Large language model real-time analysis method, device and equipment for meeting and medium

Abstract

This application provides a method, apparatus, device, and medium for real-time analysis of large language models for conferences, effectively solving the problems of existing large language models lacking real-time and accurate analysis, and failing to meet the needs of conferences. The method includes: pre-configuring a conference system, with participants accessing the target conference via a real-time communication channel; during the target conference, extracting various textual information, multi-dimensional conference information, and reference documents input by participants from the target conference database in parallel; fusing the various textual information, multi-dimensional conference information, and reference documents to obtain composite prompts for the target conference, and inferring the composite prompts based on a pre-set large language model to obtain inference results; parsing the inference results to determine the current conference status of the target conference, and pushing guidance suggestions from the inference results to all conference clients via the real-time communication channel.

Description

Technical Field

[0001] This application relates to the field of real-time conference analysis technology, and more specifically, to methods, apparatus, devices, and media for real-time analysis of large language models for conferences. Background Technology

[0002] With the development of artificial intelligence technology, Large Language Models (LLM) are widely used in the field of meeting minutes and summaries. Existing tools such as Lark and Otter.ai all rely on a core solution where, after the meeting, the speech-to-text transcript is submitted to the LLM to generate a one-time meeting summary. Their systems mainly consist of four modules: a speech acquisition module passively records participants' speech throughout the meeting and transmits it to a speech recognition module without any analysis or real-time feedback; a speech recognition module uses ASR technology to transcribe the speech into a meeting transcript, which is then stored in a text storage module; the text storage module submits the complete transcript to the LLM processing module after the meeting; and the LLM processing module extracts key points from the transcript, generates a summary, and then the summary output module provides feedback to the user, who can only view the summary after the meeting.

[0003] Since the LLM processing module only intervenes after the meeting and does not perform any analysis during the meeting, it has two major drawbacks: First, the lack of real-time analysis during the meeting makes it impossible to detect deviations in discussion in a timely manner. When participants deviate from the agenda, they can only rely on themselves to notice, resulting in wasted time. Second, the agent-based analysis lacks business context and can only summarize the speech in words, which is out of touch with the actual business objectives and cannot meet the needs of the meeting. Summary of the Invention

[0004] In view of this, the purpose of this application is to provide a method, apparatus, equipment and medium for real-time analysis of large language models for conferences, which effectively solves the problem that existing large language models lack real-time and accurate analysis and cannot meet the needs of conferences.

[0005] In a first aspect, embodiments of this application provide a method for real-time analysis of a large language model for meetings, the method comprising: A pre-configured conference system allows participants to access the target conference by using their conference clients and being assigned corresponding real-time communication channels based on the configured system; the conference system includes a conference database. During the target meeting, various text information, multi-dimensional meeting information, and reference documents of the target meeting are extracted in parallel from the target meeting database based on the meeting client held by the participants. By integrating the various textual information, multi-dimensional meeting information, and reference documents, a composite prompt word for the target meeting is obtained, and the inference result is obtained by inferring the composite prompt word based on a pre-set large language model; The reasoning results are analyzed to determine the current meeting status of the target meeting, and guidance suggestions from the reasoning results are pushed to all meeting clients of the target meeting through a real-time communication channel.

[0006] In conjunction with the first aspect, this application provides a first possible implementation of the first aspect, wherein the composite prompt words for the target meeting are obtained by integrating the various textual information, multi-dimensional meeting information, and reference documents, including: The composite prompt words are pre-configured, including system prompt words and user prompt words, and the corresponding processing methods are set; The processing method described above is used to process the various text information, multi-dimensional meeting information, and reference documents to obtain the system prompt words and user prompt words.

[0007] In conjunction with the first aspect, this application provides a second possible implementation of the first aspect, wherein performing the processing method to obtain the system prompt and user prompt includes: The multi-dimensional meeting information, the roles set in the large language model, and the content of the reference documents are injected and concatenated into the preset system prompt word template; Add analysis instructions obtained by integrating multiple reasoning dimensions to the system prompt word template to generate the system prompt word.

[0008] In conjunction with the first aspect, this application provides a third possible implementation of the first aspect, wherein obtaining the reasoning result based on the compound prompt word using a preset large language model includes: Based on multiple inference dimensions in the system prompts, semantic analysis is performed on the user prompts in the composite prompts to obtain the analysis results; The inference result is obtained by cross-referencing the analysis results with the multi-dimensional meeting information in the system prompts.

