execution agents; the device includes: a management agent module 410 and multiple execution agent modules 420.

[0230] The management intelligent agent module 410 is used to obtain the information to be verified; The management agent module 410 is used to schedule the multiple execution agents to perform information verification based on the information to be verified. The plurality of execution intelligent agent modules 420 are used to generate a verification report based on the information to be verified; the verification report includes at least two of the following information: The verification result of the information to be verified; Statement information, wherein the statement information is a proposition to be verified extracted from the information to be verified; The chain of evidence for the stated information, the chain of evidence including at least one piece of evidence information; The verification result of the stated information; Summary information, which is a summary of the verification process for the information to be verified.

[0231] In some embodiments, the plurality of execution agent modules include a fact extraction agent module, a fact verification agent module, and a report generation agent module; The fact extraction agent module is used to obtain at least one statement based on the information to be verified; The fact verification agent module is used to invoke at least one tool for each statement in the at least one statement to obtain the evidence chain of the statement and / or the verification result of the statement; The report generation agent module is used to generate a verification report based on at least one of the statement information, the evidence chain of the statement information, and the verification result of the statement information.

[0232] In some embodiments, the fact verification agent module is configured to, for the i-th statement in the at least one statement, invoke at least one tool to query and obtain verification information for the i-th statement, the verification information including at least one piece of evidence information, the source of the at least one piece of evidence information, and the confidence level of the at least one piece of evidence information; based on the verification information of the i-th statement, obtain the evidence chain of the i-th statement and / or the verification result of the i-th statement; let i = i + 1, and repeat the execution from the step of invoking at least one tool to query and obtain the verification information of the i-th statement until the verification of the at least one statement is completed, where i is a positive integer.

[0233] In some embodiments, the at least one tool includes at least one of the following: The search tool is used to search for at least one of the web pages, documents, and evidence information related to the stated information. Web scraping tools are used to scrape web pages and / or document content to obtain evidentiary information; A knowledge base is used to record content related to historical information that needs to be verified.

[0234] In some embodiments, the information verification system further includes a global shared memory and multiple local team memories, wherein each of the multiple local team memories corresponds one-to-one with the multiple executing agents; The fact extraction agent module is used to write the at least one statement into the global shared memory; The fact-verifying agent module is configured to: read the at least one statement from the global shared memory; invoke the at least one tool for each statement in the at least one statement to obtain at least one piece of evidence corresponding to each statement; write the at least one piece of evidence into the local team memory of the fact-verifying agent; determine the evidence chain and / or the verification result of each statement based on the at least one piece of evidence corresponding to each statement stored in the local team memory of the fact-verifying agent; and write the evidence chain and / or the verification result of each statement into the global shared memory. The report generation agent module is used to generate a verification report based on the information stored in the global shared memory.

[0235] In some embodiments, the management agent module 410 is used to schedule each of the plurality of execution agents based on the information stored in the global shared memory.

[0236] In some embodiments, the management agent module 410 is used for at least one of the following: When the fact extraction agent completes writing the at least one statement to the global shared memory, the fact verification agent is scheduled. When the number of evidence information in the evidence chain written by the fact-verifying agent to the global shared memory reaches a first quantity, the report-generating agent is scheduled. When the number of verification results written to the global shared memory by the fact-verifying agent reaches a second quantity, the report-generating agent is scheduled.

[0237] In some embodiments, the management agent module 410 is configured to initialize the global shared memory in response to the acquisition of the information to be verified.

[0238] In some embodiments, the management agent module 410 is used to update the knowledge base based on information in the global shared memory, the knowledge base being used to record content related to historical information to be verified.

[0239] In some embodiments, the fact extraction agent module is used to obtain at least one statement and the weights of each statement in the at least one statement based on the information to be verified, wherein the weights of each statement are used to indicate the importance of each statement in the information to be verified.

[0240] In some embodiments, the fact verification agent module is configured to invoke at least one tool for the i-th statement in the at least one statement to query the verification information of the i-th statement under multiple query conditions. The multiple query conditions are restrictions on querying the verification information, and the query conditions include at least one of the following aspects of restrictions: time range; evidence set; search engine; and modality of the verification information.

