An artistic entrepreneurship scheme anti-risk performance simulation evaluation method and system

Abstract

This application relates to a simulation evaluation method and system for the risk resistance performance of an art entrepreneurship project. The method includes: acquiring multi-source heterogeneous data of the target art entrepreneurship project and performing feature extraction and structuring processing to obtain audience profile feature data; clustering and role assignment of the audience profile feature data to generate virtual audience agents with differentiated feedback tendencies; adversarially interpreting and evaluating the descriptive text of the art entrepreneurship project to obtain initial negative feedback data; quantifying the emotional intensity and analyzing the dissemination intention of the initial negative feedback data to obtain weighted negative feedback data; and performing multiple rounds of iterative optimization and impact response on the descriptive text of the art entrepreneurship project to obtain simulation evaluation results characterizing the risk resistance performance of the project. This method addresses the problems of traditional evaluations, such as strong subjectivity, limited samples, inability to simulate the spread of negative evaluations, and difficulty in quantifying risk resistance performance, thereby improving the scientific rigor and reliability of the evaluation and providing a basis for decision-making in art entrepreneurship projects.

Description

Technical Field

[0001] This invention belongs to the field of data processing, and in particular relates to a method and system for simulating and evaluating the risk resistance performance of an art entrepreneurship program. Background Technology

[0002] With the development of the cultural and creative industries, the number of art-related entrepreneurial projects is increasing, making feasibility assessment of art-related entrepreneurial plans a crucial step in project decision-making. Currently, the evaluation of art-related entrepreneurial plans mainly relies on traditional methods such as industry expert review, market research questionnaires, and simple financial forecasting models.

[0003] In traditional methods, expert reviewers typically conduct subjective assessments of art-related entrepreneurial proposals based on their own experience, or they collect data on the target audience's preferences through questionnaires, and then combine this with financial indicators to qualitatively judge the feasibility of the proposal. These methods rely heavily on human experience and limited sample feedback, making the evaluation process highly subjective and random.

[0004] However, current traditional evaluation methods have the following problems: On the one hand, expert review is limited by personal cognitive boundaries and aesthetic preferences, making it difficult to fully cover the differentiated feedback tendencies of different audience groups, resulting in insufficient representativeness and objectivity of the evaluation results; on the other hand, traditional methods can only obtain limited static feedback and cannot simulate the spread and diffusion effect of negative evaluations among the audience and their dynamic impact on the overall risk resistance of the plan, thus making it difficult to conduct a systematic and quantitative simulation evaluation of the risk resistance of art entrepreneurship plans in the real market environment. Summary of the Invention

[0005] Therefore, it is necessary to provide a simulation evaluation method and system for assessing the risk resistance of art entrepreneurship programs in response to the above-mentioned technical problems.

[0006] Firstly, this application provides a simulation evaluation method for the risk resistance performance of an art-based entrepreneurship program, including:

[0007] S1. Obtain multi-source heterogeneous data of the target art entrepreneurship plan, and perform feature extraction and structuring on the multi-source heterogeneous data to obtain audience profile feature data; wherein, the multi-source heterogeneous data includes the description text of the art entrepreneurship plan;

[0008] S2. Cluster and assign roles to the audience profile feature data to generate multiple virtual audience agents with differentiated feedback tendencies;

[0009] S3. Conduct adversarial interpretation and evaluation of the description text of the art entrepreneurship plan to obtain initial negative feedback data;

[0010] S4. Quantify the emotional intensity and analyze the willingness to spread for each negative feedback in the initial negative feedback data to obtain weighted negative feedback data;

[0011] S5. The description text of the art entrepreneurship plan is iteratively optimized and subjected to impact response in multiple rounds to obtain simulation evaluation results; among them, the simulation evaluation results are used to characterize the risk resistance performance of the plan.

[0012] In one embodiment, S1 includes:

[0013] S11. Extract keywords and semantic role labels from the description text of the art entrepreneurship plan to obtain the core element vector of the plan.

[0014] S12. Extract and normalize the audience's historical behavior data from multi-source heterogeneous data to obtain the audience behavior feature matrix.

[0015] S13. Perform feature cross-interaction and embedding fusion on the core element vector of the scheme and the audience behavior feature matrix to obtain audience profile feature data.

[0016] In one embodiment, S2 includes:

[0017] S21. Cluster the audience profile feature data, determine the optimal number of clusters based on the silhouette coefficient, and obtain K audience cluster centers;

[0018] S22. Based on K audience cluster centers, perform adversarial boundary extension on each audience cluster to obtain the adversarial audience cluster boundary; where the expression for the adversarial audience cluster boundary is:

[0019]

[0020] In the formula, For the first A boundary vector of an adversarial audience cluster. Let k be the center vector of the audience cluster. For the adversarial offset coefficient, For the first The standard deviation vector of samples within each audience cluster;

[0021] S23. Based on the boundaries of the adversarial audience clusters and the preset role allocation strategy, perform role tag mapping and proxy instantiation for each adversarial audience cluster to generate virtual audience proxies.

[0022] In one embodiment, S3 includes:

[0023] S31. Deconstruct and identify the intent of the text describing the art entrepreneurship plan sentence by sentence to obtain the semantic unit sequence of the plan;

[0024] S32. Based on the semantic unit sequence of the scheme and the profile features of each virtual audience agent, calculate the attention weight of each virtual audience agent to each semantic unit to obtain the attention weight matrix;

[0025] S33. Based on the attention weight matrix, negative tendency scoring and feedback statement generation are performed on the semantic units of each scheme to obtain initial negative feedback data.

[0026] In one embodiment, S4 includes:

[0027] S41. Extract emotional features and assign intensity values ​​to each negative feedback in the initial negative feedback data to obtain the emotional intensity value of each negative feedback.

[0028] S42. Input the emotional intensity value into the propagation dynamics model, simulate the propagation probability and propagation level of each negative feedback, and obtain the propagation influence coefficient of each negative feedback; whereby the expression for the propagation probability is:

[0029]

[0030] In the formula, For the first The probability of a negative feedback loop propagating in a single propagation event. For the first The emotional intensity value of each negative feedback. For the propagation sensitivity coefficient, This is the propagation threshold parameter;

[0031] S43. Weight and fuse the emotional intensity value and the dissemination influence coefficient to calculate the comprehensive weight of each negative feedback, and obtain weighted negative feedback data.

[0032] In one embodiment, S5 includes:

[0033] S51. Modularly decompose and analyze the text describing the art entrepreneurship plan to obtain a module dependency graph.

