Online meeting processing method, electronic device, storage medium, and program product

Abstract

This application provides a method, electronic device, storage medium, and program product for processing online meetings. The method includes: real-time detection of the emotional state of each participant in an online meeting scenario and acquisition of meeting event information for the online meeting scenario; clustering the emotional states of each participant based on the occurrence time and type of their emotional state, obtaining clustering results, where each cluster includes multiple participants with the same type of emotional state; determining the cause of the same type of emotional state included in each cluster based on the meeting event information of the online meeting scenario; determining a feedback strategy for the same type of emotional state based on the cause; and executing the feedback strategy.

Description

Technical Field

[0001] This application relates to the technical field of video conferencing, and includes, but is not limited to, an online conferencing processing method, electronic device, storage medium, and program product. Background Technology

[0002] In related technologies, video streams of participants in online meeting scenarios can be acquired. By analyzing the participants' facial expressions in real time, the individual emotional state of the participants can be determined, and feedback can be provided to the speaker based on the individual emotional state and preset rules. However, feedback strategies determined solely based on individual emotional states have the disadvantages of low accuracy and practicality. Summary of the Invention

[0003] This application provides an online meeting processing method, electronic device, storage medium, and program product.

[0004] The technical solution of this application embodiment is implemented as follows: This application provides a method for processing online meetings, the method including: Real-time detection of the emotional state of each participant in an online meeting scenario, and acquisition of meeting event information for the online meeting scenario; Based on the time of occurrence of each participant's emotional state in the online meeting scenario and the type of each participant's emotional state, the emotional states of each participant are clustered to obtain clustering results. Each cluster in the clustering results includes the same type of emotional state of multiple participants. Based on the meeting event information of the online meeting scenario, determine the cause of the same type of emotional state included in each cluster; Based on the stated causes, determine feedback strategies for the same type of emotional state; Execute the aforementioned feedback strategy.

[0005] This application also provides an electronic device, which includes a processor and a memory for storing a computer program that can run on the processor; wherein the processor is used to run the computer program to perform any of the above-described online meeting processing methods.

[0006] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements any of the above-described online meeting processing methods.

[0007] This application also provides a computer program product, including a computer program or computer executable instructions, which, when executed by a processor, implement any of the above-described online meeting processing methods.

[0008] As can be seen, the embodiments of this application can cluster the emotional states of each participant to obtain clustering results. Since each cluster in the clustering results includes the same type of emotional state of multiple participants, it can be considered that each cluster in the clustering results can represent the same type of emotional state generated by multiple participants. That is, each cluster in the clustering results can represent the same type of emotional state event generated by the group. Therefore, the embodiments of this application can generate targeted feedback strategies for the emotional state of the group. Compared with the related technologies that only provide feedback to the speaker based on the individual emotional state and preset rules, the embodiments of this application can generate feedback strategies applicable to the emotional state of the group, which has the advantages of high accuracy and practicality. Attached Figure Description

[0009] Figure 1 This is a flowchart illustrating an online meeting processing method according to an embodiment of this application. Figure 2 This is a schematic diagram of the processing architecture for online meetings in an embodiment of this application; Figure 3 This is a flowchart of the clustering process proposed in the embodiments of this application; Figure 4 This is a flowchart for determining the most likely cause in the embodiments of this application; Figure 5 This is a flowchart illustrating the sorting of feedback strategies in an embodiment of this application; Figure 6 This is a schematic diagram of the speaker's interface in an embodiment of this application; Figure 7 This is a schematic diagram of a post-meeting analysis report according to an embodiment of this application; Figure 8 This is a schematic diagram of the structure of the online meeting processing device according to an embodiment of this application; Figure 9 This is a schematic diagram of the structure of an electronic device according to an embodiment of this application. Detailed Implementation

[0010] One proposed solution involves providing feedback to speakers in online meeting scenarios. This solution acquires video streams of participants in the online meeting, performs real-time facial expression analysis on these streams to determine individual participants' emotional states, and provides feedback to the speaker based on these emotional states and preset rules. For example, it could generate a prompt such as "Some participants appear confused." This solution primarily relies on a single video data stream, employs traditional machine learning algorithms to extract individual emotional features, and generates a judgment result through thresholding, thereby producing feedback information.

[0011] The solutions proposed by related technologies have at least the following problems: 1) Processing only single-modal data (such as video streams) cannot comprehensively utilize multi-source data such as audio, presentations, and interactive behaviors, resulting in insufficient comprehensiveness and accuracy of sentiment analysis.

[0012] 2) Focusing on individual emotion analysis while ignoring the characteristics of group responses makes it difficult to capture statistically significant collective emotional fluctuations.

[0013] 3) The correlation analysis between emotion and content is too simplistic, relying solely on timestamps to analyze the correlation between emotion and content. This fails to quantify the causal impact of meeting events or content on emotional responses, thus reducing the accuracy of the correlation analysis between emotion and content.

[0014] 4) The generation of feedback strategies relies on static rules or templates and cannot be dynamically generated according to the actual situation. The feedback strategies lack situational adaptability, which to some extent limits the accuracy and practicality of the feedback strategies.

[0015] To address the technical problems existing in related technologies, this application proposes a technical solution based on embodiments of the present application. This application embodiment can provide speakers with real-time, specific, and actionable feedback suggestions by deeply analyzing the collective emotional responses of participants in online meetings or online presentations and accurately associating these responses with multimodal presentation content. This application embodiment can solve the problems of vague, delayed, and impractical feedback mechanisms by establishing a causal inference model that represents the causal relationship between collective emotions and content events, thereby giving speakers stronger insight and the ability to adjust immediately.

[0016] The embodiments of this application will be further described in detail below with reference to the accompanying drawings and examples. It should be understood that the embodiments provided herein are merely illustrative of the embodiments of this application and are not intended to limit the embodiments of this application. Furthermore, the embodiments provided below are some embodiments for implementing this application, and not all embodiments for implementing this application. Unless otherwise specified, the technical solutions described in the embodiments of this application can be implemented in any combination.

[0017] It should be noted that, in the embodiments of this application, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a method or apparatus that includes a list of elements includes not only the elements expressly described, but also other elements not expressly listed, or elements inherent to implementing the method or apparatus. Without further limitations, an element defined by the phrase "comprising a..." does not exclude the presence of other related elements (e.g., steps in the method or units in the apparatus, such as portions of circuitry, processors, programs, or software, etc.) in the method or apparatus that includes that element.

