Conference summary generation method, electronic device, storage medium and program product
By segmenting the conference text and processing large language models, accurate conference summary is generated, and the automation and accuracy problems of conference summary generation in the existing technology are solved, and an efficient summary generation process is achieved.
Patent Information
- Application Number
- CN202411912403.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-24
- Publication Date
- 2025-04-29
AI Technical Summary
It is difficult for the existing technology to automate, efficiently and accurately generate meeting summary of online audio and video conferences, especially when key personnel fail to attend the meeting, they cannot quickly grasp the core content of the meeting.
By obtaining the conference text and dividing it into multiple text fragments, a large language model is used to generate fragment summary and text deduplication, and finally a full text abstract is generated, combining the first and second prompt words to guide the model to output an accurate conference summary.
It realizes the automation, efficient and accurate generation of conference summary, avoids information omissions or misunderstandings, retains important details and discussion points of the conference, and ensures the accuracy and coherence of the summary.
Smart Images
Figure CN119903835A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of deep learning, and in particular to a conference summary generation method, electronic equipment, storage medium and program product. Background Art
[0002] In the context of the increasing popularity of digital office, the rapid development of VoIP (Voice over Internet Protocol) technology and advanced audio and video communication technology has greatly promoted the widespread application of online audio and video conferencing systems. These systems not only optimize the communication efficiency within the enterprise, but also break geographical boundaries, making remote collaboration possible. However, with the increase in the frequency of meetings and the diversity of participants, a significant problem has gradually emerged: when some key personnel fail to attend the meeting for some reason, they often need to quickly grasp the core content of the meeting in order to connect with subsequent work or make decisions. Therefore, how to achieve the automatic, efficient and accurate generation of meeting summaries based on the content of the meeting has become an urgent problem to be solved. Summary of the invention
[0003] The purpose of the embodiments of the present invention is to provide a conference summary generation method, electronic device and storage medium, which can realize automatic, efficient and accurate generation of conference summaries based on conference content through conference text segmentation and a large language model.
[0004] In order to solve the above technical problems, an embodiment of the present invention provides a conference summary generation method, comprising:
[0005] Acquire a conference text, divide the conference text into multiple text segments, and the contents of each text segment do not overlap with each other; input each text segment and a first prompt word into a large language model in sequence to obtain a segment summary corresponding to each text segment; perform text deduplication on each segment summary to obtain a segment abstract corresponding to each text segment; input each segment abstract and a second prompt word into the large language model to obtain a full-text abstract corresponding to the conference text.
[0006] An embodiment of the present invention further provides an electronic device, including:
[0007] at least one processor; and,
[0008] a memory communicatively connected to the at least one processor; wherein,
[0009] The memory stores instructions that can be executed by the at least one processor. The instructions are executed by the at least one processor to enable the at least one processor to perform the steps of the conference summary generation method as described above.
[0010] The embodiment of the present invention further provides a computer-readable storage medium storing a computer program, and when the computer program is executed by a processor, the steps of the conference summary generation method described above are implemented.
[0011] An embodiment of the present invention further provides a computer program product, characterized in that it includes computer instructions, and when the computer instructions are executed by a processor, the steps of the conference summary generation method described above are implemented.
[0012] In addition, before obtaining the conference text, the method as described above also includes: obtaining conference audio / video content through an audio / video acquisition device; converting the conference audio / video content into a preset format, and performing automatic speech recognition on the conference audio / video content in the preset format to obtain the conference text; wherein the preset format is a format supported by automatic speech recognition.
[0013] In addition, as described above, each of the text fragments is input into the large language model in sequence together with the first prompt word according to the order of the text fragments to obtain a fragment summary corresponding to each of the text fragments, including: selecting a text fragment to be processed according to the order of the text fragments, and inputting the text fragment to be processed and the first prompt word into the large language model; the large language model outputs the fragment summary corresponding to the text fragment to be processed; wherein the first prompt word is at least used to instruct the large language model to extract the content title of the text fragment to be processed, refine the key points of the text fragment to be processed, and output the content title and the key points in the form of a list as the fragment summary corresponding to the text fragment to be processed.
