Intelligent intelligence reorganizing method of enhanced context window based on multi-intelligence cooperation

Abstract

The invention provides an intelligent intelligence reorganizing method of an enhanced context window based on multi-intelligence cooperation. The method comprises the following steps: firstly, performing theme classification; and the Master agent allocates a theme to each group of Slaver agents, each group of Slaver agents obtains information from the materials, and the information is returned to the Master agent for repeated information conflict resolution. When the repeated information threshold value is lower than a preset value, the Master agent guides each Slaver agent to generate an intelligence result; and ending the output of a certain type of theme intelligence achievement, giving a final response according to the requirement of the prompt template, and when all the Slaver intelligent agents receive an editing completion instruction, completing a complete editing process. According to the method, the tasks are split according to the subject terms, and the compiling tasks are processed by the Slavers of different subjects, so that the context window utilization rate of a large language model is improved, and data can be completely and stably processed when the intelligence material quantity is relatively large.

Description

Technical Field

[0001] The invention belongs to the field of intelligent intelligence, and relates to an intelligent intelligence compilation method based on multi-intelligent collaboration enhanced context window. Background Art

[0002] In the field of intelligent intelligence, the existing intelligent intelligence compilation has undergone a transformation from manual compilation to machine-assisted compilation. The current machine-assisted compilation generally implements specific functions based on technologies such as data mining, graph analysis, and machine learning. Users need to select and adjust the corresponding tools according to specific intelligence compilation tasks and needs. However, in this case, there is a lack of linkage between the auxiliary tools for intelligence compilation, and it is still necessary to rely on manual compilation to connect intelligence from multiple sources, and the existing auxiliary tools have poor information processing capabilities when the amount of material is large. This is mainly because the machine-assisted compilation process lacks a highly intelligent scheduling center to process the information generated by the collaboration of multiple application tools, and the traditional machine learning method relies on the call of a large language model to process the context window of information, which is not competent for the intelligent compilation of larger batches of data information. Summary of the invention

[0003] 1. Technical problems to be solved: Existing intelligent intelligence compilation methods cannot effectively and collaboratively process data from multiple intelligent processing units, and the context window for accepting information input relies on the large language model that is called, making it incapable of intelligent compilation of larger batches of data information.

[0004] 2. Technical solution: In order to solve the above problems, the present invention provides an intelligent intelligence compilation method based on multi-intelligent collaboration enhanced context window, which includes the following steps.

[0005] Step S01: Through intelligence templates and reporting keywords, the system uses word embedding to classify topics.

[0006] Step S02: Create a Master agent and multiple groups of Slaver agents. The Master agent assigns topics to each group of Slaver agents based on topic classification. The Slaver agents will perform specific intelligent reporting tasks based on the specific keywords assigned to them.

[0007] Step S03: After the topic assignment is completed, each group of Slaver agents obtains information from the materials for the topic and returns it to the Master agent to resolve duplicate information conflicts, ensuring that the source of the materials for the results generated by each group of Slaver agents is fully covered.

[0008] Step S04: When the Master agent finds that the duplicate information threshold is lower than the preset value, the Master agent instructs each group of Slaver agents to generate intelligence results; the Slaver agent ends the output of a certain type of subject intelligence results, and the Master agent gives a final response according to the prompt template requirements, which is used to command the Slaver to regenerate the results or complete the reporting of the assigned subject. When all Slaver agents receive the completion report instruction, a complete reporting process is completed.

[0009] The Master agent is capable of intelligent subject classification, automatic related information retrieval, intelligent generation, and intelligent integration.

[0010] In step S01, the intelligence template defines the output format of the final report.

[0011] In step S02, the specific intelligent reporting task compiles relevant information from the perspectives of multiple different subject terms.

[0012] Each of the Slaver agents can fully utilize the full context window of the large language model application when processing the topic subtask.

[0013] In step S03, the conflict is that different Slaver agents use the same source to generate information on different topics. The Master agent uses the following method to resolve the conflict of duplicate information: the Master agent uses text similarity and source tracking methods to summarize the information and compare it with the source of the material. Then the Master agent instructs the Slaver agents with duplicate information conflicts to use different sources of material to generate intelligence.

