A method and system for accurately evaluating international communication effects based on a large language model

Abstract

The application discloses a kind of international spread effect accurate evaluation method, system based on large language model, it is related to data processing technical field, the method includes: S1, read media data, and the sensitive word identification and data cleaning of the media data are carried out;S2, construct spread diffusion model, obtain the media data after cleaning in step S1, obtain spread path analysis value information by graph neural network and time series analysis, identify abnormal node and feedback to the spread diffusion model;S3, obtain the spread behavior of sensitive word in step S1 and classify, calculate the influence degree of each kind of spread behavior;S4, the data obtained from step S2 and step S3 are analyzed side by side, obtain each kind of influence factor to realize data fitting, calculate spread effect value;S5, the result obtained from step S4 is input into Qwen big model, generates evaluation report, and carries out visual output.The application realizes multi-dimensional spread effect evaluation by the evaluation index of multiple modules combination, and then gives the suggestion of optimizing spread effect and prediction by the accurate analysis capability of integrated big model.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, and particularly relates to a method and system for accurately evaluating international communication effects based on a large language model. BACKGROUND

[0002] With the rapid development of the Internet and social media platforms, the speed and scope of information dissemination have greatly increased. Especially in the context of globalization, cross-border communication and the formation of international public opinion have become more complex. The evaluation of international communication effects not only needs to consider traditional communication paths and audience characteristics, but also needs to integrate more extensive data analysis and social network analysis methods. The application of big data technology and artificial intelligence (AI) algorithms provides unprecedented opportunities for this field, which can deeply mine the internal laws of information dissemination and help predict communication effects and optimize communication strategies. In this context, a large model-based communication effect accurate evaluation system has emerged, providing a more refined and intelligent evaluation means for global communication.

[0003] Traditional methods of evaluating international communication effects mainly rely on statistical data and manual analysis, and the evaluation process often focuses on the quantitative results of a single communication path, such as analyzing media reprint volume, reading volume, and other indicators to evaluate the influence of communication content and audience coverage. However, these methods have some limitations. On the one hand, they often fail to accurately capture the entire process of information dissemination, lacking dynamic tracking of communication paths and communication nodes; on the other hand, communication behavior analysis relies on manual experience or simple classification models, ignoring the complex relationship between communication content and communication path, and failing to comprehensively evaluate multi-dimensional factors in the communication process. In addition, existing technologies also lack efficient data cleaning and processing methods, and the integration and processing of massive data are not fine enough. Due to the lack of support for deep learning and intelligent analysis, the accuracy and real-time performance of traditional methods are poor, and they cannot cope with the rapidly changing international communication environment.

[0004] Although existing technologies can evaluate communication effects to some extent, they have the disadvantage of being unable to comprehensively evaluate multi-dimensional information of communication paths, communication nodes, and communication behaviors, and lack effective mechanisms to analyze abnormal nodes and changes in communication intensity during the communication process. Existing methods often simplify the communication process into static statistical data, and cannot dynamically track communication paths and key behavior nodes in the communication process, resulting in poor timeliness and accuracy of communication strategy adjustment. In addition, traditional technologies also have great limitations in processing large-scale data, lacking deep semantic understanding and behavior recognition. SUMMARY

[0005] In view of the problems of the prior art, the purpose of the present application is to provide a large language model-based international communication effect accurate evaluation method and system, which aims to enhance the prediction ability of complex communication phenomena through data cleaning, multi-dimensional communication path analysis, accurate communication behavior classification and quantitative analysis, and real-time evaluation of communication effect, so as to more comprehensively and accurately evaluate the communication effect and provide data support for the optimization of communication strategy, and has significant technical innovation and practical value.

[0006] In order to achieve the above-mentioned goal, the present application provides the following technical solutions:

[0007] The large language model-based international communication effect accurate evaluation method of the present application comprises:

[0008] S1, reading media data and performing sensitive word identification and data cleaning on the media data;

[0009] S2, constructing a communication diffusion model, obtaining the cleaned media data in step S1, obtaining communication path values and other information through a graph neural network and time series analysis, identifying abnormal nodes and feeding back to the communication diffusion model;

[0010] S3, obtaining the communication behavior of the sensitive words in step S1 and classifying them, and calculating the influence degree of each type of communication behavior;

[0011] S4, analyzing the data obtained in steps S2 and S3, obtaining various influence factors to realize data fitting, and determining the communication effect value;

[0012] S5, inputting the result obtained in step S4 into the Qwen large model, generating an automatic report, and performing visual output, and the system gives a prediction and suggestion.

[0013] In a preferred scheme, step S1 has the following specific process:

[0014] S11, constructing a data acquisition module to obtain text data such as posts, news and comments of foreign mainstream social media, news websites and video websites, and generating an original data set;

[0015] S12, using the Qwen large model to identify sensitive words in the original data set obtained in step S11, and filtering out words and sentences irrelevant to the analysis target;

[0016] S13, constructing a semantic model based on the vector relationship between sensitive words and sensitive words, and using vector decomposition technology to analyze the association relationship between sensitive words in step S12;

[0017] S14, using the Qwen large model to perform semantic analysis in the data cleaning process, removing noise information, merging similar sentences, and ensuring the accuracy and quality of the data.

[0018] In step S12, compared with the traditional rule matching method, the potential sensitive content can be more accurately identified according to the context, and the words and sentences irrelevant to the analysis target are filtered out, so as to refine the data content.

[0019] In step S13, a semantic model is constructed based on the vector relationship between sensitive words, and the identified sensitive words are processed by keyword identification to determine a plurality of theme words, and the theme words are weighted according to the importance coefficients of the theme words. The weights of the theme words reflect their relative importance in media dissemination, which is comprehensively evaluated by multiple factors such as frequency of occurrence and user interaction.

[0020] Preferably, in step S2, the specific process is as follows:

[0021] S21: The weighted sensitive words and theme word feature elements in step S1 are regarded as dissemination nodes;

[0022] S22: The text content of the dissemination nodes is processed by the Qwen large model to convert these elements into time series data, and the propagation relationship and propagation strength between different nodes are captured through graph neural network (GNN) and time series analysis, the propagation path value is calculated, and the overall effect of the propagation activity is evaluated, including the propagation breadth, depth and interaction level of the content;

[0023] S23: The propagation path analysis is further improved by analyzing and finding abnormal nodes, and the state of the nodes is analyzed in detail to identify abnormal nodes and take corresponding corrective measures;

[0024] S24: Based on the propagation relationship, propagation strength and propagation path value obtained in steps S22 and S23, a network relationship directed graph of each sensitive word node and each information is drawn, and combined with the abnormal analysis function, a propagation diffusion model is constructed.

