Conference summary generation method, electronic equipment and readable storage medium
By obtaining the interactive pair text and Q&A text in the meeting record text, associating the spokesperson's content and replacing the interactive pair text, the problem of semantic discontinuity of meeting minutes is solved, and the accuracy of generating meeting minutes is improved.
Patent Information
- Application Number
- CN202411287063.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-13
- Publication Date
- 2024-12-06
AI Technical Summary
When generating meeting minutes, the existing technology directly compresses the speech content of non-leaders, resulting in the problem of semantic discontinuity of meeting minutes.
By obtaining the interactive pair text in the meeting record text, obtaining the Q&A text to associate the context semantics of the speaker, and replacing the interactive pair text with the Q&A text, generating meeting minutes to ensure the context relevance of the speech content.
It solves the problem of semantic discontinuity in meeting minutes, and improves the accuracy and consistency of the content of the generated meeting minutes.
Smart Images

Figure CN119088952A_ABST
Abstract
Description
Technical Field
[0001] The present application belongs to the field of artificial intelligence, and specifically relates to a method for generating meeting minutes, an electronic device, and a readable storage medium. Background Art
[0002] With the popularization of smart office, the issue of how to intelligently generate meeting minutes is often involved in conference application scenarios.
[0003] In the process of generating meeting minutes, the relevant technology will first distinguish the identity type of the speaker in the meeting record text (leader or non-leader), and then compress the speech content of the non-leader in the meeting record text to obtain a compressed content text, and then use the compressed content text to replace the speech content of the non-leader in the meeting record text, and obtain the meeting minutes based on this.
[0004] In the related art, this method of generating meeting minutes directly compresses the speech content of non-leaders, and the compressed content is prone to missing, resulting in semantic discontinuity in the obtained meeting minutes. Summary of the invention
[0005] The embodiments of the present application provide a method for generating meeting minutes, an electronic device, and a readable storage medium, which can solve the problem of semantic discontinuity in meeting minutes obtained by related technologies.
[0006] In a first aspect, an embodiment of the present application provides a method for generating meeting minutes, the method comprising:
[0007] Acquire a conference record text, wherein the conference record text includes speech contents of multiple speakers;
[0008] Acquire an interactive pair text from the conference record text, wherein the interactive pair text includes speech content of at least two speakers;
[0009] Based on the interactive text, obtaining a question-and-answer text corresponding to the interactive text, wherein the question-and-answer text is used to associate contextual semantics between speech contents of at least two speakers in the interactive text;
[0010] Using the question-and-answer text to replace the interaction text in the conference record text to obtain a first record text;
[0011] Generate meeting minutes based on the first record text.
[0012] In a second aspect, an embodiment of the present application provides an electronic device, comprising a processor and a memory, wherein the memory stores a program or instruction running on the processor, and when the program or instruction is executed by the processor, the steps of the method described in the first aspect are implemented.
[0013] In a third aspect, an embodiment of the present application provides a computer-readable storage medium, on which a program or instruction is stored, and when the program or instruction is executed, the steps of the method described in the first aspect are implemented.
[0014] In a fourth aspect, an embodiment of the present application provides a computer program product, wherein the computer program product includes a computer program, and when the computer program is executed by a processor, the steps of the method described in the first aspect are implemented.
[0015] At least one of the above technical solutions provided in the embodiments of the present application can achieve the following technical effects:
[0016] In an embodiment of the present application, a conference record text is obtained, wherein the conference record text includes the speech contents of multiple speakers; an interactive pair text is obtained from the conference record text, wherein the interactive pair text includes the speech contents of at least two speakers; based on the interactive pair text, a question and answer text corresponding to the interactive pair text is obtained, wherein the question and answer text is used to associate the contextual semantics between the speech contents of the at least two speakers in the interactive pair text; the interactive pair text in the conference record text is replaced with the question and answer text to obtain a first record text; and based on the first record text, a meeting minutes is generated. In this way, in the process of generating the meeting minutes, by obtaining the question and answer text corresponding to the interactive pair text, wherein the question and answer text is used to associate the contextual semantics between the speech contents of the at least two speakers in the interactive pair text, and replacing the interactive pair text in the conference record text with the question and answer text, the relevance of the contextual semantics of the speech contents is ensured, thereby solving the problem of semantic discontinuity in the meeting minutes obtained by the related art to a certain extent. BRIEF DESCRIPTION OF THE DRAWINGS
[0017] Figure 1 It is a flowchart of a method for generating meeting minutes provided in an embodiment of the present application.
[0018] Figure 2 It is a flowchart of the process of generating meeting minutes based on the first record text in an embodiment of the present application.
[0019] Figure 3 It is a flowchart of a method for generating meeting minutes provided in an embodiment of the present application.
[0020] Figure 4-1 It is a schematic diagram of a meeting record text provided in an embodiment of the present application.
[0021] Figure 4-2 It is a schematic diagram of a meeting record text provided in an embodiment of the present application.
[0022] Figure 4-3 It is a schematic diagram of a first record text provided in an embodiment of the present application.
[0023] Figure 4-4 It is a schematic diagram of another first record text provided in an embodiment of the present application.
[0024] Figure 5 It is a structural block diagram of a meeting minutes generating device provided in an embodiment of the present application.
[0025] Figure 6 It is a structural block diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION
[0026] The following will be combined with the drawings in the embodiments of the present application to clearly describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.
