Intelligent conference implementation method and device, equipment and storage medium

Abstract

The invention discloses an intelligent conference implementation method and device, equipment and a storage medium, relates to the technical field of artificial intelligence, is applied to an intelligent conference management system, and comprises the following steps: analyzing conference content information provided by a conference initiator based on a preset long-term conference memory bank by using a preset conference planning agent; generating a target conference agenda arrangement so as to start a conference based on the target conference agenda arrangement; in the conference proceeding process, the conference discussion content is monitored and guided in cooperation with each conference process coordination agent through a preset dynamic coordination memory library, so that conference record content is obtained; generating a conference discussion result by using a preset decision support agent and a task conversion agent based on the conference record content; the conference discussion result comprises task execution arrangement for the conference content information. Therefore, intelligent management of the whole process of the conference can be realized.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a method, device, equipment and storage medium for implementing an intelligent conference. Background Art

[0002] As modern enterprises expand in size and their organizational structures become more complex, meetings have become an important form of decision-making and collaboration. However, traditional meeting management methods have many problems, which seriously affect meeting efficiency and decision-making quality.

[0003] The conference management systems in the existing technology mainly have the following deficiencies: lack of experience accumulation and reuse mechanism. Most conference management systems only provide basic scheduling and conference room booking functions, and are unable to extract valuable experience from historical meetings. Each meeting needs to be planned from scratch, resulting in a lot of repetitive work and no way to learn from past successful experiences; the management of the meeting process is extensive. The existing system lacks intelligent support during the meeting and cannot identify problems such as discussions deviating from the topic and inappropriate decision-making timing in real time. Meeting records rely on manual work, which easily misses key information and makes it difficult to form structured knowledge accumulation; post-meeting execution tracking is weak, and meeting resolutions often remain at the level of written records. There is a lack of effective task decomposition and tracking mechanism, unclear task allocation, and difficult to monitor execution progress, resulting in difficulty in implementing meeting results; lack of intelligent collaboration support. Existing systems are mostly simple integrations of single functional modules. There is a lack of coordination between modules, and management strategies cannot be adjusted according to the dynamics of the meeting. The system cannot learn and adapt to the characteristics of different types of meetings, and it is difficult to provide personalized management solutions.

[0004] Therefore, how to achieve intelligent management of the entire process needs to be solved. Summary of the Invention

[0005] In view of this, the purpose of the present invention is to provide a method, device, equipment and storage medium for implementing intelligent conferencing, which can realize intelligent management of the entire process. The specific scheme is as follows:

[0006] In a first aspect, the present application discloses a method for implementing an intelligent conference, which is applied to an intelligent conference management system, comprising:

[0007] Analyzing the meeting content information provided by the meeting initiator using a preset meeting planning agent and based on a preset long-term meeting memory to generate a target meeting agenda, so as to start the meeting based on the target meeting agenda;

[0008] During the meeting, the preset dynamic collaborative memory library coordinates the various meeting process coordination agents to monitor and guide the meeting discussion content to obtain the meeting minutes;

[0009] Based on the content of the meeting minutes, and using a preset decision support agent and a task conversion agent, a meeting discussion result is generated; the meeting discussion result includes a task execution arrangement for the meeting content information.

[0010] Optionally, the method of using a preset meeting planning agent and analyzing the meeting content information provided by the meeting initiator based on a preset long-term meeting memory to generate a target meeting agenda includes:

[0011] Obtain conference content information provided by the conference initiator;

[0012] The meeting theme in the meeting content information is analyzed by a preset meeting planning agent and a preset general text vector model to obtain a target semantic vector; the preset meeting planning agent uses a preset artificial intelligence model as an inference engine;

[0013] Matching an initial reference meeting from each historical meeting record in a preset long-term meeting memory using the target semantic vector, and determining a target reference meeting from the initial historical meetings based on the multi-dimensional meeting information in the meeting content information;

[0014] The target meeting agenda is generated by the preset meeting planning agent and utilizing the multi-dimensional meeting information and the meeting process of the target reference meeting.

[0015] Optionally, the monitoring and guiding of the conference discussion content by coordinating various conference process coordination agents through a preset dynamic collaborative memory library includes:

[0016] Using the preset dynamic collaborative memory library to coordinate preset topics, the intelligent agent conducts real-time speech processing and semantic analysis on the current meeting discussion content to obtain the speaker's identity and real-time speech text content;

[0017] Based on the preset topic guidance agent and the target meeting agenda, the agent obtains the current topic information, and uses the preset dynamic collaborative memory to obtain the meeting speech information of the target meeting time period; the target meeting time period is a historical time period adjacent to the speech time period corresponding to the real-time speech text content;

[0018] The preset topic guides the intelligent agent to match the real-time speech text content with the current topic information, the conference speech information, and the identity of the conference speaker to obtain a relevance score;

[0019] If the relevance score is lower than a preset relevance threshold, a guide is generated based on the current topic information and the identity of the conference speaker to prompt the current conference speaker to adjust the current conference discussion content.

[0020] Optionally, the monitoring and guiding of the conference discussion content by coordinating various conference process coordination agents through a preset dynamic collaborative memory library includes:

[0021] Using a rule engine in a preset progress reminder agent to set a discussion time threshold for each discussion topic in the target meeting agenda; the discussion time threshold is determined based on the discussion time corresponding to each discussion topic;

[0022] When the discussion time of the current discussion topic in the meeting discussion content reaches the discussion time threshold, the preset progress is used to remind the intelligent agent and call the preset large-scale intensive language model to generate corresponding prompt content.

