Management system for comprehensive treatment of smart community

Abstract

The invention relates to the technical field of community management, and discloses a management system for comprehensive management of a smart community, comprising: a multi-modal data feature acquisition and fusion module; a topic evolution analysis module; a cognitive model construction module; a guide strategy generation module; and an emotion influence evaluation module. According to the community monitoring information management system and method, scientific and precise management of community monitoring information is achieved by fusing multi-modal monitoring information analysis, cognitive intervention and emotion regulation technologies, the scientificity, initiative and precision of monitoring information management are effectively improved, harmonious and stable development of communities is promoted, and technical support is provided for constructing a community environment containing mutual trust.

Description

Technical Field

[0001] The present invention relates to the field of community management technology, and more specifically, to a management system for comprehensive management of smart communities. Background Art

[0002] As the construction of smart cities is further advanced, the problem of monitoring information management in communities, as the basic unit of urban governance, is becoming increasingly prominent. Currently, the main technical problems in community monitoring information management are as follows: Traditional monitoring information monitoring systems mainly rely on text analysis technology and cannot effectively integrate multimodal data such as images and audio, resulting in a one-sided and incomplete understanding of monitoring information. These systems can often only capture surface text features and cannot deeply explore emotional features and evolution trends, limiting the ability to deeply analyze community monitoring information. Existing monitoring information guidance systems mainly adopt passive response strategies, focusing on intervention at the level of information dissemination, and lack the ability to actively guide at the cognitive and emotional levels. This passive management model makes the monitoring information guidance effect limited and cannot effectively cope with the complex and changing monitoring information environment. Existing monitoring information analysis technology does not have a good grasp of the evolution rules of topics, has limited predictive ability, and is difficult to achieve pre-intervention. This leads to community managers often being in a passive situation of making up for the loss, unable to effectively warn and intervene before the monitoring information spreads. Community governance lacks refined guidance strategies for different groups of people, and often adopts a one-size-fits-all approach, which cannot effectively deal with emotional polarization and cognitive bias. This extensive governance approach ignores the individual differences of community residents and reduces the pertinence and effectiveness of monitoring information guidance.

[0003] Therefore, there is an urgent need for a smart community monitoring information governance technology solution that can integrate multimodal information, achieve proactive intervention, accurate prediction, and differentiated guidance, so as to improve the scientificity, foresight, and accuracy of community monitoring information governance. Summary of the invention

[0004] The present invention provides a management system for the comprehensive management of smart communities, which solves technical problems in related technologies such as one-sided monitoring of monitoring information, passive guidance, limited prediction capabilities, and extensive strategies.

[0005] The present invention provides a management system for comprehensive management of smart communities, including: Multimodal data feature collection and fusion module, which is used to process community monitoring information data using a multimodal feature extraction network and generate representation vectors by fusing multimodal information through a cross-modal attention algorithm; Topic evolution analysis module, used to build a topic life cycle evolution model, calculate topic activity indicators and predict its future change trend; The cognitive model building module is used to build a cognitive model of community residents and identify cognitive bias types through the Bayesian belief network; Guidance strategy generation module, used to stratify community populations and generate targeted guidance strategies based on cognitive models and topic predictions; The emotional impact assessment module is used to apply the emotional computing model to evaluate the emotional impact of the guiding content, establish a multi-dimensional emotional computing model to finely characterize the emotional state of community monitoring information, and realize closed-loop management of monitoring information governance.

[0006] Furthermore, in the multimodal data feature acquisition and fusion module, the attention weight calculation formula for cross-modal attention fusion is: ;in, Respectively represent the mode Features, is the cosine similarity function between modalities, Representing modality For modal The attention weight of represents the exponential function; ;in, is the global weight parameter of each mode, which is dynamically adjusted according to different community monitoring information scenarios. is the final fused feature vector.

[0007] Furthermore, in the multimodal data feature collection and fusion module, an improved dynamic topic model is applied to process time series data. The model introduces the time dimension to make the topic distribution parameters change smoothly over time. The calculation formula is: ;in, , express The topic distribution parameters at the moment, is a noise term with a mean of 0 and a variance of The normal distribution with parameters Dynamically adjust according to the speed of topic changes, Represents the identity matrix.