[0034] S52. Based on weighted negative feedback data, perform negative impact transmission and accumulation calculations on the solution module dependency graph to obtain the cumulative impact value for each solution module; wherein, the expression for the cumulative impact value is:

[0035]

[0036] In the formula, For the first The cumulative impact value of each solution module, This is the set of feedback corresponding to the weighted negative feedback data. For the first The overall weight of each negative feedback item For the first Negative feedback item and number The associated indicator variables of each solution module, For the first A collection of predecessor modules for each solution module. The impact attenuation coefficient is... For the first Impact transmission rate of each front-drive module;

[0037] S53. Based on the cumulative impact value and the preset iterative optimization strategy, the scheme module is adjusted and the response is evaluated in multiple rounds. When the rate of change of the impact value between two consecutive rounds is less than the convergence threshold, the iteration is stopped and the simulation evaluation results are obtained.

[0038] In one embodiment, S33 includes:

[0039] S331. Match and calculate the sensitivity of the semantic units of the schemes to obtain the negative sensitivity score of each semantic unit of the schemes.

[0040] S332. Based on the negative sensitivity score and the attention weight matrix, calculate the comprehensive negative score for each virtual audience agent to obtain the negative score vector; where the expression for the comprehensive negative score is:

[0041]

[0042] In the formula, For the first The overall negative rating of a virtual audience agent. For the first The virtual audience agent for the first The attention weight of each semantic unit in the scheme For the first Negative sensitivity score of semantic units of each scheme. This represents the total number of semantic units in the scheme.

[0043] S333. When the overall negative score exceeds the preset negative threshold, generate the corresponding negative feedback statement through the preset template, and summarize all generated negative feedback statements into initial negative feedback data.

[0044] Secondly, this application also provides a simulation evaluation system for the risk resistance performance of art entrepreneurship programs, including:

[0045] The data acquisition and feature extraction module is used to acquire multi-source heterogeneous data of the target art entrepreneurship plan, and to extract and structure the features of the multi-source heterogeneous data to obtain audience profile feature data; among which, the multi-source heterogeneous data includes the descriptive text of the art entrepreneurship plan;

[0046] The adversarial audience generation module is used to cluster and assign roles to audience profile feature data, generating multiple virtual audience agents with differentiated feedback tendencies.

[0047] The multi-dimensional feedback simulation module is used to perform adversarial interpretation and evaluation of the description text of the art entrepreneurship plan, and obtain initial negative feedback data;

[0048] The feedback diffusion and intensity quantification module is used to quantify the emotional intensity and analyze the willingness to spread each negative feedback in the initial negative feedback data to obtain weighted negative feedback data.

[0049] The risk resistance simulation module is used to perform multiple rounds of iterative optimization and impact response on the description text of the art entrepreneurship plan to obtain simulation evaluation results; among which, the simulation evaluation results are used to characterize the risk resistance performance of the plan.

[0050] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the method described in the first aspect.

[0051] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described in the first aspect.

[0052] The aforementioned method and system for simulating and evaluating the risk resistance of art entrepreneurship schemes, through structured processing of multi-source heterogeneous data and precise construction of audience profiles, can avoid the cognitive limitations and aesthetic biases of expert reviewers, thereby improving the objectivity and comprehensiveness of the evaluation. By generating virtual audience agents and interpreting them adversarially, it can comprehensively uncover potential risk points, compensating for the shortcomings of traditional methods in overlooking risks. Through the quantification of negative feedback emotions, simulation of dissemination, and cumulative calculation of module impacts, it can achieve a dynamic and systematic quantitative evaluation of the scheme's risk resistance performance. Through multiple rounds of iterative optimization, it can provide technical support for risk response, improve the scientific rigor and reliability of art entrepreneurship scheme evaluations, and provide a basis for project decision-making. Attached Figure Description

[0053] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0054] Figure 1 This is a flowchart illustrating a simulation evaluation method for the risk resistance performance of an art entrepreneurship program in one embodiment.

[0055] Figure 2 This is a schematic diagram of the structure of a simulation evaluation system for the risk resistance performance of an art entrepreneurship program, as shown in one embodiment. Detailed Implementation

[0056] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0057] In one embodiment, reference Figure 1 The document presents a flowchart illustrating a simulation evaluation method for the risk resistance performance of an art entrepreneurship program provided in this application. This embodiment uses the application of this method to a simulation evaluation terminal (hereinafter referred to as the terminal) as an example. It is understood that this method can also be applied to a server, and further to a system including both a terminal and a server, and is implemented through interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0058] S1. Obtain multi-source heterogeneous data of the target art entrepreneurship plan, and extract and structure the features of the multi-source heterogeneous data to obtain audience profile feature data.

[0059] For example, the simulation evaluation terminal acquires multi-source heterogeneous data on the target art entrepreneurship plan, and performs feature extraction and structuring processing on the multi-source heterogeneous data to ultimately obtain audience profile feature data. This multi-source heterogeneous data may include the descriptive text of the art entrepreneurship plan, as well as publicly available audience preference data from the industry, audience feedback data from similar art entrepreneurship projects, and macro-trend data from the cultural consumption market. Data acquisition sources can be achieved through various methods, such as developing and acquiring publicly available network resources based on HTTP / HTTPS protocols, connecting to industry database API interfaces, and receiving compliant data provided by the plan submitter, ensuring the comprehensiveness and compliance of the data. The descriptive text of the art entrepreneurship plan may include project positioning, art product form, operating model, profit path, and definition of the target audience.

[0060] For example, when the simulation evaluation terminal extracts features from multi-source heterogeneous data, it can use Natural Language Processing (NLP) technology to process text data by word segmentation and stop word removal, and use feature engineering technology to preprocess non-text data by discretization and normalization. It can also use redundant information filtering algorithms, missing data completion algorithms, and data format unification algorithms supported by hardware acceleration units to complete structured processing, and finally extract core features related to the audience to generate standardized audience profile feature data.

[0061] S2. Cluster and assign roles to the audience profile feature data to generate multiple virtual audience agents with differentiated feedback tendencies.

[0062] For example, the simulation evaluation terminal, supported by hardware computing power, can perform clustering operations and role assignment on the acquired audience profile feature data, generating multiple virtual audience agents with differentiated feedback tendencies. During the clustering process, the simulation evaluation terminal can employ an unsupervised clustering algorithm, using a GPU-accelerated unit to calculate the Euclidean distance between different samples in the audience profile feature data, quantifying sample similarity, and grouping samples with similar features into one category, thus achieving precise stratification of the audience group. To ensure the rationality of the clustering results, the simulation evaluation terminal can introduce the silhouette coefficient as an evaluation criterion for the clustering effect. By traversing different numbers of clusters, the clustering effect evaluation algorithm is called to select the number of clusters corresponding to the optimal silhouette coefficient as the optimal number of clusters, thereby obtaining K audience cluster centers, each representing an audience group with similar characteristics. The silhouette coefficient can be used as an indicator to evaluate the clustering effect, with a value range of [-1, 1]. The closer the coefficient is to 1, the better the clustering effect; the closer it is to -1, the worse the clustering effect.