[0018] The online meeting processing method provided in this application includes a series of steps, but the online meeting processing method provided in this application is not limited to the steps described. Similarly, the online meeting processing device provided in this application includes a series of modules, but the device provided in this application is not limited to the modules explicitly described, and may also include modules that need to be set up for obtaining relevant information or processing based on information.

[0019] Figure 1 This is a flowchart of an online meeting processing method according to an embodiment of this application, such as... Figure 1 As shown, the process includes: Step 101: Real-time detection of the emotional state of each participant in the online meeting scenario, and acquisition of meeting event information in the online meeting scenario.

[0020] In practical applications, multi-source heterogeneous data streams can be collected in online meeting scenarios. By analyzing these streams, the emotional state of each participant and meeting event information can be determined. (Refer to...) Figure 2 The multi-source heterogeneous data stream includes real-time video streams and real-time audio streams acquired through the input layer, as well as presentation or screen sharing data acquired through the input layer. Here, the real-time video stream and real-time audio stream are video and audio streams acquired in real time in the online meeting scenario, and the presentation or screen sharing data is data in the online meeting scenario. The format of the presentation or screen sharing data can be PPT, PDF, etc.

[0021] In some embodiments, to obtain the emotional state of each participant, a continuous video stream can be acquired using the front-facing camera of the participant's electronic device. Facial images are then captured in each frame of the video stream, and facial recognition and expression analysis techniques are used to analyze these images to determine each participant's emotional state. The principle of facial recognition and expression analysis is as follows: facial features are captured through a camera, and action units or emotion tags are extracted for individual emotion monitoring.

[0022] Reference Figure 2 The system can receive real-time video streams from the group emotion event detection module in the core processing module. The group emotion event detection module can use facial recognition and expression analysis algorithms to process each frame of the real-time video stream and extract feature vectors that can represent individual emotional states, such as the activation intensity, valence, and arousal of facial action units.

[0023] Understandably, any participant in an online meeting may have different emotional states at different times. Therefore, the emotional state of each participant in an online meeting may include multiple emotional states at different times.

[0024] Meeting event information in online meeting scenarios can include one or more of the following: meeting content information and meeting interaction behavior. Regarding the implementation of obtaining meeting event information in online meeting scenarios, in some embodiments, the real-time audio stream of the online meeting can be collected through audio devices such as microphones on the electronic devices used by participants, and presentation slides or screen-sharing data from the electronic devices used by participants can be obtained. The meeting content information can include one or more of the following: content obtained by analyzing the real-time audio stream, and content from the presentation slides or screen-sharing data. In some embodiments, meeting interaction behavior generated by participants while using electronic devices can also be obtained; here, participants can be speakers or participants other than presenters, and meeting interaction behavior can represent human-computer interaction operations related to the meeting, such as slide switching, mouse clicks, highlighting or zooming of control areas, etc.

[0025] In some embodiments, speech-to-text technology can be used to transcribe the audio stream of the online meeting into text in real time. Natural Language Processing (NLP) technology can be used to segment the text, perform syntactic analysis and extract keywords. Optical Character Recognition (OCR) technology can be used to parse the presentation content or screen sharing data of the online meeting and extract the text and image elements of the presentation or screen sharing data.

[0026] Reference Figure 2The core task of the multi-module content analysis module within the core processing module is to transform unstructured data into machine-readable meeting event information with precise timestamps. For example, it can convert the content of an unstructured presentation into machine-readable meeting event information with precise timestamps. Exemplarily, the multi-module content analysis module can perform the following operations: transcribing the speaker's audio stream into a time-stamped transcript in real time using speech-to-text technology; performing word segmentation, syntactic analysis, and key term extraction on the transcribed text using natural language processing technology; performing optical character recognition and layout analysis on each page of the presentation, extracting elements such as text, images, and charts, and quantifying their information density and visual complexity; and simultaneously recording the speaker's human-computer interaction with the presentation in real time, such as switching to the Nth slide at time t1 or highlighting a chart on the page at time t2. All this information is integrated into a series of discrete, time-stamped meeting event information. In an online meeting scenario, meeting event information can be denoted as C, where C = {(c1,t1,A1),(c2,t2,A2),...,(cn,tn,An)}, where, when i takes values ​​from 1 to n, Ci represents the identifier of the i-th meeting event information, ti represents the time when the i-th meeting event information is generated, and Ai represents the set of vectors describing the attributes of the i-th meeting event information.

[0027] As can be seen, the embodiments of this application can introduce a multimodal fusion mechanism into the data processing flow, which can significantly improve the comprehensiveness and accuracy of sentiment analysis by integrating video, audio and interactive behavior data, and improve data utilization compared with a single-modal system.

[0028] Step 102: Based on the time of occurrence of each participant's emotional state in the online meeting scenario and the type of each participant's emotional state, cluster the emotional states of each participant to obtain clustering results. Each cluster in the clustering results includes multiple participants with the same type of emotional state.

[0029] In this embodiment, individual emotional data is not analyzed in isolation or simply superimposed. Instead, a mechanism is designed and implemented to accurately identify and extract statistically significant "Group Emotional Events (GEEs)" from scattered, instantaneous individual emotional responses. This mechanism is the premise and foundation for subsequent analysis of the causes of emotional states. For example, performing this step can generate clustering results, where each cluster represents a group emotional event.

[0030] Step 103: Based on the meeting event information in the online meeting scenario, determine the cause of the same type of emotional state included in each cluster.

[0031] For example, refer to Figure 2 Steps 102 to 103 can be executed by the association attribution module within the core processing module. In practical applications, the group emotion event detection module can output the emotional state of each participant in the online meeting scenario to the association attribution module, and the multi-module content analysis module can output meeting event information of the online meeting scenario to the association attribution module. The association attribution module can execute step 102 based on the emotional state of each participant in the online meeting scenario to obtain the clustering result; the association attribution module can then execute step 103 based on the clustering result and the meeting event information of the online meeting scenario.

[0032] Step 104: Based on the cause, determine the feedback strategy for the same type of emotional state.

[0033] For example, refer to Figure 2 Step 104 can be executed by the strategy generation and presentation module within the core processing module. In practical applications, the attribution module can output the causes of the same type of emotional state included in each cluster to the strategy generation and presentation module.

[0034] Step 105: Implement the feedback strategy.