[0241] Figure 9 A structural block diagram of a multi-agent collaborative processing apparatus provided in an exemplary embodiment of this application is shown. The apparatus is applied to a system comprising: a management agent, multiple execution agents, a global shared memory, and multiple local team memories corresponding to the multiple execution agents. The apparatus includes: a management agent module 510 and multiple execution agent modules 520.

[0242] The management agent module 510 is used to initialize the global shared memory in response to the acquired task to be executed; The management agent module 510 is used to monitor the task status in the global shared memory and dynamically schedule the target agent from the multiple execution agents based on the changes in the task status. The target agent module among the plurality of execution agent modules 520 is used to respond to the scheduling, read task data from the global shared memory, perform processing using the corresponding local team memory, and write back the processing result and the updated task status to the global shared memory. The multiple execution agents complete the task to be executed through collaborative reading and writing of the globally shared memory.

[0243] In some embodiments, the management agent module 510 monitors the task status in the global shared memory; The management agent module 510 performs at least one of the following steps: When the task state in the global shared memory is the target state, the target agent is dynamically scheduled from the plurality of executing agents, and there is a correspondence between the target state and the target agent; When the task status in the global shared memory is that the first subtask has been completed and the score of the processing result of the first subtask is lower than the first score threshold, the target agent is rescheduled from the plurality of executing agents, and the target agent is used to execute the first subtask. When the task status in the global shared memory is the execution of the second subtask, and the feedback information of the first agent is read, including a cooperation request, the target agent is scheduled from the plurality of executing agents. The first agent is the executing agent that executes the second subtask, and the cooperation request is used to request the scheduling of other agents to assist in completing the second subtask. If the processing status of the third subtask indicated by the task status in the global shared memory meets the rescheduling conditions, the target agent is selected and scheduled from the plurality of execution agents; the rescheduling conditions include at least one of the following: the third subtask processing fails; the processing latency of the third subtask exceeds a first latency threshold; the processing progress of the third subtask is lower than a first progress threshold; the real-time load of the execution agent executing the third subtask is higher than a first load threshold.

[0244] In some embodiments, the management agent module 510, in response to the acquired task to be executed, splits the task to be executed into multiple sub-tasks; The management agent module 510 monitors the task status of each subtask in the global shared memory.

[0245] It should be noted that the apparatus provided in the above embodiments is only illustrated by the division of the above functional modules when implementing its functions. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the content structure of the device can be divided into different functional modules to complete all or part of the functions described above. In addition, the apparatus and method embodiments provided in the above embodiments belong to the same concept, and the specific implementation process can be found in the method embodiments, which will not be repeated here.

[0246] Figure 10 A structural block diagram of a server provided in an exemplary embodiment of this application is shown.

[0247] The server 1000 includes a central processing unit (CPU) 1001, a system memory 1004 including random access memory (RAM) 1002 and read-only memory (ROM) 1003, and a system bus 1005 connecting the system memory 1004 and the CPU 1001. The server 1000 also includes a basic input / output system (I / O system) 1006 to facilitate information transfer between various devices within the server, and a mass storage device 1007 for storing the operating system 1013, application programs 1014, and other program modules 1015.

[0248] The basic input / output system 1006 includes a display 1008 for displaying information and an input device 1009 for user input, such as a mouse or keyboard. Both the display 1008 and the input device 1009 are connected to the central processing unit 1001 via an input / output controller 1010 connected to the system bus 1005. The basic input / output system 1006 may also include the input / output controller 1010 for receiving and processing input from multiple other devices such as a keyboard, mouse, or electronic stylus. Similarly, the input / output controller 1010 also provides output to a display screen, printer, or other types of output devices.

[0249] The mass storage device 1007 is connected to the central processing unit 1001 via a mass storage controller (not shown) connected to the system bus 1005. The mass storage device 1007 and its associated computer-readable storage media provide non-volatile storage for the server 1000. That is, the mass storage device 1007 may include computer-readable storage media (not shown) such as a hard disk or a compact disc read-only memory (CD-ROM) drive.

[0250] Without loss of generality, the computer-readable storage medium may include computer storage media and communication media. Computer storage media include volatile and non-volatile, removable and non-removable media implemented using any method or technology for storing information such as computer-readable storage instructions, data structures, program modules, or other data. Computer storage media include RAM, ROM, erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other solid-state storage technologies, CD-ROM, digital versatile disc (DVD) or other optical storage, magnetic tape cassettes, magnetic tape, disk storage, or other magnetic storage devices. Of course, those skilled in the art will recognize that the computer storage medium is not limited to the above-mentioned types. The system memory 1004 and mass storage device 1007 described above can be collectively referred to as memory.

