Methods, devices, electronic equipment, and storage media for processing meeting content

By dynamically configuring multiple independent meeting minutes generation and evaluation models, the problems of low meeting minutes generation quality and poor stability are solved, achieving high-quality and complete meeting minutes generation.

CN123115353APending Publication Date: 2026-11-17BEIJING DAJIA INTERNET INFORMATION TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611486607.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-09-22
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

In existing technologies, the quality of meeting minutes generation is low and not stable enough, especially for long meetings where information is easily missed.

Method used

By acquiring information on the content volume, target text, and context of the meeting, multiple independent meeting minutes generation and evaluation models are dynamically configured to perform multi-dimensional evaluation and processing, generate multiple candidate meeting minutes, and select or merge the target meeting minutes based on the evaluation results.

Benefits of technology

It improved the quality and stability of meeting minutes generation, avoided the omission of key information, enhanced the completeness and accuracy of the generated minutes, and prevented model illusion.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123115353A_ABST
    Figure CN123115353A_ABST
Patent Text Reader

Abstract

This disclosure relates to a method, apparatus, electronic device, and storage medium for processing meeting content. The method includes: acquiring content volume information, target text information, and context information corresponding to the meeting content to be processed; determining an evaluation model and multiple meeting minutes generation models based on the content volume information; inputting the target text information and context information into each of the multiple meeting minutes generation models to perform meeting minutes generation processing, obtaining multiple first candidate meeting minutes information; inputting each first candidate meeting minutes information into the evaluation model, performing multi-dimensional evaluation processing respectively, obtaining evaluation results corresponding to each first candidate meeting minutes information; and determining the target meeting minutes information corresponding to the meeting content to be processed based on the first candidate meeting minutes information and the evaluation results corresponding to the first candidate meeting minutes information. This disclosure can effectively improve the quality of generated meeting minutes.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This disclosure relates to the field of artificial intelligence technology, and in particular to methods, apparatus, electronic devices and storage media for processing conference content. Background Technology

[0002] In daily office work, it's essential to transcribe meeting minutes. Manual transcription is inefficient, prone to omissions, and inconsistent in quality. With the development of large language model technology, several AI-based meeting minutes generation applications have emerged, offering functions such as transcribing meeting recordings and intelligent summarization. However, the quality of currently generated meeting minutes remains low and unstable, especially for long meetings where information is easily missed. Summary of the Invention

[0003] This disclosure provides methods, apparatus, electronic devices, and storage media for processing meeting content, to at least solve the aforementioned problems in the related art. The technical solution of this disclosure is as follows: According to a first aspect of the present disclosure, a method for processing meeting content is provided, comprising: Obtain the content volume information, target text information, and context information corresponding to the meeting content to be processed; the context information indicates the meeting background of the meeting content to be processed. Based on the content volume information, an evaluation model and multiple meeting minutes generation models are determined; the multiple meeting minutes generation models are independent of each other and are not the same. The target text information and the context information are input into each of the multiple meeting minutes generation models to perform meeting minutes generation processing, and multiple first candidate meeting minutes information are obtained. Each first candidate meeting minutes is input into the evaluation model and subjected to multi-dimensional evaluation processing to obtain the evaluation result corresponding to each first candidate meeting minutes. Based on the meeting minutes information of each first candidate meeting and the evaluation results corresponding to each first candidate meeting minutes information, the target meeting minutes information corresponding to the meeting content to be processed is determined.

[0004] Optionally, the step of inputting the target text information and the context information into each of the multiple meeting minutes generation models to perform meeting minutes generation processing, and obtaining multiple first candidate meeting minutes information, includes: The target text information and the context information are fused together to obtain target fused information; Based on the target fusion information, the target meeting type of the meeting content to be processed is determined; Determine the minutes template information corresponding to the meeting type; The target fusion information and the minutes template information are input into each meeting minutes generation model to perform meeting minutes generation processing, thereby obtaining the multiple first candidate meeting minutes information.

[0005] Optionally, determining the target meeting type of the meeting content to be processed based on the target fusion information includes: The meeting title information of the meeting content to be processed is determined from the target fusion information; The meeting title information and at least one keyword indicating the meeting type are matched to obtain the matching result; If the matching result indicates that there is a target keyword among the at least one keyword that matches the meeting title information, the target meeting type of the meeting content to be processed is determined based on the target keyword. If the matching result indicates that none of the at least one keywords matches the meeting title information, the target fusion information is input into a preset large model for semantic recognition processing to determine the target meeting type of the meeting content to be processed.

[0006] Optionally, the method further includes: If the content volume information indicates that the meeting duration is less than a first preset duration or the number of words in the target text information is less than a first preset number of words, a single meeting minutes generation model is determined; The target text information and the context information are input into a single meeting minutes generation model to perform meeting minutes generation processing, thereby obtaining the target meeting minutes information corresponding to the meeting content to be processed.

[0007] Optionally, determining the evaluation model and multiple meeting minutes generation models based on the content volume information includes: When the content volume information indicates that the meeting duration is between a first preset duration and a second preset duration, or when the number of words in the target text information is between a first preset number of words and a second preset number of words, the evaluation model and the multiple meeting minutes generation models are determined, wherein the number of multiple meeting minutes generation models is two; the first preset duration is less than the second preset duration; and the first preset number of words is less than the second preset number of words. If the content volume information indicates that the meeting duration is not less than the second preset duration, or if the number of words in the target text information is not less than the second preset number of words, then the evaluation model and the plurality of meeting minutes generation models are determined, and the number of the plurality of meeting minutes generation models is at least three.

[0008] Optionally, determining the target meeting minutes information corresponding to the meeting content to be processed based on the evaluation results corresponding to each first candidate meeting minutes information includes: If the evaluation results corresponding to each first candidate meeting minutes indicate that there is a target first candidate meeting minutes among the multiple first candidate meeting minutes that the evaluation results corresponding to the target first candidate meeting minutes meet the preset single selection condition, then the target meeting minutes are determined based on the target first candidate meeting minutes. If the evaluation results corresponding to each first candidate meeting minutes indicate that none of the first candidate meeting minutes satisfy the preset single selection condition, the first candidate meeting minutes are merged based on the evaluation results corresponding to each first candidate meeting minutes to obtain the target meeting minutes.

[0009] Optionally, the method further includes: Based on the structured target meeting minutes information, at least one to-do item is identified, along with item information and traceability information corresponding to each to-do item in the at least one to-do item. Based on the item information and traceability information corresponding to each pending item, generate item record information corresponding to each pending item and store it in the database; In response to a notification instruction for a target to-do item, the item information corresponding to the target to-do item is sent to the target object, wherein the target to-do item is any one of the at least one to-do items; Determine the execution status information corresponding to the target pending item; Based on the execution status information corresponding to the target to-do item, update the item record information corresponding to the target to-do item stored in the database.

[0010] Optionally, the method further includes: Based on the structured target meeting minutes information, at least one to-do item is identified, along with item information and traceability information corresponding to each to-do item in the at least one to-do item. Based on the item information and traceability information corresponding to each pending item, generate item record information corresponding to each pending item and store it in the database; In response to a notification instruction for a target to-do item, the item information corresponding to the target to-do item is sent to the target object, wherein the target to-do item is any one of the at least one to-do items; Determine the execution status information corresponding to the target pending item; Based on the execution status information corresponding to the target to-do item, update the item record information corresponding to the target to-do item stored in the database.

[0011] Optionally, after obtaining the target text information corresponding to the meeting content to be processed and the context information corresponding to the meeting content to be processed, the method further includes: The target text information is divided into multiple fragment text information; The text information of each segment in the multiple segment text information is fused with the context information to obtain multiple segment fused information; Each segment fusion information from the multiple segment fusion information is input into the target meeting minutes generation model for meeting minutes generation processing to obtain the third candidate meeting minutes information corresponding to each segment fusion information. The third candidate meeting minutes information corresponding to the fusion information of each segment is merged to obtain the target meeting minutes information corresponding to the meeting content to be processed.

[0012] Optionally, obtaining the target text information corresponding to the meeting content to be processed and the context information corresponding to the meeting content to be processed includes: When the meeting content to be processed is an audio file, the meeting content to be processed is subjected to speech recognition processing based on a pre-configured domain hot word list to obtain the first text information; The first text information is cleaned to obtain the target text information.