[0035] For example, step 105 can be performed by the output and interaction layer. (See reference...) Figure 2 The output and interaction layer can include a speaker's private view interface and a meeting data recording and analysis backend. After determining the feedback strategy, the strategy generation and presentation module can execute the feedback strategy by controlling the speaker's private view interface and the meeting data recording and analysis backend.

[0036] In practical applications, steps 101 to 104 can be implemented based on a processor and a communication interface. The processor can be at least one of the following: Application Specific Integrated Circuit (ASIC), Digital Signal Processor (DSP), Digital Signal Processing Device (DSPD), Programmable Logic Device (PLD), Field Programmable Gate Array (FPGA), Central Processing Unit (CPU), Controller, Microcontroller, and Microprocessor.

[0037] As can be seen, the embodiments of this application can cluster the emotional states of each participant to obtain clustering results. Since each cluster in the clustering results includes the same type of emotional state of multiple participants, it can be considered that each cluster in the clustering results can represent the same type of emotional state generated by multiple participants. That is, each cluster in the clustering results can represent the same type of emotional state event generated by the group. Therefore, the embodiments of this application can generate targeted feedback strategies for the emotional state of the group. Compared with the related technologies that only provide feedback to the speaker based on the individual emotional state and preset rules, the embodiments of this application can generate feedback strategies applicable to the emotional state of the group, which has the advantages of high accuracy and practicality.

[0038] In some embodiments, step 102 can be implemented by acquiring each emotion data point of each participant. Each emotion data point is used to characterize the attribute data of an emotion state, including the participant's identifier, the time of occurrence of the emotion state, and the type of the emotion state. After acquiring each emotion data point of each participant, the neighborhood of a first emotion data point in the online meeting scenario is determined. The first emotion data point represents any emotion data point among the various emotion data points of each participant. The neighborhood of the first emotion data point includes other emotion data points besides the first emotion data point. The other emotion data points have the same type of emotion state as the first emotion data point, and the time difference between the occurrence time of the emotion state in the other emotion data points and the occurrence time of the emotion state in the first emotion data point is less than or equal to a first threshold. After determining the neighborhood of each emotion data point in the online meeting scenario, the emotion states of each participant are clustered according to the neighborhood of each emotion data point to obtain the clustering result.

[0039] In this embodiment of the application, in order to achieve clustering of emotional states, Individual Emotional Reaction Points (IERPs) can be defined, and each IERP is an emotional data point described above. For example, each IERP is data generated based on the emotional state of a participant during the data collection and preprocessing stage, and each IERP can represent a change in the emotional state of a participant.

[0040] For example, each emotion data point can be represented as a triple Pi, Pi=(ui,ti,ei), where ui is the unique identifier (User ID) of the i-th participant, ti is the moment when the emotion state is captured, which can be represented by a precise timestamp, and ei is the type or category label representing the emotion state. For example, ei can be a predefined category label such as "positive", "negative" or "confused".

[0041] In practical applications, the group event detection module can continuously generate emotion data points, thus forming a dynamically growing IERP set. When clustering the emotional states of each participant, the distance is not calculated in the traditional multidimensional Euclidean space, but rather in a mixed-dimensional space composed of time, identity, and emotion type.

[0042] The first threshold can be set according to actual needs. For example, the first threshold can be a preset time neighborhood radius. For any data point Pi in the IERP set, its neighborhood is defined as the set of all data points Pj=(uj,tj,ej) that satisfy the following two conditions: First, temporal proximity, the absolute value of the difference between the timestamps of Pj and Pi is not greater than a preset threshold. t (i.e., |ti) tj∣≤ First, the clustering of emotion data points ensures that they are highly concentrated in the temporal dimension. Second, the consistency of emotion types, i.e., ei=ej, guarantees that the clustering results are consistent and clear in the dimension of emotion type, where i and j represent different positive integers. Of course, if an emotion data point does not satisfy the temporal proximity or the consistency of emotion type, then the emotion data point is not in the neighborhood of data point Pi.

[0043] As can be seen, when the attribute data of emotional state includes the participant's identifier, the time of occurrence of the emotional state, and the type of the emotional state, the embodiments of this application can accurately represent the same type of emotional state that occurs within a short period of time through the neighborhood of an emotional data point from the perspective of the time of occurrence of the emotional state and the type of the emotional state. Thus, based on the neighborhood of each emotional data point, the emotional states of each participant can be reasonably and accurately clustered from the time dimension and the type dimension of the emotional state.

[0044] For the implementation of clustering the emotional states of each participant based on the neighborhood of each emotional data point, in some embodiments, if the number of different participants corresponding to the neighborhood of the first emotional data point is greater than or equal to a second threshold, the first emotional data point can be determined as the core emotional data point, and the neighborhood of the core emotional data point can be added to a cluster. For each cluster, the operation of adding emotional data points to the cluster is repeated until no more emotional data points can be added to the cluster. The operation of adding emotional data points to the cluster includes: when the emotional data points in the neighborhood of the first core emotional data point are not assigned to any cluster, the emotional data points in the neighborhood of the first core emotional data point are added to the cluster. The first core emotional data point represents any core emotional data point existing in the cluster at the current time.

[0045] Here, the second threshold can be set according to actual needs. For example, the second threshold can be a preset minimum response group size, MinUsers. For any sentiment data point Pi in the IERP set, it can be determined whether it is a core sentiment data point. The determination of a core sentiment data point is not only based on the number of surrounding data points, but also on the number of responses from different participants that it can "density-reach" in the time-emotion dimension. After determining the neighborhood of sentiment data point Pi, it is possible to further count how many different participant identifiers are contained in the neighborhood of data point Pi. If the number of different participant identifiers in the neighborhood of data point Pi is greater than or equal to the preset minimum response group size, MinUsers (|{uj|Pj∈Neighborhood(Pi)}|≥MinUsers), then point Pi is determined to be a core sentiment data point, where Neighborhood(Pi) represents the neighborhood of sentiment data point Pi.

