Meeting minutes generation method and device, storage medium and electronic device

Abstract

Embodiments of the present application disclose a conference minutes generation method and device, a storage medium and an electronic device. The method comprises: obtaining a conference record set; determining conference minutes content corresponding to target conference record data in the conference record set based on a conference minutes template, a conference regulation and the conference record set; and writing the conference minutes content into the conference minutes template to generate a conference minutes file corresponding to the conference minutes template. The embodiments of the present application can improve the efficiency of conference minutes generation.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to a method, apparatus, storage medium and electronic device for generating meeting minutes. Background Technology

[0002] With the development of communication technology, especially real-time audio and video communication, audio and video conferencing is being used by more and more people. Audio and video conferencing has become widely used in remote work, remote learning, and remote communication. During the meeting, records can be taken and organized to create meeting minutes, which facilitate meeting management tasks such as meeting review and summarization. Summary of the Invention

[0003] This application provides a method, apparatus, storage medium, and electronic device for generating meeting minutes, which can simplify the process of generating meeting minutes and improve the efficiency of meeting minutes generation. The technical solution is as follows:

[0004] In a first aspect, embodiments of this application provide a method for generating meeting minutes, the method comprising:

[0005] Get the collection of meeting minutes;

[0006] Based on the meeting minutes template, meeting rules, and the set of meeting records, determine the meeting minutes content corresponding to the target meeting record data in the set of meeting records;

[0007] The meeting minutes content is written into the meeting minutes template to generate the meeting minutes file corresponding to the meeting minutes template.

[0008] Secondly, embodiments of this application provide a meeting minutes generation apparatus, the apparatus comprising:

[0009] The record set generation module is used to obtain a set of meeting records;

[0010] The minutes content determination module is used to determine the meeting minutes content corresponding to the target meeting record data based on the meeting minutes template, meeting rules, and the meeting record set.

[0011] The minutes document generation module is used to write the meeting minutes content into the meeting minutes template to generate the meeting minutes document corresponding to the meeting minutes template.

[0012] Thirdly, embodiments of this application provide a computer storage medium storing a plurality of instructions adapted for loading by a processor and executing the above-described method steps.

[0013] Fourthly, embodiments of this application provide an electronic device that may include: a processor and a memory; wherein the memory stores a computer program adapted to be loaded by the processor and to execute the above-described method steps.

[0014] The beneficial effects of the technical solutions provided in some embodiments of this application include at least the following:

[0015] In one or more embodiments of this application, an electronic device acquires a set of meeting minutes, then determines the meeting minutes content corresponding to the target meeting minutes data based on a meeting minutes template, meeting rules, and the set of meeting minutes; and then writes the meeting minutes content into the meeting minutes template to generate a meeting minutes file corresponding to the meeting minutes template. This avoids the inefficiency of methods such as manually recording meeting minutes information. By combining meeting minutes templates and meeting rules, it can quickly assist in the automatic generation of minutes of interest, and the meeting minutes generation efficiency is higher and faster, thus improving the efficiency of meeting minutes generation. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a schematic diagram of a scenario architecture for generating meeting minutes provided in an embodiment of this application;

[0018] Figure 2 This is a flowchart illustrating a meeting minutes generation method provided in an embodiment of this application;

[0019] Figure 3 This is a flowchart illustrating another method for generating meeting minutes provided in an embodiment of this application;

[0020] Figure 4 This is a schematic diagram of a model training scenario involved in the meeting minutes generation method provided in this application embodiment;

[0021] Figure 5 This is a flowchart illustrating another method for generating meeting minutes provided in an embodiment of this application;

[0022] Figure 6 This is a schematic diagram of a meeting minutes template containing historical meeting minutes content, which is involved in the meeting minutes generation method provided in this application embodiment;

[0023] Figure 7 This is a schematic diagram of a meeting minutes generation device provided in an embodiment of this application;

[0024] Figure 8 This is a schematic diagram of the structure of a minutes content determination module provided in an embodiment of this application;

[0025] Figure 9 This is a schematic diagram of another meeting minutes generation device provided in an embodiment of this application;

[0026] Figure 10 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0027] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0028] In the description of this application, it should be understood that the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance. In the description of this application, it should be noted that, unless otherwise expressly specified and limited, "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or devices. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances. Furthermore, in the description of this application, unless otherwise stated, "multiple" means two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist; for example, A and / or B can represent: A alone, A and B simultaneously, and B alone. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship.

[0029] The present application will now be described in detail with reference to specific embodiments.

[0030] Please see Figure 1 This is a schematic diagram illustrating a meeting minutes generation system provided in an embodiment of this application. Figure 1 As shown, the meeting minutes generation system may include a client cluster and a server 100.

[0031] The client cluster may include at least one client, such as Figure 1 As shown, it specifically includes client 1 corresponding to user 1, game client 2 corresponding to user 2, ..., client n corresponding to user n, where n is an integer greater than 0.

[0032] Each client in a client cluster can be an electronic device with communication capabilities, including but not limited to: wearable devices, handheld devices, personal computers, tablets, in-vehicle devices, smartphones, computing devices, or other processing devices connected to a wireless modem. Electronic devices may have different names in different networks, such as: user equipment, access terminal, user unit, user station, mobile station, mobile station, remote station, remote terminal, mobile device, user terminal, terminal, wireless communication equipment, user agent or user device, cellular phone, cordless phone, personal digital assistant (PDA), and electronic devices in 5G networks or future evolved networks.

[0033] The server 100 can be a standalone server device, such as a rack-mount, blade, tower, or cabinet-type server device, or a workstation, mainframe, or other hardware device with strong computing power; or it can be a server cluster composed of multiple servers. The servers in the service cluster can be composed in a symmetrical manner, wherein each server is functionally and functionally equivalent in the business chain, and each server can provide services independently. The independent provision of services can be understood as not requiring the assistance of other servers.

[0034] According to some embodiments, server 100 can establish a communication connection with at least one client in the client cluster, and complete the data interaction during the meeting management process based on the communication connection;

[0035] It should be noted that the server 100 establishes a communication connection with at least one client in the client cluster via a network for interactive communication. This network can be a wireless network or a wired network. Wireless networks include, but are not limited to, cellular networks, wireless LANs, infrared networks, or Bluetooth networks. Wired networks include, but are not limited to, Ethernet, universal serial bus (USB), or controller area networks. In some embodiments, technologies and / or formats including Hyper Text Markup Language (HTML), Extensible Markup Language (XML), etc., are used to represent data exchanged over the network (such as target compressed packets). Furthermore, conventional encryption technologies such as Secure Socket Layer (SSL), Transport Layer Security (TLS), Virtual Private Network (VPN), and Internet Protocol Security (IPsec) can be used to encrypt all or some links. In other embodiments, customized and / or dedicated data communication technologies can be used to replace or supplement the aforementioned data communication technologies.

[0036] Furthermore, the meeting minutes generation system embodiments provided above and the meeting minutes generation methods described in some embodiments belong to the same concept. The execution entity corresponding to the meeting minutes generation methods involved in some embodiments of this application can be the aforementioned server 100, or it can be a client in a client cluster, specifically determined based on the actual application environment. The implementation process of the meeting minutes generation system embodiments can be detailed in the following method embodiments, and will not be repeated here.

[0037] In one embodiment, such as Figure 2 As shown, a method for generating meeting minutes is proposed. This method can be implemented using a computer program and can run on a meeting minutes generation device based on the von Neumann architecture. This computer program can be integrated into applications or run as a standalone utility application.

[0038] Specifically, the method for generating these meeting minutes includes:

[0039] Step 101: Obtain the set of meeting minutes.

[0040] The meeting can be a remote work meeting, an online / offline teaching meeting, a training meeting, etc., and meeting segments are generated during the meeting. In some implementation scenarios, participating users can typically capture video image segments of participants through devices such as cameras, and capture audio segments of participants through devices such as microphones. By recording at least one meeting segment, meeting minutes can be generated.

[0041] In some implementations, the electronic device can record each speech made by each user during the conference to generate at least one conference segment. This segment may contain one of the following data types: data manually entered by the user via a display panel, data entered via an input device, audio data input via voice, or video data captured from images. Furthermore, all conference segments can constitute the conference's multimedia conference data, which can be understood as the total recorded data of the conference.

[0042] For example, during a multimedia conference, participating users can collect conference segment data (such as voice and video) from their own devices and then send it to other participating users for playback. This allows participants to communicate in real time even in a remote setting. Conference segments can be understood as records of the multimedia data generated in the aforementioned conference scenario.

[0043] Understandably, to record and differentiate meeting segments, segment feature information can be generated during the recording process. This segment feature information distinguishes a meeting segment from others, reflecting its essential characteristics. It may include at least one or more of the following: segment timestamp, segment meeting identifier (such as the segment name), speaker identification information, and segment memory size. By associating meeting segments with their corresponding segment feature information, a meeting record data set is formed. Similarly, a meeting record set containing at least one meeting record data set can be obtained.

[0044] Optionally, the set of meeting records can be represented in the form of a meeting record list, a meeting record array, a meeting record stack, or the like.

