Dynamic management method and system for international transmission data flow based on Agent agent

Abstract

The invention provides a dynamic management method and system for an international transmission data stream based on an Agent agent, and relates to the technical field of data management and cross-border compliance. According to the method, an original data pool with a semantic index is constructed through a distributed crawler, a governance instruction is generated based on multi-modal perception and a joint decision model, operations such as data filtering and desensitization are executed to form a governance result, channel division is performed according to a compliance ontology model, and continuous self-adaptive updating is performed through a multi-target Bayesian optimization driving strategy. And dynamic management and optimization of cross-platform and multi-language data streams are realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data governance and cross-border compliance, in particular to an international communication data flow dynamic management method and system based on Agent intelligent agents. BACKGROUND

[0002] Under the pattern of highly interconnected global digital media, international communication data flow has grown exponentially, with media platforms, social networks and content distribution networks running simultaneously in various countries. Massive cross-language, cross-platform and cross-time-zone data flows are simultaneously flowing into centralized monitoring and filtering systems. The industry usually relies on pre-set keyword matching, black and white lists and static data quality checks to screen, such as the invention patent with publication number CN105808604A, which proposes to improve data compliance through dirty data identification and correction, representing the typical approach of centralized rules and manual governance.

[0003] With the deepening of artificial intelligence technology, automatic semantic understanding, multi-agent collaborative decision-making and cross-domain execution framework have become a hot research direction. In recent years, some solutions have tried to abstract regulatory elements into a knowledge base or "compliance factor library" and use deep learning and encryption technology to achieve cross-border data protection and blocking. For example, the invention patent with publication number CN114760149B uses dynamic compliance factor matching and privacy identification to achieve automatic compliance control of cross-border data, providing technical support for rapid iteration of multiple regulations.

[0004] However, most existing systems are still an extension of monolithic models or rule engines: their compliance knowledge base updates and governance strategy execution chain are loose, and they cannot adapt to reorganization when global regulations change rapidly or major events occur. They also lack real-time evaluation of governance effectiveness and strategy closed-loop optimization, resulting in coexistence of strategy solidification, resource waste and compliance risks. SUMMARY

[0005] In order to overcome the shortcomings of the prior art, the purpose of the present application is to provide an international communication data flow dynamic management method and system based on Agent intelligent agents, which realizes the dynamic collection and adaptive governance of cross-platform and multi-language international communication data flow, significantly improving the accuracy, real-time and sustainable optimization ability of governance.

[0006] To achieve the above purpose, the present application provides the following scheme:

[0007] An international communication data flow dynamic management method based on Agent intelligent agents, comprising:

[0008] The distributed crawler nodes dynamically allocate crawling tasks according to an extensible hash algorithm to capture cross-platform and multi-lingual data streams, and attach source identification, timestamps, topic labels and sensitivity weights to each data stream to form a raw data pool with semantic indexing.

[0009] Autonomous agents are instantiated for each topic label in the raw data pool; the autonomous agents include a perception module, a decision module and an execution module; the perception module encodes the multi-modal content of the raw data pool using a hybrid Transformer-CNN network; the decision module outputs governance instructions through a multi-armed bandit-contextual bandit-policy gradient joint network combined with an international compliance rule base; each autonomous agent exchanges state summaries and issues synchronization instructions through a message bus, and maintains global policy consistency through a consensus algorithm;

[0010] The execution module performs processing operations including filtering, desensitization, rate limiting, reordering, multi-lingual synchronization on the raw data streams according to the governance instructions to obtain governance results;

[0011] The governance results are written to a governance result queue, and the data streams in the governance result queue are divided into fast channels, review channels and cold storage channels according to the priority of the compliance ontology model;

[0012] After each governance cycle ends, governance results are extracted from the governance result queue, and a comprehensive score is calculated based on timeliness, compliance and propagation effect;

[0013] Based on the comprehensive score, a multi-objective Bayesian optimization driven model-independent meta-learning algorithm is used to update the parameters of the decision module in small steps, and the updated parameters are written back to each autonomous agent to achieve adaptive migration and optimization of the governance strategy.

[0014] Preferably, the distributed crawler nodes dynamically allocate crawling tasks according to an extensible hash algorithm to capture cross-platform and multi-lingual data streams, and attach source identification, timestamps, topic labels and sensitivity weights to each data stream to form a raw data pool with semantic indexing, including:

[0015] A predetermined data source is obtained; the data source includes news websites, social platforms, video platforms, content distribution networks, open data interfaces and other information publishing channels, each data source corresponding to an access address, data format and access protocol;

[0016] Based on the access domain name, IP address and interface call information of the data source, a unique source identification is generated, and the platform type, language type and time zone information corresponding to the data source are recorded as metadata basis;

[0017] Based on the computing power, network bandwidth, and historical load of each crawler node, a scalable hash algorithm is used to dynamically divide the task domain, and the data source is mapped and distributed to different distributed crawler nodes according to platform type, language type, and time zone information.

[0018] Each of the distributed crawler nodes captures the target data stream in real time based on the task domain rules, and adds the source identifier and collection timestamp information to each of the collected target data streams;

[0019] For the collected target data stream, topic tags are extracted through natural language processing and topic modeling, and sensitivity weights are calculated in combination with the content sensitivity model;

[0020] The source identifier, the collection timestamp information, the topic tag, the sensitivity weight, and the target data stream are encapsulated into structured metadata and uniformly written into the original data pool.

[0021] Preferably, the perception module encodes the multimodal content of the original data pool using a hybrid Transformer-CNN network, including:

[0022] For the data streams associated with each of the aforementioned topic tags in the original data pool, text data, audio data, video data, and image data are extracted from the data streams to form a multimodal dataset;

[0023] For the text data, a pre-trained Transformer encoding model is used to perform context modeling, and the output text feature vector containing semantic relationships is generated.

[0024] For the audio data and the video data, first extract the corresponding audio feature sequence and video frame image sequence, then input them into a convolutional neural network to extract temporal and spatial local features, and output audio feature vectors and video feature vectors;

[0025] The image data is directly input into a convolutional neural network to extract spatial hierarchical features and output an image feature vector.

[0026] The text feature vector, audio feature vector, video feature vector and image feature vector are concatenated to construct a unified multimodal semantic feature matrix;

[0027] The multimodal semantic feature matrix is ​​input into the fusion layer, and the cross-modal correlation is further modeled using the fusion Transformer encoding layer to output a unified multimodal semantic vector.