[0014] In addition, the above-mentioned deduplication of text is performed on each of the fragment summaries to obtain the fragment summaries corresponding to each of the text fragments, including: dividing all of the fragment summaries into multiple fragment summary pairs in sequence according to the order of the text fragments corresponding to each of the fragment summaries; wherein, in two adjacent fragment summary pairs, the second fragment summary in the first fragment summary pair and the first fragment summary in the second fragment summary pair are the same fragment summary; performing text similarity analysis on the first fragment summary and the second fragment summary in each fragment summary pair in sequence according to the order of the multiple fragment summary pairs, removing the text in the second fragment summary that is repeated with the first fragment summary, and updating the second fragment summary; after the second fragment summary in each fragment summary pair is completely updated, the fragment summaries corresponding to each of the text fragments are obtained.
[0015] In addition, as described above, each of the fragment summaries and the second prompt word is input into the large language model to obtain a full-text summary corresponding to the conference text, including: inputting each of the fragment summaries and the second prompt word into the large language model in sequence according to the order of each of the fragment summaries; wherein the second prompt word is used to instruct the large language model to iteratively generate the full-text summary according to the input each of the fragment summaries; and the large language model outputs the full-text summary corresponding to the conference text.
[0016] In addition, the large language model as described above iteratively generates the full-text summary based on the input fragment summaries, including: selecting at least two of the fragment summaries in the order of the fragment summaries to generate a first summary summary; generating a new first summary summary based on the first summary summary and the first unselected fragment summary among the fragment summaries to update the first summary summary; and using the new first summary summary generated after all the fragment summaries are selected as the full-text summary.
[0017] In addition, after obtaining the full-text abstract corresponding to the conference text as described above, the method further includes: merging each of the segment abstracts with the full-text abstract to form a final conference abstract, and outputting the final abstract in a preset format.
[0018] Compared with the prior art, the implementation method of the present invention obtains the conference text and divides the conference text into multiple text fragments. According to the sequence of the text fragments, each text fragment is input into the large language model in sequence with the first prompt word to obtain a fragment summary corresponding to each text fragment; each fragment summary is deduplicated to obtain a fragment summary corresponding to each text fragment; each fragment summary and the second prompt word are input into the large language model to obtain a full-text summary corresponding to the conference text. The present invention uses a large language model to generate summaries. The model can deeply understand the conference text, accurately extract key information, avoid information omissions or misunderstandings that may be caused by traditional methods, and retain important details and discussion points in the meeting. The summary automatically generated by the large language model is closer to the original intention of the meeting, ensuring the accuracy and coherence of the conference summary. BRIEF DESCRIPTION OF THE DRAWINGS
[0019] Figure 1 The specific process of the conference summary generation method according to the embodiment of the present invention is as follows: Figure 1 ;
[0020] Figure 2 The specific process of the conference summary generation method according to the embodiment of the present invention is as follows: Figure 2 ;
[0021] Figure 3 is a framework flow chart of a conference summary generation method according to an embodiment of the present invention;
[0022] Figure 4 is a schematic structural diagram of an electronic device according to an embodiment of the present invention. DETAILED DESCRIPTION
[0023] In order to make the purpose, technical scheme and advantages of the embodiments of the present invention clearer, the following will be described in detail with reference to the accompanying drawings. However, it will be appreciated by those skilled in the art that in the various embodiments of the present invention, many technical details are provided in order to enable the reader to better understand the present application. However, even without these technical details and various changes and modifications based on the following embodiments, the technical scheme claimed in the present application can be implemented.
[0024] One embodiment of the present invention relates to a method for generating a conference summary, such as Figure 1 As shown, the following steps are included.
[0025] Step 101: Obtain a conference text, and divide the conference text into a plurality of text segments, wherein the contents of each text segment do not overlap with each other.
[0026] Specifically, the conference text can be obtained by recording the entire conference audio and converting it into text in real time, and this application does not impose specific restrictions on this. After obtaining the complete conference text, the conference text is segmented into multiple text segments with non-overlapping content. The specific method of text segmentation can be performed by character segmentation, semantic segmentation, recursive character text segmentation, regular expressions, etc., and this application does not impose specific restrictions on this. It should be noted that in order to ensure the semantic integrity of each text segment after segmentation, regular expressions are used to segment the conference text in this embodiment. The content of each two adjacent text segments after segmentation is connected but there is no overlapping content.
[0027] In one example, the conference text can be obtained by obtaining the conference audio / video content through an audio / video acquisition device; the conference audio / video content is converted into a preset format, and automatic speech recognition is performed on the conference audio / video content in the preset format to obtain the conference text; wherein the preset format is a format supported by automatic speech recognition.
