An artificial intelligence-based multi-dimensional permission and publishing domain management method and system

Abstract

The application relates to the technical field of content cooperation platform permission control, and discloses a multi-dimensional permission and publishing domain management and control method and system based on artificial intelligence, wherein the method comprises the following steps: obtaining an editing event stream of target content and aggregating the editing event stream into an editing behavior sequence matrix; calculating a participant activity concentration index of each time window and identifying a candidate point of a cooperation micro-stage conversion; extracting editing behavior features before and after the candidate point and identifying a cooperation micro-stage type and a content sensitivity level; searching for a permission contraction strategy and calculating a section-level permission set and a publishing visible range set of each participant; generating a section-level access mask instruction and a publishing range limitation instruction and updating in batches in an atomic operation.

Description

Technical Field

[0001] This invention relates to the field of content collaboration platform access control technology, specifically to a multi-dimensional access control method and system for publishing domains based on artificial intelligence. Background Technology

[0002] In enterprise content management and publishing platforms, content goes through multiple collaborative stages from creation to official release. The editing behavior patterns of participants at different stages are significantly different, and the sensitivity of each section of the content also changes dynamically with the progress of collaboration.

[0003] Existing systems typically configure access permissions statically based on content status tags or user roles, with the entire content serving as the minimum permission control granularity.

[0004] The above methods cannot detect the internal sensitivity shifts caused by structural changes in the editing behavior patterns of participants during the collaboration process, nor can they implement differentiated release scope restrictions for different sections of the same content. This results in highly sensitive sections still being exposed to non-essential personnel when sensitivity increases, posing a risk that core content may be obtained without authorization before official release. Summary of the Invention

[0005] This invention provides a multi-dimensional permission and publishing domain management method and system based on artificial intelligence, which solves the technical problem in related technologies that content collaboration platforms cannot adaptively adjust the segment-level permissions and publishing visibility range of participants according to the dynamic stage changes of the collaboration process.

[0006] This invention discloses a multi-dimensional permission and publishing domain management method based on artificial intelligence, comprising the following steps: The real-time editing event stream of the target content in the content collaboration platform is obtained, and the operator's identity, editing location segment identifier, operation type, operation timestamp and operation duration are extracted and aggregated into an editing behavior sequence matrix according to a preset sliding time window. For the editing behavior sequence matrix, a participant activity concentration index is calculated for each time window. The participant activity concentration index includes the Gini coefficient of editing operation volume, the proportion of core editors, and the frequency of approval operations. The concentration index sequence consisting of multiple consecutive windows is input into the change point detection model to identify the time points when structural changes occur in the concentration index sequence, and the detected change time points are marked as candidate points for collaborative micro-stage transition. For each collaborative micro-stage transition candidate point, the editing behavior features within a preset number of windows before and after the candidate point are extracted, the editing behavior features are concatenated into a feature vector, input into the micro-stage type recognition model, and the collaborative micro-stage type label and content sensitivity level are output. Based on the content sensitivity level, the permission contraction strategy is retrieved from the sensitivity-permission mapping rule base. The role attributes of the current participant are matched with the permission contraction strategy, and the segment-level permission set and the publication visibility range set of each participant are calculated. Generate segment-level access mask instructions and publishing scope restriction instructions, and update the segment permission mask and publishing scope configuration of all affected participants in batches using atomic operations.

[0007] Furthermore, the row dimension of the editing behavior sequence matrix corresponds to the participants, the column dimension corresponds to the time window, and the matrix elements are the operation type vectors of the corresponding participants within the corresponding time window; the operation type vector is a numerical vector composed of the number of operations of each type in the preset operation type set arranged in a fixed order.

[0008] Furthermore, the Gini coefficient of the amount of editing operation Calculate using the following formula: ; in, For time window The total number of participants within, and Participant indexes, For the first Each participant in the window Total number of editing operations within, For the first Each participant in the window Total number of editing operations within.