[0009] In conjunction with the first aspect, this application provides a fourth possible implementation of the first aspect, wherein the inference result is obtained by cross-comparing the analysis result with the multi-dimensional meeting information in the system prompts, including: The similarity between the core semantic direction in the analysis results and the current agenda in the multi-dimensional meeting information is calculated to determine the real-time status of the discussion; The real-time progress of the meeting is assessed based on the remaining meeting time and the current agenda from multi-dimensional meeting information, and then integrated with the real-time status of the discussion to form the reasoning result.

[0010] In conjunction with the first aspect, this application provides a fifth possible implementation of the first aspect, wherein pushing guidance suggestions from the inference results to all meeting clients of the target meeting via a real-time communication channel includes: The meeting client is pre-configured to determine the emotional state contained in the guidance suggestions; Based on the emotional state, a target visual style is matched from a pre-defined visual library to display the guidance suggestion.

[0011] In conjunction with the first aspect, this application provides a sixth possible implementation of the first aspect, wherein, during the conduct of the target meeting, the following is included: Real-time monitoring of the amount of text information in the target meeting within a preset time window, as well as the meeting agenda of the target meeting; Determine whether the quantity and the meeting agenda meet the triggering mechanism for real-time analysis of the target meeting, so as to analyze the target meeting in real time.

[0012] Secondly, embodiments of this application provide a real-time analysis device for large language models used in conferences, the device comprising: A configuration module is used to pre-configure the conference system. Participants use their conference clients to access the target conference by being assigned a corresponding real-time communication channel based on the configured conference system. The conference system includes a conference database. The input module is used to extract, in parallel, various text information, multi-dimensional meeting information, and reference documents of the target meeting from the target meeting database during the target meeting process. The reasoning module is used to fuse the various text information, multi-dimensional meeting information and reference documents to obtain the composite prompt words of the target meeting, and to reason about the composite prompt words based on a preset large language model to obtain the reasoning result; The parsing module is used to parse the inference results, determine the current meeting status of the target meeting, and push the guidance suggestions in the inference results to all meeting clients of the target meeting through a real-time communication channel.

[0013] Thirdly, embodiments of this application provide an electronic device, including: a processor, a memory, and a bus. The memory stores machine-readable instructions executable by the processor. When the electronic device is running, the processor communicates with the memory via the bus. When the machine-readable instructions are executed by the processor, they perform the steps of any of the methods described in the real-time analysis method for large language models used in conferencing.

[0014] Fourthly, embodiments of this application provide a computer-readable storage medium storing a computer program that, when executed by a processor, performs the steps of any of the methods for real-time analysis of large language models for conferencing.

[0015] This application provides a real-time analysis method for a large language model used in meetings. The method first pre-configures a meeting system. Participants access the target meeting by using their meeting clients, which are assigned corresponding real-time communication channels based on the configured meeting system. The meeting system includes a meeting database. Secondly, during the target meeting, various textual information, multi-dimensional meeting information, and reference documents of the target meeting are extracted in parallel from the target meeting database based on the participants' meeting clients. Then, the various textual information, multi-dimensional meeting information, and reference documents are fused to obtain a composite prompt word for the target meeting. The composite prompt word is then inferred based on a pre-set large language model to obtain an inference result. Finally, the inference result is parsed to determine the current meeting status of the target meeting, and guidance suggestions from the inference result are pushed to all meeting clients of the target meeting through the real-time communication channel. Based on the above methods, this application not only realizes real-time meeting analysis based on a large language model, but also injects pre-related reference document content and multi-dimensional meeting information into the prompt words, enabling the large language model to have multi-dimensional context awareness capabilities, thereby solving the problem of analysis results being out of touch with business scenarios; and pushes guidance suggestions to improve the practical value of the analysis results. Participants can obtain timely status feedback and adjust the direction of the discussion during the discussion, effectively reducing the time wasted due to discussion deviations and progress delays, and meeting the needs of the meeting. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 A flowchart illustrating a real-time analysis method for a large language model used in conferencing, provided in an embodiment of this application, is shown. Figure 2 This paper illustrates a flowchart of another real-time analysis method for large language models in conferencing, provided in an embodiment of this application. Figure 3 This paper illustrates a schematic diagram of the architecture of a conference system provided in an embodiment of this application. Figure 4This illustration shows a structural block diagram of a large language model real-time analysis device for conferencing provided in an embodiment of this application; Figure 5 A structural block diagram of an electronic device provided in an embodiment of this application is shown. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. It should be understood that the accompanying drawings in this application are for illustrative and descriptive purposes only and are not intended to limit the scope of protection of this application. Furthermore, it should be understood that the schematic drawings are not drawn to scale. The flowcharts used in this application illustrate operations implemented according to some embodiments of this application. It should be understood that the operations in the flowcharts may not be implemented in sequence, and steps without logical contextual relationships may be reversed or implemented simultaneously. In addition, those skilled in the art, guided by the content of this application, may add one or more other operations to the flowcharts, or remove one or more operations from the flowcharts.

