Conference summary generation method applied to official document format

Abstract

The invention discloses a conference summary generation method applied to an official document format, relates to the technical field of intelligent conferences, and solves the technical problems of incomplete recording, low accuracy and poor format normalization in the case of generating the conference summary in the official document format by the scheme in the prior art. The method comprises the following steps: acquiring feature information of a plurality of participants in a conference; on the basis of the identity information and the head position information, the attending concentration degree of each participant is obtained; performing text correction on the original conference text to obtain a conference text; integrating the conference text and the conference key information into a conference summary; and filling the conference key information and the attending concentration degree of each participant into an official document template to generate a conference summary. According to the invention, based on the attention duration and the speaking duration, the conference participation concentration score of each participant is calculated through a formula; and whether the head posture faces the conference concern or not can reflect the concentration degree of the participants on the conference, so that the generated conference summary can more comprehensively record the actual situation of the conference.

Description

Technical Field

[0001] The invention belongs to the technical field of intelligent conferences, and in particular is a method for generating conference minutes in an official document format. Background Art

[0002] In the operation of modern enterprises and organizations, meeting minutes are an important form of documentation that records the main content, decisions and follow-up actions of the meeting. However, traditional methods of generating meeting minutes often rely on manual recording, resulting in inefficiency, incomplete content and inconsistent quality.

[0003] The existing technology collects the voice information of the participants, converts the voice information into conference text based on voice recognition technology, and processes the conference text to obtain the minutes. However, the concentration of each senator in the meeting is different, and the degree of understanding of the content of the meeting is also different. Therefore, it is impossible to formulate reasonable work content for the senators in the subsequent meeting content; the minutes generated by the existing technology only record the speech content in the meeting, and it is difficult to obtain the concentration of the participants, resulting in the generated minutes not being able to fully reflect the actual meeting situation; the existing technology solution has high efficiency and accuracy in generating minutes in the scene of a single person speaking. In the meeting scene where multiple people speak at the same time, the noisy voice will affect the voice recognition, so the converted conference text has low content semantic accuracy and chaotic word order, and manual correction is still required later, resulting in low efficiency and accuracy in generating minutes; in addition, the existing technology solution for generating minutes in the format of official documents mostly has the problem of poor format standardization, for example, the format of the generated minutes is not uniform, and the wording in the heading and the signature is not standardized.

[0004] The present invention provides a method for generating meeting minutes in an official document format to solve the above technical problems. Summary of the invention

[0005] The present invention aims to solve at least one of the technical problems existing in the prior art; to this end, the present invention proposes a method for generating meeting minutes in an official document format, which is used to solve the problem that the prior art is difficult to obtain the concentration of participants in the meeting, resulting in the generated meeting minutes failing to fully reflect the actual meeting situation; and, in the prior art, in a meeting scenario where multiple people speak at the same time, the noisy voice will affect the voice recognition, so that the converted meeting text has low content semantic accuracy and chaotic word order, and manual correction is still required later, resulting in low efficiency and accuracy in generating meeting minutes; in addition, the prior art solution also has the technical problem that most of the meeting minutes generated in the official document format have poor format standardization.

[0006] To achieve the above object, a first aspect of the present invention provides a method for generating meeting minutes in an official document format, comprising:

[0007] S1: Obtain feature information of multiple participants in the conference, and number the multiple participants respectively; wherein the feature information includes voice information, face information, and head position information;

[0008] S2: Convert the speech information into text based on the speech recognition algorithm to obtain the original meeting text;

[0009] S3: Obtaining the identity information of each participant based on the facial information;

[0010] S4: obtaining the concentration of each participant based on the identity information and the head position information; wherein the concentration includes high concentration and low concentration;

[0011] S5: Perform text correction on the original conference text to obtain the conference text; wherein the text correction includes correcting semantics, deleting typos, and adjusting word order;

[0012] S6: Extract key information of the meeting from the meeting text based on the large language model; the key information of the meeting includes the name of the meeting topic, the meeting resolution, and the identity information of the host and several speakers;

[0013] S7: Fill in the key information of the meeting and the concentration of each participant into the official document template to generate the meeting minutes.

[0014] Preferably, the step of numbering the multiple participants respectively includes:

[0015] Get the seating order of each participant, and label each participant as i in clockwise order starting from the conference host; where the conference host belongs to the participants, i=1, 2, ..., n, where n is the total number of participants.