[0009] Furthermore, the change point detection model is a probabilistic model based on Bayesian change point detection, whose input is the concentration index sequence and output is the time position of the candidate change point and the corresponding posterior probability; before inputting the concentration index sequence into the change point detection model, a moving average smoothing process is applied to the concentration index sequence.

[0010] Furthermore, only collaborative micro-stage transition candidate points with posterior probabilities exceeding a preset threshold are subjected to editing behavior feature extraction and micro-stage type identification; candidate points with posterior probabilities not exceeding the preset threshold are directly discarded.

[0011] Furthermore, the editing behavior features include the distribution of new participant role types, the rate of change of editing section coverage, and the rate of change of annotation density; the micro-stage type recognition model is a multi-class neural network with a fully connected output layer that outputs the probability distribution of collaborative micro-stage type labels and the corresponding content sensitivity level; the micro-stage type recognition model uses the cross-entropy loss function as the optimization objective and the Adam optimization algorithm for parameter updates.

[0012] Furthermore, the sensitivity-permission mapping rule base uses a tuple of role type and sensitivity level as the index key, and stores the whitelist of accessible segments, the list of upper limit of operation permissions, and the set of target audiences for the publishing scope in the corresponding entries; when a participant has multiple roles, the intersection of the upper limit of operation permissions in the shrinkage strategy corresponding to all roles of the participant is taken as the final upper limit of permissions.

[0013] Furthermore, the atomic operation method encapsulates the writing of the segment permission mask and the writing of the publishing scope configuration of all affected participants into a single transaction commit. If any write in the transaction fails, all will be rolled back. Before generating the segment-level access mask instruction and the publishing scope restriction instruction, it is determined whether the time interval since the last permission update exceeds the preset cooldown period. If it does not exceed the time interval, the current instruction is added to the pending queue and will be merged and processed after the cooldown period ends.

[0014] This invention discloses a multi-dimensional permission and publishing domain management system based on artificial intelligence, comprising: The event acquisition module is used to acquire the real-time editing event stream of the target content, extract the operator's identity, editing location segment identifier, operation type, operation timestamp, and operation duration, and aggregate them into an editing behavior sequence matrix according to a preset sliding time window; The concentration analysis module is used to calculate the participant activity concentration index for each time window of the editing behavior sequence matrix, input the concentration index sequence composed of multiple consecutive windows into the change point detection model, and output the collaborative micro-stage transition candidate points. The micro-stage identification module is used to extract the editing behavior feature vector before and after each candidate point of the collaborative micro-stage transformation, input the feature vector into the micro-stage type identification model, and output the collaborative micro-stage type label and content sensitivity level. The strategy matching module is used to retrieve permission contraction strategies from the sensitivity-permission mapping rule base based on the content sensitivity level, match the participant role attributes with the permission contraction strategies, and calculate the segment-level permission set and the publication visibility range set for each participant. The permission update module is used to generate segment-level access mask instructions and publishing scope restriction instructions, and to write the segment permission mask and publishing scope configuration of all affected participants in batches using atomic operations.

[0015] Furthermore, before generating an instruction, the permission update module determines whether the time interval between the last permission update and the current instruction exceeds a preset cooldown period. If it does not exceed the time interval, the current instruction is added to the processing queue and processed after the cooldown period ends. When a participant has multiple roles, the strategy matching module takes the intersection of the upper limits of operation permissions in the shrinkage strategies corresponding to all roles as the final upper limit of permissions.

[0016] The beneficial effects of this invention are as follows: This invention makes the active distribution structure of participants explicit by performing window-by-window concentration index calculation on the editing behavior sequence matrix, enabling the change point detection model to capture the time points of structural changes in the distribution of editing power. It solves the technical problem that relying on coarse-grained status labels cannot perceive the sensitivity transitions within the collaboration period. It achieves the technical effect of refining the minimum granularity of permission control and publishing scope control from the entire content to content segments, and enabling highly sensitive segments to independently trigger contraction when sensitivity increases, while the original permissions of non-sensitive segments remain undisturbed. Attached Figure Description

[0017] Figure 1 This is a flowchart of a multi-dimensional permission and publishing domain management method based on artificial intelligence provided in an embodiment of the present invention. Detailed Implementation

[0018] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, some features described in the examples may be combined in other examples.