[0045] Optionally, the meeting record set consists of multiple meeting record data. In some implementations, if the meeting segments are converted to text, the meeting record data consists of the segment text content and segment feature information corresponding to the meeting segment. In other implementations, if the meeting segments are not converted to text and are only saved in the original meeting segment format (usually audio or video data format), the meeting record data consists of the meeting segment and segment feature information. The specific method depends on the actual application.

[0046] In some implementations, during the process of generating a set of meeting minutes, meeting segments can be converted into text to obtain the text content of the converted segments. Then, the text content of the segments represents the meeting segments, and the text content of the segments is associated with the segment feature information to obtain a set of meeting minutes containing all the text content of the segments and the segment feature information corresponding to the text content of the segments. At this time, the set of meeting minutes is usually in text form. For example, the set of meeting minutes may consist of the text content of the segments (such as the speech content), the speaker's identity, and the segment timestamp.

[0047] In some implementations, during the generation of the meeting record set, the meeting segments may not be converted into text. The meeting record set can be generated directly from the data format of the recorded meeting segments (usually audio or video data format). In this case, the meeting segments in the meeting record set can usually be understood as a type of multimedia data (such as audio, video, etc.). The meeting record set contains the text content of all segments and the segment feature information corresponding to the text content of the segments.

[0048] Step 102: Based on the meeting minutes template, meeting rules, and the set of meeting records, determine the meeting minutes content corresponding to the target meeting record data in the set of meeting records.

[0049] The meeting charter is typically established before the meeting begins. In some implementations, the meeting charter includes one or more of the following: meeting theme, decision-making matters, resolution process, resolution method, basic meeting information (such as time, participants, etc.), and meeting segments. It is understood that the meeting charter and the meeting process are usually not the same. By incorporating the meeting charter into this application, the meeting segment in which meeting record data is located can be quickly determined in stages such as meeting minutes generation and meeting segment data processing, thereby improving data processing efficiency and better assisting in the generation of meeting minutes documents.

[0050] The meeting minutes template is used to generate meeting minutes. In practical applications, the generated meeting minutes content can be written into the meeting minutes template. After all the meeting minutes content has been written, the meeting minutes file can be generated.

[0051] The meeting minutes template can be viewed as a reference meeting attribute and an area where the minutes content is to be written. In some implementations, the reference meeting attributes corresponding to different meeting minutes templates can be different. Multiple meeting minutes templates can be set in advance, and before the meeting starts, a user terminal, such as the meeting host, can select the meeting minutes template corresponding to the current meeting from multiple meeting minutes templates.

[0052] The meeting minutes template can include, but is not limited to, one or more of the following meeting attributes: meeting topic, meeting time, meeting location, meeting chairperson, meeting discussion issues, meeting discussion speeches, moderator's speech, summary speech, meeting action items (which can be understood as the solutions reached), meeting resolutions, and follow-up work arrangements. The specific attributes are determined by the user before the meeting based on the actual application situation, and no specific limitations are made here.

[0053] In one feasible implementation: the multiple meeting record data contained in the meeting record set can be multimedia data, that is, when generating the meeting record set, the meeting segments do not need to be converted into meeting segment text. The meeting record data consists of meeting multimedia segments (such as meeting audio / video segments). Meeting minutes can be extracted directly based on the meeting minutes template, meeting rules and the meeting record set to generate the meeting minutes content to be written corresponding to the target meeting record data.

[0054] In one feasible implementation: the multiple meeting record data contained in the meeting record set can be text-type data, that is, when generating the meeting record set, the meeting fragments are first converted into meeting fragment text. The meeting record data can be composed of meeting fragment text and fragment feature information. Meeting minutes can be extracted directly based on the meeting minutes template, meeting charter and meeting record set to generate the meeting minutes content to be written corresponding to the target meeting record data.

[0055] The target meeting record data can be understood as the target meeting record data corresponding to the target meeting segments that the meeting minutes template focuses on. This means that during the meeting recording process, the meeting minutes template may exclude certain meeting segments from its scope; that is, the substantive meeting content of these segments is not included in the meeting minutes template. For example, the reference meeting attributes of the meeting minutes template might be meeting resolution attributes or meeting follow-up work arrangement attributes. If the meeting record data corresponding to meeting segment A is content related to the moderator's speech meeting record attribute, then this meeting record data is not included in the reference for generating the meeting minutes. Furthermore, if the meeting record data for meeting segment B is content related to meeting resolution attributes or meeting follow-up work arrangement attributes, then "the meeting record data for meeting segment B" is the target meeting record data.

[0056] In this application, the meeting charter is used to quickly determine the meeting stage to which the meeting record data belongs during the meeting record data processing process. This facilitates the determination of the meeting record attribute corresponding to the meeting stage. Specifically, considering that meeting stages are usually related to meeting record attributes (e.g., if the meeting stage is a meeting resolution stage defined in the meeting charter, then the meeting record attribute is usually the meeting resolution attribute), the meeting charter assists in quickly determining the meeting record attribute corresponding to the meeting record data during the meeting record data processing process. This avoids relying solely on methods such as semantic understanding to predict the meeting record attribute of the meeting record data, improving the accuracy of data processing. Since the information contained in the meeting charter is strongly correlated with the meeting record attributes, the meeting record data obtained from the speaker's recording can usually be determined with the assistance of the meeting charter. For example, the meeting stage in the meeting charter can be predicted based on the meeting record attribute corresponding to the meeting record data. At least one attribute mapping relationship between meeting stage types and meeting record attributes can be established in advance. After determining the current meeting stage, the meeting record attribute corresponding to the current meeting stage can be determined based on the attribute mapping relationship. Alternatively, the meeting record data can be targeted for attribute determination based solely on the key charter information corresponding to the meeting charter. etc.

[0057] Understandably, determining the meeting minutes content corresponding to the target meeting record data in the meeting record set can be understood as taking each meeting record data in the meeting record set as the processing object, and filtering the target meeting record data based on the meeting minutes template, meeting rules, and each meeting record data in the meeting record set, and extracting the meeting minutes from the filtered target meeting record data, so as to obtain the meeting minutes content corresponding to the target meeting record data.

[0058] In practice, a first minutes generation model for meeting minutes extraction can be trained based on a neural network. Based on the first minutes generation model, the model can automatically identify the target meeting record data that the meeting minutes template in the meeting record set is interested in, understand the semantics of the audio and video in the target meeting record data during the meeting, and extract the relevant meeting minutes content from the target meeting record data.

[0059] Understandably, meeting minutes are essentially a type of meeting signal wave (such as audio stream, video stream, or text stream). When electronic devices begin semantic recognition of the meeting minutes, they need to preprocess the collected data to eliminate interference from environmental noise, echoes, and other factors that could affect the quality of the signal wave. In practice, this preprocessing includes endpoint detection, noise reduction, and beamforming. After preprocessing, the meeting minutes undergo post-filtering to eliminate residual signal noise (such as audio and image noise). Then, an automatic gain control algorithm is used to adjust the signal energy of the collected meeting minutes before proceeding to the next step of semantic recognition.

[0060] Understandably, before semantic recognition processing of meetings, a large amount of meeting sample data can usually be obtained to train an initial neural network model. Electronic devices can obtain all or part of the meeting sample data from existing meeting audio and video databases, and / or obtain meeting sample data recorded by recording devices in actual language environments. Furthermore, the meeting sample data includes at least the sample meeting minutes template, the sample meeting rules, and the sample meeting record set of the sample meeting. Then, the meeting feature vectors in the meeting sample data are extracted (the feature vectors can be composed of one or more of the following: high-dimensional meeting audio feature vectors, meeting image feature vectors, meeting text feature vectors, minutes template feature vectors (mainly representing the reference record attributes of the minutes template), and meeting rules feature vectors (mainly representing the key information of the meeting rules)).

[0061] Electronic devices extract meeting feature vectors from meeting sample data (which can be referred to as meeting sample feature vectors of meeting sample data). These meeting feature vectors are key semantic attributes unique to unstructured data expressed in natural language. Taking a meeting as an example, the meeting feature vector includes semantic elements such as meeting intent, meeting data topic description, underlying meeting feature meaning, minutes template attribute meaning, and meeting agenda segments. These semantic elements, categorized by signal type, can be divided into one or more high-dimensional features, including high-dimensional meeting audio feature vectors, meeting image feature vectors, meeting text feature vectors, minutes template feature vectors (primarily representing the reference recording attributes of the minutes template), and meeting agenda feature vectors (primarily representing the key information of the meeting agenda). These meeting semantic elements are features capable of expressing the key semantics of the meeting itself and various key semantics within the meeting environment.

[0062] Optionally, taking the extraction process of conference feature vectors as an example, the conference sample data is digitized and preprocessed. The digitization process includes anti-aliasing filtering, sampling, A / D conversion, etc. The preprocessing process includes sampling quantization, pre-emphasis, windowing and framing, and endpoint detection of the conference sample data.