[0027] The terms "first", "second", etc. in the specification and claims of the present application are used to distinguish similar objects, and are not used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable under appropriate circumstances, so that the embodiments of the present application can be implemented in an order other than those illustrated or described here, and the objects distinguished by "first", "second", etc. are generally of one type, and the number of objects is not limited. For example, the first object can be one or more. In addition, "and / or" in the specification and claims represents at least one of the connected objects, and the character " / " generally indicates that the objects associated with each other are in an "or" relationship.
[0028] The meeting minutes generation method provided in the embodiment of the present application can be executed by a target device, wherein the target device can be one electronic device or multiple electronic devices. In other words, the meeting minutes generation method provided in the embodiment of the present application can be executed by an electronic device, wherein the electronic device can be a terminal device such as a desktop computer, a laptop computer, a mobile phone, a tablet, or a server, such as an independent physical server or a cloud server capable of cloud computing. In the case where the meeting minutes generation method provided in the embodiment of the present application is executed by multiple electronic devices, these multiple electronic devices can form a service cluster, which cooperate with each other to complete each step.
[0029] The meeting minutes generation method provided in the embodiment of the present application can be applied to an intelligent office assistant, which can generate meeting minutes of the corresponding meeting based on the meeting record text obtained by the conference system, thereby assisting employees to achieve efficient office work.
[0030] Figure 1 : is a flow chart of a method for generating meeting minutes provided in an embodiment of the present application. Figure 1 The method for generating meeting minutes provided in the embodiment of the present application includes:
[0031] Step 110, obtaining a conference record text, wherein the conference record text includes speech contents of multiple speakers;
[0032] The embodiments of the present application do not limit the specific method of obtaining the conference record text. The conference record text can be transcribed and generated by the conference system based on the collected conference audio. The conference audio can be an online conference audio or an offline conference audio obtained by various conference audio collection devices. At the same time, the conference system also provides a voice identity recognition function, and the identity of the speaker of each text content can be marked in the conference record text transcribed and generated by the conference system.
[0033] The meeting record text obtained by the target device may also be the meeting record text transmitted by other devices, or the meeting record text input by the user. The multiple speakers included in the meeting record text may be obtained by machine marking or manual marking.
[0034] Step 120, obtaining an interactive pair text from the conference record text, wherein the interactive pair text includes speech contents of at least two speakers;
[0035] In the embodiment of the present application, interactive text extraction rules can be set. That is, in step 120, interactive text can be obtained from the conference record text according to the interactive text extraction rules. The interactive text extraction rules can be rules related to the frequency of speakers' speeches, rules for interactions between multiple speakers, or rules related to the content or duration of speakers' speeches.
[0036] Step 130, based on the interactive text, obtaining a question-and-answer text corresponding to the interactive text, wherein the question-and-answer text is used to associate contextual semantics between speech contents of the at least two speakers in the interactive text;
[0037] Wherein, in step 130, based on the interaction text, obtaining the question-answer text corresponding to the interaction text may include: inputting the interaction text and the first prompt word into a first language model to obtain the question-answer text corresponding to the interaction text; wherein the first prompt word is a prompt word used to generate the question-answer text. The first language model may be a machine learning language model.
[0038] The question and answer text in the embodiment of the present application may include a question text and an answer text, wherein the answer text is the answer to the question text, and the question text and the answer text may correspond one to one.
[0039] Step 140, using the question and answer text to replace the interaction text in the conference record text to obtain a first record text;
[0040] In the embodiment of the present application, the first record text is obtained by modifying the meeting record text.
[0041] Step 150: Generate meeting minutes based on the first record text.
[0042] In an embodiment of the present application, the first recorded text can be input into a deep learning model for language processing, and the model output can be used as the meeting minutes. Alternatively, further processing can be performed on the basis of the first recorded text to obtain a second recorded text, and then the second recorded text can be input into a deep learning model for language processing, and the model output can be used as the meeting minutes.
[0043] The method for generating minutes of a meeting provided in an embodiment of the present application comprises the steps of obtaining a meeting record text, wherein the meeting record text includes the speech contents of multiple speakers; obtaining an interactive pair text from the meeting record text, wherein the interactive pair text includes the speech contents of at least two speakers; obtaining a question-and-answer text corresponding to the interactive pair text based on the interactive pair text, wherein the question-and-answer text is used to associate the contextual semantics between the speech contents of the at least two speakers in the interactive pair text; replacing the interactive pair text in the meeting record text with the question-and-answer text to obtain a first record text; and generating minutes of a meeting based on the first record text. In this way, in the process of generating minutes of a meeting, by obtaining the question-and-answer text corresponding to the interactive pair text, wherein the question-and-answer text is used to associate the contextual semantics between the speech contents of the at least two speakers in the interactive pair text, and replacing the interactive pair text in the meeting record text with the question-and-answer text, the relevance of the contextual semantics of the speech contents is ensured, thereby solving the problem of semantic discontinuity in the meeting minutes obtained by the related art to a certain extent.