[0025] Preferably, in step S3, the specific process is as follows:

[0026] S31: According to the quantitative data in the propagation process, the weighted sum of the number of different types of reprints when each sensitive word appears is calculated to provide a quantitative index for the analysis of the propagation behavior and analyze the influence of the behavior characteristics on the propagation effect;

[0027] S32: Combined with the deep learning ability of the Qwen large model, the user nodes with high influence and their propagation preferences can be identified through sentiment analysis and user behavior prediction, so as to analyze the behavior patterns of different user groups to determine the propagation behavior of sensitive words and classify them;

[0028] S33: According to the importance of the propagation behavior in step S32, the behavior is divided into core behavior and sub-core behavior, considering the influence degree of different propagation behaviors on the propagation effect of sensitive words, and the influence degree of each type of propagation behavior is calculated.

[0029] In a preferred embodiment, step S4 has the following specific process:

[0030] S41: Determine the propagation effect of the sensitive word and correct it, the average growth of the reprint volume of high and low sensitivity is different, and the reprint volume is corrected according to the sensitivity, so that the propagation effects of different sensitivities have the same dimension;

[0031] S42: Fit the reprint volume of the sensitive word by a quadratic function, and use an exponential or logarithmic function for the fitting of the reading volume and the reprint volume;

[0032] S43: Further quantify the propagation effect, calculate the correlation between the reprint frequency and the reprint volume and the reading volume;

[0033] S44: The system combines all the corrected propagation behaviors, propagation paths, reprint frequencies and correlations, and obtains the propagation effect value through the fitting result;

[0034] When analyzing the propagation result, if the propagation result meets the expectation, the next output module is entered; if the propagation result does not meet the expectation, an evaluation function needs to be constructed to feedback and optimize the propagation effect; the evaluation function can be comprehensively evaluated by the propagation effect of the sensitive word, the influence factor and other factors, and adjusted according to the feedback result.

[0035] In a preferred embodiment, step S5 has the following specific process:

[0036] An automatic report is generated by using the Qwen large model to explain the key factors, influence nodes and propagation trends in the propagation process, interactive charts are generated by combining the Tableau data visualization tool, to help users understand the propagation effect, and support multi-dimensional analysis.

[0037] The application also provides an international propagation effect precise evaluation system based on a large language model, comprising a data processing module, a propagation path analysis module, a propagation behavior analysis module, a propagation effect evaluation module and an output module.

[0038] The data processing module is configured to read media data, and perform sensitive word identification and data cleaning on the media data, and determine the cleaned media data.

[0039] The propagation path analysis module is configured to build a propagation diffusion model, obtain the cleaned media data, obtain propagation path values and other information through time series analysis, and identify abnormal nodes and feed them back to the model.

[0040] The propagation behavior analysis module is used for determining the propagation behavior of sensitive words and classifying, and calculating the influence degree of each type of propagation behavior;

[0041] The propagation effect evaluation module is used for analyzing the obtained data and obtaining various influence factors to realize data fitting and determine the propagation effect value;

[0042] The output module is used for generating an automated report and visualizing the output, and the system gives a prediction and a suggestion.

[0043] The innovation of the present application lies in that the deep learning and natural language processing capabilities of the Qwen large model can improve the intelligent level of the system, and deeply mine the complex patterns in the propagation data, so as to provide more accurate and personalized propagation effect evaluation.

[0044] Based on historical propagation data, the Qwen large model is used for effect prediction, the system gives suggestions for optimizing the propagation effect, and the evaluation model can be dynamically adjusted to cope with real-time changes in the propagation process. Through multi-dimensional analysis and evaluation of the propagation process, the problems of inaccurate and incomplete analysis in the prior art are overcome, and the transparency and evaluation accuracy of the media propagation process can be effectively improved.

[0045] The beneficial technical effects of the present application are:

[0046] (1) By integrating the precise analysis capability of the large model, combining sensitive word recognition, propagation path analysis, propagation behavior analysis and propagation effect evaluation and other modules, various factors in the propagation process are comprehensively analyzed. By using time series analysis and data fitting technology, the key nodes in the propagation path can be more accurately identified, the propagation effect can be calculated, and more accurate propagation effect evaluation results can be provided. This precision improvement provides a scientific basis for the optimization of the propagation strategy and the adjustment of the media content.

[0047] (2) Through the propagation path analysis module, the system can finely track the propagation trajectory of sensitive words, and establish a propagation diffusion model based on time series to deeply mine the internal laws of propagation. The propagation behavior analysis module classifies and quantitatively analyzes the propagation behavior of each node in the propagation path, helps to accurately identify the key influence factors, propagation behavior patterns and propagation paths in the propagation, and provides guidance for further optimizing the propagation strategy. This process ensures a deep understanding of the propagation path and behavior, and avoids the one-sidedness of relying only on traditional statistical methods.

[0048] (3) The propagation effect evaluation module combines the fitting algorithm for the number of reposts and the number of readings, and realizes the dynamic correction of the propagation effect through the correlation analysis of the propagation frequency and the propagation volume. Through real-time analysis and feedback, the system can adjust the propagation strategy in a timely manner according to the evaluation results, making the propagation process more efficient and accurate. This has important significance for international communication, especially in the context of cross-cultural communication and cross-language communication, and can effectively improve the global influence of the communication content.

[0049] (4) The output module combines the propagation path and the evaluation results of the effect to provide intuitive and clear visual output. This visual result not only facilitates the user to understand the complex information in the propagation process, but also makes the evaluation of the propagation effect more transparent and easy to monitor. Through the clear propagation path diagram and effect score diagram, decision makers can quickly grasp the implementation of the propagation strategy, facilitate targeted adjustments, and thus improve the efficiency of the propagation.

[0050] (5) With the increasing complexity of the information propagation environment, especially the international communication across national boundaries and cultures, it faces the interference and challenges of language, culture, and region. This scheme helps users cope with the uncertainty and complexity in international communication by accurately identifying and analyzing sensitive words, propagation paths, and node characteristics. The system can flexibly adjust the evaluation method of the propagation behavior, adapt to the propagation characteristics of different countries and regions, improve the evaluation ability of cross-country communication, and enhance the response ability to global public opinion. BRIEF DESCRIPTION OF DRAWINGS

[0051] Figure 1 The flowchart of the international communication effect precise evaluation method based on a large language model;

[0052] Figure 2 The sensitive word identification and theme word weighting flowchart in the data processing module;

[0053] Figure 3 The propagation path analysis flowchart;

[0054] Figure 4 The propagation effect evaluation flowchart. DETAILED DESCRIPTION

[0055] The technical solutions of the present application will be described in further detail below in combination with the drawings and specific embodiments.

