System for generating a science brief

Abstract

The application relates to a system for generating a science and technology brief, comprising a scheduling agent, a plurality of execution agents and a quality inspection agent; the scheduling agent divides a received science and technology brief generation task into a plurality of subtasks and distributes the subtasks to the plurality of execution agents; the execution agents execute the subtasks in sequence and transmit the generated execution results to the execution agents corresponding to the next subtasks and the quality inspection agent; the quality inspection agent checks the execution results based on a preset rule set; when the execution results cannot pass the check, the quality inspection agent generates a reflection instruction and transmits the reflection instruction to the scheduling agent; the scheduling agent determines a target subtask to be re-executed based on the reflection instruction and re-distributes the execution agents corresponding to the target subtask; and the scheduling agent generates the science and technology brief based on the execution results in response to reaching a preset termination condition.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence, and in particular to a system for generating scientific and technological briefings. Background Technology

[0002] With the continuous development of artificial intelligence and natural language processing technologies, the automatic generation of science and technology briefings using large language models has been gradually applied to scenarios such as summarizing science and technology information, analyzing industry trends, tracking scientific research results, and managing corporate intelligence. Existing solutions typically use a single model or multiple agents working together to complete tasks such as information collection, data extraction, content generation, and formatting, thereby improving the efficiency of science and technology briefing generation.

[0003] However, in existing technologies, multiple agents typically complete their tasks sequentially according to a pre-defined process, with subsequent processing heavily reliant on the output of the previous stage. When the content generated at a certain stage contains factual errors, outdated data, insufficient information sources, or content that does not meet requirements, these errors can easily propagate in subsequent tasks, ultimately leading to problems such as insufficient factual accuracy, poor logical consistency, and low credibility in the generated technical briefings.

[0004] On the other hand, while some existing solutions introduce result verification mechanisms, most rely on the verification module to directly modify the generated results or only perform unified checks at the final output stage, lacking dynamic feedback and rescheduling capabilities for the task execution process. When problems are found in the execution results, it is often necessary to re-execute the entire generation process or rely on manual intervention for correction, which not only increases resource consumption but also reduces the efficiency of technology briefing generation. Therefore, how to construct a technology briefing generation system that can achieve dynamic verification, reflective feedback, and rescheduling during task execution has become an urgent technical problem to be solved in this field. Summary of the Invention

[0005] To improve the accuracy, logic, and reliability of technical briefings, this application proposes a system for generating technical briefings that can dynamically reflect and verify.

[0006] The system includes a scheduling agent, multiple execution agents, and a quality inspection agent. The scheduling agent divides the received technology briefing generation task into multiple subtasks and assigns them to the multiple execution agents. The execution agents execute the multiple subtasks sequentially and transmit the generated execution results to the execution agent and the quality inspection agent corresponding to the next subtask. The quality inspection agent verifies the execution results based on a preset rule set, which includes multi-source information consistency rules, data timeliness rules, and sensitive word filtering rules. When the execution result fails the verification, the quality inspection agent generates a reflection instruction and transmits it to the scheduling agent. The scheduling agent determines the target subtask to be re-executed based on the reflection instruction and reassigns the execution agent corresponding to the target subtask. In response to reaching a preset termination condition, the scheduling agent generates a technology briefing based on the execution results and records the workflow information and verification results corresponding to the technology briefing generation task for updating the scheduling agent's task segmentation strategy or task template library.

[0007] Optionally, the scheduling agent divides the received technology briefing generation task into multiple sub-tasks based on a preset task template library, wherein the preset task template library includes multiple task templates for different technology fields.

[0008] Optionally, the plurality of subtasks includes at least two of the following: hot topic information collection subtask, data extraction subtask, opinion summarization subtask, report writing subtask, and formatting subtask.

[0009] Optionally, the multi-source information consistency rules include:

[0010] For the target factual information in the execution result, corresponding factual data are obtained from multiple information sources;

[0011] Consistency indicators are determined based on factual data from the multiple information sources;

[0012] The consistency index is used to determine whether the target factual information passes the verification.