[0028] Specifically, Automatic Speech Recognition (ASR) is a technology that converts human speech into text. It allows computers to understand human speech and convert it into a text form that can be processed and understood by computers. In this embodiment, the audio / video content of the meeting can be obtained through audio and video acquisition devices such as cameras and recorders at the beginning of the meeting. After the meeting, the acquisition is completed, and the collected audio / video content of the meeting is converted into a preset format, such as m4a, mp3, mp4, mpeg, mpga, wav, webm and other formats supported by automatic speech recognition. After obtaining the conference audio / video content in the preset format, automatic speech recognition is performed on it to obtain the conference text.
[0029] Step 102: According to the sequence of the text segments, the text segments are input into the large language model in sequence together with the first prompt word to obtain a segment summary corresponding to each text segment.
[0030] Specifically, a large language model (LLM) refers to a deep learning model trained with a large amount of text data that can generate natural language text or understand the meaning of language text. Large language models can handle a variety of natural language tasks, such as text classification, question answering, and dialogue, and are an important path to artificial intelligence.
[0031] After completing the text segmentation and obtaining multiple text segments, this embodiment uses the large language model and the multiple text segments to generate a summary of the conference text. Since the conference text records the content according to the timeline of the conference, the multiple text segments obtained by segmenting the conference text also have a time sequence. Each text segment is input into the large language model in sequence with the first prompt word to obtain a segment summary corresponding to each text segment.
[0032] In the field of artificial intelligence (AI), "prompt" refers to text or instructions that provide input to a model to guide it to produce a specific output. It is a text paragraph provided by the user when interacting with the model, which is used to describe the information, answer, text, etc. that the user wants to obtain from the model. The purpose of the prompt is to guide the model to produce the desired response in order to better control the output generated by the model.
[0033] In this embodiment, each text segment and a first prompt are sequentially input into a large language model, and the first prompt guides the output content of the large language model to obtain a segment summary corresponding to each text segment.
[0034] In an example, step 102 can be specifically implemented through the following steps: select the text segment to be processed according to the sequence of the text segments, and input the text segment to be processed and the first prompt word into the large language model; the large language model outputs a segment summary corresponding to the text segment to be processed; wherein the first prompt word is at least used to instruct the large language model to extract the content title of the text segment to be processed, refine the key points of the text segment to be processed, and output the content title and the key points in the form of a list as the segment summary corresponding to the text segment to be processed.
[0035] Specifically, according to the order of each text segment in the conference text, the first text segment is selected as the text segment to be processed, and the text segment to be processed and the first prompt word are input into the large language model together. The large model extracts the content title of the text segment to be processed according to the instruction of the first prompt word, refines the key points of the text segment to be processed, and outputs the content title and key points in the form of a list as a segment summary corresponding to the text segment to be processed. After obtaining the segment summary corresponding to the first text segment, the second text segment is used as the text segment to be processed, and the text segment to be processed and the first prompt word are input into the large language model together to obtain the segment summary corresponding to the second text segment. Repeat the above steps until each text segment is input into the large language model, and obtain the segment summary corresponding to each text segment. Each segment summary is output in the form of a list of "content title + key points" according to the instruction of the first prompt word.
[0036] Step 103: De-duplicate the text of each segment summary to obtain a segment summary corresponding to each text segment.
[0037] Specifically, since the same topic may be discussed repeatedly during the meeting, there may be semantically repeated content after the meeting text is segmented. In this embodiment, after obtaining the segment summaries corresponding to each text segment, the segment summaries after deduplication are performed on each segment summary, and the segment summaries after deduplication are used as segment summaries to obtain the segment summaries corresponding to each text segment. It should be noted that the semantically repeated content described in this embodiment does not contradict the non-overlapping content of each text segment described in step 101. In order to avoid confusion, an example is given here. The meeting text A is divided into text segment 1, text segment 2, and text segment 3 in order. Among them, text segment 1 and text segment 2 are connected, which can be understood as the last sentence of text segment 1 and the first sentence of text segment 2 are two adjacent sentences in the meeting text, and text segment 1 and text segment 2 have no text overlap. However, it is entirely possible that there is semantic repetition between text segment 1 and text segment 2. Therefore, in this embodiment, deduplication of text for each segment summary refers to removing the text content corresponding to the part of each segment summary that has semantic repetition with other segment summaries.
[0038] In one example, the text of each segment summary is deduplicated to obtain the segment summary corresponding to each text segment, such as Figure 2 As shown, the following steps are included:
[0039] Step 1031: Divide all the segment summaries into multiple segment summary pairs in sequence according to the order of the text segments corresponding to each segment summary; wherein, in two adjacent segment summary pairs, the second segment summary in the first segment summary pair and the first segment summary in the second segment summary pair are the same segment summary.