[0008] Furthermore, in the topic evolution analysis module, the calculation formula for calculating and predicting topic activity is: ;in, is the current time point, is the time window, is the time decay weight, is a comprehensive function, is the number of messages, To participate in user characteristics, For emotional polarity; is the topic activity index; time decay weight The calculation formula is: ;in, is the attenuation coefficient, The time when the message was released.

[0009] Furthermore, in the cognitive model building module, the belief update calculation formula of the Bayesian belief network is: ;in, Indicates the state of belief, Indicates the information received. Indicates receiving information The belief probability after In the belief state Receive information The conditional probability of represents the prior belief probability, Display information The marginal probability of .

[0010] Furthermore, in the cognitive model building module, a correction strategy mapping matrix is ​​established for the four main cognitive bias types: confirmation bias, framing effect, anchoring effect and availability bias. : ;in, is the cognitive bias feature vector, is the generated correction strategy vector.

[0011] Furthermore, in the guidance strategy generation module, the multidimensional emotion calculation model controls the three-dimensional emotion space based on valence awakening, and its calculation formula is: ;in, Display content emotional state, represents the valence dimension, represents the awakening dimension, Represents the control dimension; the emotion influences the prediction model, and its calculation formula is: ;in, Display content For users emotional impact, is the user feature weight matrix, For users The feature vector of .

[0012] Furthermore, it also includes intervention effect evaluation and feedback adjustment, and evaluates the guidance effect by calculating the changes in the monitoring information status before and after the intervention: ;in, represents the intervention effect evaluation value, and Respectively represent the The status of the indicator, Indicates The weight of the indicator, Indicates the number of evaluation metrics.

[0013] Furthermore, in the emotion impact assessment module, emotion enhancement adopts the following formula: ;in, is the emotion feature extraction function, is the emotion enhancement coefficient, is the final feature vector after emotion enhancement, is the fused feature vector.

[0014] The present invention provides a storage medium storing non-temporary computer-readable instructions. When the non-temporary computer-readable instructions are executed by a computer, the modules in the above-mentioned management system for comprehensive management of smart communities can be executed.

[0015] The beneficial effects of the present invention are as follows: the present invention improves the accuracy of emotion recognition by 23% through multimodal monitoring information feature fusion technology, and combines the dynamic modeling of the topic life cycle to make the monitoring information prediction accuracy reach 76% (an increase of 31%), leading to pre-intervention 12 hours earlier; the cognitive model based on the Bayesian network improves the effectiveness of the intervention plan by 65%, and the success rate of cognitive bias correction increases from 37% to 68%, realizing the transformation from one-size-fits-all to segmented and precise governance, and forming a closed-loop governance mechanism of prediction-guidance-shaping. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a module diagram of the management system for comprehensive management of smart communities of the present invention; Figure 2 is a flowchart of the steps in the multimodal data feature acquisition and fusion module of the present invention; Figure 3 is a flowchart of steps in the topic evolution analysis module of the present invention; Figure 4 is a flowchart of the steps in the cognitive model building module of the present invention; Figure 5 is a flowchart of steps in the guidance strategy generation module of the present invention; Figure 6 It is a flow chart of the steps in the emotion impact assessment module of the present invention. DETAILED DESCRIPTION

[0017] The subject matter described herein will now be discussed with reference to example implementations. It should be understood that the discussion of these implementations is only to enable those skilled in the art to better understand and implement the subject matter described herein, and the functions and arrangements of the elements discussed may be changed without departing from the scope of protection of the present specification. Various examples may omit, replace, or add various processes or components as needed. In addition, the features described in some examples may also be combined in other examples.