[0046] Once a core sentiment data point Pi is identified, it serves as the "seed" for constructing a new cluster. Initially, this cluster contains Pi and all data points within its neighborhood. Subsequently, each member of this newly added cluster is checked. If any member is also a core sentiment data point, then all points within that member's neighborhood (if not yet assigned to any cluster) are also added to the current cluster. This process continues iteratively or recursively until all reaction points reachable from the core point density are merged into the same cluster. Each cluster represents a complete set of sentiment data points that can represent a collective emotional fluctuation. In the entire dataset consisting of all IERPs, IERPs that are not covered by the neighborhood of any core sentiment data point are considered random, unrepresentative "noise points" and are ignored in subsequent analysis. Finally, each successfully constructed cluster is abstractly encapsulated into a Group Emotional Event (GEE) object, i.e., GEE=(Te,E,G). In this process, the emotion type E is directly derived from the common emotion label ei shared by all members within the cluster; the event timestamp Te is defined as the weighted center or median of the timestamps ti of all data points within the cluster, to most accurately represent the moment of the group's emotional outburst; and the group profile G is the set of identifiers ui of all participants in the cluster, clearly defining the specific personnel involved in this emotional event. Through the above process, this embodiment of the application can robustly and efficiently extract structured group emotional events that are of significant indicative importance to the content of a speech from massive, fragmented individual data, providing high-quality input for achieving accurate causal attribution.

[0047] For example, refer to Figure 3After detecting the emotional state of each participant in a real-time online meeting scenario, the system iterates through the emotional data points. When the currently iterated emotional data point is represented as Pi, it counts the number M of different participants corresponding to Pi's neighborhood. It then checks if M is greater than or equal to MinUsers. If M is less than MinUsers, Pi is marked as a noise point and skipped, and the process of iterating through the emotional data points continues. If M is greater than or equal to MinUsers, Pi is identified as the core emotional data point, and clusters are constructed based on these core emotional data points. Finally, each constructed cluster is encapsulated as a GEE object.

[0048] It can be seen that when the number of different participants corresponding to the neighborhood of the first emotion data point is greater than or equal to the second threshold, the emotion data points in the neighborhood of the first emotion data point can be considered to represent the common emotion state generated by a large number of participants. The common emotion state generated by a large number of participants is a representative emotion state in the online meeting scenario. Therefore, based on the common emotion state generated by a large number of participants, the emotion state can be clustered more reasonably.

[0049] Regarding the implementation of step 103, in some embodiments, the meeting event information to be analyzed can be determined from the meeting event information in the online meeting scenario based on the generation time of the same type of emotional state included in each cluster; the generation cause of the same type of emotional state included in each cluster can be determined based on one or more of the following: the novelty of the content of the meeting event information to be analyzed, the complexity of the content of the meeting event information to be analyzed, and the intensity of attention guidance brought about by the interactive behavior represented by the meeting event information to be analyzed.

[0050] For example, the meeting event information to be analyzed can be determined from the meeting event information in an online meeting scenario by using the weighted center or median of the occurrence time of the same type of emotional state included in each cluster.

[0051] For example, when the meeting event information to be analyzed includes the i-th meeting event information, the novelty of the content of the i-th meeting event information can be represented by f(Ni), the complexity of the content of the i-th meeting event information can be represented by f(Ci), and the intensity of attention guidance brought about by the interactive behavior represented by the i-th meeting event information can be represented by f(Ii).

[0052] f(Ni) is a quantitative function representing the novelty of the content. f(Ni) can be used to assess the novelty of information introduced in meeting content or interactions. For example, if the information introduced in the i-th meeting event is a technical term appearing for the first time in this meeting, the value of f(Ni) will be large; conversely, if the information introduced in the i-th meeting event is a recurring concept, the value of f(Ni) will be small. The value of f(Ni) can be obtained by comparing it with historical data of past meeting content or interactions.

[0053] f(Ci) is a quantification function representing content complexity, which assesses the cognitive load required to understand meeting content or interactive behaviors from multiple dimensions. For example, for text content, it can be measured by syntactic tree depth and terminology density; for Neri of presentation pages, it can be measured by information entropy, the number of visual elements, and the clutter of their arrangement.

[0054] f(Ii) is a quantification function representing interaction salience, which assesses the strength of attentional guidance brought about by interactive behaviors in a meeting. For example, a slide change, zooming in or highlighting a key area will result in a higher interaction salience score than a silent state without interaction.

[0055] Understandably, in online meeting scenarios, emotional states are usually triggered by meeting event information in the short period of time preceding their occurrence. Therefore, based on the occurrence time of the same type of emotional state included in each cluster, it is beneficial to reasonably determine the meeting event information to be analyzed in the online meeting scenario, and to accurately determine the cause of the same type of emotional state by analyzing the meeting event information to be analyzed.

[0056] In some embodiments, the time period during which the same type of emotional state is generated in each cluster can be used to determine the time period during which the meeting event information leading to the same type of emotional state occurs. In the meeting event information of the online meeting scenario, the meeting event information to be analyzed is determined based on the time period during which the same type of emotional state is generated in each cluster.

[0057] For example, the end time of the occurrence period is the weighted center or median of the occurrence times of the same type of emotional state. The duration of the occurrence period can be set according to actual needs; for example, the duration of the occurrence period can be denoted as... , The value ranges from 3 to 5 seconds. When the weighted center or median of the occurrence time of the same type of emotional state is Te, the occurrence time period is defined as a time interval. .

[0058] In practical applications, the attribution module can be triggered upon detecting a GEE object. Instead of matching emotional states with meeting content or interactions at the same time, the attribution module determines the time period of meeting events that lead to the same type of emotional state based on a "causal time window" hypothesis. This "causal time window" hypothesis states that a participant's emotional state is typically triggered by meeting content or interactions occurring a short period prior to the emotional state. In this embodiment, a time interval can be determined. Then, the time interval Information on all meeting events that occurred during the meeting was used as a candidate reason.

[0059] To identify the most likely trigger from numerous subsequent event information, this application proposes a quantitative evaluation model based on Causal Association Strength (CAS). For the i-th meeting event information, the quantitative evaluation model can calculate the CAS score based on Ai, and the formula for calculating the CAS score is formula (1).

[0060] CAS(ci)=wN f(Ni)+wC f(Ci)+wI f(Ii) (1) Here, wN, wC, and wI represent the weighting coefficients of f(Ni), f(Ci), and f(Ii), respectively. wN, wC, and wI can be preset weighting coefficients or dynamically adjusted weighting coefficients. wN, wC, and wI respectively represent the importance of content novelty, content complexity, and meeting interaction in evoking specific emotional responses. The values ​​of wN, wC, and wI can be configured according to the meeting type or speaker preferences.

[0061] After calculating the CAS score of all meeting event information in the meeting event information to be analyzed, the meeting event information with the highest CAS score (cmpc) can be identified as the cause of the same type of emotional state. That is, the meeting event information with the highest CAS score is identified as the most probable cause (MPC) of the Gee object. For example, refer to Figure 4 The process for determining the most likely cause of a group emotional event in this embodiment includes: after detecting a Gee object, defining a time interval. The process involves identifying the meeting event information to be analyzed, calculating the CAS score for each meeting event information, determining the most likely cause of the group emotional event, and outputting the most likely cause.