[0040] Specifically, since each text segment has a sequence in the conference text, the segment summaries corresponding to each text segment also have a sequence corresponding to each text segment. All segment summaries are divided into multiple segment summary pairs according to the sequence of each segment summary; wherein, in two adjacent segment summary pairs, the second segment summary in the first segment summary pair and the first segment summary in the second segment summary pair are the same segment summary. For example, conference text A is divided into text segment 1, text segment 2, and text segment 3 in sequence, and after being processed by the large language model, segment summary 1 corresponding to text segment 1, segment summary 2 corresponding to text segment 2, and segment summary 3 corresponding to text segment 3 are obtained. By dividing segment summary 1, segment summary 2, and segment summary 3, two segment summary pairs can be obtained, the first segment summary pair contains segment summary 1 and segment summary 2, and the second segment summary pair contains segment summary 2 and segment summary 3. In other words, if N segment summaries are divided, N-1 segment summary pairs can be obtained, and in every two adjacent segment summary pairs, the second segment summary in the first segment summary pair and the first segment summary in the second segment summary pair are the same segment summary. The first and second fragment summaries in each fragment summary pair are determined by the order of the fragment summaries. In each fragment summary pair, the text fragment corresponding to the first fragment summary precedes the text fragment corresponding to the second fragment summary in the conference text.
[0041] Step 1032: Perform text similarity analysis on the first segment summary and the second segment summary in each segment summary pair in the order of the multiple segment summary pairs, remove the text in the second segment summary that is repeated with the first segment summary, and update the second segment summary.
[0042] Specifically, text similarity analysis is performed on the first fragment summary and the second fragment summary in each fragment summary pair in the order of multiple fragment summary pairs, text in the second fragment summary that is semantically repeated with the first fragment summary is removed, and the second fragment summary is updated.
[0043] Specifically, conference text A is divided into text segment 1, text segment 2, and text segment 3 in order. After being processed by the large language model, segment summary 1, segment summary 2, and segment summary 3 are obtained. After division, segment summary pair 1 and segment summary pair 2 are obtained. Segment summary pair 1 includes segment summary 1 (the first segment summary) and segment summary 2 (the second segment summary), and segment summary pair 2 includes segment summary 2 (the first segment summary) and segment summary 3 (the second segment summary). Since the time sequence of segment summary pair 1 in the conference text is overall earlier than that of segment summary pair 2, segment summary pair 1 is processed first. Text similarity analysis is performed on segment summary 1 and segment summary 2 in segment summary pair 1, and the text in segment summary 2 that is semantically repeated with segment summary 1 is removed to obtain an updated segment summary 2 (the second segment summary). At this time, fragment summary 2 (the first fragment summary) in fragment summary pair 2 is updated, and then text similarity analysis is performed on the updated fragment summary 2 and fragment summary 3 (the second fragment summary) in fragment summary pair 2, and the text in fragment summary 3 (the second fragment summary) that is semantically repeated with the updated fragment summary 2 is removed to obtain updated fragment summary 3 (the second fragment summary)... and so on, until the second fragment summary in each fragment summary pair is completely updated.
[0044] Step 1033: After the second segment summary in each segment summary pair is completely updated, the segment summary corresponding to each text segment is obtained.
[0045] Specifically, after the second segment summary in each segment summary pair is completely updated, each segment summary (except the first segment summary) is deduplicated once, and the deduplicated segment summary is used as the segment summary to obtain the segment summary corresponding to each text segment.
[0046] Step 104: Input each segment summary and the second prompt word into the large language model to obtain a full text summary corresponding to the conference text.
[0047] Specifically, after obtaining the fragment summaries corresponding to each text fragment, each fragment summary and a second prompt word are input into the large language model, and the large language model generates and outputs the full text summary corresponding to the conference text using each fragment summary according to the instruction of the second prompt word.
[0048] In one example, step 104 of inputting each fragment summary and the second prompt word into the large language model to obtain the full-text summary corresponding to the conference text specifically includes the following steps: inputting each fragment summary and the second prompt word into the large language model in sequence according to the order of each fragment summary; wherein the second prompt word is used to instruct the large language model to iteratively generate the full-text summary according to the input fragment summaries; and the large language model outputs the full-text summary corresponding to the conference text.