[0044] Figure 2 This is a flowchart of an example process of generating meeting minutes based on the first record text in an embodiment of the present application. Figure 2 In the embodiment of the present application, the process of generating the meeting minutes based on the first record text may include:
[0045] Step 210, obtaining a semantically compressed text of the target speaker's speech content in the first recorded text;
[0046] In an embodiment of the present application, the target speaker may be a speaker who meets the set target conditions. The target condition may be a speaker adjacent to the question and answer text. If the question and answer text is a copy, the target speaker may be a speaker adjacent to the question and answer text. The target speaker may also be a speaker between two adjacent question and answer texts in the first recorded text. Each of the two adjacent question and answer texts includes a question text and an answer text. The question text may consist of a question, and the answer question may consist of the answer to this question.
[0047] In an embodiment of the present application, the target condition may also be a condition related to the speaker's speech content or speech duration. The target speaker may be a speaker whose speech content has more characters than a threshold or whose speech duration exceeds a preset value.
[0048] In the embodiment of the present application, the speech content of the target speaker is compressed to obtain a semantically compressed text of the speech content of the target speaker. That is, the content of the semantically compressed text is less than the speech content of the target speaker.
[0049] Step 220, determining the weight of the semantically compressed text of the target speaker;
[0050] Step 230, based on the semantically compressed text of the target speaker and the weight of the semantically compressed text of the target speaker, the first recorded text is updated to obtain a second recorded text;
[0051] Step 240: Generate meeting minutes based on the second record text.
[0052] The meeting minutes generation method provided in the embodiment of the present application takes into account the semantically compressed text of the target speaker and its weight in the process of generating the meeting minutes, and then updates the first record text according to the semantically compressed text of the target speaker and its weight to obtain the second record text. In this way, it is possible to adjust and highlight the key or hot speech content based on the determined weight of the semantically compressed text, thereby improving the accuracy of the generated meeting minutes. Especially for large meetings or meetings attended by multiple users, it can avoid the problem that the generated meeting minutes have too much content, lack of emphasis, and limited adaptability.
[0053] In an embodiment of the present application, the step 210 of obtaining a semantically compressed text of the speech content of the target speaker in the first recorded text includes: obtaining the speech content and a second prompt word of the target speaker in the first recorded text, wherein the target speaker is the speaker between two adjacent question and answer texts in the first recorded text; inputting the speech content of the target speaker and the second prompt word into a second language model to obtain a semantically compressed text of the speech content of the target speaker; wherein the second prompt word is a prompt word used to extract keywords or summarize the speech content.
[0054] In an embodiment of the present application, determining the weight of the semantically compressed text of the target speaker in step 220 includes: obtaining the similarity between the semantically compressed text of the target speaker and an adjacent question and answer text; and determining the weight of the semantically compressed text of the target speaker based on the similarity.
[0055] In an embodiment of the present application, the target speaker is a speaker between the first question and answer text and the second question and answer text in the first recorded text, and the first question and answer text and the second question and answer text are adjacent question and answer texts of the target speaker. The process of obtaining the similarity between the semantically compressed text of the target speaker and the adjacent question and answer text may include: obtaining a first vector, a second vector and a third vector; wherein the first vector is a vector representation of the semantically compressed text of the target speaker, the second vector is a vector representation of the answer text in the first question and answer text, and the third vector is a vector representation of the question text in the second question and answer text; obtaining a first similarity value between the first vector and the second vector; obtaining a second similarity value between the first vector and the third vector. In an embodiment of the present application, the first question and answer text, the second question and answer text and the semantically compressed text of the target speaker may be arranged in the following order, i.e., the first question and answer text-the semantically compressed text of the target speaker-the second question and answer text. The question and answer text may include a question text and an answer text, wherein the question text may be before the answer text, i.e., the question and answer text is arranged and laid out in the order of question text-answer text.
[0056] In an embodiment of the present application, the process of determining the weight of the semantically compressed text of the target speaker based on the similarity described above may include: if the difference between the first similarity value and the second similarity value is greater than or equal to the standard value, the weight of the semantically compressed text of the target speaker is determined to be a first value; if the difference between the first similarity value and the second similarity value is less than the standard value, the weight of the semantically compressed text of the target speaker is determined to be a second value. The first value may be 1, and the second value may be 0.
[0057] In an embodiment of the present application, the first record text is updated based on the semantically compressed text of the target speaker and the weight of the semantically compressed text of the target speaker in step 230, including: when the weight of the semantically compressed text of the target speaker is a first value, replacing the speech content of the target speaker in the first record text with the semantically compressed text of the target speaker; when the weight of the semantically compressed text of the target speaker is a second value, deleting the speech content of the target speaker in the first record text.
[0058] In an embodiment of the present application, the process of generating meeting minutes based on the second record text described in step 240 may include: obtaining a third prompt word; inputting the second record text and the third prompt word into a third language model to obtain meeting minutes corresponding to the meeting record text; wherein the third prompt word is a prompt word for organizing the meeting minutes.
[0059] Figure 3 : is a flow chart of a method for generating meeting minutes provided in an embodiment of the present application. Figure 3 The meeting minutes generation method provided in the embodiment of the present application may include the following steps:
[0060] Step 310, obtaining a conference record text, wherein the conference record text includes the speech contents of multiple speakers;
[0061] The meeting minutes generation method provided in the embodiment of the present application can be applied to an intelligent office assistant, which can generate meeting minutes of the corresponding meeting based on the meeting record text obtained by the conference system, thereby assisting employees to achieve efficient office work.
