Public opinion risk assessment method and electronic device

Abstract

This application provides a method and electronic device for assessing public opinion risks. The method includes repeatedly executing multiple rounds of content distribution assessment operations based on a target event and a public opinion risk control objective. Upon reaching a preset threshold number of executions for the content distribution assessment operations, a public opinion risk assessment result is determined based on the assessment results of all rounds. The public opinion risk assessment method and electronic device of this application can effectively predict and assess public opinion risks.

Description

Technical Field

[0001] This application relates to the field of risk analysis technology, and in particular to a public opinion risk assessment method and electronic device. Background Technology

[0002] Public opinion risks can not only lead to reputational damage, decreased trust, and customer churn, but may also trigger a chain reaction of consequences such as compliance reviews, business disruptions, supply chain fluctuations, and increased collaboration costs. Current public opinion risk monitoring relies heavily on volume statistics, keyword alerts, or manual verification, resulting in post-event summaries with significant delays, making it difficult to identify early signs of escalating risks in a timely manner. Furthermore, the evaluation dimensions lack structured characterization and quantitative measurement of key features such as information evolution paths, changes in opinion divergences, and trends in emotional fluctuations. Therefore, it is difficult to generate interpretable and verifiable trend assessments and predictions for specific events, and thus cannot effectively support the need for early warning and dynamic management throughout the entire event lifecycle. Summary of the Invention

[0003] In view of this, the purpose of this application is to propose a public opinion risk assessment method and electronic device.

[0004] To achieve the above objectives, this application provides a method for assessing public opinion risk, including:

[0005] Based on the public opinion risk management objective of the target event, multiple rounds of content distribution evaluation are performed cyclically. Each round of content distribution evaluation includes: Based on the target distribution parameters of this round, the public opinion risk management target, and the sentiment data of all target users towards the target event, multiple pieces of content to be distributed in this round are determined from the content pool pre-built based on the target event; From all the target users, determine the distribution objects corresponding to each round of distribution content, and distribute each round of distribution content to the corresponding distribution objects to obtain the user interaction data for this round. Based on all the content distributed in this round received by each distribution recipient, update the sentiment data of the corresponding distribution recipient; Based on the user interaction data of this round, the evaluation results of this round of content distribution evaluation operation are calculated; Based on the evaluation results of this round of content distribution evaluation, the target distribution parameters for the next round of content distribution evaluation will be determined. In response to the content distribution assessment operation reaching a preset threshold number of executions, the public opinion risk assessment result is determined based on the assessment results of all rounds.

[0006] Optionally, based on the target distribution parameters of this round, the public opinion risk management target, and the sentiment data of all target users regarding the target event, multiple pieces of content for this round of distribution are determined from a content pool pre-built based on the target event, including: Obtain the preset sentiment data corresponding to each distributed content in the content pool; Based on the sentiment tendency data of all target users regarding the target event, the sentiment tendency baseline data is calculated; For each piece of content distributed in the content pool, the suitability data corresponding to each piece of content is calculated based on the corresponding preset sentiment tendency data, the sentiment tendency benchmark data, and the public opinion risk control target. Based on the target distribution parameters for this round and the adaptation data corresponding to each distribution content in the content pool, multiple distribution contents for this round are determined from the content pool.

[0007] Optionally, determining multiple distribution content for this round from the content pool based on the target distribution parameters for this round and the adaptation data corresponding to each distribution content in the content pool includes: Based on the historical distribution records of each content in the content pool, the freshness data of each content is determined. Based on the target distribution parameters for this round, the adaptation data and freshness data corresponding to each distribution content in the content pool, multiple distribution contents for this round are determined from the content pool.

[0008] Optionally, determining the distribution target corresponding to each round of distribution content from all target users includes: Obtain the preset sentiment data corresponding to each piece of content distributed in this round; Candidate users are determined from all the target users; For each piece of content distributed in this round, the sentiment matching data between it and each candidate user is calculated based on the corresponding suitability data. Based on the sentiment matching data between each piece of content distributed in this round and each candidate user, the distribution targets corresponding to each piece of content distributed in this round are determined.

[0009] Optionally, determining the distribution target corresponding to each round of distribution content based on the sentiment matching data between each round of distribution content and each candidate user includes: For each piece of content distributed in this round, based on its sentiment matching data with each candidate user, all candidate users are divided into a first user set and a second user set. The sentiment matching data of candidate users in the first user set is higher than that of candidate users in the second user set. Based on the target proportion for this round, the distribution targets are determined from the first user set and the second user set.

[0010] Optionally, after calculating the evaluation result of the content distribution evaluation operation based on the user interaction data of this round, the method further includes: Based on the evaluation results of this round of content distribution evaluation, the target proportion for the next round of content distribution evaluation will be determined.

[0011] Optionally, for each piece of content distributed in this round, based on the corresponding suitability data, the emotional matching data between it and each candidate user is calculated, including: Based on the historical distribution records of each piece of content in this round, determine the freshness data of each piece of content in this round. For each piece of content distributed in this round, matching data between it and each candidate user is calculated based on the corresponding freshness and suitability data.

[0012] Optionally, the calculation of the evaluation result of the content distribution evaluation operation based on the user interaction data of this round includes: Based on the user interaction data of this round, the distribution effect evaluation data is calculated using a pre-built large language model; Based on the user interaction data of this round, the correlation data between the distributed content and the user interaction data, the content interaction rate data, and the user coverage data of this round are calculated. Calculate the deviation between the current sentiment data of all target users and the aforementioned public opinion risk management target; Based on the distribution effect evaluation data, the correlation data, the content interaction rate data, the user coverage data, and the deviation data, the evaluation results of this round of content distribution evaluation operation are determined.

[0013] Optionally, the correlation data is calculated using the following formula, including: ; ; ; in, For related data, The normalized entropy is N, where N is the total number of contents distributed in this round. This is the Shannon entropy corresponding to this round. For the first This round of distribution content, For the first This round of distribution content The percentage of interactions, =No. This round of distribution content The corresponding number of interactions / the total number of interactions for all content distributed in this round.

[0014] Based on the same inventive concept, this disclosure also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable by the processor, wherein the processor implements the method described above when executing the computer program.

[0015] As can be seen from the above, the public opinion risk assessment method and electronic device provided in this application achieve quantitative assessment, trend judgment and risk warning of public opinion risks related to the event by constructing a closed-loop assessment mechanism of "content pool - target user (simulated user) - multi-round assessment - feedback iteration" for the target event, without relying on or with weak reliance on real public opinion data. By selecting content from the content pool based on target distribution parameters, determining distribution targets, and acquiring user interaction data, this approach continuously updates user sentiment data through multiple iterations. It calculates the evaluation results for each round and adaptively adjusts the target distribution parameters for the next round, thereby improving evaluation coverage and stability while reducing the randomness and lag of single-round evaluations. By introducing and tracking user sentiment data (such as sentiment orientation and attitude orientation), it allows for a more granular characterization and quantitative comparison of changes in users' psychological state before and after exposure to information related to the target event. This identifies early risk signals such as emotional fluctuations, attitude shifts, and escalating disagreements, providing more interpretive evaluation evidence beyond relying solely on interaction behavior statistics. This effectively improves the accuracy and sensitivity of judging the evolution trend and direction of public opinion risks. After the evaluation rounds reach a preset threshold, the evaluation results of each round are summarized to output a public opinion risk assessment conclusion, providing a basis for risk monitoring, early warning, and subsequent strategy optimization. This application achieves effective prediction and assessment of public opinion risks for events and effectively improves the response efficiency to public opinion risks, better supporting the needs for risk early warning and dynamic management throughout the entire lifecycle of target events. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in this application or related technologies, the drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the drawings described below are only embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a schematic diagram of a public opinion risk assessment method according to an embodiment of this application; Figure 2 This is a schematic diagram of a public opinion risk assessment device according to an embodiment of this application; Figure 3 This is a schematic diagram of an electronic device according to an embodiment of this application. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with specific embodiments and the accompanying drawings.