[0016] Preferably, the acquiring the concentration of each participant based on the identity information and the head position information includes:

[0017] A1: Extract the identity information and head position information of participant i;

[0018] A2: Obtain the concentration score of each participant based on identity information and head position information;

[0019] A3: Determine whether the attendee concentration score is greater than a preset score threshold; if yes, mark the attendee concentration of attendee i as high concentration; if no, mark the attendee concentration of attendee i as low concentration.

[0020] Preferably, the acquiring of the meeting concentration score of each participant based on the identity information and the head position information includes:

[0021] B1: Obtain the head posture of participant i from the head position information;

[0022] B2: Get the speaker's speaking time;

[0023] B3: Obtain the duration of the participant i's head posture facing the conference focus point and mark it as the focus duration; the conference focus point includes the current speaker, the conference screen and the conference materials;

[0024] B4: Calculate the participant i's concentration score ZZDi through the formula ZZDi=(a×GZSi) / FYS; where FYS is the speaker's speaking time, GZSi is the participant i's attention time, and a is an influence coefficient greater than 0.

[0025] It should be noted that when a participant is a speaker, the participant's concentration score is not calculated.

[0026] The present invention obtains the length of time that the participant's head posture is directed toward the meeting focus point, that is, the attention time and the speaker's speaking time; and obtains the meeting concentration score of each participant through a formula based on the attention time and the speaking time; because whether the head posture is directed toward the meeting focus point can reflect the participant's concentration level on the meeting, the calculated meeting concentration score can facilitate the subsequent acquisition of the corresponding meeting concentration based on the meeting concentration score, and the meeting concentration is used to generate meeting minutes, which is conducive to the generated meeting minutes being able to more comprehensively record the actual situation of the meeting.

[0027] Preferably, the text correction of the original conference text includes:

[0028] C1: Extract the original conference text and obtain the modal particle database;

[0029] C2: Determine whether the modal particles in the original conference text belong to the modal particle database; if yes, remove the corresponding modal particles; if no, jump to C3;

[0030] C3: Mark the original conference text without modal particles as the first text;

[0031] C4: Correct the typos in the first text to obtain the second text;

[0032] C5: Adjust the word order of the second text to the normal word order to obtain the conference text.

[0033] It should be noted that the modal particle database is a collection of modal particles that are not allowed to appear in common official documents, such as: "ah", "yeah", "ne", "bar", etc.

[0034] Preferably, adjusting the word order of the second text to a normal word order includes:

[0035] D1: Extract the second text and the voice information of the participants;

[0036] D2: dividing the second text into a number of continuous text segments based on the voice information of the participants;

[0037] D3: Mark the text segment as j; where j = 1, 2, ..., m, and m is the total number of text segments;

[0038] D4: Calculate the semantic relevance between text segment j and the rest of the text segments;

[0039] D5: Determine whether the semantic relevance is greater than a preset relevance threshold;

[0040] If yes, the corresponding text segment is added to the queue of text segments to be adjusted;

[0041] If not, the word order of the corresponding text segment remains unchanged;

[0042] D6: Arrange the text segments in the queue of text segments to be adjusted in descending order according to their corresponding semantic relevance;

[0043] D7: Extract the text segment with the highest semantic relevance in the queue of text segments to be adjusted, and adjust it to the position of the word order of text segment j plus one.

[0044] The present invention calculates the semantic relevance between each text segment and the remaining text segments, and adjusts the text segment with the highest semantic relevance to the position of the word order of the corresponding text segment plus one; it solves the problem of chaotic word order in the prior art of conference text generated by using speech recognition when multiple people speak at the same time, makes the logic and orderliness between the context of the word order in the generated conference text clearer, and is conducive to improving the accuracy of generating meeting minutes.

[0045] Preferably, the dividing the second text into a plurality of continuous text segments based on the voice information of the participants includes:

[0046] E1: Obtain some continuous voice information of all participants;

[0047] E2: Calculate the pause intervals between adjacent speech messages in sequence;

[0048] E3: Determine whether the pause interval is greater than a preset pause threshold; if yes, divide the text information corresponding to the adjacent voice information into two text segments; if no, merge the text information corresponding to the adjacent voice information into one text segment.

[0049] Preferably, extracting key meeting information from the meeting text based on the large language model includes:

[0050] G1: Preprocess the conference text;

[0051] G2: Segment the conference text based on word segmentation software to obtain several words;

[0052] G3: Tag the parts of speech of several words based on the part-of-speech tagging tool; the parts of speech include nouns, adjectives and verbs;

[0053] G4: Extract the conference text based on the feature extraction algorithm to obtain high-frequency words and short sentences;

[0054] G5: Integrate high-frequency words and short sentences into key information for the meeting.