[0019] Furthermore, the described embodiments are merely some, not all, of the embodiments of this application. The components of the embodiments of this application described and illustrated herein can typically be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0020] It should be noted that the term "comprising" will be used in the embodiments of this application to indicate the presence of the features declared thereafter, but does not exclude the addition of other features.

[0021] Existing LLM meeting support tools (such as Lark Notes) primarily involve transcribing speech into text and submitting it to an LLM database to generate meeting minutes. These tools include four main modules but lack AI analysis during the meeting. They suffer from drawbacks such as difficulty in detecting deviations in discussion and analysis that is detached from the business context, thus failing to meet meeting needs.

[0022] Based on this, embodiments of this application provide a method, apparatus, device, and medium for real-time analysis of large language models for conferences, which are described below through embodiments.

[0023] Example 1 To facilitate understanding of this embodiment, a real-time analysis method for large language models used in conferencing, disclosed in this application embodiment, will first be described in detail. For example... Figure 1The diagram shown illustrates a real-time analysis method for large language models used in conferences. Figure 2 The diagram illustrates another flowchart of a real-time analysis method for large language models used in conferences. This application provides a real-time analysis method for large language models used in conferences, the method comprising: S101. The conference system is pre-configured, and participants access the target conference by using their conference clients to be assigned corresponding real-time communication channels based on the configured conference system; the conference system includes a conference database. S102. During the target meeting, extract in parallel from the target meeting database various text information, multi-dimensional meeting information, and reference documents of the target meeting entered by the participants based on their meeting clients. S103. By integrating the various textual information, multi-dimensional meeting information, and reference documents, a composite prompt word for the target meeting is obtained, and the inference result is obtained by inferring the composite prompt word based on a preset large language model. S104. Analyze the reasoning result, determine the current meeting status of the target meeting, and push the guidance suggestions in the reasoning result to all meeting clients of the target meeting through the real-time communication channel.

[0024] In step S101, as Figure 3 As shown, this application pre-configures a conference system, meaning the system has conference clients corresponding to the number of participants. Before the conference begins, the system allocates a real-time communication channel to each client. These channels are created using a publish / subscribe mechanism based on the conference system's real-time communication module. Each conference has its own dedicated real-time communication channel. One or more participants subscribe to this channel via their conference clients using the SSE protocol, thus establishing the channel. The configured conference system then allocates the corresponding real-time communication channel for accessing the target conference. These real-time communication channels also handle the push notifications of various real-time events: newly submitted text messages from participants, notifications of participants joining or leaving the conference, guidance suggestions generated by real-time AI analysis, and synchronized updates of conference information (such as agenda adjustments).

[0025] When participants connect to the conference system via a conference client, the real-time communication module converts the connection into a persistent SSE connection, maintaining a mapping table in memory with the conference identifier as the key and the connection set as the value. The real-time communication module sends heartbeat messages to each SSE connection at fixed intervals to prevent connections from being closed due to timeouts; it automatically removes the connection from the connection set when a conference client disconnects, and automatically cancels subscriptions to a channel when the connection set for a conference is empty. When participants leave messages, the message processing module in the conference system connects to the real-time communication module to receive concurrent text messages from multiple participants. After performing a rate limit check on the text messages, it stores the text messages, along with the participant identifier and timestamp, in the conference system's conference database and broadcasts them to the conference clients of all participants in the conference via the publish / subscribe channel of the real-time communication module.

