Method and device for automatically analyzing and responding to network public opinion events based on large model agent

Abstract

The present disclosure provides a network public opinion event automatic research and judgment coping method and device based on a large model intelligent agent, which comprises: obtaining public opinion data of a public opinion event at a current time node; performing situation awareness analysis on the public opinion data of the current time node to obtain situation data of the current time node; based on the situation data of the current time node and basic information of the public opinion event and element information of a plurality of historical key time nodes, using a pre-trained public opinion coping large model to generate coping measures for the current time node; and based on the coping measures for the current time node, automatically coping with the public opinion event. The present disclosure can automatically generate coping measures for the public opinion event and automatically cope with the public opinion event through the public opinion large model, and the entire public opinion research and judgment coping process can not require human participation, and the automation is comprehensively realized through the large model intelligent agent, thereby greatly improving the accuracy and response speed of public opinion research and judgment coping.

Description

Technical Field

[0001] This disclosure relates to the field of artificial intelligence technology, and in particular to a method and apparatus for automatically assessing and responding to online public opinion events based on a large model intelligent agent. Background Technology

[0002] With the development of the internet, the impact of online public opinion events on society has become increasingly prominent. Although public opinion event monitoring and analysis technologies have made significant progress over the years, providing increasingly important technical support for the assessment and response to public opinion events, current technical means for assessing and responding to public opinion events are mainly concentrated in the assessment stage. The decision-making and response stages are primarily user-driven and involve manual decision-making and implementation by professionals. Faced with the rapidly changing online public opinion landscape, existing public opinion assessment and response technologies appear slow to respond, especially in the face of sudden public opinion events. Under the current model, after manual assessment and decision-making, the optimal opportunity for response is often missed. Therefore, how to improve the automation and response speed of public opinion event response has become a pressing technical problem for those skilled in the art. Summary of the Invention

[0003] In view of this, this disclosure proposes an automatic judgment and response method and device for online public opinion events based on large model intelligent agents, which can greatly improve the automation level and response speed of public opinion response.

[0004] According to the first aspect of this disclosure, an automatic assessment and response method for online public opinion events based on a large-scale intelligent agent model is provided, including: Obtain public opinion data for the current time point of the public opinion event; Situational awareness analysis is performed on the public opinion data at the current time point to obtain the situational data at the current time point; Based on the situational data at the current time point, the basic information of the public opinion event, and the element information of multiple historical key time points, a pre-trained public opinion response model is used to generate response measures for the current time point. Based on the response measures at the current time point, the response to the public opinion event will be carried out automatically.

[0005] In one possible implementation, when generating response measures for the current time point using a pre-trained public opinion response model based on the current time point's situational data, the basic information of the public opinion event, and the element information of historical key time points, the measures include: Based on the current situation data, the basic information of the public opinion event, and the element information of historical key time nodes, a time series data for decision-making is constructed. The decision-making basis time series data is input into the public opinion response model to generate the response measures and expected response effects for the current time node. The response measures include the types of response measures and the parameters of the response measures.

[0006] In one possible implementation, when constructing time-series data for decision-making based on the current time-point situational data, the basic information of the public opinion event, and the element information of historical key time points, the following is included: By combining the element information of multiple historical key time nodes in chronological order, the sequence data of historical key nodes at the current time node is obtained. The basic information of the public opinion event, the historical key node sequence data at the current time point, and the situation data at the current time point are combined in sequence to construct the time series data of the decision-making basis.

[0007] In one possible implementation, the large-scale public opinion response model is trained based on the basic information of each historical public opinion event and the element information of each key time node.

[0008] In one possible implementation, after performing situational awareness analysis on the public opinion data at the current time point to obtain the situational data at the current time point, the method further includes: Based on the situational data at the current time point, determine whether it is necessary to respond to the public opinion event at the current time point; When it is determined that a response is needed to the public opinion event at the current time point, the operation is then performed based on the situational data at the current time point, the basic information of the public opinion event, and the element information of historical key time points, using a pre-trained public opinion response model to generate response measures for the current time point.

[0009] In one possible implementation, when automatically responding to the public opinion event based on the response measures at the current time point, it includes: Based on the response measures at the current time point, generate an action task list; Based on the action task list, the response to the public opinion event will be carried out automatically.

[0010] In one possible implementation, when generating the action task list based on the response measures at the current time point, it includes: The response measures at the current time point are analyzed to obtain the types and parameters of the response measures; Obtain an action task template that matches the type, and fill the action task template with the response parameters to obtain the action task list of the response measures.

[0011] According to a second aspect of this disclosure, an automatic assessment and response device for public opinion events is provided, comprising: The public opinion data acquisition module is used to acquire public opinion data of public opinion events at the current time point; The public opinion situation awareness module is used to perform situation awareness analysis on the public opinion data at the current time point to obtain the situation data at the current time point. The decision planning module is used to generate response measures for the current time node based on the situation data at the current time node, the basic information of the public opinion event, and the element information of multiple historical key time nodes, using a pre-trained public opinion response model. The action response module automatically responds to the public opinion event based on the response measures at the current time point.