[0056] EMBODIMENT

[0057] As shown in Figure 1 , an international communication effect precise evaluation method based on a large language model includes:

[0058] Step 1, data processing module: read media data, and perform sensitive word identification and data cleaning on the media data;

[0059] S11: Collecting post, news and comments and other text data from foreign mainstream social media, news websites and video websites, constructing a data collection module to generate an original data set;

[0060] In this scheme, the main task of the data processing module is to extract key sensitive words and topic words from the media data, and through semantic analysis and cleaning operation, to provide reliable data support for subsequent propagation path analysis and effect evaluation; Figure 2 The sensitive word recognition and topic word weighting process in the data processing module;

[0061] S12: First, the Qwen large model in the text recognition unit identifies sensitive words from the data set obtained in S11, filters out words and sentences irrelevant to the analysis target, and refines the data content; Specifically, after the sensitive words are identified, a plurality of topic words are determined, and the topic words are weighted according to the importance coefficients of the topic words; The weight of these topic words reflects their relative importance in media propagation, which may be evaluated by multiple factors such as frequency of occurrence, user interaction, etc.

[0062] S13: In the process of identifying sensitive words and topic words, the semantic analysis unit is responsible for analyzing the association between the sensitive words; To achieve this goal, vector decomposition technology is used to construct a semantic model based on the vector relationship between sensitive words and sensitive words;

[0063] This process can be described by the following formula:

[0064] A=W*S

[0065] Where A is the set of sensitive word entities, W is the sensitive word vector conversion matrix, and S is the semantic representation matrix of the sensitive words; Through this formula, the semantic information of the sensitive words is converted into vector representation, and the W matrix describes the semantic relationship between different sensitive words; Specifically, each row of the W matrix represents the semantic features of a sensitive word, and the S matrix maps each sensitive word to a semantic space, so that the similarity between words and words can be calculated, and their association can be determined;

[0066] The cleaning unit uses the Qwen large model to further optimize data quality in the data cleaning process, especially when processing data, it can eliminate redundant information and noise, and ensure that the subsequent analysis is based on clean and high-quality data;

[0067] Step 2, propagation path analysis module: build a propagation diffusion model, obtain the cleaned media data in step 1, obtain the propagation path value and other information through graph neural network and time series analysis, and identify abnormal nodes and feed back to the model; Figure 3A flowchart for propagation path analysis;

[0068] S21: The weighted sensitive words and topic word features in step S1 are regarded as propagation nodes; by converting these propagation nodes into time series data, the timeliness and evolution trend in the propagation process can be reflected;

[0069] S22: The text content of the propagation nodes is processed by the Qwen large model, and these elements are converted into time series data; through graph neural network (GNN) and time series analysis, the propagation relationship and propagation strength of information between different nodes are captured, the propagation path effect value is calculated, and the overall effect of the propagation activity is evaluated, including the propagation breadth, depth and interaction level of the content;

[0070] The propagation path analysis unit tracks the propagation process of information according to these time series data and evaluates the diffusion of information at each node; the propagation path analysis unit will further evaluate the path effect of the entire propagation process, which can be quantified by the following formula:

[0071] O = Q·A·M·N·I

[0072] Where O represents the propagation path analysis value, Q is the number of publications, A is the number of propagations on different media platforms, M is the total number of reprints, N is the reading volume of the initial published content, and I is the propagation path value; through this formula, the overall effect of the propagation activity can be comprehensively evaluated, including the propagation breadth, depth and interaction level of the content; the propagation path value I is further determined by the propagation path analysis unit, and the specific calculation method is:

[0073] I = L·U

[0074] Where I represents the propagation path value, L is the number of propagation levels, and U is the propagation strength value; the number of propagation levels L reflects the hierarchical number of information diffusion, while the propagation strength U is closely related to factors such as the propagation rate and participation of information at each node; through the comprehensive evaluation of these two parameters, the potential and effectiveness of information propagation can be further revealed;

[0075] For example, assuming that the number of media content publications Q is 10, the number of propagations on multiple platforms A is 50, the total number of reprints M is 100, the initial reading volume N is 500, and the propagation path value I is 2, the propagation effect value P can be calculated by the above formula; if the number of propagation levels L is 3 and the propagation strength U is 0.8, then the propagation path value I is:

[0076] I = 3*0.8 = 2.4

[0077] Combining these information, we can get the propagation path effect value P1, thus providing a quantitative basis for subsequent propagation effect evaluation and optimization;

[0078] S23: Further perfecting the propagation path analysis by analyzing and finding abnormal nodes, conducting detailed analysis on the state of the nodes, identifying abnormal nodes and taking corresponding corrective measures; In the propagation path analysis unit, one of the core steps to determine the propagation path is to conduct detailed analysis on the state of the nodes in order to identify abnormal nodes and take corresponding corrective measures; The effectiveness of the propagation path and the accurate evaluation of the propagation path effect depend on the behavior and propagation of each node; Specifically, the state of each node reflects the performance of information in the propagation process, and the abnormality of these states may affect the entire propagation process; Therefore, the abnormality degree of the node needs to be calculated and analyzed in order to discover problems early and intervene;

[0079] Abnormality degree of node j It can be calculated by the following formula:

[0080] ρ j =k ij *(F j -R j )*S j

[0081] Wherein, ρ j represents the abnormality degree of the jth node, k ij is the number of propagations from the ith node to the jth node, F j is the appearance frequency value of the jth node in the propagation path, R j is the existence frequency value of the jth node in the propagation path, S j is the influence factor of the sensitive word type corresponding to the jth node; By calculating the abnormality degree of the node, we can judge whether the node is abnormal; The higher the abnormality degree, the more likely it is that the node has problems in the propagation process, such as lower than expected propagation, insufficient sensitive word propagation, or deviation of the propagation path from the expected direction;

[0082] For example, assume that in a certain propagation process, the number of propagations k ij of the jth node is 30, the appearance frequency F j of the node is 0.6, the existence frequency R j is 0.5, and the influence factor S j of the sensitive word type is 1.2, then the abnormality degree ρ j of the node will be:

[0083] ρ j =30*(0.6-0.5)*1.2=30*0.1*1.2=3.6

[0084] If the calculated abnormality degree value ρ jIf the threshold is exceeded, it indicates that the node may have a propagation problem and further analysis and correction of the abnormal cause are needed.