[0063] In this embodiment, the first minutes generation model is a neural network model. This neural network model is composed of densely interconnected simple nonlinear simulation processing elements of many nodes, mimicking a system model of biological neurons. The neural network model is formed by connecting the input of at least one node to the output of every other node, similar to the synaptic connections of real neurons. Each neuron expresses a specific output function, i.e., an activation function, and the connection between every two neurons contains a connection strength, i.e., a weighted value acting on the signal passing through that connection. In this embodiment, by inputting meeting sample data into the neural network model for training, a trained first minutes generation model can be obtained. This first minutes generation model has the ability to extract key meeting information features, summarize semantic knowledge, identify target meeting record data of interest to the meeting minutes template, extract meeting minutes content, and learn and memorize. Typically, the information or knowledge learned by the neural network model is stored in the connection matrix between each unit node.

[0064] Optionally, the first summary generation model can be implemented by fitting one or more of the following models: Convolutional Neural Network (CNN), Deep Neural Network (DNN), Recurrent Neural Networks (RNN), Embedding model, Gradient Boosting Decision Tree (GBDT) model, Logistic Regression (LR) model, etc.

[0065] During the training process of the first minutes generation model, when the electronic device extracts the meeting feature vectors (composed of one or more of the following: meeting audio feature vectors, meeting image feature vectors, meeting text feature vectors, minutes template feature vectors, and meeting charter feature vectors) from the meeting sample data, it annotates the meeting feature vectors. The annotation can be understood as annotating the sample key information (standard minutes content, target record identifier (indicating target meeting record data)) corresponding to the meeting sample data. The meeting feature vectors are then input into the initial neural network model for training. By training the neural network model based on the annotated meeting sample data, a trained first minutes generation model can be obtained.

[0066] It should be noted that after the first minutes generation model is trained, in the actual application stage, the electronic device can input the meeting minutes template, meeting rules and the set of meeting records into the first minutes generation model, and the first minutes generation model can output the meeting minutes content corresponding to the target meeting record data in the set of meeting records.

[0067] In some embodiments, for the first minutes generation model, it is necessary to identify the meeting record attributes of the target meeting record data during the process of generating meeting minutes content, so as to distinguish whether they are reference record attributes of interest to the meeting minutes template. Based on this, the meeting minutes content output by the first minutes generation model can also carry writing instruction information to instruct the meeting minutes content to be written to the corresponding area of ​​the meeting minutes template. The writing instruction information includes at least the meeting record attributes.

[0068] Understandably, in some implementations, electronic devices can directly extract meeting minutes content without converting meeting segment data into text, using a pre-trained first minutes generation model. This simplifies the minutes generation process, improves efficiency, and saves computational resources by incorporating the meeting transcript into the extraction process. This accelerates the model extraction process, saves extraction time, and facilitates quick differentiation of meeting record attributes to determine whether the minutes are the type of interest in the meeting minutes template. The entire minutes generation process is significantly simplified, and the accuracy of meeting minutes generation is also improved.

[0069] Step 103: Write the meeting minutes content into the meeting minutes template to generate the meeting minutes file corresponding to the meeting minutes template.

[0070] Understandably, a meeting minutes template can be viewed as consisting of reference meeting attributes and the corresponding area for the minutes content to be written. In some implementations, different meeting minutes templates may correspond to different reference meeting attributes. Multiple meeting minutes templates can be pre-set, and before the meeting begins, the user, such as the meeting host, selects the meeting minutes template corresponding to the current meeting from among the multiple templates. After determining the meeting minutes content corresponding to the target meeting record data in the set of meeting records, the electronic device can write the meeting minutes content into the meeting minutes template. Specifically, the electronic device can obtain the meeting record attributes corresponding to the meeting minutes content and then write the meeting minutes content into the area corresponding to the meeting record attributes in the meeting minutes file. In some implementations, the meeting minutes content is usually the meeting minutes content corresponding to multiple target meeting record data. In this case, the electronic device writes the meeting minutes content into the meeting minutes template one by one based on the output sequence of the meeting minutes content. After completing the writing of the meeting minutes content, the meeting minutes template can be used as the meeting minutes file.

[0071] Understandably, the meeting minutes involved in this application are the result of extracting and simplifying the content of meeting record data using methods such as natural language understanding and semantic analysis. Compared to meeting record data, meeting minutes are typically smaller in size, more concise, and semantically refined, and are not a direct record of meeting record data in plain text form.

[0072] In this embodiment, the electronic device acquires a set of meeting minutes, then, based on a meeting minutes template, meeting rules, and the set of meeting minutes, determines the meeting minutes content corresponding to the target meeting minutes data within the set of meeting minutes; and then writes the meeting minutes content into the meeting minutes template to generate a meeting minutes file corresponding to the template. This avoids the inefficiency of methods such as manually recording meeting minutes information. By combining meeting minutes templates and meeting rules, it can quickly assist in the automatic generation of minutes of interest, and the meeting minutes generation efficiency is higher and faster, thus improving the efficiency of meeting minutes generation; furthermore, it can directly extract meeting minutes content from non-text meeting minutes data, saving text conversion time.

[0073] Please see Figure 3 , Figure 3 This is a flowchart illustrating another embodiment of a meeting minutes generation method proposed in this application. Specifically:

[0074] Step 201: Obtain the set of meeting minutes.

[0075] For details, please refer to the method steps in other embodiments of this application, which will not be repeated here.

[0076] Step 202: Based on the meeting charter and each meeting record data in the meeting record set, determine the meeting record attribute corresponding to each meeting record data.

[0077] In some embodiments, meeting rules are typically set for the meeting before it begins, such as meeting rules specified by a user for the meeting before it begins.

[0078] In some embodiments of this application, by incorporating the meeting charter as a reference, the charter stage in which the meeting record data is located within the entire meeting charter can be quickly determined in stages such as the generation of meeting minutes content and the processing of meeting segment data. Since the charter stage is usually strongly correlated with the attributes of the meeting record, the attributes of the meeting record can be quickly determined based on the determined charter stage to improve data processing efficiency and better assist in the generation of meeting minutes documents.

[0079] The meeting record attributes are attribute types determined through semantic understanding of the meeting record data and can be used for recording meeting minutes. In this application, meeting record attributes correspond to the reference record attributes of the meeting minutes template. Meeting record attributes can be understood as the type of meeting information to which the meeting record data belongs. In some embodiments, the meeting record attributes can be understood as one or more attribute types determined through semantic understanding of the meeting record data, such as meeting topic, meeting time, meeting location, meeting chairperson, meeting discussion issues, meeting discussion speeches, moderator's speeches, summary speeches, meeting action items (which can be understood as the solutions reached), meeting resolutions, and follow-up work arrangements.

[0080] In one feasible implementation: meeting record attributes can be quickly determined based on the meeting charter; in specific implementation, the meeting segment in the meeting charter can be determined based on each meeting record data in the meeting record set, and then the first meeting record attribute corresponding to the meeting segment can be obtained, and the first meeting record attribute can be used as the meeting record attribute corresponding to the meeting record data.

[0081] Understandably, meeting minutes are typically set up before the meeting begins, and the entire meeting schedule is based on the meeting minutes. The meeting minutes include information such as the meeting theme, decision-making items, resolution process, resolution method, and basic meeting information (such as time and participants). Therefore, when determining the meeting record attributes corresponding to meeting record data, it is possible to determine the stage of the meeting minutes data in the meeting minutes. Since the stage of the minutes is strongly correlated with the meeting record attributes, the meeting record attributes corresponding to the stage of the minutes can be quickly obtained based on the mapping relationship between the stage of the minutes and the specific meeting record attributes, which is the first meeting record attribute.

[0082] The first meeting record attribute can be understood as the meeting record attribute determined based on the meeting charter. In some implementations, meeting record data consists of meeting record data and segment feature information such as segment timestamps, segment meeting identifiers, speaker identity information, segment meeting type information, and meeting session information. The charter session in the meeting charter can be determined at least based on the segment timestamps in the meeting record data. For example, when setting up the meeting charter, a reference time (such as the reference time for the resolution process) is set for each charter session. The matching charter session can be determined based on the segment timestamps in the meeting record data. Alternatively, the charter session can be determined at least based on the meeting session information in the meeting record data. Meeting session information is similar to charter session information, so the corresponding charter session can be determined based on the meeting session information.

[0083] In one feasible implementation: the first meeting record attribute can be quickly determined based on the meeting charter. At the same time, in order to improve the accuracy of the meeting record attribute determination, semantic recognition can be performed on the meeting record data to determine the second meeting record attribute during the process of determining the first meeting record attribute.

[0084] Optionally, key semantic features can be extracted from the meeting record data, and the second meeting record attribute corresponding to the meeting record data can be identified based on the key semantic features. The key semantic features are semantic key attributes unique to unstructured data expressed in natural language. Key semantic features include, but are not limited to, key meeting word information, meeting word frequency distribution information, grammatical level entity meeting information, semantic level meeting topic, etc.

[0085] In practical applications, a first attribute determination model based on a neural network can be trained. Typically, the model training process involves acquiring a large amount of conference sample data (as well as sample conference record data) to train the first attribute determination model. This can be done by obtaining all or part of the conference sample data from an existing conference database, and / or by acquiring conference sample data recorded in a real-world language environment using recording devices. Then, sample feature vectors are extracted from the conference sample data (feature vectors can consist of high-dimensional conference audio feature vectors, conference image feature vectors, and conference text feature vectors). When the sample feature vectors of the conference sample data are obtained, they are labeled. This labeling can be understood as labeling the standard attributes of the records corresponding to the conference sample data. The conference feature vectors are then input into the initial first attribute determination model for training. By training the first attribute determination model based on the labeled conference sample data, a well-trained first attribute determination model can be obtained.