[0023] Optionally, the preset dynamic collaborative memory library is used to coordinate with each conference process coordination agent to monitor and guide the conference discussion content to obtain the meeting record content, including:

[0024] Extract the intelligent agent through the preset key points and use the preset Chinese lexical analysis tool to perform named entity recognition on the meeting discussion content to obtain the target entity, and use the preset dynamic collaborative memory to obtain the current speaker information, current speech content and current topic background;

[0025] Extracting current meeting data from the current speaker information, the current speech content and the current topic background through the preset key point extraction agent;

[0026] Generate corresponding meeting record content based on the target entity and the current meeting data.

[0027] Optionally, generating the meeting discussion results based on the meeting record content and using a preset decision support agent and a task conversion agent includes:

[0028] Analyze the logical relationship in the content of the meeting minutes based on a preset decision support agent and construct a corresponding decision argument diagram;

[0029] Generate a corresponding meeting decision result through the preset decision support agent and the decision argument diagram;

[0030] The preset task conversion agent is used to decompose the meeting decision results into corresponding tasks to be executed, and determine the task execution arrangements corresponding to the tasks to be executed.

[0031] Optionally, after generating the meeting discussion results based on the meeting minutes and using a preset decision support agent and task conversion agent, the method further includes:

[0032] A meeting minutes is generated based on the meeting record content and the meeting discussion results, and the preset long-term meeting memory is updated based on the meeting minutes and the meeting content information.

[0033] In a second aspect, the present application discloses an intelligent conference implementation device, which is applied to an intelligent conference management system, comprising:

[0034] A meeting content analysis module is used to analyze the meeting content information provided by the meeting initiator using a preset meeting planning agent and based on a preset long-term meeting memory to generate a target meeting agenda so as to start the meeting based on the target meeting agenda;

[0035] The meeting management module is used to monitor and guide the meeting discussion content during the meeting by coordinating various meeting process coordination agents through a preset dynamic collaborative memory library to obtain the meeting minutes;

[0036] The discussion result generation module is used to generate the meeting discussion results based on the meeting record content and using the preset decision support agent and task conversion agent; the meeting discussion results include task execution arrangements for the meeting content information.

[0037] In a third aspect, the present application discloses an electronic device, comprising:

[0038] Memory, used to store computer programs;

[0039] A processor is used to execute the computer program to implement the aforementioned intelligent conference implementation method.

[0040] In a fourth aspect, the present application discloses a computer-readable storage medium for storing a computer program, which implements the aforementioned intelligent conference implementation method when executed by a processor.

[0041] As can be seen, in this application, a preset conference planning agent is used to analyze the conference content information provided by the conference initiator based on a preset long-term conference memory to generate a target conference agenda so that the meeting can be started based on the target conference agenda. During the meeting, the preset dynamic collaborative memory is used to coordinate with each conference process coordination agent to monitor and guide the conference discussion content to obtain the meeting record content. Based on the meeting record content, the preset decision support agent and task transformation agent are used to generate the meeting discussion results. The meeting discussion results include the task execution arrangement for the meeting content information. That is, based on the two memory architectures of the long-term conference memory and the dynamic collaborative memory, different preset agents are dispatched to manage and promote the conference process. In this way, the problems of the existing conference management system, such as lack of experience accumulation, extensive process management, and low intelligence, are solved, and intelligent management of the entire conference process is realized, which significantly improves the meeting efficiency and decision-making quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are merely embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0043] Figure 1 This is a flow chart of a method for implementing an intelligent conference disclosed in this application;

[0044] Figure 2 This is a schematic diagram of the structure of an intelligent conference implementation device disclosed in this application;

[0045] Figure 3 This is a structural diagram of an electronic device disclosed in this application. DETAILED DESCRIPTION

[0046] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0047] In recent years, although some AI-based conference assistance tools have emerged, such as automatic transcription and intelligent summaries, these tools are still isolated functional points and lack systematic solutions. More importantly, existing tools fail to fully utilize the capabilities of large language models in semantic understanding, logical reasoning, and content generation, and lack a mechanism to effectively combine short-term meeting information with long-term organizational knowledge. They cannot solve the core problems of meeting management: how to make each meeting draw on historical experience, how to provide intelligent support during the meeting, and how to ensure that the results of the meeting are effectively transformed into action. Therefore, this application will specifically introduce a method for implementing intelligent meetings that can solve the above problems.

[0048] See also Figure 1 As shown, the embodiment of the present application discloses a method for implementing an intelligent conference, which is applied to an intelligent conference management system, including:

[0049] Step S11: Utilize the preset meeting planning agent and analyze the meeting content information provided by the meeting initiator based on the preset long-term meeting memory to generate a target meeting agenda, so as to start the meeting based on the target meeting agenda.

[0050] In this embodiment, it should be noted that the present invention constructs a dual memory architecture, including two parts: a long-term conference memory and a dynamic collaborative memory. The long-term conference memory is implemented using a graph database, which organizes core entities such as meetings, personnel, topics, decisions and tasks, and their relationship networks through knowledge graph technology. Specifically, a large language model is used to perform semantic understanding and entity extraction of conference texts, and a vectorized model is used to build semantic retrieval capabilities to support multi-dimensional association queries and similarity matching. Dynamic collaborative memory is implemented based on an in-memory database, using a multi-level data structure divided by time slices. It not only records real-time information during the meeting, but also realizes instant communication between intelligent agents through a publish-subscribe mechanism. The two forms of memory are organically connected through a systematic information screening and migration mechanism to ensure that short-term experience can be effectively transformed into long-term knowledge assets.

