Intelligent conference information processing method, device and storage medium

Abstract

The present application discloses an intelligent conference information processing method, device and storage medium. The present application solution detects the status of the conference application, and when the application switches from the foreground to the background, triggers the transcription function, extracts key information, generates fragmented conference summary information and displays it to the user, so that even if the user cannot participate in the entire video conference, he or she can still know the core content of the conference during the missed period, which greatly improves the user's work efficiency, reduces the risk of missing important information, and enhances the user's experience.

Description

Technical Field

[0001] The present application relates to the field of intelligent information processing technology, and in particular to an intelligent conference information processing method, device and storage medium. Background Art

[0002] In existing video conferencing systems, users usually need to participate in the entire meeting to obtain complete meeting information. However, due to time, energy or other reasons, users may be interrupted during the meeting and unable to participate in the entire meeting, resulting in missing some important and key information.

[0003] Although some existing conference application systems provide recording and broadcasting functions, these functions usually only provide complete transcribed videos for users to play back, and do not provide targeted solutions for processing phased summaries or key information. Therefore, it is urgent to develop a system that can intelligently and automatically extract the corresponding key information and form a summary, so that users can quickly obtain the key content of the meeting during a certain period even if they are not in the meeting at a certain stage, so as to improve the user experience. Summary of the invention

[0004] Based on this, the present application proposes an intelligent conference processing solution to address the above-mentioned problem, so that even if the user is not in a meeting period, they can quickly obtain the meeting content of the missed period, thereby improving the user experience.

[0005] On the one hand, the present application provides an intelligent conference processing method, the method comprising:

[0006] Detect the status of the current conference application;

[0007] If the conference application is switched from the foreground to the background, the current conference information is transcribed into text information in real time;

[0008] extracting at least one key information of the text information at a preset time interval;

[0009] Detecting whether the conference application is switched to the foreground;

[0010] If yes, generating segment meeting summary information based on the key information;

[0011] The fragmented conference summary information is displayed to the user.

[0012] Specifically, the real-time transcription of the current conference information into text information includes: if the conference application is switched from the foreground to the background, triggering the start of a preset audio listener or API; capturing the conference audio stream in real time;

[0013] The audio stream is converted into text in real time through speech recognition algorithms.

[0014] Further, extracting at least one key information of the text information at a preset time interval includes:

[0015] Preprocess the text information; use the TF-IDF algorithm and the NPL algorithm to identify different topics or discussion points in the text; segment the preprocessed text content according to the topic or discussion point identification results; filter the segmented content that meets the requirements to form at least one key information; wherein one key information corresponds to one topic segment.

[0016] Furthermore, the generating of the fragment meeting summary information based on the key information includes:

[0017] Input each of the key information as a paragraph into a pre-trained text summary generation model to obtain the summary information;

[0018] Among them, the pre-trained text summary generation model includes:

[0019] Encode the input paragraph based on the BERT encoder and generate the corresponding context vector;

[0020] Aggregate the context vectors in each paragraph to generate a vector for each topic;

[0021] Calculate the weight of each topic through the linear layer and softmax function;

[0022] According to the topic weight and sentence content, the most representative sentences are selected as candidate sentences for the summary;

[0023] Merge the selected candidate sentences into a coherent summary text.

[0024] Furthermore, the method further comprises:

[0025] Detect whether a preset event occurs; if so, trigger the conference application to switch to the foreground, generate second conference summary information, the second summary information includes the event information; and display the second summary information to the user.

[0026] Specifically, the detecting whether a preset event occurs includes:

[0027] Analyze the voice or text content in the meeting, identify the user name or identifier based on NLP, and determine whether someone mentions the current user;

[0028] and / or,

[0029] Get the meeting schedule and current user role information, and based on the current meeting progress, predict whether the user's speech preset time has arrived.

[0030] Furthermore, the fragmented conference summary information is displayed to the user, including:

[0031] The summary information is displayed in a floating small window or card form.