[0049] Specifically, since each text segment has a sequence in the conference text, the segment summaries corresponding to each text segment also have a sequence corresponding to each text segment. In the order of each segment summary, each segment summary and the second prompt word are input into the large language model in sequence. Specifically, at least one segment summary and the second prompt word are input each time, and the second prompt word instructs the large model to summarize and output at least one segment summary input, and then summarize and output the output with the next segment summary, and iteratively output the result.
[0050] In one example, the large language model iteratively generates a full-text summary based on the input fragment summaries, including: selecting at least two fragment summaries in the order of the fragment summaries to generate a first summary; generating a new first summary summary based on the first summary summary and the first fragment summary that is not selected in the fragment summaries to update the first summary summary; and using the new first summary summary generated after all the fragment summaries are selected as the full-text summary.
[0051] Specifically, the present embodiment generates a full-text summary by summarizing all the fragment summaries, and the process of generating the full-text summary is performed iteratively. For example, conference text A is divided into text fragment 1, text fragment 2, and text fragment 3 in order, corresponding to fragment summary 1, fragment summary 2, and fragment summary 3, respectively. Fragment summary 1 and fragment summary 2 are selected in the order of fragment summary 1, fragment summary 2, and fragment summary 3 and input into the large language model. The large language model summarizes fragment summary 1 and fragment summary 2 to obtain a first summary, and then fragment summary 3 is input into the large language model; the large language model summarizes the first summary and fragment summary 3 to generate a new first summary. At this point, all the fragment summaries of conference text A have been selected, and they are summarized with fragment summary 3 to generate a new first summary, which is then output as the full-text summary.
[0052] In another example, in actual application, after obtaining the full-text abstract corresponding to the conference text, this embodiment further includes: merging each segment abstract with the full-text abstract to form a conference abstract, and outputting it in a preset format.
[0053] Specifically, in order to let the user intuitively understand the key points of the meeting, each segment summary and the full text summary can be output to the user as the final meeting summary. The preset format of the meeting summary output can be: full text summary + segment summary, and this application does not make specific restrictions on this.
[0054] In order to better understand this solution, the overall process of this solution is summarized as follows: Figure 3 shown.
[0055] During a meeting, the audio / video content of the meeting can be obtained through an audio / video acquisition device, and the audio / video content of the meeting can be converted into a format supported by automatic speech recognition (ASR) and then automatically recognized to obtain a meeting text; the meeting text is segmented to obtain text segment 1, text segment 2...text segment n, and each text segment and a first prompt word are input into a large language model (LLM) to obtain a segment summary 1, a segment summary 2...segment summary n; each segment summary is deduplicated to obtain a segment summary 1, a segment summary 2...segment summary n, and the segment summary 1, the segment summary 2...segment summary n and a second prompt word are input into a large language model (LLM) to generate a full-text summary, and the segment summary 1, the segment summary 2...segment summary n and the full-text summary are output as the meeting summary.
[0056] Compared with the related art, this embodiment obtains the conference text and divides the conference text into multiple text fragments. According to the sequence of the text fragments, each text fragment is input into the large language model in sequence with the first prompt word to obtain a fragment summary corresponding to each text fragment; each fragment summary is deduplicated to obtain a fragment summary corresponding to each text fragment; each fragment summary and the second prompt word are input into the large language model to obtain a full-text summary corresponding to the conference text. The present invention uses a large language model to generate summaries. The model can deeply understand the conference text, accurately extract key information, avoid information omissions or misunderstandings that may be caused by traditional methods, and retain important details and discussion points in the meeting. The summary automatically generated by the large language model is closer to the original intention of the meeting, ensuring the accuracy and coherence of the content.
[0057] Another embodiment of the present invention relates to an electronic device, such as Figure 4 As shown, it includes at least one processor 202; and a memory 201 that is communicatively connected to the at least one processor 202; wherein the memory 201 stores instructions that can be executed by the at least one processor 202, and the instructions are executed by the at least one processor 202 so that the at least one processor 202 can execute any of the above method embodiments.
[0058] Among them, the memory 201 and the processor 202 are connected in a bus manner, and the bus may include any number of interconnected buses and bridges, and the bus connects one or more processors 202 and various circuits of the memory 201 together. The bus can also connect various other circuits such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art, and therefore, are not further described herein. The bus interface provides an interface between the bus and the transceiver. The transceiver can be one element or multiple elements, such as multiple receivers and transmitters, providing a unit for communicating with various other devices on a transmission medium. The data processed by the processor 202 is transmitted on a wireless medium through an antenna, and further, the antenna also receives data and transmits the data to the processor 202.
