Method for dynamic allocation and management of multi-seat permissions of digital conference system server
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SHENZHEN HAIWEI HENGTAI INTELLIGENT TECH CO LTD
- Filing Date
- 2026-03-27
- Publication Date
- 2026-06-19
Smart Images
Figure CN122247973A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of computer technology, specifically relating to a method for dynamic allocation and management of multi-seat permissions in a digital conferencing system server. Background Technology
[0002] With the widespread adoption of remote work and hybrid meeting models, digital conferencing systems have become a core infrastructure supporting enterprise decision-making, collaborative innovation, and efficient communication. Modern digital conferencing platforms generally support multi-seat access, role differentiation, and access control. Their basic architecture typically includes modules for participant authentication, speaking rights allocation, content sharing, and operation authorization. However, current mainstream systems still employ static preset mechanisms for access management, requiring meeting organizers to manually configure fixed permission levels for each seat before the meeting, such as "host," "speaker," and "observer," lacking the ability to proactively model the dynamic progress of the meeting and group interaction behavior.
[0003] Dynamic allocation of multi-seat permissions is a key element in improving meeting efficiency. It aims to adjust the operational permissions and interactive capabilities of each seat in real time based on meeting objectives, agenda stages, and participant characteristics. Ideally, the mechanism should balance speaking opportunities, guide discussion focus, suppress redundant interference, and promote high-quality decision output. However, existing technologies struggle to quantitatively evaluate the actual effectiveness of different permission configuration schemes before the meeting begins. This forces organizers to rely on subjective experience or past practices, making it impossible to predict whether specific strategies will lead to unfocused discussions, suppression of key opinions, or inefficient decision-making.
[0004] Existing access control systems generally suffer from the following problems: a lack of modeling capabilities for participants' historical behavior, professional background, and interaction tendencies, making it impossible to construct a virtual simulation environment that closely resembles real-world interaction logic; the absence of efficient policy search and optimization algorithms, hindering the rapid identification of high-performance configuration schemes within a vast space of permission combinations; and the lack of a dynamic feedback mechanism covering the entire meeting process, making it impossible to generate composite access control policies that accommodate both initial settings and mid-meeting adjustments. These problems are particularly pronounced in highly complex formal meeting scenarios involving multiple stakeholders, easily leading to agenda loss, difficulty in reaching consensus, or resource waste. There is an urgent need for an access control policy simulation and optimization method that integrates digital twins and intelligent search to achieve data-driven scientific decision support. Summary of the Invention
[0005] This invention provides a method for dynamic allocation and management of multi-seat permissions in a digital conferencing system server. It aims to address the technical problem that meeting organizers cannot predict the impact of different permission allocation schemes on meeting efficiency, participation, and decision-making quality before the meeting, and can only rely on experience to set permissions. The method constructs a virtual simulation model driven by historical meeting data to quantitatively evaluate and prioritize multiple candidate permission allocation schemes before the meeting officially begins. This provides meeting organizers with interpretable and verifiable permission configuration suggestions. Furthermore, during the meeting, the method dynamically adjusts the operation permissions of each seat based on real-time interactive behavior, achieving adaptive coordination between permission policies and meeting progress status.
[0006] This invention provides a method for dynamic allocation and management of multi-seat permissions in a digital conferencing system server, including: Obtain a dataset of historical similar meetings for the target meeting; A meeting effectiveness prediction model is constructed based on the historical similar meeting dataset; the target meeting basic parameters are received from the meeting organizer. Multiple candidate seat permission allocation schemes are generated based on the target meeting's basic parameters; Each candidate seat allocation scheme and the target meeting basic parameters are input into the meeting effectiveness prediction model to obtain the corresponding predicted meeting effectiveness evaluation index value. Based on the preset meeting effectiveness priority weight vector, the predicted meeting effectiveness evaluation index values of all candidate seat permission allocation schemes are weighted and summed to generate the comprehensive effectiveness score of each scheme. The schemes are then sorted in descending order of comprehensive effectiveness score and the recommended permission allocation scheme sequence is output. During the formal operation of the meeting, real-time data streams of interactive behaviors at each seat are collected; Based on the interactive behavior data stream, calculate the current meeting process state feature vector; The current meeting process state feature vector is compared with the initial recommended permission allocation scheme. When the interaction behavior of any seat deviates from the preset behavior baseline threshold, the permission dynamic adjustment mechanism is triggered. The dynamic permission adjustment mechanism generates permission fine-tuning instructions for specific seats based on a pre-trained permission adjustment strategy network. The permission fine-tuning instruction is sent to the meeting service execution unit to complete the real-time update of the target seat's operation permissions and simultaneously record the adjustment event to the meeting audit log.