[0062] The conference record text can be transcribed and generated by the conference system based on the collected conference audio. The conference audio can be online conference audio or offline conference audio. At the same time, the conference system also provides a voice identity recognition function, and the identity of the speaker of each text content can be marked in the transcribed conference record text.
[0063] Step 320, obtaining an interactive pair text from the conference record text, wherein the interactive pair text includes speech contents of at least two speakers;
[0064] The server of the intelligent office assistant can extract the interactive pair text from the acquired meeting record text based on the acquired meeting record text and the set interactive pair text extraction rules. In the embodiment of the present application, the number of interactive pair texts can be multiple.
[0065] Figure 4-1 This is an example of a partial text snippet of a meeting record. Figure 4-1As shown in the figure, in this example, the conference record text includes multiple speakers, namely speaker 1, speaker 2, and speaker 3. According to the set interaction pair text extraction rules, the extracted interaction pair text is in the dashed box in the figure below. The interaction pair text includes at least two speakers, and the text length of each speaker's speech content is relatively short, or the speech duration of each speaker is relatively short.
[0066] In an embodiment of the present application, the interactive text extraction rule may include an extraction rule for the starting position of the interactive text and an extraction rule for the ending position of the interactive text. The extraction rule for the starting position of the interactive text may be set as needed. For example, the extraction rule for the starting position of the interactive text may include any of the following:
[0067] If the speech text length or speech duration of the speaker before the current speaker is greater than the set value, and the speech text length or speech duration of the current speaker and the speaker after the current speaker is less than the set value, the speech text of the current speaker is used as the starting position for extraction.
[0068] If the speech text length or speech duration of the speaker before the current speaker is greater than the set value, and the speech text length or speech duration of the current speaker is less than the set value, and the speech text length or speech duration of the speaker after the current speaker is greater than the set value, the speech text of the current speaker is used as the starting position for extraction.
[0069] In the embodiment of the present application, the extraction rule of the end position of the interaction text can be set as needed. For example, the extraction rule of the end position of the interaction text can include any of the following:
[0070] If the speech text length or speech duration of the speaker before the current speaker is less than the set value, and the speech text length or speech duration of the speaker after the current speaker is greater than the set value, the speech text corresponding to the current speaker is used as the end position of extraction.
[0071] If the speech text length or speech duration of the speaker before the current speaker is greater than the set value, and the speech text length or speech duration of the current speaker and the speaker after the current speaker is less than the set value, the speech text corresponding to the last speaker whose speech text length or speech duration is less than the set value will be used as the end position of extraction.
[0072] If the speech text length or speech duration of the speaker before the current speaker is greater than the set value, and the speech text length or speech duration of the current speaker is less than the set value, and the speech text length or speech duration of the speaker after the current speaker is greater than the set value, the speech text of the first speaker after the current speaker will be used as the end position of extraction.
[0073] Figure 4-1 and Figure 4-2 They are two examples of meeting record texts. The dotted box part in the figure is the interactive text obtained according to the set interactive text extraction rules.
[0074] Step 330, based on the interactive text, obtaining a question-and-answer text corresponding to the interactive text, wherein the question-and-answer text is used to associate contextual semantics between speech contents of the at least two speakers in the interactive text;
[0075] In an embodiment of the present application, multiple interaction pair texts and the first prompt word can be respectively input into the first language model to obtain the question and answer relationship text of the speaker included in each interaction pair, which is referred to as question and answer text hereinafter.
[0076] The first prompt word is a prompt word used to generate a question and answer relationship text.
[0077] The specific structure of the prompt words is as follows:
[0078]
Prompt words
[0079] #Role:
[0080] You are a text conversion expert who is good at converting natural conversation input text into high-quality question and answer text.
[0081] #Skills:
[0082] -Perform semantic understanding on the input natural dialogue text, determine the question content and answer content included in the dialogue text, and list the question and answer text included in the text.
[0083] -Please focus on brevity and clarity, effectively conveying the main ideas and emotional level while using simple and clear language.
[0084] - Familiar with various professional terms and can express text professionally.
[0085] #Constraint:
[0086] - You must strictly adhere to the accuracy and precision of the information and must not expand the information provided by users.
[0087] -It needs to fully reflect the information of natural conversation text, and highlight the key points without missing any information.
[0088] -Just organize and summarize the information, and make adjustments to the incorrect and unsmooth sentences in the text.
[0089] -Use a formal tone and be sure not to use colloquial language.
[0090] -The key points of the question should be concise and clear, and the answer information should not be over-explained.
[0091] #Workflows:
[0092] 1. Understand the semantics of the input natural dialogue text and determine the question content and answer content included in the dialogue text;
[0093] 2. Arrange the questions and answers for semantic understanding in the following format. When there are multiple question and answer texts, list them in order according to the natural semantic relationship:
[0094] - Question: Question text.
[0095] -Answer: Answer text.
[0096] Specifically, a plurality of interaction pair texts and the first prompt word are input into a first language model (a large model), and a question-answer text included in each interaction pair text is generated by the first language model.
[0097] Step 340, using the question and answer text to replace the interaction text in the conference record text to obtain a first record text;
[0098] The interaction pair texts determined in step 320 are replaced by the generated multiple interaction pair texts, thereby generating a first recorded text. Figure 4-3 As shown, the generated question-and-answer text replaces the original interactive text, and is then sequentially spliced with the non-interactive text in the original meeting record text to form a new meeting record text as the first record text. Figure 4-3 The replacement of one interaction pair text is merely taken as an example. If there are multiple interaction pair texts, and multiple or multiple groups of question-answer texts are obtained based on the interaction pair texts, the interaction pair texts can be replaced by all the question-answer texts.