[0063] For example, based on K audience cluster centers, the simulation evaluation terminal can expand the feature boundary of each audience cluster by introducing an adversarial offset coefficient, thereby covering the extreme feedback tendencies that such audiences may have. The simulation evaluation terminal can assign corresponding role tags to each adversarial audience cluster according to the adversarial audience cluster boundary and the preset role allocation strategy, combined with the characteristics of various audience clusters, through a role tag mapping algorithm. Then, through proxy instantiation technology, the features of each cluster are bound to the role tags to generate virtual audience proxies with independent feedback logic and differentiated feedback tendencies, accurately simulating the feedback behavior of different audience groups in the real market and improving the realism of risk simulation.

[0064] S3. Conduct adversarial interpretation and evaluation of the description text of the art entrepreneurship plan to obtain initial negative feedback data.

[0065] For example, the simulation evaluation terminal can use semantic parsing algorithms to perform adversarial interpretation and evaluation of the description text of an art entrepreneurship plan, generating initial negative feedback data. Adversarial interpretation uses technical means to uncover potential information in the plan description text that may trigger negative evaluations from the audience, avoiding overlooking potential risk points.

[0066] Specifically, the simulation evaluation terminal first uses a text decomposition algorithm to decompose the description text of the art entrepreneurship plan sentence by sentence, breaking the text into multiple independent semantic units. Each semantic unit corresponds to a specific content module of the plan. Then, a natural language intent recognition algorithm is used to analyze the core intent expressed by each semantic unit, clarify the possible audience feedback direction that the unit may trigger, and finally obtain a standardized sequence of plan semantic units.

[0067] For example, the simulation evaluation terminal can calculate the attention weight of each virtual audience agent to each semantic unit based on the generated semantic unit sequence of the scheme and the profile features of each virtual audience agent, combined with the role tags and feedback tendencies of the virtual audience agents, through a weight calculation algorithm. The calculation of attention weight is based on the correlation between the audience profile features and the semantic unit, and the correlation coefficient is quantified through a correlation analysis algorithm. The higher the correlation coefficient, the greater the attention weight. A standardized attention weight matrix can be generated, which can clearly represent the degree of attention of different virtual audience agents to different content of the scheme.

[0068] For example, the simulation evaluation terminal can score the semantic units of each scheme based on the attention weight matrix and the negative tendency scoring algorithm. The scoring criteria include multiple dimensions such as the rationality, feasibility and market acceptance of the content involved in the semantic unit. Then, based on the negative tendency scoring results and combined with the attention weight of the virtual audience agent, a negative feedback statement that conforms to the feedback tendency of the agent is generated through the feedback statement generation algorithm. All negative feedback statements generated by the virtual audience agents are summarized and deduplicated to obtain the initial negative feedback data, ensuring the comprehensiveness and accuracy of the negative feedback data.

[0069] S4. Quantify the emotional intensity and analyze the willingness to spread for each negative feedback in the initial negative feedback data to obtain weighted negative feedback data.

[0070] For example, the simulation evaluation terminal can quantify the emotional intensity and analyze the willingness to spread each negative feedback in the initial negative feedback data to obtain weighted negative feedback data. During the emotional intensity quantification process, the simulation evaluation terminal can use emotion analysis technology to extract the emotional characteristics of each negative feedback, identify the types of negative emotions contained in the feedback, such as dissatisfaction, doubt, and denial, and then, through preset emotional intensity assignment rules, transform the abstract negative emotions into quantifiable emotional intensity values. These emotional intensity values ​​can intuitively reflect the intensity of each negative feedback.

[0071] For example, the simulation evaluation terminal can input the emotional intensity value of each negative feedback into the propagation dynamics model to simulate the propagation probability and propagation level of each negative feedback. The propagation dynamics model is a mathematical model that can be used to simulate the process of information propagation in a group. Based on the characteristics of the information itself and the attributes of the group, it can simulate the propagation probability, propagation range, and propagation level of the information. Through the simulation of the propagation dynamics model, the propagation influence coefficient of each negative feedback is obtained. This coefficient comprehensively considers the propagation probability and propagation level, and can characterize the propagation ability and influence range of the negative feedback in the audience group.

[0072] For example, the simulation evaluation terminal can weight and fuse the emotion intensity value and the dissemination influence coefficient, set reasonable weight allocation rules, calculate the comprehensive weight of each negative feedback, and the higher the comprehensive weight, the greater the impact of the negative feedback on the risk resistance performance of the art entrepreneurship plan. All negative feedback and their comprehensive weights are associated and stored to obtain weighted negative feedback data.

[0073] S5. The description text of the art entrepreneurship plan is iteratively optimized and subjected to impact response in multiple rounds to obtain simulation evaluation results.

[0074] For example, the simulation evaluation terminal can perform multiple rounds of iterative optimization and impact response simulation on the description text of an art entrepreneurship plan to obtain simulation evaluation results that characterize the plan's risk resistance performance. For example, the simulation evaluation terminal can use a modular decomposition algorithm to modularize the description text of the art entrepreneurship plan, breaking it down into multiple interconnected plan modules according to the plan's function and logic, such as art product modules, operation modules, profit modules, and marketing modules. Then, through association analysis technology, graph theory analysis algorithms are used to identify the dependencies between each plan module. For example, the profit module depends on the execution effect of the operation module, and the marketing module affects the audience acceptance of the product module, constructing a plan module dependency graph that clearly presents the connection paths and influence relationships between each module.

[0075] For example, the simulation evaluation terminal can use weighted negative feedback data to perform negative impact transmission and accumulation calculations on the module dependency graph using an impact transmission algorithm to determine the module corresponding to each negative feedback. By associating the indicator variable with the k-th module, and combining the impact transmission rate and impact attenuation coefficient of each module in the predecessor module set, the cumulative impact value of each module is calculated using a cumulative impact value calculation algorithm. The cumulative impact value can intuitively reflect the comprehensive impact degree of the module on the negative feedback. The impact of the predecessor module will be transmitted to the current module through the dependency relationship, and the impact intensity will attenuate during the transmission process.