[0085] The analysis and search of abnormal nodes is an important step to further improve the propagation path analysis. First, the abnormal degree of each node needs to be calculated and compared with other nodes. If the abnormal degree of a node is significantly higher than that of other nodes, the node can be marked as an abnormal node. Then, the reasons for the abnormal node need to be further investigated, which may include but are not limited to insufficient propagation amount of sensitive words, poor information transmission path, unreasonable node attribute setting, etc. After finding the reasons for the abnormal nodes, the system needs to correct these abnormal nodes. For example, it may be necessary to increase the information propagation frequency of the node, adjust the propagation relationship between nodes, or optimize the propagation content and sensitive word setting involved in the node;

[0086] During the preprocessing unit process, the system will first determine the frequency value of the key word, the type of sensitive word, the propagation path, and the propagation path. These data are crucial for subsequent propagation path analysis. By collecting and organizing these information, time series data can be constructed to accurately show the propagation changes of sensitive words at different time nodes. For example, assuming that the frequency of a sensitive word changes over time within a day, the system can display it in time series to show the propagation situation and path of the sensitive word in different propagation paths. The display form of these data not only helps to visualize the propagation process, but also provides detailed historical data support for subsequent analysis;

[0087] For example, assuming that the propagation path of a sensitive word is as follows: in the first hour, the number of propagation paths of the sensitive word is 3 times; in the second hour, the number of propagation paths increases to 5 times; in the third hour, the number of propagation paths is 7 times. By arranging these data in chronological order, the system can accurately track the dynamic changes of the sensitive word propagation and help users analyze the propagation effect and path;

[0088] Through node state analysis, abnormal degree calculation, propagation path adjustment, and propagation data preprocessing, the system can effectively identify and optimize the problems in the propagation path. These analysis processes not only can find abnormal nodes, but also can provide data support for the optimization of propagation strategy. Through these accurate evaluation and adjustment, the propagation path can be more in line with the expectation, thereby improving the effect and accuracy of information propagation;

[0089] S24: Draw the network relationship directed graph of each sensitive word node and each information based on the propagation relationship, propagation strength, and propagation path value obtained in S22 and S23, and construct a propagation diffusion model in combination with the abnormal analysis function.

[0090] By drawing a network diagram, nodes (such as individuals, media, platforms, etc.) and their connections (edges) can be displayed, which is very suitable for representing complex communication networks; the size of the node represents the influence or propagation of the node, and the color is used to distinguish different types of nodes or propagation states; the thickness of the edge can represent the propagation strength or the frequency of information flow;

[0091] And the directed graph can clearly represent the direction of information flow, that is, from one node to another node, indicating that information propagates from the source node to the target node; such a graph can visually display the path and directionality of information propagation; combined with the time series data converted by the Qwen model, the propagation of information over time on different nodes can be displayed, and the dynamic changes and trends of information propagation can be analyzed by the large model; finally, combined with the abnormal analysis function, the abnormal node data is fed back to the large model in real time, and the propagation diffusion model is constructed;

[0092] Step 3, propagation behavior analysis module: obtain the propagation behavior of the sensitive word in step 1 and classify it, and calculate the influence degree of each type of propagation behavior;

[0093] S31: According to the quantitative data in the propagation process, the weighted sum of the number of different types of reprints at each occurrence of the sensitive word is calculated, which provides a quantitative index for the analysis of the propagation behavior and analyzes the influence of its behavior characteristics on the propagation effect; In the propagation behavior analysis module, the core task is to accurately analyze each type of behavior in the propagation path to determine the propagation behavior of the sensitive word and classify it; propagation behavior refers to how the sensitive word or information spreads and propagates in the network at a specific time point and node, and how users interact with it; Therefore, the analysis of the propagation behavior not only focuses on the quantitative data in the propagation process, but also focuses on the influence of its behavior characteristics on the propagation effect; The analysis of the propagation behavior first needs to determine the performance of each propagation behavior through the following formula:

[0094]

[0095] Among them, β i represents the reprint propagation effect of the sensitive word, T ik represents the number of the kth type in the number of reprints of the ith occurrence of the sensitive word, P ik represents the total number of reprints of the ith occurrence of the sensitive word, and w represents the number of types in the propagation behavior; Through this formula, the weighted sum of the number of different types of reprints at each occurrence of the sensitive word can be calculated, and a quantitative index for the analysis of the propagation behavior is provided;

[0096] For example, assume that the number of reprints of the sensitive word at the first occurrence is 1000, of which there are two types of reprints, the number of type A reprints is 600, and the number of type B reprints is 400; then for this propagation behavior, the formula can be calculated as follows:

[0097] β i = (600 * 1) + (400 * 1) = 1000

[0098] In this way, the propagation effect of sensitive words can be quantified, helping to analyze which types of propagation behavior have a greater impact on the propagation effect;

[0099] S32: Combined with the deep learning capabilities of the Qwen large model, it can identify high-influence user nodes and their propagation preferences through sentiment analysis and user behavior prediction, analyze the behavior patterns of different user groups, and determine the propagation behavior of sensitive words and classify them;

[0100] S33: Further, according to the importance of the propagation behavior in S32, the behavior is divided into core behavior and secondary core behavior; considering the influence degree of different propagation behaviors on the propagation effect of sensitive words, the influence degree of each type of propagation behavior is calculated; core behavior refers to behavior that has a significant impact on the propagation effect, while secondary core behavior has a relatively small impact on the propagation effect; the purpose of behavior classification is to help identify key driving factors in the propagation process and optimize the propagation strategy;

[0101] In the influence analysis unit, the analysis of propagation behavior is not limited to the number of occurrences and types, but further considers the influence degree of different propagation behaviors on the propagation effect of sensitive words; the influence degree of propagation behavior can be calculated by the following formula:

[0102]

[0103] Where U j represents the propagation behavior influence factor of the jth node, F j represents the propagation behavior characteristics of the jth node, R j represents the propagation behavior characteristics of the node, C j represents the number of occurrences of the sensitive word, and E j represents the average increase in the number of reposts for the first occurrence of the sensitive word; through this formula, the influence degree of propagation behavior can be quantified, and it can be analyzed which behaviors have a greater impact on the propagation effect;

[0104] For example, if the propagation behavior characteristics F j of the jth node is 50, the propagation behavior characteristics R j is 30, the number of occurrences of the sensitive word C j is 1000 times, and the average increase in the number of reposts for the first occurrence of the sensitive word E j is 200 times, then the propagation behavior influence factor U j can be calculated as:

[0105]

[0106] The value indicates that the propagation behavior of the jth node has a greater impact on the propagation effect of the sensitive word, specifically in the relationship between the change of the propagation behavior characteristics and the growth of the sensitive word;

[0107] Next, in the propagation behavior analysis calculation process, the system also needs to analyze the behavior characteristics of the sensitive word to further quantify the influence of the propagation behavior; the calculation formula is as follows:

[0108]

[0109] Wherein, a represents the quantitative influence factor of the propagation behavior effect of the sensitive word, F j represents the propagation behavior characteristics of the jth node, G j represents the average growth of the reprint volume brought by the high-sensitivity propagation behavior; through the formula, the system can comprehensively consider the propagation behavior of each node in the propagation process and its contribution to the propagation effect, and calculate the overall propagation effect of the sensitive word;

[0110] For example, assume that the sensitive word performs differently at multiple nodes during the propagation process, the propagation behavior characteristics of a node is 100, and the node has high sensitivity, and the growth of the reprint volume is 50; the propagation behavior characteristics of another node is 50, and the node has low sensitivity, and the growth of the reprint volume is 10; then, the propagation behavior effect of the sensitive word is calculated as follows:

[0111] a = (100 * 5) + (50 * 10) = 5000 + 500 = 5500

[0112] This indicates that the propagation effect of the sensitive word at the high-sensitivity node is much greater than that at the low-sensitivity node, thereby providing a quantitative basis for the propagation effect evaluation;