[0019] At least one embodiment of the present invention discloses a multi-dimensional permission and publishing domain management method based on artificial intelligence, see [link to relevant documentation]. Figure 1 This includes the following steps: Step 1: Obtain the editing event stream of the target content and aggregate it into an editing behavior sequence matrix.

[0020] Acquire the real-time editing event stream of the target content in the content collaboration platform. For each record in the event stream, extract the operator's identity, editing location segment identifier, operation type, operation timestamp, and operation duration. The operation type includes add, delete, modify, comment, and approval tags.

[0021] The above event streams are aggregated according to a preset sliding time window to generate an editing behavior sequence matrix. . The row dimension corresponds to the participants, the column dimension corresponds to the time window, and the matrix elements... For the first The participants in the A vector of operation types within a time window. Indexing participants, Index for time windows.

[0022] It should be noted that the operation type vector is a numerical vector formed by arranging the number of operations of each type in a preset set of operation types in a fixed order. For example, if the set of operation types includes five categories: add, delete, modify, comment, and approval mark, then the dimension of the operation type vector is 5, and each component corresponds to the number of times the corresponding participant performs the corresponding operation within that window.

[0023] In this embodiment of the application, in order to make the aggregation result robust to sudden high-frequency operations, an upper limit can be applied to the duration of the operation before it participates in the aggregation, so as to avoid abnormal pull of matrix elements by a single long-duration operation.

[0024] Step 2: Calculate the concentration index based on the edit behavior sequence matrix and identify candidate points for collaborative micro-stage transitions.

[0025] For editing behavior sequence matrix For each time window Calculate the participant activity concentration index, including the Gini coefficient of editing activity. Percentage of core editors and the frequency of approval-related operations .

[0026] in, Calculate using the following formula: ; In the formula, For time window The total number of participants within, and Participant indexes, For the first Each participant in the window The total number of editing operations within (i.e., the sum of the components of the operation type vector). For the first Each participant in the window Total number of editing operations within. The closer it is to 1, the more concentrated the editing operations are on a small number of participants.

[0027] Defined as the ratio of the number of participants whose editing activity ranks among the top preset percentages to the total number of participants. Define as window The ratio of the number of internal approval mark operations to the total number of operations.

[0028] A series of concentration indicators composed of multiple consecutive windows The input change point detection model identifies time points where significant structural changes occur in the concentration index sequence, and marks these detected time points as candidate points for collaborative micro-stage transitions. The change point detection model is a probabilistic model based on Bayesian change point detection, and its input is the concentration index sequence. The output is the time location of the candidate change point and its corresponding posterior probability.

[0029] It should be noted that the core editors are identified by the smallest set of participants whose cumulative editing operations reach a preset percentage threshold of the total number of operations in the entire window, after being sorted in descending order of the number of editing operations performed by each participant in the window.

[0030] In this embodiment of the application, in order to reduce false alarm candidate points caused by abnormal jitter of a single window, the concentration index sequence can be first subjected to moving average smoothing before being input into the change point detection model.

[0031] Step 3: Extract editing behavior features before and after candidate points and identify the type of collaborative micro-stage and content sensitivity level.

[0032] For each collaborative micro-stage transition candidate point, editing behavior features are extracted within each preset number of windows before and after the candidate point. Specifically, these features include: the distribution of new participant role types, the rate of change in edited segment coverage, and the rate of change in annotation density. The distribution of new participant role types is represented by a vector representing the percentage of new participants for each role type (regular editor, approver, read-only observer). The rate of change in edited segment coverage is defined as the ratio of the difference between the number of edited segments in the preset window after the candidate point and the number of edited segments in the corresponding window before the candidate point, to the previous value. The rate of change in annotation density is defined as the ratio of the difference between the average number of annotation operations per segment in the corresponding window before and after the candidate point, to the previous value.