[0028] Preferably, the decision-making module outputs governance instructions through a multi-armed contextual gambling machine-policy gradient joint network and in conjunction with an international compliance rule base, including:

[0029] Receive the multimodal unified semantic vector s output by the perception module, and extract the compliance constraint embedding vector c corresponding to the current propagation scenario from the international compliance rule library;

[0030] The multimodal unified semantic vector s is concatenated with the compliance constraint embedding vector c to form a context vector, which serves as the state input for the multi-armed context gambling machine;

[0031] For each governance action a in action set A, explore using a multi-armed contextual gambling machine – use the score and select several candidate actions;

[0032] The candidate action, along with the context vector, is fed into the policy gradient network. The compliance-weighted policy gradient is calculated according to the following formula, and the policy parameters are updated:

[0033]

[0034] Where θt is the policy network parameter vector; η is the learning rate; rt is the real-time propagation reward; λ is the compliance penalty coefficient; l t For compliance and violation level; a t For the selected governance action; s t c is the perception state vector; t For compliant embedding vectors; π θ (·) represents the parameterized strategy probability distribution; gradient operator

[0035] Based on the updated policy distribution, candidate actions are re-evaluated, and the optimal action is selected as the final governance instruction and output to the execution module. At the same time, the current round's state-action-benefit-penalty quadruple is written into the experience replay buffer.

[0036] Preferably, each of the autonomous agents exchanges state summaries and issues synchronization instructions via a message bus, maintaining global policy consistency through a consensus algorithm, including:

[0037] Each autonomous agent generates a state summary after completing its local governance decision; the state summary includes at least the current local policy vector, average reward value and timestamp, and is written to the message bus via publish and subscribe.

[0038] Each of the autonomous agents subscribes to external state summaries related to its own topic tags from the message bus to construct a neighborhood state set;

[0039] For the aforementioned neighborhood state set, a global consistency strategy is calculated using a weighted consensus algorithm. The specific update rules are as follows:

[0040]

[0041] in, Let N be the local policy vector of agent i in the k-th round; α be the neighborhood fusion coefficient; N be the local policy vector of agent i in the k-th round; α be the neighborhood fusion coefficient; N be the local policy vector of agent i in the k-th round; α be the local policy vector of agent i in the k-th round ... α be the local policy vector of agent i i Let w be the set of neighborhoods visible to agent i on the message bus; ij Let the trust weights from agent j to agent i satisfy...

[0042] The obtained consistent policy vector is written into the local policy cache, and a synchronization instruction is generated and broadcast to the execution module, so that each autonomous agent executes the consistent governance policy in the next decision cycle.

[0043] Preferably, the execution module performs processing operations on the original data stream according to the governance instructions, including filtering, desensitization, rate limiting, reordering, and multilingual synchronization, to obtain the governance result, including:

[0044] Parse the governance instructions, extract the operation type, execution order, target fields, and language tags, and construct an operation list;

[0045] Based on the keywords, regular expressions, and blacklist / whitelist rules in the operation list, content matching is performed on the text, image, and audio content in the original data stream to delete or block content related to the violation topic, resulting in a filtered data stream D. filter ;

[0046] For the filtered data stream D filter Fields involving identity information, geolocation information, and device identification information are anonymized. Sensitivity levels are set to implement tiered labeling and replacement, resulting in an anonymized data stream D. desens ;

[0047] Based on the bandwidth threshold and time window specified in the governance instructions, the de-identification result data stream D... desens Implement traffic shaping, dynamically adjust the data packet sending rate, and generate a rate-limited data stream D. throttle ;

[0048] Based on the priority information and timestamp in the governance instruction, the rate-limited data stream D is... throttle Reorder the data to obtain the sorted data stream D. sorted High-priority data packets are sent first, while low-priority data packets are delayed and queued.

[0049] When the operation list contains multilingual synchronization markers, based on the sorted data stream D sorted The text content is processed by a neural machine translation model and a terminology database to generate a target language version aligned with the original language content. This target language version is then encapsulated using the original media format to obtain a multilingual synchronous data stream D. multi ;

[0050] Take the latest available data in the filtered data stream D filter , the desensitized result data stream D desens , the data stream D after speed limit throttle , the sorted data stream D sorted and the multilingual synchronization data stream D multi as the final data stream, and combine the operation records, processing tags and compliance status, and summarize and package them into governance results.

[0051] Preferably, write the governance results into the governance result queue, and divide the data streams in the governance result queue into fast channels, review channels and cold storage channels according to the compliance ontology model by priority, including:

[0052] When generating the governance results, attach compliance risk scores, time urgency levels and content popularity prediction values to them, and write them into the governance result queue;

[0053] For each governance result in the governance result queue, call the compliance ontology model to verify the number of trigger clauses, and calculate the priority score

[0054] S = tanh(R + T − E − C)

[0055] where R is the compliance risk score; T is the time urgency level; E is the content popularity prediction value; C is the number of compliance clause triggers; S is the priority score normalized by the hyperbolic tangent function;

[0056] According to the priority score S and the preset threshold set {s1, s2} for channel division:

[0057] When s1 ≤ S, it is assigned to the fast channel; when s2 ≤ S < s1, it is assigned to the review channel; when S < s2, it is assigned to the cold storage channel;

[0058] Update the index in the queue with the channel label together with the governance result.

[0059] Preferably, after each governance cycle, extract the governance results from the governance result queue, and calculate the comprehensive score according to timeliness, compliance and dissemination effect, including:

[0060] Read the generation timestamp of the governance result, calculate the difference Δt from the current time, and calculate the timeliness decay factor H, and its calculation method is:

[0061] H = e -Δt / τ

[0062] where τ is the time decay constant, Δt is the time difference between the current time and the generation time of the governance result, and H represents the timeliness decay factor;

[0063] Based on the number C of violations triggered under the compliance ontology model according to the governance results, the compliance penalty factor P is calculated as follows:

[0064]

[0065] The click count V, forwarding count F, and average dwell time D of the data stream corresponding to the governance results are obtained among the target audience. The dissemination effect index E is calculated as follows:

[0066]

[0067] Based on the above indicators, the overall score S is calculated using the following formula:

[0068] S=σ(ln(1+E)HP)

[0069] Where S is the overall score, E is the dissemination effect index, H is the timeliness decay factor, P is the compliance penalty factor, and σ is the logistic function.