[0019] It should be noted that, unless otherwise defined, the technical or scientific terms used in the embodiments of this application should have the ordinary meaning understood by one of ordinary skill in the art to which this application pertains. The terms "first," "second," and similar terms used in the embodiments of this application do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed after the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are only used to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0020] In a communication environment heavily influenced by the internet and social media, public opinion events often exhibit characteristics such as rapid dissemination, wide reach, complex transmission chains, and strong emotional fluctuations. For businesses or public institutions, public opinion risks can not only lead to reputational damage, decreased trust, and customer churn, but may also trigger a chain reaction of consequences, including compliance reviews, business disruptions, supply chain fluctuations, and increased collaboration costs. Therefore, conducting continuous monitoring and early warning of public opinion risks is of great significance: on the one hand, it can help managers promptly grasp changes in the intensity and risk level of events, and identify key dissemination nodes and risk triggering factors; on the other hand, it can provide data support for subsequent risk assessment, response decisions, and resource allocation, thereby achieving early detection, early assessment, and early handling of risks.

[0021] However, existing public opinion risk monitoring methods still have significant limitations in practical applications: First, monitoring results often rely on existing volume statistics, keyword alerts, or manual verification, and are more of a post-event summary and retrospective analysis of public opinion that has already occurred, which has a significant lag and makes it difficult to capture early signals of rising risks in a timely manner; Second, most methods focus on superficial indicators such as whether an outbreak has occurred and the scale of the outbreak, lacking a structured characterization of multi-dimensional characteristics such as the evolution path of event-related information, changes in opinions, and trends in emotional fluctuations, resulting in insufficient refined assessment of the public opinion situation; Third, in scenarios facing specific events, existing solutions generally lack the ability to quantitatively assess and predict the trend of public opinion, making it difficult to generate interpretable and verifiable predictive outputs, thus failing to effectively support the needs of risk warning and dynamic control throughout the entire event lifecycle.

[0022] In view of this, this application provides a public opinion risk assessment method that can effectively predict and assess the public opinion risks and development trends of an event.

[0023] like Figure 1 As shown, the method includes: S101. Based on the target event, the public opinion risk management target is to perform multiple rounds of content distribution evaluation operations. Specifically, public opinion risk management objectives can include sentiment orientation objectives and attitudinal orientation objectives. Based on the CAC (Cognitive-Affective-Intention / Behavioral Orientation) theory, the public's understanding of information about a target event will first generate corresponding emotional responses, which will further influence their attitudinal orientation and behavioral intentions. Sentiment orientation objectives can be used to characterize early changes in public opinion risk, while attitudinal orientation objectives are used to reflect the direction of public opinion judgment and potential behavioral tendencies. For different target events, corresponding public opinion risk management objectives are dynamically configured to adapt to differences in event attributes, audience structure, and communication environment. When the public opinion trend indicators (such as sentiment orientation and attitudinal orientation) obtained from monitoring the public gradually approach and meet the public opinion risk management objectives, the public opinion trend related to the target event can be considered to be in a relatively stable and controllable state. Target events can be social events, public issues, brand / product events, service guarantee events, etc., without specific restrictions.

[0024] Furthermore, each round of content distribution evaluation includes: S201. Based on the target distribution parameters of this round, the public opinion risk management target, and the sentiment data for the target event, determine multiple content items for this round of distribution from the content pool pre-built based on the target event; Specifically, for a target event, a content pool corresponding to that target event is pre-built; the content pool is used to store multiple pieces of content related to the target event for distribution. Based on the content pool, multiple pieces of content for this round of distribution are selected for content distribution. The distribution content in the content pool can be relevant content that may arise from the target event, including but not limited to: summaries of factual information or investigation materials, event context and impact analysis, statements of key viewpoints and organization of arguments, structured comparative analysis of different viewpoints, answers to frequently asked questions, clarifications or citations of authoritative sources, etc.

[0025] Sentiment data on target users regarding a target event is used to characterize the psychological state of target users after encountering and understanding information related to the target event. Based on the CAC (Cognitive-Affective-Intention / Behavioral Tendency) theory, after forming a cognition about the target event, target users will have corresponding emotional responses, which are further reflected in relatively stable attitude orientations. Therefore, sentiment data on target users regarding a target event may include, but is not limited to, quantitative indicators such as sentiment tendency values ​​and attitude orientation values. Sentiment tendency values ​​are used to characterize the polarity and degree of the target user's emotions towards the target event; a higher sentiment tendency value indicates a more positive emotional response, and vice versa. Attitude orientation values ​​are used to characterize the target user's viewpoint orientation towards the target event; a higher attitude orientation value indicates a more supportive orientation, and vice versa.

[0026] The sentiment data of target users is dynamic and updates with changes in the information content, source, and level of understanding they encounter. For example, when content related to a target event reaches target users, their perception of the event may change, causing corresponding fluctuations in their sentiment index, attitude orientation, and other sentiment data. These fluctuations may manifest as either an increase or a decrease in the indicators, depending on factors such as the factual information, argument structure, expression style, and the target users' existing cognitive foundation in the distributed content.

[0027] The target users can be simulated users. This involves pre-constructing a simulated user group with cognitive dimensions in the absence of or without relying on real user data, and generating controllable dynamic interaction data to provide an experimental environment for subsequent distribution evaluation. Each simulated user's profile consists of two parts: basic attributes and cognitive characteristics, formally defined as User=(D,C), where D is the set of demographic characteristics and C is the set of cognitive characteristics. Cognitive characteristics C represent the emotional tendency data for the target event, such as emotional tendency values ​​and attitude orientation values, used to characterize the simulated user's emotional polarity and opinion orientation after encountering and understanding information related to the target event. Demographic characteristics include age, gender, occupation, and education level, without specific restrictions. The user simulation module can generate diverse user profiles in batches and simulate their contact, understanding, and feedback process with distributed content in a dynamic information flow scenario. This allows the emotional tendency data to be dynamically updated with information input, enabling preliminary verification and iteration of the system's core algorithm in the absence of real data. Ultimately, a simulated user set U (i.e., the target user set) containing N differentiated users can be formed, U={User1,User2,…,User…} N}

[0028] When performing content distribution evaluation, the selection and sampling process of the content to be distributed in this round can be determined based on the target distribution parameters. For the first round of content distribution evaluation, the target distribution parameters can be determined by the quantity and type distribution of the content in the content pool, the statistical characteristics of the target user group, and the public opinion risk management objectives corresponding to the target event. Target distribution parameters may include exploration rate and temperature parameters, used to adjust the diversity and randomness of sampling in the content pool. The exploration rate and temperature parameters are used to adjust the selection strategy and sampling randomness of the content to be distributed in this round, in order to control the coverage and stability of the distribution results. The exploration rate is used to determine the quantity of content to be distributed in this round; a higher exploration rate results in a larger quantity of exploratory content and a wider coverage; a lower exploration rate results in a smaller quantity of exploratory content and a preference for stable candidates. The temperature parameter is used to adjust the intensity / dispersion of randomness when sampling within the content pool, thereby affecting the divergence and diversity of the selected content to be distributed. The higher the temperature, the flatter the sampling distribution, and the more random and dispersed the results; the lower the temperature, the more concentrated the sampling distribution, the more biased the selection is towards high-scoring candidates, and the more certain and consistent the results.

