Conference summary generation method and device, computer equipment and storage medium
By preprocessing the meeting minutes text and processing the topic segmentation model, meeting minutes are generated, and poor quality of meeting minutes and topic merging in the existing technology is solved, and meeting minutes generation with unified format and rich content is achieved, and team collaboration efficiency is improved .
Patent Information
- Application Number
- CN202411002193.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-25
- Publication Date
- 2024-11-19
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The existing meeting minutes generation method cannot effectively process long meeting records, resulting in poor quality of generated summary, and the inability to merge multiple similar topics to generate summary, and the inability to generate summary for topics.
By preprocessing the meeting minutes text, dividing it into multiple text blocks, inputting it into the topic segmentation model, obtaining the topic vector, calculating the similarity and merging similar topics, and finally entering the language model to generate meeting minutes to ensure that the format is unified and the content is rich.
The generated meeting minutes are of unified format and rich content, which significantly improves the subsequent review and reference efficiency of participants and improves the efficiency and fluency of teamwork.
Smart Images

Figure CN118981532A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of data processing technology, and in particular to a method, device, computer equipment and storage medium for generating meeting minutes. Background Art
[0002] A meeting is an organized, led, and purposeful deliberative activity that is conducted at a designated time and place in accordance with a certain procedure. It is very common in office scenarios. The content of the meeting recorded during the meeting will be made into a meeting record, and a short text description containing key information of the meeting will be generated based on the meeting record, i.e., the meeting summary. However, due to the high complexity of the field of data processing technology, conventional meeting minutes methods basically summarize the entire meeting. If the meeting record is too long, the quality of the generated minutes is poor. If multiple topics are discussed in the meeting, and multiple similar topics are discussed in different time periods, the existing summary generation method cannot generate a summary after merging the topics, nor can it generate a summary for the topic, but only summarizes the content.
[0003] Therefore, how to generate high-quality, high-value meeting minutes with a unified format has become an important issue that needs to be solved urgently. Summary of the invention
[0004] In view of this, the embodiments of the present application provide a meeting minutes generation method, apparatus, computer equipment and storage medium, which can effectively solve the problem that the existing summary method cannot generate multiple summaries according to multiple topics, and the generated summary format is chaotic and the quality is poor.
[0005] In a first aspect, an embodiment of the present application provides a method for generating meeting minutes, including:
[0006] Preprocessing the meeting record text to obtain multiple meeting record text blocks;
[0007] Inputting a plurality of the conference record text blocks into a topic segmentation model to obtain topic segmentation texts;
[0008] Input the topic segmentation text into a word vector model to obtain a topic vector for each topic;
[0009] Calculating the similarity of each topic according to the topic vector of each topic, and merging topics with higher similarity than a preset similarity;
[0010] The merged topics are input into the language model to generate meeting minutes.
[0011] In some embodiments, before the preprocessing of the conference record text to obtain a plurality of conference record text blocks, the method further includes:
[0012] Acquire voice information of the video conference, perform voice recognition on the voice information, and obtain voice recognition text;
[0013] The speech recognition text is converted into a meeting minutes format to obtain the meeting record text.
[0014] In some embodiments, the step of inputting the plurality of meeting record text blocks into a topic segmentation model to obtain topic segmentation texts includes:
[0015] Inputting a plurality of the conference record text blocks into the topic segmentation model to obtain topic segmentation texts segmented according to topic types;
[0016] The topic segmented text is segmented again to obtain segmented sub-texts of the topic segmented text;
[0017] Inputting the segmented sub-texts of the topic segmented text into a word vector model to obtain a sentence vector of each segmented sub-text;
[0018] Dividing the sentence vector of each segmented sub-text according to topic type to obtain multiple sentence vectors of the same topic;
[0019] Process the multiple sentence vectors of the same topic to obtain a representation vector of the topic.
[0020] In some embodiments, calculating the similarity of each topic according to the topic vector of each topic and merging topics with higher similarity than a preset similarity includes:
[0021] Regularizing the representation vector of each topic to obtain a topic vector of each topic;
[0022] The similarity of the topic vector of each topic is calculated, and topics with similarities greater than a preset similarity are merged to obtain merged topics.