[0055] Preferably, the step of filling the key meeting information and the concentration of each participant into the document template includes:

[0056] T1: Extracting official document templates, key meeting information, and the concentration of each participant;

[0057] T2: Extract the mapping relationship between the key information of the meeting, the concentration of each participant and the corresponding unique placeholder from the database;

[0058] T3: Replace each unique placeholder in the official document template with the corresponding key meeting information and the concentration of each participant to generate the meeting minutes.

[0059] The present invention generates meeting minutes by acquiring the mapping relationship between key meeting information, the degree of concentration of each participant and the corresponding unique placeholder, and replacing each unique placeholder in the official document template with the corresponding key meeting information and the degree of concentration of each participant; since there is a one-to-one correspondence between the unique placeholder and the meeting minutes and the degree of concentration of each participant, the meeting content can be accurately filled into the corresponding official document template, which is conducive to improving the standardization of generating meeting minutes in official document format.

[0060] Preferably, the official document template is a text in an official document format, and different positions in the text where information is to be filled in are filled with corresponding unique placeholders.

[0061] Compared with the prior art, the present invention has the following beneficial effects:

[0062] 1. The present invention obtains the length of time that the participant's head posture is directed toward the meeting focus point, that is, the attention time and the speaker's speaking time; based on the attention time and the speaking time, a formula is used to calculate the participant's attendance concentration score; whether the head posture is directed toward the meeting focus point can reflect the participant's degree of concentration on the meeting, so that the calculated attendance concentration score can facilitate the subsequent acquisition of the corresponding attendance concentration according to the attendance concentration score, and the attendance concentration is used to generate meeting minutes, which is conducive to the generated meeting minutes being able to more comprehensively record the actual situation of the meeting.

[0063] 2. The present invention calculates the semantic relevance between each text segment and the remaining text segments, and adjusts the text segment with the highest semantic relevance to the position of the word order of the corresponding text segment plus one; it solves the problem of chaotic word order in the prior art of conference text generated by using speech recognition when multiple people speak at the same time, and makes the logic and orderliness between the context of the word order in the generated conference text clearer, which is conducive to improving the accuracy of generating meeting minutes.

[0064] 3. The present invention generates meeting minutes by obtaining the mapping relationship between the meeting minutes, the meeting concentration score of each participant and the corresponding unique placeholder, and replacing each unique placeholder in the official document template with the corresponding meeting minutes and the meeting concentration score of each participant; since there is a one-to-one correspondence between the unique placeholder and the meeting minutes and the meeting concentration score of each participant, the meeting content can be accurately filled into the corresponding official document template, which is conducive to improving the standardization of generating meeting minutes in official document format. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0066] Figure 1 It is an overall flow chart of the method for generating meeting minutes in the official document format in the present invention;

[0067] Figure 2 A flowchart of performing text correction on the original conference text to obtain the conference text in the present invention;

[0068] Figure 3 The present invention is a flowchart for generating meeting minutes based on key meeting information and meeting concentration. DETAILED DESCRIPTION

[0069] The technical solution of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0070] See also Figure 1-Figure 3 The first aspect of the present invention provides a method for generating meeting minutes in an official document format, comprising:

[0071] S1: Obtain feature information of multiple participants in the conference, and number the multiple participants respectively; wherein the feature information includes voice information, face information, and head position information;

[0072] S2: Convert the speech information into text based on the speech recognition algorithm to obtain the original meeting text;

[0073] S3: Obtaining the identity information of each participant based on the facial information;

[0074] S4: obtaining the concentration of each participant based on the identity information and the head position information; wherein the concentration includes high concentration and low concentration;

[0075] S5: Perform text correction on the original conference text to obtain the conference text; wherein the text correction includes correcting semantics, deleting typos, and adjusting word order;

[0076] S6: Extract key information of the meeting from the meeting text based on the large language model; the key information of the meeting includes the name of the meeting topic, the meeting resolution, and the identity information of the host and several speakers;

[0077] S7: Fill in the key information of the meeting and the concentration of each participant into the official document template to generate the meeting minutes.

[0078] It should be noted that the large language model in this embodiment adopts iFlytek's Spark large model, where the Spark large model is an NLP model that can perform complex natural language processing.

[0079] In this embodiment, multiple participants are numbered respectively, including:

[0080] Get the seating order of each participant, and label each participant as i in clockwise order starting from the conference host; where the conference host belongs to the participants, i=1, 2, ..., n, where n is the total number of participants.