[0026] In step S102, during the target meeting, this application uses AI or other intelligent agents to extract text information, multi-dimensional meeting information, and reference documents of the target meeting from the target meeting database in parallel. The text information is obtained by the AI ​​or other intelligent agents through the recent speech extraction module of the meeting system, querying the meeting database for speech records of the target meeting within the target time period, excluding automatically generated messages, retaining only the speeches of genuine participants, and formatting the speeches in chronological order as "Participant Name: The text of the speech content is presented line by line. The multi-dimensional meeting information is obtained by AI and other intelligent agents querying the meeting database for currently ongoing agenda items based on the agenda context extraction module, and extracting their titles, discussion objectives, and end times. The remaining meeting time is calculated based on the difference between the end time and the current time. The reference documents for the target meeting are obtained by AI and other intelligent agents querying the meeting database for pre-associated reference documents for the target meeting through the meeting system's reference document injection module. The reference documents are text materials uploaded and stored by the meeting host during the meeting creation stage, including but not limited to requirement documents, minutes of the previous meeting, project plans, etc. The reference document injection module also formats the retrieved document content into text blocks with identifiers.

[0027] In a specific implementation of step S102, one embodiment includes: during the target meeting, the following steps are taken: S1021. Monitor in real time the amount of text information in the target meeting within a preset time window, as well as the meeting agenda of the target meeting; S1022. Determine whether the quantity and the meeting agenda meet the triggering mechanism for real-time analysis of the target meeting, so as to analyze the target meeting in real time.

[0028] In steps S1021-S1022, during the execution of the target meeting, the conference system of this application monitors in real time the amount of text information and the meeting agenda of the target meeting within a preset time window, and determines whether the amount and the meeting agenda meet the triggering mechanism for real-time analysis of the target meeting. The triggering mechanism is for AI and other intelligent agents, which call a large language model to perform real-time analysis of the target meeting. The specific triggering mechanisms include: (a) adaptive triggering based on speech density: when the conference system detects that the number of new speeches within the preset time window exceeds a threshold, it automatically triggers analysis to provide more timely feedback during the active discussion period; (b) triggering based on agenda switching: when the conference system detects that the current time has entered the time range of the next agenda item, it automatically triggers the analysis and summary of the previous agenda item. When any triggering mechanism meets the condition, the text information, multi-dimensional meeting information, and reference documents of the target meeting are automatically extracted for real-time analysis.

[0029] The triggering mechanism also includes: (c) periodic automatic triggering: the conference system automatically triggers an analysis once at a preset fixed interval (such as every 5 minutes); (d) host manual triggering: the host manually initiates an analysis request through the client interface. In step S103, the composite prompt word construction module of the conference system of this application is connected to the agenda context extraction module, the reference document injection module, and the recent speech extraction module respectively to obtain text information, multi-dimensional conference information, and the reference document of the target conference. The module then integrates the various text information, multi-dimensional conference information, and reference document to obtain the composite prompt word for the target conference. The composite prompt word combines information from three dimensions: multi-dimensional conference information, reference document, and text information. This allows the large language model to simultaneously perceive "what should be discussed," "what is the business background," "what is actually being discussed," and "how much time is left" in a single inference. The composite prompt word construction module is connected to the large language model inference module in the conference system to obtain the inference result based on the preset large language model inference. The large language model inference module sets the temperature parameter to a low value (e.g., 0.3) to ensure the stability and consistency of the state determination result. The large language model inference module supports configuring multiple large language model services as inference backends, and automatically switches to a backup model service when the current large language model service is unavailable.

[0030] In a specific implementation of step S103, one embodiment involves fusing the various textual information, multi-dimensional meeting information, and reference documents to obtain the composite prompt words for the target meeting, including: S10311. Pre-configure the composite prompt words, including system prompt words and user prompt words, and set the corresponding processing methods; S10312. Process the various text information, multi-dimensional meeting information, and reference documents using the processing method described above to obtain the system prompt words and user prompt words.