[0032] A second aspect of the present application provides an intelligent conference information processing device, the device comprising:

[0033] The detection module is used to detect the status of the current conference application;

[0034] A transcription module, used to transcribe the current conference information into text information in real time if the conference application is switched from the foreground to the background;

[0035] An extraction module, used to extract at least one key information of the text information at a preset time interval;

[0036] A generating module, configured to generate fragment conference summary information based on the key information if it is detected that the conference application is switched to the foreground;

[0037] The display module is used to display the fragment conference summary information to the user.

[0038] The third aspect of the present application provides a computer-readable storage medium storing a computer program, which, when executed by a processor, enables the processor to perform the steps of any one of the above methods.

[0039] A fourth aspect of the present application provides a computer terminal device, comprising a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of any one of the above methods.

[0040] This application scheme detects the status of the conference application. When the application switches from the foreground to the background, it triggers the transcription function, extracts key information, generates fragmented conference summary information, and displays it to the user. Even if the user is unable to participate in the entire video conference, he or she can still know the core content of the meeting during the missed period, which greatly improves the user's work efficiency, reduces the risk of missing important information, and enhances the user experience.

[0041] Furthermore, the present application solution can generate an accurate, coherent and concise text summary of the characteristics of the meeting content, allowing users to quickly obtain the core content of the missed meeting; at the same time, the present application solution also sets a preset event reminder mechanism to prevent users from missing important events and nodes of the meeting. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] 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.

[0043] in:

[0044] Figure 1 A flow chart of an intelligent conference processing method in one embodiment;

[0045] Figure 2 It is a structural block diagram of an intelligent conference processing device in one embodiment;

[0046] Figure 3 FIG. 4 is a structural block diagram of a computer device in one embodiment. DETAILED DESCRIPTION

[0047] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not 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.

[0048] It should be noted that the terms "include", "comprises" and "have" and any variations thereof in the specification and claims of the present application and the above-mentioned drawings are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units that are not listed, or may optionally include other steps or units that are inherent to these processes, methods, products or devices. In the claims, specification and drawings of the present application, relational terms such as "first" and "second" are merely used to distinguish one entity / operation / object from another entity / operation / object, and do not necessarily require or imply any such actual relationship or order between these entities / operations / objects.

[0049] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0050] In one embodiment, if Figure 1 The figure is a flow chart of an intelligent conference processing method of the present application, the method comprising:

[0051] S10: Detect the status of the current conference application.

[0052] Specifically, when the conference application is started, the application can determine the current state of the conference application based on the current operating system environment (such as Android, iOS, Windows, etc.) and the development framework used. The status shown mainly includes whether the application is in the foreground or background, specifically including:

[0053] For the Android platform, you can register ActivityLifecycleCallbacks to listen to the lifecycle events of the application activity, or check the visibility of the current Activity to determine whether the application is in the foreground. For the iOS platform, you can use the applicationState property of UIApplication: this property provides the current state information of the application, such as UIApplicationStateActive indicates that the application is running in the foreground, and UIApplicationStateBackground indicates that the application is running in the background; or, by registering to listen for notifications, you can get notifications sent when the application enters the background and foreground. For the Windows system platform, you can use the Windows API to query the current active window or process status.

[0054] S11. If the conference application is switched from the foreground to the background, the current conference information is transcribed into text information in real time.

[0055] When the conference application is started, it is usually in the foreground. During the running process, there may be a response event that causes the application to switch to the background, such as a sudden voice call or a phone call that triggers a passive interruption, or the user suddenly has other urgent matters to deal with, which actively triggers the video application to switch to the background (such as minimizing or switching other pending applications to the foreground). In one embodiment, if the conference application is switched to the background, the microphone function in the video conference application is automatically turned off to prevent conflicts or crosstalk with other foreground applications.

[0056] In one embodiment, when it is detected that the conference application is switched from the foreground to the background, in order to prevent the user from missing important conference content during the period due to an emergency, the conference information transcription function is triggered. Specifically, the current conference information is transcribed into text information in real time, including:

[0057] Trigger the start of a preset audio listener or API to capture the conference audio stream in real time.