[0013] Optionally, the method further includes: If the meeting content to be processed is a meeting video file and the meeting duration of the meeting content to be processed is longer than a third preset duration, the meeting content to be processed is intelligently segmented to obtain multiple video sub-files; Each video sub-file in the plurality of video sub-files is transcoded one by one to obtain the corresponding audio sub-file, and the corresponding audio sub-file is uploaded to the object storage space one by one. Retrieve the audio sub-files corresponding to each video sub-file from the object storage space; Merge the audio sub-files corresponding to each video sub-file into a merged audio file; Based on the aforementioned domain hot word list, the merged audio file is subjected to speech recognition processing to obtain second text information; The second text information is cleaned to obtain the target text information.

[0014] According to a second aspect of the present disclosure, an apparatus for processing meeting content is provided, comprising: The acquisition module is configured to acquire the content volume information, target text information, and context information corresponding to the meeting content to be processed; the context information indicates the meeting background of the meeting content to be processed. The model determination module is configured to determine an evaluation model and multiple meeting minutes generation models based on the content volume information; the multiple meeting minutes generation models are independent of each other and are not the same. The meeting minutes information generation module is configured to input the target text information and the context information into each of the multiple meeting minutes generation models, perform meeting minutes generation processing, and obtain multiple first candidate meeting minutes information; The evaluation module is configured to input each first candidate meeting minutes from the plurality of first candidate meeting minutes into the evaluation model, perform multi-dimensional evaluation processing, and obtain the evaluation result corresponding to each first candidate meeting minutes. The target meeting minutes information determination module is configured to determine the target meeting minutes information corresponding to the meeting content to be processed based on the evaluation results corresponding to each first candidate meeting minutes information and the first candidate meeting minutes information.

[0015] Optionally, the meeting minutes information generation module includes: The information fusion unit is configured to perform a fusion process on the target text information and the context information to obtain target fused information; The meeting type determination unit is configured to determine the target meeting type of the meeting content to be processed based on the target fusion information; The minutes template determination unit is configured to determine the minutes template information corresponding to the meeting type. The first meeting minutes generation unit is configured to input the target fusion information and the minutes template information into each meeting minutes generation model, perform meeting minutes generation processing, and obtain the multiple first candidate meeting minutes information.

[0016] Optionally, the meeting type determination unit includes: The meeting title determination subunit is configured to determine the meeting title information of the meeting content to be processed from the target fusion information; The matching subunit is configured to perform matching processing on the meeting title information and at least one keyword indicating the meeting type, and obtain the matching result; The first determining subunit is configured to determine the target meeting type of the meeting content to be processed based on the target keyword when the matching result indicates that there is a target keyword that matches the meeting title information among the at least one keyword; The second determining subunit is configured to, when the matching result indicates that no keyword among the at least one keyword matches the meeting title information, input the target fusion information into a preset large model, perform semantic recognition processing, and determine the target meeting type of the meeting content to be processed.

[0017] Optionally, the model determination module includes: The first model configuration unit is configured to determine a single meeting minutes generation model when the content volume information indicates that the meeting duration is less than a first preset duration or the number of words in the target text information is less than a first preset number of words. The second meeting minutes generation unit is configured to input the target text information and the context information into a single meeting minutes generation model, perform meeting minutes generation processing, and obtain the target meeting minutes information corresponding to the meeting content to be processed.

[0018] Optionally, the model determination module includes: The second model configuration unit is configured to determine the evaluation model and the plurality of meeting minutes generation models when the content volume information indicates that the meeting duration is between a first preset duration and a second preset duration, or when the number of words in the target text information is between a first preset number of words and a second preset number of words; the number of the plurality of meeting minutes generation models is two; the first preset duration is less than the second preset duration; the first preset number of words is less than the second preset number of words; The third model configuration unit is configured to determine the evaluation model and the plurality of meeting minutes generation models when the content volume information indicates that the meeting duration is not less than the second preset duration, or the number of words of the target text information is not less than the second preset number of words. The number of the plurality of meeting minutes generation models is at least three.

[0019] Optionally, the target meeting minutes information determination module includes: The first determining unit is configured to determine the target meeting minutes information based on the target first candidate meeting minutes information when the evaluation result corresponding to each of the multiple first candidate meeting minutes information indicates that the evaluation result corresponding to the target first candidate meeting minutes information satisfies a preset single selection condition. The second determining unit is configured to perform a merging process on the evaluation results corresponding to each first candidate meeting minutes information to obtain the target meeting minutes information when the evaluation results corresponding to each first candidate meeting minutes information indicate that none of the evaluation results corresponding to any of the multiple first candidate meeting minutes information satisfy a preset single selection condition.

[0020] Optionally, the device includes: The to-do item determination module is configured to execute, based on the structured target meeting minutes information, to determine at least one to-do item and the item information and traceability information corresponding to each to-do item in the at least one to-do item; The storage module is configured to generate and store the corresponding task record information for each task in the database based on the task information and traceability information for each task. The push module is configured to execute a notification instruction in response to a target to-do item, and send the item information corresponding to the target to-do item to the target object, wherein the target to-do item is any one of the at least one to-do items; The execution status determination module is configured to determine the execution status information corresponding to the target pending item. The update module is configured to update the item record information corresponding to the target to-do item stored in the database based on the execution status information corresponding to the target to-do item.

[0021] Optionally, the device includes: The third meeting minutes generation unit is configured to, after obtaining the target text information corresponding to the meeting content to be processed and the context information corresponding to the meeting content to be processed, input the target text information and the context information multiple times into the target meeting minutes generation model to perform meeting minutes generation processing and obtain multiple second candidate meeting minutes information; the target meeting minutes generation model uses a different random seed each time it performs meeting minutes generation processing; The third determining unit is configured to determine the target meeting minutes information corresponding to the meeting content to be processed based on the plurality of second candidate meeting minutes information.

[0022] Optionally, the device further includes: The text segmentation unit is configured to, after obtaining the target text information corresponding to the meeting content to be processed and the context information corresponding to the meeting content to be processed, segment the target text information into multiple fragment text information. The segment fusion unit is configured to perform fusion processing on each segment text information in the plurality of segment text information and the context information to obtain multiple segment fusion information; The fourth meeting minutes generation unit is configured to input the fusion information of each segment from the multiple fusion information into the target meeting minutes generation model, perform meeting minutes generation processing, and obtain the third candidate meeting minutes information corresponding to each fusion information; The fourth determining unit is configured to perform merging processing on the third candidate meeting minutes information corresponding to the fusion information of each segment, so as to obtain the target meeting minutes information corresponding to the meeting content to be processed.

[0023] Optionally, the acquisition module includes: The first speech recognition unit is configured to perform speech recognition processing on the meeting content to be processed based on a pre-configured domain hot word list when the meeting content to be processed is a meeting audio file, so as to obtain first text information; The first text cleaning unit is configured to perform text cleaning processing on the first text information to obtain the target text information.

[0024] Optionally, the device further includes: The video segmentation unit is configured to intelligently segment the meeting content to be processed into multiple video sub-files when the meeting content to be processed is a meeting video file and the meeting duration of the meeting content to be processed is longer than a third preset duration. The video transcoding unit is configured to transcode each video sub-file in the plurality of video sub-files one by one to obtain the audio sub-files corresponding to each video sub-file, and upload the audio sub-files corresponding to each video sub-file to the object storage space one by one. The audio acquisition unit is configured to retrieve the audio sub-files corresponding to each video sub-file from the object storage space; The audio merging unit is configured to merge the audio subfiles corresponding to each video subfile into a merged audio file. The second speech recognition unit is configured to perform speech recognition processing on the merged audio file based on the domain hot word list to obtain second text information; The second text cleaning unit is configured to perform text cleaning processing on the second text information to obtain the target text information.

[0025] According to a third aspect of the present disclosure, an electronic device is provided, comprising: a processor; and a memory for storing processor-executable instructions; wherein the processor is configured to execute the instructions to implement the method for processing conference content as described in any one of the first aspects of the present disclosure.

[0026] According to a fourth aspect of the present disclosure, a computer-readable storage medium is provided that, when instructions in the computer-readable storage medium are executed by a processor of an electronic device, enables the electronic device to perform a method for processing conference content as described in any one of the first aspects of the present disclosure.

[0027] According to a fifth aspect of the present disclosure, a computer program product is provided, including computer instructions that, when executed by a processor, implement a method for processing meeting content as described in any one of the first aspects of the present disclosure.