[0062] Understandably, in online meeting scenarios, emotional states are usually triggered by meeting event information in the short period of time preceding their occurrence. Therefore, based on the occurrence time of the same type of emotional state included in each cluster, it is helpful to reasonably determine the time period of the meeting event information that led to the occurrence of the same type of emotional state, thereby helping to accurately determine the meeting event information to be analyzed based on the time period.

[0063] In some embodiments, for the implementation of step 104, feedback strategies that meet the triggering conditions can be selected from a variety of feedback strategies. The triggering conditions include at least the following: the feedback strategy matches the above-mentioned cause. Then, at least based on the recommended rating of the feedback strategies that meet the triggering conditions, a feedback strategy for the same type of emotional state is determined.

[0064] In this embodiment, the various feedback strategies can be feedback strategies from a strategy library. The strategy library is not a simple collection of "if-then" rules, but a structured, object-oriented library. Each entry in the strategy library is a "multi-attribute strategy object," and each feedback strategy in the strategy library has rich attribute definitions. These attributes make the matching process far more intelligent and accurate than keyword search.

[0065] For example, the attributes of the feedback strategy in the strategy include MPC Trigger, Intervention Type, Content Payload, Interruption Index, Estimated Effectiveness Score, and Applicability Context. The MPC Trigger defines the type of MPC and / or specific content keywords that can activate this feedback strategy. For example, the types of MPCs that activate the feedback strategy include one or more of the following: TECHNICAL_TERM, COMPLEX_DIAGRAM, DATA_TABLE, and FAST_PACE. The MPC Trigger can be used to determine whether the feedback strategy matches the aforementioned causes.

[0066] Intervention type indicates the action classification of the feedback strategy. For example, the action classification of the feedback strategy can include any of the following: sample form presentation (DISPLAY_ANNOTATION), initiating a vote (LAUNCH_POLL), video playback (PLAY_VIDEO), and changing the description method (SUGGEST_REPHRASING).

[0067] Content payload represents the specific content required to execute the feedback strategy. It can be static text such as a standard definition of a term, or a content template such as a voting question. For example, the content payload could be a question like "How well do you understand the concept '{MPC_Content}'?", where {MPC_Content} will be populated in real time.

[0068] The interruption index is a quantifiable value used to assess the degree to which a feedback strategy disrupts the meeting flow (or evolution process). For example, the interruption index ranges from 1 to 10. The interruption index for several feedback strategies is illustrated below. The interruption index for the feedback strategy "Display TIPS annotations" might be 2 (low), while the interruption index for the feedback strategy "Play 30-second video" might be 8 (high). This application embodiment uses the interruption index to take into account the negative impact of feedback strategies.

[0069] The expected effectiveness score is a rating of the feedback strategy based on historical data or expert annotations, used to predict the effectiveness of the feedback strategy for a specific MPC type. For example, for an MPC of type TECHNICAL_TERM, "Show Comments" may have a higher effectiveness score than "Play General Introductory Video".

[0070] The applicability context defines the optimal conditions for using a feedback strategy. For example, the optimal conditions for using a feedback strategy may include one or more of the following: more than 5 minutes remaining, the audience profile is that of experts or novices, and the current session is a Q&A or the introduction of a new concept. By introducing the optimal conditions for using feedback strategies, the recommendation of strategies becomes intelligent and context-appropriate.

[0071] In this embodiment, the policy triggering conditions can be set according to actual needs. For example, the policy triggering conditions further include: the applicability context corresponding to the feedback policy matches the current meeting state of the online meeting scenario. For instance, the current meeting state of the online meeting scenario may include one or more of the following: remaining meeting time, participant profiles, and the current meeting stage. In practical applications, a matching score ContextFit(S) between the feedback policy and its corresponding applicability context can also be calculated based on preset rules. ContextFit(S) is a comprehensive score that considers applicability conditions such as remaining meeting time and participant profiles.

[0072] In practical applications, when an MPC is input into the strategy generation and presentation module, the strategy generation and presentation module does not simply search for keywords. Instead, it filters out feedback strategies that match the applicable context and the current meeting state of the online meeting scenario from all feedback strategies that meet the above-mentioned generation reasons. The filtered feedback strategies form an initial candidate set, and the feedback strategies in the initial candidate set are the feedback strategies that meet the strategy triggering conditions.

[0073] It can be seen that the feedback strategies that meet the triggering conditions are the feedback strategies that match the causes mentioned above. Therefore, based on the recommended scores of the feedback strategies that meet the triggering conditions, the feedback strategies for the same type of emotional state can be reasonably determined.

[0074] To determine the implementation method of the feedback strategy for the same type of emotional state based on the recommended score of the feedback strategy that meets the policy triggering conditions, in some embodiments, the same type of emotional state, the cause of the same type of emotional state, and the meeting event information of the online meeting scenario can be processed according to the pre-trained policy generation model to obtain the target feedback strategy; and the feedback strategy for the same type of emotional state can be determined based on the recommended score of the feedback strategy that meets the policy triggering conditions and the target feedback strategy.

[0075] To address the issue that the strategy library cannot cover all novel or complex situations, in some embodiments, a pre-trained strategy generation model can be used to generate target feedback strategies. For example, the strategy generation model is a Pedagogical Strategy Generative Model (PSGM), which is typically a large language model finely tuned with data from the teaching domain (such as lesson scripts from excellent teachers and knowledge of educational psychology).

[0076] Regarding the activation conditions of the policy generation model, in some embodiments, the scores of feedback policies that meet the policy triggering conditions and are selected from the policy library can be determined based on a pre-trained scoring model. When the scores of all feedback policies selected from the policy library are less than a preset scoring threshold, it indicates that the policy library cannot provide high-quality feedback policies. At this time, the policy generation model can be activated.

[0077] In some embodiments, the activation conditions for the strategy generation model can be determined by decision routing to determine the novelty and complexity of the meeting event information to be analyzed. Based on the novelty and complexity of the meeting event information to be analyzed, and according to preset judgment rules, it is determined whether a more creative feedback strategy needs to be used. When it is determined that a more creative feedback strategy needs to be used, the strategy generation model is activated.

