CN-122019767-A - Conference summary processing method, device, equipment and storage medium

CN122019767ACN 122019767 ACN122019767 ACN 122019767ACN-122019767-A

Abstract

The embodiment of the application provides a conference summary processing method, device, equipment and storage medium, which are used for reconstructing the content of a conference summary so as to improve the readability of the conference summary. The method comprises the steps of obtaining conference texts to be processed and multi-source matching data, wherein the multi-source matching data comprises participant metadata, current conference information and historical conference data, performing speaker matching on the conference texts to be processed based on the multi-source matching data to obtain first conference summary, wherein the first conference summary comprises the text data of the conference texts to be processed and speaker information related to the text data, performing topic identification on the first conference summary to obtain a plurality of topics corresponding to the first conference summary and summary information corresponding to each topic, and generating a target conference summary based on the plurality of topics and the summary information corresponding to each topic.

Inventors

Gao Caifang
HE HONGDE
GU YIBING

Assignees

深圳TCL新技术有限公司

Dates

Publication Date: 20260512
Application Date: 20260114

Claims (10)

1. A conference summary processing method, comprising: acquiring conference text to be processed and multi-source matching data, wherein the multi-source matching data comprises participant metadata, current conference information and historical conference data; Performing speaker matching on the conference text to be processed based on the multi-source matching data to obtain a first conference summary, wherein the first conference summary comprises text data of the conference text to be processed and speaker information associated with the text data; Identifying the first meeting summary to obtain a plurality of topics corresponding to the first meeting summary and summary information corresponding to each topic; Generating a target conference summary based on the plurality of topics and summary information corresponding to each topic.
2. The method of claim 1, wherein after generating the target meeting summary based on the plurality of topics and summary information corresponding to each of the topics, the method further comprises: Performing conflict verification on the target meeting summary to obtain a verification result; And updating the target meeting summary based on the verification result.
3. The method of claim 2, wherein performing a conflict check on the target meeting summary to obtain a check result comprises: Detecting a target conflict type in the target meeting summary, wherein the target conflict type comprises identity consistency conflict, subject standing conflict and content integrity conflict; Generating a hypothetical branch for the target conflict type; calculating the confidence coefficient of the hypothesized branch; And determining the verification result based on the confidence.
4. A method according to any one of claims 1 to 3, wherein before performing topic identification on the first topic summary to obtain a plurality of topics corresponding to the first topic summary and summary information corresponding to each topic, speaker matching is performed on the conference text to be processed based on the multi-source matching data to obtain a first topic summary, the method further comprises: and extracting and supplementing action item information for the first meeting summary to obtain structured data, wherein the action item information is used for representing task content, task execution objects and task time contained in the meeting text to be processed, and the structured data is used for identifying the topics.
5. The method of claim 4, wherein extracting and supplementing action item information for the first session summary to obtain structured data comprises: performing intention recognition on the first meeting summary to obtain intention classification of each piece of text data in the first meeting summary; Determining text data to be processed based on the intent classification; carrying out grammar analysis on the text data to be processed so as to extract and supplement the task content, the task execution object and the task time; The structured data is generated based on the task content, the task execution object, and the task time.
6. A method according to any one of claims 1 to 3, wherein the identifying the first session summary to obtain a plurality of sessions corresponding to the first session summary and summary information corresponding to each session comprises: performing semantic analysis on the first meeting summary to obtain a semantic vector of text data in the first meeting summary; Performing topic clustering on the semantic vectors to obtain a plurality of initial topics; Performing topic switching detection on the semantic vector to obtain the starting time of the initial topics; Carrying out semantic similarity calculation and aggregation on the semantic vectors to obtain the complete content of the initial subjects; constructing a plurality of issues corresponding to the first meeting summary based on the start time and the complete content; extracting key nodes of text data of each of the plurality of topics; and compressing and recombining the text data of each topic by using a summary generation technology based on the key nodes so as to obtain summary information of each topic.
7. A method according to any of claims 1 to 3, wherein speaker matching the conference text to be processed based on the multi-source matching data to obtain a first conference summary comprises: based on at least one speaker matching operation, obtaining a candidate speaker set corresponding to each piece of text data of the conference text to be processed; determining a speaker corresponding to each piece of text data based on the candidate speaker set corresponding to each piece of text data so as to obtain the first meeting summary; the at least one speaker matching operation includes: constructing a role vector library based on the participant metadata and the historical conference data; Acquiring semantic vectors corresponding to each text data of the conference text to be processed; performing similarity matching from the role vector library based on the semantic vector to obtain a first candidate speaker corresponding to each piece of text data of the conference text to be processed; And/or; Carrying out named entity recognition on the conference text to be processed to obtain name information in the conference text to be processed; Determining a second candidate speaker corresponding to each piece of text data in the conference text to be processed based on a co-reference resolution method and the title information; And/or; Performing dialogue interactive recognition on text data in the conference text to be processed to obtain a third candidate speaker corresponding to each text data segment in the conference text to be processed; And/or; performing domain term identification on text data in the conference text to be processed to obtain a domain term set; matching each domain term in the domain term set to obtain a fourth candidate speaker from a domain term library, wherein the domain term library is used for representing the association relation between the domain term and the participant role; And/or; Performing speaker matching on each piece of text data of the conference text to be processed based on time sequence viscosity characteristics and a global optimal allocation method to obtain a fifth candidate speaker; Wherein the first candidate speaker, the second candidate speaker, the third candidate speaker, the fourth candidate speaker, and the fifth candidate speaker are included in the candidate speaker set.
8. A conference summary processing apparatus, comprising: The acquisition module is used for acquiring conference text to be processed and multi-source matching data, wherein the multi-source matching data comprises participant metadata, current conference information and historical conference data; The processing module is used for carrying out speaker matching on the conference text to be processed based on the multi-source matching data to obtain a first conference summary, wherein the first conference summary comprises text data of the conference text to be processed and speaker information related to the text data, the first conference summary is identified to obtain a plurality of issues corresponding to the first conference summary and summary information corresponding to each issue, and a target conference summary is generated based on the plurality of issues and the summary information corresponding to each issue.
9. A computer device, the computer device comprising: one or more processors; Memory, and One or more applications, wherein the one or more applications are stored in the memory and configured to be executed by the processor to implement the method of any of claims 1 to 7.
10. A computer readable storage medium, having stored thereon a computer program, the computer program being loaded by a processor to perform the steps of the method of any of claims 1 to 7.