[0076] For example, the simulation evaluation terminal can adjust and optimize the solution modules with high impact values ​​based on the cumulative impact value and the preset iterative optimization strategy through the module optimization algorithm. The optimization direction may include improving the module content, adjusting the implementation strategy, etc. Then, the impact response of the optimized solution modules is evaluated, the adjusted cumulative impact value is calculated, and the above optimization and evaluation process is repeated. When the rate of change of impact value between two consecutive rounds is less than the preset convergence threshold, the iteration stops. The data such as the cumulative impact value of each module and the overall impact level of the solution obtained at this time constitute the simulation evaluation result. The simulation evaluation result can be used to accurately characterize the risk resistance performance of art entrepreneurship solutions and provide technical support for art entrepreneurship solution decision-making.

[0077] The aforementioned simulation evaluation method for the risk resistance performance of art entrepreneurship projects accurately constructs audience profiles by extracting features and structuring multi-source heterogeneous data, thus reducing over-reliance on human experience and effectively improving the objectivity and comprehensiveness of the evaluation. By generating virtual audience agents with differentiated feedback tendencies and conducting adversarial interpretations, the method comprehensively uncovers potential risk points of the project, compensating for the shortcomings of traditional evaluation methods that easily overlook risks. By quantifying the emotional intensity of negative feedback, analyzing dissemination intentions, and calculating the cumulative impact of modules, combined with multiple rounds of iterative optimization, the method can achieve a dynamic and systematic evaluation of the project's risk resistance performance, providing reliable technical support for project risk response. Overall, it improves the scientificity and reliability of art entrepreneurship project evaluation and provides accurate and feasible basis for project decision-making.

[0078] In an optional embodiment, S1 includes:

[0079] S11. Extract keywords and semantic role labels from the description text of the art entrepreneurship plan to obtain the core element vector of the plan.

[0080] Optionally, the simulation evaluation terminal uses text processing algorithms to extract keywords and label semantic roles in the description text of the art entrepreneurship plan, obtaining a vector of core elements of the plan. Keyword extraction can employ the TF-IDF (Term Frequency-Inverse Document Frequency) algorithm, which extracts keywords that can represent the core content of the plan from the text and removes irrelevant and redundant words. Semantic role labeling can use natural language processing technology, which identifies the semantic role of each keyword in the text, such as subject, object, action, and attribute, through a semantic role labeling model, clarifying the semantic relationships between keywords. Furthermore, vector embedding technology can be used to transform the extracted keywords and semantic roles into a numerical vector of core elements of the plan, enabling the numerical representation of textual information.

[0081] S12. Extract and normalize the behavioral features of audience historical behavior data from multi-source heterogeneous data to obtain the audience behavior feature matrix.

[0082] Optionally, the simulation evaluation terminal can extract and normalize the audience's historical behavior data from multi-source heterogeneous data to obtain an audience behavior feature matrix. The audience's historical behavior data can include their cultural consumption records, art product browsing records, and feedback records for similar projects. During the behavior feature extraction process, the simulation evaluation terminal can use behavior feature extraction algorithms to extract core behavioral features such as the audience's consumption preferences, feedback tendencies, and focus areas. Then, using normalization techniques, such as the min-max normalization algorithm, it transforms behavioral features of different dimensions and magnitudes into feature data with a unified value range, eliminating the influence of dimensions and improving the consistency of data processing. Finally, it constructs an audience behavior feature matrix, where each row corresponds to an audience sample and each column corresponds to a behavioral feature.

[0083] S13. Perform feature cross-interaction and embedding fusion on the core element vector of the scheme and the audience behavior feature matrix to obtain audience profile feature data.

[0084] Optionally, the simulation evaluation terminal can use a feature fusion algorithm to perform feature cross-fertilization and embedding fusion of the core element vectors of the solution and the audience behavior feature matrix to obtain audience profile feature data. Feature cross-fertilization can be achieved by constructing feature interaction terms to explore the correlation between the core elements of the solution and the audience behavior features, such as the matching degree between the audience behavior features and the product form of the solution. Embedding fusion can use deep learning embedding technology to map the core element vectors of the solution and the audience behavior feature matrix to the same feature space through an embedding model. It can also achieve deep fusion of the two types of features through fusion algorithms, and finally obtain audience profile feature data that can comprehensively represent the characteristics of the relationship between the audience and the solution, thereby improving the accuracy of the audience profile.

[0085] In an optional embodiment, S2 includes:

[0086] S21. Cluster the audience profile feature data, determine the optimal number of clusters based on the silhouette coefficient, and obtain K audience cluster centers.

[0087] Optionally, the simulation evaluation terminal can cluster the audience profile feature data, determine the optimal number of clusters based on the silhouette coefficient, and obtain K audience cluster centers. The clustering process can use the K-means clustering algorithm. Initially, multiple different numbers of clusters K are set. Clustering operations are performed on the audience profile feature data, and the silhouette coefficient corresponding to each number of clusters is calculated. The K value with the largest silhouette coefficient is selected as the optimal number of clusters. At this point, the clustering result can most reasonably divide the audience group, and the audience in each cluster has highly similar profile features. Then, the mean of all sample features in each cluster is calculated to obtain K audience cluster centers, and each cluster center represents the core feature of this type of audience.

[0088] S22. Based on K audience cluster centers, perform adversarial boundary expansion on each audience cluster to obtain the adversarial audience cluster boundary.

[0089] Alternatively, the expression for the adversarial audience cluster boundary can be:

[0090]

[0091] In the above expression, For the first A boundary vector for an adversarial audience cluster is used to define the possible feature range of this type of audience; Let be the center vector of the k-th audience cluster, representing the core characteristics of this type of audience; This is the resistance offset coefficient, used to adjust the extent of boundary expansion. A reasonable value range can be preset based on industry experience and evaluation needs. For the first The standard deviation vector of samples within an audience cluster is used to characterize the dispersion of audience characteristics within that cluster. The larger the standard deviation, the greater the difference in characteristics of that type of audience, and the boundary expansion needs to be adjusted accordingly.

[0092] For example, the simulation evaluation terminal can calculate the adversarial audience cluster boundary vector based on K audience cluster centers using the above expression, and perform adversarial boundary extension on each audience cluster to obtain the adversarial audience cluster boundary.

[0093] S23. Based on the boundaries of the adversarial audience clusters and the preset role allocation strategy, perform role tag mapping and proxy instantiation for each adversarial audience cluster to generate virtual audience proxies.