[0251] The memory stores one or more programs, which are configured to be executed by one or more central processing units 1001. The one or more programs contain instructions for implementing the above method embodiments, and the central processing unit 1001 executes the one or more programs to implement the methods provided by the various method embodiments described above.

[0252] According to various embodiments of this application, the server 1000 can also be connected to a remote computer device on a network, such as the Internet. That is, the server 1000 can be connected to the network 1012 via a network interface unit 1011 connected to the system bus 1005, or the network interface unit 1011 can be used to connect to other types of networks or remote computer device systems (not shown).

[0253] The memory further includes one or more programs stored in the memory, and the one or more programs include steps executed by the server in the method provided in the embodiments of this application.

[0254] Figure 11 A structural block diagram of a terminal provided in an exemplary embodiment of this application is shown.

[0255] The terminal 1100 can be a portable mobile terminal, also referred to as a mobile terminal in this embodiment. Examples include: smartphones, tablets, MP3 players (Moving Picture Experts Group Audio Layer III), and MP4 players (Moving Picture Experts Group Audio Layer IV). The terminal 1100 may also be referred to as user equipment, portable terminal, or other names.

[0256] Typically, terminal 1100 includes a processor 1101 and a memory 1102.

[0257] Processor 1101 may include one or more processing cores, such as a quad-core processor, an octa-core processor, etc. Processor 1101 may be implemented using at least one hardware form selected from DSP (Digital Signal Processing), FPGA (Field Programmable Gate Array), and PLA (Programmable Logic Array). Processor 1101 may also include a main processor and a coprocessor. The main processor, also known as a CPU (Central Processing Unit), is used to process data in the wake-up state; the coprocessor is a low-power processor used to process data in the standby state. In some embodiments, processor 1101 may integrate a GPU (Graphics Processing Unit), which is responsible for rendering and drawing the content to be displayed on the screen. In some embodiments, processor 1101 may also include an AI (Artificial Intelligence) processor, which is used to handle computational operations related to machine learning.

[0258] The memory 1102 may include one or more computer-readable storage media, which may be tangible and non-transitory. The memory 1102 may also include high-speed random access memory and non-volatile memory, such as one or more disk storage devices or flash memory devices. In some embodiments, the non-transitory computer-readable storage media in the memory 1102 are used to store at least one instruction, which is executed by the processor 1101 to implement the information verification method based on multi-agent cooperation provided in the embodiments of this application, and / or, the processing method based on multi-agent cooperation.

[0259] In some embodiments, the terminal 1100 may also optionally include: a peripheral device interface 1103 and at least one peripheral device. Specifically, the peripheral device includes at least one of: a radio frequency circuit 1104, a touch display screen 1105, a camera 1106, an audio circuit 1107, and a power supply 1108.

[0260] Peripheral device interface 1103 can be used to connect at least one I / O (Input / Output) related peripheral device to processor 1101 and memory 1102. In some embodiments, processor 1101, memory 1102 and peripheral device interface 1103 are integrated on the same chip or circuit board; in some other embodiments, any one or two of processor 1101, memory 1102 and peripheral device interface 1103 can be implemented on separate chips or circuit boards, which is not limited in this embodiment.

[0261] The radio frequency (RF) circuit 1104 is used to receive and transmit RF (Radio Frequency) signals, also known as electromagnetic signals. The RF circuit 1104 communicates with communication networks and other communication devices via electromagnetic signals. The RF circuit 1104 converts electrical signals into electromagnetic signals for transmission, or converts received electromagnetic signals back into electrical signals. Optionally, the RF circuit 1104 includes: an antenna system, an RF transceiver, one or more amplifiers, a tuner, an oscillator, a digital signal processor, a codec chipset, a user identity module card, etc. The RF circuit 1104 can communicate with other terminals through at least one wireless communication protocol. This wireless communication protocol includes, but is not limited to: the World Wide Web, metropolitan area networks, intranets, various generations of mobile communication networks (2G, 3G, 4G, and 5G), wireless local area networks, and / or WiFi (Wireless Fidelity) networks. In some embodiments, the RF circuit 1104 may also include circuitry related to NFC (Near Field Communication), which is not limited in this application.