[0018] At least one embodiment of the present invention discloses a management system for comprehensive management of smart communities, such as Figure 1-Figure 6 As shown, including: Multimodal data feature collection and fusion module, which is used to process community monitoring information data using a multimodal feature extraction network and generate representation vectors by fusing multimodal information through a cross-modal attention algorithm; Specifically include: Step 1.1, multimodal data acquisition and preprocessing; Collect multimodal data such as text, images, and audio in the community, and perform standardized preprocessing on different modal data. Text data is converted into standard text sequences through operations such as word segmentation and stop word removal; image data is resized and normalized; and audio data is converted into spectral features. The preprocessed multimodal data is organized into structured inputs to facilitate subsequent feature extraction.

[0019] Step 1.2, modality-specific feature extraction; For each modal data after preprocessing, a dedicated feature extraction model is applied: Text data: Apply the improved pre-trained language model to extract semantic features, and the output dimension is The eigenvector of ; Image data: Apply convolutional neural network to extract visual features, the output dimension is The eigenvector of ; Audio data: Apply a one-dimensional convolutional network to extract audio features, and the output dimension is The eigenvector of ; The feature extraction process of each modality can be expressed as: ; ; ; in, , , They are feature extraction functions for text, image, and audio respectively; , , are the input data of the corresponding modes respectively.

[0020] Step 1.3, cross-modal attention fusion; The improved cross-modal attention algorithm is applied to perform weighted fusion of features of each modality to capture complementary information between modalities. The algorithm is implemented in the following way: first, the similarity matrix between features of different modalities is calculated, then the attention weight is calculated based on the similarity, and finally the feature fusion is performed according to the weight.

[0021] Specifically, in the scenario where community residents complain about property services, when residents express the content of the complaint through text descriptions, uploaded photos, and voice messages at the same time, the algorithm can automatically identify the emotional consistency and complementary information in the three modalities. For example, the text content may be calm in tone, but the emotions revealed in the voice may be more intense, and the photos provide objective evidence. Through cross-modal attention fusion, the system gives the voice modality a higher emotional weight, while retaining the semantic information of the text and the evidential information of the image, so as to accurately grasp the urgency and emotional tendency of the complaint.

[0022] The cross-modal attention weight is calculated as follows: ; in, , Represents the characteristics of different modes, is the cosine similarity function between modalities, Representing modality For modal The attention weight.

[0023] The fusion feature calculation formula is: ; in, is the global weight parameter of each mode, which can be dynamically adjusted according to different community monitoring information scenarios. For the final fusion feature vector, community monitoring information mainly includes: community residents' views and suggestions on community management, services, environment, etc.; conflicts and disputes among community residents; community environment and ecological problems, etc.

[0024] Step 1.4, emotional feature enhancement; Apply the emotion enhancement module to the fused features to highlight emotion-related features. The emotion enhancement uses the following formula: ; in, is the emotion feature extraction function, is the emotion enhancement coefficient, is the final feature vector after sentiment enhancement.

[0025] Topic evolution analysis module, used to build a topic life cycle evolution model, calculate topic activity indicators and predict its future change trend; Specifically include: Step 2.1, topic identification and clustering; Based on multimodal fusion features, an improved hierarchical clustering algorithm is used to identify community hot topics. The specific process is as follows: Calculate the similarity matrix between feature vectors ,in Indicates Article and Similarity of information: ; in, , Respectively Article and The feature vector of the information, is the cosine function of topic similarity.

[0026] Apply a hierarchical clustering algorithm with dynamic thresholds to form topic clusters ,in Indicates Topic clusters, is the number of topic clusters; Indicates Topic clusters, is the number of the topic cluster.

[0027] For each topic cluster Extract core keywords and representative content to generate topic summaries .

[0028] Step 2.2, construction of temporal topic model; For each identified topic, a temporal topic model is constructed to capture the evolution of the topic over time: Topic data is divided into time windows Segment, where Indicates time window, is the number of time windows; each time window contains data of a fixed length. In community information management systems, it is typically set to one time window every 4 hours to balance computational efficiency and the accuracy of capturing evolutionary trends.

