Multi-agent-based urban international propagation influence evaluation system and evaluation method

Abstract

The invention discloses a multi-agent-based urban international propagation influence evaluation system and evaluation method, and the system comprises a data perception layer which comprises a plurality of special data collection agents; the analysis decision-making layer is connected with the data sensing layer and is used for carrying out collaborative analysis and comprehensive evaluation on the preprocessed urban international propagation related data; and the report generation layer is connected with the analysis decision-making layer. The evaluation method comprises the following steps: realizing collaborative perception, intelligent analysis and decision-making of multi-source data through a hierarchical collaborative multi-agent architecture; meanwhile, a consensus voting algorithm and a historical case library are utilized to effectively solve the problem of conflict resolution in cross-culture assessment, and finally a dynamic assessment report with causal interpretation is generated. According to the method, collaborative perception, intelligent analysis and decision making of multi-source data are realized, and the problem of conflict resolution in cross-culture assessment is effectively solved.

Description

A Multi-Agent-Based Assessment System and Methodology for Evaluating the International Communication Influence of Cities Technical Field

[0001] This invention belongs to the field of deep interdisciplinary technology of artificial intelligence and urban communication studies, specifically involving a multi-agent-based urban international communication influence assessment system and assessment method. Background Technology

[0002] Against the backdrop of deepening globalization, the assessment of a city's international communication influence has become a core element in enhancing its soft power and optimizing its public diplomacy strategies. Technological exploration and practice in related fields have always been driven by the goal of "responding to actual needs more accurately and efficiently." Currently, two main approaches have emerged: human expert assessment and single algorithm models. Both have played a positive role in their respective application scenarios, but they also face challenges that require further optimization.

[0003] Among them, the human expert evaluation scheme (such as the questionnaire survey and media coverage statistics combined model adopted by some authoritative institutions) relies on the domain experience of professional teams and can ensure the depth and comprehensiveness of the evaluation logic. However, its dependence on static data means that the evaluation cycle usually takes several weeks (4-8 weeks in some complex scenarios), which makes it difficult to meet the dynamic response needs in scenarios such as sudden public opinion events and temporary international exchange events (such as transnational cultural exhibitions and temporary diplomatic meetings). At the same time, the subjective differences in human judgment may also affect the quantitative consistency of the evaluation results to a certain extent, which restricts the large-scale standardized evaluation.

[0004] The application of single-algorithm models (such as mainstream public opinion analysis tools) has effectively broken through the efficiency bottleneck of manual processing and provided technical support for the rapid analysis of tens of millions of data points. However, there is still room for improvement in adapting to the complex needs of urban international communication: First, the data coverage dimensions need to be expanded. Most current tools focus on processing English social media data and lack integration of key information sources such as non-Western media (such as Al Jazeera, RT, etc.), diplomatic archives, and regional academic literature, which can easily lead to an incomplete evaluation perspective. Second, the decision-making adaptability needs to be further enhanced. When there is a conflict between social media popularity and policy influence (differences between public opinion popularity and policy adoption rates in specific international events), the system often cannot resolve the contradictions autonomously and requires manual intervention for calibration, which weakens the advantages of automation to some extent. Third, the scenario adaptability of the report output needs to be improved. The templated results are difficult to fully analyze complex factors such as cultural differences (such as differences in the acceptance of "urban development issues" in different regions) and lack causal logic explanations for the evaluation conclusions. This makes the evaluation error rate in cross-cultural scenarios often remain above 30%, which is difficult to meet the refined needs of urban international communication strategy formulation.

[0005] In summary, there is currently no assessment scheme in the field that can deeply integrate knowledge from the field of international communication, efficiently process multi-source heterogeneous data, and have the ability to resolve conflicts autonomously. This makes it difficult to fully meet the comprehensive needs of urban international communication scenarios for "efficient response, accurate quantification, and cultural adaptation." Therefore, further optimizing the adaptability and intelligence level of assessment technology still has important practical significance and value. Summary of the Invention

[0006] The purpose of this invention is to provide a multi-agent-based system and method for assessing the international communication influence of cities. This invention achieves collaborative perception, intelligent analysis, and decision-making on multi-source data through a hierarchical and collaborative multi-agent architecture. Simultaneously, it effectively resolves conflict resolution issues in cross-cultural assessments by utilizing consensus voting algorithms and a historical case database, ultimately generating a dynamic assessment report with causal explanatory power.