[0262] The touch display screen 1105 is used to display a UI (User Interface). This UI may include graphics, text, icons, videos, and any combination thereof. The touch display screen 1105 also has the ability to collect touch signals on or above its surface. These touch signals can be input as control signals to the processor 1101 for processing. The touch display screen 1105 is used to provide virtual buttons and / or a virtual keyboard, also known as soft buttons and / or a soft keyboard. In some embodiments, there may be one touch display screen 1105, which is located on the front panel of the terminal 1100; in other embodiments, there may be at least two touch display screens, respectively located on different surfaces of the terminal 1100 or in a folded design; in some embodiments, the touch display screen 1105 may be a flexible display screen, located on a curved surface or folded surface of the terminal 1100. Furthermore, the touch display screen 1105 may be configured as a non-rectangular irregular shape, i.e., a non-rectangular screen. The touch display screen 1105 may be made of materials such as LCD (Liquid Crystal Display) or OLED (Organic Light-Emitting Diode).

[0263] The camera assembly 1106 is used to acquire images or videos. Optionally, the camera assembly 1106 includes a front-facing camera and a rear-facing camera. Typically, the front-facing camera is used for video calls or selfies, and the rear-facing camera is used for taking photos or videos. In some embodiments, there are at least two rear-facing cameras, which are any one of a main camera, a depth-sensing camera, and a wide-angle camera, to achieve background blurring by fusion of the main camera and the depth-sensing camera, and panoramic shooting and VR (Virtual Reality) shooting by fusion of the main camera and the wide-angle camera. In some embodiments, the camera assembly 1106 may also include a flash. The flash can be a single-color temperature flash or a dual-color temperature flash. A dual-color temperature flash is a combination of a warm light flash and a cool light flash, which can be used for light compensation at different color temperatures.

[0264] Audio circuit 1107 provides an audio interface between the user and terminal 1100. Audio circuit 1107 may include a microphone and a speaker. The microphone is used to collect sound waves from the user and the environment, converting the sound waves into electrical signals that are input to processor 1101 for processing, or input to radio frequency circuit 1104 for voice communication. For stereo sound acquisition or noise reduction purposes, multiple microphones may be used, each located at a different part of terminal 1100. The microphone may also be an array microphone or an omnidirectional microphone. The speaker converts the electrical signals from processor 1101 or radio frequency circuit 1104 into sound waves. The speaker may be a conventional diaphragm speaker or a piezoelectric ceramic speaker. When the speaker is a piezoelectric ceramic speaker, it can convert electrical signals not only into audible sound waves but also into inaudible sound waves for purposes such as distance measurement. In some embodiments, audio circuit 1107 may also include a headphone jack.

[0265] Power supply 1108 is used to power the various components in terminal 1100. Power supply 1108 can be AC ​​power, DC power, a disposable battery, or a rechargeable battery. When power supply 1108 includes a rechargeable battery, the rechargeable battery can be a wired rechargeable battery or a wireless rechargeable battery. A wired rechargeable battery is a battery that is charged via a wired line, and a wireless rechargeable battery is a battery that is charged via a wireless coil. The rechargeable battery can also be used to support fast charging technology.

[0266] In some embodiments, the terminal 1100 further includes one or more sensors 1109. The one or more sensors 1109 include, but are not limited to: an acceleration sensor 1110, a gyroscope sensor 1111, a pressure sensor 1112, an optical sensor 1113, and a proximity sensor 1114.

[0267] Accelerometer 1110 can detect the magnitude of acceleration along the three coordinate axes of a coordinate system established by terminal 1100. For example, accelerometer 1110 can be used to detect the components of gravitational acceleration along the three coordinate axes. Processor 1101 can control touchscreen 1105 to display the user interface in landscape or portrait view based on the gravitational acceleration signal acquired by accelerometer 1110. Accelerometer 1110 can also be used for collecting motion data from games or users.