[0028] The technical solutions provided in this disclosure offer at least the following beneficial effects: In the technical solution provided in this disclosure, the content volume information, target text information, and context information corresponding to the meeting content to be processed are first obtained, wherein the context information indicates the meeting background of the meeting content to be processed; this disclosure dynamically configures an evaluation model and multiple meeting minutes generation models based on the content volume information, and the multiple meeting minutes generation models are independent and different from each other, which can avoid excessive waste of resources; this disclosure obtains multiple first candidate meeting minutes information by inputting the target text information and context information into each of the multiple meeting minutes generation models for meeting minutes generation processing, and the parallel generation processing of multiple meeting minutes generation models. This approach ensures the completeness and stability of meeting minutes generation, preventing the omission of key information due to the large amount of content in a single model, thus guaranteeing the overall generation quality. This disclosure inputs the information from multiple first-candidate meeting minutes into the evaluation model, performing multi-dimensional evaluation processing to obtain the corresponding evaluation results. Based on these results, selection or merging can be performed to obtain higher-quality target meeting minutes, improving the overall quality of meeting minutes generation and effectively avoiding model illusions in the content production process.

[0029] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0030] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure, and are not intended to unduly limit this disclosure.

[0031] Figure 1 This is a schematic diagram illustrating an application environment according to an exemplary embodiment; Figure 2 This is a flowchart illustrating a method for processing meeting content according to an exemplary embodiment; Figure 3 This is a flowchart illustrating a speech recognition process for video content to be processed, according to an exemplary embodiment. Figure 4 This is a flowchart illustrating the processing of a to-do item according to an exemplary embodiment; Figure 5 This is a schematic diagram of a system architecture for implementing the above-described method for processing meeting content, according to an exemplary embodiment. Figure 6 This is a flowchart illustrating the processing flow of meeting content in a specific application scenario according to an exemplary embodiment; Figure 7 This is a block diagram illustrating a conference content processing apparatus according to an exemplary embodiment; Figure 8 This is a block diagram illustrating an apparatus for implementing a method for processing conference content, according to an exemplary embodiment. Figure 9 This is a block diagram of an apparatus for implementing a method for processing conference content, according to an exemplary embodiment. Detailed Implementation

[0032] To enable those skilled in the art to better understand the technical solutions of this disclosure, the technical solutions in the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings.

[0033] It should be noted that the terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this disclosure are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this disclosure described herein can be implemented in orders other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this disclosure as detailed in the appended claims.

[0034] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for display, data used for analysis, etc.) involved in this disclosure are all information and data authorized by the user or fully authorized by all parties.

[0035] Please see Figure 1 The diagram illustrates an application environment for a verification code request processing method according to an exemplary embodiment. The application environment may include a terminal 110, a server 120, and a terminal 130. The terminal 110 and the server 120, and the terminal 130 and the server 120, can be connected via a wired network or a wireless network.

[0036] Terminals 110 and 130 can be smartphones, tablets, laptops, desktop computers, etc., but are not limited to these. Applications (Apps) can be installed on terminals 110 and 130. These applications can be standalone applications or subroutines within standalone applications. Users of terminals 110 and 130 can log in to the applications using pre-registered user information, which may include an account and password. Server 120 can be a server providing background services for the applications in terminals 110 and 130, or it can be another server connected and communicating with the background server of the applications. It can be a single server or a server cluster consisting of multiple servers.

[0037] In an exemplary embodiment of this disclosure, terminal 110 can be used to record a meeting and obtain meeting content to be processed. The meeting content to be processed can be an audio file or a video file, and can be used to transmit the meeting content to be processed to server 120. Server 120 can determine corresponding content volume information, target text information, and context information based on the meeting content to be processed, wherein the context information can indicate the meeting background of the meeting content to be processed, such as participants, organization, meeting documents, etc. Based on the content volume information, server 120 determines an evaluation model and multiple meeting minutes generation models, wherein the multiple meeting minutes generation models are independent of each other and are different. Server 120 inputs the target text information and context information into each of the multiple meeting minutes generation models to perform meeting minutes generation processing, obtaining multiple first candidate meeting minutes information. Then, it inputs each of the first candidate meeting minutes information into the evaluation model for multi-dimensional evaluation processing to obtain the evaluation results corresponding to each first candidate meeting minutes information. Based on the first candidate meeting minutes information and the evaluation results corresponding to each first candidate meeting minutes information, the target meeting minutes information corresponding to the meeting content to be processed can be determined. The above embodiments dynamically configure the evaluation model and multiple meeting minutes generation models based on the content volume information. These multiple meeting minutes generation models are independent and different from each other, which can avoid excessive waste of resources. The parallel generation processing of multiple meeting minutes generation models can ensure the integrity and stability of meeting minutes generation, and will not cause the problem of a single model missing key information due to the large amount of content, thus ensuring the overall generation quality. Based on the evaluation results of each first candidate meeting minutes information and each first candidate meeting minutes information, selection or merging is performed to obtain higher quality target meeting minutes information, thereby improving the quality of meeting minutes generation and effectively avoiding the model illusion phenomenon in the content production process.

[0038] In another exemplary embodiment of this disclosure, the evaluation model and multiple meeting minutes generation models can also run on terminal 110, thereby enabling local generation on terminal 110.

[0039] In one exemplary embodiment of this disclosure, server 120 sends the target meeting minutes information corresponding to the finally determined meeting content to be processed to terminal 130 (such as the terminal corresponding to the participant's account) or other server (such as enterprise document service).

[0040] Figure 2 This is a flowchart illustrating a method for processing meeting content according to an exemplary embodiment, such as... Figure 2 As shown, the method may include the following steps: In step S201, the content volume information, target text information, and context information corresponding to the meeting content to be processed are obtained.

[0041] In this embodiment of the disclosure, the meeting content to be processed can be an audio file or a video file obtained by recording the meeting. The aforementioned content volume information can indicate the size of the meeting content to be processed, for example, the meeting duration, the file size of the meeting document, the number of words in the target text information obtained after speech recognition, etc.; the aforementioned context information indicates the meeting background of the meeting content to be processed, for example, it can include, but is not limited to, the identity information of the participants, the information of the meeting organization, the attribute information of the meeting document, the identification information of the associated meeting, etc.; the aforementioned target text information is the recognition result obtained after performing speech recognition processing (Automatic Speech Recognition, ASR) on the meeting content to be processed, and can indicate the meeting content information, for example, it can include, but is not limited to, the meeting topic, the meeting matters, and the speech content of each participant, etc.

[0042] In a specific embodiment, step S201 above can be implemented as follows: In step S2011, when the meeting content to be processed is a meeting audio file, speech recognition processing is performed on the meeting content to be processed based on a pre-configured domain hot word list to obtain the first text information.

[0043] The domain hot words list can include proprietary terms, product names, personal names, and other words within a specific domain or organization.

[0044] Considering that general-purpose speech recognition (ASR) models have low recognition rates for proper nouns or product names, and that retraining specialized ASR models is costly and has low reusability, the conference audio files and a pre-configured domain hot word list can be input into the general-purpose ASR model. The domain hot word list assists the ASR model in performing speech recognition processing on the conference content to be processed, enhancing the accuracy of hot word recognition, thereby improving the accuracy and quality of the first text information obtained after speech recognition.

[0045] In an optional embodiment, the speech recognition processing can be performed automatically on the meeting content to be processed after the meeting ends and the complete meeting content to be processed is recorded. Alternatively, speech recognition processing can be performed synchronously during the recording process to improve the efficiency of meeting minutes generation.

[0046] In an optional embodiment, the first text information may include audio text data organized in chronological order and corresponding to each speaker at the meeting.

[0047] In step S2012, the first text information is cleaned to obtain the target text information.

[0048] It is understandable that speeches at meetings may contain slips of the tongue, interjections, and repeated words. By performing text cleaning on the primary text information, such as removing interjections, correcting pronoun errors, and removing repeated sentences, text redundancy can be reduced and text quality improved.

[0049] In the above embodiments, the domain hot word list is used to assist the speech recognition model in performing speech recognition processing on the meeting content to be processed, thereby enhancing the accuracy of hot word recognition and improving the accuracy and text quality of the first text information obtained after speech recognition; and by performing text cleaning processing on the first text information, the text quality of the obtained target text information can be further improved.

[0050] In a specific embodiment, step S201 above can be implemented as follows: In step S2013, if the meeting content to be processed is a meeting video file and the meeting duration of the meeting content to be processed is longer than the third preset duration, the meeting content to be processed is intelligently segmented to obtain multiple video sub-files.

[0051] The third preset duration can be the maximum duration of the pre-defined video sub-files after segmentation, such as 10 minutes or 15 minutes.

[0052] Feasibly, considering that conference video files are generally large in size, directly extracting audio or performing speech recognition on the complete conference video file can easily lead to memory overflow or excessive timeouts. Therefore, the conference content to be processed can be intelligently segmented into multiple video sub-files, with the video duration of each sub-file not exceeding the aforementioned third preset duration, and the video durations of each sub-file can be unequal.