[0059] The processor 202 is responsible for managing the bus and general processing, and can also provide various functions, including timing, peripheral interfaces, voltage regulation, power management and other control functions. The memory 201 can be used to store data used by the processor 202 when performing operations.
[0060] Another embodiment of the present invention relates to a computer-readable storage medium storing a computer program, which implements any of the above method embodiments when executed by a processor.
[0061] That is, those skilled in the art can understand that all or part of the steps in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a program, and the program is stored in a storage medium, including several instructions to enable a device (which can be a single-chip microcomputer, chip, etc.) or a processor to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.
[0062] The present application also provides a computer program product, including computer instructions, which, when executed by a processor, implement the above Figure 1 The various processes of any of the conference summary generation method embodiments shown can achieve the same technical effect, and will not be described again here to avoid repetition.
[0063] Those skilled in the art will appreciate that the above-mentioned embodiments are specific examples for implementing the present invention, and in actual applications, various changes may be made thereto in form and detail without departing from the spirit and scope of the present invention.
Claims
1. A method for generating a conference summary, characterized in that: include: Acquire a conference text, and divide the conference text into a plurality of text segments, wherein the contents of the text segments do not overlap with each other; According to the order of the text fragments, the text fragments are input into the large language model together with the first prompt word in sequence to obtain a fragment summary corresponding to each text fragment; Deduplication is performed on each of the segment summaries to obtain a segment summary corresponding to each of the text segments; Each of the segment summaries and the second prompt word is input into the large language model to obtain a full-text summary corresponding to the conference text.
2. The method according to claim 1, characterized in that The obtaining of the conference text comprises: Acquire conference audio / video content through audio and video acquisition equipment; Converting the conference audio / video content into a preset format, performing automatic speech recognition on the conference audio / video content in the preset format, and obtaining the conference text; The preset format is a format supported by automatic speech recognition.
3. The method according to claim 1, characterized in that According to the order of the text segments, the text segments are sequentially input into the large language model together with the first prompt word to obtain a segment summary corresponding to each text segment, including: Selecting a text segment to be processed according to the order of the text segments, and inputting the text segment to be processed and the first prompt word into the large language model; The large language model outputs the segment summary corresponding to the text segment to be processed; Among them, the first prompt word is at least used to instruct the large language model to extract the content title of the text fragment to be processed, refine the key points of the text fragment to be processed, and output the content title and the key points in a list form as the fragment summary corresponding to the text fragment to be processed.
4. The method according to claim 3, characterized in that Deduplication of text is performed on each of the segment summaries to obtain a segment summary corresponding to each of the text segments, including: According to the order of the text segments corresponding to the segment summaries, all the segment summaries are divided into a plurality of segment summary pairs; wherein, in two adjacent segment summary pairs, the second segment summary in the first segment summary pair and the first segment summary in the second segment summary pair are the same segment summary; Performing text similarity analysis on the first and second fragment summaries in each fragment summary pair in order of the plurality of fragment summary pairs, removing text in the second fragment summary that is repeated with the first fragment summary, and updating the second fragment summary; After the second segment summary in each segment summary pair is completely updated, the segment summary corresponding to each text segment is obtained.
5. The method according to claim 4, characterized in that The step of inputting each of the segment summaries and the second prompt word into the large language model to obtain a full text summary corresponding to the conference text includes: Inputting each of the segment summaries and the second prompt word into the large language model in sequence according to the order of each of the segment summaries; wherein the second prompt word is used to instruct the large language model to iteratively generate the full text summary according to each of the input segment summaries; The large language model outputs the full-text summary corresponding to the conference text.
6. The method according to claim 5, characterized in that The large language model iteratively generates the full-text summary according to each of the input segment summaries, including: Selecting at least two of the fragment summaries according to the order of the fragment summaries to generate a first summary; generating a new first summary according to the first summary and the first unselected segment summary in each of the segment summaries to update the first summary; The new first abstract generated after all the segment abstracts are selected is summarized as the full-text abstract.
7. The method according to claim 1, characterized in that After obtaining the full text abstract corresponding to the conference text, the method further includes: merging each of the segment abstracts with the full text abstract to form a final conference abstract, and outputting the final conference abstract in a preset format.
8. An electronic device, characterized in that: include: at least one processor; as well as, a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the steps of the conference summary generation method according to any one of claims 1 to 7.
9. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the conference summary generation method according to any one of claims 1 to 7 are implemented.
10. A computer program product, characterized in that The method comprises computer instructions, which, when executed by a processor, implement the steps of the conference summary generation method according to any one of claims 1 to 7.