[0094] Optionally, the simulation evaluation terminal, based on the boundaries of adversarial audience clusters and a preset role allocation strategy, can generate virtual audience agents by mapping role tags and instantiating agents for each adversarial audience cluster using role mapping algorithms and agent instantiation technology. The preset role allocation strategy can assign corresponding role tags to each cluster based on the characteristics of the audience clusters and the audience type of the art entrepreneurship project, such as core audience, potential audience, and opposing audience. These role tags determine the feedback tendency of the virtual audience agents. During agent instantiation, the boundary characteristics and role tags of each adversarial audience cluster are bound to preset feedback logic. An agent generation algorithm generates virtual audience agents with independent decision-making capabilities that can simulate real audience feedback behavior. Each virtual audience agent possesses differentiated feedback tendencies, ensuring the comprehensiveness of the simulation evaluation.

[0095] In an optional embodiment, S3 includes:

[0096] S31. The text describing the art entrepreneurship plan is broken down sentence by sentence and the intent is identified to obtain the semantic unit sequence of the plan.

[0097] Optionally, the simulation evaluation terminal can use text decomposition algorithms and intent recognition algorithms to decompose the description text of the art entrepreneurship plan sentence by sentence and identify intent, obtaining a sequence of semantic units for the plan. During sentence-by-sentence decomposition, the simulation evaluation terminal can break down the plan description text into multiple independent sentences according to punctuation and semantic logic using text segmentation algorithms. Each sentence is then semantically decomposed to extract core semantic information, forming independent semantic units. Each semantic unit corresponds to a specific content point of the plan, such as product pricing, operational cycle, and artistic style. Intent recognition can employ intent recognition algorithms from natural language processing. Based on a pre-set intent category library, the intent recognition model analyzes the core intent expressed by each semantic unit, clarifying the possible audience feedback direction that the unit might elicit. Audience feedback directions can include positive, negative, and neutral. Finally, all semantic units are arranged in text order to obtain a sequence of semantic units for the plan.

[0098] S32. Based on the semantic unit sequence of the scheme and the profile features of each virtual audience agent, calculate the attention weight of each virtual audience agent for each semantic unit to obtain the attention weight matrix.

[0099] Optionally, the simulation evaluation terminal can calculate the attention weight matrix by using a weight calculation algorithm to determine the attention weight of each virtual audience agent to each semantic unit based on the sequence of semantic units in the scheme and the profile features of each virtual audience agent. The calculation of attention weights can be based on the correlation between audience profile features and semantic units, using a correlation analysis algorithm, such as the Pearson correlation coefficient algorithm, to calculate the correlation coefficient between the profile features of each virtual audience agent and each semantic unit. A higher correlation coefficient indicates a higher degree of attention from the virtual audience agent to the semantic unit. The calculated attention weights are then organized according to the correspondence between virtual audience agents and semantic units to generate an attention weight matrix. Each row of the matrix can correspond to a virtual audience agent, each column can correspond to a semantic unit in the scheme, and the matrix elements can represent the attention weight of the corresponding virtual audience agent for that semantic unit.

[0100] S33. Based on the attention weight matrix, negative tendency scoring and feedback statement generation are performed on the semantic units of each scheme to obtain initial negative feedback data.

[0101] Optionally, the simulation evaluation terminal can use a negative scoring algorithm and a feedback statement generation algorithm to score the semantic units of each scheme with negative tendencies and generate feedback statements based on the attention weight matrix, thereby obtaining initial negative feedback data.

[0102] Optionally, the negative scoring algorithm can adopt a fine-grained negative scoring model based on deep learning, combined with a preset negative risk assessment index system, to quantitatively score each semantic unit of the solution from multiple core dimensions such as feasibility, market acceptance, artistic innovation and operational sustainability. During the scoring process, the corresponding weight values ​​in the attention weight matrix can be introduced to adjust the weights of semantic units with different attention levels of different virtual audience agents, so as to ensure that the scoring results can match the feedback tendencies of different audience groups and avoid the bias caused by single-dimensional scoring.

[0103] Optionally, the feedback statement generation algorithm can be based on a pre-trained natural language generation model, combined with the role tags, feedback tendencies, and attention weights of each virtual audience agent, and call a preset negative feedback statement template library. The template library covers different types of negative evaluations, such as questioning feasibility, dissatisfaction with pricing strategies, disapproval of artistic expression, and concerns about operational risks. It can also adaptively adjust the tone and expression details of the statements based on the negative rating results of semantic units to ensure that the generated negative feedback statements conform to the behavioral logic and language habits of the corresponding virtual audience agent. After generating all negative feedback statements, the simulation evaluation terminal can remove duplicate or highly similar feedback content through a deduplication algorithm, and then use a data verification algorithm to filter out valid feedback that conforms to the preset format and evaluation logic. Finally, the initial negative feedback data can be obtained by summarizing the data.

[0104] In an optional embodiment, S4 includes:

[0105] S41. Extract emotional features and assign intensity values ​​to each negative feedback in the initial negative feedback data to obtain the emotional intensity value of each negative feedback.

[0106] Optionally, the simulation evaluation terminal can use sentiment analysis algorithms to extract sentiment features and assign intensity values ​​to each negative feedback in the initial negative feedback data, obtaining a sentiment intensity value for each negative feedback. Sentiment feature extraction can employ sentiment analysis technology, using a negative sentiment dictionary matching algorithm and feature extraction model to identify the negative sentiment words, interjections, and other sentiment features contained in each negative feedback, clarifying the type and intensity of the negative emotion. During intensity assignment, based on a preset sentiment intensity level standard, a sentiment intensity assignment algorithm can assign a corresponding sentiment intensity value to each negative feedback based on the intensity of the negative emotion. A higher sentiment intensity value indicates a stronger degree of negativity in the feedback, directly reflecting the adverse impact of the feedback on the solution.

[0107] S42. Input the emotional intensity value into the propagation dynamics model, simulate the propagation probability and propagation level of each negative feedback, and obtain the propagation influence coefficient of each negative feedback.

[0108] Alternatively, the expression for the propagation probability can be:

[0109]

[0110] In the above expression, For the first The propagation probability of a negative feedback in a single propagation event, with a value range of [0,1]; For the first The emotional intensity value of each negative feedback. This is the propagation sensitivity coefficient, used to moderate the effect of emotion intensity on the propagation probability. The larger the value, the more significant the impact of emotional intensity on the probability of transmission. For the propagation threshold parameter, when hour, A value close to 0 indicates that the negative feedback is difficult to spread. hour, Follow The increase in the value indicates that the stronger the negative emotion, the higher the probability of its spread.

[0111] Optionally, the simulation evaluation terminal can input the emotion intensity value into the propagation dynamics model to simulate the propagation probability and propagation level of each negative feedback, and obtain the propagation influence coefficient of each negative feedback. The propagation dynamics model simulates the propagation process of negative feedback in a virtual audience agent group based on the audience network structure and information propagation law. Through model simulation, the propagation level of each negative feedback is obtained, that is, the depth and scope of propagation. Combined with the propagation probability, the propagation influence coefficient is calculated. The higher the coefficient, the stronger the propagation ability and the wider the scope of influence of the negative feedback.