[0051] In this embodiment, the method of using a preset conference planning agent and analyzing the conference content information provided by the conference initiator based on a preset long-term conference memory to generate a target conference agenda includes: obtaining the conference content information provided by the conference initiator; analyzing the conference topic in the conference content information using a preset general text vector model through a preset conference planning agent to obtain a target semantic vector; the preset conference planning agent using a preset artificial intelligence model as an inference engine; matching an initial reference meeting from each historical meeting record in the preset long-term conference memory using the target semantic vector, and determining a target reference meeting from the initial historical meeting based on the multi-dimensional conference information in the conference content information; and generating a target conference agenda through the preset conference planning agent and using the multi-dimensional conference information and the meeting process of the target reference meeting. That is, the preset conference planning agent uses a large model to analyze historical meeting data and effect evaluation to generate an agenda that incorporates best practices.

[0052] Specifically, a pre-configured meeting planning agent serves as the overall coordinator of the system, using the DeepSeek-V3 model (a large language model) as its inference engine. This model excels in Chinese language comprehension and logical reasoning, making it particularly well-suited for complex meeting planning tasks. During the pre-meeting preparation phase, the planning agent first conducts a deep analysis of the meeting topic, generating a semantic vector for the topic using the BAAI General Embedding (BGE) model. It then searches the vector database for historical meetings with the highest similarity. This search not only considers topic similarity but also uses graph queries in Neo4j (a high-performance graph database) to obtain multi-dimensional information such as attendee overlap and time proximity. The most relevant reference meetings are then selected based on a comprehensive evaluation. Based on this retrieved historical experience, the planning agent then uses DeepSeek-V3 to generate agenda suggestions. The system provides the model with basic information about the current meeting, including the topic, objectives, estimated duration, and attendees, as well as analysis of agenda patterns from similar historical meetings. The model is then instructed to generate a complete agenda, including the names of each topic, time allotment, discussion focus, suggested speakers, and transitions between topics. Practice has shown that the model can effectively understand the characteristics of meeting types and generate a logical agenda. For example, for product review meetings, the model typically arranges requirements confirmation, solution presentation, feasibility analysis, risk assessment, and decision-making discussions, allocating time appropriately based on the complexity of the topic.

[0053] Step S12: During the meeting, the preset dynamic collaborative memory library is used to coordinate with each meeting process coordination agent to monitor and guide the meeting discussion content to obtain the meeting record content.

[0054] In this embodiment, during the meeting, each agent communicates asynchronously through the publish-subscribe function of the memory database, and adopts a formatted information transmission format to ensure the integrity and routability of the message. The system enables the agents to perceive each other's working status through a state synchronization mechanism, and dynamically adjust their own behavior to avoid processing bottlenecks. Among them, the meeting process coordination agent includes a preset topic guidance agent, a preset progress reminder agent, and a preset key points extraction agent. The preset topic guidance agent combines speech recognition and semantic analysis to monitor the discussion focus in real time and provide gentle guidance when necessary. The preset progress reminder agent collaborates with the rule engine and the large language model to ensure the efficiency of the meeting while avoiding interrupting key discussions. The preset key points extraction agent uses a combination of named entity recognition of the speech analysis model and deep understanding of the large model to efficiently and accurately extract the key points of the meeting. Furthermore, the collaboration between the various agents is achieved through a carefully designed communication and coordination mechanism. All agents communicate asynchronously through the publish-subscribe functionality of Redis (Remote Dictionary Server). Messages are formatted in JSON-RPC (a stateless and lightweight remote procedure call protocol) and contain information such as the sender's identity, recipient list, message type, specific content, priority, and timestamp. This design ensures message integrity while supporting flexible routing and prioritization. Furthermore, each agent maintains its own status information, including the current task being processed, the length of its pending queue, the last active time, and a processing capacity assessment, which is updated in Redis every second. Other agents can query this status in real time and adjust their behavior accordingly. For example, if the processing queue of the key point extraction agent becomes too long, the topic guidance agent will appropriately reduce the frequency of generating guidance suggestions to avoid processing bottlenecks.

[0055] In this embodiment, the preset dynamic collaborative memory library is used to coordinate various conference process coordination agents to monitor and guide the conference discussion content, including: using the preset dynamic collaborative memory library to coordinate the preset topic guiding agent to perform real-time voice processing and semantic analysis on the current conference discussion content to obtain the conference speaker identity and real-time speech text content of the current conference speaker; based on the preset topic guiding agent and the target conference agenda, the current topic information is obtained, and the conference speech information of the target conference time period is obtained using the preset dynamic collaborative memory library; the target conference time period is a historical time period adjacent to the speech time period corresponding to the real-time speech text content; the preset topic guiding agent matches the real-time speech text content with the current topic information, the conference speech information, and the conference speaker identity to obtain a relevance score; if the relevance score is lower than the preset relevance threshold, a guide statement is generated based on the current topic information and the conference speaker identity to prompt the current conference speaker to adjust the current conference discussion content.