[0081] For example, assuming that there are 4 participants in total, the conference host is marked as 1, and the remaining participants are marked as participant 2, participant 3, and participant 4 in a clockwise order starting from the conference host.

[0082] In this embodiment, the concentration of each participant is obtained based on the identity information and the head position information, including:

[0083] A1: Extract the identity information and head position information of participant i;

[0084] A2: Obtain the concentration score of each participant based on identity information and head position information;

[0085] A3: Determine whether the attendee concentration score is greater than a preset score threshold; if yes, mark the attendee concentration of attendee i as high concentration; if no, mark the attendee concentration of attendee i as low concentration.

[0086] Exemplarily, the meeting concentration score of participant 3 is set to 0.84, and the preset score threshold is 0.8; since the meeting concentration score of participant 3 is greater than the preset score threshold, the meeting concentration of participant 3 is marked as high concentration.

[0087] In this embodiment, the concentration score of each participant is obtained based on the identity information and the head position information, including:

[0088] B1: Obtain the head posture of participant i from the head position information;

[0089] B2: Get the speaker's speaking time;

[0090] B3: Obtain the duration of the participant i's head posture facing the conference focus point and mark it as the focus duration; the conference focus point includes the current speaker, the conference screen and the conference materials;

[0091] B4: Calculate the participant i's concentration score ZZDi through the formula ZZDi=(a×GZSi) / FYS; where FYS is the speaker's speaking time, GZSi is the participant i's attention time, and a is an influence coefficient greater than 0.

[0092] It should be noted that when a participant is a speaker, the participant's concentration score is not calculated.

[0093] Exemplarily, the influence coefficient a is set to 1.2; the attention time of participant 3 GZS3 = 30 min, the speaker is participant 2, and the speaker's speaking time FYS = 43 min; the participant 3's concentration score ZZD3≈0.84 is calculated through the formula.

[0094] The present invention obtains the length of time that the participant's head posture is directed toward the meeting focus point, that is, the attention time and the speaker's speaking time; and obtains the meeting concentration score of each participant through a formula based on the attention time and the speaking time; because whether the head posture is directed toward the meeting focus point can reflect the participant's concentration level on the meeting, the calculated meeting concentration score can facilitate the subsequent acquisition of the corresponding meeting concentration based on the meeting concentration score, and the meeting concentration is used to generate meeting minutes, which is conducive to the generated meeting minutes being able to more comprehensively record the actual situation of the meeting.

[0095] In this embodiment, text correction is performed on the original conference text to obtain the conference text, including:

[0096] C1: Extract the original conference text and obtain the modal particle database;

[0097] C2: Determine whether the modal particles in the original conference text belong to the modal particle database; if yes, remove the corresponding modal particles; if no, jump to C3;

[0098] C3: Mark the original conference text without modal particles as the first text;

[0099] C4: Correct the typos in the first text to obtain the second text;

[0100] C5: Adjust the word order of the second text to the normal word order to obtain the conference text.

[0101] It should be noted that the modal particle database is a collection of modal particles that are not allowed to appear in common official documents, such as: "ah", "yeah", "ne", "bar", etc.

[0102] In this embodiment, adjusting the word order of the second text to a normal word order includes:

[0103] D1: Extract the second text and the voice information of the participants;

[0104] D2: dividing the second text into a number of continuous text segments based on the voice information of the participants;

[0105] D3: Mark the text segment as j; where j = 1, 2, ..., m, and m is the total number of text segments;

[0106] D4: Calculate the semantic relevance between text segment j and the rest of the text segments;

[0107] D5: Determine whether the semantic relevance is greater than a preset relevance threshold;

[0108] If yes, the corresponding text segment is added to the queue of text segments to be adjusted;

[0109] If not, the word order of the corresponding text segment remains unchanged;

[0110] D6: Arrange the text segments in the queue of text segments to be adjusted in descending order according to their corresponding semantic relevance;

[0111] D7: Extract the text segment with the highest semantic relevance in the queue of text segments to be adjusted, and adjust it to the position of the word order of text segment j plus one.

[0112] The present invention calculates the semantic relevance between each text segment and the remaining text segments, and adjusts the text segment with the highest semantic relevance to the position of the word order of the corresponding text segment plus one; it solves the problem of chaotic word order in the prior art of conference text generated by using speech recognition when multiple people speak at the same time, makes the logic and orderliness between the context of the word order in the generated conference text clearer, and is conducive to improving the accuracy of generating meeting minutes.