[0268] The gyroscope sensor 1111 can detect the orientation and rotation angle of the terminal 1100. The gyroscope sensor 1111 can work in conjunction with the accelerometer sensor 1110 to collect the user's 3D movements on the terminal 1100. Based on the data collected by the gyroscope sensor 1111, the processor 1101 can perform the following functions: motion sensing (e.g., changing the UI based on the user's tilt), image stabilization during shooting, game control, and inertial navigation.

[0269] The pressure sensor 1112 can be disposed on the side bezel of the terminal 1100 and / or on the lower layer of the touch display screen 1105. When the pressure sensor 1112 is disposed on the side bezel of the terminal 1100, it can detect the user's grip signal on the terminal 1100 and perform left / right hand recognition or quick operation based on the grip signal. When the pressure sensor 1112 is disposed on the lower layer of the touch display screen 1105, it can control the operable controls on the UI interface based on the user's pressure operation on the touch display screen 1105. The operable controls include at least one of button controls, scroll bar controls, icon controls, and menu controls.

[0270] Optical sensor 1113 is used to collect ambient light intensity. In one embodiment, processor 1101 can control the display brightness of touch screen 1105 based on the ambient light intensity collected by optical sensor 1113. Specifically, when the ambient light intensity is high, the display brightness of touch screen 1105 is increased; when the ambient light intensity is low, the display brightness of touch screen 1105 is decreased. In another embodiment, processor 1101 can also dynamically adjust the shooting parameters of camera assembly 1106 based on the ambient light intensity collected by optical sensor 1113.

[0271] The proximity sensor 1114, also known as a distance sensor, is typically located on the front of the terminal 1100. The proximity sensor 1114 is used to detect the distance between the user and the front of the terminal 1100. In one embodiment, when the proximity sensor 1114 detects that the distance between the user and the front of the terminal 1100 is gradually decreasing, the processor 1101 controls the touchscreen display 1105 to switch from a screen-on state to a screen-off state; when the proximity sensor 1114 detects that the distance between the user and the front of the terminal 1100 is gradually increasing, the processor 1101 controls the touchscreen display 1105 to switch from a screen-off state to a screen-on state.

[0272] Those skilled in the art will understand that Figure 11 The structure shown does not constitute a limitation on terminal 1100 and may include more or fewer components than shown, or combine certain components, or use different component arrangements.

[0273] In an exemplary embodiment, this application provides a chip including programmable logic circuits and / or program instructions, which, when running on a computer device, are used to implement the information verification method based on multi-agent collaboration provided in the above method embodiments, and / or the processing method based on multi-agent collaboration.

[0274] In an exemplary embodiment, a non-transitory computer-readable storage medium is also provided, on which a computer program is stored, which, when executed by a processor, implements the above-described information verification method based on multi-agent cooperation and / or the processing method based on multi-agent cooperation.

[0275] In an exemplary embodiment, a computer program product is also provided, which, when executed by a processor, is used to implement the above-described information verification method based on multi-agent collaboration, and / or the processing method based on multi-agent collaboration.

[0276] It should be noted that the collection and processing of relevant data (including evidence information) in this application should strictly comply with the requirements of relevant national laws and regulations, obtain the informed consent or separate consent of the personal information subject, and carry out subsequent data use and processing within the scope of laws and regulations and the authorization of the personal information subject.

[0277] It should be understood that "multiple" as used herein refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. Furthermore, the step numbers described herein are merely illustrative of one possible execution order. In some other embodiments, the steps may not be executed in numerical order, such as two steps with different numbers being executed simultaneously, or two steps with different numbers being executed in the reverse order of the illustration. This application does not limit this.

[0278] The above description is merely an optional embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. An information verification system, characterized in that, The information verification system includes a management agent and multiple execution agents; The management agent is used to acquire information to be verified; and based on the information to be verified, to schedule the multiple execution agents to perform information verification. The plurality of execution agents are used to generate a verification report based on the information to be verified; the verification report includes at least two of the following information: The verification result of the information to be verified; Statement information, wherein the statement information is a proposition to be verified extracted from the information to be verified; The chain of evidence for the stated information, the chain of evidence including at least one piece of evidence information; The verification result of the stated information; Summary information, which is a summary of the verification process for the information to be verified.