[0013] Optionally, the data timeliness rules include:

[0014] Retrieve the data publication time corresponding to the data referenced in the execution result;

[0015] Compare the data release time with the current time;

[0016] When the time difference exceeds the preset timeliness threshold, the execution result is determined to have failed the verification.

[0017] Optionally, the rethinking instruction includes at least one of the following: a verification failure location identifier; a verification failure type; a verification failure reason; and optimization suggestions generated based on the verification failure reason.

[0018] Optionally, the preset termination condition includes at least one of the following:

[0019] All execution results passed the verification.

[0020] The number of times the scheduling agent replans the target sub-task based on the reflection instruction reaches a preset threshold;

[0021] Received manual termination instruction.

[0022] Compared with existing technologies, the system for generating technical briefings provided in this application has at least the following advantages:

[0023] (1) This application sets up a scheduling agent, multiple execution agents and a quality inspection agent, and establishes a collaborative working mechanism that combines task scheduling, result verification and reflection feedback. When the quality inspection agent finds a problem with the execution result, it generates a reflection instruction, and the scheduling agent re-plans the target sub-task and reassigns the corresponding execution agent, instead of directly modifying the execution result. This enables closed-loop feedback control in the process of generating the science and technology briefing, and improves the flexibility and scalability of the task execution process.

[0024] (2) By introducing a verification mechanism based on multi-source information consistency rules, data timeliness rules and sensitive word filtering rules in each execution stage, this application can promptly detect problems such as insufficient factual basis, outdated cited data and non-compliance with requirements during the generation of science and technology briefings, reduce the spread of erroneous information to subsequent processing procedures, and improve the accuracy, reliability and content quality of the generated results.

[0025] (3) This application records the workflow information and verification results corresponding to the task of generating the science and technology briefing, and uses the recorded information to update the task decomposition strategy or task template library of the scheduling agent, so that the system can continuously optimize the subsequent task scheduling method according to the historical execution situation, improve the task planning ability and processing efficiency in the process of generating science and technology briefings in different science and technology fields, and enhance the system's adaptability to different application scenarios. Attached Figure Description

[0026] Figure 1 This is a schematic diagram of a system for generating scientific and technological briefings according to an embodiment of this application.

[0027] Figure 2 This is a flowchart of the system execution for generating a technical briefing according to the embodiments of this application. Detailed Implementation

[0028] The system for generating scientific and technological briefings provided in this application will be further described below with reference to the accompanying drawings and specific embodiments. It should be understood that the following embodiments are only used to explain this application and are not intended to limit the scope of protection of this application.

[0029] Figure 1 This illustration shows a schematic diagram of a system architecture for generating scientific and technological briefings according to an embodiment of this application. Figure 2 This illustration shows a workflow diagram for generating scientific and technological briefings according to an embodiment of this application.

[0030] like Figure 1 As shown, the system provided in this application mainly includes a scheduling agent 1, multiple execution agents 2, and a quality inspection agent 3. The scheduling agent receives the technology briefing generation task and performs task segmentation and scheduling. The multiple execution agents execute the sub-tasks assigned by the scheduling agent according to the sequential relationship between the sub-tasks. The quality inspection agent verifies the execution results generated by each execution agent according to a preset rule set, and generates a reflection instruction to feed back to the scheduling agent when the verification fails, so that the scheduling agent can replan the target sub-tasks and reassign the corresponding execution agents.

[0031] In this embodiment, the scheduling agent establishes task allocation relationships with multiple execution agents, and the execution agents establish result transmission relationships based on task dependencies. Each execution agent also establishes a result verification relationship with the quality inspection agent. When the quality inspection agent determines that the execution result meets the preset rules, the scheduling agent continues to execute subsequent tasks; when the quality inspection agent determines that the execution result does not meet the preset rules, it generates a reflection instruction and returns it to the scheduling agent, which then redetermines the target sub-tasks to be executed and reassigns tasks, thereby forming a closed-loop collaborative mechanism of task scheduling, task execution, result verification, reflection feedback, and rescheduling.