[0029] S202. Determine the distribution objects corresponding to each round of distribution content from all target users, and distribute each round of distribution content to the corresponding distribution objects to obtain user interaction data for this round. Specifically, a single piece of content distributed in this round can be distributed to multiple target users, and a single target user can receive multiple pieces of content; there are no specific restrictions. For example, the content distributed in this round... The distribution targets can include target users User1 and User2; target user User1 can receive the content distributed in this round. , , When the target user is a simulated user, the distribution recipients are also simulated users. After receiving the corresponding distributed content, the simulated user can generate interactive responses related to that content through a user behavior simulation process, thereby forming user interaction data. User interaction data may include, but is not limited to, records of behaviors such as liking, commenting, forwarding / sharing, collecting, clicking / viewing, dwell time, and no interaction, as well as associated information such as timestamps, content identifiers, and user identifiers related to these behavior records. The above interaction data can be used to characterize the user feedback results of this round of distribution and provide input for subsequent evaluation and iteration.

[0030] S203. Based on all the content distributed in this round received by each distribution object, update the sentiment data of the corresponding distribution object; Specifically, for each distribution target, the set of distribution content they have encountered in this round can be summarized, and combined with the interaction behavior and feedback results of the distribution target on each distribution content, the sentiment data of the target event can be updated, including the update of sentiment value, attitude value and other values.

[0031] S204. Based on the user interaction data of this round, calculate the evaluation result of the content distribution evaluation operation of this round; S205. Based on the evaluation results of this round of content distribution evaluation, determine the target distribution parameters for the next round of content distribution evaluation. Specifically, the evaluation results of this round of content distribution assessment can be compared with the public opinion risk management objectives to determine whether the current assessment has met expectations, whether there is insufficient coverage or excessive fluctuations, and the target distribution parameters for the next round can be adaptively adjusted accordingly.

[0032] The content of steps S201-S205 can be executed iteratively using a pre-built large language model. A large language model (LLM) is a generative model trained on large-scale text data, capable of semantic understanding, information summarization, and content generation from input text. An LLM can receive structured or semi-structured system state descriptions and output structured decision results, thus supporting tasks such as multi-source information fusion, distributed content filtering, object allocation, and evaluation parameter updates. The large language model is a pre-trained model. During its pre-training phase, it utilizes large-scale corpora for self-supervised learning, enabling the model to acquire language structure modeling capabilities, semantic association capabilities, and knowledge representation capabilities. Furthermore, training samples relevant to the task of this scheme can be introduced, allowing the model to understand the system state description and generate output results according to a pre-defined structure. This provides the model with the capabilities for content filtering, object allocation, interactive response generation, evaluation result summarization, and target distribution parameter updates adapted to steps S201-S205. After the above training, the large language model can be deployed in the system as a callable module. In each round, it reads the current system state and outputs structured decision results to support the multi-round distribution and evaluation process. It is especially suitable for simulation verification and scheme comparison under the condition of lack of real data.

[0033] S102. In response to the content distribution assessment operation reaching a preset threshold number of executions, determine the public opinion risk assessment result based on the assessment results of all rounds.

[0034] Specifically, the preset threshold number of times can be set to 10, 20, 30, 40, 50, 60, 80, 100 times, or other values, without any specific restrictions. The public opinion risk assessment result can be a comprehensive judgment of the public opinion situation of the target event within the assessment period, used to characterize the overall level of public opinion risk and its evolution characteristics; specifically, it may include, but is not limited to, the following: public opinion risk level or risk score (e.g., outputting low / medium / high risk according to preset classification rules), the changing trend of key assessment indicators (e.g., the trend curve or stage inflection point of reaching and interacting intensity, negative / positive sentiment, attitude orientation change, degree of divergence change, no-interaction ratio change, etc. with each round), risk warning signals and triggering reasons (e.g., the triggering rounds and characteristic items corresponding to abnormal fluctuations, sudden increase and spread, and intensification of divergence); and suggested outputs for subsequent handling and tracking (e.g., user groups that need to be focused on, key content categories, risk-sensitive periods, etc.). The results of public opinion risk assessment can also provide trend predictions or conclusions on future trends. For example, based on the rate of change and fluctuation of multiple assessment indicators, the direction of risk increase / decrease in the short term can be predicted, and a basis can be provided for subsequent monitoring frequency or assessment strategy adjustments.

[0035] After obtaining the public opinion risk assessment results for the target event, the content pool can be dynamically adjusted based on these results to improve the security and controllability of subsequent public opinion management and information dissemination. Specifically, based on the risk level, indicator trends, and triggering reasons, the content categories or expressions related to risk fluctuations in the content pool (such as incomplete information, highly ambiguous expressions, lack of authoritative source support, content that is easily misinterpreted, or content that may cause controversy) can be identified. Corresponding content can then be updated, supplemented, or replaced. This includes supplementing factual evidence and source explanations, adding key background and limiting conditions, improving answers to common questions and clarifications, correcting inaccurate expressions, and deleting or removing content that does not meet compliance requirements or is too risky, thereby reducing the possibility of uncertainty and the spread of misunderstandings. Furthermore, the public opinion risk assessment results can serve as a reference for subsequent information dissemination and communication, helping to develop more robust content review and dissemination strategies. This includes identifying key issues to focus on, prioritizing clarification materials and Q&A statements, establishing necessary review processes and controlling the release pace, and conducting risk assessments and compliance checks before release to avoid new public opinion risks caused by inappropriate information expression, unclear facts, or semantic ambiguity.

[0036] In this application, a closed-loop evaluation mechanism is constructed for target events, consisting of "content pool - target users (simulated users) - multi-round evaluation - feedback iteration". This mechanism enables quantitative assessment, trend analysis and risk warning of public opinion risks related to events without relying on or with weak reliance on real public opinion data. By selecting content from the content pool based on target distribution parameters, determining distribution targets, and acquiring user interaction data, this approach continuously updates user sentiment data through multiple iterations. It calculates the evaluation results for each round and adaptively adjusts the target distribution parameters for the next round, thereby improving evaluation coverage and stability while reducing the randomness and lag of single-round evaluations. By introducing and tracking user sentiment data (such as sentiment orientation and attitude orientation), it allows for a more granular characterization and quantitative comparison of changes in users' psychological state before and after exposure to information related to the target event. This identifies early risk signals such as emotional fluctuations, attitude shifts, and escalating disagreements, providing more interpretive evaluation evidence beyond relying solely on interaction behavior statistics. This effectively improves the accuracy and sensitivity of judging the evolution trend and direction of public opinion risks. After the evaluation rounds reach a preset threshold, the evaluation results of each round are summarized to output a public opinion risk assessment conclusion, providing a basis for risk monitoring, early warning, and subsequent strategy optimization. This application achieves effective prediction and assessment of public opinion risks for events and effectively improves the response efficiency to public opinion risks, better supporting the needs for risk early warning and dynamic management throughout the entire lifecycle of target events.