[0033] The above features are concatenated into a feature vector and input into the micro-stage type recognition model. The micro-stage type recognition model is a multi-classification neural network, trained based on labeled micro-stage sequences of historically completed content. Further, the input layer of the micro-stage type recognition model receives the concatenated feature vector, and the output layer is... The fully connected layer outputs the probability distribution of type labels for the current collaborative micro-stage and the corresponding content sensitivity level; during training, the cross-entropy loss function is used as the optimization objective. The optimization algorithm updates its parameters. The sensitivity level is a value taken from a discrete, ordered set of labels.

[0034] It should be noted that the collaboration micro-stage type tag refers to the enumeration of typical stage types manually marked during the historical content collaboration process, such as open discussion stage, restricted approval stage, high-level approval stage, etc. The tagging source is the collaboration record of historical completed content and the corresponding final status information.

[0035] In this embodiment of the application, in order to handle the case where the posterior probability of the candidate point is low, a probability threshold filtering can be set. Only the candidate points whose posterior probability exceeds the preset threshold are subjected to feature extraction and recognition in step 3, and the remaining candidate points are directly discarded to reduce invalid model inference calls.

[0036] Step 4: Retrieve the permission contraction strategy and calculate the segment-level permission set and the publication visibility range set for each participant.

[0037] Based on the output content sensitivity level, a multi-dimensional permission contraction strategy applicable to the current micro-stage is retrieved from the sensitivity-permission mapping rule base. This strategy defines the range of accessible content segments, upper limits of operation permissions, and publishing scope restrictions for each participant role under the corresponding sensitivity level.

[0038] Match the role attributes of all current participants with the retrieved contraction strategies, and calculate the set of segment-level permissions to be contracted or maintained for each participant. and the set of visible ranges for publication .in, Identify participants, For participants An enumerated set of operation permission limits for each content section. For participants The union of the target audience sets in each segment.

[0039] It should be noted that the shrinkage strategy in the sensitivity-permission mapping rule base uses a tuple (role type, sensitivity level) as the index key, and the corresponding entries store the whitelist of accessible segments, the list of operation permission limits, and the set of target audiences for the publishing scope.

[0040] In this embodiment of the application, in order to support the situation where the same participant has multiple roles, the intersection of the upper limit of operation permissions in the shrinkage strategy corresponding to all roles of the participant can be taken as the final upper limit of permissions, so as to ensure that the permission shrinkage is not relaxed when multiple roles are superimposed.

[0041] Step 5: Generate section-level access mask instructions and publication range restriction instructions, and update them in batches using atomic operations.

[0042] For participants requiring access control reduction, generate segment-level access mask instructions and publishing scope restriction instructions. The segment-level access mask instruction changes the access status of the participant in the high-sensitivity segment from editable or viewable to masked. The publishing scope restriction instruction narrows the publishing scope parameters of the corresponding segment to the publishing visibility range set determined in step 4. ,in Identify participants.

[0043] The permission mask and publishing scope configurations for all affected participants are updated in batches using atomic operations. Existing permissions and publishing scopes for non-sensitive segments remain unchanged in this update.

[0044] It should be noted that the atomic operation method refers to encapsulating the write of the permission mask of all affected participants and the write of the publication scope configuration into a single transaction commit. If any write in the transaction fails, the entire transaction is rolled back to avoid inconsistencies in permission status caused by partial updates.

[0045] In this embodiment of the application, in order to deal with the situation where the micro-stage type switches continuously in a short period of time, a cooling window judgment can be introduced before generating the instruction. If the time interval between the last permission update does not exceed the preset cooling time, the current instruction is added to the pending queue and processed after the cooling window ends, so as to reduce the impact of frequent writes on the permission configuration service.