[0053] Feasibly, during the recording of the conference video file, if the recording duration reaches a duration node corresponding to the third preset duration (such as the recording duration being one time or two times the third preset duration), the current timestamp or the current recording duration can be automatically marked as a segmentation point so that the complete conference video file can be segmented based on the marked segmentation point in the subsequent process, thereby improving the segmentation efficiency.

[0054] Alternatively, during the recording of the conference video file, in response to the marking operation triggered by the terminal account, the timestamp of the marking operation or the recording duration at the time of the marking operation can be marked as a segmentation point. This allows the complete conference video file to be segmented based on the marked segmentation point in subsequent processing, which can meet other needs of the terminal account for video segments.

[0055] Feasibly, during the meeting recording process, the video files can be automatically saved and recorded locally on the terminal when the recording time reaches the time node corresponding to the third preset time, directly obtaining the aforementioned multiple video sub-files.

[0056] In step S2014, each video sub-file in the multiple video sub-files is transcoded one by one to obtain the corresponding audio sub-file, and the corresponding audio sub-file is uploaded to the object storage space one by one.

[0057] The transcoding process mentioned above refers to extracting the audio track from the video sub-file to obtain the corresponding audio sub-file.

[0058] It is feasible to store the unstructured data of video sub-files and audio sub-files as objects. The objects contain the file data itself, metadata, and globally unique identifiers, so that the data can be stored and retrieved in the object storage space.

[0059] Feasibly, the segmented video sub-files and transcoded audio sub-files can be resumed from the point of interruption (a technique that allows uploading or downloading files from the point of interruption after a file transfer is interrupted, avoiding repeated transfers, improving transfer efficiency and saving bandwidth) to object storage space, which can be a distributed cluster of storage nodes.

[0060] Alternatively, abnormal files can be cleaned up from the multiple audio sub-files after transcoding, such as removing audio sub-files containing only ambient noise or duplicate audio sub-files, thereby reducing the processing load of the speech recognition process and improving processing efficiency.

[0061] In step S2015, the audio sub-files corresponding to each video sub-file are obtained from the object storage space.

[0062] In step S2016, the audio subfiles corresponding to each video subfile are merged into a merged audio file.

[0063] Alternatively, steps S2014-S2016 can be implemented as follows: First, upload each video sub-file from the multiple video sub-files to the object storage space one by one to release the storage space of the local terminal. Then, the server retrieves each video sub-file from the object storage space, performs transcoding processing on each one to obtain the audio sub-file corresponding to each video sub-file, merges the audio sub-files corresponding to each video sub-file into a merged audio file, and uploads the audio sub-files corresponding to each video sub-file and the final merged audio file to the object storage space.

[0064] In step S2017, based on the domain hot word list, speech recognition processing is performed on the merged audio file to obtain the second text information.

[0065] In step S2018, the second text information is cleaned to obtain the target text information.

[0066] Steps S2017-S2018 can be referred to the aforementioned embodiments, and will not be repeated here.

[0067] In one exemplary embodiment, such as Figure 3 As shown, based on the metadata of the meeting video file, such as the total video duration, format, and file size, the duration is determined. If the total video duration is not higher than the third preset duration (e.g., 30 minutes), the meeting video file can be directly transcoded into MP3 (Moving Picture Experts Group Audio Layer III) format meeting audio file. If the total video duration is higher than the third preset duration (e.g., 30 minutes), the segmentation process begins. Optionally, the meeting video file can be divided into segments of 30 minutes each, resulting in multiple video sub-files. Each video sub-file is then transcoded and uploaded to object storage space. The server can retrieve and merge the audio sub-files from the object storage space to obtain a merged audio file. Finally, the complete audio file is input into the speech recognition (ASR) model for speech recognition processing. In addition, a domain hot word list can be used to bias the speech recognition, which can improve the recognition rate of specific domain hot words and obtain the target text information mentioned above.

[0068] In the above embodiments, the meeting content to be processed for the type of meeting video file will be segmented into video segments. This can avoid the memory overflow or timeout problems that are easily caused by transcoding a complete video file at once, and can also reduce the performance requirements of video processing. By using object storage, distributed storage and retrieval of video sub-files and audio sub-files can be realized, which can improve the efficiency of file transcoding, transmission and storage.

[0069] In step S202, an evaluation model and multiple meeting minutes generation models are determined based on the content volume information.

[0070] In this embodiment of the disclosure, multiple meeting minutes generation models are independent of each other, that is, multiple meeting minutes generation models execute meeting minutes generation processing in parallel. The generation time is the word time of the meeting minutes generation model with the slowest generation processing, rather than the sum of the time of multiple meeting minutes generation models. The processing time perceived by the user is close to that of the single-model scheme.

[0071] In this embodiment, the multiple meeting minutes generation models are different, specifically in terms of different model structures, temperature parameters, random seeds, or prompt word templates. Each meeting minutes generation model can be a deep learning model used to generate meeting minutes type text, obtained by training a Large Language Model (LLM) on a first training dataset. The first training dataset may include text information from historical meetings, context information, and corresponding structured meeting minutes information.

[0072] In this embodiment of the disclosure, the evaluation model can be a deep learning model obtained by training a large language model with a second training dataset for multi-dimensional evaluation of text quality. The second training dataset can include multiple meeting minutes. The evaluation model acts as a judge, and its evaluation dimensions can include, but are not limited to, text completeness, generation accuracy (especially checking whether the model is hallucinating based on the comparison results with the meeting text obtained after speech recognition), structural standardization of the meeting minutes, and the rationality of the tasks that can be parsed.

[0073] In this embodiment of the disclosure, unlike the fixed configuration of the number and combination of models, the number or combination of models can be dynamically adjusted according to the content volume information corresponding to the meeting content to be processed, so as to realize the on-demand allocation of processing resources and avoid excessive waste of processing resources.

[0074] In a specific embodiment, step S202 can be implemented as follows: In step S2021, when the content volume information indicates that the meeting duration is between the first preset duration and the second preset duration, or when the number of words in the target text information is between the first preset number of words and the second preset number of words, an evaluation model and multiple meeting minutes generation models are determined, and the number of multiple meeting minutes generation models is two; the first preset duration is less than the second preset duration; and the first preset number of words is less than the second preset number of words.

[0075] In step S2022, if the content volume information indicates that the meeting duration is not less than the second preset duration, or if the number of words in the target text information is not less than the second preset number of words, an evaluation model and multiple meeting minutes generation models are determined, and the number of multiple meeting minutes generation models is at least three.

[0076] For example, the first preset duration can be 30 minutes, and the second preset duration can be 90 minutes; the first preset word count is 5,000 words, and the second preset word count is 20,000 words.

[0077] In the above embodiments, by dynamically adjusting the number or combination of models according to the content volume information corresponding to the meeting content to be processed, processing resources can be allocated on demand, avoiding excessive waste of processing resources.

[0078] In one specific embodiment, the above method may further include: In step S301, if the content volume information indicates that the meeting duration is less than the first preset duration or the number of words in the target text information is less than the first preset number of words, a single meeting minutes generation model is determined.

[0079] Among them, a single meeting minutes generation model can be any one of the above-mentioned multiple meeting minutes generation models.

[0080] In step S302, the target text information and context information are input into a single meeting minutes generation model to perform meeting minutes generation processing, thereby obtaining the target meeting minutes information corresponding to the meeting content to be processed.

[0081] In the above embodiments, when the content of the meeting to be processed is relatively small, a single-model approach can be used to generate meeting minutes, which can reduce the overhead of processing resources and shorten the overall processing time.

[0082] In step S203, the target text information and context information are input into each of the multiple meeting minutes generation models to perform meeting minutes generation processing, and multiple first candidate meeting minutes information are obtained.

[0083] In this embodiment of the disclosure, target text information and context information are input into each meeting minutes generation model, thereby enabling more accurate meeting minutes generation processing based on multi-source information that integrates target text information and context information. This improves the accuracy of each first candidate meeting minutes information, enhances the relevance of each first candidate meeting minutes information to the meeting business, and facilitates the subsequent accurate identification of the responsible parties for pending tasks.

[0084] In one specific embodiment, step S203 may include: In step S2031, the target text information and context information are fused to obtain target fused information.

[0085] In step S2032, the target meeting type of the meeting content to be processed is determined based on the target fusion information.

[0086] For example, meeting types may include, but are not limited to, interview meetings, performance review meetings, product seminars, business progress meetings, internal personnel meetings, and external liaison meetings.

[0087] In step S2033, the minutes template information corresponding to the meeting type is determined.