[0037] In some embodiments, the determination of multiple pieces of content for distribution in this round from a pre-built content pool based on the target event, based on the target distribution parameters of this round, the public opinion risk management target, and the sentiment data of all target users regarding the target event, includes: Obtain the preset sentiment data corresponding to each distributed content in the content pool; Based on the sentiment tendency data of all target users regarding the target event, the sentiment tendency baseline data is calculated; For each piece of content distributed in the content pool, the suitability data corresponding to each piece of content is calculated based on the corresponding preset sentiment tendency data, the sentiment tendency benchmark data, and the public opinion risk control target. Based on the target distribution parameters for this round and the adaptation data corresponding to each distribution content in the content pool, multiple distribution contents for this round are determined from the content pool.

[0038] Specifically, for each piece of distributed content in the content pool, based on its recorded text content, expressive characteristics, etc., a number of labels are pre-configured to represent the emotional polarity and attitude orientation that the content may embody, thus obtaining preset sentiment tendency data. The sentiment tendency baseline data can be the average of the sentiment tendency data of all target users for the target event.

[0039] Adaptation data for each piece of content distributed. It is calculated using the following formula: ; in, Preset sentiment data corresponding to the content being distributed. As the baseline data for sentiment tendency, For the purpose of public opinion risk management, As the benchmark coefficient, The target weight coefficient, It can be set to 0.7. It can be set to 0.3 or other values; there are no specific restrictions. To reflect the degree to which content that is consistent with the direction but exceeds the target value contributes to the achievement of the target, the case of correct direction but overshoot can be processed by intervalization, that is, when... Exceeding When overshoot occurs, adjustments can be made based on the overshoot amplitude. Perform assignment or mapping to 0.4~1.0; when Not exceeding At that time, it was considered detrimental to achieving the goal. Assign or map to 0.1 to 0.25.

[0040] When the public opinion risk management objectives include both sentiment orientation objectives and attitude orientation objectives, the sentiment orientation data of target users based on the target event also includes sentiment orientation values ​​and attitude orientation values. The preset sentiment orientation data corresponding to the distributed content also includes preset sentiment orientation values ​​and preset attitude orientation values. The sentiment orientation benchmark data also includes the first benchmark data of the corresponding sentiment orientation and the second benchmark data of the corresponding attitude orientation. At this point, the preset sentiment tendency value and preset attitude orientation value corresponding to each distributed content in the content pool are obtained; based on the sentiment tendency values ​​of all target users towards the target event, the first benchmark data is calculated (the first benchmark data can be the average of the sentiment tendency values ​​of all target users towards the target event); for each distributed content in the content pool, the sentiment fit degree corresponding to each distributed content is calculated according to the corresponding preset sentiment tendency value, the first benchmark data, and the sentiment tendency target; based on the attitude orientation values ​​of all target users towards the target event, the second benchmark data is calculated (the second benchmark data can be the average of the attitude orientation values ​​of all target users towards the target event); for each distributed content in the content pool, the attitude fit degree corresponding to each distributed content is calculated according to the corresponding preset attitude orientation value, the second benchmark data, and the attitude orientation target; fit degree data is calculated based on sentiment fit degree and attitude fit degree; based on the target distribution parameters of this round and the fit degree data corresponding to each distributed content in the content pool, multiple distributed content for this round are determined from the content pool. The fit degree data is obtained through sentiment fit degree. Attitude fit The following formula can be used for calculation: ) / 2; ; ; in, The preset sentiment value corresponding to the distributed content. This serves as the primary baseline data. For the emotional inclination target, The preset attitude value corresponding to the distributed content. This serves as the second baseline data. For attitude-oriented goals.

[0041] In some embodiments, determining multiple distribution content items for this round from the content pool based on the target distribution parameters for this round and the adaptation data corresponding to each distribution content item in the content pool includes: Based on the historical distribution records of each content in the content pool, the freshness data of each content is determined. Based on the target distribution parameters for this round, the adaptation data and freshness data corresponding to each distribution content in the content pool, multiple distribution contents for this round are determined from the content pool.

[0042] Specifically, for each piece of content distributed in the content pool, its distribution and interaction data in historical rounds are used to calculate freshness data, which characterizes whether the content has recently reappeared, whether it has been sufficiently exposed, and whether it still has timely value. This freshness data is used as one of the criteria for subsequent content filtering and sorting. Freshness data can be a numerical or rank-based indicator, used to characterize the recentity and repetition of the distributed content. For example, content with a longer time since its most recent distribution, fewer historical distributions, and a lower rate of repeated reach can have its freshness data set to higher; conversely, if content is distributed frequently in a short period of time, appears repeatedly in multiple rounds, or has already covered a large number of users, its freshness data can be set to lower to reduce the redundancy and fatigue effect caused by repeated distribution.

[0043] The number of content items (K) to be distributed in this round is determined by the target distribution parameters. Then, the content in the content pool is sorted and filtered: first, the content is sorted in descending order based on fit data; then, freshness data is used as a ranking weight or secondary ranking criterion to further rearrange the content; when content with the same fit data or falling into the same level range appears, the content with higher freshness data is prioritized. Through this sorting and filtering, K content items for this round of distribution, prioritizing fit while also considering content freshness, can be obtained. This ensures that the content distributed in this round matches the public opinion risk management objectives, reduces information redundancy and user fatigue risks caused by high-frequency repetitive distribution, improves the diversity and timeliness of content coverage, and makes the subsequently acquired user interaction data more representative and stable, thereby improving the effectiveness and credibility of the content distribution evaluation results.

[0044] In some embodiments, determining the distribution target corresponding to each round of distribution content from all target users includes: Obtain the preset sentiment data corresponding to each piece of content distributed in this round; Candidate users are determined from all the target users; For each piece of content distributed in this round, the sentiment matching data between it and each candidate user is calculated based on the corresponding suitability data. Based on the sentiment matching data between each piece of content distributed in this round and each candidate user, the distribution targets corresponding to each piece of content distributed in this round are determined.

[0045] Specifically, the target user base may be large (e.g., a massive number of users). Calculating the "content distribution - target user" matching relationship directly for each of these users can easily lead to computational overhead that increases quadratically with the user base, resulting in a performance bottleneck. To address this, a Neural Combinatorial Optimizer (NCO) can be introduced to reduce the computational burden of a large user base. When the number of target users is large, a user random sampling strategy can be used to select a predetermined number of users (e.g., 100 users) as candidate users from all target users. Subsequent matching and allocation calculations can then be performed based on these candidate users. Conversely, when the number of target users is small or computational resources allow, all target users can be directly identified as candidate users to improve coverage and evaluation sufficiency; no specific restrictions apply.

[0046] The following formula can be used to calculate the sentiment matching data between the content distributed in this round and the candidate users. : ; in, This refers to the compatibility data corresponding to the content distributed in this round. To determine the degree of alignment between the content distributed in this round and the stance of the candidate users, The emotional match between the content distributed in this round and the candidate users; =1.0-min(1.0,|Stance post -Stance user | / 2.0), =1.0-min(1.0,|Sent post -Sent user | / 2.0), Stance post Stance represents the preset attitude value for the content distributed in this round. user Sent represents the candidate user's attitude towards the target event. post Sent is the preset sentiment value for the content distributed in this round. user This represents the sentiment index of candidate users towards the target event. The activity level of candidate users can be determined based on their historical interaction data, with a value ranging from 0 to 1. , , , These are the preset first coefficient, preset second coefficient, preset third coefficient, and preset fourth coefficient, respectively. It can be set to 0.5, or it can be set to other values. It can be set to 0.2, or it can be set to other values. It can be set to 0.2, or it can be set to other values. It can be set to 0.1 or other values; there are no specific restrictions.