[0007] The technical solution of this invention: A city's international communication influence assessment system based on multi-agent intelligence, comprising: a data perception layer, which includes multiple dedicated data acquisition agents for real-time acquisition and preprocessing of city international communication-related data from multiple heterogeneous data sources; an analysis and decision-making layer, connected to the data perception layer, for collaborative analysis and comprehensive evaluation of the preprocessed city international communication-related data; the analysis and decision-making layer includes: multiple domain analysis agents for parallel analysis and local evaluation of the data, and outputting local evaluation values ​​and corresponding confidence levels; and a central decision-making module, connected to the multiple domain analysis agents, for... The central decision-making module receives the local evaluation values ​​and confidence levels of each local evaluation value. It includes: a consensus voting module for weighted fusion based on the local evaluation values ​​and confidence levels, and triggering a conflict resolution mechanism when conflicts are detected between local evaluation values; a historical case library storing records of historical international communication events; when the conflict resolution mechanism is triggered, the central decision-making module calls the historical case library for analogical reasoning to resolve conflicts and generate a comprehensive evaluation result; and a report generation layer connected to the analysis and decision-making layer, including a natural language generation engine and a visualization module, for automatically generating a structured evaluation report based on the comprehensive evaluation result.

[0008] In the aforementioned multi-agent-based urban international communication influence assessment system, the dedicated data collection agents include social media crawler agents, policy database docking agents, and news semantic parsing agents; the heterogeneous data sources include social media data sources, international news data sources, diplomatic archive data sources, and academic literature data sources; and the urban international communication-related data include multilingual data, cross-media text data, diplomatic archive data, and treaty data.

[0009] In the aforementioned multi-agent-based urban international communication influence assessment system, the domain analysis agents include an influence calculation agent for quantifying urban node centrality, a cultural fit assessment agent, and a public opinion sentiment analysis agent for identifying media biases.

[0010] In the aforementioned multi-agent-based urban international communication influence assessment system, the historical case database stores historical case records of multiple international communication events, and each record contains at least a historical event feature vector and a corresponding conflict resolution.

[0011] The aforementioned evaluation method for a multi-agent-based urban international communication influence assessment system includes the following steps: Step S1: Collecting multi-source heterogeneous data through multiple dedicated data acquisition agents in the data perception layer and preprocessing it; Step S2: Performing parallel analysis and local evaluation on the preprocessed urban international communication-related data through multiple domain analysis agents in the analysis and decision-making layer to obtain local evaluation values ​​and corresponding confidence levels; Step S3: Using the consensus voting algorithm in the consensus voting module, weighted fusion of each local evaluation value and corresponding confidence level is performed. When the consensus voting algorithm detects a conflict between local evaluation values, it calls the historical case library for analogical reasoning to resolve the conflict and form a comprehensive evaluation result; Step S4: Using the natural language generation engine and visualization module in the report generation layer, the comprehensive evaluation result is transformed into a structured report containing causal explanations and recommendations and output.

[0012] In the aforementioned evaluation method, the influence calculation agent focuses on preprocessed urban international communication-related data, quantifies the node centrality in the data, and outputs a local evaluation value and corresponding confidence level for the urban international communication influence dimension. The cultural adaptability evaluation agent calls the Hofstede model to analyze the cross-cultural acceptability of the urban international communication-related data content, and outputs a local evaluation value and corresponding confidence level for the cross-cultural acceptability of content by regional cultural groups. The public opinion sentiment analysis agent uses the LSTM-Attention model to identify implicit biases in the preprocessed urban international communication-related data, and corrects the sentiment tendency judgment based on the biases, thereby outputting a local evaluation value and corresponding confidence level for the international public opinion sentiment tendency dimension.

[0013] In the aforementioned evaluation method, step S3, the consensus voting algorithm includes the following steps: Step S3.1: Receive the local evaluation values ​​output by each domain analysis agent. and confidence level ; Evaluation values ​​for each local area Perform weighted fusion to obtain the initial weighted score. Step S3.2: Compare each local evaluation value with the initial weighted score. If the maximum deviation exceeds the dynamic confidence threshold, a conflict is identified and a resolution mechanism is triggered. The historical case library is called for analogical reasoning to generate a weight adjustment factor. Update weights Recalculate the final comprehensive evaluation results .