[0023] In some embodiments, inputting the merged topics into a language model to generate meeting minutes includes:
[0024] Inputting the merged topics into a language model to obtain topic summaries of each topic;
[0025] Using regular expressions to process each of the topic summaries in order to obtain a topic summary in a preset format;
[0026] Generate meeting minutes based on the preset topic summary in the template format.
[0027] In some embodiments, the topic vector of each topic is calculated by the following formula:
[0028]
[0029] in,
[0030] In the formula, v represents the topic vector, v topic The characterization vector representing the topic; p = 2; ∈ is the preset value; |v i | represents the i-th element in the vector, i = 1, 2, 3, ..; n is the vector v topic Each element in .
[0031] In some embodiments, the similarity of each of the topics is calculated using the following formula:
[0032]
[0033] Among them, cos similarity (θ) represents similarity, A and B are the topic vectors for which similarity needs to be calculated, A i and B i is each element in the vector, n represents the length of the two vectors A and B, and n is the preset value; cos similarity The value range of (θ) is the preset range.
[0034] In a second aspect, an embodiment of the present application provides a device for generating meeting minutes, including:
[0035] A conference record text block generation module is used to pre-process the conference record text to obtain multiple conference record text blocks;
[0036] A topic segmentation text processing module is used to input a plurality of the conference record text blocks into a topic segmentation model to obtain topic segmentation text;
[0037] A topic vector generation module, used for inputting the topic segmentation text into a word vector model to obtain a topic vector for each topic;
[0038] A topic merging module, used to calculate the similarity of each topic according to the topic vector of each topic, and merge topics with a similarity higher than a preset value;
[0039] The meeting minutes generation module is used to input the merged topics into the language model to generate meeting minutes.
[0040] In a third aspect, an embodiment of the present application provides a computer device, comprising a processor and a memory, wherein the memory stores a computer program, and the processor is used to execute the computer program to implement the meeting minutes generation method described in the first aspect above.
[0041] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium storing a computer program. When the computer program is executed on a processor, the method for generating meeting minutes described in the first aspect above is implemented.
[0042] The embodiments of the present application have the following beneficial effects: by preprocessing the meeting record text, a plurality of meeting record text blocks are obtained, so that the plurality of meeting record text blocks can be input into the topic segmentation model; the topic segmentation model further divides each meeting record text block into smaller text blocks according to the content thereof, thereby obtaining topic segmentation text; the topic segmentation text is input into the word vector model; the key information in the topic segmentation text is captured by the word vector model to obtain the topic vector of each topic; finally, the similarity of each topic is calculated according to the topic vector of each topic, and topics with a similarity higher than a preset similarity are merged; the merged topics are input into the language model; the language model generates corresponding meeting minutes according to the input topic content, so that the generated meeting minutes have a unified format, rich content and high value, which greatly facilitates the subsequent review and reference of the participants, and significantly improves the efficiency and fluency of team collaboration. BRIEF DESCRIPTION OF THE DRAWINGS
[0043] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.
[0044] Figure 1 A framework / application scenario diagram of the method for generating meeting minutes according to an embodiment of the present application is shown;
[0045] Figure 2 A flow chart of the method for generating meeting minutes according to an embodiment of the present application is shown;
[0046] Figure 3 A flow chart showing a method for generating meeting minutes according to another embodiment of the present application is shown;
[0047] Figure 4 A flow chart showing a method for generating meeting minutes according to another embodiment of the present application is shown;
[0048] Figure 5 A flow chart showing a method for generating meeting minutes according to another embodiment of the present application is shown;
[0049] Figure 6 A flow chart showing a method for generating meeting minutes according to another embodiment of the present application is shown;
[0050] Figure 7 A structural schematic diagram of a method for generating meeting minutes according to an embodiment of the present application is shown. DETAILED DESCRIPTION
[0051] The technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments.
[0052] The components of the embodiments of the present application generally described and shown in the drawings herein may be arranged and designed in various configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the application claimed for protection, but merely represents the selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without making creative work belong to the scope of protection of the present application.
[0053] Hereinafter, the terms "including", "having" and their cognates that can be used in various embodiments of the present application are intended only to indicate specific features, numbers, steps, operations, elements, components or a combination of the foregoing items, and should not be understood as first excluding the existence of one or more other features, numbers, steps, operations, elements, components or a combination of the foregoing items or increasing the possibility of one or more features, numbers, steps, operations, elements, components or a combination of the foregoing items. In addition, the terms "first", "second", "third" and the like are only used to distinguish descriptions and cannot be understood as indicating or implying relative importance.