[0047] In some embodiments, determining the distribution target corresponding to each round of distribution content based on the sentiment matching data between each round of distribution content and each candidate user includes: For each piece of content distributed in this round, based on its sentiment matching data with each candidate user, all candidate users are divided into a first user set and a second user set. The sentiment matching data of candidate users in the first user set is higher than that of candidate users in the second user set. Based on the target proportion for this round, the distribution targets are determined from the first user set and the second user set.

[0048] Specifically, a "content-candidate user" matching matrix can be constructed based on the sentiment matching data between the content distributed in this round and the candidate users. A preliminary allocation scheme is then generated under adaptive user capacity constraints to limit the capacity of individual users, avoid over-allocation, and ensure sufficient candidate coverage for each piece of content. Specifically, a greedy allocation strategy can be adopted: first, the content distributed in this round is sorted according to the average matching degree of each piece of content in the candidate user set to determine the content processing order; then, for each piece of content distributed in this round, candidate users are selected sequentially from high to low sentiment matching data until the upper limit of the candidate user capacity corresponding to that content is reached. By introducing the above sorting and item-by-item allocation mechanism, the bias caused by a fixed order can be reduced, preventing some content from being prioritized or ignored for a long time. Finally, a high-scoring candidate pool composed of highly matched users can be formed for each piece of content; the high-scoring candidate pool corresponds to the first user set, and the remaining candidate users correspond to the second user set.

[0049] Candidate user capacity limit It can be defined as: N posts N represents the number of contents distributed in this round. users Total number of candidate users.

[0050] The target ratio for this round is used to define the proportion of distribution targets selected from the first user set and the second user set. The number of distribution targets for each piece of content can be constrained and configured by a preset distribution quota. The preset distribution quota indicates the number of target users to be allocated to a single piece of content in this round; it is a pre-set parameter, and its specific value can be configured according to coverage requirements and computing resources, for example, it can be set to 20, 30, etc., but is not limited. Based on the preset distribution quota and the target ratio for this round, the number of users selected from the first user set and the second user set for each piece of content can be determined. For example, when the target ratio is 6:4 and the preset distribution quota is 20, then 12 target users are selected from the first user set as distribution targets, and 8 candidate users are selected from the second user set as distribution targets, forming the distribution targets corresponding to the content distributed in this round.

[0051] By mixing the distribution targets into the first and second user sets according to the target proportion of this round, cross-group coverage can be introduced while ensuring matching accuracy, thereby significantly improving the heterogeneity and coverage breadth of the distribution targets. This reduces coverage bias caused by excessive concentration of distribution results on highly matched users, alleviating the homogeneous dissemination caused by information circulating within local user groups for extended periods. Furthermore, it increases the probability of content reaching users with different interests and interaction habits, enhancing the connectivity and cross-circle dissemination capabilities of information, and making subsequent user interaction data more representative and generalizable.

[0052] In some embodiments, after calculating the evaluation result of the current round of content distribution evaluation operation based on the current round of user interaction data, the method further includes: Based on the evaluation results of this round of content distribution evaluation, the target proportion for the next round of content distribution evaluation will be determined.

[0053] Specifically, the results of this round of evaluation can be compared with the objectives of public opinion risk management. Taking into account changes in coverage, diversity indicators, interactive feedback distribution, and sentiment data, the selection ratio of the first and second user sets can be adaptively adjusted: when the evaluation results show insufficient coverage or excessively concentrated user feedback, the proportion of the second user set can be appropriately increased to enhance diversity and cross-group coverage; when the evaluation results show insufficient matching effect or poor evaluation stability, the proportion of the first user set can be appropriately increased to improve overall matching quality and evaluation consistency. This allows the target ratio for the next round to be dynamically updated based on evaluation feedback, supporting the gradual convergence of multiple rounds of evaluation and outputting more stable evaluation conclusions.

[0054] In some embodiments, the step of calculating the sentiment matching data between each piece of content distributed in this round and each candidate user based on the corresponding suitability data includes: Based on the historical distribution records of each piece of content in this round, determine the freshness data of each piece of content in this round. For each piece of content distributed in this round, matching data between it and each candidate user is calculated based on the corresponding freshness and suitability data.

[0055] Specifically, the freshness of the content distributed in this round can also be used to calculate the sentiment matching data between the content and each candidate user. At this point, the sentiment matching data... It can be calculated using the following formula: ; in, The freshness factor for the content distributed in this round can be determined based on the historical distribution records of the content. This reduces the probability of repeated distribution of recently frequently exposed content and increases the entry opportunities for less exposed content, while ensuring matching priority. The freshness factor can be assigned a value based on the post's exposure in historical rounds. For example: undistributed content is assigned a value of 1.0; if the content was distributed in the most recent round, the freshness factor drops to 0.3; if the interval since the most recent distribution is 5-10 rounds, the freshness factor increases to 0.6; if the interval exceeds 10 rounds, the freshness factor increases to 0.8, but remains lower than the value for undistributed content, reflecting the higher novelty priority of historically unexposed content. This method allows for a simultaneous consideration of content suitability and novelty when calculating candidate user matching data, improving the coverage diversity and evaluation representativeness of the distribution results.

[0056] In some embodiments, calculating the evaluation result of the current round of content distribution evaluation operation based on the current round of user interaction data includes: Based on the user interaction data of this round, the distribution effect evaluation data is calculated using a pre-built large language model; Based on the user interaction data of this round, the correlation data between the distributed content and the user interaction data, the content interaction rate data, and the user coverage data of this round are calculated. Calculate the deviation between the current sentiment data of all target users and the aforementioned public opinion risk management target; Based on the distribution effect evaluation data, the correlation data, the content interaction rate data, the user coverage data, and the deviation data, the evaluation results of this round of content distribution evaluation operation are determined.

[0057] Specifically, the large language model receives aggregated statistical results of user interaction data from the current round as input to conduct high-level qualitative analysis of distribution effectiveness, depicting the overall trend and pattern of distribution behavior. The aggregated statistical results include macro-level indicators such as total interactions, average interactions per post, number of participating users, content coverage, and their trends across rounds. The aggregated statistical results can be summary data from the most recent preset rounds (e.g., the last five rounds), meaning they include not only user interaction data from the current round but also user interaction data from the previous four rounds. Upon receiving the aggregated statistical results, the large language model outputs distribution effectiveness evaluation data, i.e., an overall effectiveness index (standardized to 0-1), used to measure the overall performance of content distribution in this round.

[0058] The correlation between the content distributed in this round and user interaction data was calculated using the following formula, including: ; ; ; in, For related data, The normalized entropy is N, where N is the total number of contents distributed in this round. This is the Shannon entropy corresponding to this round. For the first This round of distribution content, For the first This round of distribution content The percentage of interactions, =No. This round of distribution content The corresponding number of interactions / the total number of interactions for all content distributed in this round.

[0059] Interaction volume can include likes, comments, and the number of users reached / distributed. The total interaction volume across all distributed content can be normalized into a probability distribution to characterize the distribution of interactions across different content. Shannon entropy measures the dispersion of interactions across content dimensions. It can be normalized to the theoretical maximum entropy value (related to the number of content distributed in this round) to obtain normalized entropy. Higher normalized entropy indicates more dispersed interactions and more diverse content combinations; lower normalized entropy indicates interactions are more concentrated on a few types of content. Relevance data, an approximate mutual information indicator, reflects the concentration of correlation between distributed content and user interactions. A higher indicator usually indicates interactions are more concentrated on specific content or a specific subset of users; a lower indicator indicates a more balanced and dispersed distribution of content and interactions. Thus, relational data can characterize the correlation structure and concentration / dispersion characteristics of the distributed content and user interaction data in this round.