[0014] In step S04, the method by which the Master agent guides each group of Slaver agents to generate intelligence results is as follows: for a keyword, after the Master finds that the Slaver who is responsible for compiling a certain type of subject report has used a certain type of material, it will guide the Slaver who is responsible for compiling another type of subject report to avoid using the previous material, thereby ensuring high information coverage.

[0015] 3. Beneficial effects: The present invention provides an intelligent intelligence compilation method based on multi-intelligent collaboration and enhanced context window. Through the Master-Slaver mechanism, the intelligence data generated by multiple agents can be more effectively coordinated and summarized. At the same time, by splitting tasks according to keywords, the Slavers of different topics handle the compilation tasks respectively, which improves the context window utilization rate of the large language model, and can process data completely and stably when the amount of intelligence material is large. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a flow chart of the technical solution of the present invention. DETAILED DESCRIPTION

[0017] The present invention is described in detail below with reference to the accompanying drawings and embodiments.

[0018] like Figure 1 As shown, an intelligent intelligence compilation method based on enhanced context window of multi-intelligent collaboration includes the following steps.

[0019] Step S01: Through intelligence templates and reporting keywords, the system uses word embedding to classify topics to avoid generating reporting content that is irrelevant to the topic.

[0020] For example, if you specify the "technology" category and then compile information related to "Xiaomi", the generated content will only be related to Xiaomi company and will not be mixed with Xiaomi information in the fields of botany or agriculture.

[0021] Among them, the intelligence template defines the output format of the final report, such as logical segmentation, keywords, etc. The keywords can be specified by the user.

[0022] Step S02: Create a Master agent and multiple groups of Slaver agents. The Master agent assigns topics to each group of Slaver agents based on topic classification. The Slaver agents will perform specific intelligent reporting tasks based on the specific keywords assigned to them.

[0023] Step S03: After the topic assignment is completed, each group of Slaver agents obtains information from the materials for the topic and returns it to the Master agent to resolve duplicate information conflicts, ensuring that the source of the materials for the results generated by each group of Slaver agents is fully covered.

[0024] Step S04: When the Master agent finds that the duplicate information threshold is lower than the preset value, the Master agent instructs each group of Slaver agents to generate intelligence results; the Slaver agent ends the output of a certain type of subject intelligence results, and the Master agent gives a final response according to the prompt template requirements, which is used to command the Slaver to regenerate the results or complete the reporting of the assigned subject. When all Slaver agents receive the completion report instruction, a complete reporting process is completed.

[0025] The Master agent is capable of intelligent topic classification, automatic related information retrieval, intelligent generation, intelligent integration, etc. It is similar to a management and aggregation node and can assign topics to each group of Slaver agents based on keywords.

[0026] In one embodiment, in step S02, the specific intelligent reporting task compiles relevant information from the perspectives of multiple different keywords, such as historical background, geopolitics, and international environment.

[0027] In one embodiment, relevant information on the conflict between Country A and Country B is compiled from the perspectives of historical background, geopolitics, and international environment. Since the Master agent splits the overall task into different subtasks according to the keywords, the historical background, geopolitics, and international environment are each assigned to Slaver for processing. Each Slaver can fully use the complete context window of the large language model application when processing a specific subject subtask. The traditional method is to hand over all topics of the entire task to the large model application for processing at one time. Therefore, compared with the traditional method, the keywords are split and processed separately, which can be equivalently regarded as the context window being expanded.

[0028] On the basis of the previous embodiment, in step S03, after completing the task assignment by keyword, each group of Slavers obtains information from news or encyclopedia knowledge materials according to the topic, completes the assigned intelligent reporting task, and returns the processing results to the Master agent to resolve duplicate information "conflicts", avoiding different Slaver agents using the same material source to generate reporting information on different topics, resulting in a waste of large language model context windows.

[0029] For example, when compiling and reporting on the conflict between country A and country B, or topics related to geopolitics and international relations, it is very likely that the Slaver agent will use the same source of material to generate information, so that the final two parts of the report will be identical. In this case, the Master agent resolves the conflict of duplicate information; the specific method is: the Master agent uses text similarity, source tracking and other methods to summarize the reported information and compare it with the source of material, and then the Master agent guides the Slaver agent with duplicate information conflict to generate intelligence using different sources of material. Ensure that the source of material generated by each group of Slaver agents is comprehensive.