[0014] In the aforementioned evaluation method, in step S3.2, the analogical reasoning retrieves the Top-K most similar historical case records by calculating the weighted similarity between the current event feature vector and the historical event feature vector, and generates the weight adjustment factor based on the conflict resolution of these historical case records.

[0015] In the aforementioned evaluation method, the current event feature vector and the historical event feature vector contain three core dimensions: event type, cultural background label, and data source distribution entropy value; the weighted similarity calculation adopts a linear combination of weighted cosine similarity and Jaccard keyword similarity.

[0016] In the aforementioned evaluation method, in step S4, the natural language generation engine generates a causal chain explanation based on the comprehensive evaluation results and the communication rule base, and the visualization module dynamically generates a heat map and confidence interval annotations, outputting an interactive structured report.

[0017] Compared with existing technologies, the present invention has the following beneficial effects: 1. The data perception layer of the present invention integrates multi-source heterogeneous data through multiple dedicated data acquisition agents, breaking through the limitations of a single data source, ensuring the comprehensiveness of the evaluation perspective, and improving efficiency through parallel acquisition and preprocessing, adapting to dynamic evaluation needs; the analysis and decision-making layer relies on the collaborative analysis of multi-domain analysis agents, combined with the consensus voting of the central decision-making module and the historical case library, to autonomously detect and resolve evaluation conflicts without human intervention, reducing subjective errors and improving evaluation accuracy; the report generation layer uses a natural language generation engine and visualization module to automatically generate structured reports with causal explanations, eliminating template-based presentations, reducing understanding costs, and providing practical decision support for cities to optimize their international communication strategies.

[0018] 2. This invention integrates heterogeneous data sources such as social media, international news, diplomatic archives, and academic literature through three types of dedicated intelligent agents: social media crawlers, policy database integration, and news semantic parsing. It covers data types such as multilingual and cross-media texts and relies on a federated learning framework to complete data preprocessing. While ensuring data comprehensiveness, it protects sensitive information through a differential privacy mechanism, avoids single points of failure, and solves the dual problems of "one-sided evaluation perspective" and "data security risks".

[0019] 3. This invention analyzes the decision-making level through three types of intelligent agents: influence calculation, cultural fit assessment, and public opinion sentiment analysis, to achieve multi-dimensional parallel evaluation. Combining consensus voting algorithms and historical case databases, it can autonomously detect and evaluate conflicts. By generating weight adjustment factors through analogical reasoning combining weighted cosine similarity and Jaccard keyword similarity, conflicts can be resolved without human intervention, thus significantly reducing cross-cultural evaluation errors.

[0020] 4. This invention uses a natural language generation engine combined with a communication studies rule base to generate reports containing causal chain explanations and strategic suggestions. It also features a visualization module to generate heatmaps and confidence interval annotations, and supports interactive drill-down of raw data. Compared to templated reports, this invention better meets the needs of refined strategy formulation in cross-cultural scenarios and helps users quickly understand evaluation conclusions. Attached Figure Description

[0021] Figure 1 is a flowchart of the method of the present invention. Detailed Implementation

[0022] The present invention will be further described below with reference to the accompanying drawings and embodiments, but this should not be construed as limiting the present invention.

[0023] Example: A multi-agent-based system for assessing the international communication influence of cities, comprising: a data perception layer, which includes multiple dedicated data collection agents, including a social media crawler agent, a policy database interface agent, and a news semantic analysis agent. All dedicated data collection agents adopt a unified interface specification (standardized input / output data structures) and a differentiated kernel design. Each agent inherits from the base class AgentBase, but designs core algorithms according to the task: the social media crawler agent integrates a multilingual BERT-Mini model, supports real-time sentiment analysis in 50+ languages, and balances data bias through dynamic sampling strategies (e.g., increasing the sampling weight of Arabic posts by 30%); the policy database interface agent embeds a diplomatic knowledge graph, automatically associates UN treaty clauses with city policy texts, and introduces a geopolitical attenuation factor when calculating policy influence. The specific formula is as follows: ;in, For policy similarity, For national diplomatic distance; The attenuation coefficient is the attenuation coefficient. The values ​​are determined based on the following: 1. Historical data fitting: By using regression analysis of international communication cases in 300+ cities worldwide from 2016 to 2024, a mapping relationship between "diplomatic distance and policy influence attenuation rate" is established; 2. Domain expert experience: Adjustments are made for different event types (such as sports events and cultural exhibitions). In this example, the value needs to be increased by 20%-30% for "cultural communication" events (because cultural differences have a more significant impact on policy acceptance); 3. Adaptation to city characteristics: A smaller value (0.1-0.3) is used for cities with frequent international exchanges (such as Shanghai and Guangzhou), and a smaller value is used for cities that are conducting international communication for the first time. Take the larger value (0.6-1.0).