[0032] The following is combined Figure 2 right Figure 1 The technical briefing generation process executed by the system shown is described in detail.

[0033] In step S101, the scheduling agent 1 receives the technology briefing generation task and divides the technology briefing generation task into multiple sub-tasks. It can be understood that the division here includes both dividing the overall task into multiple sequentially executed steps and planning the task content of different sub-tasks during the division.

[0034] In this embodiment, the technology briefing generation task can be input by a user terminal, or it can originate from a preset scheduled task or be invoked by other business systems. After receiving the technology briefing generation task, the scheduling agent parses the task and completes task segmentation according to the preset task template library.

[0035] The preset task template library can include multiple task templates for different technology fields, such as templates for artificial intelligence, semiconductors, biomedicine, and new energy. When a task to generate a technology briefing in a corresponding field is received, the scheduling agent can match the appropriate task template and break down the task.

[0036] In one embodiment, the scheduling agent can, based on a preset task template, divide the technology briefing generation task into at least two of the following sub-tasks for a specific technology field: hot topic information collection, data extraction, viewpoint summarization, report writing, and formatting. Those skilled in the art will understand that, depending on actual business needs, it can also be divided into other types of sub-tasks, and this application does not limit this division.

[0037] It should be noted that the above-mentioned sub-tasks for collecting hot information, extracting data, summarizing viewpoints, writing reports, and formatting are only examples. In actual applications, other sub-tasks such as literature retrieval, chart generation, abstract generation, and risk analysis can be added according to the type and requirements of the scientific briefing. Correspondingly, the number and functional division of the executing agents can also be adjusted according to specific application needs, and are not limited here.

[0038] In step S102, the scheduling agent assigns multiple subtasks to the corresponding execution agents, which then execute the subtasks in sequence and transmit the execution results to the execution agent and quality inspection agent corresponding to the next subtask.

[0039] In this embodiment, different executing agents can have different processing capabilities, or they can be configured with different prompts from the same basic model to form different roles. This application does not limit this.

[0040] During execution, each execution agent executes tasks sequentially according to the dependencies between multiple subtasks. For example, after the hot topic information collection subtask is completed, an information set containing news links, paper titles, publication dates, etc., is generated and transmitted to the execution agent responsible for data extraction for further processing. Simultaneously, the hot topic information collection results are also sent to the quality control agent for verification. Similarly, after data extraction is completed, the execution results are sent to both the viewpoint summarization execution agent and the quality control agent; the viewpoint summarization results are sent to both the report writing execution agent and the quality control agent; and the report writing results are sent to both the formatting execution agent and the quality control agent, thus achieving phased quality control during the execution process.

[0041] In step S103, the quality inspection intelligent agent verifies the execution result based on a preset rule set.

[0042] In this embodiment, the preset rule set includes multi-source information consistency rules, data timeliness rules, and sensitive word filtering rules.

[0043] Among them, the multi-source information consistency rule is used to verify whether the target factual information in the execution result is supported by multiple information sources. For example, when the execution result contains the target factual information "a certain company has released a new generation of artificial intelligence chips," the quality inspection agent can access multiple information sources such as official website announcements, industry media reports, and public databases to obtain corresponding factual data, and determine the consistency index based on the degree of consistency among multiple information sources. When the consistency index reaches a preset threshold, the target factual information is deemed to have passed the verification; otherwise, the target factual information is deemed to have insufficient credibility.

[0044] Furthermore, data timeliness rules are used to determine whether the data referenced in the execution result meets the time requirements. For example, for a task that requires generating a briefing on science and technology developments over the past seven days, the quality inspection agent obtains the publication time corresponding to the referenced data and compares the publication time with the current time. When the time difference exceeds the preset timeliness threshold (seven days), it is determined that the data does not meet the timeliness requirements, and the corresponding execution result fails the verification.