[0056] Specifically, the pre-set topic steering agent is responsible for maintaining the focus of the meeting discussion. Its implementation involves two key components: real-time speech processing and semantic analysis. The speech recognition system uses an open-source speech recognition toolkit model that supports streaming processing and can convert speech to text with extremely low latency. It should be noted that the system calls the recognition interface every three seconds to obtain the text content during that period and simultaneously uses voiceprint recognition technology to determine the speaker's identity. The system uses the Qwen3-14B model to analyze the recognized text for topic relevance. Regarding the implementation of topic relevance determination, the system provides the main model with information about the current topic, a list of topic keywords, and the last three minutes of discussion content. The model is instructed to assess the relevance of the discussion to the topic and output a relevance score from 0 to 100. The model also determines the type of deviation: completely irrelevant, partially relevant, or overly involved. When the model determines that steering is needed, it generates a guide of no more than 50 words. These guides follow specific principles: first, they affirm the value of the current discussion and then gently guide the topic back to the main topic, avoiding abrupt interruptions. For example, when a technical discussion goes too deep into details, the system may generate guiding suggestions such as "This technical point is indeed important. It is recommended to record it and discuss it separately. Why don't we first determine the overall implementation plan?"

[0057] In this embodiment, the preset dynamic collaborative memory repository coordinates various conference process coordination agents to monitor and guide meeting discussion content. This includes: utilizing a rules engine within a preset progress reminder agent to set a discussion time threshold for each discussion topic in the target meeting agenda; the discussion time threshold is determined based on the discussion duration of each discussion topic; and when the discussion time for the current discussion topic in the meeting reaches the discussion time threshold, the preset progress reminder agent invokes a preset large-scale intensive language model to generate a corresponding reminder. Specifically, the preset progress reminder agent utilizes a combination of a rules engine and a large-scale intensive language model. The rules engine monitors time progress in real time and maintains a state machine for each preset topic, including four states: not started, in progress, about to time out, and timed out. When a topic's usage time reaches 80% of its planned duration, the system enters an alert state. At this point, the system invokes the Qwen3-14B model (a question-and-answer model) to generate appropriate reminder content. When generating reminders, the model comprehensively considers factors such as the importance of the current topic, the number of remaining topics, and the level of discussion activity. If a critical decision-making discussion is currently underway, the model generates a more tactful reminder and may even suggest an appropriate extension of the time. These adjustments are based on historical experience data - the system obtains relevant information from its long-term memory bank, and forcibly interrupting key discussions often leads to a decline in decision-making quality.

[0058] In this embodiment, the preset dynamic collaborative memory library is used to coordinate various conference process coordination agents to monitor and guide the conference discussion content to obtain the meeting record content, including: using a preset key point extraction agent and using a preset Chinese lexical analysis tool to perform named entity recognition on the conference discussion content to obtain a target entity, and using the preset dynamic collaborative memory library to obtain the current speaker information, current speech content and current topic background; extracting current meeting data from the current speaker information, the current speech content and the current topic background through the preset key point extraction agent; and generating corresponding meeting record content based on the target entity and the current meeting data.

[0059] Specifically, the system first uses the LAC (Lexical Analysis of Chinese) model for named entity recognition, accurately extracting key entities such as names of people, organizations, and products. Compared to large models like Qwen3-14B, the LAC model significantly improves processing efficiency and reduces overall algorithmic overhead. For numerical information, carefully designed regular expressions are used to recognize various types of numbers, including amounts, percentages, and dates with Chinese units. After extracting basic information, the system then uses the DeepSeek-V3 model for in-depth understanding and structuring.

[0060] The core of key point extraction lies in identifying the importance and relevance of information. The system provides DeepSeek-V3 with speaker information, speech content, and the background of the current topic, requiring the model to identify the nature of the core viewpoint (support, opposition, or suggestion), extract key data and facts, evaluate the importance of the information (1-5 points), and determine whether it is relevant to the decision. The model output includes structured information such as key point type, specific content, importance score, emotional tendency, and related entities. Key points with an importance of 4 points or more will be immediately synchronized to the dynamic collaborative memory to ensure that other intelligent agents can obtain key information in a timely manner. It is particularly noteworthy that the model will refer to historical experience when evaluating importance - if some seemingly ordinary information points have triggered important decisions or subsequent actions in historical meetings, the system will increase their importance score accordingly.

[0061] Step S13: Based on the content of the meeting minutes, and using a preset decision support agent and a task conversion agent, generate a meeting discussion result; the meeting discussion result includes a task execution arrangement for the meeting content information.

[0062] In this embodiment, at the end of a meeting, a pre-set decision support agent and a pre-set task conversion agent can collaboratively generate the final meeting discussion results. The pre-set decision support agent constructs an argument graph structure, identifies opinion relationships and consensus levels through semantic analysis, and facilitates decision-making at the appropriate time. The pre-set task conversion agent, based on the organizational knowledge graph, decomposes decision intelligence into actionable tasks and optimizes personnel allocation.