[0112] S43. Weight and fuse the emotional intensity value and the dissemination influence coefficient to calculate the comprehensive weight of each negative feedback, and obtain weighted negative feedback data.

[0113] Optionally, the simulation evaluation terminal can use a weighted fusion algorithm to weight and fuse the emotion intensity value and the propagation influence coefficient, calculating the comprehensive weight of each negative feedback to obtain weighted negative feedback data. During the weighted fusion process, the weight percentages of the emotion intensity value and the propagation influence coefficient are preset. A weighted summation algorithm can be used to fuse the two to obtain the comprehensive weight of each negative feedback. The comprehensive weight takes into account both the intensity and propagation ability of the negative feedback, characterizing the impact of each negative feedback on the risk resistance of the art entrepreneurship plan. All negative feedback and their corresponding comprehensive weights are associated and stored to generate weighted negative feedback data.

[0114] In an optional embodiment, S5 includes:

[0115] S51. Modularly decompose and analyze the text describing the art entrepreneurship plan to obtain the module dependency graph.

[0116] Optionally, the simulation evaluation terminal can perform modular decomposition and correlation analysis on the description text of the art entrepreneurship plan to obtain a module dependency graph. Modular decomposition breaks down the plan description text into multiple independent but related modules according to the plan's functional logic and implementation process. Each module corresponds to a core aspect of the plan, such as the art creation module, product development module, marketing module, profit module, and operation management module. Correlation analysis can employ graph theory to identify the dependencies between the modules, including direct and indirect dependencies. For example, the profit module directly depends on the execution effect of the operation management module and indirectly depends on the promotion effect of the marketing module. By treating each module as a node and the dependencies between modules as edges, a module dependency graph can be constructed, clearly showing the mutual influence relationships between the modules.

[0117] S52. Based on the weighted negative feedback data, perform negative impact transmission and accumulation calculations on the solution module dependency graph to obtain the cumulative impact value of each solution module.

[0118] Alternatively, the expression for the cumulative impact value can be:

[0119]

[0120] In the above expression, For the first The cumulative impact value of each module can characterize the overall impact of negative feedback on that module; This is the set of feedback corresponding to the weighted negative feedback data. For the first The overall weight of each negative feedback item For the first Negative feedback item and number The associated indicator variable of the first scheme module, if the first... Negative feedback item and number Each solution module is related, then ,otherwise ; For the first The set of predecessor modules for the first scheme module, that is, for the first scheme module Modules that have direct dependencies on each other; This is the impact attenuation coefficient, which can be used to characterize the degree of attenuation of impacts during transmission between modules. The larger the value, the slower the impact decay. For the first The impact transmission rate of a preceding module can be used to characterize the proportion of impact transmitted from the preceding module to the current module.

[0121] For example, the simulation evaluation terminal can perform negative impact transmission and accumulation calculation on the module dependency graph of the scheme based on weighted negative feedback data. The simulation evaluation terminal can calculate the cumulative impact of each module through the above expression to obtain the cumulative impact value of each scheme module. The calculation of cumulative impact fully considers the direct impact and indirect transmission impact of negative feedback.

[0122] S53. Based on the cumulative impact value and the preset iterative optimization strategy, the scheme module is adjusted and the response is evaluated in multiple rounds. When the rate of change of the impact value between two consecutive rounds is less than the convergence threshold, the iteration is stopped and the simulation evaluation results are obtained.

[0123] Optionally, the simulation evaluation terminal can perform multiple rounds of adjustment and response evaluation on the scheme modules based on the cumulative impact value and a preset iterative optimization strategy. Iteration stops when the rate of change of the impact value between two consecutive rounds is less than a convergence threshold, yielding the simulation evaluation result. The iterative optimization strategy can prioritize adjustments to modules with higher cumulative impact values ​​based on the magnitude of the cumulative impact value. Adjustments can include improving module content, optimizing implementation processes, and supplementing risk response measures. After each adjustment, the cumulative impact value of each module is recalculated, and the rate of change of the impact value between the current and previous rounds is compared. When the rate of change is less than a preset convergence threshold, it indicates that the scheme's risk resistance performance has stabilized, and iteration stops. The data obtained at this point, including the cumulative impact value of each module, the overall impact level of the scheme, and the number of iterations, collectively constitute the simulation evaluation result, which can be used to comprehensively and quantitatively characterize the risk resistance performance of the art entrepreneurship scheme.

[0124] In an optional embodiment, S33 includes:

[0125] S331. Match and calculate the sensitivity of the semantic units of the schemes to obtain the negative sensitivity score of each scheme semantic unit.

[0126] Optionally, the simulation evaluation terminal can match and calculate the sensitivity of semantic units of the solutions to obtain a negative sensitivity score for each semantic unit. During the matching process, the simulation evaluation terminal can match each semantic unit of the solution with a preset negative risk keyword library to identify risk points in the semantic unit that may trigger negative feedback. The sensitivity calculation can be based on factors such as the severity of the risk points and the audience's attention. The simulation evaluation terminal can use a quantitative scoring algorithm to assign a corresponding negative sensitivity score to each semantic unit of the solution. The higher the negative sensitivity score, the more likely the semantic unit is to trigger negative feedback from the audience.

[0127] S332. Based on the negative sensitivity score and attention weight matrix, calculate the comprehensive negative score for each virtual audience agent to obtain the negative score vector.

[0128] Alternatively, the expression for the overall negative score can be:

[0129]

[0130] In the above expression, For the first The overall negative rating of a virtual audience agent can be used to characterize the degree of negative evaluation of the art entrepreneurship program by that agent. For the first The virtual audience agent for the first The attention weights of the semantic units of each scheme are derived from the attention weight matrix; For the first Negative sensitivity score of semantic units of each scheme. This represents the total number of semantic units in the scheme.

[0131] For example, the simulation evaluation terminal can calculate the comprehensive negative score for each virtual audience agent based on the negative sensitivity score and the attention weight matrix, using the above expression, combined with the focus of the virtual audience agent and the negative risk of each semantic unit, thus obtaining a negative score vector.

[0132] S333. When the overall negative score exceeds the preset negative threshold, generate the corresponding negative feedback statement through the preset template, and summarize all generated negative feedback statements into initial negative feedback data.