[0029] The improved Dynamic Topic Model (DTM) is applied to process time series data. The model introduces the time dimension to make the topic distribution parameters change smoothly over time, thereby capturing the evolution of the topic. In practical applications, the model can effectively handle the evolution of sensitive topics such as community property fee adjustments, for example, from the initial policy notification stage, to the resident question stage, to the opinion differentiation stage, and finally to the consensus formation stage, fully tracking the characteristics of each stage of the topic life cycle. The calculation formula is: ; in, express The topic distribution parameters at the moment, is a noise term with a mean of 0 and a variance of The normal distribution of Represents the identity matrix. Dynamically adjust according to the speed of topic change, set larger values ​​for rapidly changing emergencies (such as community safety accidents), and set smaller values ​​for slowly evolving long-term topics (such as community environmental improvement).

[0030] The variational inference algorithm is used to optimize the model parameters and extract the changes in the topic distribution within each time window. The algorithm gradually approximates the posterior distribution in an iterative manner, including calculating the expectation of the variational parameters and maximizing the objective function. In the analysis of community monitoring information, the number of iterations is usually set to 100 or the convergence threshold is set to 0.001 to ensure that the model converges and meets the real-time requirements.

[0031] Step 2.3, topic activity calculation and prediction; Calculate the topic activity index and predict its future change trend: The formula for calculating topic activity is: ; in, is the current time point, is the time window, is the time decay weight, is a comprehensive function, is the number of messages, To participate in user characteristics, For emotional polarity, It is an indicator of topic activity.

[0032] Time decay weight The calculation formula is: ; in, is the attenuation coefficient, The time when the message was released.

[0033] Based on historical activity data ,in, Indicates Activity indicators, is the number of activity indicators; a long short-term memory network (LSTM) is used to build a prediction model: ; in, Respectively represent Activity indicators, is the input window size of the LSTM model; To predict the activity of the topic at the next moment, are the LSTM model parameters.

[0034] Combined with the predicted topic activity, determine the life cycle stage of the topic (germination stage, outbreak stage, stable stage, decline stage).

[0035] Step 2.4, topic evolution path analysis; Based on the historical evolution data of the topic, analyze its potential evolution path: Constructing a topic semantic evolution graph , where the node Indicates the topic status, edge Indicates the state transition relationship.

[0036] Calculate the state transition probability matrix ,in Indicates from the state Transfer to state The specific calculation method of state transition probability is: ; in, Indicates the state of historical data Transfer to state The number of observations, Represents the total number of states, Indicates from the state Transfer to state The number of observations.

[0037] For sparse data, smoothing is used: ; in, It is a smoothing parameter, usually set to a small positive number between 0.01 and 0.1 to avoid zero probability problems. In actual applications, the system calculates the matrix based on the historical data of topic evolution in the last 30 days and updates it every 24 hours to adapt to the dynamic characteristics of community monitoring information.

[0038] Apply the Markov process to simulate the future evolution path of the topic and output the possible evolution direction and its probability distribution: based on the state transition probability matrix , from the current topic status Start by iterating and calculating the future The state distribution of the step: ; in, is the probability distribution vector of the current state (the current state is 1, and the others are 0), Represents the transition probability matrix Power, Indicates that from the current state Departure, passing Steps after reaching state The probability of is the current time point, is the predicted number of steps.

[0039] For each possible evolution path ,in, Respectively represent status, is the number of states; calculate their probabilities:

[0040] Screening probability is greater than the threshold (usually set to 0.05) are taken as valid prediction results and are output in descending order of probability.

[0041] The system updates the evolution path prediction results every 12 hours and triggers instant updates at key nodes (such as mutation points of topic activity) to ensure the timeliness of the prediction.

[0042] The cognitive model building module is used to build a cognitive model of community residents and identify cognitive bias types through the Bayesian belief network; Specifically include: Step 3.1: User feature analysis and portrait construction; Build user profiles based on historical behavior data of community users: Collect basic user attributes (such as age group, occupation type, etc.) and behavioral data (such as post content, reply frequency, active time periods, etc.).

[0043] Apply unsupervised learning methods to reduce the dimension of user features: ; in, For users The original eigenvector of is the dimension reduction function, is the feature vector after dimensionality reduction.