[0031] In steps S10311-S10312, this application pre-configures the composite prompt words, including system prompt words and user prompt words. The system prompt words are extracted from multi-dimensional meeting information and reference documents of the target meeting, while the user prompt words are based on text information input by participants. Therefore, the composite prompt words integrate system and user prompt words, taking into account both business context and real-time needs of participants, thus solving the problem of existing AI analysis being detached from business scenarios. System prompt words are injected with pre-set materials such as meeting agendas and requirement documents, giving LLM business context awareness capabilities and avoiding mere literal summarization. User prompt words incorporate real-time input from participants, making the analysis more relevant to the meeting's needs and improving the relevance and practicality of the minutes. Furthermore, this application also sets corresponding processing methods for the composite prompt words, including system prompt words and user prompt words, and executes these processing methods to process the various text information, multi-dimensional meeting information, and reference documents to obtain the system prompt words and user prompt words.

[0032] The compound prompt word construction module directly uses the text from the recent speech extraction module as the content of the user prompt word, keeping the line-by-line format of "participant name: speech content" unchanged, and provides it as input material for the large language model to be analyzed.

[0033] In a specific implementation of step S10312, one embodiment is as follows: The processing method is executed to obtain the system prompt and user prompt, including: A1. Inject the multi-dimensional meeting information, the roles set in the large language model, and the content of the reference documents into the preset system prompt word template; A2. Add the analysis instructions obtained by integrating multiple reasoning dimensions to the system prompt word template to generate the system prompt word.

[0034] In steps A1-A2, for system prompt words, the composite prompt word construction module of this application pre-sets a system prompt word template. According to the system prompt word template, the role of the large language model is first set as "experienced meeting host" as a behavioral constraint for the analysis task. Then, the current agenda title, discussion objectives, and remaining time from the multi-dimensional meeting information output by the agenda context extraction module are concatenated into the system prompt word target in a structured text format. Next, the reference document content output by the reference document injection module is inserted into the system prompt word template in an independent text block format with start and end identifiers, so that the large language model can distinguish reference materials from other instructions. Finally, an analysis instruction is attached to the end of the system prompt word template. The analysis instruction is set based on reasoning dimensions such as whether it is focused on the objective, whether it deviates from the topic, and whether it is lagging behind. It requires the large language model to judge whether the target meeting is focused on the objective, whether it deviates from the topic, and whether it is lagging behind. In the specific implementation of step S103, another embodiment is as follows: The reasoning result is obtained by reasoning the compound prompt word based on a preset large language model, including: S10321. Based on the multiple inference dimensions in the system prompt words, perform semantic analysis on the user prompt words in the compound prompt words to obtain the analysis results; S10322. Cross-compare the analysis results with the multi-dimensional meeting information in the system prompts to obtain the reasoning results.

[0035] In steps S10321-S10322, when the large language model reasoning module of this application reasons about the compound prompt words based on the preset large language model, it performs semantic analysis on the user prompt words in the compound prompt words according to the multiple reasoning dimensions corresponding to the system prompt word analysis instructions in the compound prompt words to obtain the analysis results. Specifically, the large language model first parses all user prompt words sentence by sentence, identifies the core nouns, key verbs and semantic relationships in the text, and filters out invalid and redundant information. Then, through semantic clustering, keyword extraction and contextual association analysis, it integrates multiple user prompt words recently entered by the participants, mines the common semantics and core demands, and extracts the core semantic direction of recent speeches, finally forming an accurate analysis result. Based on the analysis result, it clearly defines the core topics and key directions of the participants' discussion, providing accurate and reliable core basis for real-time meeting analysis, deviation detection and progress control. Then, it cross-compares the analysis result with the multi-dimensional meeting information in the system prompt words to obtain the reasoning result, thereby completing the reasoning for the compound prompt words, determining the real-time discussion status and real-time progress of the target meeting, and thus obtaining the reasoning result.

[0036] In a specific implementation of step S10322, one embodiment involves cross-comparing the analysis results with the multi-dimensional meeting information in the system prompts to obtain the inference results, including: B1. Calculate the similarity between the core semantic direction in the analysis results and the current agenda in the multi-dimensional meeting information to determine the real-time status of the discussion; B2. The real-time progress of the meeting is assessed based on the remaining meeting time and the current agenda in the multi-dimensional meeting information, and integrated with the real-time status of the discussion to form the reasoning result.