[0086] It should be noted that after the first attribute determination model is trained, in the actual application stage, each meeting record data in the meeting record set is input into the first attribute determination model, and the first attribute determination model outputs the second meeting record attribute corresponding to each meeting record data.

[0087] Understandably, the meeting record attributes corresponding to the meeting record data can then be determined based on the first meeting record attribute and the second meeting record attribute. In specific implementation, attribute fitting can be performed based on the first meeting record attribute and the second meeting record attribute to generate the meeting record attributes corresponding to the final meeting record data; for example, the intersection of the first meeting record attribute and the second meeting record attribute can be taken to obtain the meeting record attribute; or, for another example, the union of the first meeting record attribute and the second meeting record attribute can be taken to obtain the meeting record attribute; the attribute fitting method can be determined based on the actual application environment.

[0088] In one feasible implementation, the process of determining meeting record attributes can also be entirely based on a second attribute determination model. This model can perform semantic recognition on the meeting record data based on the data input and determine attributes based on the meeting rules, finally providing a comprehensive output. In practical applications, each meeting record data in the meeting record set and the meeting rules can be input into the second attribute determination model, which outputs the meeting record attributes corresponding to each set of meeting record data.

[0089] Both the second attribute determination model and the first attribute determination model can be created based on neural networks. During training, the second attribute determination model obtains meeting sample data (including sample meeting record data and sample meeting charter), and then labels the meeting sample data. The labeling can be understood as labeling the record standard attributes corresponding to the meeting sample data. The meeting feature vector is then input into the initial second attribute determination model for training. By training the second attribute determination model based on the labeled meeting sample data, a trained second attribute determination model can be obtained.

[0090] In some implementations, the first attribute determination model can be the same model as the second attribute determination model, or they can be different models.

[0091] Step 203: Based on the meeting record attributes and meeting minutes template, determine the meeting minutes content corresponding to the target meeting record data in the meeting record set.

[0092] Understandably, once the electronic device determines the meeting record attributes corresponding to the meeting record data, it can then determine the target meeting record data from the set of meeting records based on the meeting minutes template, and thus determine the meeting minutes content corresponding to the target meeting record.

[0093] In one feasible implementation: the reference record attribute corresponding to the meeting minutes template can be determined, and then it can be determined whether the meeting record attribute of each meeting record data matches the reference record attribute in the meeting record set; for example, it can be determined whether the reference record attribute contains the meeting record attribute. If the reference record attribute contains the meeting record attribute, it is determined to match; otherwise, if the reference record attribute does not contain the meeting record attribute, it is determined not to match.

[0094] If a target meeting record data exists in the set of meeting records that matches the attribute of the reference record, then the meeting minutes content corresponding to the target meeting record data is determined.

[0095] Furthermore, non-target meeting record data in the meeting record set can be ignored, for example, non-target meeting record data can be deleted.

[0096] Understandably, electronic devices can determine the meeting minutes content corresponding to target meeting record data in the following ways:

[0097] Meeting minutes can be extracted from target meeting record data based on a pre-trained neural network-based second minutes generation model;

[0098] Input the target meeting record data into the second minutes generation model, and output the meeting minutes content corresponding to the target meeting record data.

[0099] The second summary generation model can also be implemented by fitting one or more of the following models: Convolutional Neural Network (CNN), Deep Neural Network (DNN), Recurrent Neural Networks (RNN), Embedding model, Gradient Boosting Decision Tree (GBDT) model, Logistic Regression (LR) model, etc.

[0100] The initial second minutes generation model can be trained in advance. The training can be done by acquiring a large amount of sample meeting record data to train the initial neural network model. Electronic devices can acquire all or part of the sample meeting record data from existing meeting databases, and / or acquire sample meeting record data recorded by recording devices in actual language environments. Furthermore, the sample meeting record data includes at least the meeting record data of sample meetings. Then, the meeting feature vectors in the sample meeting record data are extracted (the feature vectors can be composed of one or more of the following: high-dimensional meeting audio feature vectors, meeting image feature vectors, meeting text feature vectors, minutes template feature vectors (mainly representing the reference record attributes of the minutes template), and meeting charter feature vectors (mainly representing the key information of the meeting charter).

[0101] Electronic devices extract meeting feature vectors from sample meeting transcript data (which can be referred to as meeting sample feature vectors of meeting sample data). These meeting feature vectors are key semantic attributes unique to unstructured data expressed in natural language. Taking a meeting as an example, the meeting feature vector includes meeting intent, meeting data topic description, and underlying meeting feature meanings, among other semantic elements. These semantic elements can be categorized by signal type into one or more high-dimensional types, such as meeting audio feature vectors, meeting image feature vectors, and meeting text feature vectors. These semantic elements are features that express the key semantics of the meeting itself and the key semantics within the meeting environment.

[0102] In this embodiment, the meeting sample data is input into the neural network model for training, which yields a second minutes generation model. This second minutes generation model has the ability to extract meeting minutes information features, summarize semantic knowledge, extract meeting minutes content, and learn and memorize. Typically, the information or knowledge learned by the neural network model is stored in the connection matrix between each unit node.

[0103] In some embodiments, the model training process involved in this application can be deployed on the server or in the cloud. After the model training is completed, the corresponding model file is installed on the electronic device to improve the utilization of device resources and save the computing resources of the electronic device.

[0104] Step 204: Write the meeting minutes content into the meeting minutes template to generate the meeting minutes file corresponding to the meeting minutes template.

[0105] For details, please refer to the method steps in other embodiments of this application, which will not be repeated here.

[0106] Step 205: Collect sample data containing target meeting record data and meeting minutes content, and use the calibration service to perform minutes calibration processing on the sample data to obtain the minutes-calibrated sample calibration data.

[0107] Understandably, after the meeting minutes document or the meeting minutes content corresponding to the target meeting record data is generated, a calibration service can be invoked to perform expert calibration on the meeting minutes document or the meeting minutes content. For example, the expert or user end corresponding to the calibration service can perform error correction, polishing and other minutes calibration processing on the data in the meeting minutes document or the meeting minutes content. The meeting minutes calibration document or meeting minutes calibration content after the minutes calibration processing has been generated. At this time, sample calibration data can be generated based on the "meeting minutes calibration document or meeting minutes calibration content". In some implementations, the sample calibration data may include the meeting minutes calibration document or meeting minutes calibration content, the target meeting record data, the meeting minutes content, etc.

[0108] Optional, such as Figure 4 As shown, Figure 4 It is a schematic diagram of a scenario involving model training. Figure 4In this process, electronic devices (server, client, or other devices) can collect at least one meeting segment from the meeting process based on the meeting application service to generate a meeting record set. The meeting record set contains meeting record data corresponding to each meeting segment. Based on the meeting minutes template and meeting rules, the electronic devices execute the meeting minutes generation method involved in this application to generate meeting minutes files. The electronic devices can call the calibration service to output the meeting minutes content or meeting minutes calibration file on the display interface. The expert or user end can read the target meeting record data based on the calibration service and make corresponding corrections and modifications based on calibration experience. The electronic devices can collect data throughout the correction and modification process to generate sample calibration data, which is convenient for subsequent error correction of the relevant minutes generation model.

[0109] Optionally, the above operations can be repeated during each meeting: during each meeting, sample data containing target meeting record data and meeting minutes content is collected, and the sample data is calibrated using a calibration service to obtain a large amount of calibrated sample data; after the total amount of calibrated sample data reaches the set amount, error training is performed on relevant models such as the second and first minutes generation models based on the calibrated sample data to improve the accuracy of the models such as the second and first minutes generation models.

[0110] Step 206: Train the second summary generation model based on the sample calibration data to obtain the trained second summary generation model.

[0111] To ensure accuracy in generating meeting minutes, the second minutes generation model can be trained using sample calibration data to adjust its parameters. The expected output value (meeting minutes calibration file or meeting minutes calibration content) is determined based on the sample calibration data. Multiple target meeting record data points (i.e., the input data of the original meeting minutes file or meeting minutes content) from the sample calibration data are used as input to the second minutes generation model. This yields the actual output value for each round of the second minutes generation model. During this process, based on the expected output value, the actual output value, and a preset loss function, it is determined whether the second minutes generation model has converged in the current training phase. If converged, it indicates that the error between each actual result (actual output value) and the corresponding output result label (expected output value) has reached the expected value; for example, the output of the preset loss function is less than a preset threshold. In this case, the second minutes generation model's output on the meeting record data is relatively accurate. If it has not converged, it indicates that the error between each actual result and the corresponding evaluation result label (the labeled expected output value) has not reached the expected value. At this point, the error between the expected and actual output values ​​is backpropagated from the output layer, and the connection weights and thresholds of each layer are adjusted along the output path. After training is complete, the second summary generative model is obtained.

[0112] In one feasible implementation, the expected error can be the mean square error (MSE) between the actual output value and the expected output value. The mean square error (MSE) can be calculated using the following formula:

[0113]

[0114] Where m is the number of output nodes, p is the number of training samples, is the expected output value, and is the actual output value.