[0046] On the other hand, this invention also proposes a multi-dimensional permission and publishing domain management system based on artificial intelligence, including: The event acquisition module is used to acquire the real-time editing event stream of the target content, extract each field and aggregate them into an editing behavior sequence matrix according to the sliding time window, which corresponds to the processing in step 1; The concentration analysis module is used to calculate the participant activity concentration index for each time window of the editing behavior sequence matrix, and input the concentration index sequence into the change point detection model to output the collaborative micro-stage transition candidate points, corresponding to the processing in step 2. The micro-stage identification module is used to extract the feature vectors of editing behavior before and after each collaborative micro-stage transition candidate point, and input the feature vectors into the micro-stage type identification model to output the collaborative micro-stage type label and content sensitivity level, corresponding to the processing in step 3. The strategy matching module is used to retrieve shrinkage strategies from the sensitivity-permission mapping rule base based on the content sensitivity level, match them with the role attributes of the participants, calculate the segment-level permission set and the publication visibility range set for each participant, which corresponds to the processing in step 4. The permission update module is used to generate segment-level access mask instructions and publishing scope restriction instructions, and to write the permission mask and publishing scope configuration of all affected participants in batches using atomic operations, corresponding to the processing in step 5.

[0047] By way of example, the system proposed in this invention may also include various features and combinations thereof in the method embodiments, which will not be elaborated here.

[0048] This invention makes the previously invisible active participant distribution structure explicit by performing window-by-window calculations of the Gini coefficient, the proportion of core editors, and the frequency of approval operations on the editing behavior sequence matrix. This allows the change point detection model to capture the time points of structural changes in the distribution of editing power, thus overcoming the limitation of relying on coarse-grained status labels, which cannot perceive sensitivity transitions within the collaboration period. Based on this, the micro-stage type recognition model maps detected transition candidate points to specific content sensitivity levels. Changes in sensitivity levels directly trigger the linked generation of section-level access mask instructions and publishing scope restriction instructions, refining the minimum granularity of permission control and publishing scope control from the entire content to content sections. High-sensitivity sections can independently trigger contraction when sensitivity increases, while the original permissions and collaboration processes of non-sensitive sections remain unaffected, thereby maintaining collaboration continuity while ensuring content security.

[0049] The embodiments of the present invention have been described above. However, the embodiments are not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make more equivalent embodiments under the guidance of the present embodiments, and all of them are within the protection scope of the present embodiments.

Claims

1. A multi-dimensional permission and publishing domain management method based on artificial intelligence, characterized in that, Includes the following steps: The real-time editing event stream of the target content in the content collaboration platform is obtained, and the operator's identity, editing location segment identifier, operation type, operation timestamp and operation duration are extracted and aggregated into an editing behavior sequence matrix according to a preset sliding time window. For the editing behavior sequence matrix, a participant activity concentration index is calculated for each time window. The participant activity concentration index includes the Gini coefficient of editing operation volume, the proportion of core editors, and the frequency of approval operations. The concentration index sequence consisting of multiple consecutive windows is input into the change point detection model to identify the time points when the concentration index sequence undergoes structural changes. The detected change time points are marked as candidate points for collaborative micro-stage transition. For each collaborative micro-stage transition candidate point, the editing behavior features within a preset number of windows before and after the candidate point are extracted, the editing behavior features are concatenated into a feature vector, input into the micro-stage type recognition model, and the collaborative micro-stage type label and content sensitivity level are output. Based on the content sensitivity level, the permission contraction strategy is retrieved from the sensitivity-permission mapping rule base. The role attributes of the current participant are matched with the permission contraction strategy, and the segment-level permission set and the publication visibility range set of each participant are calculated. Generate segment-level access mask instructions and publishing scope restriction instructions, and update the segment permission mask and publishing scope configuration of all affected participants in batches using atomic operations.

2. The multi-dimensional permission and publishing domain management method based on artificial intelligence according to claim 1, wherein, The row dimension of the edit behavior sequence matrix corresponds to the participants, the column dimension corresponds to the time window, and the matrix elements are the operation type vectors of the corresponding participants within the corresponding time window; the operation type vector is a numerical vector composed of the number of operations of each type in the preset operation type set arranged in a fixed order.