[0037] In steps B1-B2, based on the analysis results, this application calculates the similarity between the core semantic direction in the analysis results and the discussion target in the current agenda of the multi-dimensional meeting information to determine the real-time status of the discussion. If the main semantic direction of recent speeches is consistent with the discussion target, the discussion status is determined to be on_track, i.e., the meeting is proceeding normally. If the topic of recent speeches deviates from the field involved in the discussion target (e.g., a large number of discussions about "marketing promotion" appear in the "requirements review" agenda), it is determined to be off_topic, i.e., the meeting has deviated from the topic. In addition, the large language model combines the remaining time information in the system prompt words to evaluate the discussion progress: if the discussion has not deviated from the topic but has not yet touched on the core decision-making matters of the discussion target, and the remaining time is insufficient, such as less than 5 minutes, it is determined to be behind_schedule, i.e., the progress is lagging. The large language model also generates guidance suggestions based on the three inferred meeting states, namely, meeting proceeding normally, meeting deviating from the topic, and progress lagging, and generates inference results based on the status judgment fields on_track, off_topic, and behind_schedule corresponding to meeting proceeding normally, meeting deviating from the topic, and progress lagging, as well as the guidance suggestion field (thought). The status judgment fields can also include "repetitive" for "discussion getting stuck in a rut" and "consensus_reached" for "consensus reached," or they can be reduced to two types (normal / abnormal), depending on the application scenario. For scenarios where the meeting is proceeding normally, encouraging content or summaries of current key points will be generated, judged as a positive and gentle encouraging emotional state, ensuring the guiding tone aligns with the normal progress of the meeting. For scenarios where the meeting deviates from the topic, gentle reminders to return to the intended topic will be generated, judged as a calm and tactful gentle reminder emotional state, avoiding overly harsh reminders that might cause resistance from participants. For scenarios where the meeting is lagging behind, more urgent guidance will be generated, judged as a serious and urgent urgent reminder emotional state, highlighting the urgency of the delay through the differentiation of tone intensity, urging participants to speed up the discussion, focus on the meeting agenda, and ensure the target meeting proceeds as planned.

[0038] In step S104, the result parsing module of the conference system of this application connects with the large language model inference module to receive the inference result output by the large language model inference module, parse the inference result, extract the guidance suggestion field (thought) and the status determination field (status), and determine the current conference status of the target conference based on the status determination field in the inference result. If JSON parsing fails, the original output text of the model is used as the guidance suggestion, and the status is set to a neutral value. The guidance suggestion in the inference result is pushed to all conference clients of the target conference through the result push module of the conference system via the real-time communication channel for display, and to remind the participants holding the conference client to conduct the conference according to the guidance suggestion. That is, the conference system completes this real-time analysis process. The conference system repeats the above real-time analysis process of the target conference until the target conference ends. This application takes conference management as the anchor point and, through the triple innovation of prompt word engineering, state machine design, and communication architecture, collaboratively constructs a new paradigm of AI conference collaboration that is "business-understandable, process-interventionable, and system-trustworthy".

[0039] In a specific implementation of step S104, one embodiment involves pushing guidance suggestions from the inference results to all meeting clients of the target meeting via a real-time communication channel, including: S1041. Pre-configure the meeting client to enable the meeting client to determine the emotional state contained in the guidance suggestion; S1042. Based on the emotional state, match the target visual style from a pre-set visual library to display the guidance suggestion.

[0040] In steps S1041-S1042, when configuring the conference system, this application also configures the corresponding conference client. For example, an emotion state recognition model is configured in the conference client to perform semantic analysis on the text content of the guidance suggestion, so that the conference client can determine the emotion state contained in the guidance suggestion, namely, encouragement, gentle reminder, or urgent reminder. After the recognition is completed, the client will quickly match the corresponding target visual style from the pre-set visual library based on the recognized emotion state, and finally display the guidance suggestion with the matched visual style, so that the participants can intuitively perceive the emotional tendency of the guidance suggestion through visual perception, thereby improving the communication efficiency and acceptance of the guidance suggestion. If the meeting proceeds normally, the guidance suggestion is "The current discussion is on track. Please keep it up and focus on the details of the requirements review." The client recognizes this suggestion as an encouraging emotional state and will match a warm yellow background, rounded gray border, smiley face icon, and regular font from the visual library to display the guidance suggestion in this visual style. If the meeting deviates from the topic, the guidance suggestion is "Friendly reminder: The current agenda is the requirements review. We suggest you return to the core topic to avoid going off-topic." The client recognizes this as a gentle reminder emotional state and will match a light blue background, simple thin border, ordinary prompt icon, and regular font. If the meeting is behind schedule, the guidance suggestion is "Attention! The meeting is 15 minutes behind schedule. Please speed up the discussion and prioritize finalizing the core decisions." The client recognizes this as an urgent reminder emotional state and will match a light red background, thick border, warning triangle icon, and bold font to make participants intuitively feel the pressure of the schedule.