[0007] Preferably, the process of constructing the historical similar meeting dataset includes: Filter historical meeting records from the digital conferencing system database where the meeting topic category tags match the target meeting with a preset similarity threshold; Extract the participant identity attribute vector from each historical meeting record; Cluster each historical meeting record by seat role to form a standardized role template library; Map the actual permission assignments in each historical meeting record to the standardized role template library to generate a role-permission mapping table; Link the meeting process interaction logs and meeting performance evaluation indicators for each historical meeting record to complete the dataset labeling.
[0008] Preferably, the process of constructing the meeting effectiveness prediction model includes: A multilayer perceptron architecture is adopted, and its input layer receives a first feature vector generated by encoding the target meeting basic parameters and a second feature vector generated by encoding the candidate seat permission allocation scheme. The first feature vector is formed by concatenating the meeting topic category, expected number of participants, number of preset topics and preset meeting duration range into dense vectors of fixed dimensions through an embedding layer; The second feature vector is formed by one-hot encoding of the permission levels of each seat in multiple operation dimensions and then flattening and splicing them in the order of the seats. The hidden layers of the multilayer perceptron adopt a multilayer fully connected structure, and the activation function is the modified linear unit; The output layer contains multiple independent regression heads, each corresponding to a predicted value of a meeting effectiveness evaluation index; the model training uses a mean squared error loss function and an adaptive momentum estimation optimizer.
[0009] Preferably, the process of generating multiple candidate seat permission allocation schemes includes: Based on the set of participant identity attributes in the target meeting's basic parameters, an initial role label is assigned to each participant; according to the initial role label, the corresponding baseline permission configuration is retrieved from the pre-stored role-permission baseline policy library; Based on the aforementioned baseline permission configuration, the permission levels of at least two types of operation permissions for ordinary participants are perturbed to generate multiple permission states. Combine all perturbation variables to generate multiple candidate seat permission allocation schemes; Eliminate schemes with conflicting permissions, defined as the same seat being simultaneously granted both agenda-progressing trigger permission and screen-sharing operation prohibition.
[0010] Preferably, the preset meeting effectiveness priority weight vector is set by the meeting organizer through a slider control in the solution evaluation interface. The slider control corresponds to each meeting effectiveness evaluation index, and the sum of all slider values is normalized to 1; If the meeting organizer does not manually set the weights, the default weight vector will be used, in which the resolution achievement rate has the highest weight.
[0011] Preferably, the real-time acquisition of the interactive behavior data stream is achieved through a behavior monitoring agent deployed on the conference service node; The behavior monitoring agent captures the operation instruction packets sent by each client, performs protocol parsing on the operation instruction packets, and extracts the operation type identifier, operation target object identifier, and operation timestamp. Aggregate the operation events of the same seat within a continuous time window into a single interaction behavior record; The interaction records are pushed to the permission dynamic adjustment engine via a message queue.
[0012] Preferably, the preset behavior baseline threshold is determined based on the statistical distribution of interaction behaviors of seats with the same role in a historical similar meeting dataset; For each role type and each interaction behavior metric, calculate its mean and standard deviation in historical data; The preset behavior baseline threshold is set as an interval consisting of the mean plus or minus the standard deviation; When real-time interactive behavior data exceeds the upper limit of the interval, it is judged as overactive behavior; When the value is less than the lower limit of the interval, it is judged as passive participation behavior.
[0013] Preferably, the permission adjustment strategy network adopts a dual-branch convolutional neural network structure, and the input is the difference vector between the current meeting process state feature vector and the initial permission configuration vector; The first branch handles scenarios of overactive behavior and outputs suggestions for limiting permissions; the second branch handles scenarios of passive participation and outputs suggestions for incentivizing permissions. The two branches share the underlying feature extraction module, while the high-level decision-making module is independent. The network output contains adjustment information in multiple dimensions, with each dimension corresponding to the adjustment direction and magnitude of an operation permission dimension; The final permission fine-tuning command only retains the adjustment suggestions for the dimension with the largest absolute value of the adjustment range, and does not make adjustments to the other dimensions.
[0014] Preferably, after receiving the permission fine-tuning instruction, the meeting service execution unit first verifies whether the target seat's current session status is in an active connection state; If the connection is active, push a permission change notification message to the target seat client and update the seat permission status table in the memory of the conference server. If the target seat is currently disconnected, the permission fine-tuning command will be temporarily stored in the pending queue and will be automatically applied when it reconnects. All permission change operations generate structured log entries.