Description

Conference summary processing method, device, equipment and storage medium Technical Field The application relates to the field of computers, in particular to a conference summary processing method, a conference summary processing device, conference summary processing equipment and a conference summary processing storage medium. Background The current automatic meeting summary generation technology focuses on two directions of abstract extraction and keyword extraction, and basic information extraction can be completed, but has obvious defects in multiple aspects in practical application. For example, identity alignment is inaccurate, anonymous speaker and semantic roles thereof cannot be precisely matched, subjects are insufficient, generated contents are more easily listed according to time sequence, content reconstruction driven by subjects is not realized, so that a user is difficult to capture core conclusions quickly, implicit information mining is imperfect, recognition effects on pain points, requirements and action items in a conference are poor, responsibilities and deadlines corresponding to the action items are always missed, a self-correction and audit mechanism is lacking, contradictory information in the conference is faced, backtracking correction capability is not provided, meanwhile, systematic audit flow and related evidence chain support are lacking, data fusion is insufficient, multi-source enterprise data such as project data, system data and the like are not fully integrated, and the accuracy and the intelligent level of summary generation are difficult to be further improved. Therefore, there is a need for a meeting summary processing method that can improve the readability of meeting summary. Disclosure of Invention The embodiment of the application provides a conference summary processing method, device, equipment and storage medium, which are used for reconstructing the content of a conference summary so as to improve the readability of the conference summary. The technical scheme adopted by the invention for solving the problems is as follows: in a first aspect, the present application provides a method for processing a meeting summary, including: acquiring conference text to be processed and multi-source matching data, wherein the multi-source matching data comprises participant metadata, current conference information and historical conference data; Performing speaker matching on the conference text to be processed based on the multi-source matching data to obtain a first conference summary, wherein the first conference summary comprises text data of the conference text to be processed and speaker information associated with the text data; Identifying the first meeting summary to obtain a plurality of topics corresponding to the first meeting summary and summary information corresponding to each topic; Generating a target meeting summary based on the plurality of topics and summary information corresponding to each topic. In some embodiments of the present application, after generating the target meeting summary based on the plurality of topics and summary information corresponding to each of the topics, the method further comprises: performing conflict verification on the target meeting summary to obtain a verification result; And updating the target meeting summary based on the verification result. In some embodiments of the present application, the performing conflict verification on the target meeting summary to obtain a verification result includes: Detecting a target conflict type in the target meeting summary, wherein the target conflict type comprises identity consistency conflict, subject position conflict and content integrity conflict; Generating a hypothetical branch for the target conflict type; Calculating the confidence of the hypothesized branch; The verification result is determined based on the confidence level. In some embodiments of the present application, before performing topic identification on the first meeting summary to obtain a plurality of topics corresponding to the first meeting summary and summary information corresponding to each topic, performing speaker matching on the conference text to be processed based on the multi-source matching data to obtain a first meeting summary, the method further includes: and extracting and supplementing action item information for the first meeting summary to obtain structured data, wherein the action item information is used for representing task content, task execution objects and task time contained in the to-be-processed meeting text, and the structured data is used for identifying the topics. In some embodiments of the application, performing action item information extraction and supplementation to the first session summary to obtain structured data comprises: performing intention recognition on the first meeting summary to obtain intention classification of each piece of text data in the first meeting sum