[0113] Through these steps of propagation behavior analysis, the system can comprehensively analyze the key behaviors in the propagation path, the influence relationship between nodes, the classification of the propagation behavior, and its influence on the propagation effect, thereby optimizing the propagation strategy and improving the precise propagation effect of information;

[0114] Step 4, propagation effect evaluation module: analyze the data obtained in steps 2 and 3, obtain various influence factors to realize data fitting, and determine the propagation effect value;

[0115] S41: determine the propagation effect of the sensitive word and correct it; the average growth of the reprint volume is different for high-sensitivity and low-sensitivity, and the reprint volume is corrected according to the sensitivity, so that the propagation effects of different sensitivities have the same dimension;

[0116] Correction of the sensitive word:

[0117] In the evaluation of the propagation effect of sensitive words, the average growth of the number of reposts of sensitive words with high sensitivity and low sensitivity is different; in the propagation process, sensitive words with high sensitivity are usually more likely to obtain greater growth in the number of reposts, while sensitive words with low sensitivity may face smaller propagation expansion; therefore, the number of reposts needs to be corrected according to the sensitivity, so that the propagation effects of different sensitivities have the same dimension;

[0118] The correction formula is:

[0119]

[0120] Among them, S j represents the behavior correction value of the jth node, R j represents the number of reposts of the sensitive word corresponding to the jth node, T j represents the number of each type in the number of reposts corresponding to the jth node, Q j represents the sensitivity coefficient of the node; through the formula, the system can adjust the propagation behavior correction value of different sensitivity levels, and then ensure that the propagation quantity growth of high and low sensitivity has consistent dimension;

[0121] For example, assuming that the number of sensitive word reposts of a node is 100 times, type A accounts for 70 times, and type B accounts for 30 times, and the sensitivity coefficient Q j of the node is 1.5; then, the corrected behavior value can be calculated as:

[0122]

[0123] In this way, through the correction method, the propagation growth of type A with high sensitivity and type B with low sensitivity is unified and adjusted;

[0124] S42: Fit the number of reposts of the sensitive word by a quadratic function, and use an exponential or logarithmic function for the fitting of the reading volume and the number of reposts;

[0125] The focus of the work in the propagation effect evaluation module is to calculate the propagation effect of the sensitive word by fitting and correcting; first, fit the number of reposts of the sensitive word and the behavior analysis result, and then correct and correlate it with the reading volume, the number of reposts, etc.; by calculating the correlation between the number of reposts and the reading volume, a precise propagation effect value can be obtained;

[0126] Fitting of the number of reposts and the reading volume:

[0127] In the evaluation of the propagation effect, the number of reposts of the sensitive word is fitted by a quadratic function to help identify the trend of its propagation; for the fitting of the reading volume and the number of reposts, an exponential or logarithmic function is used, and such fitting method can reflect the expansion speed and form of information in the propagation process; Figure 4The propagation effect evaluation flowchart;

[0128] Further quantify the propagation effect, calculate the correlation between the reprint frequency and the reprint volume and the reading volume; the calculation formula of the reprint frequency is:

[0129]

[0130] Wherein, F represents the reprint frequency of the sensitive word, A represents the reprint volume of the sensitive word, and B represents the reading volume of the sensitive word; through this formula, the reprint frequency of each sensitive word in the propagation process can be calculated, and the propagation effect can be further evaluated according to the frequency;

[0131] The correlation calculation of the propagation effect is:

[0132] In order to further quantify the propagation effect, we also need to calculate the correlation between the reprint frequency and the reprint volume and the reading volume, and the formula is as follows:

[0133]

[0134] Wherein, τ represents the correlation between the reprint volume and the reading volume, A represents the reprint volume, B represents the reading volume, and F represents the reprint frequency; the function of this formula is to calculate the correlation through the relationship between the reprint volume and the reading volume, so as to determine the effectiveness of the propagation effect;

[0135] For example, assuming that the reprint volume of the sensitive word is 5000 times, the reading volume is 10000 times, and the reprint frequency F is 0.5; according to the formula, the correlation is calculated as:

[0136]

[0137] S43: Finally, the system combines all the corrected propagation paths, propagation behaviors, reprint frequencies and correlations, and obtains the final propagation effect value through the fitting result; when analyzing the propagation result, if the propagation result meets the expectation, it enters the next output module; if the propagation result does not meet the expectation, it needs to build an evaluation function to feedback and optimize the propagation effect; the evaluation function can be evaluated comprehensively through the propagation effect of the sensitive word, the influence factor and other factors, and adjusted according to the feedback result;

[0138] In the propagation effect analysis unit, the analysis and evaluation of the propagation process are mainly realized by combining the analysis of the propagation path and the propagation behavior in S2 and S3, and then obtaining various influence factors; this process covers multiple steps, from determining the influence factor, the propagation distance to the calculation of the propagation intensity and the propagation order, which directly affects the final propagation effect evaluation; the influence factors include the importance coefficient of the sensitive word, the number of sensitive words cited, the appearance frequency of the sensitive word, etc., which reflect the various influences on the information in the propagation process;

[0139] Specifically, the impact factor can be determined using the following formula:

[0140]

[0141] Where σ represents the impact factor, D is the set of all impact factors, and w d It is the keyword importance coefficient of each sensitive word, c d U is the number of times the keyword is cited, K is the total number of times the sensitive word appears, F is the number of attribute types of the sensitive word, T is the number of times the sensitive word is published, and T is the number of channels in the dissemination process. This formula can comprehensively consider the influence of multiple factors and evaluate the strength of the role of each sensitive word in the dissemination process.

[0142] Next, determining the dissemination distance is crucial for assessing the scope and effectiveness of the dissemination; the dissemination distance reflects the coverage of information from its source to its final audience, and can be calculated using the following formula:

[0143] shi(μ=E·C

[0144] Where μ represents the propagation distance, E is the propagation effect value of different social media platforms, and C is the number of channels or pathways involved in the propagation process; this formula can quantify the depth and breadth of information propagation across different platforms.