[0054] Unless otherwise defined, all terms (including technical terms and scientific terms) used herein have the same meanings as those generally understood by those skilled in the art to which the various embodiments of the present application belong. The terms (such as those defined in generally used dictionaries) will be interpreted as having the same meanings as the contextual meanings in the relevant technical field and will not be interpreted as having idealized meanings or overly formal meanings unless clearly defined in the various embodiments of the present application.
[0055] In conjunction with the accompanying drawings, some embodiments of the present application are described in detail below. In the absence of conflict, the following embodiments and features in the embodiments can be combined with each other.
[0056] Considering that the conventional method of taking minutes of a meeting is to summarize the entire meeting, if the minutes are too long, the quality of the minutes generated will be poor. If multiple topics are discussed in the meeting, and multiple similar topics are discussed in different time periods, the existing summary generation method cannot generate a summary after merging the topics, and cannot generate a summary for the topic. Therefore, the present application proposes a method, device, computer equipment and storage medium for generating minutes of a meeting, which first pre-processes the meeting minutes text to obtain multiple meeting minutes text blocks, so that the multiple meeting minutes text blocks can be input into the topic segmentation model. The topic segmentation model further divides each meeting minutes text block into smaller text blocks according to the content, thereby obtaining topic segmentation text. Then, the topic segmentation text is input into the word vector model, and the key information in the topic segmentation text is captured by the word vector model to obtain the topic vector of each topic. Finally, the similarity of each topic is calculated according to the topic vector of each topic, and the topics with higher than the preset similarity are merged. The merged topics are input into the language model, and the language model will generate corresponding meeting minutes based on the input topic content. The meeting minutes generated by the above method have a unified format, rich content and high teaching value, which greatly facilitates the subsequent review and reference of the participants and significantly improves the efficiency and fluency of team collaboration.
[0057] The method for generating meeting minutes is described below in conjunction with some specific embodiments.
[0058] Figure 1 A schematic diagram showing an application environment of the method for generating meeting minutes of an embodiment of the present application is shown. Specifically, the method for generating meeting minutes of the present application is applied in a computer system, and the system includes: Figure 1 The client and server shown in the figure communicate with each other through the network. The client, also known as the user end, refers to the program corresponding to the server and providing local services to the client. The client can be installed on, but not limited to, various personal computers, laptops, smart phones, tablet computers, portable wearable devices, etc. The server can be implemented as an independent server or a server cluster consisting of multiple servers.
[0059] Figure 2 A flow chart of a method for generating meeting minutes of an embodiment of the present application is shown. The method for generating meeting minutes comprises the following steps:
[0060] Step S100, pre-processing the conference record text to obtain a plurality of conference record text blocks.
[0061] Specifically, the preprocessing process may include the following steps: First, clean the meeting records and remove the useless information therein, such as extra spaces, line breaks, tab characters, HTML tags, and special characters, etc.; Second, split the continuous text into a sequence of words, phrases or sentences; Third, split the long text into paragraphs or text blocks according to the theme or logic; At the same time, convert the vocabulary, symbols, abbreviations, dates, numbers, etc. in the text into a unified format; In addition, remove the words that frequently appear in the text but contribute little to the analysis, such as function words like "de", "shi", "zai", etc.; Finally, merge multiple short sentences that appear consecutively and are from the same speaker and have a length of less than 10 characters into one sentence. Additionally, for extremely long meeting record texts, we use the line break character "\n" and the full stop "。" as the splitting points to strictly split them into multiple text blocks with no more than 4096 characters for subsequent processing.
[0062] It should be noted that the above length value and the character number value of the text block are both examples, and can be adaptively adjusted according to the specific situation.
[0063] Step S200, input multiple said meeting record text blocks into the topic segmentation model to obtain topic-segmented text.
[0064] Specifically, the topic segmentation model is a pre-trained machine learning model. In this application, this model adopts the T5-Encoder architecture. When we input multiple meeting record text blocks into this topic segmentation model, the model can further split each text block into smaller text blocks according to the content of each text block. After being processed by the topic segmentation model, the output result is called "topic-segmented text".