[0115] Optionally, the first minutes generation model involved in the meeting minutes generation process can also be trained based on the sample calibration data to obtain the trained first minutes generation model;

[0116] The difference between training the first minutes generation model and training the second minutes generation model lies in the different input data. The input data for the first minutes generation model consists of multiple "target meeting record data (i.e., the input data of the original meeting minutes document or meeting minutes content) + meeting rules + meeting minutes template". The rest of the model training process is similar to that of the second minutes generation model, and will not be described in detail here.

[0117] In this embodiment, the electronic device acquires a set of meeting records, then determines the meeting minutes content corresponding to the target meeting record data based on the meeting minutes template, meeting rules, and the set of meeting records. The meeting minutes content is then written into the meeting minutes template to generate the meeting minutes file corresponding to the template. This avoids the inefficiency of methods such as manually recording meeting minutes. By combining the meeting minutes template and meeting rules, the automatic generation of relevant minutes content can be quickly assisted, resulting in higher efficiency and speed. Furthermore, meeting minutes content can be directly extracted from non-text meeting record data, saving text conversion time. Collecting sample calibration data facilitates subsequent error correction of the relevant minutes generation model, improving the accuracy of meeting minutes generation. Finally, during the meeting minutes generation process, the device can fit the written minutes content and ignore irrelevant or unrelated data, improving the intelligence of meeting minutes generation.

[0118] Please see Figure 5 , Figure 5 This is a flowchart illustrating another embodiment of a meeting minutes generation method proposed in this application. Specifically:

[0119] Step 301: Obtain the meeting minutes collection;

[0120] For details, please refer to the method steps in other embodiments of this application, which will not be repeated here.

[0121] Step 302: Based on the meeting minutes template, meeting rules, and the set of meeting records, determine the target meeting record data in the set of meeting records;

[0122] According to some embodiments, the meeting record set contains multiple meeting record data. The process of determining the meeting minutes content corresponding to the target meeting record data is usually carried out one by one. That is, each one is determined based on "meeting record data + meeting minutes template + meeting charter" to determine whether it is the target meeting record data, and then the meeting minutes content corresponding to the target meeting record data is determined.

[0123] Considering that during the entire meeting, there may be pairs of meeting minutes that contain the same or partially the same semantic content, in order to improve the efficiency of meeting minutes generation, we can check whether the meeting minutes files that have already been written with meeting minutes content contain the content of the current meeting minutes data. If so, we can process the minutes content accordingly.

[0124] According to some embodiments, determining the target meeting record data from the meeting record set can be as follows:

[0125] 1. Based on the meeting charter and each meeting record data in the meeting record set, determine the meeting record attributes corresponding to each meeting record data;

[0126] 2. Determine the reference record attributes corresponding to the meeting minutes template;

[0127] 3. Determine whether the meeting record attributes of each meeting record data in the meeting record set match the reference record attributes;

[0128] 4. If the target meeting record attribute matches the reference record attribute in the set of meeting record data, then the target meeting record data can be determined.

[0129] Understandably, assuming there are n meeting record data in the meeting record set, the process usually starts from the first meeting record data in the set, determines whether the meeting record data is the target meeting record data, and then obtains the next meeting record data... until the last meeting record data.

[0130] It can be represented as:

[0131] s1. Obtain the current i-th meeting record data (i is between 1 and n), and determine the meeting record attributes corresponding to the i-th meeting record data;

[0132] s2. Determine the reference record attributes corresponding to the meeting minutes template;

[0133] s3. Determine whether the meeting record attribute of the i-th meeting record data matches the reference record attribute in the meeting record set;

[0134] s4. If the meeting record attribute of the i-th meeting record data matches the reference record attribute, then the i-th meeting record data can be determined as the target meeting record data; then, steps s303, 304, and 305 are executed; and,

[0135] If the meeting record attributes of the i-th meeting record data do not match the reference record attributes, it means that the i-th meeting record data is not the content of the meeting minutes template and can be ignored.

[0136] S5, then execute s1, s2, s3; or, then execute s1, s3;

[0137] Step 303: Obtain the content of historical meeting minutes from the meeting minutes template.

[0138] The historical meeting minutes content can be understood as follows: the meeting minutes content corresponding to each target meeting record data is added and written to the meeting minutes template one by one. Assuming that the current target meeting record data is the xth target meeting record data, the historical meeting minutes content is all or part of the meeting minutes content corresponding to the 1st to the (x-1)th target meeting record data.

[0139] Understandably, once the first target meeting record data is written into the meeting minutes template, the meeting minutes template for writing the first target meeting record data can be regarded as the meeting minutes template for writing the historical meeting minutes content. After all the target meeting record data in the meeting record set has completed writing the meeting minutes content into the meeting minutes template, the final meeting minutes file is generated.

[0140] Step 304: Perform semantic analysis on the target meeting record data based on the content of the historical meeting minutes, and determine the meeting minutes content corresponding to the target meeting record data based on the semantic analysis results.

[0141] Understandably, semantic analysis is used to analyze whether the target meeting record data exists in the content of historical meeting minutes in order to further generate the corresponding meeting minutes content.

[0142] In this application, semantic analysis processing may involve obtaining the historical meeting vector corresponding to the content of historical meeting minutes, and extracting the target meeting vector from the target meeting record data.

[0143] In some implementations, the acquisition of historical meeting vectors and target meeting vectors is based on a neural network model. The neural network model can embed the content of historical meeting minutes and the target meeting record data into a high-dimensional space, such as 100-500 dimensions in some implementations. In the new high-dimensional space, the content of historical meeting minutes and the target meeting record data are represented as meeting vectors.

[0144] Then, similarity parameters can be calculated based on the historical meeting vectors and the target meeting vectors. The similarity parameter can be similarity or similarity distance. For example, the similarity distance can be calculated using at least one of the following formulas: Euclidean distance, Manhattan distance, cosine distance, correlation coefficient distance, etc.

[0145] It is understandable that obtaining the historical meeting vector corresponding to the historical meeting minutes does not require reprocessing the historical meeting minutes. Since the historical meeting minutes are derived from the output data of the minutes generation model (such as the first minutes generation model and the second minutes generation model), the processing of the input data by the minutes generation model will generate the historical meeting vector of the previous target meeting record data. The electronic device can directly obtain the historical meeting vector from the model cache data corresponding to the minutes generation model.

[0146] Understandably, whether the semantics of the historical meeting minutes match the target meeting record data can be determined by setting a parameter range for the calculated similarity parameters. If the calculated similarity parameters fall outside the parameter range, a semantic analysis result of the first result type is obtained; if the calculated similarity parameters fall within the parameter range, a semantic analysis result of the second result type is obtained.

[0147] Further analysis is then conducted:

[0148] If the semantic analysis result is of the first result type, then the newly added content of the minutes corresponding to the target meeting record data is determined, and the newly added content of the minutes is used as the meeting minutes content to execute the step of writing the meeting minutes content into the meeting minutes template; the semantic analysis result of the first result type is that the semantics of the historical meeting minutes content does not match the target meeting record data.

[0149] This can be understood as follows: the semantic analysis result is the first result type. The target meeting record data at this point is equivalent to the historical meeting minutes content that needs to be added to the meeting minutes template. That is, the added minutes content is used as the meeting minutes content to write the meeting minutes content into the meeting minutes template; for example... Figure 6 As shown, Figure 6 This is a schematic diagram of a meeting minutes template that includes the content of historical meeting minutes. Figure 6 For example, if the historical meeting minutes record 6 meeting minutes, namely 2 minutes corresponding to the meeting time, participants, and discussion items, and 3 minutes corresponding to the action items, then it is necessary to add a new meeting minute based on the 6 meeting minutes. This means writing the "meeting minutes corresponding to the target meeting record data" in the target area outside the historical meeting minutes content area. In this case, the meeting minutes content is the newly added meeting minutes content. For example, based on the "meeting minutes corresponding to the target meeting record data", a third meeting minutes content is added in the "discussion items".

[0150] If the semantic analysis result is of the second result type, then based on the target meeting record data, the updated content of the meeting minutes is determined for the historical meeting minutes content, and the updated content of the meeting minutes is used as the meeting minutes content to execute the step of writing the meeting minutes content into the meeting minutes template; the semantic analysis result of the second result type is the semantic matching between the historical meeting minutes content and the target meeting record data.

[0151] Immediately, the semantic analysis result is of the second result type. The target meeting record data at this point is equivalent to the existence of semantically similar content in historical meeting minutes. Therefore, the electronic device, using the historical meeting minutes as a reference, updates the semantically similar content within those minutes, such as adding, deleting, or modifying the original minutes based on the updated content. Figure 6 As shown, Figure 6 This is a schematic diagram of a meeting minutes template that includes the content of historical meeting minutes. Figure 6 For example, if the historical meeting minutes record 6 meeting minutes, namely the meeting time, participants, 2 meeting minutes corresponding to the discussion items, and the meeting minutes corresponding to the action items, then it is necessary to update the content of at least one of the 6 meeting minutes. This update should be based on the "meeting minutes content related to the historical meeting minutes". For example, the first meeting minute of the "discussion items" should be updated based on the "meeting minutes content related to the historical meeting minutes".

