Conference record summarization and induction method based on voice

Abstract

The invention discloses a conference record summarization and induction method based on voice, relates to the technical field of conference record summarization, and solves the technical problem of low service quality caused by the fact that conference summarization is not optimized according to the use habits of a user in the prior art. The conference voice file is collected through the voice collection device and transmitted to the system, the identity verification information of the user is collected, and the identity of the user is verified according to the identity verification information; converting the conference voice into a conference text, identifying and extracting a structural theme and related keywords in the conference text, and counting the number of the keywords to obtain theme information; generating a conference summary according to the theme information; and extracting a revision record of the user for the conference summary, extracting revision theme features in the revision record, marking the revision theme features as personalized records, adjusting theme correlation coefficients according to the personalized records, constructing and updating a user portrait, and regenerating and optimizing the conference summary.

Description

Technical Field

[0001] The invention belongs to the field of conference record summarization and relates to Internet of Things technology, and specifically is a conference record summarization and induction method based on voice. Background Art

[0002] In the modern office environment, meetings are an important way to exchange information. With the rise of remote work, the need to record and summarize meeting content is increasing. The traditional manual recording method is not only time-consuming and laborious, but also easy to miss important information. The meeting record summary function based on speech recognition and large model technology came into being, aiming to improve the efficiency and accuracy of meeting records through automation.

[0003] The invention patent with application number 2024101952439 discloses a meeting record summary generation method, terminal device and storage medium. The invention creates a voice channel for each meeting member and sets a hook mechanism in the voice channel to generate a complete meeting summary based on the voice conference text collected by all voice channels; when obtaining the meeting summary, the invention collects the voice text of each member to analyze and generate the meeting summary. In actual application, the generated meeting summary may not meet the user's psychological expectations and needs to be changed, requiring a more personalized system service.

[0004] The present invention provides a conference record summarizing and induction method based on speech 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 voice-based conference record summary and induction method to solve the technical problem that the prior art does not optimize the conference summary according to the user's usage habits, resulting in low service quality.

[0006] To achieve the above object, the first aspect of the present invention provides a method for summarizing and generalizing conference records based on speech, comprising:

[0007] Step 1: Collect the conference audio through the audio collection device and collect the user's identity verification information;

[0008] Step 2: Verify the user's identity through identity verification information;

[0009] Step 3: Convert the conference speech into conference text; identify and extract the topic information in the conference text, divide the conference text into different topic structures according to the topic information and generate a conference summary;

[0010] Step 4: The user revises the meeting summary to obtain a revised text and a revision record, generates a personalized record based on the revised record, and optimizes the meeting summary based on the personalized record to generate an optimized meeting summary.

[0011] It should be noted that the conference voice and identity verification information of the mobile phone in the present invention are collected under the premise of the consent of the conference participants.

[0012] Preferably, verifying the identity of the user through identity verification information includes:

[0013] A1: Extract identity verification information from a pre-stored database containing registration and usage records; the identity verification information is the mobile phone number entered by the user and the input verification code received;

[0014] A2: Determine whether the verification code you input is the same as the verification code automatically generated by the system; if yes, jump to A3; if no, the identity verification fails;

[0015] A3: Determine whether the mobile phone number is included in the pre-stored database; if so, extract the usage record library corresponding to the mobile phone number in the pre-stored database; if not, create a new user's usage record library and bind the usage record to the mobile phone number.

[0016] Preferably, the identifying and extracting the subject information in the conference text includes:

[0017] Extract meeting texts;

[0018] The structural themes in the conference text are extracted through natural language processing technology, and the keywords related to the structural themes and the corresponding keyword quantities are extracted; the structural themes, keywords and keyword quantities are combined to generate topic information.

[0019] Preferably, dividing the conference text into different topic structures according to the topic information and generating a conference summary includes:

[0020] Extract subject information;

[0021] Convert keywords and structural topics into keyword vectors and topic vectors, and calculate the cosine similarity between keyword vectors and topic vectors to obtain text relevance;

[0022] The topic relevance WBpq between the keyword numbered p and the structural topic numbered q is calculated by the formula WBpq=Sp×Wpq×α; wherein Sp represents the value after normalization of the number of occurrences of the keyword numbered p, and Wpq represents the text relevance between the keyword numbered p and the structural topic numbered q; q=1, 2, 3…n, n is a positive integer; p=1, 2, 3…m, m is a positive integer; α is the topic relevance coefficient, which is initially set to 1;

[0023] The keywords under each topic are sorted according to the degree of topic relevance, and the structural topics are sorted according to the order in which they appear in the conference text to obtain a conference summary.