[0099] Step 350, obtaining a semantically compressed text of the speech content of the target speaker in the first recorded text; the target speaker is a speaker between the first question and answer text and the second question and answer text in the first recorded text, and the first question and answer text and the second question and answer text are adjacent question and answer texts;
[0100] In an embodiment of the present application, the speech content and the second prompt word of the target speaker (including the speaker between adjacent question and answer texts) can be respectively input into the second language model to semantically compress the speech content to obtain the semantically compressed text of the target speaker.
[0101] like Figure 4-4 As shown, if the target speaker is a speaker whose speech content has more than the threshold number of characters or whose speech duration has exceeded the preset value, the speech content corresponding to each target speaker may be the speech content corresponding to speaker 1, speaker 3, and speaker 2. Of course, the target speakers may also include speakers between adjacent question and answer texts. In this case, there are multiple target speakers' speech content between adjacent question and answer texts. At this time, the speech content of multiple target speakers may be extracted as input together with the second prompt word and input into the second language model. The second language model performs semantic compression on the speech content, thereby obtaining the semantically compressed text corresponding to each target speaker.
[0102] The second prompt word may be a prompt word used for keyword extraction and speech summary.
[0103] Step 360, determining the weight of the semantically compressed text of the target speaker;
[0104] In an embodiment of the present application, the similarity between the semantically compressed text of each target speaker and the adjacent question and answer text can be calculated, and the weight of the semantically compressed text of each target speaker (i.e., the sentiment weight, which can be 0 or 1) can be determined based on the similarity.
[0105] The specific process can be as follows:
[0106] The embedding model is used to obtain vector representations of semantically compressed text and the question and answer texts in each question and answer text. For semantically compressed text, question text, and answer text, their corresponding vector representations can be obtained through the embedding model respectively for subsequent similarity calculations.
[0107] Calculate the similarity between the vector representation of each semantically compressed text and its adjacent question text vector representation and answer text vector representation; the similarity calculation can be calculated using the cosine similarity calculation formula. Figure 4-4 , taking the target speaker as an example, question 1 and answer 1 constitute the first question and answer text, question 2 and answer 2 constitute the second question and answer text, and the target speaker is speaker 3. The first similarity value is obtained by calculating the similarity between the vector representation of the semantically compressed text of the target speaker (i.e., the first vector) and the vector representation of the answer text (i.e., answer 1) in the question and answer text before its text position (i.e., the second vector); and the second similarity value is obtained by calculating the similarity between the vector representation of the question text (i.e., question 2) in the question and answer text after its text position (i.e., the third vector).
[0108] If the difference between the first similarity value and the second similarity value is less than the set standard value, that is, close to zero, the sentiment weight of the semantically compressed text of the target speaker is 0, otherwise the sentiment weight of the semantically compressed text of the target speaker is 1. If the adjacent question and answer texts include semantically compressed texts of at least two target speakers, the first similarity value is first calculated based on the vector representation of the semantically compressed text at the first text position (for example, the frontmost position), and then the second similarity value is calculated based on the vector representation of the semantically compressed text at the second text position (for example, the backmost position). If the difference between the first similarity value and the second similarity value is less than the set standard value, that is, close to zero, the sentiment weight of the semantically compressed texts of the at least two speakers is 0, otherwise the sentiment weight of the semantically compressed text of the speaker is 1. Furthermore, the similarity of the vector representations of the semantically compressed texts of at least two target speakers can be calculated to obtain a third similarity value. If the third similarity value is greater than a set threshold, the sentiment weight of the semantically compressed text that meets the specified conditions (for example, the semantically compressed text with the most text content) is 1, and the sentiment weight of the semantically compressed text that does not meet the specified conditions is 0.
[0109] Step 370, based on the semantically compressed text of the target speaker and the weight of the semantically compressed text of the target speaker, the first recorded text is updated to obtain a second recorded text;
[0110] In an embodiment of the present application, when the sentiment weight of the semantically compressed text is 1, the speech content corresponding to the identity of the speaker of the semantically compressed text in the first recorded text is replaced by the semantically compressed text.
[0111] When the sentiment weight of the semantically compressed text is 0, the speech content of the speaker identity corresponding to the semantically compressed text in the first recorded text is deleted.
[0112] Combination Figure 4-4 For example, the sentiment weight of the semantically compressed text corresponding to speaker 3 is 0. At this time, the speech content (conference record text fragment) corresponding to speaker 3 is deleted in the first record text. The sentiment weight of the semantically compressed text corresponding to speaker 2 is 1. At this time, in the first record text, the speech content of speaker 2 is updated using the semantically compressed text of speaker 2.
[0113] Step 380: Generate meeting minutes based on the second record text.
[0114] Finally, the updated second record text and the third prompt word are input into the third language model to generate the meeting minutes corresponding to the meeting record text.
[0115] The third prompt word is a prompt word used to organize the meeting minutes. An example of the third prompt word is as follows:
[0116]
Prompt words
[0117] #Role:
[0118] You are a professional CEO secretary who is good at organizing the input meeting minutes into high-quality meeting minutes, ensuring that the meeting objectives and action plans are clear.