[0012] According to a third aspect of this disclosure, an automatic assessment and response device for public opinion events is provided, comprising: a processor; a memory for storing processor-executable instructions; wherein the processor is configured to execute the method described in the first aspect of this disclosure.

[0013] According to a fourth aspect of this disclosure, a non-volatile computer-readable storage medium is provided that stores computer program instructions thereon, wherein the computer program instructions, when executed by a processor, implement the method described in the first aspect of this disclosure.

[0014] This disclosure provides a method, apparatus, device, and storage medium for automatic assessment and response to online public opinion events based on a large-scale intelligent model. The method includes: acquiring public opinion data of the event at the current time point; performing situational awareness analysis on the public opinion data at the current time point to obtain situational data; generating response measures for the current time point using a pre-trained large-scale public opinion response model based on the situational data, basic information of the public opinion event, and element information from multiple historical key time points; and automatically responding to the public opinion event based on the response measures. The large-scale public opinion model in this disclosure can automatically generate response measures for public opinion events and automatically respond to the events based on these measures. The entire public opinion response process can be fully automated without human intervention, significantly improving the response speed.

[0015] Other features and aspects of this disclosure will become clear from the following detailed description of exemplary embodiments with reference to the accompanying drawings. Attached Figure Description

[0016] The accompanying drawings, which are included in and form part of this specification, illustrate exemplary embodiments, features, and aspects of this disclosure together with the specification and serve to explain the principles of this disclosure.

[0017] Figure 1 A flowchart is shown for an automatic assessment and response method for online public opinion events based on a large model intelligent agent according to an embodiment of the present disclosure; Figure 2 A flowchart illustrating an example of an automatic assessment and response method for online public opinion events based on a large model intelligent agent according to an embodiment of the present disclosure; Figure 3 A schematic block diagram of an automatic network public opinion event analysis and response device based on a large model intelligent agent according to an embodiment of the present disclosure is shown. Figure 4 This diagram illustrates an automatic analysis and response device for online public opinion events based on a large model intelligent agent, according to an embodiment of the present disclosure. Detailed Implementation

[0018] Various exemplary embodiments, features, and aspects of this disclosure will now be described in detail with reference to the accompanying drawings. The same reference numerals in the drawings denote elements that have the same or similar functions. Although various aspects of the embodiments are shown in the drawings, they are not necessarily drawn to scale unless specifically indicated otherwise.

[0019] The term “exemplary” as used herein means “serving as an example, embodiment, or illustration.” Any embodiment illustrated herein as “exemplary” is not necessarily to be construed as superior to or better than other embodiments.

[0020] Furthermore, to better illustrate this disclosure, numerous specific details are set forth in the following detailed description. Those skilled in the art will understand that this disclosure can be practiced without certain specific details. In some instances, methods, means, components, and circuits well known to those skilled in the art have not been described in detail in order to highlight the main points of this disclosure.

[0021] <Method Implementation> Figure 1 The flowchart illustrates an automatic assessment and response method for online public opinion events based on a large-scale intelligent agent according to an embodiment of the present disclosure. This method is implemented by an automatic assessment and response system for public opinion events, such as... Figure 1 As shown, the method includes steps S1100-S1400.

[0022] S1100, Obtain public opinion data for the current time point of the public opinion event. Public opinion data refers to the collection of information related to the current online public opinion event, gathered from various online public opinion spheres (such as Weibo, news websites, short video platforms, etc.). This public opinion data can include at least one of the following: text data, multimedia data, user interaction data, metadata, and event-related data. Text data can be, for example, news articles, Weibo posts, user comments in forums, etc.; multimedia data can be images, short video clips, live stream recordings, audio content, etc.; user interaction data can be the number of Weibo reposts, likes, number of followers of a user account, and their location, etc.; metadata can be when the public opinion data was published, on which platform it was published, the corresponding webpage link, and the reposting path of this public opinion data, etc.; event-related data includes keywords related to the current public opinion event (such as "product quality of a certain brand") and trending topic tags (such as trending topics on Weibo), etc.

[0023] Before acquiring public opinion data, it is necessary to set the data acquisition parameters. These parameters can include at least one of the following: data source, data type, and update frequency. The data source specifies the platform on which the public opinion data is collected. The data type specifies the data type to be collected, which can be at least one of text, images, or videos. The update frequency specifies the frequency at which the public opinion data is collected.

[0024] After setting the parameters for acquiring public opinion data, the data can be obtained according to these parameters. Specifically, based on the set data source, the corresponding tool is called to crawl the public opinion data. For example, if the data source is a platform with an open API (such as Weibo), the data is directly acquired through the platform's open API interface. Alternatively, if the data source is a website without an API (such as some forums), the behavior of a real person browsing the webpage is simulated to "scrape" the public opinion data from the page. During the data crawling process, the data will be crawled according to the specified frequency and data type.