[0024] Preferably, generating a personalized record according to the revision record includes:

[0025] Extract the revision records of the meeting summary made by the user;

[0026] Determine whether there is a usage record in the user's usage record library; if yes, extract the user portrait in the usage record library, analyze the user's revision usage intention in combination with the revision record and the user portrait, and extract the revision theme features; if no, extract the revision theme features in the revision record;

[0027] Marks the revision topic feature as a personalized record.

[0028] Preferably, the step of optimizing the meeting summary according to the personalized record to generate an optimized meeting summary includes:

[0029] Extract personalized notes and meeting summaries;

[0030] Modify and optimize the topic correlation coefficient when generating the meeting summary based on personalized records, build a personalized user portrait, and regenerate and optimize the meeting summary.

[0031] Compared with the prior art, the present invention has the following beneficial effects:

[0032] 1. The present invention collects conference voice files through a voice collection device and transmits them to the system, collects user identity authentication information, verifies the identity of the user based on the identity authentication information, and ensures the consistency and integrity of user data by binding the mobile phone number to the usage record, and facilitates later data management and analysis, can track user usage behavior and preferences, and also reduces the difficulty of operation; converts conference voice into conference text, identifies and extracts structural topics and related keywords in the conference text, and counts the number of keywords to obtain topic information; converts keywords and structural topics into text vectors, and calculates the topic relevance between keywords and structural topics, and generates a conference summary based on the topic relevance, which can more accurately identify the main content of the meeting, and the generated conference summary is more comprehensive and accurate, so that the conference summary has better structured characteristics, which is convenient for readers to quickly understand the key points of the meeting, reduces the time and energy of manual sorting, and improves work efficiency; extracts the user's revision record of the conference summary, extracts the revision topic features in the revision record and marks the revision topic features as personalized records, adjusts the topic relevance coefficient according to the personalized record, constructs and updates user portraits, regenerates and optimizes the conference summary, which is more in line with the user's personalized needs, and continuously optimizes the conference summary based on the user's revision feedback, improves the quality and accuracy of the summary, and improves work efficiency;

[0033] 2. The present invention converts keywords and structural topic texts into text vectors, calculates the cosine similarity between the keyword vector and the topic vector to obtain text relevance, and calculates the topic relevance between the keywords and the structural topics in combination with the set topic relevance coefficient. The meeting summary can be sorted according to the size of the topic relevance and the order in which the structural topics appear in the meeting text. The main content of the meeting can be identified more accurately, and the key points of the meeting can be grasped for rapid understanding when reading, thereby improving the quality of the meeting summary and improving work efficiency and decision-making support capabilities. When the user makes revisions to the meeting summary, the topic relevance coefficient is optimized and adjusted according to the revision record, and the optimized meeting summary is regenerated to be more in line with the user's reading habits and preferences, thereby improving the user experience and the quality of the meeting summary. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] 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.

[0035] Figure 1 The following is a flowchart of an embodiment of the present invention.

[0036] Figure 2 The figure is a complete flow chart of verifying the identity of a user in one embodiment of the present invention.

[0037] Figure 3 This is a workflow diagram for modifying and optimizing topic relevance coefficients based on personalized records in one embodiment of the present invention. DETAILED DESCRIPTION

[0038] 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.

[0039] See also Figure 1-Figure 3 The first embodiment of the present invention provides a method for summarizing and generalizing conference records based on speech, comprising:

[0040] Step 1: Collect conference audio and user's identity verification information through audio collection equipment.

[0041] Exemplarily, voice information during the meeting is collected through a voice collection device (such as a microphone), and the mobile phone number entered by the user when logging in and the input verification code entered by the user are collected.

[0042] Step 2: Verify the user's identity through identity verification information.