2. The system according to claim 1, characterized in that, The plurality of execution agents include a fact extraction agent, a fact verification agent, and a report generation agent; The fact extraction agent is used to obtain at least one statement based on the information to be verified. The fact-verification agent is used to invoke at least one tool for each statement in the at least one statement to obtain the evidence chain of the statement and / or the verification result of the statement; The report generating agent is used to generate a verification report based on at least one of the statement information, the evidence chain of the statement information, and the verification result of the statement information.

3. The system according to claim 2, characterized in that, The fact verification agent is configured to, for the i-th statement in the at least one statement, invoke at least one tool to query and obtain the verification information of the i-th statement, wherein the verification information includes at least one piece of evidence information, the source of the at least one piece of evidence information, and the confidence level of the at least one piece of evidence information; and based on the verification information of the i-th statement, obtain the evidence chain of the i-th statement and / or the verification result of the i-th statement. Let i = i + 1, and repeat the step of calling at least one tool to query and obtain the verification information of the i-th statement information until the verification of at least one statement information is completed, where i is a positive integer.

4. The system according to claim 2 or 3, characterized in that, The at least one tool includes at least one of the following: The search tool is used to search for at least one of the web pages, documents, and evidence information related to the stated information. Web scraping tools are used to scrape web pages and / or document content to obtain evidentiary information; A knowledge base is used to record content related to historical information that needs to be verified.

5. The system according to any one of claims 2 to 4, characterized in that, The information verification system also includes a global shared memory and multiple local team memories, with each of the multiple local team memories corresponding one-to-one with the multiple executing intelligent agents; The fact extraction agent is used to write the at least one statement into the global shared memory; The fact-verifying agent is configured to read the at least one statement from the global shared memory; for each statement in the at least one statement, invoke the at least one tool to obtain at least one piece of evidence corresponding to each statement; write the at least one piece of evidence into the local team memory of the fact-verifying agent; and, based on the at least one piece of evidence corresponding to each statement stored in the local team memory of the fact-verifying agent, determine the evidence chain corresponding to each statement and / or the verification result of each statement. Write the evidence chain corresponding to each statement and / or the verification result of each statement into the global shared memory; The report-generating agent is used to generate a verification report based on the information stored in the global shared memory.

6. The system according to claim 5, characterized in that, The management agent is used to schedule each of the multiple execution agents based on the information stored in the global shared memory.

7. The system according to claim 6, characterized in that, The management agent is used for at least one of the following: When the fact extraction agent completes writing the at least one statement to the global shared memory, the fact verification agent is scheduled. When the number of evidence information in the evidence chain written by the fact-verifying agent to the global shared memory reaches a first quantity, the report-generating agent is scheduled. When the number of verification results written to the global shared memory by the fact-verifying agent reaches a second quantity, the report-generating agent is scheduled.

8. The system according to any one of claims 5 to 7, characterized in that, The management agent is used to initialize the global shared memory in response to the acquisition of the information to be verified.

9. The system according to any one of claims 5 to 7, characterized in that, The management agent is used to update the knowledge base based on the information in the global shared memory. The knowledge base is used to record content related to historical information to be verified.

10. The system according to any one of claims 2 to 9, characterized in that, The fact extraction agent is used to obtain at least one statement and the weight of each statement in the at least one statement based on the information to be verified. The weight of each statement is used to indicate the importance of each statement in the information to be verified.

11. The system according to any one of claims 3 to 9, characterized in that, The fact-verifying agent is used to invoke at least one tool to query the verification information of the i-th statement in the at least one statement, respectively, under multiple query conditions. The multiple query conditions are restrictions on querying the verification information, and the query conditions include at least one of the following aspects: time range; evidence set; search engine; and modality of verification information.

12. An information verification method based on multi-agent cooperation, characterized in that, The method is applied to a system comprising: a management agent and multiple execution agents; The management agent acquires the information to be verified. The management agent, based on the information to be verified, schedules the multiple execution agents to perform information verification; The plurality of execution agents generate a verification report based on the information to be verified; the verification report includes at least two of the following information: The verification result of the information to be verified; Statement information, wherein the statement information is a proposition to be verified extracted from the information to be verified; The chain of evidence for the stated information, the chain of evidence including at least one piece of evidence information; The verification result of the stated information; Summary information, which is a summary of the verification process for the information to be verified.