[0065] Step S300, input the said topic-segmented text into the word vector model to obtain the topic vectors of each topic.
[0066] Specifically, input the topic-segmented text obtained in step S200 into the word vector model to obtain the topic vectors of each topic. Among them, the word vector model used is the Dmeta-embedding-zh model, which is a machine learning model that can convert vocabulary or entire text segments into vector representations in a high-dimensional space. During the conversion process, the Dmeta-embedding-zh model will capture the key information in each topic-segmented text and encode it into a vector with a fixed length, that is, the "topic vector".
[0067] Step S400, calculate the similarity of each said topic according to the topic vectors of each topic, and merge the topics with a similarity higher than the preset similarity.
[0068] For example, the similarity of topics can be calculated by calculating cosine similarity, Euclidean distance, Jaccard similarity, etc. While setting a preset similarity threshold based on experimental data, it can also be adjusted according to actual needs, and topics with higher similarity than the preset similarity can be merged to reduce redundancy and duplication in the data.
[0069] Step S500: input the merged topics into a language model to generate meeting minutes.
[0070] Specifically, the merged topics are input into the language model. The language model can be selected according to the actual situation, such as Qwen1.5-7B-chat, or models based on the Transformer structure, such as the GPT series (GPT-2, GPT-3, etc.) and BERT variants (such as T5, BART, etc.). When setting the generated text for the above model, the corresponding parameters need to be configured, such as the generated length (that is, the number of words or sentences that the meeting minutes should contain), the temperature value (used to control the degree of randomness of the generated text), and the prefix text (if a specific beginning or format needs to be added before the generated meeting minutes). Subsequently, the generation mechanism is triggered, and the language model will generate the corresponding meeting minutes based on the input topic content.
[0071] Furthermore, the generated meeting minutes may need to be formatted and corrected for grammatical errors to improve their quality. In addition, user feedback can be collected to understand the effect of the generated meeting minutes in actual applications, and the model can be continuously optimized and adjusted based on this feedback to continuously improve its performance and applicability.
[0072] The meeting minutes generation method of the embodiment of the present application obtains multiple meeting minutes text blocks by preprocessing the meeting minutes text, inputs the multiple meeting minutes text blocks into the topic segmentation model to obtain topic segmentation text, and then inputs the topic segmentation text into the word vector model to obtain the topic vector of each topic. Finally, the similarity of each topic is calculated according to the topic vector of each topic, and topics with higher than a preset similarity are merged, and the merged topics are input into the language model to generate meeting minutes, which greatly facilitates the subsequent review and reference of the participants and improves the efficiency and fluency of team collaboration.
[0073] In an optional embodiment, if Figure 3 As shown, in step S100, before the conference record text is preprocessed to obtain a plurality of conference record text blocks, the following is also included:
[0074] Step S101, obtaining voice information of a video conference, performing voice recognition on the voice information, and obtaining a voice recognition text.
[0075] For example, the voice information comes from the speech of a single participant or multiple participants in a video conference. The system captures and extracts the voice information from the audio stream of the video conference by collecting audio signals, denoising, enhancing, etc. The acoustic features of the voice signal (such as pitch, volume, timbre, etc.) are analyzed by the speech recognition model and matched with the predefined vocabulary library or language model to generate speech recognition text.
[0076] Step S102, converting the speech recognition text into a meeting minutes format to obtain a meeting record text.
[0077] Specifically, before text conversion, the format requirements of the meeting minutes must be clarified first, such as the meeting time, location, participants, main topics of the meeting, discussion content, decision results, and subsequent planned actions, etc. Then, the speech recognition text is converted according to the meeting minutes format to obtain the initial meeting record text.
[0078] The meeting minutes generation method of the embodiment of the present application converts and organizes the speech recognition text according to the meeting minutes format, and finally generates a clear and accurate meeting record text.
[0079] In an optional embodiment, if Figure 4 As shown, in step S200, the plurality of conference record text blocks are input into the topic segmentation model to obtain topic segmentation texts, including:
[0080] Step S201: input a plurality of the conference record text blocks into the topic segmentation model to obtain topic segmentation texts segmented according to topic types.
[0081] Specifically, in this application, the topic segmentation model adopts the T5-Encoder architecture. When we input multiple meeting record text blocks into the topic segmentation model, the model can further segment each text block into smaller text blocks around specific topics or themes based on the content of each text block. After being processed by the topic segmentation model, the topic segmentation text segmented by topic type is obtained.