[0015] Preferably, the participant identity attribute vector includes job level, professional field tag, historical participation frequency, and historical speaking activity level; The standardized role template library includes roles such as host, speaker, recorder, general participant, and observer; The meeting effectiveness evaluation indicators include total meeting duration, average depth of discussion on topics, frequency of effective speeches, resolution achievement rate, and participant satisfaction rating.
[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention introduces a meeting effectiveness prediction model based on historical data, enabling meeting organizers to quantitatively evaluate the potential effects of different permission allocation schemes before the meeting, thus eliminating the need for blind configuration based on subjective experience. 2. By building a standardized role template library and a role-permission baseline strategy library, the structured and reusable nature of permission allocation has been achieved; 3. By collecting interactive behavior data in real time and establishing behavioral baseline thresholds, abnormal deviations in seat participation status can be accurately identified; 4. By deploying a network of permission adjustment policies, fine-grained, single-dimensional, and small-step dynamic fine-tuning of permission policies is achieved, avoiding brute-force permission deprivation or granting. 5. The entire method forms a closed-loop management process of "pre-meeting simulation - in-meeting monitoring - dynamic optimization - audit traceability", which improves the intelligence level of the digital conference system and ensures the coordinated optimization of meeting efficiency, participation fairness and decision-making quality. Attached Figure Description
[0017] Figure 1 This is a schematic diagram of the overall technical solution architecture of the present invention; Figure 2 This is a schematic diagram of the core principle framework of the meeting effectiveness prediction model driven by historical meeting data in this invention; Figure 3 This is a flowchart illustrating the logical process of generating a candidate seat authority allocation scheme and ranking it based on comprehensive performance evaluation in this invention. Figure 4 This is a logical flowchart of the data stream acquisition of interactive behavior and the calculation of process state feature vectors during the meeting operation phase in this invention. Figure 5 This is a schematic diagram of the permission fine-tuning mechanism framework based on behavior baseline deviation detection and permission dynamic adjustment strategy network in this invention; Figure 6 This is a schematic diagram of the multi-level interaction relationship and data flow between the digital conference system server and each seat client in this invention. Detailed Implementation
[0018] refer to Figures 1 to 6 This invention provides a method for dynamic allocation and management of multi-seat permissions in a digital conferencing system server. It aims to address the technical problem that meeting organizers cannot predict the impact of different permission allocation schemes on meeting efficiency, participation, and decision-making quality before the meeting, and can only rely on experience to set permissions. The method constructs a virtual simulation model driven by historical meeting data to quantitatively evaluate and prioritize multiple candidate permission allocation schemes before the meeting officially begins. This provides meeting organizers with interpretable and verifiable permission configuration suggestions. Furthermore, during the meeting, the method dynamically adjusts the operation permissions of each seat based on real-time interactive behavior, achieving adaptive coordination between permission policies and meeting progress status.
[0019] The method includes the following steps: Obtain a dataset of historical similar meetings for the target meeting; A meeting effectiveness prediction model is constructed based on the historical similar meeting dataset; the target meeting basic parameters are received from the meeting organizer. Multiple candidate seat permission allocation schemes are generated based on the target meeting's basic parameters; Each candidate seat allocation scheme and the target meeting basic parameters are input into the meeting effectiveness prediction model to obtain the corresponding predicted meeting effectiveness evaluation index value. Based on the preset meeting effectiveness priority weight vector, the predicted meeting effectiveness evaluation index values of all candidate seat permission allocation schemes are weighted and summed to generate the comprehensive effectiveness score of each scheme. The schemes are then sorted in descending order of comprehensive effectiveness score and the recommended permission allocation scheme sequence is output. During the formal operation of the meeting, real-time data streams of interactive behaviors at each seat are collected; Based on the interactive behavior data stream, calculate the current meeting process state feature vector; The current meeting process state feature vector is compared with the initial recommended permission allocation scheme. When the interaction behavior of any seat deviates from the preset behavior baseline threshold, the permission dynamic adjustment mechanism is triggered. The dynamic permission adjustment mechanism generates permission fine-tuning instructions for specific seats based on a pre-trained permission adjustment strategy network. The permission fine-tuning instruction is sent to the meeting service execution unit to complete the real-time update of the target seat's operation permissions and simultaneously record the adjustment event to the meeting audit log.
[0020] In the above method, step S1 involves obtaining a historical similar meeting dataset for the target meeting. This historical similar meeting dataset includes meeting metadata, seat role configuration information, seat operation permission allocation records, meeting process interaction logs, and meeting performance evaluation metrics for multiple completed meetings.