[0063] In this embodiment, generating meeting discussion results based on the meeting minutes using a pre-set decision support agent and a task transformation agent involves: analyzing the logical relationships within the meeting minutes using a pre-set decision support agent and constructing a corresponding decision argument graph; generating the corresponding meeting decision results using the pre-set decision support agent and the decision argument graph; and decomposing the meeting decision results into corresponding pending tasks using a pre-set task transformation agent and determining the execution schedule for each pending task. Specifically, a graph structure is constructed and maintained, where nodes represent different viewpoints or proposals, and edges represent the logical relationships between viewpoints. Identifying relationship types relies on the semantic understanding capabilities of the large model. The system inputs pairs of viewpoints into the model and is asked to determine whether the relationship between them is supportive, opposing, complementary, or substitutive. The model not only provides relationship types but also confidence scores and justifications. By continuously analyzing speeches during the meeting, the system gradually constructs a complete argument graph, clearly displaying the logical flow of each party's viewpoints. The construction of the argument graph also incorporates learning from historical decision patterns. The system retrieves similar decision scenarios from its long-term memory to identify common argumentation patterns and decision pitfalls. Consensus is calculated using a comprehensive assessment method. The system not only counts explicit statements, such as "I agree" and "I disagree," but also identifies implicit attitudes through semantic analysis. For example, when someone offers a supplementary suggestion, it typically indicates partial agreement; proposing an alternative may indicate disagreement. The final consensus is calculated using a weighted approach, with explicit support accounting for 70% and implicit support accounting for 30%. When consensus exceeds a preset threshold (typically 75%), the system determines that the decision phase can be entered. Specifically, determining the timing of a decision requires more comprehensive consideration. The system invokes the DeepSeek-V3 model, taking as input the decision topic, discussion duration, number of participants, current support, and a summary of the last five minutes of discussion. The model is instructed to assess whether the discussion is sufficient and whether the decision criteria have been met. The model analyzes whether key issues have been addressed, whether key stakeholders have expressed their views, and whether there are any major unresolved disagreements. Based on this analysis, if the time is not ripe for a decision, the agent identifies areas requiring further discussion. After a decision is reached, a pre-set task conversion agent ensures that the meeting decision is translated into concrete action. The system uses the Qwen3-14B model to intelligently decompose decision content. The model identifies all required actions, specifies the target for each action, extracts constraints such as time and quality, and recommends appropriate individuals based on the personnel information in the knowledge graph.

[0064] In this embodiment, after generating meeting discussion results based on the meeting minutes using a pre-set decision support agent and task conversion agent, the system further includes: generating meeting minutes based on the meeting minutes and the meeting discussion results, and updating the pre-set long-term meeting memory based on the meeting minutes and the meeting content information. Specifically, after each meeting, an experience extraction process is automatically triggered. By comparing the meeting plan with the actual execution, deviations and areas for improvement are identified. This experience data is not only updated in the long-term memory but also fed back to each agent to optimize its decision-making model. For example, if a certain type of topic is found to frequently exceed the time limit, the meeting planning agent will automatically increase the time budget for this topic in subsequent similar meetings. Through this continuous learning and improvement, the system gradually accumulates the organization's meeting wisdom, making each meeting more efficient than the previous one. In this way, by establishing a comprehensive experience accumulation and reuse mechanism, not only can high-value content in dynamic memory be transferred to the long-term memory, but each agent can also refer to historical experience when handling current tasks. For example, the topic guidance agent will learn the discussion patterns of specific topic types, and the decision support agent will identify common patterns and potential pitfalls in historical decision-making.

[0065] In addition, it should be noted that in this application, the output of all intelligent agents is presented in the form of suggestions, and it is up to the meeting host to decide whether to adopt them. The system will learn the preferences of the host and the meeting culture of the organization, and gradually optimize the content and timing of the suggestions. Through continuous learning and improvement, the system can adapt to the needs of different types of meetings and provide organizations with increasingly accurate meeting management support. In this way, not only can the efficiency and quality of a single meeting be improved, but more importantly, a systematic meeting knowledge management system can be established. The system can learn from each meeting, transform individual experience into organizational wisdom, and continuously improve meeting management capabilities, truly realizing the intelligent management goal of learning from experience and evolving in practice.

[0066] The system's complete operational process is illustrated using a product technology proposal review meeting as an example. The meeting initiator created a "Next-Generation Intelligent Customer Service System Technology Proposal Review" meeting, set it to 2.5 hours, and invited eight relevant personnel from product, technology, and operations departments. The meeting planning agent immediately began work, using semantic search and graph query to locate 12 relevant technology review meetings from the past year. Analysis revealed that successful technology proposal reviews typically include five key steps: requirements confirmation, architecture design, technical details, deployment plan, and risk assessment. Based on this model and the specifics of the meeting, the system generated a detailed agenda and identified the need to invite security experts to participate in the risk assessment phase. The agents worked collaboratively throughout the meeting. During the architecture design discussion, when the topic shifted toward specific code implementation details, the topic guidance agent promptly identified the issue and generated guidance advice: "Code-level discussions are valuable. We recommend first finalizing the overall architecture, with implementation details discussed later in the meeting." During the risk assessment phase, the decision support agent constructed an argument diagram to clearly illustrate the different viewpoints and their rationale regarding the use of open source components versus in-house core modules, helping participants fully understand each side's position. As the discussion neared its conclusion, the decision-support agent determined that decision-making conditions were met and recommended a vote. The meeting ultimately resulted in three key decisions, which the task conversion agent then decomposed into 12 specific tasks, including prototype development, performance testing, and security assessment. Each task clearly defined responsible individuals, collaborators, completion standards, and deadlines. After the meeting, the system not only generated detailed meeting minutes but also updated the learnings from the meeting into its long-term memory. In particular, key learning points, such as "microservice architecture selection" and "open source component usage decisions," along with the underlying rationale for these decisions, were fully documented. These valuable insights will serve as a valuable reference for similar meetings in the future. Through this continuous learning and accumulation, the system's meeting management capabilities continue to improve, truly achieving the intelligent management goal of learning from experience and evolving through practice.