[0088] It is feasible to pre-configure corresponding minutes template information for different meeting types. The minutes template information can indicate the format, font, text structure, word count, language, text style, or copy recipients, copy process, and archiving process of the meeting minutes information for the corresponding meeting type.

[0089] In an optional embodiment, step S2032 may include: In step S20321, the meeting title information of the meeting content to be processed is determined from the target fusion information.

[0090] In step S20322, the meeting title information and at least one keyword indicating the meeting type are matched to obtain the matching result.

[0091] Feasibly, rule-based matching can be performed on the meeting title information and at least one preset keyword indicating the meeting type to attempt to find target keywords that are the same as or similar to the meeting title information among the at least one keyword. That is, the matching result can indicate whether there are target keywords that are the same as or similar to the meeting title information among the at least one keyword, wherein the overlap between the characters of the target keyword and the characters of the meeting title information meets a preset threshold condition.

[0092] In step S20323, if the matching result indicates that at least one keyword matches the meeting title information, the target meeting type of the meeting content to be processed is determined based on the target keyword.

[0093] In step S20324, if the matching result indicates that no keyword in at least one keyword matches the meeting title information, the target fusion information is input into a preset large model for semantic recognition processing to determine the target meeting type of the meeting content to be processed.

[0094] When the type of meeting content to be processed cannot be determined through simple and efficient rule matching, the target meeting type can be determined by using a pre-set large model for semantic recognition (such as an LLM model).

[0095] Alternatively, the extracted meeting title information and target text information can be input into a preset large model for semantic recognition processing to obtain the aforementioned target meeting type.

[0096] In the above embodiments, the target meeting type is first determined directly based on the meeting title information using a simple and efficient rule matching method. If it cannot be determined, a more accurate and comprehensive semantic recognition is performed using a preset large model to determine the target meeting type. While ensuring that the target meeting type is determined, the number of calls to the preset large model can be reduced, the overhead of processing resources can be reduced, and the processing efficiency can be improved.

[0097] In step S2034, the target fusion information and the minutes template information are input into each meeting minutes generation model to perform meeting minutes generation processing, and multiple first candidate meeting minutes information are obtained.

[0098] It is feasible to input the target fusion information and the minutes template information into each meeting minutes generation model, so that each meeting minutes generation model can perform semantic recognition and reconstruction of the template fusion information into the corresponding first candidate meeting minutes information according to the requirements of the minutes template information.

[0099] In the above embodiments, the meeting minutes template information corresponding to the meeting type to which the meeting content to be processed belongs can assist in the generation and processing of meeting minutes, and improve the standardization of the generated results and their adaptability to the meeting type.

[0100] In step S204, the first candidate meeting minutes information from multiple first candidate meeting minutes information is input into the evaluation model, and multi-dimensional evaluation processing is performed to obtain the evaluation results corresponding to each first candidate meeting minutes information.

[0101] In this embodiment of the disclosure, the evaluation dimensions may include, but are not limited to, text integrity, generation accuracy (especially checking whether the model is experiencing illusions based on the comparison results with the meeting text obtained after speech recognition), the structural regularity of the meeting minutes, and the rationality of the to-do items that can be parsed. That is, the evaluation results corresponding to each first candidate meeting minutes information can indicate the performance of the corresponding first candidate meeting minutes information in the above-mentioned dimensions.

[0102] In step S205, based on the information of each first candidate meeting minutes and the evaluation results corresponding to each first candidate meeting minutes, the target meeting minutes information corresponding to the meeting content to be processed is determined.

[0103] In this embodiment, based on the evaluation results corresponding to each first candidate meeting minutes, one first candidate meeting minutes can be selected as the target meeting minutes, or the target meeting minutes can be obtained by merging the first candidate meeting minutes. The text quality of the target meeting minutes determined through multi-model generation and multi-dimensional evaluation is improved, and model illusion can be effectively avoided.

[0104] In one specific embodiment, step S205 above may include: In step S2051, if the evaluation results corresponding to each first candidate meeting minutes indicate that there is a target first candidate meeting minutes among the multiple first candidate meeting minutes that the evaluation results corresponding to the target first candidate meeting minutes meet the preset single selection condition, the target meeting minutes are determined based on the target first candidate meeting minutes.

[0105] Among them, the aforementioned preset single selection condition can indicate that the evaluation results in each dimension are superior to other first candidate meeting minutes information.

[0106] In step S2052, if the evaluation results corresponding to each first candidate meeting minutes indicate that none of the first candidate meeting minutes satisfy the preset single selection condition, the first candidate meeting minutes are merged based on the evaluation results corresponding to each first candidate meeting minutes to obtain the target meeting minutes.

[0107] It is feasible to select the top-ranking first-candidate meeting minutes from the various first-candidate meeting minutes information and use them as a benchmark. If, in one evaluation dimension, there are other first-candidate meeting minutes information that are relatively better than the top-ranking first-candidate meeting minutes information in the overall evaluation result, then the top-ranking first-candidate meeting minutes information in the overall evaluation result can be fine-tuned in the above evaluation dimension by using the other first-candidate meeting minutes information, so as to obtain the above-mentioned target meeting minutes information.

[0108] It is feasible to select the first candidate meeting minutes information with the highest textual accuracy from among the various first candidate meeting minutes information and use it as a benchmark.

[0109] In the above embodiments, based on the evaluation results corresponding to each first candidate meeting minutes, one first candidate meeting minutes can be selected as the target meeting minutes, or the first candidate meeting minutes can be merged to obtain the target meeting minutes. The text quality of the target meeting minutes is improved, and model illusion can be effectively avoided.

[0110] In a specific embodiment, after obtaining the target text information corresponding to the meeting content to be processed and the context information corresponding to the meeting content to be processed, the method further includes: In step S401, the target text information and context information are input into the target meeting minutes generation model multiple times to perform meeting minutes generation processing, resulting in multiple second candidate meeting minutes information; the target meeting minutes generation model uses a different random seed each time it performs meeting minutes generation processing.

[0111] The random seed is used to initialize the random number generator, thereby ensuring the diversity and discriminativeness of the multiple second candidate meeting minutes information obtained when only a single target meeting minutes generation model is used.

[0112] In step S402, the target meeting minutes information corresponding to the meeting content to be processed is determined based on multiple second candidate meeting minutes information.

[0113] Step S402 can refer to the foregoing embodiments, using an evaluation model to determine the target meeting minutes information from multiple second candidate meeting minutes information, which will not be repeated here.

[0114] In the above embodiments, by using a single model to generate multiple rounds under different random seeds, it can be applied to terminals or servers with relatively insufficient resources, and can also reduce the cost of model training and application.

[0115] In one specific embodiment, after step S205, the above method may further include: In step S2061, based on the structured target meeting minutes information, at least one to-do item and the item information and traceability information corresponding to each to-do item in the at least one to-do item are determined.

[0116] Feasibly, the target meeting minutes information can be structured data, such as using JSON Schema format. Structured target meeting minutes information can consist of a summary, key points of the meeting, detailed records of the topics, decisions, to-do items, risk considerations, etc.

[0117] Feasible, such as Figure 4As shown, intelligent parsing of the target meeting minutes information can be performed on the pending tasks (e.g., using a large language model to perform intelligent parsing and extraction), and the task information and traceability information of each pending task can be extracted. The task information may include, but is not limited to, the content of the corresponding pending task, the relevant responsible party (supporter or executor), and the deadline. The traceability information may include, but is not limited to, the location information of the task information in the target meeting minutes information, the audio timestamp in the pending meeting content, etc. The location information can be the relative character offset of the character recording the task information in the target meeting minutes information. The position of the first character in the target meeting minutes information (sourceTextStartIndex) or the position of the last character in the target meeting minutes information (sourceTextEndIndex) can be used as the relative character offset to achieve accurate indexing of the characters recording the task information.

[0118] In step S2062, based on the item information and traceability information corresponding to each pending item, item record information corresponding to each pending item is generated and stored in the database.

[0119] Feasible, such as Figure 4 As shown, the task record information corresponding to each task can include not only the task information and traceability information, but also the execution status information, which is initialized to pending. It can also include the corresponding meeting identifier (meetingId) as an index field for the task record information. The notified or remindNotified fields can also be set to indicate whether to trigger automatic reminders or timed reminders.

[0120] In step S2063, in response to a notification instruction for a target to-do item, the item information corresponding to the target to-do item is sent to the target object. The target to-do item is any one of at least one to-do item.

[0121] Feasible, such as Figure 4 As shown, notification instructions can be triggered manually by the meeting creator or administrator, or automatically by the system. The target object refers to the relevant responsible party indicated by the target matter information, which may include supporting parties, implementing parties, or parties with informed consent.