[0021] The process of constructing a historical similar meeting dataset includes: filtering historical meeting records from the digital meeting system database where the meeting topic category label matches the target meeting with a preset similarity threshold.
[0022] The preset similarity threshold is set to 0.75, and the cosine similarity algorithm is used to calculate the matching degree of the keyword vectors of the meeting topic. The participant identity attribute vector is extracted from each historical meeting record. The identity attribute vector includes job level, professional field tag, historical participation frequency, and historical speaking activity.
[0023] Job levels are divided into four categories: senior managers, middle managers, junior employees, and external experts; professional field tags use predefined industry knowledge graph node identifiers; historical attendance frequency is calculated based on the number of times the same type of meeting was attended in the past 12 months; historical speaking activity is defined as the average number of words spoken within the unit's meeting duration.
[0024] Each historical meeting record is clustered by seat role to form a standardized role template library. The standardized role template library includes five roles: moderator, speaker, recorder, general participant, and observer. The clustering process uses the K-means algorithm, with identity attribute vectors as input features. The initial cluster centers are determined by typical role samples labeled by experts. The actual permission allocation in each historical meeting record is mapped to the standardized role template library to generate a role-permission mapping table.
[0025] The mapping rule is as follows: if the function performed by a seat in a historical meeting overlaps more than 80% with the responsibilities described in the standard role template, then that seat is classified into that role category. The meeting process interaction logs for each historical meeting record are linked with meeting performance evaluation metrics to complete the dataset annotation. Meeting performance evaluation metrics include total meeting duration, average depth of discussion on topics, frequency of effective speeches, resolution achievement rate, and participant satisfaction rating.
[0026] The average discussion depth of each topic is calculated by semantic analysis of the speech text using natural language processing technology to calculate the average information entropy of the speech content under each topic; the effective speech frequency is defined as the number of speeches whose content contains substantive viewpoints or data support; the resolution achievement rate is equal to the number of resolutions successfully passed divided by the total number of resolutions submitted; the participant satisfaction score is derived from an anonymous questionnaire survey after the meeting, with a value range of 1 to 5.
[0027] In the above method, step S2 involves constructing a meeting effectiveness prediction model based on the historical similar meeting dataset. The construction process of the meeting effectiveness prediction model includes: employing a multilayer perceptron architecture, where the input layer receives a first feature vector generated by encoding the target meeting's basic parameters and a second feature vector generated by encoding the candidate seat permission allocation scheme.
[0028] The first feature vector is formed by concatenating the meeting topic category, expected participant range, number of preset topics, and preset meeting duration range into dense vectors of fixed dimensions through an embedding layer. The meeting topic category is encoded using one-hot encoding and then input into an embedding layer with a dimension of 100. The expected participant range is divided into four levels: less than 10 people, 10 to 30 people, 30 to 50 people, and more than 50 people. After one-hot encoding, the input into an embedding layer with a dimension of 50 is formed. The number of preset topics is directly normalized to the range of 0 to 1. The preset meeting duration range is divided into four levels: less than 30 minutes, 30 to 60 minutes, 60 to 120 minutes, and more than 120 minutes. After one-hot encoding, the input into an embedding layer with a dimension of 50 is formed.
[0029] The second feature vector is formed by flattening and concatenating the permission levels of each seat across five operational dimensions using one-hot encoding, arranged in seat order. The five operational dimensions include shared document editing, audio / video speaking control, voting initiation, agenda advancement triggering, and screen sharing. The permission levels for each operational dimension are divided into four levels: prohibited, read-only, requestable, and self-executable, with a corresponding one-hot encoding length of 4.
[0030] If the maximum number of seats in the meeting is 50, then the length of the second feature vector is 1000. The hidden layers of the multilayer perceptron adopt a 3-layer fully connected structure, with 512, 256, and 128 neurons in each layer, respectively, and the activation function is a modified linear unit. The output layer contains 5 independent regression heads, corresponding to the predicted values of total meeting duration, average discussion depth of topics, effective speaking frequency, resolution achievement rate, and participant satisfaction rating.
[0031] The model training employed a mean squared error loss function, an adaptive momentum estimation optimizer, and a learning rate of 0.001, with 50 training epochs. An early stopping strategy was used during training, terminating training when the validation set loss failed to decrease for five consecutive epochs. The model input data was standardized before training, with each feature dimension having a mean of 0 and a variance of 1.
[0032] In the above method, step S3 involves receiving the target meeting basic parameters input by the meeting organizer. These parameters include the meeting topic category, expected number of participants, preset agenda list, preset meeting duration, and a set of participant identity attributes. The meeting topic category is selected from a predefined topic classification tree, which contains three levels of nodes: the root node represents the industry category, and the leaf nodes represent specific business scenarios. The expected number of participants is an integer ranging from 3 to 100.