[0067] In conjunction with the above examples, the long-term meeting memory and dynamic collaboration memory used in this application are explained again. First, the construction of the long-term meeting memory adopts knowledge graph technology, and Neo4j is selected as the graph database engine. The system first defines an ontology model containing five core entities: meetings, personnel, topics, decisions, and tasks. In order to achieve efficient semantic understanding and information extraction, the system uses the Qwen3-14B large language model to process meeting texts. In the entity recognition stage, the system inputs meeting record fragments into the Qwen3-14B model and guides the model to extract structured information through carefully designed prompts. The prompts include requiring the model to recognize the names, departments, and positions of personnel, extract topic names and discussion points, identify the content and type of decisions, and decompose task descriptions, responsible persons, and deadlines. The model outputs structured data in JSON format, which is stored in the graph database using Neo4j's batch import interface after format verification. To support semantic retrieval capabilities, the system uses the BGE-m3 vectorization model, which is a semantic representation model specially optimized for Chinese. In its implementation, the system segments meeting transcripts based on semantic completeness, limiting each segment to 512 characters. These segments are then converted into high-dimensional dense vectors using a BGE model. These vectors are stored in a vector database, forming an efficient similarity search index. When searching for relevant historical meetings, the system similarly converts the query text into vectors, calculates cosine similarity to identify the most relevant meeting segments, and then uses these segments to locate the complete meeting transcript. Dynamic collaborative memory is implemented using a Redis in-memory database with a multi-layered data structure. The system divides the meeting process into 5-minute time slices, with each time slice corresponding to a Redis hash table that stores all important information from that time period. The hash table contains fields such as a list of discussion points, analysis outputs from each agent, provisional consensus, and to-do items. To enable real-time communication between agents, the system utilizes Redis's publish-subscribe mechanism. Each agent subscribes to a specific message channel. When an agent produces important output, it notifies other relevant agents by publishing a message. This short-term memory not only facilitates real-time collaboration within the current meeting but also provides raw material for the formation of long-term memory. After the meeting, the system will screen and refine the content in dynamic memory and transfer high-value information to the long-term memory bank, thus achieving an organic connection between the two forms of memory.

[0068] It should be noted here that when implementing smart meetings for the first time, first, analyze the target business process and identify key decision points and information flows; second, design a domain ontology model and define core entities, attributes, and relationships; then, configure the roles of intelligent agents and clarify their respective responsibilities and collaboration modes; then, initialize the memory library and import historical data and domain knowledge; finally, deploy and operate the system and continuously optimize it in practice.

[0069] It can be seen that in this embodiment, a preset conference planning agent is used to analyze the conference content information provided by the conference initiator based on a preset long-term conference memory to generate a target conference agenda so that the meeting can be started based on the target conference agenda. During the meeting, the preset dynamic collaborative memory is used to coordinate with each conference process coordination agent to monitor and guide the conference discussion content to obtain the meeting record content. Based on the meeting record content, the preset decision support agent and task conversion agent are used to generate the meeting discussion results. The meeting discussion results include the task execution arrangement for the meeting content information. That is, based on the two memory architectures of the long-term conference memory and the dynamic collaborative memory, different preset agents are dispatched to manage and promote the conference process. In this way, the problems of the existing conference management system, such as lack of experience accumulation, extensive process management, and low intelligence, are solved, and intelligent management of the entire conference process is realized, which significantly improves the meeting efficiency and decision-making quality.

[0070] refer to Figure 2 The embodiment of the present application further discloses an intelligent conference implementation device, which is applied to an intelligent conference management system, including:

[0071] A meeting content analysis module 11 is configured to analyze the meeting content information provided by the meeting initiator using a preset meeting planning agent and based on a preset long-term meeting memory to generate a target meeting agenda, so as to start the meeting based on the target meeting agenda;

[0072] The meeting management module 12 is used to monitor and guide the meeting discussion content during the meeting by coordinating the meeting process agents through a preset dynamic collaborative memory library to obtain the meeting record content;

[0073] The discussion result generation module 13 is used to generate the meeting discussion results based on the meeting record content and using the preset decision support agent and task conversion agent; the meeting discussion results include task execution arrangements for the meeting content information.

[0074] It can be seen that in this embodiment, based on the two memory architectures of the long-term conference memory library and the dynamic collaborative memory library, different preset intelligent agents are dispatched to manage and promote the conference process. In this way, the problems of the existing conference management system such as lack of experience accumulation, extensive process management, and low intelligence level are solved, and intelligent management of the entire conference process is realized, which significantly improves the meeting efficiency and decision-making quality.

[0075] In some specific embodiments, the conference content analysis module 11 may specifically include:

[0076] A conference content acquisition unit, configured to acquire conference content information provided by the conference initiator;

[0077] A conference theme analysis unit is configured to analyze the conference theme in the conference content information using a preset conference planning agent and a preset general text vector model to obtain a target semantic vector; the preset conference planning agent uses a preset artificial intelligence model as an inference engine;

[0078] a historical meeting retrieval unit, configured to match an initial reference meeting from each historical meeting record in a preset long-term meeting memory using the target semantic vector, and determine a target reference meeting from the initial historical meetings based on the multi-dimensional meeting information in the meeting content information;

[0079] An agenda security generation unit is used to generate a target meeting agenda arrangement through the preset meeting planning agent, and using the multi-dimensional meeting information and the meeting process of the target reference meeting.