[0058] Pre-set an audio listener or API in the video conferencing APP. When the user switches the APP to the background, call to start the audio listener or API to continue capturing the audio stream.

[0059] The audio stream is converted into text in real time through a speech recognition algorithm. Specifically, the application pre-integrates a third-party real-time transcription service (such as Google Cloud Speech-to-Text, IBM Watson, etc.) or integrates an open source speech recognition library to convert the audio stream into text. The above services provide API interfaces that can be easily integrated into the APP.

[0060] S12. Extract at least one key information of the text information at a preset time interval.

[0061] In one embodiment, after the current conference audio is transcribed into text in real time, important information of the text is extracted at preset time intervals. The preset time interval can be fixed, such as 10s, 30s, 60s, etc., and can also be adjusted according to the amount of current transcribed text.

[0062] Specifically, extracting at least one key information of the text information at a preset time interval includes:

[0063] S121. Preprocess the text information.

[0064] Specifically, the initial transcribed text contains a lot of noise, which directly affects the subsequent recognition results. Therefore, the initial transcribed text needs to be preprocessed, including: using natural language processing technology (NLP) to clean, segment, and tag the transcribed text, remove irrelevant words, correct erroneous words and punctuation marks, and obtain preprocessed text.

[0065] S122. Identify different topics or discussion points in the text.

[0066] After preprocessing the initial transcription text, natural language processing technology is used to identify the preprocessed text and identify different topics or discussion points in the text.

[0067] Furthermore, the preprocessed text is used to identify different topics or discussion points in the text based on natural language processing technology. This includes:

[0068] The text is divided into individual words or phrases, and keywords or phrases in the text are extracted based on the TF-IDF method, or word embedding (such as Word2Vec, GloVe, etc.) is used to convert the text into a vector representation. The TF-IDF algorithm is used to calculate the weight of each word in the preprocessed text information, and keywords are extracted according to the weight sorting, and sentences or phrases containing keywords with weights greater than preset values ​​are extracted to extract useful key information. Clustering or classification algorithms are then used to identify different topics or discussion points in the text. For example, in one embodiment, the K-means clustering algorithm can be used to cluster the text according to topics, or classification algorithms such as support vector machines (SVM) and naive Bayes can be used to classify the text into different topic categories.

[0069] S123. Segment the preprocessed text content according to the identification results of the topic or discussion point.

[0070] Specifically, according to the above steps, the keywords of each topic are analyzed to determine their occurrence positions in the text. The density or frequency of keywords is calculated using the TF-IDF algorithm to identify the core area of ​​the topic, and then the semantic similarity between adjacent sentences is calculated using cosine similarity. If the similarity suddenly decreases, it is determined that the topic has changed. Based on the results of the above keyword analysis and sentence similarity evaluation, the topic boundaries in the text are determined. These boundaries can be between sentences, between paragraphs, or at a more fine-grained position. Finally, based on the detected topic boundaries, the text is divided into different paragraphs or sub-texts, and the segmentation results are output: the segmented text is output in a clear manner, such as marking the topic label of each paragraph or generating a new text file.

[0071] S123, screening the segmented contents that meet the requirements to form at least one key information text;

[0072] Specifically, after the text content is segmented, it is determined whether the size of each segmented text is greater than a preset threshold. If so, the segmented text is used as a key information, wherein one key information corresponds to one topic segment.

[0073] S13: Detect whether the conference application is switched to the foreground.

[0074] Specifically, after a conference application is switched to the background due to an emergency event, the user may return to the conference after the event is processed. Therefore, it is necessary to detect in real time whether the conference application is switched to the foreground.

[0075] S14: If yes, generate segment meeting summary information based on the key information.

[0076] Specifically, if the conference application is switched back to the foreground, that is, the user returns to the conference, the generation of the conference summary information during this period can be triggered. At the same time, in order to prevent the user from switching the application to the background due to an erroneous operation, or to prevent the conference application from switching to the background due to a short event, an embodiment of the present application also includes: judging the time difference between the application switching to the foreground and the previous switching to the background, if the time difference is less than a preset value, it is determined to be a false trigger operation, at which time, the summary information is not generated, and the summary information generation operation is performed only if the time difference is greater than or equal to the preset value.