[0033] The pre-set agenda list is entered item by item by the meeting organizer, with each agenda item including a title and a brief description. The pre-set meeting duration is in minutes, ranging from 15 to 300. The participant identity attribute set is filled in by the meeting organizer for each participant, including name, job title, professional field tag, and whether they are a speaker. All parameters are submitted through a form on the meeting creation interface, validated by the front end, and then transmitted to the server.
[0034] In the above method, step S4 is to generate multiple candidate seat permission allocation schemes based on the target meeting basic parameters.
[0035] The process of generating multiple candidate seat permission allocation schemes includes: assigning an initial role label to each participant based on the participant identity attribute set in the target meeting basic parameters.
[0036] The initial role assignment rules are as follows: if a participant is marked as a keynote speaker, they will be assigned the keynote speaker role; if a participant is a senior manager and is not marked as a keynote speaker, they will be assigned the moderator role; if a participant is a middle manager and their professional field is highly relevant to the meeting topic, they will be assigned the recorder role; all other participants will be assigned the general participant role; if the meeting organizer invites additional external personnel and does not assign any other roles, they will be assigned the observer role.
[0037] Based on the initial role label, retrieve the corresponding baseline permission configuration from the pre-stored role-permission baseline policy library.
[0038] In the role-permission baseline policy library, the host role has the highest permissions across all five operation dimensions; the speaker role has the highest permissions in shared document editing, audio and video speaking control, and screen sharing operations, and can request permissions in initiating voting and triggering agenda progression; the recorder role has the highest permissions in shared document editing, and has read-only or prohibited permissions in the other dimensions; the ordinary participant role has can request permissions in audio and video speaking control, and is prohibited by default in the other dimensions; the observer role has prohibited permissions in all dimensions.
[0039] Based on the aforementioned baseline permission configuration, three independent permission level perturbations are applied to the document editing permission, voting initiation permission, and agenda advancement trigger permission for ordinary participants. Each perturbation generates three permission states: high, medium, and low. A high permission state indicates autonomous execution, a medium permission state indicates requestable execution, and a low permission state indicates prohibited execution. Combining all perturbation variables generates 27 candidate seat permission allocation schemes. Schemes with permission conflicts are eliminated. A permission conflict is defined as the same seat being simultaneously granted agenda advancement trigger permission and a screen sharing operation prohibited state. Permission conflict detection involves traversing the permission configuration vector of each seat to check for combined patterns; if a pattern exists, the scheme is eliminated.
[0040] In the above method, step S5 involves inputting the permission allocation scheme for each candidate seat and the basic parameters of the target meeting into the meeting effectiveness prediction model to obtain the corresponding predicted meeting effectiveness evaluation index value.
[0041] The input process includes: encoding the basic parameters of the target meeting into a first feature vector, encoding the candidate seat permission allocation scheme into a second feature vector, and then concatenating them and inputting them into the trained multilayer perceptron model.
[0042] The model inference process is executed in a GPU-accelerated environment, with a single inference taking less than 50 milliseconds. The output consists of 5 predicted values, corresponding to the total meeting duration (in minutes), average topic discussion depth (in information entropy), effective speaking frequency (in times / hour), resolution achievement rate (dimensionless, 0 to 1), and participant satisfaction score (in points).
[0043] In the above method, step S6 is to perform a weighted summation of the predicted meeting effectiveness evaluation index values of all candidate seat permission allocation schemes based on a preset meeting effectiveness priority weight vector, generate a comprehensive effectiveness score for each scheme, and output a recommended permission allocation scheme sequence in descending order of comprehensive effectiveness score.
[0044] The preset meeting effectiveness priority weight vector is set by the meeting organizer through a slider control in the solution evaluation interface. The slider control corresponds to 5 meeting effectiveness evaluation indicators. The value of each slider ranges from 0 to 1, and the sum of all slider values is normalized to 1.
[0045] If the meeting organizer does not manually set the weights, the default weight vector will be used. In the default weight vector, the weight of the resolution achievement rate is 0.4, the weight of the average discussion depth of the topics is 0.3, the weight of the total meeting duration is -0.2, the weight of the effective speaking frequency is 0.2, and the weight of the participant satisfaction rating is 0.1.
[0046] The overall performance score is calculated as follows: The overall performance score is the sum of the normalized or original values of each meeting performance evaluation indicator and their corresponding weights.