[0060] Content interaction rate data is used to characterize the average interaction level of this round of content. Specifically, the average interaction per post (avg) is first calculated. actions :avg actions =actions total / num posts actions total num represents the total number of interactions in this round. posts The number of content distributed in this round; then normalized to an ideal value of 10 interactions per post, and the result is limited to the range of 0-1: interaction score =min(1.0, avg actions / 10.0), interaction score This refers to content interaction rate data.

[0061] User coverage data is used to characterize the actual interaction coverage among the users assigned in this round. Specifically, it represents the number of deduplicated users actually assigned (or reached) in this round.users_count When )>0, first calculate the proportion of actual participating users to actual allocated users: coverage = unique users / assigned users_coun Among them, unique users This represents the number of unique users who interacted in this round, assigned. users_count This represents the number of unique users actually allocated (or reached) in this round; subsequently, the proportion of actual participating users to actual allocated users is cropped to a range of 0-1 to obtain user coverage data. score :coverage score =min(1.0,max(0.0,coverage)).

[0062] After obtaining correlation data, content interaction rate data, and user reach data, these can be weighted and fused to obtain deterministic effect evaluation data. It can be calculated using the following formula: coverage score ; in, This refers to content interaction rate data; Related data; coverage score For user coverage data, , , These are the first weighting factor, the second weighting factor, and the third weighting factor, respectively. It can be set to 0.5. It can be set to 0.3. Can be set to , , , Other values ​​can also be set; there are no specific restrictions.

[0063] Deviation data between current sentiment data of all target users and public opinion risk management objectives It can be calculated using the following formula: ; ; ; in, The overall average stance of all target users, i.e., the average attitude value of all target users towards the target event; target The attitude orientation objective for public opinion risk management.current This is the overall sentiment mean of all target users, calculated by averaging the sentiment inclinations of all target users towards the target event. target This refers to the sentiment orientation target within the public opinion risk management objectives. After each recipient receives the distributed content, their sentiment orientation data changes; that is, both their attitude and sentiment values ​​towards the target event are updated accordingly. Therefore, by using the updated sentiment orientation data and calculating the deviation data from the public opinion risk management objectives, we can more timely and granularly quantify changes in the public opinion landscape. This provides more interpretive assessment criteria beyond interactive behavior statistics, effectively identifying trends of public opinion converging towards or deviating from the target range, quickly capturing early risk signals such as emotional fluctuations, attitude shifts, and escalating disagreements, and improving the sensitivity and accuracy of public opinion risk trend analysis and early warning.

[0064] Finally, the evaluation results of this round of content distribution evaluation are calculated using the following formula: ; in, This is the final score for this round of content distribution evaluation, i.e., the evaluation result for this round of content distribution evaluation. For distribution effectiveness evaluation data, For deterministic effect evaluation data, For deviation data, , , These are the first factor coefficient, the second factor coefficient, and the third factor coefficient, respectively. It can be set to 0.3. It can be set to 0.3. Can be set to , , , Other values ​​can also be set; there are no specific restrictions.

[0065] Once the evaluation results of this round of content distribution evaluation are obtained, the target distribution parameters for the next round of content distribution evaluation can be determined based on these results. For example, if the evaluation result of this round of content distribution evaluation is lower than 0.4, the exploration rate for the next round can be increased.

[0066] In addition to providing comprehensive conclusions such as risk level / risk score, trend and inflection point, the results of public opinion risk assessment can further include a set of quantitative indicators to characterize interaction intensity, dissemination, coverage, system operation and user experience quality, supporting multi-dimensional assessment and comparative analysis of the public opinion situation. These quantitative indicators can be organized by category as participation indicators, influence indicators, dissemination scope indicators, system performance indicators, user experience indicators, and actual effect indicators.

[0067] Engagement metrics measure the depth and breadth of user interaction and can include engagement rate and interaction diversity. Engagement rate = number of comments / number of unique users, reflecting the depth of engagement among core users; where the number of comments is the total number of comments generated by all users, and the number of unique users is the number of duplicate users who participated in the interaction. Interaction diversity = number of interaction types / total number of interactions, assessing the richness of user behavior patterns; the number of interaction types is the number of different actions a user can perform (such as liking, commenting, sharing, etc.), and the total number of interactions is the sum of the number of times all types of interactions were performed.

[0068] Influence metrics are used to assess the strength of content's influence across a user network, and can include average influence. Average influence is calculated as follows: first, the ratio of users reached to total interactions is calculated for each post; then, this ratio is averaged across all distributed content. The resulting average value quantifies the average influence strength of a single piece of content. Here, the number of users reached is the number of users who participated in any interaction (deduplicated), and the total number of interactions is the sum of all types of interactions. Influence metrics can also be used to compare the relative trends in content reach and interaction structure across different rounds or strategies.

[0069] The reach metric measures the breadth of content reach and the depth of interaction after reach, and can include the number of people reached and the depth of reach. The number of people reached can be directly counted as the total number of unique users who participated in the interaction, that is, the number of users obtained by collecting user identifiers from all interaction records and removing duplicates; the depth of reach = total number of interactions / number of people reached, which reflects the average interaction intensity generated by each user reached, and the higher the value, the more sufficient the interaction after reach.

[0070] System performance metrics are used to monitor system operating efficiency and resource consumption, and can include average response time. Average response time can be extracted from the response time series in the distribution logs and the average value is calculated. Response time can be obtained from the end time to the start time, and can be further subdivided by processing stage, such as large language model call time (time spent waiting for the model to return results), user behavior simulation time (processing time for simulating user thinking and actions), evaluation processing time (time spent in this round of evaluation process), etc., in order to locate performance bottlenecks and optimize them.

[0071] User experience metrics are used to evaluate the quality of interactive experiences from the user's perspective and can include user engagement scores. User engagement scores are calculated by assigning differentiated weights to different interactive behaviors and then calculating a weighted average. For example, a comment is worth 1.0 point, a like is worth 0.5 points, a share is worth 1.5 points, and other minor behaviors are worth 0.1 points (e.g., simply browsing or no significant interaction). This weighted average is then applied to all interaction records in the current round. A higher score generally indicates higher quality user engagement and more in-depth and valuable interaction.

[0072] Actual effectiveness metrics are used to evaluate the overall efficiency of the distribution module from a macro perspective. Actual effectiveness metric = Number of participating users / Number of published content, used to characterize the average user reach per piece of content; where the number of participating users is the number of unique users who interacted (deduplicated), and the number of published content is the number of content distributed in this round. Through the combined output of these multi-dimensional metrics, the public opinion risk assessment results can present a structured picture of the public opinion situation from multiple perspectives such as "interaction—diffusion—coverage—performance—experience—efficiency," providing a quantitative basis for trend analysis, threshold warning, and strategy adjustment.

[0073] It should be noted that the method in this embodiment can be executed by a single device, such as a computer or server. The method can also be applied in a distributed scenario, where multiple devices cooperate to complete the task. In such a distributed scenario, one of these devices may execute only one or more steps of the method in this embodiment, and the multiple devices will interact with each other to complete the method described.

[0074] It should be noted that the above description describes some embodiments of this application. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in a different order than that shown in the above embodiments and still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0075] Based on the same inventive concept, corresponding to any of the above embodiments, this application also provides a public opinion risk assessment device.