[0133] Optionally, when the overall negative score exceeds a preset negative threshold, the simulation evaluation terminal can generate corresponding negative feedback statements using a preset template, and aggregate all generated negative feedback statements into initial negative feedback data. The preset negative threshold can be set based on industry experience and evaluation needs; it can be used to determine whether the virtual audience agent will generate negative feedback. The negative feedback statement template can be preset based on different types of negative evaluations, such as questioning feasibility, dissatisfaction with pricing, and disagreement with artistic style. When the negative threshold is exceeded, the corresponding template is invoked based on the attention weight of the agent and the negative sensitivity of each semantic unit to generate negative feedback statements that conform to the agent's feedback tendency. The negative feedback statements generated by all virtual audience agents are summarized and deduplicated to finally obtain the initial negative feedback data.

[0134] The aforementioned simulation evaluation method for the risk resistance performance of art entrepreneurship programs constructs virtual audience agents with differentiated feedback tendencies and introduces adversarial boundary expansion, which can comprehensively cover the potential negative evaluation perspectives of diverse audiences and effectively overcome the subjectivity and sample bias of traditional expert reviews. Quantifying the emotional intensity and simulating the propagation dynamics of generated negative feedback can simulate the amplification and diffusion effects of negative evaluations in the real market, thus compensating for the shortcomings of traditional static evaluations in capturing propagation risks. Combining the program module dependency graph with impact transmission calculations and multi-round iterative optimization enables a systematic and quantitative simulation evaluation of the risk resistance performance of art entrepreneurship programs under dynamic negative impacts, providing entrepreneurs with a more objective and reliable basis for decision-making.

[0135] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0136] Based on the same inventive concept, this application also provides an art entrepreneurship scheme risk resistance performance simulation evaluation system for implementing the above-mentioned method for simulating and evaluating the risk resistance performance of an art entrepreneurship scheme. The solution provided by this system is similar to the solution described in the above method. Therefore, the specific limitations of one or more embodiments of the art entrepreneurship scheme risk resistance performance simulation evaluation system provided below can be found in the limitations of the art entrepreneurship scheme risk resistance performance simulation evaluation method described above, and will not be repeated here.

[0137] In one exemplary embodiment, such as Figure 2 As shown, a structural diagram of a simulation evaluation system 10 for assessing the risk resistance of an art entrepreneurship program is provided, comprising:

[0138] The data acquisition and feature extraction module 11 is used to acquire multi-source heterogeneous data of the target art entrepreneurship plan, and to extract and structure the features of the multi-source heterogeneous data to obtain audience profile feature data; wherein, the multi-source heterogeneous data includes the descriptive text of the art entrepreneurship plan;

[0139] The adversarial audience generation module 12 is used to cluster and assign roles to audience profile feature data, and generate multiple virtual audience agents with differentiated feedback tendencies.

[0140] The multi-dimensional feedback simulation module 13 is used to perform adversarial interpretation and evaluation of the description text of the art entrepreneurship plan to obtain initial negative feedback data;

[0141] The feedback diffusion and intensity quantification module 14 is used to quantify the emotional intensity and analyze the willingness to spread each negative feedback in the initial negative feedback data to obtain weighted negative feedback data.

[0142] The risk resistance simulation module 15 is used to perform multiple rounds of iterative optimization and impact response on the description text of the art entrepreneurship plan to obtain simulation evaluation results; among which, the simulation evaluation results are used to characterize the risk resistance performance of the plan.

[0143] Furthermore, the data acquisition and feature extraction module 11 can also be used for:

[0144] S11. Extract keywords and semantic role labels from the description text of the art entrepreneurship plan to obtain the core element vector of the plan.

[0145] S12. Extract and normalize the audience's historical behavior data from multi-source heterogeneous data to obtain the audience behavior feature matrix.

[0146] S13. Perform feature cross-interaction and embedding fusion on the core element vector of the scheme and the audience behavior feature matrix to obtain audience profile feature data.

[0147] Furthermore, the adversarial audience generation module 12 can also be used for:

[0148] S21. Cluster the audience profile feature data, determine the optimal number of clusters based on the silhouette coefficient, and obtain K audience cluster centers;

[0149] S22. Based on K audience cluster centers, perform adversarial boundary expansion on each audience cluster to obtain the adversarial audience cluster boundary;

[0150] S23. Based on the boundaries of the adversarial audience clusters and the preset role allocation strategy, perform role tag mapping and proxy instantiation for each adversarial audience cluster to generate virtual audience proxies.

[0151] Furthermore, the multi-dimensional feedback simulation module 13 can also be used for:

[0152] S31. Deconstruct and identify the intent of the text describing the art entrepreneurship plan sentence by sentence to obtain the semantic unit sequence of the plan;

[0153] S32. Based on the semantic unit sequence of the scheme and the profile features of each virtual audience agent, calculate the attention weight of each virtual audience agent to each semantic unit to obtain the attention weight matrix;

[0154] S33. Based on the attention weight matrix, negative tendency scoring and feedback statement generation are performed on the semantic units of each scheme to obtain initial negative feedback data.

[0155] Furthermore, the feedback diffusion and intensity quantization module 14 can also be used for:

[0156] S41. Extract emotional features and assign intensity values ​​to each negative feedback in the initial negative feedback data to obtain the emotional intensity value of each negative feedback.

[0157] S42. Input the emotional intensity value into the propagation dynamics model, simulate the propagation probability and propagation level of each negative feedback, and obtain the propagation influence coefficient of each negative feedback.

[0158] S43. Weight and fuse the emotional intensity value and the dissemination influence coefficient to calculate the comprehensive weight of each negative feedback, and obtain weighted negative feedback data.

[0159] Furthermore, the risk mitigation simulation module 15 can also be used for:

[0160] S51. Modularly decompose and analyze the text describing the art entrepreneurship plan to obtain a module dependency graph.

[0161] S52. Based on the weighted negative feedback data, perform negative impact transmission and accumulation calculations on the solution module dependency graph to obtain the cumulative impact value of each solution module.

[0162] S53. Based on the cumulative impact value and the preset iterative optimization strategy, the scheme module is adjusted and the response is evaluated in multiple rounds. When the rate of change of the impact value between two consecutive rounds is less than the convergence threshold, the iteration is stopped and the simulation evaluation results are obtained.

[0163] Furthermore, the multi-dimensional feedback simulation module 13 can also be used for:

[0164] S331. Match and calculate the sensitivity of the semantic units of the schemes to obtain the negative sensitivity score of each semantic unit of the schemes.

[0165] S332. Based on the negative sensitivity score and attention weight matrix, calculate the comprehensive negative score for each virtual audience agent to obtain the negative score vector;

[0166] S333. When the overall negative score exceeds the preset negative threshold, generate the corresponding negative feedback statement through the preset template, and summarize all generated negative feedback statements into initial negative feedback data.