[0047] The total meeting duration needs to be normalized first. The normalized value is calculated by subtracting the ratio of (the predicted total duration of the target meeting minus the shortest duration of similar historical meetings) to (the longest duration of similar historical meetings minus the shortest duration of similar historical meetings) from 1. The shortest and longest durations of similar historical meetings are the shortest and longest durations of similar meetings in the historical data, respectively. Other indicators use the original values directly, including the average depth of discussion on topics, the frequency of effective speeches, the resolution achievement rate, and the participant satisfaction score. Each indicator corresponds to a weight coefficient.
[0048] After the calculation is completed, all candidate schemes are sorted in descending order of comprehensive effectiveness score, and the top five are output as the recommended permission allocation scheme sequence. The detailed indicator comparison of each scheme is displayed on the interface.
[0049] In the above method, step S7 involves real-time collection of interactive behavior data streams from each seat during the formal operation of the meeting. These interactive behavior data streams include the percentage of speaking time, document modification frequency, voting response latency, number of agenda skip requests, and number of unauthorized operation attempts. Real-time collection of these interactive behavior data streams is achieved through a behavior monitoring proxy deployed on the meeting service node. The behavior monitoring proxy captures operation command packets sent by clients at each seat with a sampling period of 10 milliseconds.
[0050] The operation instruction packet follows a unified internal communication protocol and includes an operation type identifier, an operation target object identifier, and an operation timestamp. The operation instruction packet is parsed to extract the required fields. Operation events from the same seat within a consecutive 5-second window are aggregated into a single interaction record.
[0051] The aggregation rules are as follows: the percentage of speaking time is equal to the total speaking time of the seat within the 5-second window divided by the total window duration; the document modification frequency is the number of times the seat edits the shared document; the voting response delay is the time difference from the initiation of the vote to the submission of the seat's response, and if there is no response, it is recorded as a timeout value; the number of agenda skipping requests is the number of times the seat actively requests to skip the current topic; the number of unauthorized operation attempts is the number of times the seat attempts to perform an operation it does not have permission to perform.
[0052] The interaction records are pushed to the permission dynamic adjustment engine via a message queue, which ensures that the data is transmitted in an orderly and reliable manner.
[0053] In the above method, step S8 involves calculating the current meeting process state feature vector based on the interactive behavior data stream. The current meeting process state feature vector is a five-dimensional vector, with each dimension corresponding to the moving average of the five interactive behavior indicators mentioned above. The sliding window length is 30 seconds, and an exponentially weighted moving average algorithm is used, with the weight of the most recent five seconds of data being 0.6 and the weight of the data from the previous 25 seconds being 0.4. The feature vector is updated each time a new interactive behavior record arrives, with an update frequency greater than once per second.
[0054] In the above method, step S9 involves comparing the current meeting process state feature vector with the initial recommended permission allocation scheme. When the interaction behavior of any seat deviates from a preset behavior baseline threshold, a dynamic permission adjustment mechanism is triggered. The preset behavior baseline threshold is determined based on the statistical distribution of interaction behaviors of seats with the same role in a historical similar meeting dataset.
[0055] For each role type and each interaction behavior metric, its mean and standard deviation in historical data are calculated. A preset behavior baseline threshold is set as the interval formed by the mean plus or minus twice the standard deviation. When real-time interaction behavior data exceeds the upper limit of the interval, it is judged as overactive behavior; when it is below the lower limit, it is judged as passive participation behavior. The comparison process is as follows: for each seat in the current meeting, the corresponding behavior baseline threshold interval is queried based on its role label, and its current interaction behavior metric value is compared with the interval boundary. If any metric exceeds the interval, a dynamic permission adjustment mechanism is triggered, and the trigger reason code is recorded.
[0056] In the above method, step S10 involves the dynamic permission adjustment mechanism generating permission fine-tuning instructions for specific seats based on a pre-trained permission adjustment strategy network. The permission adjustment strategy network employs a dual-branch convolutional neural network structure, and its input is the difference vector between the current meeting process state feature vector and the initial permission configuration vector.
[0057] The difference vector is calculated by subtracting the corresponding baseline mean from the current interaction behavior index value and then dividing by the standard deviation to obtain the standardized deviation vector. The first branch handles overactive behavior scenarios and outputs permission restriction suggestions; the second branch handles passive participation behavior scenarios and outputs permission incentive suggestions. The two branches share the underlying feature extraction module, while the high-level decision-making module is independent.