[0044] Based on the feature vector after dimension reduction, the clustering algorithm is applied to divide users into Groups: ; in, Respectively represent groups, is the number of groups; each group Contains collections of users with similar characteristics.

[0045] Step 3.2, Bayesian belief network construction; For each user group, a Bayesian belief network is constructed to simulate its information reception and belief updating process: Defining the belief state space ,in, Respectively represent A state of belief, is the number of belief states; each Indicates the belief state of a specific event or opinion. In community monitoring information management, belief states can include support, opposition, neutrality, uncertainty, etc., and belief spaces are constructed for different community issues (such as property management, community renovation, neighborhood relations, etc.).

[0046] Defining Information Spaces ,in, Respectively represent Information type, is the number of information types; each Indicates the type of information that may be received. In practical applications, information types can be classified based on dimensions such as source (official channels, community leaders, ordinary residents, etc.), sentiment tendency (positive, negative, neutral), and content attributes (facts, opinions, rumors), such as official positive facts, leaders' negative opinions, etc.

[0047] Construct a conditional probability table to describe the impact of information on beliefs: Indicates that information has been received After that, hold the belief probability.

[0048] In the community property service quality evaluation scenario, the network can be used to analyze the impact of different information sources on residents' evaluation. For example, when residents receive negative feedback from neighbors, the probability of negatively evaluating the property service quality may increase from 30% to 60%; and when they receive explanations from the property company, the probability may drop to 45%. In this way, the system can quantify the impact of different information on residents' cognition.

[0049] The belief updating process follows the Bayesian formula: ; in, is the prior belief probability, obtained through historical data and initial survey; is the likelihood function, which means that Receive information under the conditions probability; To receive information The posterior belief probability.

[0050] The network simulates the cumulative effect of multiple rounds of information influence through iterative updating. In each round, the posterior probability of the previous round is used as the prior probability of the next round, thereby tracking the dynamic evolution of the belief state. In the system implementation, a combination of structural learning and parameter learning is used to automatically build and optimize the network structure and parameters.

[0051] Step 3.3, identification and classification of cognitive biases; Based on the constructed Bayesian belief network, common cognitive bias types are identified and classified: Identify cognitive biases by analyzing abnormal patterns in the user's information reception and belief updating process. The main types of cognitive biases include: Confirmation Bias: The tendency to seek out information that supports one's existing views. Anchoring Effect: Over-reliance on the first information obtained; Availability Bias: Making judgments based on easily recalled information; Framing Effect: Differences in judgment caused by different ways of presenting information; Group Think: Suppressing dissenting opinions to maintain group harmony; Constructing a cognitive bias judgment model , evaluate users On topic Types of cognitive biases: ; in, is a set of cognitive bias types, For user characteristics, For users in the topic Historical behavior, Type of cognitive bias.

[0052] Assess the degree of the identified cognitive biases, using a score between 0 and 1 to indicate the strength of the bias : ; Among them, 0 means no bias, 1 means very strong bias, For the current user, Types of cognitive biases: For the current topic.

[0053] Step 3.4, group cognitive network construction; Integrate individual cognitive models to build group cognitive networks: Constructing a community cognitive influence map ; Among them: Node Indicates community users; Indicates the influence relationship between users; weight Indicates the intensity of impact; Calculate influence weight based on user historical interaction data : ; in, is the interaction frequency, For interactive type, For emotional relevance.

[0054] Apply social network analysis methods to calculate node centrality and influence indicators and identify key opinion leaders and opinion groups.

[0055] Guidance strategy generation module, used to stratify community populations and generate targeted guidance strategies based on cognitive models and topic predictions; Specifically include: Step 4.1, guide target determination and quantification; Based on topic evolution prediction and cognitive model, determine the goal of monitoring information guidance: Define monitoring information state space ,in, Respectively represent Monitoring information status, It is the number of monitoring information states; each state represents a form of monitoring information.

[0056] Defining the target state , indicating the expected monitoring information status.