[0080] In some specific embodiments, the conference management module 12 may specifically include:

[0081] The conference content analysis unit is used to use the preset dynamic collaborative memory library to coordinate the preset topics to guide the intelligent agent to perform real-time speech processing and semantic analysis on the current conference discussion content, so as to obtain the speaker identity and real-time speech text content of the current conference speaker;

[0082] An agenda information acquisition unit is configured to guide the agent to acquire current agenda information based on the preset agenda and the target meeting agenda, and to use the preset dynamic collaborative memory to acquire meeting speech information for a target meeting time period; the target meeting time period is a historical time period adjacent to the speech time period corresponding to the real-time speech text content;

[0083] a relevance judgment unit, configured to guide the agent to match the real-time speech text content with the current topic information, the conference speech information, and the identity of the conference speaker through the preset topic, so as to obtain a relevance score;

[0084] The conference guiding unit is configured to generate a guiding statement based on the current topic information and the identity of the conference speaker if the relevance score is lower than a preset relevance threshold, so as to prompt the current conference speaker to adjust the current conference discussion content.

[0085] In some specific embodiments, the conference management module 12 may specifically include:

[0086] a duration determination unit, configured to set a discussion time threshold for each discussion topic in the target meeting agenda using a rule engine in a preset progress reminder agent; the discussion time threshold is determined based on the discussion time corresponding to each discussion topic;

[0087] The duration prompt unit is used to remind the intelligent agent through the preset progress and call the preset large-scale intensive language model to generate corresponding prompt content when the discussion time of the current discussion topic in the meeting discussion content reaches the discussion time threshold.

[0088] In some specific embodiments, the conference management module 12 may specifically include:

[0089] The content entity recognition unit is used to extract the intelligent entity through the preset key points and use the preset Chinese lexical analysis tool to perform named entity recognition on the meeting discussion content to obtain the target entity, and use the preset dynamic collaborative memory to obtain the current speaker information, current speech content and current topic background;

[0090] A data extraction unit, configured to extract current meeting data from the current speaker information, the current speech content, and the current topic background through the preset key point extraction agent;

[0091] A conference record unit is used to generate corresponding conference record content based on the target entity and the current conference data.

[0092] In some specific embodiments, the discussion result generating module 13 may specifically include:

[0093] A decision diagram generating unit, configured to analyze the logical relationships in the meeting minutes based on a preset decision support agent and construct a corresponding decision argument diagram;

[0094] A decision result determination unit, configured to generate a corresponding meeting decision result through the preset decision support agent and the decision argument diagram;

[0095] The task arrangement generation unit is used to use the preset task conversion agent to decompose the meeting decision results into corresponding tasks to be executed, and determine the task execution arrangements corresponding to the tasks to be executed.

[0096] In some specific embodiments, the intelligent conference implementation device may further include:

[0097] A database updating module is used to generate a meeting minutes based on the meeting record content and the meeting discussion results, and to update the preset long-term meeting memory based on the meeting minutes and the meeting content information.

[0098] Furthermore, the embodiment of the present application also discloses an electronic device, Figure 3 This is a structural diagram of an electronic device 20 according to an exemplary embodiment. The content in the diagram should not be considered as any limitation to the scope of application of the present application.

[0099] Figure 3 This is a schematic diagram of the structure of an electronic device 20 provided in an embodiment of the present application. The electronic device 20 may specifically include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input / output interface 25, and a communication bus 26. The memory 22 is used to store a computer program, which is loaded and executed by the processor 21 to implement the relevant steps of the intelligent conferencing implementation method disclosed in any of the aforementioned embodiments. Furthermore, the electronic device 20 in this embodiment may specifically be an electronic computer.

[0100] In this embodiment, the power supply 23 is used to provide operating voltage for each hardware device on the electronic device 20; the communication interface 24 can create a data transmission channel between the electronic device 20 and the external device. The communication protocol it follows is any communication protocol that can be applied to the technical solution of this application and is not specifically limited here; the input and output interface 25 is used to obtain external input data or output data to the outside world. Its specific interface type can be selected according to specific application needs and is not specifically limited here.

[0101] In addition, the memory 22, as a carrier for resource storage, can be a read-only memory, random access memory, disk or CD, etc. The resources stored thereon can include an operating system 221, a computer program 222, etc., and the storage method can be temporary storage or permanent storage.

[0102] The operating system 221 is used to manage and control the hardware devices on the electronic device 20 and the computer program 222. The operating system 221 can be Windows Server, NetWare, Unix, Linux, etc. In addition to including computer programs capable of implementing the intelligent conferencing implementation method performed by the electronic device 20 as disclosed in any of the aforementioned embodiments, the computer program 222 can further include computer programs capable of performing other specific tasks.

[0103] Furthermore, this application also discloses a computer-readable storage medium for storing a computer program; wherein, when executed by a processor, the computer program implements the aforementioned intelligent conferencing implementation method. The specific steps of this method can be referred to the corresponding contents disclosed in the aforementioned embodiments and will not be repeated here.

[0104] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from the other embodiments. Reference can be made to the descriptions of the identical or similar parts between the various embodiments. For the devices disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple, and the relevant parts can be referred to the descriptions of the methods.

[0105] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the above description has generally described the components and steps of each example according to their functions. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0106] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein may be implemented directly using hardware, a software module executed by a processor, or a combination of the two. The software module may be placed in random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.

[0107] Finally, it should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of additional identical elements in the process, method, article, or device comprising the element.

[0108] The above is a detailed introduction to the technical solution provided by the present application. Specific examples are used herein to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method of the present application and its core idea. At the same time, for those skilled in the art, according to the ideas of the present application, there may be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on the present application.