[0043] Exemplarily, determine whether the input verification code is the same as the verification code automatically generated by the system; the verification code automatically generated by the system is "LCK30", and the input verification code entered by the user is "LCK30", and it is determined that the input verification code is the same as the verification code automatically generated by the system, and continue to determine whether the mobile phone number is included in the pre-stored database; if it is determined that the mobile phone number entered by the user is not included in the pre-stored database, then a new usage record library bound to the mobile phone number is created.

[0044] In some other embodiments, if the input verification code is different from the verification code automatically generated by the system, the verification code is regenerated and the verification is performed again; if the mobile phone number is determined to be included in the pre-stored database after the verification is passed, it means that the user corresponding to the mobile phone number has used the system and has a corresponding usage record library, and the corresponding usage record library is extracted for subsequent operations; the pre-stored database contains the usage record library registered and used by all users, and the corresponding usage record library is matched by matching the mobile phone number.

[0045] Step 3: Convert the conference speech into conference text; identify and extract the topic information in the conference text, divide the conference text into different topic structures according to the topic information and generate a meeting summary.

[0046] Exemplarily, the collected conference audio is transcribed through speech transcription software (such as Google Cloud Translator) to obtain a conference text, and the conference text is input into a natural language processing tool (such as spaCY) to extract the structural themes in the conference text. The structural themes in the conference text are "product development", "market strategy" and "team building". The keywords corresponding to the three structural themes are extracted, namely, the keywords for "product development" are deadline, technical challenges and cost budget, the keywords for "market strategy" are marketing channels, target customers and competitors, and the keywords for "team building" are recruitment plan and training plan.

[0047] The keyword "technical challenge" in the "product development" theme appeared 5 times, "deadline" appeared 3 times, and "cost budget" appeared 2 times; the keyword "marketing channel" in the market strategy theme appeared 6 times, "target customers" appeared 5 times, and "competitors" appeared 2 times; the keyword "recruitment plan" in the team building theme appeared 4 times and "training plan" appeared 3 times.

[0048] The keyword vector and topic vector are obtained by inputting the structural topics and keywords into the word embedding model. The structural topics are numbered according to the order in which they appear in the conference text, and the keywords are numbered from large to small according to the number of times they appear in the structural topics. The cosine similarity of the keyword vector and the topic vector is calculated, and the obtained cosine similarity is marked as the text relevance.

[0049] The number of occurrences of each keyword under the structural theme was normalized. The normalized value of the keyword "technical challenge" in the "product development" theme was 0.5, the normalized value of the keyword "deadline" was 0.3, and the normalized value of the keyword "cost budget" was 0.2. The cosine similarity between "technical challenge" and "product development" was 0.56, the cosine similarity between "deadline" and "product development" was 0.34, and the cosine similarity between "cost budget" and "product development" was 0. is 0.28; through the formula WBpq=Sp×Wpq×α, we get the subject relevance WB11=0.28 between “technical challenge” and “product development”, the subject relevance WB21=0.10 between “deadline” and “product development”, and the subject relevance WB31=0.06 between “cost budget” and “product development”; sort the keywords under the structural theme of “product development” according to the size of the subject relevance; similarly calculate and sort the keywords under the structural themes of “market strategy” and “team building” to get the meeting summary.

[0050] Step 4: The user revises the meeting summary to obtain a revised text and a revision record, generates a personalized record based on the revised record, and optimizes the meeting summary based on the personalized record to generate an optimized meeting summary.

[0051] Exemplarily, a user revises a meeting summary to obtain a revision record; it is determined whether there is a usage record in the user's usage record library. In this embodiment, the user is a newly registered user with no usage record, so the revision keywords related to the corresponding topic in the revision record are extracted through a natural language processing tool (such as spaCY), and the revision keywords are tagged and classified into deletion tags, addition tags, and correction tags; the deletion keywords and the corresponding deletion numbers in the deletion tags are counted, the corresponding deletion keywords are matched, and the corresponding topic correlation coefficient is reduced according to the proportion of the deletion number to the original keyword number; the addition keywords and the corresponding addition number in the addition tag are counted, the corresponding addition keywords are matched, and the corresponding topic correlation coefficient is increased according to the proportion of the addition number to the original keyword number; the corresponding revision keywords are counted to determine whether the revision keywords are overall keyword modifications, that is, the statement expression is modified; if yes, the revision keywords replace the original keywords recorded in the system; if not, the revised keywords will be recorded in the system and the context information of the original keywords will be retained.