[0082] Step S202: segment the topic segment text again to obtain segmented sub-texts of the topic segment text.
[0083] Specifically, the topic segmented text segmented according to the topic type is segmented again into segmented sub-texts with a length less than 512. In other implementations, the segmented length can be set as needed.
[0084] Step S203: input the segmented sub-texts of the topic segmented text into a word vector model to obtain a sentence vector of each segmented sub-text.
[0085] Specifically, the segmented sub-texts with a length less than 512 are input into the Dmeta-embedding-zh word vector model to obtain several sentence vectors with a length of 768.
[0086] Among them, the Dmeta-embedding-zh word vector model is a BERT-like model trained with Chinese corpus. This model is modified based on BERT, especially adding a pooling layer. The text is encoded as an id by the BERT tokenizer and then input into the model. After encoding and pooling, the model obtains a sentence vector with a shape of (sequence_length, 768), where sequence_length is the token length of the input text.
[0087] Step S204, dividing the sentence vector of each segmented sub-text according to topic type to obtain multiple sentence vectors of the same topic.
[0088] Specifically, a classifier can be trained using an appropriate classification algorithm, such as logistic regression, decision tree, random forest, neural network, etc., to classify the sentence vector of each segmented sub-text, thereby obtaining multiple sentence vectors belonging to the same topic.
[0089] Step S205, processing the multiple sentence vectors of the same topic to obtain a representation vector of the topic.
[0090] Specifically, all sentence vectors of the same topic can be aggregated. There are many aggregation methods, among which the most commonly used is averaging (also called average pooling). That is, the values of all sentence vectors in each dimension are added together, and then divided by the number of sentence vectors to obtain a new vector, which is the representation vector of the topic.
[0091] For example, in addition to averaging, other aggregation methods can also be considered, such as maximum pooling (taking the maximum value of each dimension), minimum pooling (taking the minimum value of each dimension), or weighted sum (giving different weights according to the importance of the sentence and then summing it up). The specific usage method depends on the actual application scenario and requirements.
[0092] After obtaining the characterization vector of the topic, it may be necessary to perform standardization to ensure that the characterization vectors of different topics are numerically comparable. Standardization can be achieved by subtracting the mean of each dimension value of the characterization vector and dividing it by the standard deviation, depending on the needs of subsequent analysis.
[0093] Finally, the representation vector of the topic is output. The representation vector will be used as a representative of the topic in subsequent analysis and can be used for various tasks, such as topic clustering, topic similarity calculation, and topic trend prediction.
[0094] The meeting minutes generation method of the embodiment of the present application processes the meeting record text block and finally obtains the representation vector, so as to generate the meeting minutes with unified format, rich content and high value in the subsequent steps, which not only facilitates the subsequent review and reference of the participants, but also improves the efficiency and fluency of team collaboration.
[0095] In an optional embodiment, if Figure 5 As shown, in step S400, the similarity of each topic is calculated according to the topic vector of each topic, and topics with higher similarity than a preset similarity are merged, including:
[0096] Step S401, regularize the representation vector of each topic to obtain the topic vector of each topic.
[0097] Specifically, the topic vector can be regularized by the following formula to obtain the topic vector of each topic:
[0098]
[0099] in,
[0100] In the above formula, v represents the topic vector, v topic The characterization vector representing the topic; p = 2; ∈ = 1e-12, which is used to prevent the denominator from being 0; |v i | represents the i-th element in the vector, i = 1, 2, 3, ..; n is the vector v topic Each element in .
[0101] It is worth noting that the values of the above parameters are only examples and can be adaptively adjusted according to specific circumstances.
[0102] Step S402, calculating the similarity of the topic vector of each topic, merging topics with similarities greater than a preset similarity to obtain merged topics.
[0103] Specifically, after obtaining the regularized topic vector of each topic, the similarity of the topic vector of each topic can be calculated by the following formula:
[0104]
[0105] Among them, cos similarity (θ) represents similarity, and its value range is the preset range [0,1]. A and B are the topic vectors for which similarity needs to be calculated. i and B i Represents each element in the quantity, n represents the length of the two vectors A and B, n=768.