[0045] Sensitive word filtering rules are used to detect whether the execution result contains preset sensitive words or expressions that do not conform to the specifications. When such content is detected, the current execution result can also be determined to have failed the validation.

[0046] In step S104, it is determined whether the execution result can pass the verification.

[0047] When the execution result fails the verification, in step S105, the quality inspection agent generates a reflection instruction and sends it to the scheduling agent.

[0048] In this embodiment, the rethinking instruction may include at least one of the following: a verification failure location identifier, a verification failure type, a verification failure reason, and optimization suggestions generated based on the verification failure reason.

[0049] For example, when the execution result has insufficient information sources, the reflection instruction can include the corresponding location identifier and optimization suggestions for "supplementing information sources"; when the execution result has the problem of referencing data exceeding the preset time range, the reflection instruction can include the corresponding location identifier and optimization suggestions for "updating data"; when the execution result has logical inconsistencies or non-standard expressions, corresponding optimization suggestions can also be generated.

[0050] It should be noted that in this application, the quality inspection agent does not directly modify the execution result, but generates reflection instructions and feeds them back to the scheduling agent, which then uniformly replans and reschedules the process.

[0051] Then, the scheduling agent determines the target subtask to be re-executed based on the reflection instructions, and can reallocate the execution agent corresponding to the target subtask.

[0052] In this embodiment, after receiving the reflection instruction, the scheduling agent can determine the target subtask that needs to be re-executed based on the location and reason of the verification failure, and re-execute the aforementioned subtask splitting and allocation steps.

[0053] In one embodiment, it is not necessary to re-execute all subtasks. For example, when a verification failure occurs during the hotspot information collection phase, the scheduling agent can re-plan the hotspot information collection subtask when splitting the task; when a verification failure occurs during the data extraction phase, only the data extraction subtask can be re-planned. However, when multiple related subtasks are affected, multiple target subtasks can also be re-planned simultaneously, and the corresponding execution agents can be reallocated to continue execution, thereby reducing redundant processing and improving the efficiency of technology briefing generation.

[0054] The execution result after re-execution is sent to the corresponding next execution agent and quality inspection agent for processing. Steps S101 to S105 are repeated until the preset termination condition is met.

[0055] In step S106, in response to the achievement of the preset termination condition, the scheduling agent generates a technology briefing based on the execution result and records the workflow information and verification result corresponding to the technology briefing generation task.

[0056] In this embodiment, the preset termination condition may include at least one of the following: all execution results pass the verification; the number of times the scheduling agent replans the target sub-task based on the reflection instruction reaches a preset number threshold; or a manual termination instruction is received.

[0057] After generating the final technology briefing, the scheduling agent records the workflow information and verification results corresponding to this task, and uses the workflow information and verification results to update the task splitting strategy or task template library of the scheduling agent, so as to optimize the task splitting method and task scheduling strategy in subsequent similar technology briefing generation tasks, thereby improving the overall processing efficiency of the system and the quality of technology briefing generation.

[0058] It should be noted that, Figure 2 The process shown is only a preferred embodiment of this application. Those skilled in the art can adjust the execution order of some steps or add or remove some steps according to specific application scenarios without departing from the protection scope of this application.

[0059] The specific embodiments described above illustrate the implementation of this application. Those skilled in the art can easily understand other advantages and effects of this application from the content disclosed in this specification. Although the description of this application is presented in conjunction with preferred embodiments, this does not mean that the features of this invention are limited to these embodiments. Furthermore, to avoid confusion or obscuring the focus of this application, some specific details will be omitted in the description. It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other.

[0060] Furthermore, the various operations will be described as multiple discrete operations in a manner most conducive to understanding the illustrative embodiments; however, the order of description should not be construed as implying that these operations must depend on the order. In particular, these operations do not need to be performed in the order presented.