[0078] The workflow of the strategy generation model includes two parts: constructing the input information for the strategy generation model and dynamically generating feedback strategies. The process of constructing the input information for the strategy generation model includes: generating a prompt by structuring the current meeting context, and then inputting the prompt into the strategy generation model. The input information for the strategy generation model is relatively rich, and may include: similar emotional states, the reasons for these emotional states, and meeting event information in the online meeting scenario; for example, the input information for the strategy generation model includes: detailed information about the MPC (content, type), quantified audience confusion level, the meeting presentation topic, the core content of the current slide, the speaker's real-time speech transcription, and a pre-defined participant profile.

[0079] The process of dynamically generating feedback strategies includes: dynamically generating entirely new, highly customized feedback strategies based on input information. These strategies generated by the strategy generation model are strategies that do not exist in the strategy library. For example, for MPC "Tensor Decomposition", the strategy generation model may not generate a standard definition, but instead create a more inspiring analogy. For example, the strategy output by the strategy generation model might be: "Suggested analogy: 'Tensor decomposition' can be interpreted as 'the process of breaking down a complex multidimensional Rubik's Cube (data) into several simple strips (core elements)'", or generating a more guiding question: "Suggested question: 'We can think of the matrix we just saw as a two-dimensional table. So, how do you think we would represent the data if it had a third dimension, such as time?' This introduces the concept of tensors."

[0080] It can be seen that the target feedback strategy is a feedback strategy obtained by using a pre-trained strategy generation model based on the same type of emotional state, the cause of the same type of emotional state, and the meeting event information in the online meeting scenario. Therefore, the target feedback strategy may contain rationalized feedback strategies. Thus, by comprehensively considering the feedback strategies that meet the policy triggering conditions and the target feedback strategy, the feedback strategy for the same type of emotional state can be determined more rationally.

[0081] In this embodiment, by combining precise matching based on a policy library and dynamic generation based on context awareness, and using a multi-dimensional ranking engine, the feedback strategy presented to the speaker is ensured to be the optimal solution for the current meeting context. For example, refer to... Figure 5 The workflow of the strategy generation and presentation module can include: generating feedback strategies that meet the strategy triggering conditions through the strategy library; generating target feedback strategies through the strategy generation model; and sorting the feedback strategies that meet the strategy triggering conditions and the target feedback strategies through a multi-dimensional sorting engine.

[0082] In this embodiment, both feedback strategies from the strategy library and target feedback strategies generated by the strategy generation model are input into the multidimensional ranking engine. Simply listing all feedback strategies would place a decision-making burden on the speaker. To address this issue, in this embodiment, the multidimensional ranking engine can achieve the following goal: within seconds, present the speaker with a controllable (e.g., 3 feedback strategies) and optimized list of feedback strategies.

[0083] For example, the multidimensional ranking engine can use a multi-objective optimization function for ranking, and the multidimensional ranking engine is used to calculate the final recommendation score Score(S) for each feedback strategy. The formula for calculating Score(S) is Equation (2).

[0084] Score(S) = Effectiveness(S) Interruption(S)+ ContextFit(S) (2) Effectiveness(S) is the expected effectiveness score of the feedback strategy. For strategies generated by the strategy generation model, this score is quickly predicted by another small evaluation model. Interruption(S) is the interruption index of the feedback strategy. Interruption(S) can be a normalized value. For example, a higher value of Interruption(S) means a higher degree of interruption in the feedback strategy, and its recommendation score is more penalized. , and It's a weighting factor that can be preset by the speaker (for example, a speaker prioritizing "fluency" can increase it). The weights can be adjusted adaptively based on the meeting context.

[0085] The multidimensional ranking engine calculates the final recommended scores of the feedback strategies that meet the policy triggering conditions and the target feedback strategies. It can sort the strategies in descending order based on the final recommended scores, select the top N feedback strategies from the ranking results, and push the selected top N feedback strategies along with quantified diagnostic information to the speaker's private interface. For example, N equals 3, and the quantified diagnostic information is: "Approximately 35% of attendees may be confused by the term 'tensor decomposition'", this percentage is mapped from the fluctuation range of the confusion index in the audience state vector.

[0086] The speaker's interface can display, for example, Figure 6 The list of feedback strategies shown is a small, intelligently sorted list that reduces the speaker's cognitive load, enabling them to make high-quality decisions quickly.

[0087] This application embodiment can also generate, after the meeting ends, a group emotional event that occurred in the online meeting scenario, the cause of the group emotional event, and the feedback strategy for the group emotional event, based on these events. Figure 7 The post-meeting analysis report is shown below.

[0088] In summary, this application proposes an intelligent meeting feedback scheme based on multi-source heterogeneous data streams, which can be applied to scenarios such as metaverse, virtual meetings, and video conferencing. In this application embodiment, through multimodal data fusion, group emotional event detection, causal correlation strength assessment, and dynamic feedback strategy generation, the interactivity and effectiveness of online meetings are significantly improved. This application embodiment has at least the following advantages: 1) Introducing a multimodal fusion mechanism into the data processing workflow significantly improves the comprehensiveness and accuracy of sentiment analysis by integrating video, audio, and interactive behavior data, thereby increasing data utilization compared to a single-modal system.

[0089] 2) A group emotional event detection algorithm is proposed. By performing emotional state clustering, statistically significant group emotional events are extracted from individual emotions, improving the robustness of emotion detection compared to individual emotion analysis methods of related technologies. Furthermore, a causal association strength model is introduced, which can quantify content novelty, complexity, and interaction saliency to accurately identify the triggers of emotional states, improving the accuracy of causal attribution compared to simple timestamp matching schemes.

[0090] 3) Design a strategy generation model and a multi-dimensional sorting engine to generate highly customized feedback strategies that balance effectiveness and low interruption, improving the adaptability and practicality of feedback strategies compared to fixed template solutions.

[0091] 4) By leveraging a strategy library and context-aware mechanisms, the applicability of feedback strategies can be further optimized, reducing the cognitive load on speakers and enhancing the efficiency of meeting interactions.

[0092] This application's embodiments elevate "intelligent feedback" from a concept to a feasible solution. Through the collaborative work of a structured strategy library, a strategy generation model, and a multi-dimensional ranking engine, it ensures that what is provided to the speaker is not just suggestions, but a well-thought-out optimal strategy that fits the current context.