[0070] Preferably, based on the comprehensive score, a multi-objective Bayesian optimization-driven model-independent meta-learning algorithm is used to update the parameters of the decision module in small steps, and the updated parameters are written back to each of the autonomous agents to achieve adaptive migration and optimization of the governance strategy, including:

[0071] At the end of each governance cycle, the comprehensive score S corresponding to each autonomous agent is collected. i and its current decision parameters Constructing a historical sample set Among them, S i Let i be the overall score of the i-th autonomous agent. Let be the decision parameter vector of the i-th autonomous agent at time t;

[0072] For the historical sample set D, a Gaussian process surrogate model is trained to minimize the objective min(1-S). And based on the Gaussian process proxy model Define the multi-objective expected hypervolume improvement function EHVI(θ);

[0073] The optimal increment is obtained through multi-objective Bayesian optimization, using the following formula:

[0074] Δθ * =argmax Δθ EHVI(θ (t) +Δθ)

[0075] Where, Δθ * To maximize the optimal parameter increment obtained by EHVI(θ);

[0076] Small-step transfer is performed using a model-independent meta-learning update rule, with the following formula:

[0077] θ (t+1) =θ (t) +ηΔθ *

[0078] Where η is the step size coefficient adaptively calculated from the average prediction variance of the surrogate model; θ (t+1) This is the decision parameter vector updated with a small step size;

[0079] Update the parameter θ (t+1) The data is written back to the decision-making module corresponding to the autonomous agent to complete the policy adaptive optimization.

[0080] A dynamic management system for international communication data streams based on agent-based intelligent agents, comprising:

[0081] The data acquisition and annotation unit is used by distributed crawler nodes to dynamically allocate crawling tasks based on a scalable hash algorithm to capture cross-platform, multilingual data streams, and to attach source identifiers, timestamps, topic tags and sensitivity weights to each data stream to form a raw data pool with semantic indexes.

[0082] An agent construction and coordination unit is used to instantiate autonomous agents for each of the topic tags in the original data pool; the autonomous agent includes a perception module, a decision module, and an execution module; the perception module uses a hybrid Transformer-CNN network to encode the multimodal content of the original data pool; the decision module outputs governance instructions through a multi-armed contextual gambling machine-policy gradient joint network and in conjunction with an international compliance rule base; each of the autonomous agents exchanges state summaries and issues synchronization instructions via a message bus, and maintains global policy consistency with the help of a consensus algorithm;

[0083] The data governance execution unit is used to perform processing operations, including filtering, desensitization, rate limiting, reordering, and multilingual synchronization, on the original data stream according to the governance instructions by the execution module, and to obtain the governance result.

[0084] The priority channel partitioning unit is used to write the governance results into the governance result queue and, based on the compliance ontology model, divide the data flow in the governance result queue into fast channel, review channel and cold storage channel according to priority.

[0085] The comprehensive evaluation unit is used to extract governance results from the governance result queue after each governance cycle ends, and calculate a comprehensive score based on timeliness, compliance and dissemination effect;

[0086] The strategy optimization and update unit is used to update the parameters of the decision module in small steps based on the comprehensive score, using a multi-objective Bayesian optimization-driven model irrelevant meta-learning algorithm, and write the updated parameters back to each of the autonomous agents to achieve adaptive migration and optimization of the governance strategy.

[0087] According to specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0088] (1) By introducing distributed crawlers and scalable hash algorithms, this invention can dynamically and efficiently cover the collection of international communication data streams in multiple languages ​​and across platforms, overcoming the shortcomings of incomplete data acquisition and insufficient real-time performance under existing centralized and static rules, and ensuring the comprehensiveness and timeliness of the original data pool.

[0089] (2) This invention constructs an autonomous intelligent agent system with perception, decision-making and execution capabilities. Based on multimodal semantic coding and reinforcement learning, it dynamically generates governance instructions, replacing the existing governance methods that rely on a single rule base or static model. It has adaptive governance capabilities for complex propagation scenarios and significantly improves the accuracy and flexibility of cross-context and cross-regional data flow management.

[0090] (3) This invention constructs a closed-loop optimization link by dividing the governance results into priority channels and the comprehensive evaluation mechanism of the governance cycle, combined with multi-objective Bayesian optimization and model-independent meta-learning. This overcomes the problems of disconnect between governance effect and strategy update and strategy solidification in existing solutions, realizes dynamic migration and continuous optimization of governance strategies, and improves the long-term effectiveness and stability of governance.

[0091] (4) In this invention, each autonomous intelligent agent exchanges state summaries through a message bus and maintains global consistency, avoiding the risks of isolated single-model and fragmented strategies, ensuring the overall collaborative effect under the concurrent governance of multiple themes and tasks, and improving the robustness and adaptability of the system under the dynamic changes of the global compliance environment. Attached Figure Description

[0092] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0093] Figure 1 A flowchart of the method provided in an embodiment of the present invention;

[0094] Figure 2 This is a schematic diagram of the system structure provided in an embodiment of the present invention. Detailed Implementation

[0095] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0096] The purpose of this invention is to provide a dynamic management method and system for international communication data streams based on agent-based intelligent agents, which realizes dynamic collection and adaptive governance of cross-platform, multilingual international communication data streams, and significantly improves the accuracy, real-time performance and sustainable optimization capabilities of governance.

[0097] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0098] Figure 1 The method flowchart provided in the embodiments of the present invention is as follows: Figure 1 As shown, this invention provides a method for dynamic management of international communication data streams based on agent-based intelligent agents, including:

[0099] Step 100: Distributed crawler nodes dynamically allocate crawling tasks based on a scalable hash algorithm to capture cross-platform, multilingual data streams, and attach source identifiers, timestamps, topic tags, and sensitivity weights to each data stream to form a raw data pool with semantic indexes;

[0100] Step 200: Instantiate autonomous agents for each topic tag in the original data pool; the autonomous agent includes a perception module, a decision module, and an execution module; the perception module encodes the multimodal content of the original data pool using a hybrid Transformer-CNN network; the decision module outputs governance instructions through a multi-armed contextual gambling machine-policy gradient joint network and in conjunction with an international compliance rule base; each autonomous agent exchanges state summaries and issues synchronization instructions via a message bus, and maintains global policy consistency with the help of a consensus algorithm;

[0101] Step 300: The execution module performs processing operations on the original data stream according to the governance instructions, including filtering, desensitization, rate limiting, reordering, and multilingual synchronization, to obtain the governance results;

[0102] Step 400: Write the governance results into the governance results queue, and divide the data flow in the governance results queue into fast channel, review channel and cold storage channel according to priority based on the compliance ontology model;

[0103] Step 500: After the end of each governance cycle, extract the governance results from the governance results queue and calculate the comprehensive score based on timeliness, compliance and dissemination effect;

[0104] Step 600: Based on the comprehensive score, a multi-objective Bayesian optimization-driven model irrelevant meta-learning algorithm is used to update the parameters of the decision module in small steps, and the updated parameters are written back to each autonomous agent to achieve adaptive migration and optimization of the governance strategy.