[0119] #Skills:
[0120] - Clear and accurate expression skills, especially good at capturing the key details of the text and making precise statements
[0121] -Strong text summarization ability, able to summarize key information in the text
[0122] - Familiar with various professional terms and can express text professionally
[0123] #Constraint:
[0124] - You must strictly adhere to the accuracy and precision of the information and must not expand the information provided by users.
[0125] -It needs to fully reflect the information of natural conversation text, and highlight the key points without missing any information.
[0126] -Just organize and summarize the information, and make adjustments to the incorrect and unsmooth sentences in the text.
[0127] -Use a formal tone and be sure not to use colloquial language.
[0128] -The key points of the question should be concise and clear, and the answer information should not be over-explained.
[0129] #Workflows:
[0130] 1. Input: Meeting minutes text
[0131] 2. Arrange the meeting minutes in the following format:
[0132] - Meeting Topic: The title and purpose of the meeting.
[0133] -Meeting Date and Time: The specific date and time of the meeting.
[0134] - Participants: List all the people attending the meeting.
[0135] -Meeting Minute-taker: Indicate the person who records these contents.
[0136] -Meeting Agenda: Lists all topics and discussion points for the meeting.
[0137] -Main discussion: Detailed description of the discussion content of each topic, mainly including the questions raised, proposals, opinions, etc.
[0138] -Decisions and Action Plan: List all decisions from the meeting, along with the actions planned to be taken, along with the responsible parties and planned completion dates.
[0139] -Next Steps: List next steps or issues that need to be discussed in future meetings.
[0140] 3. Output: Output the meeting minutes with clear structure and complete description.
[0141] It should be understood that in various embodiments of the present application, the same or corresponding steps can be referenced to each other. Figure 3 The explanation of each step in can also be applied to Figure 1 and Figure 2 , or corresponding steps in .
[0142] The meeting minutes generation method provided by the embodiment of the present application obtains a meeting record text including the speech content of multiple speakers; extracts multiple interaction pair texts from the meeting record text; each interaction pair text includes at least two speaker identities; inputs the multiple interaction pair texts and the first prompt word into the model respectively, and obtains the question and answer text of the speakers included in each interaction pair; replaces the multiple interaction pair texts with the generated question and answer text to obtain a first record text; inputs the speech content and the second prompt word corresponding to each speaker between adjacent question and answer texts into the model respectively to semantically compress the speech content to obtain the semantically compressed text of each speaker; calculates the similarity between the semantically compressed text of each speaker and the adjacent question and answer text, and determines the sentiment weight (0 or 1) of the semantically compressed text of each speaker based on the similarity; based on the sentiment weight of each semantically compressed text, updates the first record text through the semantically compressed text to obtain the second record text, and inputs the second record text and the third prompt word into the model to generate meeting minutes corresponding to the meeting record text.
[0143] In the overall implementation of the solution, the focus or hot spots of the speakers during the meeting are interactively mined, so that the focus or hot spots can be adjusted and highlighted based on the determined emotional weight, thereby improving the accuracy of the generated meeting minutes. Especially for large meetings or meetings attended by multiple users, it can avoid the problem of too much content, lack of focus, and limited adaptability in the generated meeting minutes.
[0144] Compared with the technology of compressing the speech content of non-leaders, the question and answer text generated by the interactive text in the embodiment of the present application can associate the contextual semantics of all speakers, and guide the compression of the speaker's speech content through the question and answer text generated by the interactive text, which can avoid the semantic discontinuity and incompleteness of the meeting records compressed by the large model, thereby improving the accuracy of the content of the meeting minutes.
[0145] In addition, compared with the technology of analyzing and summarizing the meeting minutes text based on pre-defined rules and templates using natural language processing technology, the embodiment of the present application introduces the compression and generation capabilities of the model, and analyzes the potential interaction hotspots in the entire meeting minutes based on the semantic mining of the interactions in the meeting minutes text. The content of the meeting minutes is compressed and adjusted through the interaction hotspots, so that the generated meeting minutes can highlight the key points, be accurate in content, and be highly adaptable.
[0146] Figure 5 This is a structural block diagram of a conference minutes generation device provided in an embodiment of the present application. Figure 5 , the meeting minutes generation device provided in the embodiment of the present application may include:
[0147] The acquisition module 510 is used to acquire a conference record text, wherein the conference record text includes the speech contents of multiple speakers; acquire an interactive pair text from the conference record text, wherein the interactive pair text includes the speech contents of at least two speakers; based on the interactive pair text, acquire a question-and-answer text corresponding to the interactive pair text, wherein the question-and-answer text is used to associate the contextual semantics between the speech contents of the at least two speakers in the interactive pair text;
[0148] A replacement module 520, configured to replace the interaction pair text in the conference record text with the question and answer text to obtain a first record text;
[0149] The generating module 530 is used to generate meeting minutes based on the first record text.