[0024] All agents are trained collaboratively using a federated learning framework: Local nodes encrypt the raw data and only upload gradient updates to the central server, employing a differential privacy mechanism. Protect sensitive information and avoid single points of failure.

[0025] To address the characteristics of international communication, the system pre-configures a domain knowledge base: A cultural dimension engine integrates Hofstede's six cultural dimensions (such as the Power Distance Index (PDI) and Uncertainty Avoidance Index (UAI)) to quantify content suitability, automatically increasing the weight of the "Collective Index" for content disseminated to the Middle East; a multilingual alignment pipeline uses the mBART-50 model to achieve cross-lingual semantic alignment, resolving the semantic conflict between the Chinese character for "dragon" (an auspicious symbol) and the Western character for "evil," filtering cultural noise through an attention masking mechanism; a dynamic data source weighting module automatically adjusts the credibility weight of data sources based on the strength of the city's diplomatic relations (such as the depth of cooperation with ASEAN countries), assigning a 1.2 times base weight to reports from Singapore's *Lianhe Zaobao*. A data perception layer is used to collect and preprocess city-related international communication data in real time from multiple heterogeneous data sources. These heterogeneous data sources include social media data sources, international news data sources, diplomatic archive data sources, and academic literature data sources; the city-related international communication data includes multilingual data, cross-media text data, diplomatic archive data, and treaty data.

[0026] The analysis and decision-making layer, connected to the data perception layer, is used for collaborative analysis and comprehensive evaluation of preprocessed urban international communication-related data. The analysis and decision-making layer includes multiple domain analysis agents and a central decision-making module. The domain analysis agents include an influence calculation agent, a cultural fit assessment agent, and a public opinion sentiment analysis agent. The central decision-making module has a built-in consensus voting module and a historical case database. The historical case database stores historical case records of multiple international communication events, with each record containing at least a historical event feature vector and a corresponding conflict resolution solution. In this embodiment, the historical case database stores international events from over 300 cities worldwide from 2016 to 2024, with each record containing an event feature vector. ,in, Event type (sports / culture). Based on Hofstede's cultural background tags, The data source distribution entropy value. Conflict resolution includes historical weight adjustment records. .

[0027] The report generation layer, which is connected to the analysis and decision-making layer, includes a natural language generation engine and a visualization module. The report generation layer is used to automatically generate structured evaluation reports based on the comprehensive evaluation results.

[0028] The evaluation method of the city's international communication influence assessment system based on multi-agents is shown in Figure 1, and includes the following steps: Step S1: Collect multi-source heterogeneous data through multiple dedicated data acquisition agents in the data perception layer and perform preprocessing.

[0029] In this embodiment, taking the "Hangzhou City 2024 Subsequent International Influence Assessment" as an example, the system is deployed on an Alibaba Cloud ECS cluster (4 NVIDIA A100 80GB GPU servers), and the data sources cover Twitter, TikTok, Reuters and Xinhua.

[0030] A social media crawler agent crawls #Hangzhou2024 posts every 5 minutes, filtering noise (such as removing ad posts) using multilingual BERT, and increasing the sampling weight of Arabic comments by 30%; a news semantic parsing agent connects to 15 international media APIs, using the mBART-50 model to align the semantics of Chinese and English reports, mapping the Chinese concept of "dragon boat culture" to the Western concept of "teamwork spirit"; a policy database-connected agent extracts relevant indicators, and introduces a geopolitical attenuation factor when calculating policy influence. Based on the China-ASEAN diplomatic distance. All data is encrypted through federated learning and then standardized at the edge nodes (timestamp alignment, language unification); since the edge nodes only upload gradient updates to the central server, the leakage of sensitive data can be effectively avoided.