[0105] In this embodiment, preferably, the distributed crawler nodes dynamically allocate crawling tasks based on a scalable hash algorithm to capture cross-platform, multilingual data streams, and attach source identifiers, timestamps, topic tags, and sensitivity weights to each data stream to form a raw data pool with semantic indexes. Specifically, this includes the following steps: First, this embodiment pre-defines a set of data sources, including news websites, social media platforms, video platforms, content delivery networks, open data interfaces, and other information publishing channels, and configures a unique access address, data format, and access protocol for each data source. Based on the access domain name, IP address, or interface call information of the data source, this embodiment generates a source identifier to uniquely identify the data source, and records the specific platform type (e.g., news, social media, video), language type (e.g., Chinese, English, Japanese), and time zone information of the corresponding data source, serving as the metadata foundation for subsequent data management and processing.

[0106] Furthermore, based on the real-time computing power, network bandwidth, and historical load of each distributed crawler node, the task domain is dynamically divided using a scalable hash algorithm. According to the platform type, language type, and time zone information, the data source is mapped and allocated to different distributed crawler nodes. Each crawler node crawls the corresponding target data stream in real time according to the allocation results, and automatically appends the aforementioned source identifier and collection timestamp information to each collected data stream. Subsequently, this embodiment uses natural language processing methods to parse the collected target data stream, combines topic modeling algorithms to identify its propagation topic, extracts corresponding topic tags, and calls a content sensitivity model to calculate sensitivity weights. The source identifier, collection timestamp, topic tags, and sensitivity weights are encapsulated together with the original data stream into structured metadata, which is uniformly stored in the original data pool as the basic data for subsequent autonomous agent scheduling and governance.

[0107] Specifically, in this embodiment, preferably, the perception module uses a hybrid Transformer-CNN network to encode the multimodal content of the original data pool, specifically including the following steps: First, for the data streams associated with each of the topic tags in the original data pool, this embodiment extracts text data, audio data, video data, and image data from the data streams to form a multimodal dataset. For the text data, a Transformer encoder based on a pre-trained BERT (Bidirectional Encoder Representations from Transformers) model is used for context modeling. The BERT model contains 12 encoder layers, each layer employing 12 multi-head self-attention mechanisms, with a hidden layer dimension of 768, a maximum input sequence length of 512, and outputs a text feature vector containing semantic relationships. For the audio data and the video data, short-time Fourier transform is performed on the audio to extract the Mel spectrogram feature sequence, and keyframe sequence of 5 frames per second is extracted from the video. Both are uniformly input into a five-layer convolutional neural network (CNN). Each layer of the CNN contains convolutional kernels with 64, 128, 256, 512, and 512 channels, respectively. The kernel size is 3×3, the activation function is ReLU, and the max pooling kernel size is 2×2. The corresponding audio feature vector and video feature vector are output respectively.

[0108] Furthermore, for the image data, a convolutional neural network with the same structure as the aforementioned CNN is directly input to extract spatial hierarchical features and output an image feature vector. Subsequently, in this embodiment, the text feature vector, audio feature vector, video feature vector, and image feature vector are concatenated to construct a unified multimodal semantic feature matrix, which is then input into the fusion layer. The fusion layer is based on a Transformer encoder architecture, employing a 6-layer stacked encoder structure, with each layer containing 8 multi-head self-attention units and a hidden dimension of 512. Residual connections and layer normalization use standard configurations. This fused Transformer encoder takes the concatenated multimodal feature matrix as input and outputs a fused unified multimodal semantic vector. This unified multimodal semantic vector is used by the subsequent decision-making module of the autonomous agent to generate governance instructions, ensuring that this embodiment can accurately capture the semantic associations and propagation features of different propagation topics under multimodal data.

[0109] Preferably, the decision-making module outputs governance instructions through a multi-armed contextual gambling machine-policy gradient joint network and in conjunction with an international compliance rule base, including:

[0110] Receive the multimodal unified semantic vector s output by the perception module, and extract the compliance constraint embedding vector c corresponding to the current propagation scenario from the international compliance rule library;

[0111] The multimodal unified semantic vector s is concatenated with the compliance constraint embedding vector c to form a context vector, which serves as the state input for the multi-armed context gambling machine;

[0112] For each governance action a in action set A, explore using a multi-armed contextual gambling machine – use the score and select several candidate actions;

[0113] The candidate action, along with the context vector, is fed into the policy gradient network. The compliance-weighted policy gradient is calculated according to the following formula, and the policy parameters are updated:

[0114]

[0115] Where θt is the policy network parameter vector; η is the learning rate; rt is the real-time propagation reward; λ is the compliance penalty coefficient; l t For compliance and violation level; a t For the selected governance action; s t c is the perception state vector; t For compliant embedding vectors; π θ (·) represents the parameterized strategy probability distribution; For gradient operators;

[0116] Based on the updated policy distribution, candidate actions are re-evaluated, and the optimal action is selected as the final governance instruction and output to the execution module. At the same time, the current round's state-action-benefit-penalty quadruple is written into the experience replay buffer.

[0117] In this embodiment, preferably, the decision-making module outputs corresponding governance instructions through a multi-armed contextual gambling machine and a policy gradient joint network, combined with an international compliance rule base. Specifically, the decision-making module first receives a multimodal unified semantic vector output by the perception module. This semantic vector represents the content characteristics, dissemination tendency, and public opinion attributes of the target data stream. Simultaneously, the decision-making module automatically retrieves and extracts the embedded representations of compliance constraints from a preset international compliance rule base, based on the country and region to which the current data stream belongs and its corresponding dissemination scenario. These compliance embeddings include, but are not limited to, data privacy clauses, dissemination sensitivity levels, and content security restrictions. These are transformed into vector information recognizable by the model through rule parsing and semantic vectorization techniques. Subsequently, the multimodal unified semantic vector and the compliance constraint embedded vector are concatenated to form a context vector, which serves as the state input of the multi-armed contextual gambling machine.