[0150] The conference minutes generation device provided in the embodiment of the present application obtains a conference record text, wherein the conference record text includes the speech contents of multiple speakers; obtains an interactive pair text from the conference record text, wherein the interactive pair text includes the speech contents of at least two speakers; based on the interactive pair text, obtains a question and answer text corresponding to the interactive pair text, wherein the question and answer text is used to associate the contextual semantics between the speech contents of the at least two speakers in the interactive pair text; uses the question and answer text to replace the interactive pair text in the conference record text to obtain a first record text; and generates conference minutes based on the first record text. In this way, in the process of generating conference minutes, by obtaining the question and answer text corresponding to the interactive pair text, wherein the question and answer text is used to associate the contextual semantics between the speech contents of the at least two speakers in the interactive pair text, and using the question and answer text to replace the interactive pair text in the conference record text, the relevance of the contextual semantics of the speech contents is ensured, thereby solving the problem of semantic discontinuity in the conference minutes obtained by the related art to a certain extent.
[0151] In the process of obtaining the question-answer text corresponding to the interaction text based on the interaction text, the acquisition module 510 is specifically used to: input the interaction text and the first prompt word into a first language model to obtain the question-answer text corresponding to the interaction text; wherein the first prompt word is a prompt word used to generate the question-answer text.
[0152] In the process of generating meeting minutes based on the first record text, the generation module 530 is specifically used to: obtain the semantically compressed text of the target speaker's speech content in the first record text; determine the weight of the semantically compressed text of the target speaker; based on the semantically compressed text of the target speaker and the weight of the semantically compressed text of the target speaker, update the first record text to obtain a second record text; and generate meeting minutes based on the second record text.
[0153] In the process of obtaining the semantically compressed text of the target speaker's speech content in the first record text, the generation module 530 is specifically used to: obtain the target speaker's speech content and the second prompt word in the first record text, the target speaker is the speaker between two adjacent question and answer texts in the first record text; input the target speaker's speech content and the second prompt word into the second language model to obtain the semantically compressed text of the target speaker's speech content; wherein the second prompt word is a prompt word used to extract keywords or summarize the speech content.
[0154] In the process of determining the weight of the semantically compressed text of the target speaker, the generation module 530 is specifically used to: obtain the similarity between the semantically compressed text of the target speaker and the adjacent question and answer text; and determine the weight of the semantically compressed text of the target speaker based on the similarity.
[0155] In an embodiment of the present application, the target speaker is a speaker between the first question and answer text and the second question and answer text in the first recorded text, and the first question and answer text and the second question and answer text are adjacent question and answer texts. In the process of obtaining the similarity between the semantically compressed text of the target speaker and the adjacent question and answer text, the generation module 530 is specifically used to: obtain a first vector, a second vector and a third vector; wherein the first vector is a vector representation of the semantically compressed text of the target speaker, the second vector is a vector representation of the answer text in the first question and answer text, and the third vector is a vector representation of the question text in the second question and answer text; obtain a first similarity value between the first vector and the second vector; obtain a second similarity value between the first vector and the third vector.
[0156] In the process of determining the weight of the semantically compressed text of the target speaker based on the similarity, the generation module 530 is specifically used to: if the difference between the first similarity value and the second similarity value is greater than or equal to the standard value, determine the weight of the semantically compressed text of the target speaker as a first value; if the difference between the first similarity value and the second similarity value is less than the standard value, determine the weight of the semantically compressed text of the target speaker as a second value.
[0157] In the process of updating the first record text based on the semantically compressed text of the target speaker and the weight of the semantically compressed text of the target speaker, the generation module 530 is specifically used to: when the weight of the semantically compressed text of the target speaker is a first value, replace the speech content of the target speaker in the first record text with the semantically compressed text of the target speaker; when the weight of the semantically compressed text of the target speaker is a second value, delete the speech content of the target speaker in the first record text.
[0158] like Figure 6As shown, the embodiment of the present application also provides an electronic device 600, which can be various types of computers, etc. The electronic device 600 includes: a processor 610 and a memory 620, and the memory 620 stores a program or instruction, and the program or instruction is executed by the processor 610 to implement the steps of any of the methods described above. For example, when the program is executed by the processor 610, the following process is implemented: obtaining a conference record text, the conference record text contains the speech content of multiple speakers; obtaining an interactive pair text from the conference record text, the interactive pair text contains the speech content of at least two speakers; based on the interactive pair text, obtaining a question and answer text corresponding to the interactive pair text, the question and answer text is used to associate the contextual semantics between the speech content of at least two speakers in the interactive pair text; using the question and answer text to replace the interactive pair text in the conference record text to obtain a first record text; based on the first record text, generating a meeting minutes. In this way, in the process of generating meeting minutes, by obtaining the question and answer text corresponding to the interactive text, the question and answer text is used to associate the contextual semantics between the speech contents of at least two speakers in the interactive text, and the question and answer text is used to replace the interactive text in the meeting record text, thereby ensuring the relevance of the contextual semantics of the speech content, and thus to a certain extent solving the problem of semantic discontinuity in the meeting minutes obtained by related technologies.
[0159] An embodiment of the present application also provides a readable storage medium, on which a program or instruction is stored. When the program or instruction is executed by a processor, the steps of generating meeting minutes provided in any of the above embodiments are implemented, and the same technical effect can be achieved. To avoid repetition, it will not be repeated here.
[0160] The processor is the processor in the electronic device described in the above embodiment. The readable storage medium includes a computer readable storage medium, such as a computer read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.
[0161] An embodiment of the present application further provides a chip, which includes a processor and a communication interface, wherein the communication interface is coupled to the processor, and the processor is used to run programs or instructions to implement the various processes of the above-mentioned method embodiment, and can achieve the same technical effect. To avoid repetition, it will not be repeated here.