[0031] Step S2: By analyzing multiple domain-specific intelligent agents in the decision-making layer, parallel analysis and local evaluation are performed on the preprocessed urban international communication-related data to obtain local evaluation values. In this embodiment, the analysis and local evaluation process is as follows: The influence calculation intelligent agent, based on the improved PageRank algorithm, focuses on the preprocessed urban international communication-related data, quantifies the node centrality in the urban international communication-related data, and outputs the local evaluation value and corresponding confidence level of the urban international communication influence dimension; The cultural adaptability evaluation intelligent agent calls the Hofstede model to analyze the cross-cultural acceptability of the urban international communication-related data content, analyzes the difference in the acceptability of "ink painting elements" in Europe and America (low collective) and East Asia (high collective), and outputs the local evaluation value and corresponding confidence level of the cross-cultural acceptability of content by regional cultural groups; The public opinion sentiment analysis intelligent agent uses the LSTM-Attention model to identify implicit biases in the preprocessed urban international communication-related data, such as the stereotypical description of "Asian event organization capabilities" by Western media, and corrects the sentiment tendency judgment based on the bias, thereby outputting the local evaluation value and corresponding confidence level of the international public opinion sentiment tendency dimension.

[0032] Step S3: The consensus voting (CVA) algorithm in the consensus voting module is used to weight and fuse the local evaluation values. When the consensus voting algorithm detects a conflict, it calls the historical case library for analogical reasoning to resolve the conflict and form a comprehensive evaluation result. In this step, the consensus voting algorithm includes the following steps: Step S3.1: Receive the local evaluation values ​​output by each domain analysis agent. and confidence level ; Evaluation values ​​for each local area Perform weighted fusion to obtain the initial weighted score. Wherein, the local evaluation value and The dimensionless range is 0-100 points.

[0033] Step S3.2: Compare each local evaluation value with the initial weighted score, using the following formula: ;in, The confidence threshold is initially set at 0.85. If the maximum deviation exceeds the dynamic confidence threshold, a conflict is identified and a resolution mechanism is triggered. This mechanism calls upon the historical case library for analogical reasoning and generates a weight adjustment factor. Update weights Recalculate the final comprehensive evaluation results ,in, The range is 0-100 points.

[0034] The analogical reasoning retrieves the top-K most similar historical case records by calculating the weighted similarity between the current event feature vector and the historical event feature vectors, and generates the weight adjustment factor based on the conflict resolution methods of these historical case records. Specifically, the similarity calculation uses weighted cosine similarity. : ;in, This represents the feature weight, with a value of 0.7. This represents the keyword weight, with a value of 0.3. This is the feature vector of the current event. For historical event feature vectors, The cosine similarity represents the feature vector of the current event and the feature vector of historical events. Indicates the similarity of Jaccard keywords. This represents the set of keywords for the current event and historical cases. Although the cosine similarity and Jaccard keyword similarity calculation objects are different, the scale of the output results is the same. The value range of both is between [0,1] (cosine similarity measures the closeness of vector directions, the closer to 1, the more similar; Jaccard keyword similarity measures the ratio of the intersection and union of sets, also from 0 to 1).

[0035] Feature weights and keyword weight The setting process is based on the relative importance of the two types of information in event similarity matching. Generally, the "feature vector of the event (such as structured information like attributes and parameters)" is considered the core basis for matching and has a more critical impact on similarity, so it is given higher weight. Keywords (unstructured tag information) are supplementary information and have a relatively weaker impact, so they are given a lower weight. Feature weights and keyword weight It can also be adjusted and optimized through domain experience and experiments (such as testing the matching accuracy under different weights).

[0036] After retrieving the top-3 similar historical case records, a weighted voting process is used to generate the final version. ;in, The system normalizes the weights based on case similarity. When assessing the spread of festivals, the system automatically matches the "Istanbul Ramadan incident" case, reducing the weight of social media by 0.4 and increasing the weight of institutional data by 0.6. Institutional data is sourced from diplomatic archives (including exchange records) or academic literature (cultural communication research).