[0058] The network output contains adjustment information in five dimensions. Each dimension corresponds to the adjustment direction and magnitude of an operation permission dimension. The adjustment direction is either to increase or decrease permission, and the magnitude is 0.1, 0.2, or 0.3. The final permission fine-tuning instruction only retains the adjustment suggestion for the dimension with the largest absolute value of the adjustment magnitude, and leaves the other dimensions unchanged. For example, if the adjustment magnitude for the voting initiation permission in the output result is -0.3 (i.e., a restriction direction with a magnitude of 0.3), and the adjustment magnitudes for the other dimensions are 0, then a fine-tuning instruction of "restricting voting initiation permission" is generated, with the adjustment magnitude being a reduction of one level.
[0059] In the above method, step S11 involves sending the permission fine-tuning instruction to the conference service execution unit to complete the real-time update of the target seat's operation permissions and simultaneously record the adjustment event in the conference audit log. Upon receiving the permission fine-tuning instruction, the conference service execution unit first verifies whether the target seat's current session status is in an active connection state. If it is, it pushes a permission change notification message to the target seat's client and highlights the changed item in the conference control panel; simultaneously, it updates the seat permission status table in the conference server's memory.
[0060] If the target seat is currently disconnected, the permission fine-tuning instruction is temporarily stored in the pending queue and automatically applied when it reconnects. All permission change operations generate structured log entries, which include the change time, target seat identifier, original permission status, new permission status, trigger reason code, and operator identifier. The operator identifier here is "System Automatic" to distinguish it from manual intervention.
[0061] The system implementation of the above method includes: The historical data acquisition module is used to acquire a dataset of historical similar meetings for the target meeting. The effectiveness model building module is used to build a meeting effectiveness prediction model based on the historical similar meeting dataset; The meeting parameter receiving module is used to receive the target basic meeting parameters input by the meeting organizer; The candidate scheme generation module is used to generate multiple candidate seat permission allocation schemes based on the target meeting basic parameters; The efficiency prediction and evaluation module is used to input the permission allocation scheme of each candidate seat into the meeting efficiency prediction model to obtain the predicted meeting efficiency evaluation index value. The solution ranking and recommendation module is used to comprehensively score and rank candidate solutions based on a preset meeting effectiveness priority weight vector. The interactive behavior acquisition module is used to collect interactive behavior data streams from each seat in real time during the meeting. The process state calculation module is used to calculate the current meeting process state feature vector based on the interactive behavior data stream; The deviation detection module is used to compare the current meeting process state feature vector with the initial recommended permission allocation scheme to detect deviations in interactive behavior; The dynamic permission adjustment module is used to generate permission fine-tuning instructions based on the permission adjustment policy network when a deviation is detected. The permission execution synchronization module is used to send the permission fine-tuning instructions to the conference service execution unit and update the audit log synchronously.
[0062] Each module is deployed in a distributed computing cluster of the digital conferencing system server and communicates through an internal message bus to ensure real-time response capability under high concurrency.
[0063] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.
[0064] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.
Claims
1. A method for dynamic allocation and management of multi-seat permissions in a digital conferencing system server, characterized in that, include: Obtain a dataset of historical similar meetings for the target meeting; A meeting effectiveness prediction model is constructed based on the aforementioned historical similar meeting dataset; Receive the target meeting basic parameters input by the meeting organizer; Multiple candidate seat permission allocation schemes are generated based on the target meeting's basic parameters; Each candidate seat allocation scheme and the target meeting basic parameters are input into the meeting effectiveness prediction model to obtain the corresponding predicted meeting effectiveness evaluation index value. Based on the preset meeting effectiveness priority weight vector, the predicted meeting effectiveness evaluation index values of all candidate seat permission allocation schemes are weighted and summed to generate the comprehensive effectiveness score of each scheme. The schemes are then sorted in descending order of comprehensive effectiveness score, and a recommended permission allocation scheme sequence is output. During the formal operation of the meeting, real-time data streams of interactive behaviors at each seat are collected; Based on the interactive behavior data stream, calculate the current meeting process state feature vector; The current meeting process state feature vector is compared with the initial recommended permission allocation scheme. When the interaction behavior of any seat deviates from the preset behavior baseline threshold, the permission dynamic adjustment mechanism is triggered. The dynamic permission adjustment mechanism generates permission fine-tuning instructions for specific seats based on a pre-trained permission adjustment strategy network. The permission fine-tuning instruction is sent to the meeting service execution unit to complete the real-time update of the target seat's operation permissions and simultaneously record the adjustment event to the meeting audit log; The process of constructing the meeting effectiveness prediction model includes: A multilayer perceptron architecture is adopted, and its input layer receives a first feature vector generated by encoding the target meeting basic parameters and a second feature vector generated by encoding the candidate seat permission allocation scheme. The first feature vector is formed by concatenating the meeting topic category, expected number of participants, number of preset topics and preset meeting duration range into dense vectors of fixed dimensions through an embedding layer; The second feature vector is formed by one-hot encoding of the permission levels of each seat in multiple operation dimensions and then flattening and splicing them in the order of the seats. The hidden layers of the multilayer perceptron adopt a multilayer fully connected structure, and the activation function is the modified linear unit; The output layer contains multiple independent regression heads, each corresponding to a predicted value of a meeting effectiveness evaluation index; the model training uses a mean squared error loss function and an adaptive momentum estimation optimizer.