[0025] In one possible implementation, the parameters for acquiring public opinion data can be adaptively adjusted based on the analysis results of the public opinion response model. Specifically, when the analysis result of the public opinion response model indicates that the public opinion event requires focused attention or response, at least one of the following first public opinion data acquisition parameter adjustment strategies will be generated: expanding the data source to include collection of related data sources such as accounts and linked web pages associated with the public opinion event; expanding the data collection types to comprehensively collect various types of data related to the public opinion event, such as text, images, and videos; and increasing the update frequency of data such as comments and likes related to the public opinion event. When the analysis result of the public opinion response model indicates that the public opinion event no longer requires focused attention or that the response action has ended, a second public opinion data acquisition parameter adjustment strategy will be generated: restoring the data source, data type, and update frequency to the system default values. In this way, the public opinion data acquisition parameters can be adaptively adjusted by executing either the first or second public opinion data acquisition parameter adjustment strategy.

[0026] After obtaining the public opinion data for the current time point, step S1200 can be executed to perform situational awareness analysis on the public opinion data for the current time point, thereby obtaining the situational data for the current time point. This situational data includes basic situational information of the public opinion event at the current time point, the index values ​​of various situational indicators in a pre-constructed situational indicator system, the index values ​​of a comprehensive situational indicator that can comprehensively reflect the urgency of the public opinion event at the current time point, and the predicted development trends of each situational indicator and the comprehensive situational indicator.

[0027] When conducting situational awareness analysis on public opinion data at the current time point to obtain the situational data for that time point, the following steps may be included: First, the public opinion data at the current time point is used to extract and quantitatively analyze public opinion elements, obtaining basic situation information and indicator values ​​for various situation indicators at the current time point. The basic situation information may include at least one of the following: the causes of public opinion changes, content evolution, changes in popularity, changes in sentiment, dissemination paths, and user group characteristics. The pre-constructed situation indicator system may include at least one of the following situation indicators: dissemination degree, popularity, and sensitivity.

[0028] Second, based on the values ​​of each situation indicator at the current time point, the overall situation indicator value at the current time point is calculated. Specifically, the system pre-sets corresponding weight coefficients for each situation indicator. After calculating the values ​​of each situation indicator at the current time point, the values ​​of each situation indicator are weighted and summed according to their respective weight coefficients. The summation result is the overall situation indicator value at the current time point.

[0029] In one possible implementation, the weight coefficients corresponding to each situation indicator in the system can be dynamically adjusted and updated based on continuously accumulated data such as human feedback, response plans, and response effectiveness evaluations, through linear regression or reinforcement learning.

[0030] Third, the development trends of each situation indicator and the comprehensive situation indicator at the current time point are predicted to obtain the prediction results. Specifically, for each situation indicator that changes over time, the indicator values ​​at the current time point and at previous time points are sorted and combined in chronological order to obtain the time series of each situation indicator. Then, a fitting algorithm is used to fit the time series of each situation indicator to obtain the prediction results of the development trend of each situation indicator. Similarly, the indicator values ​​of the comprehensive situation indicator at the current time point and at previous time points are sorted and combined in chronological order to obtain the time series of the comprehensive situation indicator. A fitting algorithm is used to fit the time series of the comprehensive situation indicator to obtain the prediction results of the development trend of the comprehensive situation indicator. The fitting algorithm can be implemented using the least squares method, time series models, and neural network models, etc., and no specific limitation is made here.

[0031] The basic situation information of the public opinion event at the current time point, the index values ​​of various situation indicators, the index values ​​of the comprehensive situation indicators, and the development trend prediction results of each situation indicator and the comprehensive situation indicators together constitute the situation data of the public opinion event at the current time point.

[0032] In one possible implementation, after obtaining the situational data at the current time point, the following steps are also included: determining whether it is necessary to respond to the public opinion event at the current time point based on the situational data at the current time point. If it is determined that it is necessary to respond to the public opinion event at the current time point, then step S1300 is executed.

[0033] In this embodiment, when determining whether to respond to a public opinion event at the current time point based on the situational data, this can be achieved based on a pre-set response threshold. The response threshold can include at least one of the response thresholds corresponding to individual situational indicators and the response threshold corresponding to a comprehensive situational indicator. Specifically, if the value of any situational indicator exceeds its corresponding response threshold, and / or the value of the comprehensive situational indicator exceeds its corresponding response threshold, it is determined that a response to the public opinion event at the current time point is necessary; otherwise, no response is required.

[0034] Furthermore, when it is determined that a response is needed to a public opinion event at the current time point, this current time point is identified as a critical time point. For each identified critical time point, not only is the corresponding situational data recorded, but also the response measures and their effects are recorded after the response to the critical time point has concluded. The situational data, response measures, and effects of the critical time point are then linked and stored in the system's short-term memory module as key information elements of the critical time point, along with the basic information of the public opinion event. In other words, throughout the entire process of responding to a public opinion event, the system continuously identifies critical time points and links and stores the basic information of the public opinion event and the key information elements of each critical time point in the short-term memory module. This allows for the subsequent generation of response measures for the current time point based on the data stored in the short-term memory module and in conjunction with the situational data at the current time point. In a specific example, for a public opinion event that is currently occurring, the short-term memory module records the basic information of the public opinion event in sequence: the situation data, response measures and response effects of key time node 1, the situation data, response measures and response effects of key time node 2, ..., the situation data, response measures and response effects of key time node i, where key time node i is the previous key time node of the current moment.