[0037] The consensus voting algorithm uses a dynamic threshold mechanism ( (Adaptively adjusted to data volatility) to avoid over-triggering; in this embodiment, the conflict resolution accuracy reaches 92.7%. Specifically, in this embodiment, when an output conflict occurs (social media popularity 95 points vs. policy influence 60 points, deviation 35 points > threshold 20 points), CVA triggers conflict resolution: 1. Extract the current event feature vector. 2. Search the historical case database and calculate the similarity with the historical case record of "Tokyo cultural controversy" as 0.88; 3. Apply the weight adjustment record of this historical case record (social media weight -0.3, policy data weight +0.5), update the weight and recalculate the score to 82.4 points.

[0038] In step S3.2, the analogical reasoning retrieves the Top-K most similar historical case records by calculating the weighted similarity between the current event feature vector and the historical event feature vector, and generates the weight adjustment factor based on the conflict resolution of these historical case records.

[0039] Step S4: Using the natural language generation engine and visualization module of the report generation layer, the comprehensive evaluation results are transformed into a structured report containing causal explanations and recommendations and then output.

[0040] In this embodiment, the natural language generation engine generates a causal chain explanation (such as "low acceptance in the Middle East is due to lack of adaptation to the collective culture") based on the comprehensive evaluation results and the communication rule base (containing 80+ templates). Addressing the issue of low cultural compatibility, it outputs: "It is recommended to add content adapted to Middle Eastern culture, referring to the Dubai case (historical similarity is 0.85)." The visualization module dynamically generates heatmaps and confidence interval annotations, outputting an interactive HTML or PDF structured report.

[0041] This invention integrates heterogeneous data sources such as social media, international news, diplomatic archives, and academic literature through three types of dedicated intelligent agents: social media crawlers, policy database integration, and news semantic parsing. It covers data types such as multilingual and cross-media texts and relies on a federated learning framework to complete data preprocessing. While ensuring data comprehensiveness, it protects sensitive information through a differential privacy mechanism to avoid single points of failure, thus solving the dual problems of "one-sided evaluation perspective" and "data security risks".

[0042] This invention achieves multi-dimensional parallel evaluation by analyzing three types of intelligent agents in the decision-making layer: influence calculation, cultural fit assessment, and public opinion sentiment analysis. Combining consensus voting algorithms with a historical case database, it can autonomously detect and evaluate conflicts. By generating weight adjustment factors through analogical reasoning combining weighted cosine similarity and Jaccard keyword similarity, conflicts can be resolved without human intervention, thus significantly reducing cross-cultural evaluation errors.

[0043] This invention uses a natural language generation engine combined with a communication studies rule base to generate reports containing causal chain explanations and strategy recommendations. It also features a visualization module to generate heatmaps and confidence interval annotations, and supports interactive drill-down of raw data. Compared to templated reports, this invention better meets the needs of refined strategy formulation in cross-cultural scenarios and helps users quickly understand evaluation conclusions.

[0044] In summary, the data perception layer of this invention integrates multi-source heterogeneous data through multiple dedicated data acquisition agents, overcoming the limitations of a single data source and ensuring a comprehensive evaluation perspective. Simultaneously, parallel data acquisition and preprocessing improve efficiency and adapt to dynamic evaluation needs. The analysis and decision-making layer relies on collaborative analysis by multi-domain analysis agents, combined with consensus voting from the central decision-making module and a historical case library. It can autonomously detect and resolve evaluation conflicts without human intervention, reducing subjective errors and improving evaluation accuracy. The report generation layer, utilizing a natural language generation engine and visualization module, automatically generates structured reports with causal explanations, moving away from template-based presentations, reducing comprehension costs, and providing practical decision support for cities to optimize their international communication strategies.

Claims

1. A multi-agent-based system for assessing the international communication influence of cities, characterized in that: include: The data perception layer includes multiple dedicated data acquisition agents, which are used to collect and preprocess data related to the city's international communication from multiple heterogeneous data sources in real time. The analysis and decision-making layer, connected to the data perception layer, is used for collaborative analysis and comprehensive evaluation of preprocessed urban international communication-related data. The analysis and decision-making layer includes: multiple domain analysis agents for parallel analysis and local evaluation of the data, outputting local evaluation values ​​and corresponding confidence levels; a central decision-making module, connected to the multiple domain analysis agents, for receiving each local evaluation value and confidence level; the central decision-making module includes: a consensus voting module for weighted fusion based on the local evaluation values ​​and confidence levels, and triggering a conflict resolution mechanism when conflicts are detected between local evaluation values; a historical case library storing case records of historical international communication events; when the conflict resolution mechanism is triggered, the central decision-making module calls the historical case library for analogical reasoning to resolve conflicts and generate a comprehensive evaluation result; and a report generation layer, connected to the analysis and decision-making layer, including a natural language generation engine and a visualization module, for automatically generating a structured evaluation report based on the comprehensive evaluation result.