3. The multi-dimensional permission and publishing domain management method based on artificial intelligence according to claim 1, wherein, The Gini coefficient of the amount of editing operations Calculate using the following formula: ; in, For time window The total number of participants within, and Participant indexes, For the first Each participant in the window Total number of editing operations within, For the first Each participant in the window Total number of editing operations within.

4. The multi-dimensional permission and publishing domain management method based on artificial intelligence according to claim 1, wherein, The change point detection model is a probabilistic model based on Bayesian change point detection. Its input is the concentration index sequence, and its output is the time position of the candidate change point and the corresponding posterior probability. Before inputting the concentration index sequence into the change point detection model, a moving average smoothing process is applied to the concentration index sequence.

5. The multi-dimensional permission and publishing domain management method based on artificial intelligence according to claim 4, wherein, Editing behavior feature extraction and micro-stage type identification are performed only on collaborative micro-stage transition candidate points whose posterior probability exceeds a preset threshold; candidate points whose posterior probability does not exceed the preset threshold are directly discarded.

6. The method according to claim 1, wherein, The editing behavior features include the distribution of new participant role types, the rate of change of editing section coverage, and the rate of change of annotation density; the micro-stage type recognition model is a multi-class neural network with a fully connected output layer that outputs the probability distribution of collaborative micro-stage type labels and the corresponding content sensitivity level; the micro-stage type recognition model uses the cross-entropy loss function as the optimization objective and the Adam optimization algorithm for parameter updates.

7. The multi-dimensional permission and publishing domain management method based on artificial intelligence according to claim 1, wherein, The sensitivity-permission mapping rule base uses a tuple of role type and sensitivity level as the index key, and stores the whitelist of accessible segments, the list of operation permission limits, and the target audience set of the publishing scope in the corresponding entries; When a participant has multiple roles, the intersection of the maximum operation permissions in the shrinkage strategy corresponding to all roles of that participant is taken as the final maximum permission limit.

8. The multi-dimensional permission and publishing domain management method based on artificial intelligence according to claim 1, wherein, The atomic operation method encapsulates the writing of the segment permission mask and the writing of the publishing scope configuration of all affected participants into a single transaction commit. If any write in the transaction fails, all will be rolled back. Before generating the segment-level access mask instruction and the publishing scope restriction instruction, it is determined whether the time interval since the last permission update exceeds the preset cooldown period. If it does not exceed the time interval, the current instruction is added to the pending queue and will be merged and processed after the cooldown period ends.

9. A multi-dimensional access control and publishing scope control system based on artificial intelligence, characterized in that, include: The event acquisition module is used to acquire the real-time editing event stream of the target content, extract the operator's identity, editing location segment identifier, operation type, operation timestamp, and operation duration, and aggregate them into an editing behavior sequence matrix according to a preset sliding time window; The concentration analysis module is used to calculate the participant activity concentration index for each time window of the editing behavior sequence matrix, input the concentration index sequence composed of multiple consecutive windows into the change point detection model, and output the collaborative micro-stage transition candidate points. The micro-stage identification module is used to extract the editing behavior feature vector before and after each candidate point of the collaborative micro-stage transition, input the feature vector into the micro-stage type identification model, and output the collaborative micro-stage type label and content sensitivity level. The strategy matching module is used to retrieve permission contraction strategies from the sensitivity-permission mapping rule base based on the content sensitivity level, match the participant role attributes with the permission contraction strategies, and calculate the segment-level permission set and the publication visibility range set for each participant. The permission update module is used to generate segment-level access mask instructions and publishing scope restriction instructions, and to write the segment permission mask and publishing scope configuration of all affected participants in batches using atomic operations.

10. The multi-dimensional permission and publishing domain management system based on artificial intelligence according to claim 9, wherein, Before generating an instruction, the permission update module determines whether the time interval between the last permission update and the current instruction exceeds the preset cooldown period. If it does not exceed the preset cooldown period, the current instruction is added to the processing queue and processed after the cooldown period ends. When a participant has multiple roles, the strategy matching module takes the intersection of the upper limits of operation permissions in the shrinkage strategies corresponding to all roles as the final upper limit of permissions.

Images