[0035] The basic information of a public opinion event may include the type of the event, content tags, time and location.

[0036] Response measures can include the types of response measures and specific parameters. Types of response measures can include at least one of the following: publishing promotional materials (product promotional articles, statements regarding public opinion events, product marketing copy, etc.), publishing online comments and replies, generating public opinion response command and coordination instructions, and generating public opinion reports. Parameters of the response measures are those that indicate the specific content of the response. For example, when the response measure type is publishing online comments and replies, the corresponding parameters can include the specific website platform, link address, comment content, number of comments, and commenting account.

[0037] The expected response effectiveness can be characterized by at least one of the following indicators: the rate of change of comprehensive situation indicators, the rate of change of popularity, and the degree of improvement in sentiment. The calculation formulas for each response effectiveness indicator are as follows: Rate of change of comprehensive situation indicators = (Current comprehensive situation indicator value - Comprehensive situation indicator value before execution) / Comprehensive situation indicator value before execution; Popularity change rate = (Current popularity - Popularity before execution) / Popularity before execution; Emotional improvement = the increase in the proportion of positive emotions.

[0038] S1300, based on the current situation data, basic information of public opinion events, and element information of multiple historical key time nodes, uses a pre-trained public opinion response model to generate response measures for the current time node.

[0039] Before implementing the methods disclosed herein, a large-scale public opinion response model needs to be trained based on the basic information of each historical public opinion event and the element information of each key time node. The specific training steps are as follows: First, the system acquires basic information about each historical public opinion event and key time point information. Specifically, during the response to public opinion events, the system stores basic information about each historical public opinion event and key time point information through a long-term memory module. This allows the system to retrieve the basic information about each historical public opinion event and key time point information from the long-term memory module. The key time point information is the same as described above and will not be repeated here.

[0040] Second, based on the basic information of each historical public opinion event and the element information of each key time node, time series data of the decision-making basis for each historical public opinion event is constructed. Specifically, for each historical public opinion event, the element information of each key time node of the historical public opinion event is combined in chronological order to obtain the historical key node sequence data; by combining the basic information of the historical public opinion event and the historical key node sequence data in sequence, the time series data of the decision-making basis for each historical public opinion event can be constructed.

[0041] For example, regarding public opinion event A, its basic information is B, and the key information elements for five key time points are C, D, E, F, and G. Then, the historical key node sequence data composed of the key information elements for these five key time points is {C}. D E F G}, will combine basic information B and historical key node sequence data {C} D E F By sequentially combining G}, we obtain the time series data of the decision-making basis for historical public opinion event A as {B}. C D E F G}.

[0042] Third, training data for the large language model is constructed based on the time series data of the decision-making basis for each historical public opinion event. Specifically, the time series data of the decision-making basis for each historical public opinion event is vectorized and high-order features are extracted to obtain the high-order feature sequences of the decision-making basis for each historical public opinion event. The high-order feature sequences of the decision-making basis for all historical public opinion events constitute the training data for the large language model.

[0043] Vectorizing the time-series data of decision-making criteria for historical public opinion events involves mapping each data point in the time-series data to its corresponding vector representation. Continuing the previous example, the time-series data of decision-making criteria for historical public opinion event A is S={B C D E F G}, then S is vectorized, that is, the data B in S are sequentially vectorized. C D E F G is mapped to the corresponding vector representation. The process The vector representation sequence of S can be obtained by combining them in order. In this context, vectors represent sequences. The calculation formula is as follows: In the formula, It is a vectorization function used to map each data s in the time series data S that forms the basis of the decision to a corresponding vector representation.

[0044] It should be noted that when the data s in the time series data S used for decision-making is indicator data, the indicator value of the indicator data can be directly used as the vector representation of the indicator data.

[0045] The time series data S, which forms the basis for decision-making in historical public opinion events, is vectorized to obtain the corresponding vector representation sequence. Then, we will continue to work on the vector representation sequence. High-order feature extraction is performed to obtain the high-order feature sequence Z, which is the basis for decision-making in historical public opinion events.

[0046] In one possible implementation, the vector representation sequence is... Higher-order feature extraction is implemented using a Transformer encoder. Specifically, the Transformer encoder transforms the vector representation sequence... The vector representation of each sequence position is concatenated with its corresponding positional information and then fused to produce a higher-order vector representation for each sequence position. This higher-order vector representation retains both the semantic information of the vector representation and the temporal information of the corresponding position. Combining the higher-order vector representations of all sequence positions in sequence yields the higher-order feature sequence Z, which forms the basis for the decision, thus achieving a mapping from the input space to the model representation space. The formula for calculating the higher-order feature sequence Z is shown below: In the formula, It is a positional encoding matrix, which includes the positional information of each sequence position. The positional information of each sequence position is obtained by encoding its temporal information using sine-cosine scaling, and the absolute temporal relationship is preserved in this positional information. That is, the high-order feature sequence of the decision basis output by the encoder is a comprehensive representation that includes event semantic information and spatiotemporal features.