[0145] Propagation intensity is another key evaluation indicator, determining the speed at which information spreads during the propagation process; the value of propagation intensity U can be determined by the following formula:

[0146]

[0147] in, It is the dissemination intensity value, where F is the number of times the sensitive words are published and T is the number of channels in the dissemination path; the calculation of dissemination intensity takes into account the frequency of information release and the diversity of dissemination channels, and can reflect the explosive power and breadth of information dissemination;

[0148] The propagation level is an important indicator for measuring the hierarchy of information dissemination, typically reflecting the complexity of the path information takes from its source to its target audience. The propagation level is calculated based on the number of times sensitive words are reposted and published at each propagation node. Based on the number of reposts and publications at each propagation level, the value L of the propagation level can be quantified using the following formula:

[0149]

[0150] Where L is the propagation order, R i It is the number of reposts at the i-th level of propagation, P iis the number of publications of the i-th level of propagation; through this formula, the diffusion effect of information in different levels of propagation can be further revealed;

[0151] Finally, according to the comprehensive evaluation of the propagation distance, the propagation intensity and the propagation level, the propagation effect value P can be calculated by the following formula:

[0152]

[0153] where P is the propagation effect value, μ is the propagation distance, is the propagation intensity value, and L is the propagation level; this formula comprehensively considers the coverage, propagation rate and level depth of information propagation, and provides a comprehensive quantitative index for the propagation effect;

[0154] In analyzing the propagation results, first, the values of the various indicators of the propagation are calculated by the above formula; if the propagation results meet the expectations, then the next output link is entered; if the propagation results do not meet the expectations, then an evaluation function needs to be constructed to provide feedback and optimization for the propagation effect; the evaluation function can be comprehensively evaluated by the propagation effect of sensitive words, influence factors and other factors, and adjusted according to the feedback results;

[0155] For example, assuming that a sensitive word in the propagation process has a number of publications F of 100, a number of propagation paths T of 5, and a propagation intensity value U calculated as:

[0156] μ = 100 * 5 = 500

[0157] The propagation effect value E on different social media platforms is 1.2, the number of propagation paths C is 5, and the propagation distance D is calculated as:

[0158]

[0159] According to the propagation reprint volume and the publication volume P, assuming that the reprint volume under the first propagation level is 50 and the publication volume is 80, then the propagation level L is:

[0160] L = 50 * 80 = 4000

[0161] Finally, the propagation effect value P is:

[0162] P = 6 * 500 * 4000 = 12,000,000

[0163] Through the above calculation, the effect of the sensitive word in the entire propagation process can be comprehensively evaluated; if the propagation results are lower than expected, then the propagation strategy can be adjusted according to the feedback mechanism, the propagation path and strategy can be optimized, and the propagation effect can be improved;

[0164] The propagation effect evaluation module can effectively quantify the propagation effect by correcting and fitting the propagation path, the number of reposts, the reading volume, and the behavior characteristics of sensitive words. Through the above formulas and algorithms, the system can accurately evaluate the influence of the propagation behavior, provide a feedback mechanism, and adjust the propagation strategy to obtain the optimal propagation effect in practical applications.

[0165] Step 5, output module: input the results obtained in step 4 into the Qwen large model to generate an automated report explaining the key factors, influential nodes, and propagation trends in the propagation process, and generate interactive charts using the Tableau data visualization tool to help users understand the propagation effect and support multi-dimensional analysis.

[0166] Through accurate propagation effect evaluation, media organizations, brands, and enterprises can comprehensively understand the effectiveness of the propagation activities and identify the strengths and weaknesses in the propagation. This evaluation system can provide real-time feedback on the propagation intensity, propagation effect, and content influence during the propagation process, helping brands adjust the content, optimize the propagation path, ensure that the brand image and information can be effectively transmitted to the target audience, and improve the efficiency of brand propagation.

[0167] The system provides data support for the optimization of propagation strategies through multi-level propagation behavior analysis, propagation path tracking, and propagation effect evaluation. Based on big data propagation effect analysis and model prediction suggestions, the system can provide accurate propagation optimization suggestions for users in real time, helping them achieve precise propagation in a changing propagation environment. By continuously adjusting and optimizing the propagation strategy, the success rate of information propagation is improved, and the overall propagation effect is enhanced.

[0168] Through the highly integrated propagation effect evaluation system, combining big model technology, data processing, behavior analysis, and propagation evaluation, the system can analyze media data from multiple perspectives, accurately identify key factors and propagation paths in the propagation, and improve the accuracy and operability of the propagation effect. This not only helps users effectively evaluate the propagation effect, but also provides real-time optimization strategies to improve the efficiency and influence of information propagation worldwide.

[0169] The above embodiments are the preferred embodiments of the present application, but the embodiments of the present application are not limited by the above embodiments, and any changes, modifications, substitutions, combinations, simplifications made without departing from the spirit and principles of the present application are equivalent replacement methods and are included in the protection scope of the present application.

Claims

1. A method for accurately evaluating the effectiveness of international communication based on a large language model, characterized in that, include: S1. Read media data and perform sensitive word identification and data cleaning on the media data; S2. Construct a propagation and diffusion model, obtain the cleaned media data in step S1, obtain propagation path value information through time series analysis, and identify abnormal nodes and feed them back to the model. S3. Obtain the propagation behavior of the sensitive words in step S1 and classify them, and calculate the influence of each type of propagation behavior; S4. Analyze the data obtained in step S2 and step S3 side by side to obtain various influencing factors to achieve data fitting and determine the dissemination effect value. S5. Input the results obtained in step S4 into the Qwen large model to generate an automated report and output it in a visual format. The system provides predictions and suggestions. Step S2, the specific process is as follows: S21: Extract key elements from the cleaned media data in step S1, including sensitive words and their related elements. Keyword characteristics; S22: Use the Qwen large model to process the text content of the propagation nodes, and transform these elements into time. Sequence data, through graph neural networks (GNN) and time series analysis, captures the propagation relationship and intensity of information between different nodes, calculates the propagation path value, and evaluates the overall effect of the propagation activity, including the breadth, depth and level of content dissemination. S23: Further improve the propagation path analysis by analyzing and finding abnormal nodes, conduct detailed analysis of the node status, identify abnormal nodes and take corresponding corrective measures; S24: Based on the propagation relationship, propagation intensity, and propagation path value obtained in steps S22 and S23, draw a directed graph of the network relationship between each sensitive word node and each piece of information, and then combine it with the anomaly analysis function to construct a propagation path diffusion model; Step S4, the specific process is as follows: S41: Determine the dissemination effect of sensitive words and adjust them accordingly. The average increase in repost volume differs between high and low sensitivity. Adjust the repost volume based on sensitivity so that the dissemination effect of different sensitivities has the same dimension. S42: Fit the number of reposts of sensitive words using a quadratic function. For fitting the number of reads and reposts, use an exponential or logarithmic function. S43: Further quantify the dissemination effect by calculating the correlation between reposting frequency and the number of reposts and readership; S44: The system combines all corrected propagation behaviors, propagation paths, reprint frequencies, and relevance to obtain the propagation effect value through fitting results; When analyzing the dissemination results, if the results meet expectations, proceed to the next output module; if the results do not meet expectations, an evaluation function needs to be constructed to provide feedback and optimize the dissemination effect. The evaluation function comprehensively assesses the dissemination effect of sensitive words and influencing factors, and is adjusted based on the feedback results.