[0162] An embodiment of the present application provides a computer program product, which is stored in a storage medium. The program product is executed by at least one processor to implement the various processes of the above-mentioned method embodiment and can achieve the same technical effect. To avoid repetition, it will not be repeated here.
[0163] It should be noted that, in this article, the terms "comprise", "include" or any other variant thereof are intended to cover non-exclusive inclusion, so that the process, method, article or device including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise one..." do not exclude the presence of other identical elements in the process, method, article or device including the element. In addition, it should be pointed out that the scope of the methods and devices in the embodiments of the present application is not limited to performing functions in the order shown or discussed, and may also include performing functions in a substantially simultaneous manner or in reverse order according to the functions involved, for example, the described method may be performed in an order different from that described, and various steps may also be added, omitted, or combined. In addition, the features described with reference to certain examples may be combined in other examples.
[0164] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a computer software product, which is stored in a storage medium (such as ROM / RAM, a disk, or an optical disk), and includes a number of instructions for a terminal (which can be a mobile phone, a computer, a server, or a network device, etc.) to execute the methods described in each embodiment of the present application.
[0165] The embodiments of the present application are described above in conjunction with the accompanying drawings, but the present application is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present application, ordinary technicians in this field can also make many forms without departing from the purpose of the present application and the scope of protection of the claims, all of which are within the protection of the present application.
Claims
1. A method for generating meeting minutes, characterized in that: include: Acquire a conference record text, wherein the conference record text includes speech contents of multiple speakers; Acquire an interactive pair text from the conference record text, wherein the interactive pair text includes speech content of at least two speakers; Based on the interactive text, obtaining a question-and-answer text corresponding to the interactive text, wherein the question-and-answer text is used to associate contextual semantics between speech contents of at least two speakers in the interactive text; Using the question-and-answer text to replace the interaction text in the conference record text to obtain a first record text; Generate meeting minutes based on the first record text.
2. The method according to claim 1, characterized in that The acquiring, based on the interaction pair text, a question and answer text corresponding to the interaction pair text includes: Inputting the interaction pair text and the first prompt word into a first language model to obtain a question-answer text corresponding to the interaction pair text; The first prompt word is a prompt word used to generate a question and answer text.
3. The method according to claim 1, characterized in that Generating the meeting minutes based on the first recorded text includes: Obtaining a semantically compressed text of the target speaker's speech content in the first recorded text; determining a weight of the semantically compressed text of the target speaker; Based on the semantically compressed text of the target speaker and the weight of the semantically compressed text of the target speaker, the first recorded text is updated to obtain a second recorded text; Based on the second record text, generate meeting minutes.
4. The method according to claim 3, characterized in that The step of obtaining a semantically compressed text of the speech content of the target speaker in the first recorded text includes: Acquire the speech content and the second prompt word of the target speaker in the first recorded text, wherein the target speaker is the speaker between two adjacent question-and-answer texts in the first recorded text; Inputting the speech content of the target speaker and the second prompt word into a second language model to obtain a semantically compressed text of the speech content of the target speaker; The second prompt word is a prompt word used to extract keywords or summarize speech content.
5. The method according to claim 3, characterized in that: The determining the weight of the semantically compressed text of the target speaker comprises: Obtaining the similarity between the semantically compressed text of the target speaker and the adjacent question and answer text; Based on the similarity, a weight of the semantically compressed text of the target speaker is determined.
6. The method according to claim 5, characterized in that The target speaker is a speaker between a first question and answer text and a second question and answer text in the first recorded text, and the first question and answer text and the second question and answer text are adjacent question and answer texts; The obtaining of the similarity between the semantically compressed text of the target speaker and the adjacent question and answer text comprises: Obtain a first vector, a second vector, and a third vector; wherein the first vector is a vector representation of the semantically compressed text of the target speaker, the second vector is a vector representation of the answer text in the first question and answer text, and the third vector is a vector representation of the question text in the second question and answer text; Obtaining a first similarity value between the first vector and the second vector; A second similarity value between the first vector and the third vector is obtained.
7. The method according to claim 5, characterized in that The step of determining the weight of the semantically compressed text of the target speaker based on the similarity includes: If the difference between the first similarity value and the second similarity value is greater than or equal to a standard value, determining the weight of the semantically compressed text of the target speaker to be a first value; If the difference between the first similarity value and the second similarity value is smaller than the standard value, the weight of the semantically compressed text of the target speaker is determined to be a second value.
8. The method according to claim 3, characterized in that The updating of the first recorded text based on the semantically compressed text of the target speaker and the weight of the semantically compressed text of the target speaker includes: When the weight of the semantically compressed text of the target speaker is a first value, replacing the speech content of the target speaker in the first recorded text with the semantically compressed text of the target speaker; When the weight of the semantically compressed text of the target speaker is a second value, the speech content of the target speaker in the first recorded text is deleted.
9. An electronic device, characterized in that: The method comprises a processor and a memory, wherein the memory stores a program or instruction running on the processor, and when the program or instruction is executed by the processor, the steps of the method according to any one of claims 1 to 8 are implemented.
10. A readable storage medium, characterized in that: The readable storage medium stores a program or instruction, and when the program or instruction is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.
Citation Information
Cited By
Video conference data processing method and device, equipment, medium and product
CN119697339A
A method, apparatus, device, medium and product for processing video conference data
CN119697339B