[0076] refer to Figure 2 The device includes: The cyclical content distribution assessment module 201 is used to perform multiple rounds of content distribution assessment operations based on the public opinion risk management objectives of the target event. Each round of content distribution assessment operations includes: Based on the target distribution parameters of this round, the public opinion risk management target, and the sentiment data of all target users towards the target event, multiple pieces of content to be distributed in this round are determined from the content pool pre-built based on the target event; From all the target users, determine the distribution objects corresponding to each round of distribution content, and distribute each round of distribution content to the corresponding distribution objects to obtain the user interaction data for this round. Based on all the content distributed in this round received by each distribution recipient, update the sentiment data of the corresponding distribution recipient; Based on the user interaction data of this round, the evaluation results of this round of content distribution evaluation operation are calculated; Based on the evaluation results of this round of content distribution evaluation, the target distribution parameters for the next round of content distribution evaluation will be determined. The public opinion risk assessment module 202 is used to determine the public opinion risk assessment result based on the assessment results of all rounds when the number of times the content distribution assessment operation is executed reaches a preset threshold.

[0077] In some embodiments, the cyclic distribution evaluation module 201 is further configured to: Obtain the preset sentiment data corresponding to each distributed content in the content pool; Based on the sentiment tendency data of all target users regarding the target event, the sentiment tendency baseline data is calculated; For each piece of content distributed in the content pool, the suitability data corresponding to each piece of content is calculated based on the corresponding preset sentiment tendency data, the sentiment tendency benchmark data, and the public opinion risk control target. Based on the target distribution parameters for this round and the adaptation data corresponding to each distribution content in the content pool, multiple distribution contents for this round are determined from the content pool.

[0078] In some embodiments, determining multiple distribution content items for this round from the content pool based on the target distribution parameters for this round and the adaptation data corresponding to each distribution content item in the content pool includes: Based on the historical distribution records of each content in the content pool, the freshness data of each content is determined. Based on the target distribution parameters for this round, the adaptation data and freshness data corresponding to each distribution content in the content pool, multiple distribution contents for this round are determined from the content pool.

[0079] In some embodiments, determining the distribution target corresponding to each round of distribution content from all target users includes: Obtain the preset sentiment data corresponding to each piece of content distributed in this round; Candidate users are determined from all the target users; For each piece of content distributed in this round, the sentiment matching data between it and each candidate user is calculated based on the corresponding suitability data. Based on the sentiment matching data between each piece of content distributed in this round and each candidate user, the distribution targets corresponding to each piece of content distributed in this round are determined.

[0080] In some embodiments, determining the distribution target corresponding to each round of distribution content based on the sentiment matching data between each round of distribution content and each candidate user includes: For each piece of content distributed in this round, based on its sentiment matching data with each candidate user, all candidate users are divided into a first user set and a second user set. The sentiment matching data of candidate users in the first user set is higher than that of candidate users in the second user set. Based on the target proportion for this round, the distribution targets are determined from the first user set and the second user set.

[0081] In some embodiments, after calculating the evaluation result of the current round of content distribution evaluation operation based on the current round of user interaction data, the method further includes: Based on the evaluation results of this round of content distribution evaluation, the target proportion for the next round of content distribution evaluation will be determined.

[0082] In some embodiments, the step of calculating the sentiment matching data between each piece of content distributed in this round and each candidate user based on the corresponding suitability data includes: Based on the historical distribution records of each piece of content in this round, determine the freshness data of each piece of content in this round. For each piece of content distributed in this round, matching data between it and each candidate user is calculated based on the corresponding freshness and suitability data.

[0083] In some embodiments, the cyclic distribution evaluation module 201 is further configured to: The evaluation result of this round of content distribution evaluation operation is calculated based on the user interaction data of this round, including: Based on the user interaction data of this round, the distribution effect evaluation data is calculated using a pre-built large language model; Based on the user interaction data of this round, the correlation data between the distributed content and the user interaction data, the content interaction rate data, and the user coverage data of this round are calculated. Calculate the deviation between the current sentiment data of all target users and the aforementioned public opinion risk management target; Based on the distribution effect evaluation data, the correlation data, the content interaction rate data, the user coverage data, and the deviation data, the evaluation results of this round of content distribution evaluation operation are determined.

[0084] In some embodiments, the correlation data is calculated using the following formula, including: ; ; ; in, For related data, The normalized entropy is N, where N is the total number of contents distributed in this round. This is the Shannon entropy corresponding to this round. For the first This round of distribution content, For the first This round of distribution content The percentage of interactions, =No. This round of distribution content The corresponding number of interactions / the total number of interactions for all content distributed in this round.

[0085] For ease of description, the above devices are described in terms of function, divided into various modules. Of course, in implementing this application, the functions of each module can be implemented in one or more software and / or hardware.

[0086] The apparatus described above is used to implement the corresponding public opinion risk assessment method in any of the foregoing embodiments, and has the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0087] Based on the same inventive concept, corresponding to the methods of any of the above embodiments, this application also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the public opinion risk assessment method described in any of the above embodiments.

[0088] Figure 3 This embodiment illustrates a more specific hardware structure of an electronic device. The device may include a processor 1010, a memory 1020, an input / output interface 1030, a communication interface 1040, and a bus 1050. The processor 1010, memory 1020, input / output interface 1030, and communication interface 1040 are interconnected internally via the bus 1050.

[0089] The processor 1010 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this specification.

[0090] The memory 1020 can be implemented in the form of ROM (Read Only Memory), RAM (Random Access Memory), static storage device, dynamic storage device, etc. The memory 1020 can store the operating system and other applications. When the technical solutions provided in the embodiments of this specification are implemented by software or firmware, the relevant program code is stored in the memory 1020 and is called and executed by the processor 1010.

[0091] The input / output interface 1030 is used to connect input / output modules to realize information input and output. Input / output modules can be configured as components within the device (not shown in the figure) or externally connected to the device to provide corresponding functions. Input devices may include keyboards, mice, touchscreens, microphones, various sensors, etc., while output devices may include displays, speakers, vibrators, indicator lights, etc.

[0092] The communication interface 1040 is used to connect a communication module (not shown in the figure) to enable communication between this device and other devices. The communication module can communicate via wired means (such as USB, Ethernet cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.).

[0093] Bus 1050 includes a pathway for transmitting information between various components of the device, such as processor 1010, memory 1020, input / output interface 1030, and communication interface 1040.

[0094] It should be noted that although the above-described device only shows the processor 1010, memory 1020, input / output interface 1030, communication interface 1040, and bus 1050, in specific implementations, the device may also include other components necessary for normal operation. Furthermore, those skilled in the art will understand that the above-described device may only include the components necessary for implementing the embodiments of this specification, and not necessarily all the components shown in the figures.

[0095] The electronic devices described above are used to implement the corresponding public opinion risk assessment methods in any of the foregoing embodiments, and have the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0096] Based on the same inventive concept, corresponding to the methods of any of the above embodiments, this application also provides a non-transitory computer-readable storage medium storing computer instructions for causing the computer to execute the public opinion risk assessment method as described in any of the above embodiments.

[0097] The computer-readable medium of this embodiment includes permanent and non-permanent, removable and non-removable media, and information storage can be implemented by any method or technology. Information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transfer medium that can be used to store information accessible by a computing device.

[0098] The computer instructions stored in the storage medium of the above embodiments are used to cause the computer to execute the public opinion risk assessment method as described in any of the above embodiments, and have the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0099] Based on the same concept, corresponding to any of the above embodiments, this application also provides a computer program product, including computer program instructions, which, when run on a computer, cause the computer to perform the method described in any of the above embodiments, and have the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0100] It is understood that before using the technical solutions of the various embodiments in this disclosure, users will be informed of the type, scope of use, and usage scenarios of the personal information involved in an appropriate manner, and user authorization will be obtained.