[0106] Furthermore, the preset similarity is set to 0.9. After the similarity is calculated, topics with similarity greater than 0.9 will be merged. At the same time, in order to ensure the quality of the generated summary, topics with a length less than 512 will be merged into the adjacent topics with the smallest length.
[0107] It is worth noting that the values of the above parameters are only examples and can be adaptively adjusted according to specific circumstances.
[0108] The meeting minutes generation method of the embodiment of the present application obtains the topic vector of each topic by regularizing the characterization vector of the topic, and merges the topics with similarities greater than a preset similarity by calculating the similarity of the topic vectors of each topic to obtain merged topics, so as to generate meeting minutes with unified format, rich content and high value in subsequent steps, thereby facilitating the subsequent review and reference of the participants and improving the efficiency and fluency of team collaboration.
[0109] In an optional embodiment, if Figure 6 As shown, in step S500, the merged topics are input into a language model to generate meeting minutes, including:
[0110] Step S501: input the merged topics into a language model to obtain topic summaries of each topic.
[0111] Specifically, the merged topics are input into the language model Qwen1.5-7B-chat one by one to obtain the summary of each topic. Through the prompt engineering, the Qwen1.5-7B-chat large model is guided to generate a summary format containing three parts: "discussion topic", "discussion content" and "to-do items".
[0112] It is worth noting that the to-do section will depend on the content of the specific topic, and sometimes there may be no clear to-do.
[0113] Regarding how the Qwen1.5-7B-chat model generates a summary of a single topic, the implementation process is as follows:
[0114] Prompt words for constructing the "system" role: "You are an assistant who is good at writing summaries in a specified format. The user will provide you with a discussion record, and please write a summary based on this record. The format requirements of the summary are as follows:\nDiscussion topic: <Summarize the core topic of the discussion record here>\nDiscussion content: <Please describe in detail the main points or content of each participant in the discussion (if there are specific names in the meeting record, please give them)>\nTo-do items: <If the discussion record clearly mentions the items that need to be completed, please list them one by one. If the to-do items are not mentioned in the discussion, please reply "Not mentioned.">".
[0115] Construct the prompt words for the "user" role: "The following is a discussion record, please generate a summary based on it: \n{topic}". Among them, "{topic}" is a placeholder, which needs to be replaced with the specific topic content when actually generating the summary.
[0116] Enter the above-mentioned prompt words of the "system" and "user" roles into the Qwen1.5-7B-chat large model to obtain the summary of the corresponding topic.
[0117] Repeat the above process of generating a single topic summary until summaries of all topics are generated.
[0118] Step S502: Use regular expressions to process each of the topic summaries in order to obtain a topic summary in a preset format.
[0119] Specifically, the preset format usually refers to the format of the topic summary, for example, discussion topic, discussion content, to-do items, title, date, participants, etc.
[0120] For example, you can use regular expressions to extract the content after the keywords "discussion topic", "discussion content" and "to-do items" in order. If the content exists, delete the previous label. The to-do items of each topic are temporarily stored in a list, and then regenerate labels such as "1." and "2." after all the to-do items of all topics are extracted.
[0121] Step S503: Generate meeting minutes based on the preset topic summary in a template format.
[0122] Specifically, fill in the formatted summaries of each topic into the template, put the to-do items of all topics behind all topics and contents, and get the final summary.
[0123] The meeting minutes generation method of the embodiment of the present application inputs the merged topics into the language model to obtain the topic summaries of each topic, uses regular expressions to process each topic summary in sequence to obtain a topic summary in a preset format, and generates meeting minutes according to the topic summary in the preset format according to a template format.
[0124] Figure 7 A schematic diagram of the structure of a device for generating meeting minutes according to an embodiment of the present application is shown, and the device for generating meeting minutes includes:
[0125] The conference record text block generation module 71 is used to pre-process the conference record text to obtain multiple conference record text blocks.
[0126] The topic segmentation text processing module 72 is used to input the plurality of conference record text blocks into the topic segmentation model to obtain topic segmentation text.
[0127] The topic vector generation module 73 is used to input the topic segmentation text into the word vector model to obtain the topic vector of each topic.
[0128] The topic merging module 74 is used to calculate the similarity of each topic according to the topic vector of each topic, and merge topics with a similarity higher than a preset value.