2. The city international communication influence assessment system based on multi-agent technology according to claim 1, characterized in that: The dedicated data acquisition agent includes a social media crawler agent, a policy database docking agent, and a news semantic parsing agent; the heterogeneous data sources include social media data sources, international news data sources, diplomatic archive data sources, and academic literature data sources; the city's international communication-related data includes multilingual data, cross-media text data, diplomatic archive data, and treaty data.

3. The city international communication influence assessment system based on multi-agent technology as described in claim 2, characterized in that: The domain analysis agents include influence calculation agents, cultural fit assessment agents, and public opinion sentiment analysis agents.

4. The city international communication influence assessment system based on multi-agent technology according to claim 1, characterized in that: The historical case database stores historical case records of multiple international communication events. Each record contains at least a historical event feature vector and a corresponding conflict resolution.

5. The evaluation method of the multi-agent-based urban international communication influence assessment system according to any one of claims 1-4, characterized in that, Includes the following steps: Step S1: Collect multi-source heterogeneous data through multiple dedicated data acquisition agents in the data perception layer and perform preprocessing; S2: Perform parallel analysis and local evaluation on the preprocessed urban international communication-related data through multiple domain analysis agents in the analysis and decision-making layer to obtain local evaluation values ​​and corresponding confidence levels; S3: Perform weighted fusion of each local evaluation value and corresponding confidence level through the consensus voting algorithm in the consensus voting module. When the consensus voting algorithm detects a conflict between local evaluation values, it calls the historical case library for analogical reasoning to resolve the conflict and form a comprehensive evaluation result; S4: Transform the comprehensive evaluation result into a structured report containing causal explanations and suggestions through the natural language generation engine and visualization module of the report generation layer and output it.

6. The evaluation method according to claim 5, characterized in that: In step S2, the analysis and local evaluation process is as follows: the influence calculation agent focuses on the preprocessed urban international communication-related data, quantifies the node centrality in the urban international communication-related data, and outputs the local evaluation value and corresponding confidence level of the urban international communication influence dimension; the cultural adaptability evaluation agent calls the Hofstede model to analyze the cross-cultural acceptability of the urban international communication-related data content, and outputs the local evaluation value and corresponding confidence level of the cross-cultural acceptability of the content by regional cultural groups; the public opinion sentiment analysis agent uses the LSTM-Attention model to identify implicit biases in the preprocessed urban international communication-related data, and corrects the sentiment tendency judgment based on the bias, thereby outputting the local evaluation value and corresponding confidence level of the international public opinion sentiment tendency dimension.

7. The evaluation method according to claim 5, characterized in that: In step S3, the consensus voting algorithm includes the following steps: Step S3.1: Receive the local evaluation values ​​output by the domain analysis agents. and confidence level ； Evaluation values ​​for each local area Perform weighted fusion to obtain the initial weighted score. Step S3.2: Compare each local evaluation value with the initial weighted score. If the maximum deviation exceeds the dynamic confidence threshold, a conflict is identified and a resolution mechanism is triggered. The historical case library is called for analogical reasoning to generate a weight adjustment factor. Update weights Recalculate the final comprehensive evaluation results 。 8. The evaluation method according to claim 7, characterized in that: In step S3.2, the analogical reasoning retrieves the Top-K most similar historical case records by calculating the weighted similarity between the current event feature vector and the historical event feature vector, and generates the weight adjustment factor based on the conflict resolution of these historical case records.

9. The evaluation method according to claim 8, characterized in that: The current event feature vector and the historical event feature vector contain three core dimensions: event type, cultural background label, and data source distribution entropy value; the weighted similarity calculation adopts a linear combination of weighted cosine similarity and Jaccard keyword similarity.

10. The evaluation method according to claim 5, characterized in that: In step S4, the natural language generation engine generates a causal chain explanation based on the comprehensive evaluation results and the communication rule base, and the visualization module dynamically generates a heat map and confidence interval annotations, outputting an interactive structured report.

Images