[0122] In a manually triggered scenario, in response to the meeting creator's notification, an immediate notification message can be generated based on the task information corresponding to the target to-do item. This message can include the meeting title, the to-do item's content, and a link to the target meeting minutes, which can be used to view the meeting minutes online. After sending the immediate notification message to the target object, the `notified` field can be automatically updated to `true` to indicate that an automatic reminder for the target to-do item has been triggered, avoiding duplicate notification messages.

[0123] In an automatically triggered scenario, the system scans all pending tasks via a scheduled task. If it finds a target pending task whose deadline is a preset time (e.g., 6 hours) away from the current time, and the `remindNotified` field in the task record information is not true (meaning no scheduled reminder message has been sent yet), then a scheduled reminder message can be sent to the target object. The scheduled reminder message can also include the meeting title, the content of the target pending task, and a link to the minutes, etc. If `remindNotified` is true, it indicates that a scheduled reminder message has already been sent, and the automatic reminder for the target pending task can be skipped to avoid repeated interruptions to the associated target object.

[0124] It is feasible to make the paths for manually triggered reminders and automatically triggered reminders independent of each other, and the target object only needs to receive messages a maximum of two times.

[0125] In step S2064, the execution status information corresponding to the target pending item is determined.

[0126] Feasibly, the system will detect or receive feedback from the target object regarding the target to-do item or the aforementioned notification message or timed reminder message, and determine the execution status information corresponding to the target to-do item based on the feedback information. The execution status information can be pending, in progress, completed, or canceled.

[0127] In step S2065, the task record information corresponding to the target task stored in the database is updated based on the execution status information corresponding to the target task.

[0128] In the above embodiments, at least one to-do item is extracted based on the target meeting minutes information, and the to-do item is persistently stored. It also realizes the precise association between the to-do item and the target meeting minutes information, the automatic allocation of the to-do item, multi-path notification, and the status tracking of the to-do item. This is a full-link management of the business processing process after the meeting, realizes the linkage capability of the enterprise collaboration system, and can effectively improve the efficiency of business promotion.

[0129] In a specific embodiment, after obtaining the target text information corresponding to the meeting content to be processed and the context information corresponding to the meeting content to be processed, the method further includes: In step S501, the target text information is divided into multiple fragment text information.

[0130] It is feasible to segment the target text information according to time units or according to semantic structure (such as multiple conference topics) to obtain multiple fragments of text information.

[0131] In step S502, the text information of each segment in the multiple segment text information is fused with the context information to obtain multiple segment fused information.

[0132] In step S503, the fusion information of each segment in the multiple segment fusion information is input into the target meeting minutes generation model to perform meeting minutes generation processing, and the third candidate meeting minutes information corresponding to each segment fusion information is obtained.

[0133] The fusion processing and meeting minutes generation processing in steps S502-S503 can be referred to the aforementioned embodiments, and will not be repeated here.

[0134] In step S504, the third candidate meeting minutes information corresponding to each segment fusion information is merged to obtain the target meeting minutes information corresponding to the meeting content to be processed.

[0135] In the above embodiments, by segmenting the target text information into multiple fragments and combining them with a single-model architecture, the processing efficiency can be improved relatively, and it is also suitable for situations where there are multiple meeting topics.

[0136] In one embodiment of this disclosure, a system architecture for implementing the above-described method for processing meeting content is provided, such as... Figure 5 As shown, it includes: 1. The client layer includes three forms: recording client (which can be used to record meeting videos), web management client (which can be used to view meeting minutes and manage to-do items), and web browser plugin (used to enable artificial intelligence interaction, such as intelligent agents); 2. Access layer, used for unified API Gateway (application interface gateway), responsible for authentication, traffic limiting, and request routing; 3. Business Service Layer, including the following microservices: Meeting Management Service (for meeting creation / query / status tracking), Minutes Generation Service (for scheduling the AI ​​engine to generate meeting minutes), To-Do Management Service (for storing / updating / querying to-do items), Document Collaboration Service (for uploading meeting minutes to the enterprise document system and for permission management), Calendar Integration Service (for obtaining contextual information such as participants / time / topic), and Notification Push Service (for sending notifications and due reminders through the enterprise instant messaging system). 4. The AI ​​engine layer consists of four interconnected modules: a. Meeting type classifier: The first level is a rule engine, which is used to quickly match meeting titles with keywords. If no match is found, the second level large language model is called to perform semantic classification to obtain the meeting type and determine the corresponding template. b. Intelligent router, used to determine the number and composition of model instances based on meeting duration, text word count and other volume information (the division in the figure is only an example). c. A multi-model parallel generation module, used to generate multiple candidate minutes independently and in parallel from multiple LLM model instances; d. Evaluation module, used to evaluate multiple candidate minutes in multiple dimensions such as completeness, accuracy, structural standardization, and reasonableness of tasks through an evaluation model, and select the best or merge them to output the final structured (such as JSON format) target meeting minutes.

[0137] 5. Infrastructure layer, including MySQL (persistent storage for structured data), Redis (high-frequency data caching), BS3 object storage (suitable for storing audio and video files), and ASR engine (speech-to-text).

[0138] In one embodiment of this disclosure, a processing flow for meeting content in a specific application scenario is provided, such as... Figure 6 As shown, after the user clicks "End Recording" on the application interface of the terminal, the background automatically completes the following processing steps without user intervention: Step 1: If the meeting content to be processed is a video file and the meeting duration is longer than the third preset duration, process it in segments. If the meeting content to be processed is an audio file, proceed directly to the next step. Step 2: Based on the pre-configured domain hot word list, perform automatic speech recognition processing on the meeting audio file corresponding to the meeting content to be processed to obtain the first text information. Then, perform text cleaning processing on the first text information to obtain the target text information. Step 3: Multi-source information fusion, which integrates information from four sources as subsequent input: (1) target text information; (2) attendees, time, topic, etc. in calendar information; (3) related document content; (4) attendee information such as name, department, job title, etc. Step 4: Meeting type classification. The first level is the rule engine, which is used to quickly match meeting titles and keywords. If no match is found, the second level of the large language model is called to perform semantic classification to obtain the meeting type and determine the corresponding minutes template. Step 5: Intelligent routing determines the number and composition of model instances based on meeting duration, text volume, and other volume information (the division in the figure is only an example). The number of models is allocated according to meeting duration and text volume. Step 6: Parallel generation of multiple models. Multiple LLM instances receive the same input but use different temperature parameters or cue word variations to generate multiple candidate meeting minutes in parallel. The total time is equal to the time of the slowest single model. Step 7: Model evaluation. The evaluation model is used to evaluate multiple candidate minutes from multiple dimensions, including completeness, accuracy, structural standardization, and reasonableness of tasks. The best or combined minutes are selected to produce the final structured (e.g., JSON format) meeting minutes. Step 8: Structured output and message delivery. Perform three delivery actions: a. Convert to Markdown format and upload to the enterprise document collaboration service and set permissions; b. Extract to-do items and set reminders; c. Notify all target parties through the enterprise instant messaging application.

[0139] As can be seen from the technical solutions provided by the embodiments of this disclosure above, in a method for processing meeting content provided by this disclosure, the content volume information, target text information, and context information corresponding to the meeting content to be processed are first obtained, wherein the context information indicates the meeting background of the meeting content to be processed; this disclosure dynamically configures an evaluation model and multiple meeting minutes generation models based on the content volume information, and the multiple meeting minutes generation models are independent and different from each other, which can avoid excessive waste of resources; this disclosure obtains multiple first candidate meeting minutes information by inputting the target text information and context information into each of the multiple meeting minutes generation models and performing meeting minutes generation processing. Parallel generation of the generative model ensures the integrity and stability of the generated meeting minutes, preventing the omission of key information due to the large amount of content in a single model, thus guaranteeing the overall generation quality. This disclosure inputs the first candidate meeting minutes information from multiple first candidate meeting minutes information into the evaluation model and performs multi-dimensional evaluation processing to obtain the evaluation results corresponding to the first candidate meeting minutes information. Then, based on the first candidate meeting minutes information and the evaluation results corresponding to each first candidate meeting minutes information, selection or merging can be performed to obtain higher quality target meeting minutes information, thereby improving the quality of meeting minutes generation and effectively avoiding the model illusion phenomenon in the content production process.