[0061] Unless the context otherwise specifies, the terms “contains,” “has,” and “includes” are synonyms. The phrase “A / B” means “A or B.” The phrase “A and / or B” means “(A and B) or (A or B).”

[0062] As used herein, the terms “module” or “unit” may refer to, be, or include: application-specific integrated circuits (ASICs), electronic circuits, (shared, dedicated, or group) processors and / or memories that execute one or more software or firmware programs, combinational logic circuits, and / or other suitable components that provide the described functionality.

[0063] In the accompanying drawings, certain structural or methodological features are shown in a specific arrangement and / or order. However, it should be understood that such a specific arrangement and / or order may not be necessary. In some embodiments, these features may be arranged in a manner and / or order different from that shown in the illustrative drawings. Furthermore, the inclusion of structural or methodological features in a particular figure does not imply that such features are required in all embodiments, and in some embodiments, these features may be omitted or may be combined with other features.

[0064] It should be understood that although terms such as "first," "second," etc., may be used herein to describe various units or data, these units or data should not be limited by these terms. These terms are used merely to distinguish one feature from another. For example, without departing from the scope of the exemplary embodiments, a first feature may be referred to as a second feature, and similarly, a second feature may be referred to as a first feature.

[0065] It should be noted that in this specification, similar reference numerals and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0066] Although the invention has been illustrated and described with reference to certain preferred embodiments thereof, those skilled in the art will understand that various changes in form and detail may be made therein without departing from the spirit and scope of the invention.

Claims

1. A system for generating scientific and technological briefings, characterized in that, This includes a scheduling agent, multiple execution agents, and a quality inspection agent; The scheduling agent divides the received technology briefing generation task into multiple sub-tasks and assigns them to the multiple execution agents; The execution agent executes the multiple subtasks in sequence and transmits the generated execution results to the execution agent and the quality inspection agent corresponding to the next subtask. The quality inspection intelligent agent verifies the execution result based on a preset rule set, which includes multi-source information consistency rules, data timeliness rules, and sensitive word filtering rules. When the execution result fails the verification, the quality inspection agent generates a reflection instruction and transmits it to the scheduling agent. The scheduling agent determines the target subtask to be re-executed based on the reflection instruction, and reallocates the execution agent corresponding to the target subtask; as well as In response to the achievement of a preset termination condition, the scheduling agent generates a technology briefing based on the execution result, and records the workflow information and verification results corresponding to the technology briefing generation task for updating the task splitting strategy or task template library of the scheduling agent.

2. The system according to claim 1, characterized in that, The scheduling agent divides the received technology briefing generation task into multiple sub-tasks based on a preset task template library, wherein the preset task template library includes multiple task templates for different technology fields.

3. The system according to claim 2, characterized in that, The multiple subtasks include at least two of the following: hot topic information collection, data extraction, opinion summarization, report writing, and formatting.

4. The system according to claim 3, characterized in that, The multi-source information consistency rules include: For the target factual information in the execution result, corresponding factual data are obtained from multiple information sources; Consistency indicators are determined based on factual data from the multiple information sources; The consistency index is used to determine whether the target factual information passes the verification.

5. The system according to claim 4, characterized in that, The data timeliness rules include: Retrieve the data publication time corresponding to the data referenced in the execution result; Compare the data release time with the current time; When the time difference exceeds the preset timeliness threshold, the execution result is determined to have failed the verification.

6. The system according to claim 5, characterized in that, The reflection instruction includes at least one of the following: Verification failure location identifier; Verification failure types; Reason for verification failure; Optimization suggestions generated based on the reasons for verification failure.

7. The system according to claim 6, characterized in that, The preset termination condition includes at least one of the following: All execution results passed the verification. The number of times the scheduling agent replans the target sub-task based on the reflection instruction reaches a preset threshold; Received manual termination instruction.

Images