[0118] Furthermore, the multi-armed contextual gambling machine, for a preset set of governance actions (including filtering, desensitization, rate limiting, reordering, multilingual synchronization, etc.), calculates the exploration-exploitation score of each governance action in the current propagation context, combining the context vector, and selects several candidate actions with higher scores. Subsequently, the candidate actions and context vectors are input into the policy gradient network. Combining historical governance records and the latest feedback information, the governance strategy parameters are iteratively updated by dynamically adjusting the learning step size and reward function, and the benefit-risk balance of each candidate action is re-evaluated. Finally, the decision module selects the governance action with the best overall benefit from the candidate actions and outputs it as the final governance instruction to the execution module for execution. At the same time, the state vector corresponding to this round, the selected action, the actual governance benefit, and the corresponding violation penalty information are stored in the experience replay buffer for continuous optimization of subsequent governance strategies and acceleration of model convergence. All the above parameter data are calculated and updated in real time by the system during operation, without the need for manual preset or manual adjustment, ensuring the automation and adaptive capability of the system governance.

[0119] In this embodiment, preferably, each autonomous agent exchanges state summaries and publishes synchronization instructions through a message bus to achieve consistent coordination of the global governance strategy. Specifically, after completing its local data flow governance decision, each autonomous agent automatically generates a state summary containing the current local policy vector, the cumulative average governance benefit value within the period, and the corresponding timestamp. The state summary is written to a unified message bus through a message publish-subscribe mechanism for other autonomous agents to obtain in real time. This state summary is used to characterize the current operational status and benefit performance of the agent's governance strategy, ensuring transparent information flow within the system and providing a data foundation for subsequent global consistency calculations.

[0120] Furthermore, each autonomous agent actively subscribes to state summaries published by external agents associated with its assigned topic tags via a message bus, automatically aggregating to form a neighborhood state set. This neighborhood state set describes the local governance strategy status and benefits of other related agents within the current system. This embodiment dynamically identifies the neighborhood set based on preset rules, including but not limited to dimensions such as topic tag overlap, historical collaboration frequency, and governance effect correlation. Simultaneously, this embodiment generates trust weights for each neighborhood agent based on historical interaction data or preset rules, representing the proportion of influence of its local strategy on the current agent's globally consistent strategy update. These weights are preset during system initialization and updated in real-time during operation based on strategy convergence and governance benefits.

[0121] After acquiring the neighborhood state set, each autonomous agent calculates the updated global consensus policy using a weighted consensus algorithm and caches the calculated consensus policy vector in its local policy library for execution in the next governance cycle. Simultaneously, each autonomous agent generates a synchronization instruction based on the updated consensus policy vector and broadcasts it to the execution module via a message bus, enabling all autonomous agents to collaboratively execute data flow governance tasks according to the unified governance policy in subsequent cycles. The consensus calculation and synchronization instruction update are both completed automatically by the system, ensuring the collaborative consistency of multiple autonomous agents and the global stability of the governance effect in a distributed environment.

[0122] In this embodiment, preferably, the execution module performs processing operations on the original data stream according to the governance instructions, including filtering, desensitization, rate limiting, reordering, and multilingual synchronization. Specifically, the execution module first parses the governance instructions, extracting the corresponding operation type, execution order, target fields, and language tags, and constructs an operation list as the basis for execution. Then, in this embodiment, based on keywords, regular expressions, and blacklist / whitelist rules in the operation list, it performs content recognition and matching on the text, image, and audio content in the original data stream, deleting or blocking content related to the violating topics, forming a filtered data stream. Based on this, for sensitive fields such as identity information, geographic location information, and device identification information contained in the filtered data stream, it performs hierarchical marking and replacement operations according to preset sensitivity level rules, forming a desensitized result data stream. Further, according to the preset bandwidth threshold and time window requirements in the governance instructions, it performs traffic shaping operations on the desensitized result data stream, adjusting the data packet sending rate to obtain a rate-limited data stream. This embodiment further reorders the rate-limited data stream based on the priority information and timestamp in the governance instructions, prioritizing the transmission of high-priority data packets and delaying the transmission of low-priority data packets to obtain a sorted data stream. If the operation list contains multilingual synchronization markers, a neural machine translation model and terminology database are invoked based on the text content in the sorted data stream to generate a target language version aligned with the original language content. This target language version is then encapsulated using the original media format to obtain a multilingual synchronized data stream. Finally, this embodiment uses the latest available data from each of the above stages as the final data stream. Combined with corresponding operation records, processing tags, and compliance status information, the data is summarized, encapsulated, and stored in a governance result queue for subsequent channel partitioning and evaluation modules to access.

[0123] In this embodiment, preferably, when the governance results are generated, a compliance risk score, a timeliness urgency score, and a content popularity prediction score are automatically added to them, and these are uniformly written into the governance result queue as the basis for subsequent priority allocation. Specifically, the compliance risk score is calculated based on the historical governance records of the data stream and the severity and frequency of triggering violation rules; the timeliness urgency score is set based on the timeliness requirements of public opinion in the domain to which the data stream belongs and the content update cycle; and the content popularity prediction score is estimated by combining real-time dissemination trends, click-through rates, and repost volume data from the platform. All governance results are uniformly encapsulated into structured entries and written into the governance result queue in real time for subsequent scheduling and retrieval.

[0124] Furthermore, in this embodiment, for each governance result in the governance result queue, the compliance ontology model is invoked for content verification, the number of compliance clauses triggered is counted, and a priority score is calculated based on the compliance risk score, timeliness, predicted content popularity, and the number of triggered clauses. This score is mapped to a fixed interval using a normalization function for subsequent channel allocation. In this embodiment, based on a preset set of scoring thresholds, governance results with priority scores higher than the first threshold are allocated to the fast channel for rapid, low-latency processing; governance results with scores between the first and second thresholds are allocated to the review channel for manual review and in-depth governance; and governance results with scores lower than the second threshold are allocated to the cold storage channel for long-term archiving and subsequent traceability. This embodiment updates and stores the channel tags and governance results together in the index structure of the governance result queue, ensuring transparency and traceability throughout the data flow's lifecycle management.

[0125] In this embodiment, after each governance cycle, all governance results are extracted from the governance result queue, and a comprehensive score is calculated sequentially based on timeliness, compliance, and dissemination effect. Specifically, this embodiment first reads the generation timestamp corresponding to the governance result and compares it with the current time to obtain the time difference, which is used to characterize the time decay of the governance result. Based on a preset time decay constant, which is set according to the sensitivity of different data types to timeliness requirements, this embodiment calculates the corresponding timeliness decay factor using the aforementioned time difference, reflecting the timeliness decay trend of the governance result. The timestamp information is directly read from the governance result queue, the current time is provided by the system clock, and the time decay constant is a fixed value set based on historical data experience, requiring no manual adjustment.