[0140] Figure 7 This is a block diagram illustrating a meeting content processing apparatus according to an exemplary embodiment. (Refer to...) Figure 7 The device 700 includes: The acquisition module 710 is configured to acquire content volume information, target text information, and context information corresponding to the meeting content to be processed; the context information indicates the meeting background of the meeting content to be processed. The model determination module 720 is configured to determine an evaluation model and multiple meeting minutes generation models based on the content volume information; the multiple meeting minutes generation models are independent of each other and are not the same. The meeting minutes information generation module 730 is configured to input the target text information and the context information into each of the plurality of meeting minutes generation models, perform meeting minutes generation processing, and obtain a plurality of first candidate meeting minutes information; The evaluation module 740 is configured to input each first candidate meeting minutes from the plurality of first candidate meeting minutes into the evaluation model, perform multi-dimensional evaluation processing, and obtain the evaluation result corresponding to each first candidate meeting minutes. The target meeting minutes information determination module 750 is configured to determine the target meeting minutes information corresponding to the meeting content to be processed based on the evaluation results corresponding to each first candidate meeting minutes information and each first candidate meeting minutes information.

[0141] Optionally, the meeting minutes information generation module 730 includes: The information fusion unit is configured to perform a fusion process on the target text information and the context information to obtain target fused information; The meeting type determination unit is configured to determine the target meeting type of the meeting content to be processed based on the target fusion information; The minutes template determination unit is configured to determine the minutes template information corresponding to the meeting type. The first meeting minutes generation unit is configured to input the target fusion information and the minutes template information into each meeting minutes generation model, perform meeting minutes generation processing, and obtain the multiple first candidate meeting minutes information.

[0142] Optionally, the meeting type determination unit includes: The meeting title determination subunit is configured to determine the meeting title information of the meeting content to be processed from the target fusion information; The matching subunit is configured to perform matching processing on the meeting title information and at least one keyword indicating the meeting type, and obtain the matching result; The first determining subunit is configured to determine the target meeting type of the meeting content to be processed based on the target keyword when the matching result indicates that there is a target keyword that matches the meeting title information among the at least one keyword; The second determining subunit is configured to, when the matching result indicates that no keyword among the at least one keyword matches the meeting title information, input the target fusion information into a preset large model, perform semantic recognition processing, and determine the target meeting type of the meeting content to be processed.

[0143] Optionally, the model determination module 720 includes: The first model configuration unit is configured to determine a single meeting minutes generation model when the content volume information indicates that the meeting duration is less than a first preset duration or the number of words in the target text information is less than a first preset number of words. The second meeting minutes generation unit is configured to input the target text information and the context information into a single meeting minutes generation model, perform meeting minutes generation processing, and obtain the target meeting minutes information corresponding to the meeting content to be processed.

[0144] Optionally, the model determination module 720 includes: The second model configuration unit is configured to determine the evaluation model and the plurality of meeting minutes generation models when the content volume information indicates that the meeting duration is between a first preset duration and a second preset duration, or when the number of words in the target text information is between a first preset number of words and a second preset number of words; the number of the plurality of meeting minutes generation models is two; the first preset duration is less than the second preset duration; the first preset number of words is less than the second preset number of words; The third model configuration unit is configured to determine the evaluation model and the plurality of meeting minutes generation models when the content volume information indicates that the meeting duration is not less than the second preset duration, or the number of words of the target text information is not less than the second preset number of words. The number of the plurality of meeting minutes generation models is at least three.

[0145] Optionally, the target meeting minutes information determination module 750 includes: The first determining unit is configured to determine the target meeting minutes information based on the target first candidate meeting minutes information when the evaluation result corresponding to each of the multiple first candidate meeting minutes information indicates that the evaluation result corresponding to the target first candidate meeting minutes information satisfies a preset single selection condition. The second determining unit is configured to perform a merging process on the evaluation results corresponding to each first candidate meeting minutes information to obtain the target meeting minutes information when the evaluation results corresponding to each first candidate meeting minutes information indicate that none of the evaluation results corresponding to any of the multiple first candidate meeting minutes information satisfy a preset single selection condition.

[0146] Optionally, the device 700 includes: The to-do item determination module is configured to execute, based on the structured target meeting minutes information, to determine at least one to-do item and the item information and traceability information corresponding to each to-do item in the at least one to-do item; The storage module is configured to generate and store the corresponding task record information for each task in the database based on the task information and traceability information for each task. The push module is configured to execute a notification instruction in response to a target to-do item, and send the item information corresponding to the target to-do item to the target object, wherein the target to-do item is any one of the at least one to-do items; The execution status determination module is configured to determine the execution status information corresponding to the target pending item. The update module is configured to update the item record information corresponding to the target to-do item stored in the database based on the execution status information corresponding to the target to-do item.

[0147] Optionally, the device 700 includes: The third meeting minutes generation unit is configured to, after obtaining the target text information corresponding to the meeting content to be processed and the context information corresponding to the meeting content to be processed, input the target text information and the context information multiple times into the target meeting minutes generation model to perform meeting minutes generation processing and obtain multiple second candidate meeting minutes information; the target meeting minutes generation model uses a different random seed each time it performs meeting minutes generation processing; The third determining unit is configured to determine the target meeting minutes information corresponding to the meeting content to be processed based on the plurality of second candidate meeting minutes information.

[0148] Optionally, the device 700 further includes: The text segmentation unit is configured to, after obtaining the target text information corresponding to the meeting content to be processed and the context information corresponding to the meeting content to be processed, segment the target text information into multiple fragment text information. The segment fusion unit is configured to perform fusion processing on each segment text information in the plurality of segment text information and the context information to obtain multiple segment fusion information; The fourth meeting minutes generation unit is configured to input the fusion information of each segment from the multiple fusion information into the target meeting minutes generation model, perform meeting minutes generation processing, and obtain the third candidate meeting minutes information corresponding to each fusion information; The fourth determining unit is configured to perform merging processing on the third candidate meeting minutes information corresponding to the fusion information of each segment, so as to obtain the target meeting minutes information corresponding to the meeting content to be processed.

[0149] Optionally, the acquisition module 710 includes: The first speech recognition unit is configured to perform speech recognition processing on the meeting content to be processed based on a pre-configured domain hot word list when the meeting content to be processed is a meeting audio file, so as to obtain first text information; The first text cleaning unit is configured to perform text cleaning processing on the first text information to obtain the target text information.

[0150] Optionally, the device 700 further includes: The video segmentation unit is configured to intelligently segment the meeting content to be processed into multiple video sub-files when the meeting content to be processed is a meeting video file and the meeting duration of the meeting content to be processed is longer than a third preset duration. The video transcoding unit is configured to transcode each video sub-file in the plurality of video sub-files one by one to obtain the audio sub-files corresponding to each video sub-file, and upload the audio sub-files corresponding to each video sub-file to the object storage space one by one. The audio acquisition unit is configured to retrieve the audio sub-files corresponding to each video sub-file from the object storage space; The audio merging unit is configured to merge the audio subfiles corresponding to each video subfile into a merged audio file. The second speech recognition unit is configured to perform speech recognition processing on the merged audio file based on the domain hot word list to obtain second text information; The second text cleaning unit is configured to perform text cleaning processing on the second text information to obtain the target text information.

[0151] Regarding the apparatus in the above embodiments, the specific manner in which each module performs its operation has been described in detail in the embodiments related to the method, and will not be elaborated upon here.

[0152] In an exemplary embodiment, an electronic device is also provided, including: a processor; and a memory for storing processor-executable instructions; wherein the processor is configured to execute the instructions to implement a method for processing conference content as described in the embodiments of this disclosure.

[0153] Figure 8 This is a block diagram illustrating an electronic device for implementing a method for processing conference content, according to an exemplary embodiment. The electronic device may be a terminal, and its internal structure diagram may be as follows: Figure 8As shown, the electronic device includes a processor, memory, network interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The network interface is used to communicate with external terminals via a network connection. When the computer program is executed by the processor, it implements a method for processing meeting content. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad mounted on the device's casing, or an external keyboard, touchpad, or mouse.

[0154] Figure 9 This is a block diagram of an electronic device for implementing a method for processing meeting content, according to an exemplary embodiment. The electronic device may be a server, and its internal structure diagram may be as follows: Figure 9 As shown, this electronic device includes a processor, memory, and a network interface connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The network interface is used to communicate with external terminals via a network connection. When the computer program is executed by the processor, it implements methods for processing conference content.

[0155] Those skilled in the art will understand that Figure 8 and Figure 9 The structure shown is merely a block diagram of a portion of the structure related to the present disclosure and does not constitute a limitation on the electronic device to which the present disclosure is applied. A specific electronic device may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0156] In an exemplary embodiment, a computer-readable storage medium including instructions is also provided, which, when executed by a processor of an electronic device, enable the electronic device to perform the meeting content processing method of the embodiments of this disclosure.