[0041] Example 2 This application also provides a real-time analysis device for large language models in conferences, such as... Figure 4 The diagram shows a block diagram of a real-time large language model analysis device for conferences. This device performs functions corresponding to the steps described above in executing a real-time large language model analysis method for conferences on a terminal device. The device can be understood as a server component including a processor. The real-time large language model analysis device for conferences described in this application includes: Configuration module 401 is used to pre-configure the conference system. Participants use their conference clients to access the target conference by being assigned a corresponding real-time communication channel based on the configured conference system. The conference system includes a conference database. The input module 402 is used to extract, in parallel from the target meeting database, various text information, multi-dimensional meeting information, and reference documents of the target meeting input by participants based on their meeting clients during the target meeting process. The reasoning module 403 is used to fuse the various text information, multi-dimensional meeting information and reference documents to obtain the composite prompt words of the target meeting, and to reason about the composite prompt words based on a preset large language model to obtain the reasoning result; The parsing module 404 is used to parse the inference result, determine the current meeting status of the target meeting, and push the guidance suggestions in the inference result to all meeting clients of the target meeting through the real-time communication channel.

[0042] In one feasible implementation, the inference module includes: The composite prompt words are pre-configured, including system prompt words and user prompt words, and the corresponding processing methods are set; The processing method described above is used to process the various text information, multi-dimensional meeting information, and reference documents to obtain the system prompt words and user prompt words.

[0043] In one feasible implementation, the inference module further includes: The multi-dimensional meeting information, the roles set in the large language model, and the content of the reference documents are injected and concatenated into the preset system prompt word template; Add analysis instructions obtained by integrating multiple reasoning dimensions to the system prompt word template to generate the system prompt word.

[0044] In one feasible implementation, the inference module also includes: Based on multiple inference dimensions in the system prompts, semantic analysis is performed on the user prompts in the composite prompts to obtain the analysis results; The inference result is obtained by cross-referencing the analysis results with the multi-dimensional meeting information in the system prompts.

[0045] In one feasible implementation, the inference module further includes: The similarity between the core semantic direction in the analysis results and the current agenda in the multi-dimensional meeting information is calculated to determine the real-time status of the discussion; The real-time progress of the meeting is assessed based on the remaining meeting time and the current agenda from multi-dimensional meeting information, and then integrated with the real-time status of the discussion to form the reasoning result. In one feasible implementation, the parsing module includes: The meeting client is pre-configured to determine the emotional state contained in the guidance suggestions; Based on the emotional state, a target visual style is matched from a pre-defined visual library to display the guidance suggestion.

[0046] In one feasible implementation, the input module includes: Real-time monitoring of the amount of text information in the target meeting within a preset time window, as well as the meeting agenda of the target meeting; Determine whether the quantity and the meeting agenda meet the triggering mechanism for real-time analysis of the target meeting, so as to analyze the target meeting in real time.

[0047] Example 3 This application also provides an electronic device, such as Figure 5 As shown, it includes: a processor 501, a memory 502, and a bus 503. The memory 502 stores machine-readable instructions that can be executed by the processor 501. When the electronic device is running, the processor 501 and the memory 502 communicate through the bus 503. When the machine-readable instructions are executed by the processor 501, the steps of any one of the methods for real-time analysis of a large language model for conferencing are performed.

[0048] Example 4 This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, performs the steps of any one of the methods for real-time analysis of a large language model for conferencing.

[0049] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems and devices described above can be referred to the corresponding processes in the method embodiments, and will not be repeated here. In the several embodiments provided in this application, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed mutual coupling or direct coupling or communication connection can be through some communication interfaces; the indirect coupling or communication connection of devices or modules can be electrical, mechanical, or other forms.