[0077] Specifically, the generating the fragment meeting summary information based on the key information includes:

[0078] Each of the key information is input as a paragraph into a pre-trained text summary generation model to obtain the summary information.

[0079] The pre-trained text summary generation model includes: input layer, BERT encoder, weight distribution module, sentence selection module, summary generation module and post-processing module. Specifically, it includes:

[0080] Input layer: The input layer receives the input conference text data. In this application solution, the text is first divided into multiple topics or discussion points, each of which is used as an input paragraph of the model. The input layer further decomposes each paragraph into a sequence of words or subwords and converts them into vector representations through word embedding.

[0081] BERT encoder: Encode the input paragraph based on the BERT encoder and generate the corresponding context vector. The BERT model is a Transformer-based encoder model that captures context information in the text through a self-attention mechanism. This application generates the corresponding context vector representation by inputting the word or subword sequence converted by word embedding into the BERT encoder.

[0082] Weight allocation module: The weight allocation module is responsible for assigning weights to each topic, which reflects the importance of the topic in the meeting and its contribution to the subsequent summary generation. The weight allocation module in one implementation scheme of the present application mainly includes two parts: aggregation and weight calculation, specifically including:

[0083] Aggregation: Aggregate the context vectors in each paragraph to generate a vector for each topic. To get an overall representation of a topic, different aggregation strategies can be used, such as mean pooling, max pooling, or attention pooling, to aggregate the vectors of all words or subwords in the topic into a single vector.

[0084] Weight calculation: The weight of each topic is calculated through a linear layer and a softmax function. Specifically, the aggregated topic vector is transformed through a linear layer (fully connected layer), which can learn the mapping relationship between the topic vector and the weight. Then the softmax function is used for normalization: In order to ensure that the sum of the weights of all topics is 1, the softmax function is used to normalize the result after the linear transformation.

[0085] Sentence selection module: mainly selects the most representative sentences as candidate sentences for the summary based on the above-mentioned topic weights and sentence content. Specifically, each sentence is first encoded to obtain a vector representation of the sentence, and then the importance score of each sentence is calculated by combining the topic weight and sentence vector. The score calculation can comprehensively consider factors such as the length of the sentence, the frequency of keyword occurrence, and the relevance to the topic. Finally, sentences with scores higher than the threshold are selected as candidate sentences for subsequent summary generation.

[0086] Summary generation module: merges the selected candidate sentences into a coherent summary text. In one embodiment, a Transformer-based sequence-to-sequence model structure is adopted, in which the encoder part of the model receives the vector representation of the candidate sentence, generates a context vector containing the candidate sentence information through the self-attention mechanism and position encoding, and the decoder part of the model serves as the output layer, and gradually generates the summary text according to the context vector. The decoder uses the attention mechanism to focus on the candidate sentence information related to the current generation position to generate a more accurate and coherent summary.

[0087] Post-processing module: The post-processing module further optimizes and corrects the generated summary. It first uses sentence similarity calculation to remove redundant sentences in the summary to ensure the simplicity of the summary. Then, through text generation technology, the summary text is polished and corrected to improve the readability and fluency of the summary. This can include correcting grammatical errors, adjusting sentence structure, adding appropriate conjunctions, etc.

[0088] The above model of this application can capture the deep semantic information and contextual relationships in the conference text through the BERT encoder, so as to understand the text content more accurately. In addition, the model can automatically judge the importance of different topics in the meeting through the topic weight distribution module, so that the summary can highlight the key content more. And adopt a sentence selection mechanism: the sentence selector can select the most representative sentences as candidate sentences for the summary based on the topic weight and sentence content, ensuring the accuracy and information content of the summary.