[0157] In an exemplary embodiment, a computer program product is also provided, including computer instructions that, when executed by a processor, implement the method for processing meeting content in the embodiments of this disclosure.

[0158] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. This computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and RAMbus dynamic RAM (RDRAM), etc.

[0159] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this disclosure are indicated by the following claims.

[0160] It should be understood that this disclosure is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this disclosure is limited only by the appended claims.

Claims

1. A method for processing meeting content, characterized in that, The method includes: Obtain the content volume information, target text information, and context information corresponding to the meeting content to be processed; the context information indicates the meeting background of the meeting content to be processed. Based on the content volume information, an evaluation model and multiple meeting minutes generation models are determined; the multiple meeting minutes generation models are independent of each other and are not the same. The target text information and the context information are input into each of the multiple meeting minutes generation models to perform meeting minutes generation processing, and multiple first candidate meeting minutes information are obtained. Each first candidate meeting minutes is input into the evaluation model and subjected to multi-dimensional evaluation processing to obtain the evaluation result corresponding to each first candidate meeting minutes. Based on the meeting minutes information of each first candidate meeting and the evaluation results corresponding to each first candidate meeting minutes information, the target meeting minutes information corresponding to the meeting content to be processed is determined.

2. The method according to claim 1, characterized in that, The process involves inputting the target text information and the context information into each of the multiple meeting minutes generation models to perform meeting minutes generation processing, thereby obtaining multiple first candidate meeting minutes information, including: The target text information and the context information are fused together to obtain target fused information; Based on the target fusion information, the target meeting type of the meeting content to be processed is determined; Determine the minutes template information corresponding to the meeting type; The target fusion information and the minutes template information are input into each meeting minutes generation model to perform meeting minutes generation processing, thereby obtaining the multiple first candidate meeting minutes information.

3. The method according to claim 2, characterized in that, The step of determining the target meeting type of the meeting content to be processed based on the target fusion information includes: The meeting title information of the meeting content to be processed is determined from the target fusion information; The meeting title information and at least one keyword indicating the meeting type are matched to obtain the matching result; If the matching result indicates that there is a target keyword among the at least one keyword that matches the meeting title information, the target meeting type of the meeting content to be processed is determined based on the target keyword. If the matching result indicates that none of the at least one keywords matches the meeting title information, the target fusion information is input into a preset large model for semantic recognition processing to determine the target meeting type of the meeting content to be processed.

4. The method according to claim 1, characterized in that, The method further includes: If the content volume information indicates that the meeting duration is less than a first preset duration or the number of words in the target text information is less than a first preset number of words, a single meeting minutes generation model is determined; The target text information and the context information are input into a single meeting minutes generation model to perform meeting minutes generation processing, thereby obtaining the target meeting minutes information corresponding to the meeting content to be processed.

5. The method according to claim 1, characterized in that, The process of determining the evaluation model and multiple meeting minutes generation models based on the content volume information includes: When the content volume information indicates that the meeting duration is between a first preset duration and a second preset duration, or when the number of words in the target text information is between a first preset number of words and a second preset number of words, the evaluation model and the multiple meeting minutes generation models are determined, wherein the number of multiple meeting minutes generation models is two; the first preset duration is less than the second preset duration; and the first preset number of words is less than the second preset number of words. If the content volume information indicates that the meeting duration is not less than the second preset duration, or if the number of words in the target text information is not less than the second preset number of words, then the evaluation model and the plurality of meeting minutes generation models are determined, and the number of the plurality of meeting minutes generation models is at least three.

6. The method according to claim 1, characterized in that, The process of determining the target meeting minutes information corresponding to the meeting content to be processed, based on the evaluation results corresponding to each first candidate meeting minutes information, includes: If the evaluation results corresponding to each first candidate meeting minutes indicate that there is a target first candidate meeting minutes among the multiple first candidate meeting minutes that the evaluation results corresponding to the target first candidate meeting minutes meet the preset single selection condition, then the target meeting minutes are determined based on the target first candidate meeting minutes. If the evaluation results corresponding to each first candidate meeting minutes indicate that none of the first candidate meeting minutes satisfy the preset single selection condition, the first candidate meeting minutes are merged based on the evaluation results corresponding to each first candidate meeting minutes to obtain the target meeting minutes.

7. The method according to claim 1, characterized in that, The method further includes: Based on the structured target meeting minutes information, at least one to-do item is identified, along with item information and traceability information corresponding to each to-do item in the at least one to-do item. Based on the item information and traceability information corresponding to each pending item, generate item record information corresponding to each pending item and store it in the database; In response to a notification instruction for a target to-do item, the item information corresponding to the target to-do item is sent to the target object, wherein the target to-do item is any one of the at least one to-do items; Determine the execution status information corresponding to the target pending item; Based on the execution status information corresponding to the target to-do item, update the item record information corresponding to the target to-do item stored in the database.

8. The method according to claim 1, characterized in that, After obtaining the target text information corresponding to the meeting content to be processed and the context information corresponding to the meeting content to be processed, the method further includes: The target text information and the context information are input multiple times into the target meeting minutes generation model to perform meeting minutes generation processing, resulting in multiple second candidate meeting minutes information; the target meeting minutes generation model uses a different random seed each time it performs meeting minutes generation processing; Based on the multiple second candidate meeting minutes, the target meeting minutes corresponding to the meeting content to be processed are determined.

9. The method according to claim 1, characterized in that, After obtaining the target text information corresponding to the meeting content to be processed and the context information corresponding to the meeting content to be processed, the method further includes: The target text information is divided into multiple fragment text information; The text information of each segment in the multiple segment text information is fused with the context information to obtain multiple segment fused information; Each segment fusion information from the multiple segment fusion information is input into the target meeting minutes generation model for meeting minutes generation processing to obtain the third candidate meeting minutes information corresponding to each segment fusion information. The third candidate meeting minutes information corresponding to the fusion information of each segment is merged to obtain the target meeting minutes information corresponding to the meeting content to be processed.

10. The method according to claim 1, characterized in that, The process of obtaining the target text information corresponding to the meeting content to be processed and the context information corresponding to the meeting content to be processed includes: When the meeting content to be processed is an audio file, the meeting content to be processed is subjected to speech recognition processing based on a pre-configured domain hot word list to obtain the first text information; The first text information is cleaned to obtain the target text information.

11. The method according to claim 10, characterized in that, The method further includes: If the meeting content to be processed is a meeting video file and the meeting duration of the meeting content to be processed is longer than a third preset duration, the meeting content to be processed is intelligently segmented to obtain multiple video sub-files; Each video sub-file in the plurality of video sub-files is transcoded one by one to obtain the corresponding audio sub-file, and the corresponding audio sub-file is uploaded to the object storage space one by one. Retrieve the audio sub-files corresponding to each video sub-file from the object storage space; Merge the audio sub-files corresponding to each video sub-file into a merged audio file; Based on the aforementioned domain hot word list, the merged audio file is subjected to speech recognition processing to obtain second text information; The second text information is cleaned to obtain the target text information.

12. A device for processing meeting content, characterized in that, The device includes: The acquisition module is configured to acquire the content volume information, target text information, and context information corresponding to the meeting content to be processed; the context information indicates the meeting background of the meeting content to be processed. The model determination module is configured to determine an evaluation model and multiple meeting minutes generation models based on the content volume information; the multiple meeting minutes generation models are independent of each other and are not the same. The meeting minutes information generation module is configured to input the target text information and the context information into each of the multiple meeting minutes generation models, perform meeting minutes generation processing, and obtain multiple first candidate meeting minutes information; The evaluation module is configured to input each first candidate meeting minutes from the plurality of first candidate meeting minutes into the evaluation model, perform multi-dimensional evaluation processing, and obtain the evaluation result corresponding to each first candidate meeting minutes. The target meeting minutes information determination module is configured to determine the target meeting minutes information corresponding to the meeting content to be processed based on the evaluation results corresponding to each first candidate meeting minutes information and the first candidate meeting minutes information.

13. An electronic device, characterized in that, include: processor; Memory used to store the processor's executable instructions; The processor is configured to execute the instructions to implement the method for processing meeting content as described in any one of claims 1 to 11.

14. A computer-readable storage medium, characterized in that, When the instructions in the computer-readable storage medium are executed by the processor of the electronic device, the electronic device is able to perform the method of processing conference content as described in any one of claims 1 to 11.

15. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instructions are executed by the processor, they implement the method for processing the meeting content according to any one of claims 1 to 11.