[0047] It should be noted here that the Transformer encoder captures long-term dependencies between historical nodes through a self-attention mechanism, as shown in the following formula: in It is the input vector representation sequence Query, key-value matrix, It is the scaling factor.

[0048] Fourth, by training the large language model with training data, a large public opinion response model can be obtained. Specifically, the large language model is trained using training data from each historical public opinion event. This enables the large language model to plan response measures for the current time point and predict their effects based on the basic information of the public opinion event, the element information of each key time point that has occurred, and the situation data of the current time point. The large language model with the above capabilities is the public opinion response model disclosed in this publication.

[0049] In one possible implementation, to improve the accuracy of training large language models, a multi-task loss function is designed. This multi-task loss function is obtained by weighted summation of the losses from the types of public opinion response measures, response parameters, and expected response effects. The types of response measures represent a typical classification problem, and the loss function is designed based on the cross-entropy loss function to measure the difference between the probability distribution of the model's output and the probability distribution of the true classification labels. The expected response effect represents a typical regression problem, and the loss function is designed based on the mean squared error loss function to measure the average of the squared differences between the model's predicted values ​​and the true values.

[0050] In one possible implementation, the multi-task loss function The calculation formula is as follows: In the formula, To address the cross-entropy loss of different measures, To address the loss of parameters, The mean squared error loss represents the expected response effect. This represents the task weight coefficient, the magnitude of which is set according to specific requirements. Among these, the cross-entropy loss is calculated based on the type of response measure. The calculation formula is as follows: In the formula, These are the true labels for the types of response measures for i samples. To predict probabilities.

[0051] Response measures parameter loss The calculation formula is as follows: In the formula, t It is a time step. It is a time step t target words, It is a time step t All previous target words, H It is the model's hidden state (including contextual information). The representation model is based on historical sequences. Predict the probability of the current word.

[0052] Mean squared error loss in predicting expected response effects The calculation formula is as follows: In the formula, This is the situation index predicted by the model. This is the actual value.

[0053] In one possible implementation, during the training and optimization of a large language model, the optimizer uses the AdamW optimizer to update the model parameters, decoupling weight decay from gradient updates and improving generalization ability; it sets a hierarchical learning rate, with a smaller learning rate (e.g., 1e-5) for the bottom layer parameters of the encoder and a larger learning rate (e.g., 1e-3) for the top layer and output layer; and it uses a cosine annealing scheduler to dynamically adjust the learning rate to accelerate convergence.

[0054] After training and obtaining a large-scale public opinion response model, this model can be used to generate response measures for the current time point. Specifically, this may include the following steps: First, based on the current situational data, the basic information of the public opinion event, and the element information of historical key time nodes, a time series data for decision-making is constructed. Specifically, the basic information of the public opinion event and the element information of historical key time nodes are read from the aforementioned short-term memory module; the element information of multiple historical key time nodes is combined in chronological order to obtain the historical key node sequence data at the current time node; the basic information of the public opinion event, the historical key node sequence data at the current time node, and the situational data at the current time node are combined sequentially to construct the time series data for decision-making.

[0055] For example, regarding public opinion event A, its basic information is B, the element information of the five key time nodes that have occurred is C, D, E, F, and G, the situational data at the current time node is H, and the historical key node sequence data at the current time node, composed of the element information of the five key time nodes, is {C}. D E F G}, will sequence the basic information B and the historical key node sequence data at the current time point {C} D E F The time series data {B} is obtained by sequentially combining G} and the situation data H at the current time point to obtain the decision-making basis time series data {B} at the current time point. C D E F G H}.

[0056] Second, the decision-making data is vectorized and high-order features are extracted before being input into a pre-trained public opinion response model. This allows the model to generate structured response measures and expected outcomes for the current time point based on the context vectors and through a multi-task decoding mechanism. The response measures can be one or multiple.

[0057] In one possible implementation, the large-scale public opinion response model outputs response measures as a sequence of high-order features through a pre-constructed first Softmax function. The formula for this first Softmax function is as follows: In the formula, For the corresponding countermeasures of high-order feature sequences, This is a classification weight matrix for various response measures. , Z represents the number of measure categories (e.g., "comment guidance", "offline handling", etc.), and Z is the high-order feature sequence output by the Transformer Encoder. , where n is the length of the feature sequence Z. For the feature dimensions of a large model.

[0058] The first Softmax function described above maps the high-order feature sequence output by the Transformer Encoder to a probability distribution of various response measures, representing the recommendation probability of each response measure. For example, the model might output a probability of 0.7 for "comment guidance" and 0.3 for "offline handling," ultimately recommending the TopK measures with the highest probabilities.