13. A processing method based on multi-agent cooperation, characterized in that, The method is applied to a system comprising: a management agent, multiple executive agents, a global shared memory, and multiple local team memories corresponding to the multiple executive agents; the method includes: In response to the acquired task to be executed, the management agent initializes the global shared memory; The management agent monitors the task status in the global shared memory and dynamically schedules the target agent from the multiple execution agents based on changes in the task status; The target agent responds to the scheduling by reading task data from the global shared memory, performing processing using the corresponding local team memory, and writing the processing result and the updated task status back to the global shared memory. The multiple execution agents complete the task to be executed through collaborative reading and writing of the globally shared memory.

14. The method according to claim 13, characterized in that, The management agent monitors the task status in the global shared memory and dynamically schedules the target agent from the plurality of execution agents based on changes in the task status, including: The management agent monitors the task status in the global shared memory; The management agent performs at least one of the following steps: When the task state in the global shared memory is the target state, the target agent is dynamically scheduled from the plurality of executing agents, and there is a correspondence between the target state and the target agent; When the task status in the global shared memory is that the first subtask has been completed and the score of the processing result of the first subtask is lower than the first score threshold, the target agent is rescheduled from the plurality of executing agents, and the target agent is used to execute the first subtask. When the task status in the global shared memory is the execution of the second subtask, and the feedback information of the first agent is read, including a cooperation request, the target agent is scheduled from the plurality of executing agents. The first agent is the executing agent that executes the second subtask, and the cooperation request is used to request the scheduling of other agents to assist in completing the second subtask. If the processing status of the third subtask indicated by the task status in the global shared memory meets the rescheduling conditions, the target agent is selected and scheduled from the plurality of execution agents; the rescheduling conditions include at least one of the following: the third subtask processing fails; the processing latency of the third subtask exceeds a first latency threshold; the processing progress of the third subtask is lower than a first progress threshold; the real-time load of the execution agent executing the third subtask is higher than a first load threshold.

15. The method according to claim 13 or 14, characterized in that, The method further includes: In response to the acquired task to be executed, the management agent breaks down the task to be executed into multiple sub-tasks; The management agent monitors the task status in the global shared memory, including: The management agent monitors the task status of each subtask in the global shared memory.

16. An information verification device, characterized in that, The apparatus is applied to a system comprising: a management agent and multiple execution agents; the apparatus includes: The management intelligent agent module is used to obtain the information to be verified; The management agent module is used to schedule the multiple execution agents to perform information verification based on the information to be verified. The plurality of execution intelligent agent modules are used to generate a verification report based on the information to be verified; the verification report includes at least two of the following information: The verification result of the information to be verified; Statement information, wherein the statement information is a proposition to be verified extracted from the information to be verified; The chain of evidence for the stated information, the chain of evidence including at least one piece of evidence information; The verification result of the stated information; Summary information, which is a summary of the verification process for the information to be verified.

17. A processing device based on multi-agent cooperation, characterized in that, The device is applied to a system comprising: a management agent, multiple executive agents, a global shared memory, and multiple local team memories corresponding to the multiple executive agents; the device includes: The management agent module is used to initialize the global shared memory in response to the acquired task to be executed; The management agent module is used to monitor the task status in the global shared memory and dynamically schedule the target agent from the multiple execution agents based on the changes in the task status. The target intelligent agent is used to respond to the scheduling, read task data from the global shared memory, perform processing using the corresponding local team memory, and write back the processing result and the updated task status to the global shared memory. The multiple execution agents complete the task to be executed through collaborative reading and writing of the globally shared memory.

18. A computer device, characterized in that, The computer device includes a processor and a memory, the memory storing a computer program, the computer program being loaded and executed by the processor to implement the information verification method based on multi-agent cooperation as described in claim 12, and / or the processing method based on multi-agent cooperation as described in any one of claims 13 to 15.

19. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which is loaded and executed by a processor to implement the information verification method based on multi-agent cooperation as described in claim 12, and / or the processing method based on multi-agent cooperation as described in any one of claims 13 to 15.

20. A computer program product, characterized in that, The computer program product includes a computer program stored in a computer-readable storage medium, and a processor reads and executes the computer program from the computer-readable storage medium to implement the information verification method based on multi-agent cooperation as described in claim 12, and / or the processing method based on multi-agent cooperation as described in any one of claims 13 to 15.

Images