[0152] In practice, the target historical meeting minutes that semantically match the target meeting record data can be obtained from the historical meeting minutes content, and then meeting minutes content can be generated for the historical meeting minutes content. At this time, the meeting minutes content is the difference update content.

[0153] Step 305: Write the meeting minutes content into the meeting minutes template to generate the meeting minutes file corresponding to the meeting minutes template.

[0154] For details, please refer to the method steps in other embodiments of this application, which will not be repeated here.

[0155] Step 306: In the meeting minutes file, associate the meeting minutes content with the target meeting record data and the fragment attributes of the target meeting record data.

[0156] Understandably, the generated meeting minutes content is strongly correlated with the target meeting record data (such as audio conference segments, video conference segments, and text conference segments), and the generated meeting minutes content is strongly correlated with the segment attributes of the target meeting record data. Therefore, in the meeting minutes file, the meeting minutes content can be associated with at least one of the target meeting record data and the segment attributes of the target meeting record data. This association can be achieved through hyperlinks, arrays, address mappings, or other means to link the "meeting minutes content" with "at least one of the target meeting record data" and / or "the segment attributes of the target meeting record data." In practical applications, meeting minutes establish a correspondence with meeting text records, meeting video records, and meeting audio records, facilitating quick meeting review by subsequent users based on the meeting minutes.

[0157] In one feasible implementation, the meeting minutes file can associate the meeting minutes content with the target meeting record data and the fragment attributes of the target meeting record data;

[0158] In one feasible implementation, the meeting minutes file can associate the content of the meeting minutes with the fragment attributes of the target meeting record data;

[0159] In one feasible implementation, the meeting minutes file can associate the content of the meeting minutes with the fragment attributes of the target meeting record data;

[0160] Understandably, in the process of generating meeting minutes, by associating the content of the meeting minutes with its corresponding "target meeting record data" and / or "fragment attributes," and since one or more meeting minutes with the same meeting record attribute are usually recorded in the same area of ​​the meeting minutes file, it is possible to quickly browse and locate all content with the same user focus (which can be understood as the same meeting record attribute) based on the meeting minutes file, making it convenient for subsequent users to quickly review the meeting based on the meeting minutes.

[0161] Step 306: Determine the criticality of the meeting content in the meeting record data in the meeting record set, and perform data management on the meeting record data based on the criticality of the content.

[0162] The criticality of meeting content is used to measure the importance of meeting record data to the meeting. In some embodiments, based on meeting record data with different criticalities of meeting content, data management can be performed on the associated meeting record data in the final generated meeting minutes file, such as compressing or deleting meeting record data with low criticality.

[0163] Understandably, determining the criticality of meeting content in the meeting record data set can be achieved by: extracting key semantic information from the meeting record data, calculating the semantic similarity parameter between the key semantic information and the current meeting minutes file, and using the semantic similarity parameter as the criticality of the meeting content in the meeting record data.

[0164] The key semantic information can be understood as the key semantic features in the meeting record data, such as keyword features, key phrase features, key sentence features, etc. In practical applications, meeting record data such as audio and video data types and text data types are represented by mapping to high-dimensional feature space vectors in the form of feature vectors. For example, key semantic feature vectors can be extracted based on neural network models.

[0165] In one specific implementation, the meeting minutes file is generated based on the meeting minutes content corresponding to multiple target meeting data. Therefore, the meeting minutes vectors corresponding to the meeting minutes content written in the meeting minutes file during the model processing stage (such as the processing stage of outputting meeting minutes content using the first and second minutes models) can be obtained. It can be understood that, assuming the meeting minutes file contains n meeting minutes content, the number of meeting minutes vectors obtained is n. That is to say, at least one meeting minutes vector corresponding to the model processing stage when the meeting minutes file is generated can be obtained. Then, these meeting minutes vectors are clustered to obtain cluster center vectors.

[0166] Optionally, semantic similarity parameters can be calculated based on cluster center vectors and key semantic feature vectors. In specific implementations, the semantic similarity distance between cluster center vectors and key semantic feature vectors can be calculated. The semantic similarity distance can be calculated using at least one of the following formulas: Euclidean distance formula, Manhattan distance formula, cosine distance formula, correlation coefficient distance formula, etc.

[0167] Understandably, a threshold parameter can be set for the criticality of meeting content. If the criticality of the meeting content is less than or equal to the threshold parameter, it can be considered that the current meeting record data is not the key content of this meeting, and the meeting record data can be compressed or deleted to save storage space and save bandwidth costs for user video conferencing. If the criticality of the meeting content is greater than the threshold parameter, the meeting record data will not be processed.

[0168] Understandably, step 306 can be executed before determining the meeting minutes content corresponding to the target meeting record data, that is, before judging the target meeting record data for the i-th meeting record data. If the keyness of the meeting content is less than or equal to the parameter threshold, the current meeting record data is not the key content of this meeting, and the meeting record data can be compressed or deleted. In this way, the amount of data processing can be greatly reduced when generating the meeting minutes content corresponding to the target meeting record data, and the efficiency of meeting minutes generation can be improved.

[0169] In this embodiment of the application, the electronic device obtains a set of meeting records, and then, based on the meeting minutes template, the meeting charter, and the set of meeting records, determines the meeting minutes content corresponding to the target meeting record data in the set of meeting records; then, the meeting minutes content is written into the meeting minutes template to generate a meeting minutes file corresponding to the meeting minutes template. This approach avoids the inefficiencies of manual meeting minutes recording. By combining meeting minutes templates and meeting rules, it can quickly and automatically generate relevant minutes, resulting in higher efficiency and speed. Furthermore, it can directly extract meeting minutes content from non-textual meeting records, saving text conversion time. Collecting sample calibration data facilitates subsequent error correction of the minutes generation model, improving accuracy. During minutes generation, it can fit content from different sources, ignoring irrelevant or template-unrelated data, enhancing intelligence. Instead of extracting summaries or keywords, it uses a pre-trained minutes generation model for language simplification and reorganization, resulting in a more concise generation method. Finally, it establishes a correspondence between meeting minutes and text, video, and audio records, allowing users to quickly review meetings based on the minutes.

[0170] The following are embodiments of the apparatus described in this application, which can be used to execute the embodiments of the method described in this application. For details not disclosed in the apparatus embodiments of this application, please refer to the embodiments of the method described in this application.

[0171] Please see Figure 7 This illustration shows a schematic diagram of a meeting minutes generation apparatus provided in an exemplary embodiment of this application. The meeting minutes generation apparatus can be implemented as all or part of an apparatus through software, hardware, or a combination of both. The meeting minutes generation apparatus 1 includes a record set generation module 11, a minutes content determination module 12, and a minutes file generation module 13.

[0172] Record set generation module 11 is used to obtain a set of meeting records;

[0173] The minutes content determination module 12 is used to determine the meeting minutes content corresponding to the target meeting record data in the meeting record set based on the meeting minutes template, the meeting charter, and the meeting record set.

[0174] The minutes document generation module 13 is used to write the meeting minutes content into the meeting minutes template to generate the meeting minutes document corresponding to the meeting minutes template.

[0175] Optionally, the minutes content determination module 12 is specifically used for:

[0176] Based on the meeting charter and each meeting record in the meeting record set, determine the meeting record attributes corresponding to each meeting record; based on the meeting record attributes and the meeting minutes template, determine the meeting minutes content corresponding to the target meeting record data in the meeting record set; or...

[0177] The meeting minutes template, meeting rules, and meeting record set are input into the first minutes generation model, and the meeting minutes content corresponding to the target meeting record data in the meeting record set is output.

[0178] Optionally, the minutes content determination module 12 is specifically used for:

[0179] Based on each meeting record data in the meeting record set, determine the meeting segment in the meeting charter for each meeting record data, obtain the first meeting record attribute corresponding to the meeting segment, and use the first meeting record attribute as the meeting record attribute corresponding to the meeting record data; and / or,

[0180] Each meeting record data in the meeting record set is input into the first attribute determination model, and the second meeting record attribute corresponding to each meeting record data is output; the meeting record attribute corresponding to the meeting record data is determined based on the first meeting record attribute and the second meeting record attribute.

[0181] Optionally, the minutes content determination module 12 is specifically used for:

[0182] Determine the reference record attribute corresponding to the meeting minutes template, and determine whether the meeting record attribute of each meeting record data in the meeting record set matches the reference record attribute;

[0183] If a target meeting record data exists in the set of meeting records that matches the attribute of the reference record, then the meeting minutes content corresponding to the target meeting record data is determined.

[0184] Optionally, the minutes content determination module 12 is specifically used for:

[0185] Input the target meeting record data into the second minutes generation model, and output the meeting minutes content corresponding to the target meeting record data.

[0186] Optionally, the meeting minutes generation device 1 is specifically used for:

[0187] Collect sample data containing target meeting record data and meeting minutes content, and use a calibration service to perform minutes calibration processing on the sample data to obtain the minutes-calibrated sample calibration data.

[0188] The first and / or second minutes generation models are trained based on sample calibration data to obtain the trained first and / or second minutes generation models.