[0059] In one possible implementation, the method for generating the response parameters is as follows: The response parameters (platform, link, content) in this example are a text generated through autoregressive decoding. The function projects the decoder output as a probability distribution, representing the probability of word selection in the current step. After processing through n Transformer layers, the model utilizes the final hidden state. To predict the next word The probability distribution.

[0060] The probability distribution function designed in this example combines the final hidden state with another weight matrix. The probabilities are multiplied and then passed through the Softmax function to obtain the probability distribution. Finally, the word with the highest probability is obtained as the prediction result for the current word using the argMax function.

[0061] Here This represents the i-th word of the response parameter given the context. These are the hidden states of the model. It is a weight matrix.

[0062] In one possible implementation, the method for generating the expected response effect prediction is as follows: By using joint probability modeling, we can predict the measures to be taken. The model can predict subsequent changes in public opinion. For example, it might predict that "comment guidance" will lead to a decrease in popularity. Emotional enhancement .

[0063] Input: High-order features output by the encoder and recommended response measures .

[0064] Output: and The features are concatenated or weighted and fused; the mean prediction result is obtained through a multilayer perceptron, where... It is a higher-order feature of the current node. It is the embedding vector of the classification result. It is the predicted average response.

[0065] In one possible implementation, each response can include various levels of severity, such as aggressive, moderate, and conservative, which the user can select based on their specific needs. In this implementation, the user-selected response is automatically executed as the final response.

[0066] To improve the accuracy of the response measures output by the public opinion response model, new training data will be constructed periodically based on the data of new public opinion events that have recently ended, and the public opinion response model will be continuously optimized and trained based on the new training data.

[0067] Based on supervised pre-training, reinforcement learning is used for fine-tuning to ensure that the generated actions (i.e., responses) maximize the expected cumulative response effect under a given public opinion event state. This example uses a PPO (Proximal Policy Optimization)-based algorithm to optimize the model, maximizing the long-term reward of the generated responses. The model first undergoes random trial and error (exploration), recording which measures score high, gradually reducing low-scoring actions and increasing high-scoring actions, and awarding rewards based on the degree of matching with the true answer, thus learning to select the optimal measures in complex public opinion environments.

[0068] Through reinforcement learning, the fine-tuning of the large model transforms abstract human preferences into the driving force for updating model parameters through a closed loop of "human feedback → reward model → policy optimization". It can automatically select the optimal measures (with the best effect and lowest cost) based on real-time public opinion status.

[0069] S1400, based on the response measures at the current time point, automatically respond to the aforementioned public opinion event. Specifically, this may include the following steps: First, based on the response measures at the current time point, an action task list is generated. Specifically, the response measures at the current time point are parsed to obtain the response measure types and parameters; action task templates matching the types are obtained, and the response measure parameters are filled into the action task templates to obtain the action task list of the response measures.

[0070] It should be noted that the system pre-builds a matching action task template for each type of response measure and establishes a relationship between the type and the corresponding action task template. In this way, after parsing out the type from the response measures at the current time node, the action task template matching that type can be obtained according to the pre-built relationship. Then, the parsed response measure parameters can be filled into the corresponding positions of the action task template to obtain the action task list of the response measures.

[0071] For example, if the current response is to post an online comment, the action task list generated based on the response parameters is as follows: { "Publication Platform": "XX News", "XX Weibo"; Original post title: "Clarification Statement Regarding Incident X"; "Review content": "XXXXXXXXX"; "Execution Priority": "P0"; "Expected Results": "Positive sentiment increased by 20%, and popularity decreased by 35% within 3 hours"; } Second, based on the action task list, the response to public opinion events is automated. Specifically, a specific execution tool is configured for each type of response measure. Thus, after the action task list is generated, the corresponding execution tool can be automatically retrieved, and each action in the action task list can be executed through the execution tool, thereby automatically responding to public opinion events.

[0072] When response measures include issuing press releases, posting online comments and replies, generating public opinion response command and coordination instructions, and generating public opinion reports: For issuing press releases, configure an automatic press release publishing tool; for posting online comments and replies, configure an automatic online comment posting tool; for generating public opinion reports, configure an automatic public opinion report generation tool; and for generating public opinion response command and coordination instructions, configure a command and coordination tool for internet management departments. This way, the various action task lists can be automatically executed by calling the corresponding execution tools.

[0073] Furthermore, the public opinion response model can not only predict the response measures at the current time point, but also predict the expected effects of those measures. After the response measures are automatically executed, the effectiveness of the measures can be evaluated to obtain the actual results. By comparing the actual results with the expected results, the effectiveness of the response measures can be assessed. Based on the effectiveness of the response measures, the model parameters of the public opinion response model can be continuously optimized to improve the effectiveness of the response measures generated by the model.

[0074] To clearly illustrate the method of this disclosure, the following is combined with... Figure 2 The example further illustrates the method disclosed herein. Specifically, the automatic assessment and response method for this public opinion event includes the following steps: Step 1: Public opinion data acquisition: The intelligent agent continuously acquires public opinion information from the Internet using the public opinion data acquisition module and inputs it into the public opinion situation perception module in the system.