[0113] In this embodiment, the second text is divided into a number of continuous text segments based on the voice information of the participants, including:

[0114] E1: Obtain some continuous voice information of all participants;

[0115] E2: Calculate the pause intervals between adjacent speech messages in sequence;

[0116] E3: Determine whether the pause interval is greater than a preset pause threshold; if yes, divide the text information corresponding to the adjacent voice information into two text segments; if no, merge the text information corresponding to the adjacent voice information into one text segment.

[0117] Exemplarily, several continuous voice messages of all participants are obtained, assuming that the pause interval between the 4th voice message and the 5th voice message is 5s, and the pause threshold is set to 2s; since the pause interval is longer than the preset pause threshold, the text information corresponding to the 4th voice message and the 5th voice message is divided into two text segments.

[0118] In this embodiment, key meeting information is extracted from the meeting text based on the large language model, including:

[0119] G1: Preprocess the conference text;

[0120] G2: Segment the conference text based on word segmentation software to obtain several words;

[0121] G3: Tag the parts of speech of several words based on the part-of-speech tagging tool; the parts of speech include nouns, adjectives and verbs;

[0122] G4: Extract the conference text based on the feature extraction algorithm to obtain high-frequency words and short sentences;

[0123] G5: Integrate high-frequency words and short sentences into key information for the meeting.

[0124] Exemplarily, preprocessing of the conference text includes unifying character encoding, deleting redundant spaces, and restoring abbreviations in the text; segmenting the conference text based on jieba word segmentation software to obtain several words; tagging the parts of speech of several words based on the LTP part-of-speech tagging tool; extracting the conference text through the TF-IDF (term frequency-inverse document frequency) feature extraction algorithm to obtain high-frequency words and short sentences; integrating the high-frequency words and short sentences into key information of the meeting.

[0125] In this embodiment, key meeting information and the concentration of each participant are filled into the document template, including:

[0126] T1: Extracting official document templates, key meeting information, and the concentration of each participant;

[0127] T2: Extract the mapping relationship between the key information of the meeting, the concentration of each participant and the corresponding unique placeholder from the database;

[0128] T3: Replace each unique placeholder in the official document template with the corresponding key meeting information and the concentration of each participant to generate the meeting minutes.

[0129] The present invention generates meeting minutes by acquiring the mapping relationship between key meeting information, the degree of concentration of each participant and the corresponding unique placeholder, and replacing each unique placeholder in the official document template with the corresponding key meeting information and the degree of concentration of each participant; since there is a one-to-one correspondence between the unique placeholder and the meeting minutes and the degree of concentration of each participant, the meeting content can be accurately filled into the corresponding official document template, which is conducive to improving the standardization of generating meeting minutes in official document format.

[0130] In this embodiment, the official document template is a text in an official document format, and different positions in the text where information is to be filled in are filled with corresponding unique placeholders.

[0131] Part of the data in the above formula is calculated by removing the dimension and taking its numerical value. The formula is a formula closest to the actual situation obtained by software simulation of a large amount of collected data; the preset parameters and preset thresholds in the formula are set by technicians or experts in the field according to the actual situation or obtained through simulation of a large amount of data. For example, the influence coefficient a in this embodiment is determined according to the importance of the content of the meeting.

[0132] Working principle of the present invention:

[0133] The present invention obtains feature information of multiple participants in a meeting; converts voice information into text based on a voice recognition algorithm to obtain an original meeting text; obtains identity information of each participant based on face information; obtains the concentration of each participant based on the identity information and head position information; performs text correction on the original meeting text to obtain a meeting text; extracts key meeting information from the meeting text based on a large language model; integrates the meeting text and the key meeting information into meeting minutes; fills the meeting minutes and the concentration of each participant into an official document template to generate meeting minutes.

[0134] The above embodiments are only used to illustrate the technical method of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical method of the present invention.

Claims

1. A method for generating meeting minutes in an official document format, characterized in that: include: S1: Obtain feature information of multiple participants in the conference, and number the multiple participants respectively; wherein the feature information includes voice information, face information, and head position information; S2: Convert the speech information into text based on the speech recognition algorithm to obtain the original meeting text; S3: Obtaining the identity information of each participant based on the facial information; S4: obtaining the concentration of each participant based on the identity information and the head position information; wherein the concentration includes high concentration and low concentration; S5: Perform text correction on the original conference text to obtain the conference text; wherein the text correction includes correcting semantics, deleting typos, and adjusting word order; S6: Extract key information of the meeting from the meeting text based on the large language model; the key information of the meeting includes the name of the meeting topic, the meeting resolution, the identity information of the host and several speakers; S7: Fill in the key information of the meeting and the concentration of each participant into the official document template to generate the meeting minutes.