[0129] The meeting minutes generation module 75 is used to input the merged topics into the language model to generate meeting minutes.
[0130] Optionally, the meeting minutes generating device further includes:
[0131] The voice recognition text acquisition module is used to acquire the voice information of the video conference, perform voice recognition on the voice information, and obtain the voice recognition text.
[0132] The conference record text acquisition module is used to convert the speech recognition text into a conference minutes format to obtain a conference record text.
[0133] Optionally, the topic segmentation text processing module includes:
[0134] The topic segmentation text acquisition unit is used to input the plurality of conference record text blocks into the topic segmentation model to obtain the topic segmentation text segmented according to topic type.
[0135] The segmented sub-text obtaining unit is used to segment the topic segmented text again to obtain the segmented sub-texts of the topic segmented text.
[0136] The sentence vector acquisition unit is used to input the segmented sub-texts of the topic segmented text into a word vector model to obtain a sentence vector of each segmented sub-text.
[0137] The sentence vector processing unit is used to divide the sentence vector of each segmented sub-text according to the topic type to obtain multiple sentence vectors of the same topic.
[0138] The representation vector acquisition unit is used to process the multiple sentence vectors of the same topic to obtain the representation vector of the topic.
[0139] Optionally, the topic merging module includes:
[0140] The representation vector processing unit is used to regularize the representation vector of each topic to obtain the topic vector of each topic.
[0141] The topic merging unit is used to calculate the similarity of the topic vector of each topic, merge the topics with similarity greater than a preset similarity, and obtain a merged topic.
[0142] Optionally, the meeting minutes generating module includes:
[0143] The topic summary acquisition unit is used to input the merged topics into the language model to obtain the topic summary of each topic.
[0144] The topic summary processing unit is used to process each of the topic summaries in sequence using a regular expression to obtain a topic summary in a preset format.
[0145] The meeting minutes generating unit is used to generate the meeting minutes according to the template format of the topic summary in the preset format.
[0146] It can be understood that the device of this embodiment corresponds to the meeting minutes generation method of the above embodiment, and the optional items in the above embodiment are also applicable to this embodiment, so they will not be described repeatedly here.
[0147] The present application also provides a computer device. Exemplarily, the computer device includes a processor and a memory, wherein the memory stores a computer program, and the processor runs the computer program to enable the terminal device to execute the functions of each module in the above-mentioned voiceprint registration method or the above-mentioned voiceprint registration device.
[0148] Among them, the processor can be an integrated circuit chip with signal processing capabilities. The processor can be a general-purpose processor, including a central processing unit (CPU), a graphics processing unit (GPU) and a network processor (NP), a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA) or at least one of other programmable logic devices, discrete gates or transistor logic devices, and discrete hardware components. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc., which can implement or execute the disclosed methods, steps and logic block diagrams in the embodiments of the present application.
[0149] The memory may be, but is not limited to, a random access memory (RAM), a read only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), etc. The memory is used to store a computer program, and the processor may execute the computer program accordingly after receiving an execution instruction.
[0150] The present application also provides a computer-readable storage medium for storing the computer program used in the above-mentioned computer device. For example, the computer-readable storage medium may include, but is not limited to, various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.
[0151] In several embodiments provided in the present application, it should be understood that the disclosed devices and methods can also be implemented in other ways. The device embodiments described above are merely schematic. For example, the flowcharts and structure diagrams in the accompanying drawings show the possible architecture, functions and operations of the devices, methods and computer program products according to multiple embodiments of the present application. In this regard, each box in the flowchart or block diagram can represent a module, a program segment or a part of a code, and the module, a program segment or a part of a code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in an alternative implementation, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the structure diagram and / or the flow diagram, and the combination of boxes in the structure diagram and / or the flow diagram, can be implemented with a dedicated hardware-based system that performs a specified function or action, or can be implemented with a combination of dedicated hardware and computer instructions.
[0152] In addition, the functional modules or units in the various embodiments of the present application may be integrated together to form an independent part, or each module may exist separately, or two or more modules may be integrated to form an independent part.
[0153] If the functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a smart phone, a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in the various embodiments of the present application.
[0154] The above description is only a specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application.