[0075] Step 2: Public Opinion Situation Awareness: The intelligent agent public opinion situation awareness module analyzes, processes, and assesses the network public opinion data acquired from the input system.

[0076] Step 3: Planning and Decision-Making: Based on the public opinion situation perception and assessment results, the intelligent agent planning and decision-making module uses the reasoning and planning capabilities of the public opinion response model to determine whether public opinion response measures are needed. If public opinion response is needed, the planning and decision-making module calls the public opinion response model to generate a network public opinion response decision plan and automatically decomposes the decision plan to form a specific action task list.

[0077] Step 4: Response Actions: Based on the public opinion response decision-making plan and action task list, the agent's response action module proactively discovers and calls upon appropriate external tools and corresponding resources to complete the response action tasks.

[0078] Returning to step 1: continuously acquire public opinion data from the Internet, perceive changes in public opinion trends, and evaluate the effectiveness of response actions, continuously optimize planning decisions and response actions, and ultimately achieve the goal of public opinion response.

[0079] This example demonstrates that the entire public opinion response process was handled by... Figure 2 The implementation of large-scale intelligent models means that all the above steps can be achieved through a single large-scale intelligent model, which can automatically complete the entire analysis and response to public opinion events without human intervention, thereby greatly improving the accuracy and response speed of public opinion analysis and response.

[0080] This disclosure provides an automated method for assessing and responding to online public opinion events based on a large-scale intelligent model. The method includes: acquiring public opinion data of the event at the current time point; performing situational awareness analysis on the public opinion data at the current time point to obtain situational data; generating response measures for the current time point using a pre-trained large-scale public opinion response model based on the situational data, the basic information of the public opinion event, and element information from multiple historical key time points; and automatically responding to the public opinion event based on the response measures. The large-scale public opinion model in this disclosure can automatically generate response measures for public opinion events and automatically respond to the events based on these measures. The entire public opinion response process can be fully automated without human intervention, thanks to the large-scale intelligent model, thereby significantly improving the accuracy and response speed of public opinion response.

[0081] <Device Embodiment> Figure 3 This diagram illustrates a schematic block diagram of an automatic assessment and response device for online public opinion events based on a large-scale intelligent agent, according to an embodiment of the present disclosure. Figure 3 As shown, the device 100 includes: The public opinion data acquisition module 110 is used to acquire public opinion data of public opinion events at the current time point; The public opinion situation awareness module 120 is used to perform situation awareness analysis on the public opinion data at the current time point to obtain the situation data at the current time point. The decision planning module 130 is used to generate response measures for the current time point based on the current situation data, basic information of public opinion events, and element information of multiple historical key time points, using a pre-trained public opinion response model. The action response module 140 automatically responds to public opinion events based on the response measures at the current time point.

[0082] <Equipment Example> Figure 4 This diagram illustrates an automated network public opinion event analysis and response device based on a large-scale intelligent agent, according to an embodiment of the present disclosure. Figure 4 As shown, the automatic assessment and response device 200 for public opinion events includes a processor 210 and a memory 220 for storing executable instructions of the processor 210. The processor 210 is configured to implement any of the aforementioned automatic assessment and response methods for public opinion events when executing the executable instructions.

[0083] It should be noted here that the number of processors 210 can be one or more. Furthermore, the automatic analysis and response device 200 for public opinion events in this embodiment may also include an input device 230 and an output device 240. The processors 210, memory 220, input device 230, and output device 240 can be connected via a bus or other means, which are not specifically limited here.

[0084] The memory 220, as a computer-readable storage medium, can be used to store software programs, computer-executable programs, and various modules, such as the program or module corresponding to the automatic assessment and response method for public opinion events in this embodiment of the present disclosure. The processor 210 executes various functional applications and data processing of the automatic assessment and response device 200 for public opinion events by running the software programs or modules stored in the memory 220.

[0085] Input device 230 can be used to receive input digital numbers or signals. These signals may include key signals related to user settings and function control of the device / terminal / server. Output device 240 may include a display device such as a screen.

[0086] <Storage Medium Examples> According to a fourth aspect of this disclosure, a non-volatile computer-readable storage medium is also provided, on which computer program instructions are stored, which, when executed by processor 210, implement the automatic judgment and response method for any of the aforementioned public opinion events.