2. A method for generating meeting minutes in an official document format according to claim 1, characterized in that: The method of numbering multiple participants respectively includes: Get the seating order of each participant, and label each participant as i in clockwise order starting from the conference host; where the conference host belongs to the participants, i=1, 2, ..., n, where n is the total number of participants.

3. A method for generating meeting minutes in an official document format according to claim 2, characterized in that: The obtaining of the concentration of each participant based on the identity information and the head position information includes: A1: Extract the identity information and head position information of participant i; A2: Obtain the concentration score of each participant based on identity information and head position information; A3: Determine whether the attendee concentration score is greater than a preset score threshold; if yes, mark the attendee concentration of attendee i as high concentration; if no, mark the attendee concentration of attendee i as low concentration.

4. A method for generating meeting minutes in an official document format according to claim 3, characterized in that: The obtaining of the concentration score of each participant based on the identity information and the head position information includes: B1: Obtain the head posture of participant i from the head position information; B2: Get the speaker's speaking time; B3: Obtain the duration of the participant i's head posture facing the conference focus point and mark it as the focus duration; the conference focus point includes the current speaker, the conference screen and the conference materials; B4: The concentration score of each participant is obtained by calculating the ratio of the product of each participant's attention time and influence coefficient to the speaker's speaking time.

5. The method for generating meeting minutes in official document format according to claim 1, characterized in that: The text correction of the original conference text includes: C1: Extract the original conference text and obtain the modal particle database; C2: Determine whether the modal particles in the original conference text belong to the modal particle database; if yes, remove the corresponding modal particles; if no, jump to C3; C3: Mark the original conference text without modal particles as the first text; C4: Correct the typos in the first text to obtain the second text; C5: Adjust the word order of the second text to the normal word order to obtain the conference text.

6. A method for generating meeting minutes in an official document format according to claim 5, characterized in that: The step of adjusting the word order of the second text to a normal word order includes: D1: Extract the second text and the voice information of the participants; D2: dividing the second text into a number of continuous text segments based on the voice information of the participants; D3: Mark the text segment as j; where j = 1, 2, ..., m, and m is the total number of text segments; D4: Calculate the semantic relevance between text segment j and the rest of the text segments; D5: Determine whether the semantic relevance is greater than a preset relevance threshold; If yes, the corresponding text segment is added to the queue of text segments to be adjusted; If not, the word order of the corresponding text segment remains unchanged; D6: Arrange the text segments in the queue of text segments to be adjusted in descending order according to their corresponding semantic relevance; D7: Extract the text segment with the highest semantic relevance in the queue of text segments to be adjusted, and adjust it to the position of the word order of text segment j plus one.

7. A method for generating meeting minutes in an official document format according to claim 6, characterized in that: The dividing the second text into a plurality of continuous text segments based on the voice information of the participants includes: E1: Obtain some continuous voice information of all participants; E2: Calculate the pause intervals between adjacent speech messages in sequence; E3: Determine whether the pause interval is greater than a preset pause threshold; if yes, divide the text information corresponding to the adjacent voice information into two text segments; if no, merge the text information corresponding to the adjacent voice information into one text segment.

8. The method for generating meeting minutes in official document format according to claim 1, characterized in that: The key information of the meeting is extracted from the meeting text based on the large language model, including: G1: Preprocess the conference text; G2: Segment the conference text based on word segmentation software to obtain several words; G3: Tag the parts of speech of several words based on the part-of-speech tagging tool; the parts of speech include nouns, adjectives and verbs; G4: Extract the conference text based on the feature extraction algorithm to obtain high-frequency words and short sentences; G5: Integrate high-frequency words and short sentences into key information for the meeting.

9. The method for generating meeting minutes in an official document format according to claim 1, characterized in that: The key information of the meeting and the concentration of each participant are filled into the official document template, including: T1: Extracting official document templates, key meeting information, and the concentration of each participant; T2: Extract the mapping relationship between the key information of the meeting, the concentration of each participant and the corresponding unique placeholder from the database; T3: Replace each unique placeholder in the official document template with the corresponding key meeting information and the concentration of each participant to generate the meeting minutes.

10. The method for generating meeting minutes in official document format according to claim 1, characterized in that: An official document template is a text in an official document format, and different positions in the text where information is to be filled in are filled with corresponding unique placeholders.

Images