Claims

1. A method for implementing an intelligent conference, characterized in that: Applied to intelligent conference management systems, including: Analyzing the meeting content information provided by the meeting initiator using a preset meeting planning agent and based on a preset long-term meeting memory to generate a target meeting agenda, so as to start the meeting based on the target meeting agenda; During the meeting, the preset dynamic collaborative memory library coordinates the various meeting process coordination agents to monitor and guide the meeting discussion content to obtain the meeting minutes; Based on the content of the meeting minutes, and using a preset decision support agent and a task conversion agent, a meeting discussion result is generated; the meeting discussion result includes a task execution arrangement for the meeting content information.

2. The method for implementing an intelligent conference according to claim 1, wherein: The method of using a preset meeting planning agent and analyzing the meeting content information provided by the meeting initiator based on a preset long-term meeting memory bank to generate a target meeting agenda includes: Obtain conference content information provided by the conference initiator; The meeting theme in the meeting content information is analyzed by a preset meeting planning agent and a preset general text vector model to obtain a target semantic vector; the preset meeting planning agent uses a preset artificial intelligence model as an inference engine; Matching an initial reference meeting from each historical meeting record in a preset long-term meeting memory using the target semantic vector, and determining a target reference meeting from the initial historical meetings based on the multi-dimensional meeting information in the meeting content information; The target meeting agenda is generated by the preset meeting planning agent and utilizing the multi-dimensional meeting information and the meeting process of the target reference meeting.

3. The method for implementing an intelligent conference according to claim 1, wherein: The method of coordinating the conference process coordination agents to monitor and guide the conference discussion content through the preset dynamic collaborative memory library includes: Using the preset dynamic collaborative memory library to coordinate preset topics, the intelligent agent conducts real-time speech processing and semantic analysis on the current meeting discussion content to obtain the speaker's identity and real-time speech text content; Based on the preset topic guidance agent and the target meeting agenda, the agent obtains the current topic information, and uses the preset dynamic collaborative memory to obtain the meeting speech information of the target meeting time period; the target meeting time period is a historical time period adjacent to the speech time period corresponding to the real-time speech text content; The preset topic guides the intelligent agent to match the real-time speech text content with the current topic information, the conference speech information, and the identity of the conference speaker to obtain a relevance score; If the relevance score is lower than a preset relevance threshold, a guide is generated based on the current topic information and the identity of the conference speaker to prompt the current conference speaker to adjust the current conference discussion content.

4. The method for implementing an intelligent conference according to claim 1, wherein: The method of coordinating the conference process coordination agents to monitor and guide the conference discussion content through the preset dynamic collaborative memory library includes: Using a rule engine in a preset progress reminder agent to set a discussion time threshold for each discussion topic in the target meeting agenda; the discussion time threshold is determined based on the discussion time corresponding to each discussion topic; When the discussion time of the current discussion topic in the meeting discussion content reaches the discussion time threshold, the preset progress is used to remind the intelligent agent and call the preset large-scale intensive language model to generate corresponding prompt content.

5. The method for implementing an intelligent conference according to claim 1, wherein: The preset dynamic collaborative memory library coordinates each conference process coordination agent to monitor and guide the conference discussion content to obtain the meeting record content, including: Extract the intelligent agent through the preset key points and use the preset Chinese lexical analysis tool to perform named entity recognition on the meeting discussion content to obtain the target entity, and use the preset dynamic collaborative memory to obtain the current speaker information, current speech content and current topic background; Extracting current meeting data from the current speaker information, the current speech content and the current topic background through the preset key point extraction agent; Generate corresponding meeting record content based on the target entity and the current meeting data.

6. The method for implementing an intelligent conference according to claim 1, wherein: The generating of the meeting discussion results based on the meeting record content and using the preset decision support agent and task conversion agent includes: Analyze the logical relationship in the content of the meeting minutes based on a preset decision support agent and construct a corresponding decision argument diagram; Generate a corresponding meeting decision result through the preset decision support agent and the decision argument diagram; The preset task conversion agent is used to decompose the meeting decision results into corresponding tasks to be executed, and determine the task execution arrangements corresponding to the tasks to be executed.

7. The method for implementing an intelligent conference according to any one of claims 1 to 6, characterized in that: After generating the meeting discussion results based on the meeting minutes and using the preset decision support agent and task conversion agent, the method further includes: A meeting minutes is generated based on the meeting record content and the meeting discussion results, and the preset long-term meeting memory is updated based on the meeting minutes and the meeting content information.

8. An intelligent conference implementation device, characterized in that: Applied to intelligent conference management systems, including: A meeting content analysis module is used to analyze the meeting content information provided by the meeting initiator using a preset meeting planning agent and based on a preset long-term meeting memory to generate a target meeting agenda so as to start the meeting based on the target meeting agenda; The meeting management module is used to monitor and guide the meeting discussion content during the meeting by coordinating various meeting process coordination agents through a preset dynamic collaborative memory library to obtain the meeting minutes; The discussion result generation module is used to generate the meeting discussion results based on the meeting record content and using the preset decision support agent and task conversion agent; the meeting discussion results include task execution arrangements for the meeting content information.

9. An electronic device, characterized in that: include: Memory, used to store computer programs; A processor, configured to execute the computer program to implement the intelligent conference implementation method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that Used to store a computer program, which, when executed by a processor, implements the intelligent conference implementation method according to any one of claims 1 to 7.

Images