[0050] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0051] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0052] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a processor-executable, non-volatile, computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, a platform server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, ROM, RAM, magnetic disks, or optical disks.

[0053] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A real-time analysis method for large language models used in conferences, characterized in that, The method includes: A pre-configured conference system allows participants to access the target conference by using their conference clients and being assigned corresponding real-time communication channels based on the configured system; the conference system includes a conference database. During the target meeting, various text information, multi-dimensional meeting information, and reference documents of the target meeting are extracted in parallel from the target meeting database based on the meeting client held by the participants. By integrating the various textual information, multi-dimensional meeting information, and reference documents, a composite prompt word for the target meeting is obtained, and the inference result is obtained by inferring the composite prompt word based on a pre-set large language model; The reasoning results are analyzed to determine the current meeting status of the target meeting, and guidance suggestions from the reasoning results are pushed to all meeting clients of the target meeting through a real-time communication channel.

2. The method according to claim 1, characterized in that, By integrating the various textual information, multi-dimensional meeting information, and reference documents, a composite prompt for the target meeting is obtained, including: The composite prompt words are pre-configured, including system prompt words and user prompt words, and the corresponding processing methods are set; The processing method described above is used to process the various text information, multi-dimensional meeting information, and reference documents to obtain the system prompt words and user prompt words.

3. The method according to claim 2, characterized in that, By performing the aforementioned processing method, the system prompt and user prompt are obtained, including: The multi-dimensional meeting information, the roles set in the large language model, and the content of the reference documents are injected and concatenated into the preset system prompt word template; Add analysis instructions obtained by integrating multiple reasoning dimensions to the system prompt word template to generate the system prompt word.

4. The method according to claim 1, characterized in that, The reasoning results obtained based on the compound prompt words inference using a pre-built large language model include: Based on multiple inference dimensions in the system prompts, semantic analysis is performed on the user prompts in the composite prompts to obtain the analysis results; The inference result is obtained by cross-referencing the analysis results with the multi-dimensional meeting information in the system prompts.

5. The method according to claim 4, characterized in that, By cross-referencing the analysis results with the multi-dimensional meeting information in the system prompts, the inference results are obtained, including: The similarity between the core semantic direction in the analysis results and the current agenda in the multi-dimensional meeting information is calculated to determine the real-time status of the discussion; The real-time progress of the meeting is assessed based on the remaining meeting time and the current agenda from multi-dimensional meeting information, and then integrated with the real-time status of the discussion to form the reasoning result.

6. The method according to claim 1, characterized in that, The guidance suggestions from the inference results are pushed to all meeting clients of the target meeting via a real-time communication channel, including: The meeting client is pre-configured to determine the emotional state contained in the guidance suggestions; Based on the emotional state, a target visual style is matched from a pre-defined visual library to display the guidance suggestion.

7. The method according to claim 1, characterized in that, During the target meeting, the following were included: Real-time monitoring of the amount of text information in the target meeting within a preset time window, as well as the meeting agenda of the target meeting; Determine whether the quantity and the meeting agenda meet the triggering mechanism for real-time analysis of the target meeting, so as to analyze the target meeting in real time.

8. A real-time analysis device for a large language model used in conferences, characterized in that, The device includes: A configuration module is used to pre-configure the conference system. Participants use their conference clients to access the target conference by being assigned a corresponding real-time communication channel based on the configured conference system. The conference system includes a conference database. The input module is used to extract, in parallel, various text information, multi-dimensional meeting information, and reference documents of the target meeting from the target meeting database during the target meeting process. The reasoning module is used to fuse the various text information, multi-dimensional meeting information and reference documents to obtain the composite prompt words of the target meeting, and to reason about the composite prompt words based on a preset large language model to obtain the reasoning result; The parsing module is used to parse the inference results, determine the current meeting status of the target meeting, and push the guidance suggestions in the inference results to all meeting clients of the target meeting through a real-time communication channel.

9. An electronic device, characterized in that, include: The device includes a processor, a memory, and a bus. The memory stores machine-readable instructions executable by the processor. When the electronic device is running, the processor communicates with the memory via the bus. When the machine-readable instructions are executed by the processor, they perform the steps of a real-time analysis method for a large language model for conferencing as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, performs the steps of a real-time analysis method for a large language model for conferencing as described in any one of claims 1 to 7.

Images