2. The method according to claim 1, characterized in that, The process of constructing the historical similar conference dataset includes: Filter historical meeting records from the digital conferencing system database where the meeting topic category tags match the target meeting with a preset similarity threshold; Extract the participant identity attribute vector from each historical meeting record; Cluster each historical meeting record by seat role to form a standardized role template library; Map the actual permission assignments in each historical meeting record to the standardized role template library to generate a role-permission mapping table; Link the meeting process interaction logs and meeting performance evaluation indicators for each historical meeting record to complete the dataset labeling.
3. The method according to claim 2, characterized in that, The process of generating multiple candidate seat permission allocation schemes includes: Based on the set of participant identity attributes in the target meeting's basic parameters, an initial role label is assigned to each participant; according to the initial role label, the corresponding baseline permission configuration is retrieved from the pre-stored role-permission baseline policy library; Based on the aforementioned baseline permission configuration, the permission levels of at least two types of operation permissions for ordinary participants are perturbed to generate multiple permission states. Combine all perturbation variables to generate multiple candidate seat permission allocation schemes; Schemes with conflicting permissions are eliminated. A conflicting permission is defined as the same user being simultaneously granted both agenda-progressing permission and screen-sharing permission being disabled.
4. The method according to claim 3, characterized in that, The preset meeting effectiveness priority weight vector is set by the meeting organizer through a slider control in the solution evaluation interface. The slider control corresponds to each meeting effectiveness evaluation index, and the sum of all slider values is normalized to 1; If the meeting organizer does not manually set the weights, the default weight vector will be used, in which the resolution achievement rate has the highest weight.
5. The method according to claim 4, characterized in that, The real-time acquisition of the interactive behavior data stream is achieved through a behavior monitoring agent deployed on the conference service node; The behavior monitoring agent captures the operation instruction packets sent by each client, performs protocol parsing on the operation instruction packets, and extracts the operation type identifier, operation target object identifier, and operation timestamp. Aggregate the operation events of the same seat within a continuous time window into a single interaction behavior record; The interaction records are pushed to the permission dynamic adjustment engine via a message queue.
6. The method according to claim 5, characterized in that, The preset behavior baseline threshold is determined based on the statistical distribution of interaction behaviors of seats with the same role in historical similar meeting datasets; For each role type and each interaction behavior metric, calculate its mean and standard deviation in historical data; The preset behavior baseline threshold is set as an interval consisting of the mean plus or minus the standard deviation; When real-time interactive behavior data exceeds the upper limit of the interval, it is judged as overactive behavior; When the value is less than the lower limit of the interval, it is judged as passive participation behavior.
7. The method according to claim 6, characterized in that, The permission adjustment strategy network adopts a dual-branch convolutional neural network structure, and the input is the difference vector between the current meeting process state feature vector and the initial permission configuration vector; The first branch handles scenarios involving overactive behavior and outputs suggestions for permission restrictions. The second branch handles scenarios involving passive participation and outputs permission incentive suggestions. The two branches share the underlying feature extraction module, while the high-level decision-making module is independent. The network output contains adjustment information in multiple dimensions, with each dimension corresponding to the adjustment direction and magnitude of an operation permission dimension; The final permission fine-tuning command only retains the adjustment suggestions for the dimension with the largest absolute value of the adjustment range, and does not make adjustments to the other dimensions.
8. The method according to claim 7, characterized in that, Upon receiving the permission fine-tuning instruction, the meeting service execution unit first verifies whether the target seat's current session status is in an active connection state. If the connection is active, push a permission change notification message to the target seat client and update the seat permission status table in the memory of the conference server. If the target seat is currently disconnected, the permission fine-tuning command will be temporarily stored in the pending queue and will be automatically applied when it reconnects. All permission change operations generate structured log entries.
9. The method according to claim 8, characterized in that, The participant identity attribute vector includes job level, professional field label, historical attendance frequency, and historical speaking activity level; The standardized role template library includes roles such as host, speaker, recorder, general participant, and observer; The meeting effectiveness evaluation indicators include total meeting duration, average depth of discussion on topics, frequency of effective speeches, resolution achievement rate, and participant satisfaction rating.