[0126] Furthermore, this embodiment calls the established compliance ontology model to perform rule verification on the data stream corresponding to each governance result, counts the number of violation clauses triggered during the governance process, and calculates the corresponding compliance penalty factor based on the number of triggers to reflect the compliance risk level of the data stream. Regarding the evaluation of the dissemination effect, this embodiment obtains the click volume, forwarding volume, and audience dwell time data of the data stream through a third-party data interface connected to the target audience platform. This data is used to measure its dissemination effect and generate a dissemination effect index accordingly. Finally, this embodiment performs unified normalization processing on the timeliness decay factor, compliance penalty factor, and dissemination effect index to calculate a comprehensive score. This comprehensive score serves as the basis for subsequent governance strategy optimization and data channel division.

[0127] In this embodiment, preferably, after each governance cycle, the embodiment collects the comprehensive score of each autonomous agent in the current cycle, as well as the governance decision parameters used by each autonomous agent in that cycle, and uniformly constructs a historical sample set. The comprehensive score is calculated based on the aforementioned timeliness, compliance, and propagation effect, and is directly obtained from the output of the evaluation module. The governance decision parameters of each autonomous agent are derived from the decision module currently being executed, and these parameters are used to characterize the state of the governance strategy network in that cycle. This embodiment establishes a correspondence between the comprehensive score and the governance parameters, continuously accumulating them periodically to form a historical sample dataset, which is used to support the subsequent strategy optimization process.

[0128] Furthermore, this embodiment constructs a Gaussian process proxy model based on the historical sample set. This proxy model aims to minimize the governance effect and strategy discrepancy, evaluates the impact of different parameter configurations on the governance effect, and defines a multi-objective expected improvement function to balance governance effect and compliance risk. Guided by this proxy model, this embodiment automatically solves for the optimal parameter adjustment increment using a Bayesian optimization method, combined with the prediction performance of each historical sample in the model. The step size adjustment coefficient is adaptively set based on the proxy model's prediction variance of the samples, ensuring the stability and effectiveness of small step size updates. Finally, the obtained updated parameters are written back to the decision modules of each corresponding autonomous agent, replacing the original parameters to achieve adaptive migration and continuous optimization of the governance strategy in the next cycle.

[0129] Furthermore, the comprehensive score in this embodiment is obtained in real time through the timeliness, compliance, and dissemination effect of the governance results. The decision parameters are derived from the policy network output used by each autonomous agent in the current cycle, while the historical sample set is automatically accumulated according to the governance cycle. The training data for the Gaussian process proxy model is directly provided by the aforementioned historical sample set, and the hyperparameters in the model are automatically optimized based on the fitting effect of the sample set through cross-validation or maximum likelihood estimation. The multi-objective expectation improvement function combines the balance objectives of governance effect and compliance risk, and is automatically determined by the prediction error and distribution fitted by the proxy model based on historical data. The step size adjustment coefficient is adaptively set according to the uncertainty of the proxy model's prediction of the current parameters, i.e., the variance, without the need for a preset fixed value.

[0130] In this embodiment, preferably, after completing the calculation of the comprehensive score of the governance results and updating the parameters of the decision module, this embodiment further synchronously generates a traceable report containing the governance results of the data streams, rule triggering items, autonomous agent status, and parameter update records. The report summarizes the final processing results of each data stream in the governance result queue, records the applicable governance instructions, the corresponding triggered compliance rule clause number and trigger count, the status summary of the autonomous agent within this period, including strategy parameters, performance and status synchronization information, and historical sample data and final update results during the governance strategy parameter update process of this period. All of the above data is uniformly packaged into a structured report file, associated with a timestamp and a unique data stream identifier, and stored in an audit archive database. This provides data support for external compliance audits, effect tracking, and subsequent iterative optimization, ensuring that the entire governance process is transparent, verifiable, and reproducible.

[0131] Corresponding to the above methods, such as Figure 2 As shown, this embodiment also provides a dynamic management system for international communication data streams based on agent-based intelligent agents, including:

[0132] The data acquisition and annotation unit is used by distributed crawler nodes to dynamically allocate crawling tasks based on a scalable hash algorithm to capture cross-platform, multilingual data streams, and to attach source identifiers, timestamps, topic tags and sensitivity weights to each data stream to form a raw data pool with semantic indexes.

[0133] An agent construction and coordination unit is used to instantiate autonomous agents for each of the topic tags in the original data pool; the autonomous agent includes a perception module, a decision module, and an execution module; the perception module uses a hybrid Transformer-CNN network to encode the multimodal content of the original data pool; the decision module outputs governance instructions through a multi-armed contextual gambling machine-policy gradient joint network and in conjunction with an international compliance rule base; each of the autonomous agents exchanges state summaries and issues synchronization instructions via a message bus, and maintains global policy consistency with the help of a consensus algorithm;

[0134] The data governance execution unit is used to perform processing operations, including filtering, desensitization, rate limiting, reordering, and multilingual synchronization, on the original data stream according to the governance instructions by the execution module, and to obtain the governance result.

[0135] The priority channel partitioning unit is used to write the governance results into the governance result queue and, based on the compliance ontology model, divide the data flow in the governance result queue into fast channel, review channel and cold storage channel according to priority.

[0136] The comprehensive evaluation unit is used to extract governance results from the governance result queue after each governance cycle ends, and calculate a comprehensive score based on timeliness, compliance and dissemination effect;

[0137] The strategy optimization and update unit is used to update the parameters of the decision module in small steps based on the comprehensive score, using a multi-objective Bayesian optimization-driven model irrelevant meta-learning algorithm, and write the updated parameters back to each of the autonomous agents to achieve adaptive migration and optimization of the governance strategy.

[0138] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple; relevant parts can be referred to the method section.