[0167] In one embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps of the above-described method for simulating and evaluating the risk resistance performance of an art entrepreneurship program.

[0168] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0169] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0170] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. A simulation evaluation method for the risk resistance performance of an art entrepreneurship program, characterized in that, The method includes: S1. Obtain multi-source heterogeneous data of the target art entrepreneurship plan, and perform feature extraction and structuring on the multi-source heterogeneous data to obtain audience profile feature data; wherein, the multi-source heterogeneous data includes the descriptive text of the art entrepreneurship plan; S2. Cluster and assign roles to the audience profile feature data to generate multiple virtual audience agents with differentiated feedback tendencies; S3. The text describing the art entrepreneurship plan is subjected to adversarial interpretation and evaluation to obtain initial negative feedback data; S4. Perform emotional intensity quantification and dissemination intention analysis on each negative feedback in the initial negative feedback data to obtain weighted negative feedback data; S5. Perform multiple rounds of iterative optimization and impact response on the description text of the art entrepreneurship plan to obtain simulation evaluation results; wherein, the simulation evaluation results are used to characterize the risk resistance performance of the plan.

2. The method according to claim 1, characterized in that, S1 includes: S11. Extract keywords and semantic role labels from the description text of the art entrepreneurship plan to obtain the core element vector of the plan; S12. Extract and normalize the audience historical behavior data in the multi-source heterogeneous data to obtain the audience behavior feature matrix. S13. Perform feature cross-interaction and embedding fusion on the core element vector of the scheme and the audience behavior feature matrix to obtain audience profile feature data.

3. The method according to claim 2, characterized in that, S2 includes: S21. Cluster the audience profile feature data, determine the optimal number of clusters based on the silhouette coefficient, and obtain K audience cluster centers; S22. Based on the K audience cluster centers, perform adversarial boundary expansion on each audience cluster to obtain the adversarial audience cluster boundary; wherein, the expression of the adversarial audience cluster boundary is: In the formula, For the first A boundary vector of an adversarial audience cluster. Let k be the center vector of the audience cluster. For the adversarial offset coefficient, For the first The standard deviation vector of samples within each audience cluster; S23. Based on the boundaries of the adversarial audience clusters and the preset role allocation strategy, perform role tag mapping and proxy instantiation for each adversarial audience cluster to generate the virtual audience proxy.

4. The method according to claim 1, characterized in that, S3 includes: S31. The text describing the art entrepreneurship plan is broken down sentence by sentence and the intent is identified to obtain a sequence of semantic units of the plan. S32. Based on the semantic unit sequence of the scheme and the profile features of each virtual audience agent, calculate the attention weight of each virtual audience agent for each semantic unit to obtain the attention weight matrix; S33. Based on the attention weight matrix, perform negative tendency scoring and feedback statement generation on each of the scheme semantic units to obtain the initial negative feedback data.

5. The method according to claim 1, characterized in that, S4 includes: S41. Extract emotional features and assign intensity values ​​to each negative feedback in the initial negative feedback data to obtain the emotional intensity value of each negative feedback. S42. Input the emotional intensity value into the propagation dynamics model to simulate the propagation probability and propagation level of each negative feedback, and obtain the propagation influence coefficient of each negative feedback; wherein, the expression for the propagation probability is: In the formula, For the first The probability of a negative feedback loop propagating in a single propagation event. For the first The emotional intensity value of each negative feedback. For the propagation sensitivity coefficient, This is the propagation threshold parameter; S43. The emotional intensity value and the propagation influence coefficient are weighted and fused to calculate the comprehensive weight of each negative feedback, thereby obtaining the weighted negative feedback data.

6. The method according to claim 1, characterized in that, S5 includes: S51. Modularly decompose and analyze the text describing the art entrepreneurship plan to obtain a module dependency graph. S52. Based on the weighted negative feedback data, perform negative impact transmission and accumulation calculations on the solution module dependency graph to obtain the cumulative impact value of each solution module; wherein, the expression for the cumulative impact value is: In the formula, For the first The cumulative impact value of each solution module, This is the set of feedback corresponding to the weighted negative feedback data. For the first The overall weight of each negative feedback item For the first negative feedback item and the first The associated indicator variables of each solution module, For the first A collection of predecessor modules for each solution module. The impact attenuation coefficient is... For the first Impact transmission rate of each front-drive module; S53. Based on the cumulative impact value and the preset iterative optimization strategy, the scheme module is adjusted and evaluated in multiple rounds. When the rate of change of the impact value between two consecutive rounds is less than the convergence threshold, the iteration is stopped, and the simulation evaluation result is obtained.

7. The method according to claim 4, characterized in that, S33 includes: S331. Match and calculate the sensitivity of the semantic units of the scheme to obtain the negative sensitivity score of each semantic unit of the scheme; S332. Based on the negative sensitivity score and the attention weight matrix, calculate the comprehensive negative score for each virtual audience agent to obtain a negative score vector; wherein, the expression for the comprehensive negative score is: In the formula, For the first The overall negative rating of a virtual audience agent. For the first The virtual audience agent for the first The attention weight of each semantic unit in the scheme For the first Negative sensitivity score of semantic units of each scheme. This represents the total number of semantic units in the scheme. S333. When the overall negative score exceeds the preset negative threshold, a corresponding negative feedback statement is generated through a preset template, and all the generated negative feedback statements are summarized into the initial negative feedback data.

8. A simulation evaluation system for the risk resistance performance of an art entrepreneurship program, characterized in that, The system includes: The data acquisition and feature extraction module is used to acquire multi-source heterogeneous data of the target art entrepreneurship plan, and to extract and structure the features of the multi-source heterogeneous data to obtain audience profile feature data; wherein, the multi-source heterogeneous data includes the descriptive text of the art entrepreneurship plan; An adversarial audience generation module is used to cluster and assign roles to the audience profile feature data to generate multiple virtual audience agents with differentiated feedback tendencies. A multi-dimensional feedback simulation module is used to perform adversarial interpretation and evaluation of the description text of the art entrepreneurship plan to obtain initial negative feedback data; The feedback diffusion and intensity quantification module is used to quantify the emotional intensity and analyze the willingness to spread each negative feedback in the initial negative feedback data to obtain weighted negative feedback data. The scheme risk resistance simulation module is used to perform multiple rounds of iterative optimization and impact response on the description text of the art entrepreneurship scheme to obtain simulation evaluation results; wherein, the simulation evaluation results are used to characterize the scheme's risk resistance performance.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the method of any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 7.

Images