2. The method for accurately evaluating the international communication effect based on a large language model according to claim 1, characterized in that, Step S1, the specific process is as follows: S11: Build a data acquisition module to obtain post, news, and comment text data from foreign social media, news websites, and video websites, and generate raw datasets; S12: Use the Qwen large model to identify sensitive words in the original dataset obtained in step S11 and filter out words and phrases that are irrelevant to the analysis target; S13: Construct a semantic model based on the vector relationship between sensitive words, and use vector decomposition technology to analyze the association relationship between sensitive words in step S12; S14: Utilize the Qwen large model to perform semantic analysis during data cleaning, remove noise information, merge similar statements, and ensure data accuracy and quality.

3. The method for accurately evaluating the international communication effect based on a large language model as described in claim 1, Its features are, Step S22, the specific process is as follows: The Qwen large model is used to process the text content of the propagation nodes, and these elements are transformed into time series data. Through graph neural network (GNN) and time series analysis, the propagation relationship and propagation intensity of information between different nodes are captured, the propagation path effect value is calculated, and the overall effect of the propagation activity is evaluated, including the breadth, depth and interaction level of the content. The propagation path analysis unit tracks the propagation process of information based on this time series data and evaluates the spread of information at each node. The propagation path analysis unit further evaluates the path effect of the entire propagation process, quantifying it using the following formula: Where p represents the dissemination path effect value, Q is the number of times it was published, A is the number of times it was disseminated across different media platforms, M is the total number of times it was reposted, N is the number of reads of the initial published content, and I is the dissemination path value. This formula comprehensively evaluates the overall effect of the dissemination campaign, including the breadth, depth, and level of interaction of the content. The dissemination path value I is further determined by the dissemination path analysis unit, and the specific calculation method is as follows: Wherein, I represents the propagation path value, L represents the propagation level, and U represents the propagation intensity value; the propagation level L reflects the number of levels of information diffusion, while the propagation intensity U is closely related to the propagation rate and participation factors of information at each node; through the comprehensive evaluation of these two parameters, the potential and effectiveness of information propagation are further revealed.

4. The method for accurately evaluating the international communication effect based on a large language model as described in claim 1, Its features are, Step S23, the specific process is as follows: The propagation path analysis is further improved by analyzing and identifying abnormal nodes. The state of nodes is analyzed in detail to identify abnormal nodes and take corresponding corrective measures. In the propagation path analysis unit, one of the core steps in determining the propagation path is to conduct a detailed analysis of the state of nodes in order to identify abnormal nodes and take corresponding corrective measures. The effectiveness of the propagation path and the accurate evaluation of the propagation path effect depend on the behavior and propagation situation of each node. The state of each node reflects the performance of information during the propagation process, and anomalies in these states may affect the entire propagation process; therefore, the degree of anomaly of the nodes needs to be calculated and analyzed in order to detect problems early and intervene. Anomaly level of nodes The calculation is performed using the following formula: in, This indicates the degree of abnormality of the j-th node. It represents the number of propagations from the i-th node to the j-th node. It is the frequency of the j-th node during propagation. It is the frequency value of the j-th node during propagation. It is the influence factor of the sensitive word type corresponding to the j-th node; by calculating the degree of abnormality of the node, it is determined whether the node is abnormal; the higher the degree of abnormality, the more likely there is a problem with the node in the propagation process.

5. The method for accurately evaluating the effectiveness of international communication based on a large language model as described in claim 1. Its features are, Step S3, the specific process is as follows: S31: Based on the quantitative data in the dissemination process, calculate the weighted sum of the number of reposts of different types each time a sensitive word appears, providing a quantitative indicator for the analysis of dissemination behavior and analyzing the impact of its behavioral characteristics on the dissemination effect; S32: Combining the deep learning capabilities of the Qwen large model, it can perform sentiment analysis and user behavior prediction. The test identifies high-influence user nodes and their propagation preferences, thereby analyzing the behavioral patterns of different user groups to determine and classify the propagation behavior of sensitive words. S33: Based on the importance of the propagation behavior in step S32, the behaviors are divided into core behaviors and secondary core behaviors. Considering the impact of different communication behaviors on the spread of sensitive words, calculate the impact of each type of communication behavior.

6. The method for accurately evaluating the effectiveness of international communication based on a large language model as described in claim 5. Its features are, Step S3 is as follows: S31: Based on the quantitative data during the dissemination process, calculate the weighted sum of the number of reposts of different types each time a sensitive word appears, providing a quantitative indicator for the analysis of dissemination behavior and analyzing the impact of its behavioral characteristics on the dissemination effect; in the dissemination behavior analysis module, the core task is to accurately analyze various behaviors in the dissemination path to determine and classify the dissemination behavior of sensitive words; dissemination behavior refers to how sensitive words or information spread and propagate in the network at specific points in time and nodes, and how users interact with them; therefore, the analysis of dissemination behavior not only focuses on the quantitative data during the dissemination process, but also on the impact of its behavioral characteristics on the dissemination effect; the analysis of dissemination behavior first needs to determine the performance of each dissemination behavior through the following formula: in, This indicates the spread of sensitive words. This represents the proportion of the k-th type among the number of reposts of the i-th occurrence of a sensitive word. Let w represent the total number of reposts for the i-th occurrence of the sensitive word, and w represent the number of types in the dissemination behavior. Using this formula, the weighted sum of the number of reposts for different types is calculated each time the sensitive word appears, thus providing a quantitative indicator for the analysis of dissemination behavior. S32: Combining the deep learning capabilities of the Qwen large model, it can identify highly influential user nodes and their propagation preferences through sentiment analysis and user behavior prediction, thereby analyzing the behavioral patterns of different user groups to determine and classify the propagation behavior of sensitive words. S33: Based on the importance of the communication behaviors in S32, the behaviors are divided into core behaviors and secondary core behaviors; the impact of different communication behaviors on the spread of sensitive words is considered, and the impact of each type of communication behavior is calculated; core behaviors refer to behaviors that have a significant impact on the spread effect, while secondary core behaviors have a relatively small impact on the spread effect; the purpose of behavior classification is to help identify key driving factors in the communication process and optimize communication strategies. In the impact analysis unit, the analysis of dissemination behaviors is not limited to their frequency and type, but also further considers the degree of influence of different dissemination behaviors on the spread of sensitive words; the degree of influence of dissemination behaviors is calculated using the following formula: in, This represents the propagation behavior influence factor of the j-th node. It is the frequency of the j-th node during propagation. It is the frequency value of the j-th node during propagation. Indicates the number of times sensitive words appear. This formula represents the average increase in the number of reposts when a sensitive word first appears. It quantifies the impact of dissemination behaviors and analyzes which behaviors have a relatively large influence on the dissemination effect. Next, in the process of calculating the effect of the dissemination behavior, the system also needs to analyze the behavioral characteristics of sensitive words to further quantify the influence of the dissemination behavior; the formula for calculating the effect of the dissemination behavior is as follows: in, The quantitative impact factor representing the spread of sensitive words. It is the frequency of the j-th node during propagation. This represents the average increase in reposts resulting from highly sensitive dissemination behaviors. Using this formula, the system comprehensively considers the dissemination behavior of each node in the dissemination process and its contribution to the dissemination effect, and calculates the overall dissemination effect of sensitive words. Through these steps of communication behavior analysis, the system can comprehensively analyze key behaviors in the communication path, the influence relationships between nodes, the classification of communication behaviors and their impact on communication effectiveness, thereby optimizing communication strategies and improving the accuracy of information dissemination.