[0052] After modifying the topic correlation coefficient and system records, regenerate the optimized meeting summary.

[0053] 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 technical personnel in this field according to actual conditions or obtained through simulation of a large amount of data.

[0054] Working principle of the present invention:

[0055] The present invention collects conference voices through a voice collection device, collects user identity verification information, and verifies the user's identity based on the identity verification information; converts conference voices into conference texts, identifies and extracts structural topics and keywords in the conference texts, and counts the number of keywords to integrate them into topic information; converts keywords and structural topics in the topic information into keyword vectors and topic vectors, and calculates cosine similarity, and calculates the topic relevance between the structural topics and the keywords in combination with the set topic correlation coefficient, and obtains a conference summary by sorting the topic relevance and the order in which the structural topics appear in the conference text; collects the user's revision records of the conference summary, and extracts personalized records in the revision records, adjusts the topic correlation coefficient based on the personalized records and constructs a user portrait, and regenerates an optimized conference summary.

[0056] 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 summarizing and summarizing conference records based on speech, characterized in that: include: Step 1: Collect the conference audio through the audio collection device and collect the user's identity verification information; Step 2: Verify the user's identity through identity verification information; Step 3: Convert the conference speech into conference text; identify and extract the topic information in the conference text, divide the conference text into different topic structures according to the topic information and generate a conference summary; Step 4: The user revises the meeting summary to obtain a revised text and a revision record, generates a personalized record based on the revised record, and optimizes the meeting summary based on the personalized record to generate an optimized meeting summary.

2. The method for summarizing and generalizing conference records based on voice according to claim 1 is characterized in that: Verifying the user's identity through identity verification information includes: A1: Extract identity verification information from a pre-stored database containing registration and usage records; the identity verification information is the mobile phone number entered by the user and the input verification code received; A2: Determine whether the verification code you input is the same as the verification code automatically generated by the system; if yes, jump to A3; if no, the identity verification fails; A3: Determine whether the mobile phone number is included in the pre-stored database; if so, extract the usage record library corresponding to the mobile phone number in the pre-stored database; if not, create a new user's usage record library and bind the usage record to the mobile phone number.

3. The method for summarizing and generalizing conference records based on voice according to claim 1, characterized in that: The identifying and extracting the subject information in the conference text includes: Extract meeting texts; The structural themes in the conference text are extracted through natural language processing technology, and the keywords related to the structural themes and the corresponding keyword quantities are extracted; the structural themes, keywords and keyword quantities are combined to generate topic information.

4. The method for summarizing and generalizing conference records based on speech according to claim 1, characterized in that: The conference text is divided into different topic structures according to the topic information and a conference summary is generated, including: Extract subject information; Convert keywords and structural topics into keyword vectors and topic vectors, and calculate the cosine similarity between keyword vectors and topic vectors to obtain text relevance; The topic relevance WBpq between the keyword numbered p and the structural topic numbered q is calculated by the formula WBpq=Sp×Wpq×α; wherein Sp represents the value after normalization of the number of occurrences of the keyword numbered p, and Wpq represents the text relevance between the keyword numbered p and the structural topic numbered q; q=1, 2, 3…n, n is a positive integer; p=1, 2, 3…m, m is a positive integer; α is the topic relevance coefficient, which is initially set to 1; The keywords under each topic are sorted according to the degree of topic relevance, and the structural topics are sorted according to the order in which they appear in the conference text to obtain a conference summary.

5. The method for summarizing and generalizing conference records based on voice according to claim 1, characterized in that: The generating of personalized records according to the revision records includes: Extract the revision records of the meeting summary made by the user; Determine whether there is a usage record in the user's usage record library; if yes, extract the user portrait in the usage record library, analyze the user's revision usage intention in combination with the revision record and the user portrait, and extract the revision theme features; if no, extract the revision theme features in the revision record; Marks the revision topic feature as a personalized record.

6. The method for summarizing and generalizing conference records based on voice according to claim 1, characterized in that: The step of optimizing the meeting summary according to the personalized record to generate an optimized meeting summary includes: Extract personalized notes and meeting summaries; Modify and optimize the topic correlation coefficient when generating the meeting summary based on personalized records, build a personalized user portrait, and regenerate and optimize the meeting summary.

Images