[0139] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for dynamic management of international communication data streams based on agent-based intelligent agents, characterized in that, include: Distributed crawler nodes dynamically allocate crawling tasks based on a scalable hash algorithm to capture cross-platform, multilingual data streams, and attach source identifiers, timestamps, topic tags, and sensitivity weights to each data stream to form a raw data pool with semantic indexes; Autonomous agents are instantiated for each topic tag in the original data pool; each autonomous agent includes a perception module, a decision-making module, and an execution module; the perception module encodes the multimodal content of the original data pool using a hybrid Transformer-CNN network; the decision-making module outputs governance instructions through a multi-armed contextual gambling machine-policy gradient joint network and in conjunction with an international compliance rule base; each autonomous agent exchanges state summaries and issues synchronization instructions via a message bus, and maintains global policy consistency with the help of a consensus algorithm; The execution module performs processing operations, including filtering, desensitization, rate limiting, reordering, and multilingual synchronization, on the original data stream according to the governance instructions to obtain the governance result; The governance results are written into the governance results queue, and the data flow in the governance results queue is divided into fast channel, review channel and cold storage channel according to priority based on the compliance ontology model; At the end of each governance cycle, the governance results are extracted from the governance result queue, and a comprehensive score is calculated based on timeliness, compliance, and dissemination effect. Based on the comprehensive score, a multi-objective Bayesian optimization-driven model irrelevant meta-learning algorithm is used to update the parameters of the decision module in small steps, and the updated parameters are written back to each of the autonomous agents to achieve adaptive migration and optimization of the governance strategy.

2. The method for dynamic management of international communication data streams based on agent-based intelligent agents according to claim 1, characterized in that, Distributed crawler nodes dynamically allocate crawling tasks based on a scalable hash algorithm to capture cross-platform, multilingual data streams. Each data stream is appended with a source identifier, timestamp, topic tag, and sensitivity weight, forming a raw data pool with semantic indexing, including: Obtain a preset data source; the data source includes news websites, social platforms, video platforms, content delivery networks, open data interfaces and other information publishing channels, and each data source corresponds to an access address, data format and access protocol; Based on the access domain name, IP address and interface call information of the data source, a unique source identifier is generated, and the platform type, language type and time zone information of the corresponding data source are recorded as the metadata basis; Based on the computing power, network bandwidth, and historical load of each crawler node, a scalable hash algorithm is used to dynamically divide the task domain, and the data source is mapped and distributed to different distributed crawler nodes according to platform type, language type, and time zone information. Each of the distributed crawler nodes captures the target data stream in real time based on the task domain rules, and adds the source identifier and collection timestamp information to each of the collected target data streams; For the collected target data stream, topic tags are extracted through natural language processing and topic modeling, and sensitivity weights are calculated in combination with the content sensitivity model; The source identifier, the collection timestamp information, the topic tag, the sensitivity weight, and the target data stream are encapsulated into structured metadata and uniformly written into the original data pool.

3. The method for dynamic management of international communication data streams based on agent-based intelligent agents according to claim 1, characterized in that, The perception module uses a hybrid Transformer-CNN network to encode the multimodal content of the original data pool, including: For the data streams associated with each of the aforementioned topic tags in the original data pool, text data, audio data, video data, and image data are extracted from the data streams to form a multimodal dataset; For the text data, a pre-trained Transformer encoding model is used to perform context modeling, and the output text feature vector containing semantic relationships is generated. For the audio data and the video data, first extract the corresponding audio feature sequence and video frame image sequence, then input them into a convolutional neural network to extract temporal and spatial local features, and output audio feature vectors and video feature vectors; The image data is directly input into a convolutional neural network to extract spatial hierarchical features and output an image feature vector. The text feature vector, audio feature vector, video feature vector and image feature vector are concatenated to construct a unified multimodal semantic feature matrix; The multimodal semantic feature matrix is ​​input into the fusion layer, and the cross-modal correlation is further modeled using the fusion Transformer encoding layer to output a unified multimodal semantic vector.

4. The method for dynamic management of international communication data streams based on agent-based intelligent agents according to claim 1, characterized in that, The decision-making module outputs governance instructions through a multi-armed contextual gambling machine-policy gradient joint network and in conjunction with an international compliance rule base, including: Receive the multimodal unified semantic vector s output by the perception module, and extract the compliance constraint embedding vector c corresponding to the current propagation scenario from the international compliance rule library; The multimodal unified semantic vector s is concatenated with the compliance constraint embedding vector c to form a context vector, which serves as the state input for the multi-armed context gambling machine; For each governance action a in action set A, explore using a multi-armed contextual gambling machine – use the score and select several candidate actions; The candidate action, along with the context vector, is fed into the policy gradient network. The compliance-weighted policy gradient is calculated according to the following formula, and the policy parameters are updated: Where θt is the policy network parameter vector; η is the learning rate; rt is the real-time propagation reward; λ is the compliance penalty coefficient; l t For compliance and violation level; a t For the selected governance action; s t c is the perception state vector; t For compliant embedding vectors; π θ (·) represents the parameterized strategy probability distribution; For gradient operators; Based on the updated policy distribution, candidate actions are re-evaluated, and the optimal action is selected as the final governance instruction and output to the execution module. At the same time, the current round's state-action-benefit-penalty quadruple is written into the experience replay buffer.

5. The method for dynamic management of international communication data streams based on agent-based intelligent agents according to claim 1, characterized in that, Each of the autonomous agents exchanges state summaries and issues synchronization instructions via a message bus, maintaining global policy consistency through a consensus algorithm, including: Each autonomous agent generates a state summary after completing its local governance decision; the state summary includes at least the current local policy vector, average reward value and timestamp, and is written to the message bus via publish and subscribe. Each of the autonomous agents subscribes to external state summaries related to its own topic tags from the message bus to construct a neighborhood state set; For the aforementioned neighborhood state set, a global consistency strategy is calculated using a weighted consensus algorithm. The specific update rules are as follows: in, Let N be the local policy vector of agent i in the k-th round; α be the neighborhood fusion coefficient; N be the local policy vector of agent i in the k-th round; α be the neighborhood fusion coefficient; N be the local policy vector of agent i in the k-th round; α be the local policy vector of agent i in the k-th round ... α be the local policy vector of agent i i Let w be the set of neighborhoods visible to agent i on the message bus; ij Let the trust weights from agent j to agent i satisfy... Write the obtained consistent policy vector into the local policy cache, and generate a synchronization instruction to broadcast to the execution module, so that each autonomous agent executes the unified governance policy in the next decision cycle.