7. The method for accurately evaluating the international communication effect based on a large language model as described in claim 1, Its features are, Step S4 is as follows: S41: Determine the dissemination effect of sensitive words and adjust them accordingly; the average increase in repost volume differs between high and low sensitivity, so adjust the repost volume based on sensitivity to ensure that the dissemination effect of different sensitivities has the same dimension. When evaluating the spread of sensitive words, the average increase in reposts differs between words with high and low sensitivity. During the spread process, sensitive words with high sensitivity usually achieve a larger increase in reposts, while those with low sensitivity may face a smaller spread. Therefore, it is necessary to adjust the repost volume based on sensitivity to ensure that the spread effects of different sensitivities have the same dimension. The corrected formula is: in, This represents the behavior correction value of the j-th node. This represents the number of times the sensitive word corresponding to the j-th node has been reposted. This represents the quantity of each type in the total number of reprints corresponding to the j-th node. This represents the sensitivity coefficient of the node; through this formula, the system adjusts the propagation behavior correction value for different sensitivity levels, thereby ensuring that the propagation volume growth of high sensitivity and low sensitivity has consistent dimensions. S42: Fit the number of reposts of sensitive words using a quadratic function. For fitting the number of reads and reposts, use an exponential or logarithmic function. The focus of the dissemination effect evaluation module is to calculate the dissemination effect of sensitive words through fitting and correction. First, the repost volume and behavioral analysis results of sensitive words are fitted, and then they are corrected and correlated with the reading volume and repost frequency. By calculating the correlation between the repost volume and the reading volume, an accurate dissemination effect value can be obtained. In evaluating the dissemination effect, a quadratic function is used to fit the number of reposts of sensitive words to help identify their dissemination trend; for fitting the number of reads and reposts, an exponential or logarithmic function is used. Such fitting methods can reflect the speed and form of information expansion during the dissemination process. To further quantify the dissemination effect, the correlation between reposting frequency and the number of reposts and views was calculated; the formula for calculating reposting frequency is: Where F represents the reposting frequency of the sensitive word, a represents the reposting volume of the sensitive word, and b represents the reading volume of the sensitive word; this formula is used to calculate the reposting frequency of each sensitive word during the dissemination process, and the dissemination effect is further evaluated based on this frequency; To further quantify the dissemination effect, it is also necessary to calculate the correlation between the frequency of reposting and the number of reposts and views, as shown in the following formula: Where P represents the correlation between the number of reposts and the number of reads, a represents the number of reposts of the sensitive words, b represents the number of reads of the sensitive words, and F represents the frequency of reposts of the sensitive words; the purpose of this formula is to calculate the correlation between the number of reposts and the number of reads in order to determine the effectiveness of the dissemination effect. S43: Finally, the system combines all the corrected propagation paths, propagation behaviors, reposting frequencies, and relevance to obtain the final propagation effect value through fitting results. When analyzing the propagation results, if the propagation results meet expectations, the system proceeds to the next output module. If the propagation results do not meet expectations, an evaluation function needs to be constructed to provide feedback and optimization on the propagation effect. The evaluation function comprehensively evaluates the propagation effect of sensitive words and influencing factors, and is adjusted based on the feedback results. In the dissemination effect analysis unit, the analysis and evaluation of the dissemination process is mainly achieved by combining the analysis of dissemination paths and behaviors in S2 and S3, and then obtaining various influencing factors. This process covers multiple steps, from determining influencing factors and dissemination distance to calculating dissemination intensity and dissemination level, all of which directly affect the final dissemination effect evaluation. Influencing factors include the importance coefficient of sensitive words, the number of times sensitive words are cited, and the frequency of occurrence of sensitive words. These factors reflect the various influences that information is subjected to during the dissemination process. The impact factor is determined using the following formula: in, This represents the value of the impact factor. It is the collection of all influencing factors. It is the keyword importance coefficient of each sensitive word. It is the number of times the keywords are cited. K is the total number of times the sensitive word appears, F is the number of attribute types of the sensitive word, T is the reposting frequency of the sensitive word, and T is the number of paths in the propagation process. This formula can comprehensively consider the influence of multiple factors and evaluate the strength of the role of each sensitive word in the propagation process. Next, determining the dissemination distance is crucial for assessing the scope and effectiveness of the dissemination; the dissemination distance reflects the coverage of information from its source to its final audience, and is calculated using the following formula: Where D' is the propagation distance, E is the propagation effect value of different social media platforms, and C is the number of channels or pathways involved in the propagation process; this formula quantifies the depth and breadth of information propagation across different platforms. Propagation intensity is another key evaluation indicator, determining the speed at which information spreads during the propagation process; the value of propagation intensity U is determined by the following formula: Wherein, U is the intensity value of dissemination, F is the reposting frequency of sensitive words, and T is the number of channels in the dissemination path; the calculation of dissemination intensity takes into account the frequency of information release and the diversity of dissemination channels, and can reflect the explosive power and breadth of information dissemination. The propagation level is an important indicator for measuring the hierarchy of information dissemination, typically reflecting the complexity of the path information takes from its source to its target audience. The propagation level is calculated based on the number of times sensitive words are reposted and published at each propagation node. Based on the number of reposts and publications at each propagation level, the value L of the propagation level is quantified using the following formula: Where L is the propagation level, This is the number of reposts at the i-th level of propagation. This represents the amount of information disseminated at the i-th level of propagation; this formula can further reveal the diffusion effect of information at different levels of propagation. Finally, based on a comprehensive evaluation of propagation distance, propagation intensity, and propagation level, the propagation path effect value p is calculated using the following formula: Where p is the propagation path effect value, D' is the propagation distance, U is the propagation intensity value, and L is the propagation level; this formula comprehensively considers the coverage, propagation rate, and level depth of information propagation, providing a comprehensive quantitative indicator for propagation effect; When analyzing the dissemination results, the values ​​of various dissemination indicators are first calculated using the formulas mentioned above. If the dissemination results meet expectations, the next output stage is initiated. If the dissemination results do not meet expectations, an evaluation function needs to be constructed to provide feedback and optimize the dissemination effect. The evaluation function comprehensively assesses the dissemination effect of sensitive words and influencing factors, and is adjusted based on the feedback results.

8. A precise evaluation system for international communication effects based on a large language model, characterized in that, The method for accurately evaluating the international communication effect based on a large language model, as described in any one of claims 1 to 7, is adopted.

Images