[0093] While mainstream platforms in related technologies possess basic online meeting interaction functions, they remain at a relatively rudimentary stage in terms of in-depth audience state perception and real-time, intelligent speech strategy assistance. They generally lack the ability to transform discrete individual emotional states into structured group events, and also lack the closed-loop capability to further generate high-value feedback strategies. This application's embodiments construct an intelligent feedback strategy generation scheme through a group emotional event detection mechanism, a multimodal attribution model based on causal time windows, and a dual-track feedback system built from a strategy library and a strategy generation model.

[0094] Remote work, online learning, and online activities constitute a large and continuously growing market. The solutions in this application can serve the core needs of this market: improving communication efficiency and experience quality. The application scenarios of this application cover almost all online interactive meetings that rely on demonstrations and explanations, including daily corporate meetings, large industry summits, online higher education classrooms, professional skills training, and online product launches. For example, in online education scenarios, this application can help teachers gain real-time insight into students' points of confusion and adjust the pace and method of instruction immediately through recommended teaching feedback strategies, significantly improving teaching quality and completion rates. Furthermore, the multi-source heterogeneous data fusion, group state perception, and intelligent interaction technologies involved in this application are key technological supports for constructing a metaverse space rich in "life" and "social presence," ensuring that the real users behind the virtual avatars can obtain a more realistic and efficient communication experience.

[0095] Those skilled in the art will understand that, in the above-described method of the specific implementation, the order in which each step is written does not imply a strict execution order and does not constitute any limitation on the implementation process. The specific execution order of each step should be determined by its function and possible internal logic.

[0096] Based on the online meeting processing method proposed in the foregoing embodiments, this application also proposes an online meeting processing apparatus, such as... Figure 8 As shown, the device includes: The first processing module 801 is used to detect the emotional state of each participant in the online meeting scenario in real time and obtain the meeting event information of the online meeting scenario; The second processing module 802 is used to cluster the emotional states of each participant in the online meeting scenario based on the time of occurrence of each participant's emotional state and the type of each participant's emotional state, to obtain clustering results; to determine the cause of the same type of emotional state included in each cluster based on the meeting event information of the online meeting scenario; to determine a feedback strategy for the same type of emotional state based on the cause; and to execute the feedback strategy; wherein each cluster in the clustering results includes the same type of emotional state of multiple participants.

[0097] In some embodiments, the second processing module 802 clusters the emotional states of each participant based on the time of occurrence of each participant's emotional state in the online meeting scenario and the type of each participant's emotional state, obtaining clustering results, including: Each emotional data point of each participant is obtained. Each emotional data point is used to characterize the attribute data of an emotional state. The attribute data of the emotional state includes the participant's identifier, the time of occurrence of the emotional state, and the type of the emotional state. Determine the neighborhood of a first emotion data point in the online meeting scenario. The first emotion data point represents any emotion data point among the emotion data points of each participant. The neighborhood of the first emotion data point includes other emotion data points besides the first emotion data point. The other emotion data points have the same type of emotion state as the first emotion data point, and the time difference between the occurrence time of the emotion state in the other emotion data points and the occurrence time of the emotion state in the first emotion data point is less than or equal to a first threshold. Based on the neighborhood of each emotion data point, the emotional states of each participant are clustered to obtain the clustering results.

[0098] In some embodiments, the second processing module 802 clusters the emotional states of each participant based on the neighborhood of each emotional data point to obtain clustering results, including: If the number of different participants corresponding to the neighborhood of the first emotion data point is greater than or equal to the second threshold, the first emotion data point is determined as the core emotion data point, and the neighborhood of the core emotion data point is added to a cluster. For each cluster, the operation of adding emotion data points to the cluster is repeated until no more emotion data points can be added to the cluster. The operation of adding emotion data points to the cluster includes: when the emotion data points in the neighborhood of the first core emotion data point are not assigned to any cluster, the emotion data points in the neighborhood of the first core emotion data point are added to the cluster. The first core emotion data point represents any core emotion data point existing in the cluster at the current time.

[0099] In some embodiments, the second processing module 802 determines the cause of the same type of emotional state included in each cluster based on the meeting event information of the online meeting scenario, including: Based on the occurrence time of the same type of emotional state included in each cluster, the meeting event information to be analyzed is determined from the meeting event information in the online meeting scenario; The cause of the same type of emotional state included in each cluster is determined based on one or more of the following: the novelty of the content of the meeting event information to be analyzed, the complexity of the content of the meeting event information to be analyzed, and the intensity of attention guidance brought about by the interactive behavior represented by the meeting event information to be analyzed.

[0100] In some embodiments, the second processing module 802 determines the meeting event information to be analyzed based on the generation time of the same type of emotional state included in each cluster, from the meeting event information in the online meeting scenario, including: Based on the occurrence time of the same type of emotional state included in each cluster, determine the time period of the meeting event information that led to the occurrence of the same type of emotional state; In the meeting event information of the online meeting scenario, the meeting event information to be analyzed is determined according to the time period in which it occurs.

[0101] In some embodiments, the second processing module 802 determines a feedback strategy for the same type of emotional state based on the cause of the emotion, including: Among multiple feedback strategies, a feedback strategy that meets the strategy triggering conditions is selected, wherein the strategy triggering conditions include at least: the feedback strategy matches the cause of occurrence; Based at least on the recommended scores of the feedback strategies that meet the policy triggering conditions, a feedback strategy for the same type of emotional state is determined.

[0102] In some embodiments, the second processing module 802 determines a feedback strategy for the same type of emotional state based at least on the recommended score of the feedback strategy that meets the strategy triggering condition, including: Based on the pre-trained strategy generation model, the same type of emotional state, the cause of the same type of emotional state, and the meeting event information of the online meeting scenario are processed to obtain the target feedback strategy. Based on the feedback strategies that meet the policy triggering conditions and the recommended scores of the target feedback strategies, a feedback strategy for the same type of emotional state is determined.

[0103] In practical applications, the first processing module 801 and the second processing module 802 can be implemented based on a processor.

[0104] It should be noted that the description of the above device embodiments is similar to the description of the above method embodiments, and has similar beneficial effects. For technical details not disclosed in the device embodiments of this application, please refer to the description of the method embodiments of this application for understanding.

[0105] It should be noted that, in the embodiments of this application, if the above-described methods are implemented as software functional modules and sold or used as independent products, they can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiments of this application, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a terminal, server, etc.) to execute all or part of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), magnetic disks, or optical disks. Thus, the embodiments of this application are not limited to any specific hardware and software combination.

[0106] Correspondingly, this application embodiment further provides a computer program product, the computer program product including computer executable instructions, which are used to implement any of the online meeting processing methods provided in this application embodiment.