[0089] In addition, the above-mentioned model of the present application uses conference text data with summary labels as training samples during the model training process. By defining a suitable loss function (such as cross entropy loss or ROUGE loss), the model is trained using optimization algorithms such as gradient descent. During the training process, the model adjusts parameters through the back propagation algorithm to minimize the loss function and improve the accuracy of summary generation. In the evaluation stage, an independent test set is used to evaluate the model. Evaluation indicators include ROUGE score, BLEU score, etc., which are used to measure the similarity and quality between the summary generated by the model and the real summary.

[0090] At the same time, in order to allow the weight allocation module to allocate weights more accurately, the present application pre-uses a large-scale public text summary dataset, and / or uses some abstracts, meeting minutes, etc. containing summary information as training data, so that there is not enough information in the training data to distinguish the importance of different topics.

[0091] S15: Display the fragmented conference summary information to the user.

[0092] In one embodiment, the present application solution adopts the method of displaying the summary information in a floating small window or card. For example, the summary information is displayed in the floating small window or card in the following manner:

[0093]

[0094] Furthermore, the floating small window or card can be set with one or more interactive function buttons to facilitate user closing and other operations.

[0095] The above-mentioned implementation scheme of the present application detects the status of the conference application. When the application switches from the foreground to the background, it triggers the transcription function, extracts key information, generates fragmented conference summary information, and displays it to the user. Even if the user cannot participate in the entire video conference, he or she can still know the core content of the meeting during the missed period, which greatly improves the user's work efficiency, reduces the risk of missing important information, and enhances the user experience.

[0096] In another embodiment, the method of the present application further comprises:

[0097] S16, detecting whether a preset event occurs.

[0098] When a key event occurs in a meeting, such as when a user is mentioned, when it is the user's turn to speak, or when the user needs to participate in a speech, the user hopes to be notified in time and quickly return to the meeting. Therefore, in one embodiment of the present application, an event trigger mechanism is designed to enable the video conferencing APP to intelligently perceive and respond to these key events when in the background. Specifically, the detection of whether a preset event occurs includes:

[0099] Analyze the voice or text content in the meeting, identify the user name or identifier based on NLP, and determine whether someone mentions the current user;

[0100] and / or,

[0101] Get the meeting schedule and current user role information, and based on the current meeting progress, predict whether the user's speech preset time has arrived.

[0102] S17: If yes, trigger the conference application to switch to the foreground, and generate second conference summary information, where the second summary information includes the event information.

[0103] If the above event is detected, it means that the current conference node now or will soon require the current user to participate, and the conference is actively triggered to switch to the foreground, and a second summary information including event information and summary information during the application switching to the background is generated. In one embodiment, the first summary information during the application switching to the background can be generated by the aforementioned summary generation method, and then the first summary information is merged with the description of the triggering event to form the second summary information.

[0104] S18: Display the second summary information to the user.

[0105] Similarly, the second summary information displayed to the user can be displayed in a floating small window or card. In one embodiment, the present application scheme also determines whether the triggering event is related to the topic. If so, the related topics are sorted and displayed in the front. For example, during the discussion of related topics, when a user is mentioned and speaks about a certain topic, the related topics are sorted and displayed in the front as follows:

[0106]

[0107] The above implementation scheme of the present application can not only generate an accurate, coherent and concise text summary of the characteristics of the meeting content, allowing users to quickly obtain the core content of the missed meeting, but also set up a preset event reminder mechanism to prevent users from missing important events and nodes of the meeting.

[0108] In one embodiment, Figure 2 As shown, the second aspect of the present application provides an application device, the device comprising:

[0109] The detection module is used to detect the status of the current conference application;

[0110] A transcription module, used to transcribe the current conference information into text information in real time if the conference application is switched from the foreground to the background;

[0111] An extraction module, used to extract at least one key information of the text information at a preset time interval;

[0112] A generating module, configured to generate fragment conference summary information based on the key information if it is detected that the conference application is switched to the foreground;

[0113] The display module is used to display the fragment conference summary information to the user.