6. The method for dynamic management of international communication data streams based on agent-based intelligent agents according to claim 1, characterized in that, Use the execution module to perform processing operations including filtering, desensitization, speed limiting, reordering, and multilingual synchronization on the original data stream according to the governance instruction, and obtain governance results, including: Parse the governance instruction, extract the operation type, execution order, target field, and language label, and construct an operation list; Based on the keywords, regular expressions, and blacklist / whitelist rules in the operation list, content matching is performed on the text, image, and audio content in the original data stream to delete or block content related to the violation topic, resulting in a filtered data stream D. filter ; For the filtered data stream D filter Fields involving identity information, geolocation information, and device identification information are anonymized. Sensitivity levels are set to implement tiered labeling and replacement, resulting in an anonymized data stream D. desens ; Based on the bandwidth threshold and time window specified in the governance instructions, the de-identification result data stream D... desens Implement traffic shaping, dynamically adjust the data packet sending rate, and generate a rate-limited data stream D. throttle ; Based on the priority information and timestamp in the governance instruction, the rate-limited data stream D is... throttle Reorder the data to obtain the sorted data stream D. sorted High-priority data packets are sent first, while low-priority data packets are delayed and queued. When the operation list contains multilingual synchronization markers, based on the sorted data stream D sorted The text content is processed by a neural machine translation model and a terminology database to generate a target language version aligned with the original language content. This target language version is then encapsulated using the original media format to obtain a multilingual synchronous data stream D. multi ; The filtered data stream D filter Desensitization result data stream D desens Data stream D after rate limiting throttle Sorted data stream D sorted and multilingual synchronous data stream D multi The latest available data is used as the final data stream, which is then combined with operation records, processing tags, and compliance status to summarize and encapsulate the governance results.

7. The method for dynamic management of international communication data streams based on agent-based intelligent agents according to claim 1, characterized in that, Write the governance result into the governance result queue, and divide the data stream in the governance result queue into a fast channel, a review channel, and a cold storage channel according to the compliance ontology model by priority, including: When generating the governance result, attach a compliance risk score, a timeliness urgency degree, and a content popularity prediction value to it, and write it into the governance result queue; For each governance result in the governance result queue, call the compliance ontology model to verify the number of triggered clauses, and calculate the priority score accordingly S = tanh(R + T - E - C) Where, R is the compliance risk score; T is the timeliness urgency degree; E is the content popularity prediction value; C is the number of triggered compliance clauses; S is the priority score normalized by the hyperbolic tangent function; Perform channel division according to the priority score S and the preset threshold set {s1, s2}: Allocate to the fast channel when s1 ≤ S; allocate to the review channel when s2 ≤ S < s1; allocate to the cold storage channel when S < s2; Update the index of the channel label together with the governance result in the queue.

8. The method for dynamic management of international communication data streams based on agent-based intelligent agents according to claim 1, characterized in that, After each governance cycle ends, extract the governance result from the governance result queue, and calculate the comprehensive score according to timeliness, compliance, and dissemination effect, including: Read the generation timestamp of the governance result, and calculate the timeliness decay factor H based on the time difference Δt from the current moment. The calculation method is: H=e -Δt / τ Where, τ is the time decay constant, Δt is the time difference between the current moment and the generation time of the governance result, and H represents the timeliness decay factor; Based on the number of violated clauses C triggered by the governance result under the compliance ontology model, calculate the compliance penalty factor P. The calculation method is: Obtain the click volume V, forwarding volume F, and average stay time D of the data stream corresponding to the governance result among the target audience, and calculate the dissemination effect index E. The calculation method is: Integrate each index and calculate the comprehensive score S according to the following formula: S = σ(ln(1 + E)H - P) Where, S is the comprehensive score, E is the dissemination effect index, H is the timeliness decay factor, P is the compliance penalty factor, and σ is the logistic function.

9. The method for dynamic management of international communication data streams based on agent-based intelligent agents according to claim 1, characterized in that, Based on the comprehensive score, use the multi-objective Bayesian optimization-driven model-agnostic meta-learning algorithm to perform a small-step update on the parameters of the decision module, and write the updated parameters back to each autonomous agent to achieve the adaptive migration and optimization of the governance policy, including: At the end of each governance cycle, the comprehensive score S corresponding to each autonomous agent is collected. i and its current decision parameters Constructing a historical sample set Among them, S i Let i be the overall score of the i-th autonomous agent. Let be the decision parameter vector of the i-th autonomous agent at time t; For the historical sample set D, a Gaussian process surrogate model is trained to minimize the objective min(1-S). And based on the Gaussian process proxy model Define the multi-objective expected hypervolume improvement function EHVI(θ); Obtain the optimal increment through multi-objective Bayesian optimization. The formula is: Dth * =argmax Δθ EHVI(θ (t) +Δθ) Where, Δθ * To maximize the optimal parameter increment obtained by EHVI(θ); Perform a small-step migration using the model-agnostic meta-learning update rule. The formula is: i (t+1) =θ (t) +ηΔθ * Where η is the step size coefficient adaptively calculated from the average prediction variance of the surrogate model; θ (t+1) This is the decision parameter vector updated with a small step size; Update the parameter θ (t+1) The data is written back to the decision-making module corresponding to the autonomous agent to complete the policy adaptive optimization.

10. A dynamic management system for international communication data streams based on agent-based intelligent agents, characterized in that, Including: The data acquisition and annotation unit is used by distributed crawler nodes to dynamically allocate crawling tasks based on a scalable hash algorithm to capture cross-platform, multilingual data streams, and to attach source identifiers, timestamps, topic tags and sensitivity weights to each data stream to form a raw data pool with semantic indexes. An agent construction and coordination unit is used to instantiate autonomous agents for each of the topic tags in the original data pool; the autonomous agent includes a perception module, a decision module, and an execution module; the perception module uses a hybrid Transformer-CNN network to encode the multimodal content of the original data pool; the decision module outputs governance instructions through a multi-armed contextual gambling machine-policy gradient joint network and in conjunction with an international compliance rule base; each of the autonomous agents exchanges state summaries and issues synchronization instructions via a message bus, and maintains global policy consistency with the help of a consensus algorithm; The data governance execution unit is used to perform processing operations, including filtering, desensitization, rate limiting, reordering, and multilingual synchronization, on the original data stream according to the governance instructions by the execution module, and to obtain the governance result. The priority channel partitioning unit is used to write the governance results into the governance result queue and, based on the compliance ontology model, divide the data flow in the governance result queue into fast channel, review channel and cold storage channel according to priority. The comprehensive evaluation unit is used to extract governance results from the governance result queue after each governance cycle ends, and calculate a comprehensive score based on timeliness, compliance and dissemination effect; The strategy optimization and update unit is used to update the parameters of the decision module in small steps based on the comprehensive score, using a multi-objective Bayesian optimization-driven model irrelevant meta-learning algorithm, and write the updated parameters back to each of the autonomous agents to achieve adaptive migration and optimization of the governance strategy.

Images