[0107] Accordingly, this application embodiment further provides a computer storage medium storing computer-executable instructions, which are used to implement any of the online meeting processing methods provided in the above embodiments.

[0108] This application also provides an electronic device. Figure 9 This is a schematic diagram of the composition structure of an electronic device provided in an embodiment of this application, as shown below. Figure 9 As shown, the electronic device 90 may include: Memory 901 is used to store executable instructions; The processor 902 is used to execute executable instructions stored in the memory 901 to implement any of the above-described online meeting processing methods.

[0109] The processor 902 mentioned above can be at least one of ASIC, DSP, DSPD, PLD, FPGA, CPU, controller, microcontroller, and microprocessor.

[0110] The aforementioned computer-readable storage medium and memory 901 may be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), a magnetic random access memory (FRAM), a flash memory, a magnetic surface memory, an optical disc, or a compact disc read-only memory (CD-ROM), etc.; or it may be various terminals that include one or any combination of the above-mentioned memories, such as mobile phones, computers, tablet devices, personal digital assistants, etc.

[0111] In some embodiments, the functions or modules of the apparatus provided in this application can be used to perform the methods described in the above method embodiments. The specific implementation can be referred to the description of the above method embodiments, and for the sake of brevity, it will not be repeated here.

[0112] The description of the various embodiments above tends to emphasize the differences between the various embodiments. The similarities or similarities between them can be referred to, and for the sake of brevity, they will not be repeated here.

[0113] The methods disclosed in the various method embodiments provided in this application can be arbitrarily combined to obtain new method embodiments without conflict.

[0114] The features disclosed in the various product embodiments provided in this application can be arbitrarily combined without conflict to obtain new product embodiments.

[0115] The features disclosed in the various method or device embodiments provided in this application can be arbitrarily combined without conflict to obtain new method or device embodiments.

[0116] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0117] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims. All of these forms are within the protection scope of this application.

Claims

1. A method for processing online meetings, characterized in that, The method includes: Real-time detection of the emotional state of each participant in an online meeting scenario, and acquisition of meeting event information for the online meeting scenario; Based on the time of occurrence of each participant's emotional state in the online meeting scenario and the type of each participant's emotional state, the emotional states of each participant are clustered to obtain clustering results. Each cluster in the clustering results includes the same type of emotional state of multiple participants. Based on the meeting event information of the online meeting scenario, determine the cause of the same type of emotional state included in each cluster; Based on the stated causes, determine feedback strategies for the same type of emotional state; Execute the aforementioned feedback strategy.

2. The method according to claim 1, characterized in that, The emotional states of each participant in the online meeting are clustered based on the time of occurrence and the type of their emotional state, resulting in clustering results, including: Each emotional data point of each participant is obtained. Each emotional data point is used to characterize the attribute data of an emotional state. The attribute data of the emotional state includes the participant's identifier, the time of occurrence of the emotional state, and the type of the emotional state. Determine the neighborhood of a first emotion data point in the online meeting scenario. The first emotion data point represents any emotion data point among the emotion data points of each participant. The neighborhood of the first emotion data point includes other emotion data points besides the first emotion data point. The other emotion data points have the same type of emotion state as the first emotion data point, and the time difference between the occurrence time of the emotion state in the other emotion data points and the occurrence time of the emotion state in the first emotion data point is less than or equal to a first threshold. Based on the neighborhood of each emotion data point, the emotional states of each participant are clustered to obtain the clustering results.

3. The method according to claim 2, characterized in that, The step of clustering the emotional states of each participant based on the neighborhood of each emotional data point to obtain clustering results includes: If the number of different participants corresponding to the neighborhood of the first emotion data point is greater than or equal to the second threshold, the first emotion data point is determined as the core emotion data point, and the neighborhood of the core emotion data point is added to a cluster. For each cluster, the operation of adding emotion data points to the cluster is repeated until no more emotion data points can be added to the cluster. The operation of adding emotion data points to the cluster includes: when the emotion data points in the neighborhood of the first core emotion data point are not assigned to any cluster, the emotion data points in the neighborhood of the first core emotion data point are added to the cluster. The first core emotion data point represents any core emotion data point existing in the cluster at the current time.

4. The method according to claim 1, characterized in that, The step of determining the cause of the same type of emotional state in each cluster based on the meeting event information of the online meeting scenario includes: Based on the occurrence time of the same type of emotional state included in each cluster, the meeting event information to be analyzed is determined from the meeting event information in the online meeting scenario; The cause of the same type of emotional state included in each cluster is determined based on one or more of the following: the novelty of the content of the meeting event information to be analyzed, the complexity of the content of the meeting event information to be analyzed, and the intensity of attention guidance brought about by the interactive behavior represented by the meeting event information to be analyzed.

5. The method according to claim 4, characterized in that, The step of determining the meeting event information to be analyzed based on the occurrence time of the same type of emotional state included in each cluster, in the meeting event information of the online meeting scenario, includes: Based on the occurrence time of the same type of emotional state included in each cluster, determine the time period of the meeting event information that led to the occurrence of the same type of emotional state; In the meeting event information of the online meeting scenario, the meeting event information to be analyzed is determined according to the time period in which it occurs.

6. The method according to claim 1, characterized in that, The step of determining a feedback strategy for the same type of emotional state based on the cause includes: Among multiple feedback strategies, a feedback strategy that meets the strategy triggering conditions is selected, wherein the strategy triggering conditions include at least: the feedback strategy matches the cause of occurrence; Based at least on the recommended scores of the feedback strategies that meet the policy triggering conditions, a feedback strategy for the same type of emotional state is determined.

7. The method according to claim 6, characterized in that, The step of determining a feedback strategy for the same type of emotional state based at least on the recommended rating of the feedback strategy that meets the strategy triggering conditions includes: Based on the pre-trained strategy generation model, the same type of emotional state, the cause of the same type of emotional state, and the meeting event information of the online meeting scenario are processed to obtain the target feedback strategy. Based on the feedback strategies that meet the policy triggering conditions and the recommended scores of the target feedback strategies, a feedback strategy for the same type of emotional state is determined.

8. An electronic device, characterized in that, The electronic device includes a processor and a memory for storing computer programs capable of running on the processor; wherein, The processor is used to run the computer program to perform the method according to any one of claims 1 to 7.

9. A computer storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the method described in any one of claims 1 to 7.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the method described in any one of claims 1 to 7.

Images