[0030] In step S04, when the Master agent finds that the duplicate information threshold is lower than the preset value, it means that the information collected by the Master agent is incomplete, which indirectly indicates that many Slavers use the same source of materials to generate reports. At this time, the Master agent will guide each Slaver group to generate intelligence results. Taking the data on the conflict between country A and country B as an example, when the Master agent finds that the Slaver agent that is compiling geopolitical topics has used a certain type of material, it will instruct the Slaver agent responsible for compiling international relations to avoid using the previous material, thereby ensuring high information coverage; finally, when the Slaver agent finishes the output of a certain type of subject intelligence results, the Master agent will give a final response according to the prompt template requirements, which is used to command the Slaver to regenerate the results or complete the reporting of the assigned topic. When all Slavers receive the command to complete the reporting, a complete reporting process is considered completed.

[0031] The present invention provides an intelligent intelligence compilation method based on multi-intelligent collaboration and enhanced context window. When executing intelligent compilation tasks, the system defines intelligent agents with multiple purposes, which can perform intelligent subject classification, automatic related information retrieval, intelligent generation, intelligent integration and other tasks. Through the Master-Slaver mechanism, the system incorporates the intelligent agents that perform branch processing tasks into the Slaver node and accepts the unified scheduling management of the Master node. At the same time, the Master node also carries the tasks of classification and aggregation, so that the entire system can more effectively coordinate and aggregate the intelligence data generated by multiple intelligent agents; in addition, the system splits the tasks according to the subject words, and the Slavers of different themes handle the compilation tasks respectively, which improves the context window utilization rate of the large language model, and can also process data completely and stably when the amount of intelligence materials is large.

Claims

1. An intelligent intelligence compilation method based on multi-intelligent collaboration and enhanced context window, characterized in that: The following steps are involved: Step S01: Based on the intelligence template and reporting keywords, the system uses word embedding to classify the topics; Step S02: Create a Master agent and multiple groups of Slaver agents. The Master agent assigns a topic to each group of Slaver agents according to the topic classification. The Slaver agents will perform specific intelligent reporting tasks according to the specific topic words assigned to them. Step S03: After the topic assignment is completed, each group of Slaver agents obtains information from the materials for the topic and returns it to the Master agent to resolve duplicate information conflicts, ensuring that the source of the materials generated by each group of Slavers is fully covered; Step S04: When the Master agent finds that the duplicate information threshold is lower than the preset value, the Master agent instructs each Slaver group to generate intelligence results; the Slaver agent ends the output of a certain type of subject intelligence results, and the Master agent gives a final response according to the prompt template requirements, which is used to command the Slaver to regenerate the results or complete the reporting of the assigned subject. When all Slaver agents receive the completion report instruction, a complete reporting process is completed.

2. The method according to claim 1, characterized in that: The Master agent is capable of intelligent topic classification, automatic related information retrieval, intelligent generation, and intelligent integration. It is a management and aggregation node that assigns topics to each group of Slaver agents based on keywords.

3. The method according to claim 1, characterized in that: In step S01, the intelligence template defines the output format of the final report, including logical segments and keywords, and the keywords are specified by the user.

4. The method according to claim 1, characterized in that: In step S02, the specific intelligent reporting task compiles relevant information from the perspectives of multiple different subject terms.

5. The method according to claim 1, characterized in that: In step S02, each of the Slaver agents can fully use the complete context window of the large language model application when processing the topic subtask.

6. The method according to claim 1, characterized in that: In step S03, the conflict is: different Slaver agents use the same material source to generate reporting information on different topics, and the method for the Master agent to resolve the conflict of duplicate information is: the Master agent summarizes the reporting information and compares it with the material source through text similarity and source tracking methods, and then the Master agent guides the Slaver agents that have duplicate information conflicts to use different material sources to generate intelligence.

7. The method according to claim 1, characterized in that: In step S04, the method by which the Master agent guides each group of Slaver agents to generate intelligence results is as follows: for a keyword, after the Master finds that the Slaver who is responsible for compiling a certain type of subject report has used a certain type of material, it will guide the Slaver who is responsible for compiling another type of subject report to avoid using the previous material, thereby ensuring high information coverage.

Images