[0087] The various embodiments of this disclosure have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical applications, or technological improvements to the embodiments in the market, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A method for automatically assessing and responding to online public opinion events based on a large-scale intelligent agent model, characterized in that, include: Obtain public opinion data for the current time point of the public opinion event; Specifically, three data acquisition parameters—data source, data collection type, and update frequency—are configured. Based on these parameters, public opinion data for the current time point is automatically acquired. These data acquisition parameters are adaptively adjusted according to the analysis results of the public opinion response model. When the analysis results indicate that a public opinion event requires focused attention or response, the following actions are taken: expanding the data source, expanding the data collection type, and increasing the update frequency of the public opinion data. When the analysis results indicate that the public opinion event no longer requires focused attention or that the response action has ended, the following actions are taken: restoring the data source, data collection type, and update frequency of the public opinion data to their system default values. Situational awareness analysis is performed on the public opinion data at the current time node to obtain the situational data at the current time node. The situational data at the current time node includes basic situational information such as the causes of public opinion changes, content evolution, changes in popularity, changes in sentiment, dissemination paths, and user group characteristics. The index values ​​of situational indicators such as dissemination degree, popularity, and sensitivity are also included. The index value of the comprehensive situational indicator is obtained by weighted summation of the index values ​​of the situational indicators of dissemination degree, popularity, and sensitivity. The development trends of each situational indicator and the comprehensive situational indicator are also considered. Based on the current time point's situational data, the basic information of the public opinion event, and the element information of multiple historical key time points, a pre-trained public opinion response model is used to generate response measures and expected responses for the current time point. The basic information of the public opinion event and the element information of each historical key time point are read from a short-term memory module. The element information of each historical key time point includes its situational data, response measures, and response results. The historical key node sequence data, composed of the basic information of the public opinion event and the element information of each historical key time point, is sequentially combined with the current time point's situational data to construct a decision-making basis time series data. This decision-making basis time series data is vectorized to obtain a corresponding vector representation sequence. The vector representation of each sequence position in the vector representation sequence is concatenated with the corresponding position information to obtain a high-order vector representation of each sequence position. All high-order vector representations of sequence positions are sequentially combined to obtain a high-order feature sequence of the decision-making basis, which is then input into the public opinion response model to generate the response measures and expected responses for the current time point. The system analyzes the response measures at the current time point to obtain the types and parameters of the response measures, retrieves the action task templates that match the types of response measures, fills the action task templates with the response measure parameters, and obtains the action task list of the response measures. It automatically calls the execution tools corresponding to the types of response measures to execute each action in the action task list, thus automatically responding to public opinion events.

2. The method according to claim 1, characterized in that, The aforementioned public opinion response model is trained based on basic information of various historical public opinion events and element information of key time nodes.

3. The method according to claim 1, characterized in that, After performing situational awareness analysis on the public opinion data at the current time point to obtain the situational data at the current time point, the process also includes: Based on the situational data at the current time point, determine whether it is necessary to respond to the public opinion event at the current time point; When it is determined that a response is needed to the public opinion event at the current time point, the operation is then performed based on the situational data at the current time point, the basic information of the public opinion event, and the element information of historical key time points, using a pre-trained public opinion response model to generate response measures for the current time point.

4. An automatic judgment and response device for online public opinion events based on a large-scale intelligent agent model, characterized in that, include: The public opinion data acquisition module is used to acquire public opinion data of public opinion events at the current time point; Specifically, three data acquisition parameters—data source, data collection type, and update frequency—are configured. Based on these parameters, public opinion data for the current time point is automatically acquired. These data acquisition parameters are adaptively adjusted according to the analysis results of the public opinion response model. When the analysis results indicate that a public opinion event requires focused attention or response, the following actions are taken: expanding the data source, expanding the data collection type, and increasing the update frequency of the public opinion data. When the analysis results indicate that the public opinion event no longer requires focused attention or that the response action has ended, the following actions are taken: restoring the data source, data collection type, and update frequency of the public opinion data to their system default values. The public opinion situation awareness module is used to perform situation awareness analysis on the public opinion data at the current time node to obtain the situation data at the current time node. The situation data at the current time node includes basic situation information such as the causes of public opinion changes, content evolution, changes in popularity, changes in sentiment, dissemination paths, and user group characteristics. It also includes the index values ​​of situation indicators such as dissemination degree, popularity, and sensitivity, and the index value of the comprehensive situation indicator obtained by weighted summation of the index values ​​of dissemination degree, popularity, and sensitivity, as well as the development trends of each situation indicator and the comprehensive situation indicator. The decision-making and planning module, based on the current time-node situation data, the basic information of the public opinion event, and the element information of multiple historical key time nodes, uses a pre-trained public opinion response model to generate response measures and expected responses for the current time-node. The basic information of the public opinion event and the element information of each historical key time node are read from the short-term memory module. The element information of each historical key time node includes its situation data, response measures, and response effects. The historical key node sequence data, composed of the basic information of the public opinion event and the element information of each historical key time node, is sequentially combined with the current time-node situation data to construct a decision-based time-series data. This decision-based time-series data is vectorized to obtain a corresponding vector representation sequence. The vector representation of each sequence position in the vector representation sequence is concatenated with the corresponding position information to obtain a high-order vector representation of each sequence position. All high-order vector representations of sequence positions are sequentially combined to obtain a high-order feature sequence of the decision-based data, which is then input into the public opinion response model to generate the response measures and expected responses for the current time-node. The action response module analyzes the response measures at the current time point to obtain the types and parameters of the response measures, obtains the action task templates that match the types of response measures, fills the action task templates with the response measure parameters, and obtains the action task list of the response measures; it automatically calls the execution tools corresponding to the types of response measures to execute each action in the action task list, and automatically realizes the response to public opinion events.

Images