[0114] In one embodiment, the present application provides a computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, the processor performs the following steps:

[0115] Detect the status of the current conference application;

[0116] If the conference application is switched from the foreground to the background, the current conference information is transcribed into text information in real time;

[0117] extracting at least one key information of the text information at a preset time interval;

[0118] Detecting whether the conference application is switched to the foreground;

[0119] If yes, generating segment meeting summary information based on the key information;

[0120] The fragmented conference summary information is displayed to the user.

[0121] In one embodiment, Figure 3 The present application provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor performs some steps:

[0122] Detect the status of the current conference application;

[0123] If the conference application is switched from the foreground to the background, the current conference information is transcribed into text information in real time;

[0124] extracting at least one key information of the text information at a preset time interval;

[0125] Detecting whether the conference application is switched to the foreground;

[0126] If yes, generating segment meeting summary information based on the key information;

[0127] The fragmented conference summary information is displayed to the user.

[0128] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the program can be stored in a non-volatile computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0129] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0130] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A method for processing intelligent conference information, characterized in that: The method comprises: Detect the status of the current conference application; If the conference application is switched from the foreground to the background, the current conference information is transcribed into text information in real time; extracting at least one key information of the text information at a preset time interval; Detecting whether the conference application is switched to the foreground; If yes, generating segment meeting summary information based on the key information; Detect whether a preset event occurs; if so, trigger the conference application to switch to the foreground, generate second summary information, the second summary information includes event information, and the second summary information also includes information obtained by integrating the first summary information with the description of the preset event; the first summary information is a fragment conference summary information; The second summary information is displayed to the user. The step of displaying the second summary information to the user includes displaying the second summary information in a floating window or card format.

2. The method according to claim 1, characterized in that The real-time transcription of the current meeting information into text information includes: If the conference application is switched from the foreground to the background, a preset audio listener or API is triggered to start; Capture conference audio streams in real time; The audio stream is converted into text in real time through speech recognition algorithms.

3. The method according to claim 1, characterized in that Extracting at least one key information of the text information at a preset time interval includes: Preprocess the text information; Use TF-IDF algorithm and NPL algorithm to identify different topics or discussion points in the text; According to the identification results of topics or discussion points, the preprocessed text content is segmented; Filter the segmented content that meets the requirements to form at least one key information; Among them, one key information corresponds to one topic segment.

4. The method according to claim 3, characterized in that The generating of the fragment meeting summary information based on the key information includes: Input each of the key information as a paragraph into a pre-trained text summary generation model to obtain the summary information; Among them, the pre-trained text summary generation model includes: Encode the input paragraph based on the BERT encoder and generate the corresponding context vector; Aggregate the context vectors in each paragraph to generate a vector for each topic; Calculate the weight of each topic through the linear layer and softmax function; According to the topic weight and sentence content, the most representative sentences are selected as candidate sentences for the summary; Merge the selected candidate sentences into a coherent summary text.

5. The method according to claim 1, characterized in that The detecting whether a preset event occurs includes: Analyze the voice or text content in the meeting, identify the user name or identifier based on NLP, and determine whether someone mentions the current user; and / or, Get the meeting schedule and current user role information, and based on the current meeting progress, predict whether the user's speech preset time has arrived.

6. An intelligent conference information processing device, characterized in that: The device comprises: The detection module is used to detect the status of the current conference application; A transcription module, used to transcribe the current conference information into text information in real time if the conference application is switched from the foreground to the background; An extraction module, used to extract at least one key information of the text information at a preset time interval; A generation module, configured to generate fragmented conference summary information based on the key information if it is detected that the conference application is switched to the foreground; detect whether a preset event occurs; if so, trigger the conference application to switch to the foreground, and generate second summary information, wherein the second summary information includes event information, and the second summary information also includes information obtained by fusing the first summary information with the description of the preset event; the first summary information is fragmented conference summary information; The display module is used to display the second summary information to the user, the summary information is the second summary information, and the step of displaying the second summary information to the user includes using a floating small window or a card form for floating display.

7. A computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, the processor is caused to perform the steps of the method according to any one of claims 1 to 5.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the method according to any one of claims 1 to 5.

Images