[0189] Optional, such as Figure 8 As shown, the minutes content determination module 12 includes:

[0190] Content acquisition unit 121 is used to determine target meeting record data in the meeting record set and acquire historical meeting minutes content from the meeting minutes template;

[0191] The semantic analysis unit 122 is used to perform semantic analysis processing on the target meeting record data based on the content of the historical meeting minutes, and to determine the meeting minutes content corresponding to the target meeting record data based on the semantic analysis results.

[0192] Optionally, the semantic analysis unit 122 is specifically used for:

[0193] If the semantic analysis result is of the first result type, then the meeting minutes content corresponding to the target meeting record data is determined; the semantic analysis result of the first result type is that the historical meeting minutes content does not semantically match the target meeting record data.

[0194] If the semantic analysis result is of the second result type, then the meeting minutes content for the historical meeting minutes content is determined based on the target meeting record data; the semantic analysis result of the second result type is the semantic matching between the historical meeting minutes content and the target meeting record data.

[0195] Optional, such as Figure 9 As shown, the meeting minutes generation device 1 further includes:

[0196] The data association module 14 is used to associate the meeting minutes content with at least one of the target meeting record data and / or fragment attributes of the target meeting record data in the meeting minutes file.

[0197] Optionally, the meeting minutes generation device 1 is further configured to:

[0198] Determine the criticality of meeting content in the meeting record data set, and perform data management on the meeting record data based on the criticality of the content.

[0199] Optionally, the meeting minutes generation device 1 is further configured to:

[0200] Extract key semantic information from the meeting record data, calculate the semantic similarity parameter between the key semantic information and the meeting minutes file, and use the semantic similarity parameter as the keyness of the meeting content in the meeting record data.

[0201] It should be noted that the meeting minutes generation device provided in the above embodiments is only illustrated by the division of the above functional modules when executing the meeting minutes generation method. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. In addition, the meeting minutes generation device and the meeting minutes generation method embodiments provided in the above embodiments belong to the same concept, and the implementation process is detailed in the method embodiments, which will not be repeated here.

[0202] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0203] In this embodiment, the electronic device generates a set of meeting records corresponding to each meeting segment by acquiring at least one meeting segment from the meeting; then, based on the meeting minutes template, the meeting charter, and the set of meeting records, it determines the meeting minutes content corresponding to the target meeting record data in the set of meeting records; and then writes the meeting minutes content into the meeting minutes template to generate a meeting minutes file corresponding to the meeting minutes template. This system avoids the inefficiency of manual meeting minutes generation. By combining meeting minutes templates and meeting rules, it can quickly and automatically generate minutes content relevant to the template, resulting in higher efficiency and speed. Furthermore, it can directly extract meeting minutes content from non-textual meeting records, saving text conversion time. Collecting sample calibration data facilitates subsequent error correction of the minutes generation model, improving accuracy. During generation, it can fit content from different sources, ignoring irrelevant or template-unrelated data, enhancing intelligence. Instead of extracting summaries or keywords, it uses a pre-trained minutes generation model for language simplification and reorganization, resulting in a more concise generation method. Finally, it establishes a correspondence between meeting minutes and text, video, and audio records, allowing users to quickly review meetings.

[0204] This application also provides a computer storage medium that can store multiple instructions, which are adapted to be loaded and executed by a processor as described above. Figures 1-6 The meeting minutes generation method described in the illustrated embodiment can be found in the following documentation for a detailed execution process. Figures 1-6 The specific details of the illustrated embodiments will not be elaborated here.

[0205] This application also provides a computer program product storing at least one instruction, which is loaded and executed by the processor as described above. Figures 1-6 The meeting minutes generation method described in the illustrated embodiment can be found in the following documentation for a detailed execution process. Figures 1-6 The specific details of the illustrated embodiments will not be elaborated here.

[0206] Please see Figure 10 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Figure 10 As shown, the electronic device 1000 may include: at least one processor 1001, at least one network interface 1004, a user interface 1003, a memory 1005, and at least one communication bus 1002.

[0207] The communication bus 1002 is used to realize the connection and communication between these components.

[0208] The user interface 1003 may include a display screen and a camera. Optionally, the user interface 1003 may also include a standard wired interface and a wireless interface.

[0209] The network interface 1004 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface).

[0210] The processor 1001 may include one or more processing cores. The processor 1001 connects to various parts within the server 1000 using various interfaces and lines, and performs various functions and processes data by running or executing instructions, programs, code sets, or instruction sets stored in the memory 1005, and by calling data stored in the memory 1005. Optionally, the processor 1001 may be implemented using at least one hardware form of Digital Signal Processing (DSP), Field-Programmable Gate Array (FPGA), or Programmable Logic Array (PLA). The processor 1001 may integrate one or a combination of several of the following: Central Processing Unit (CPU), Graphics Processing Unit (GPU), and modem. The CPU primarily handles the operating system, user interface, and applications; the GPU is responsible for rendering and drawing the content required for display; and the modem handles wireless communication. It is understood that the modem may also not be integrated into the processor 1001 and may be implemented as a separate chip.

[0211] The memory 1005 may include random access memory (RAM) or read-only memory. Optionally, the memory 1005 may include a non-transitory computer-readable storage medium. The memory 1005 can be used to store instructions, programs, code, code sets, or instruction sets. The memory 1005 may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for at least one function (such as touch function, sound playback function, image playback function, etc.), instructions for implementing the above-described method embodiments, etc.; the data storage area may store data involved in the above-described method embodiments, etc. Optionally, the memory 1005 may also be at least one storage device located remotely from the aforementioned processor 1001. Figure 10 As shown, the memory 1005, which serves as a computer storage medium, may include an operating system, a network communication module, a user interface module, and a meeting minutes generation application.

[0212] exist Figure 10 In the illustrated electronic device 1000, the user interface 1003 is mainly used to provide an input interface for the user and to acquire user input data; while the processor 1001 can be used to call the meeting minutes generation application stored in the memory 1005 and specifically perform the following operations:

[0213] Get the collection of meeting minutes;

[0214] Based on the meeting minutes template, meeting rules, and the set of meeting records, determine the meeting minutes content corresponding to the target meeting record data in the set of meeting records;

[0215] The meeting minutes content is written into the meeting minutes template to generate the meeting minutes file corresponding to the meeting minutes template.

[0216] In one embodiment, when the processor 1001 executes the process of determining the meeting minutes content corresponding to the target meeting minutes data in the meeting minutes set based on the meeting minutes template, meeting charter, and meeting record set, the following steps are specifically performed:

[0217] Based on the meeting charter and each meeting record in the meeting record set, determine the meeting record attributes corresponding to each meeting record; based on the meeting record attributes and the meeting minutes template, determine the meeting minutes content corresponding to the target meeting record data in the meeting record set; or...

[0218] The meeting minutes template, meeting rules, and meeting record set are input into the first minutes generation model, and the meeting minutes content corresponding to the target meeting record data in the meeting record set is output.

[0219] In one embodiment, when the processor 1001 executes the step of determining the meeting record attribute corresponding to each meeting record data based on the meeting charter and each meeting record data in the meeting record set, it specifically performs the following steps:

[0220] Based on each meeting record data in the meeting record set, determine the meeting segment in the meeting charter for each meeting record data, obtain the first meeting record attribute corresponding to the meeting segment, and use the first meeting record attribute as the meeting record attribute corresponding to the meeting record data; and / or,

[0221] Each meeting record data in the meeting record set is input into the first attribute determination model, and the second meeting record attribute corresponding to each meeting record data is output; the meeting record attribute corresponding to the meeting record data is determined based on the first meeting record attribute and the second meeting record attribute.

[0222] In one embodiment, when the processor 1001 executes the step of determining the meeting minutes content corresponding to the target meeting record data in the meeting record set based on the meeting record attributes and the meeting minutes template, it specifically performs the following steps:

[0223] Determine the reference record attribute corresponding to the meeting minutes template, and determine whether the meeting record attribute of each meeting record data in the meeting record set matches the reference record attribute;

[0224] If a target meeting record data exists in the set of meeting records that matches the attribute of the reference record, then the meeting minutes content corresponding to the target meeting record data is determined.

[0225] In one embodiment, when the processor 1001 executes the meeting minutes content corresponding to the determined target meeting record data, it specifically performs the following steps:

[0226] Input the target meeting record data into the second minutes generation model, and output the meeting minutes content corresponding to the target meeting record data.

[0227] In one embodiment, when the processor 1001 executes the meeting minutes generation method, it specifically performs the following steps:

[0228] Collect sample data containing target meeting record data and meeting minutes content, and use a calibration service to perform minutes calibration processing on the sample data to obtain the minutes-calibrated sample calibration data.

[0229] The first and / or second minutes generation models are trained based on sample calibration data to obtain the trained first and / or second minutes generation models.

[0230] In one embodiment, when the processor 1001 executes the step of determining the meeting minutes content corresponding to the target meeting minutes data in the meeting minutes set, it specifically performs the following steps:

[0231] Determine the target meeting record data from the meeting record set and retrieve the historical meeting minutes content from the meeting minutes template;

[0232] Based on the content of the historical meeting minutes, semantic analysis is performed on the target meeting record data, and the meeting minutes content corresponding to the target meeting record data is determined based on the semantic analysis results.