[0101] For example, upon receiving a user's active request, a prompt message is sent to the user to explicitly inform them that the requested operation will require the acquisition and use of the user's personal information. This allows the user to independently choose, based on the prompt message, whether to provide personal information to the software or hardware such as electronic devices, applications, servers, or storage media performing the operations of this disclosed technical solution.

[0102] As an optional but not limited implementation, in response to a user's active request, sending a prompt message to the user can be done via a pop-up window, where the prompt message can be presented in text format. Furthermore, the pop-up window can also include a selection control allowing the user to choose "agree" or "disagree" to provide personal information to the electronic device.

[0103] It is understood that the above notification and user authorization process are merely illustrative and do not constitute a limitation on the implementation of this disclosure. Other methods that comply with relevant laws and regulations may also be applied to the implementation of this disclosure.

[0104] Those skilled in the art should understand that the discussion of any of the above embodiments is merely exemplary and is not intended to imply that the scope of this application is limited to these examples; under the concept of this application, the technical features of the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other variations of different aspects of the embodiments of this application as described above, which are not provided in detail for the sake of brevity.

[0105] Additionally, to simplify the description and discussion, and to avoid obscuring the embodiments of this application, the well-known power / ground connections to integrated circuit (IC) chips and other components may or may not be shown in the provided drawings. Furthermore, the apparatus may be shown in block diagram form to avoid obscuring the embodiments of this application, and this also takes into account the fact that the details of the implementation of these block diagram apparatuses are highly dependent on the platform on which the embodiments of this application will be implemented (i.e., these details should be fully understood by those skilled in the art). While specific details (e.g., circuits) have been set forth to describe exemplary embodiments of this application, it will be apparent to those skilled in the art that the embodiments of this application can be implemented without these specific details or with variations thereof. Therefore, these descriptions should be considered illustrative rather than restrictive.

[0106] Although this application has been described in conjunction with specific embodiments thereof, many substitutions, modifications, and variations of these embodiments will be apparent to those skilled in the art from the foregoing description. For example, other memory architectures (e.g., dynamic RAM (DRAM)) may be used with the embodiments discussed.

[0107] The embodiments of this application are intended to cover all such substitutions, modifications, and variations that fall within the broad scope of the claims of this application. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the embodiments of this application should be included within the protection scope of this application.

Claims

1. A method for assessing public opinion risk, characterized in that, include: Based on the public opinion risk management objective of the target event, multiple rounds of content distribution evaluation are performed cyclically. Each round of content distribution evaluation includes: Based on the target distribution parameters of this round, the public opinion risk management target, and the sentiment data of all target users towards the target event, multiple pieces of content to be distributed in this round are determined from the content pool pre-built based on the target event; From all the target users, determine the distribution objects corresponding to each round of distribution content, and distribute each round of distribution content to the corresponding distribution objects to obtain the user interaction data for this round. Based on all the content distributed in this round received by each distribution recipient, update the sentiment data of the corresponding distribution recipient; Based on the user interaction data of this round, the evaluation results of this round of content distribution evaluation operation are calculated; Based on the evaluation results of this round of content distribution evaluation, the target distribution parameters for the next round of content distribution evaluation will be determined. In response to the content distribution assessment operation reaching a preset threshold number of executions, the public opinion risk assessment result is determined based on the assessment results of all rounds.

2. The public opinion risk assessment method according to claim 1, characterized in that, Based on the target distribution parameters for this round, the public opinion risk management objectives, and the sentiment data of all target users regarding the target event, multiple pieces of content for this round of distribution are determined from a content pool pre-built based on the target event, including: Obtain the preset sentiment data corresponding to each distributed content in the content pool; Based on the sentiment tendency data of all target users regarding the target event, the sentiment tendency baseline data is calculated; For each piece of content distributed in the content pool, the suitability data corresponding to each piece of content is calculated based on the corresponding preset sentiment tendency data, the sentiment tendency benchmark data, and the public opinion risk control target. Based on the target distribution parameters for this round and the adaptation data corresponding to each distribution content in the content pool, multiple distribution contents for this round are determined from the content pool.

3. The public opinion risk assessment method according to claim 2, characterized in that, Based on the target distribution parameters for this round and the adaptation data corresponding to each distribution content in the content pool, multiple distribution contents for this round are determined from the content pool, including: Based on the historical distribution records of each content in the content pool, the freshness data of each content is determined. Based on the target distribution parameters for this round, the adaptation data and freshness data corresponding to each distribution content in the content pool, multiple distribution contents for this round are determined from the content pool.

4. The public opinion risk assessment method according to claim 2, characterized in that, The step of determining the distribution targets corresponding to each round of content distribution from all target users includes: Obtain the preset sentiment data corresponding to each piece of content distributed in this round; Candidate users are determined from all the target users; For each piece of content distributed in this round, the sentiment matching data between it and each candidate user is calculated based on the corresponding suitability data. Based on the sentiment matching data between each piece of content distributed in this round and each candidate user, the distribution targets corresponding to each piece of content distributed in this round are determined.

5. The public opinion risk assessment method according to claim 4, characterized in that, The process of determining the distribution targets corresponding to each round of distribution content based on the sentiment matching data between each current round of distribution content and each candidate user includes: For each piece of content distributed in this round, based on its sentiment matching data with each candidate user, all candidate users are divided into a first user set and a second user set. The sentiment matching data of candidate users in the first user set is higher than that of candidate users in the second user set. Based on the target proportion for this round, the distribution targets are determined from the first user set and the second user set.

6. The public opinion risk assessment method according to claim 5, characterized in that, After calculating the evaluation result of the content distribution evaluation operation based on the user interaction data of this round, the process further includes: Based on the evaluation results of this round of content distribution evaluation, the target proportion for the next round of content distribution evaluation will be determined.

7. The public opinion risk assessment method according to claim 4, characterized in that, For each piece of content distributed in this round, based on the corresponding suitability data, its sentiment matching data with each candidate user is calculated, including: Based on the historical distribution records of each piece of content in this round, determine the freshness data of each piece of content in this round. For each piece of content distributed in this round, matching data between it and each candidate user is calculated based on the corresponding freshness and suitability data.

8. The public opinion risk assessment method according to claim 1, characterized in that, The evaluation result of this round of content distribution evaluation operation is calculated based on the user interaction data of this round, including: Based on the user interaction data of this round, the distribution effect evaluation data is calculated using a pre-built large language model; Based on the user interaction data of this round, the correlation data between the distributed content and the user interaction data, the content interaction rate data, and the user coverage data of this round are calculated. Calculate the deviation between the current sentiment data of all target users and the aforementioned public opinion risk management target; Based on the distribution effect evaluation data, the correlation data, the content interaction rate data, the user coverage data, and the deviation data, the evaluation results of this round of content distribution evaluation operation are determined.

9. The public opinion risk assessment method according to claim 8, characterized in that, The correlation data is calculated using the following formula, including: ； ； ； in, For related data, The normalized entropy is N, where N is the total number of contents distributed in this round. This is the Shannon entropy corresponding to this round. For the first This round of distribution content, For the first This round of distribution content The percentage of interactions, =No. This round of distribution content The corresponding number of interactions / the total number of interactions for all content distributed in this round.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the program, it implements the method as described in any one of claims 1 to 9.

Images