Claims
1. A method for generating meeting minutes, characterized in that: include: Preprocessing the meeting record text to obtain multiple meeting record text blocks; Inputting a plurality of the conference record text blocks into a topic segmentation model to obtain topic segmentation texts; Input the topic segmentation text into a word vector model to obtain a topic vector for each topic; Calculating the similarity of each topic according to the topic vector of each topic, and merging topics with higher similarity than a preset similarity; The merged topics are input into the language model to generate meeting minutes.
2. The method for generating meeting minutes according to claim 1, characterized in that: Before the conference record text is preprocessed to obtain a plurality of conference record text blocks, the method further includes: Acquire voice information of the video conference, perform voice recognition on the voice information, and obtain voice recognition text; The speech recognition text is converted into a meeting minutes format to obtain the meeting record text.
3. The method for generating meeting minutes according to claim 1, characterized in that: The step of inputting the plurality of conference record text blocks into the topic segmentation model to obtain topic segmentation texts includes: Inputting a plurality of the conference record text blocks into the topic segmentation model to obtain topic segmentation texts segmented according to topic types; The topic segmented text is segmented again to obtain segmented sub-texts of the topic segmented text; Inputting the segmented sub-texts of the topic segmented text into the word vector model to obtain a sentence vector of each segmented sub-text; Dividing the sentence vector of each segmented sub-text according to topic type to obtain multiple sentence vectors of the same topic; Process the multiple sentence vectors of the same topic to obtain a representation vector of the topic.
4. The method for generating meeting minutes according to claim 3, characterized in that: The calculating the similarity of each topic according to the topic vector of each topic and merging topics with a similarity higher than a preset similarity includes: Regularizing the representation vector of each topic to obtain a topic vector of each topic; The similarity of the topic vector of each topic is calculated, and topics with similarities greater than a preset similarity are merged to obtain merged topics.
5. The method for generating meeting minutes according to claim 1, characterized in that: The merged topics are input into the language model to generate meeting minutes, including: Inputting the merged topics into a language model to obtain topic summaries of each topic; Using regular expressions to process each of the topic summaries in order to obtain a topic summary in a preset format; Generate meeting minutes based on the preset topic summary in the template format.
6. The method for generating meeting minutes according to claim 3, characterized in that: The topic vector of each topic is calculated by the following formula: in, In the formula, v represents the topic vector, v topic The characterization vector representing the topic; p = 2; ∈ is the preset value; |v i | represents the i-th element in the vector, i = 1, 2, 3, ..; n is the vector v topic Each element in .
7. The method for generating meeting minutes according to claim 1, characterized in that: The similarity of each topic is calculated using the following formula: Among them, cos similarity (θ) represents similarity, A and B are the topic vectors for which similarity needs to be calculated, A i and B i are the i-th elements in vectors A and B respectively, and n represents the length of the vector.
8. A device for generating meeting minutes, characterized in that: include: A conference record text block generation module is used to pre-process the conference record text to obtain multiple conference record text blocks; A topic segmentation text processing module is used to input a plurality of the conference record text blocks into a topic segmentation model to obtain topic segmentation text; A topic vector generation module, used for inputting the topic segmentation text into a word vector model to obtain a topic vector for each topic; A topic merging module, used to calculate the similarity of each topic according to the topic vector of each topic, and merge topics with a similarity higher than a preset value; The meeting minutes generation module is used to input the merged topics into the language model to generate meeting minutes.
9. A computer device, characterized in that: The computer device comprises a processor and a memory, wherein the memory stores a computer program, and the processor is configured to execute the computer program to implement the method for generating meeting minutes according to any one of claims 1 to 7.
10. A computer storage medium, characterized in that: The computer program is stored therein, and when the computer program is executed on a processor, the method for generating meeting minutes according to any one of claims 1 to 7 is implemented.
Citation Information
Patent Citations
News information presentation method based on deep learning and related equipment
CN111813936A
Simmary processing method and device, equipment and storage medium
CN112861510A
Text processing method and device, electronic equipment and computer readable storage medium
CN114492375A
Abstract generation method and device, equipment and medium
CN114722836A
Event context generation method, electronic equipment and storage medium
CN116361468A
Cited By
Real-time conference summary generation method and device, equipment, medium and computer program product
CN119647422A
Real-time conference abstract generation method, apparatus, device, medium, and computer program product
CN119647422B
Information processing method and device, readable storage medium and program product
CN120263848A