[0233] In one embodiment, when the processor 1001 determines the meeting minutes content corresponding to the target meeting record data based on the semantic analysis result of the semantic analysis processing, it specifically performs the following steps:

[0234] If the semantic analysis result is of the first result type, then the meeting minutes content corresponding to the target meeting record data is determined; the semantic analysis result of the first result type is that the historical meeting minutes content does not semantically match the target meeting record data.

[0235] If the semantic analysis result is of the second result type, then the meeting minutes content for the historical meeting minutes content is determined based on the target meeting record data; the semantic analysis result of the second result type is the semantic matching between the historical meeting minutes content and the target meeting record data.

[0236] In one embodiment, when the processor 1001 executes the meeting minutes generation method, it specifically performs the following steps:

[0237] In the meeting minutes file, the meeting minutes content is associated with at least one of the target meeting record data and / or fragment attributes of the target meeting record data.

[0238] In one embodiment, the processor 1001 further includes the following when executing the meeting minutes generation method:

[0239] Determine the criticality of meeting content in the meeting record data set, and perform data management on the meeting record data based on the criticality of the content.

[0240] In one embodiment, when the processor 1001 executes the step of determining the criticality of meeting content in the meeting record data set, it specifically performs the following steps:

[0241] Extract key semantic information from the meeting record data, calculate the semantic similarity parameter between the key semantic information and the meeting minutes file, and use the semantic similarity parameter as the keyness of the meeting content in the meeting record data.

[0242] In this embodiment, the electronic device generates a set of meeting records corresponding to each meeting segment by acquiring at least one meeting segment from the meeting; then, based on the meeting minutes template, the meeting charter, and the set of meeting records, it determines the meeting minutes content corresponding to the target meeting record data in the set of meeting records; and then writes the meeting minutes content into the meeting minutes template to generate a meeting minutes file corresponding to the meeting minutes template. This system avoids the inefficiency of manual meeting minutes generation. By combining meeting minutes templates and meeting rules, it can quickly and automatically generate minutes content relevant to the template, resulting in higher efficiency and speed. Furthermore, it can directly extract meeting minutes content from non-textual meeting records, saving text conversion time. Collecting sample calibration data facilitates subsequent error correction of the minutes generation model, improving accuracy. During generation, it can fit content from different sources, ignoring irrelevant or template-unrelated data, enhancing intelligence. Instead of extracting summaries or keywords, it uses a pre-trained minutes generation model for language simplification and reorganization, resulting in a more concise generation method. Finally, it establishes a correspondence between meeting minutes and text, video, and audio records, allowing users to quickly review meetings.

[0243] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory, or random access memory, etc.

[0244] The above-disclosed embodiments are merely preferred embodiments of this application and should not be construed as limiting the scope of this application. Therefore, any equivalent variations made in accordance with the claims of this application shall still fall within the scope of this application.

Claims

1. A method of generating a meeting summary, characterized by, The method comprises: acquiring a meeting record set; determining a meeting record attribute corresponding to each meeting record data in the meeting record set based on a meeting procedure and the meeting record set; determining meeting summary content corresponding to target meeting record data in the meeting record set based on the meeting record attribute and a meeting summary template, the meeting procedure being used to determine a meeting session in which the meeting record data in the meeting record set is located to generate the meeting summary content based on the meeting session; writing the meeting summary content into the meeting summary template to generate a meeting summary file corresponding to the meeting summary template; The determination of the meeting record attribute corresponding to each meeting record data based on the meeting procedure and the meeting record set comprises: determining a meeting session of each meeting record data in the meeting record set based on the meeting record set, and obtaining a first meeting record attribute corresponding to the meeting session based on a mapping relationship between a reference session and a meeting record attribute; inputting each meeting record data in the meeting record set into a first attribute determination model, extracting key semantic features based on the first attribute determination model, and identifying a second meeting record attribute corresponding to the meeting record data based on the key semantic features to output the second meeting record attribute corresponding to each meeting record data, wherein the key semantic features comprise key meeting word information, meeting word frequency distribution information, syntax-level entity meeting information and semantic-level meeting theme; determining the meeting record attribute corresponding to each meeting record data based on the first meeting record attribute and the second meeting record attribute.

2. The method of claim 1, wherein, The method further comprises: inputting the meeting summary template, the meeting procedure and the meeting record set into a first summary generation model to output meeting summary content corresponding to target meeting record data in the meeting record set; or determining target meeting record data in the meeting record set based on the meeting summary template, the meeting procedure and the meeting record set, and inputting the target meeting record data into a second summary generation model to output meeting summary content corresponding to the target meeting record data.

3. The method of claim 1, wherein, The determination of the meeting summary content corresponding to the target meeting record data in the meeting record set based on the meeting record attribute and the meeting summary template comprises: determining a reference record attribute corresponding to the meeting summary template, and determining whether the meeting record attribute of each meeting record data in the meeting record set matches the reference record attribute; if the target meeting record attribute of the target meeting record data matches the reference record attribute in the meeting record set, the meeting summary content corresponding to the target meeting record data is determined.

4. The method of claim 2, wherein, The method further comprises: collecting sample data containing target meeting record data and meeting summary content, and performing summary calibration processing on the sample data by using a calibration service to obtain sample calibration data after summary calibration processing; training a first summary generation model based on the sample calibration data to obtain a trained first summary generation model; or, training a second summary generation model based on the sample calibration data to obtain a trained second summary generation model; or, training a first summary generation model and a second summary generation model based on the sample calibration data to obtain a trained first summary generation model and a trained second summary generation model.

5. The method of claim 1, wherein, The determining the meeting summary content corresponding to the target meeting record data in the meeting record set comprises: determining target meeting record data in the meeting record set, and obtaining historical meeting summary content in the meeting summary template; performing semantic analysis processing on the target meeting record data based on the historical meeting summary content, and determining the meeting summary content corresponding to the target meeting record data based on a semantic analysis result.

6. The method of claim 5, wherein, The determining the meeting summary content corresponding to the target meeting record data based on the semantic analysis result of the semantic analysis processing comprises: if the semantic analysis result is a first result type, determining summary new content corresponding to the target meeting record data, and performing the step of writing the meeting summary content into the meeting summary template by taking the summary new content as the meeting summary content; the semantic analysis result of the first result type is that the semantics of the historical meeting summary content do not match the semantics of the target meeting record data; if the semantic analysis result is a second result type, determining summary update content for the historical meeting summary content based on the target meeting record data, and performing the step of writing the meeting summary content into the meeting summary template by taking the summary update content as the meeting summary content; the semantic analysis result of the second result type is that the semantics of the historical meeting summary content match the semantics of the target meeting record data.

7. The method of claim 1, wherein, The method further comprises: in the meeting summary file, associating the meeting summary content with the target meeting record data and the segment attribute of the target meeting record data; or, in the meeting summary file, associating the meeting summary content with the segment attribute of the target meeting record data; or, in the meeting summary file, associating the meeting summary content with the segment attribute of the target meeting record data.

8. The method of claim 1, wherein, After the step of writing the meeting summary content into the meeting summary template to generate a meeting summary file corresponding to the meeting summary template, the method further comprises: determining a meeting content key degree of meeting record data in the meeting record set, and performing data management on the meeting record data based on the content key degree.

9. The method of claim 8, wherein, The determining a meeting content key degree of meeting record data in the meeting record set comprises: extracting key semantic information of the meeting record data, calculating a semantic similarity parameter between the key semantic information and the meeting summary file, and taking the semantic similarity parameter as the meeting content key degree of the meeting record data.

10. A meeting summary generation apparatus characterized by comprising: The apparatus comprises: a record set generation module configured to obtain a meeting record set; The minutes content determination module is configured to determine a conference record attribute corresponding to each conference record data in the conference record set based on the conference regulations and the conference record attribute; and determine conference minutes content corresponding to target conference record data in the conference record set based on the conference record attribute and a conference minutes template, wherein the conference regulations are used to determine a conference session in which the conference record data in the conference record set is located, and the conference session is used to generate the conference minutes content. The minutes file generation module is configured to write the conference minutes content into the conference minutes template to generate a conference minutes file corresponding to the conference minutes template. The determination of the conference record attribute corresponding to each conference record data based on the conference regulations and the conference record attribute includes: determining a conference session of each conference record data in the conference record set in the conference regulations, and obtaining a first conference record attribute corresponding to the conference session based on a mapping relationship between a reference session and a conference record attribute. Each conference record data in the conference record set is input into a first attribute determination model, and a second conference record attribute corresponding to the conference record data is recognized based on key semantic features extracted by the first attribute determination model, so as to output the second conference record attribute corresponding to each conference record data, wherein the key semantic features include key conference word information, conference word frequency distribution information, syntax-level entity conference information, and semantic-level conference theme. The conference record attribute corresponding to each conference record data is determined based on the first conference record attribute and the second conference record attribute.

11. A computer storage medium, characterized in that The computer storage medium stores a plurality of instructions, and the instructions are suitable for being loaded and executed by the processor.

12. An electronic device, comprising: The computer storage medium stores a plurality of instructions, and the instructions are suitable for being loaded and executed by the processor. The computer storage medium stores a plurality of instructions, and the instructions are suitable for being loaded and executed by the processor.

Images