[0057] Quantify the gap between the goal and the current state : ; in, and Respectively represent the current state and the target state in The values ​​in the dimensions, is the weight coefficient of the corresponding dimension. Indicates the current state. Indicates the target state.

[0058] Step 4.2: Crowd stratification and key node identification; Based on the constructed group cognitive network, the community population is stratified and key influencing nodes are identified: According to the user's position and characteristics in the cognitive network, users are divided into multiple levels: Core opinion leaders (highly influential users); Secondary disseminator layer (moderately influential users); General participant layer (low-impact users); Potential audience (observers but not active participants); For each level, calculate the user influence index : ; in, is the centrality index, is the interactive response rate, is the activity indicator, , , is the weight coefficient.

[0059] Select the most influential person at each level Users are key nodes, forming a set of priority targets for guiding strategies. : ; in, Indicates Layers of people, The number of key nodes selected for this layer, For the current user, Indicates the user influence index Sort by.

[0060] Step 4.3, design of cognitive bias correction strategy; Build corresponding correction strategies for the identified cognitive bias types: Construct a cognitive bias-correction strategy mapping table : ; in, Types of cognitive biases: A collection of correction strategies.

[0061] For Confirmation Bias, adopt the following correction strategy: Provide multi-angle information sources; Introduce counterexample information; Design a balanced argument framework; For the Anchoring Effect, the following correction strategies are adopted: Provide multiple anchor information; Reduce the weight of primary information; Reinforce accurate and up-to-date information; For availability bias, the following correction strategies are adopted: Provide statistical data and factual evidence; Reduce emotional case presentations; Add systematic information summary; Targeting specific user groups and topics , choose the optimal correction strategy : ; in, Representing a group On topic The main types of cognitive biases in Representation strategy Applicable to groups and topics expected effect.

[0062] Step 4.4, multi-level guidance scheme generation; Integrate the above analysis results to generate a multi-level, differentiated guidance plan: Construct a guidance program matrix based on population stratification, key node identification and cognitive bias correction strategies : ; in, Indicates the target population, Indicates related topics, represents the optimal strategy, Represents a collection of content formats. represents the set of communication channels, Represents a schedule collection.

[0063] Differentiate content forms, dissemination channels and time arrangements for different levels of people: Core opinion leaders: provide in-depth analysis materials and prioritize implementation through direct communication channels; Secondary communicator layer: providing simplified content that can be spread, through social media and major community platforms, with the second highest priority; General participant layer: providing concise opinions and facts, through public information channels, and regular priority implementation; Potential audience layer: Provide background information and basic facts, set through environmental information, and implement with low priority.

[0064] Generate a complete guidance plan sequence, including detailed information such as execution steps, responsible units, and effect evaluation indicators.

[0065] The emotional impact assessment module is used to apply the emotional computing model to evaluate the emotional impact of the guidance content, establish a multi-dimensional emotional computing model to finely characterize the emotional state of community monitoring information, and realize the closed-loop management of monitoring information governance; Specifically include: Step 5.1, emotional computing model construction; Establish a multi-dimensional sentiment computing model to accurately describe the sentiment state of community monitoring information: Defining a multidimensional emotional space ,in, Indicates emotional valence, ranging from [-1,1], Indicates the emotional arousal, ranging from [0,1], Indicates the sense of control, ranging from [0,1].

[0066] This multidimensional emotion representation method can distinguish subtle emotional differences, such as anger (negative valence, high arousal, high sense of control) and fear (negative valence, high arousal, low sense of control), which is difficult to achieve in traditional positive-negative binary emotion analysis.

[0067] Constructing the emotional state vector , indicating the emotional characteristics of specific content.

[0068] Training emotional computing model based on multimodal fusion features : ; in, is the emotional state vector, Sentiment-enhanced feature vector.

[0069] The emotion computing model adopts a multi-layer perceptron structure, including three hidden layers, each containing 128, 64 and 32 neurons respectively, using the ReLU activation function, and the last layer using the Tanh or Sigmoid activation function (different activation functions are used for different dimensions). The model is trained through supervised learning and uses annotated multi-dimensional emotion datasets for parameter optimization.

[0070] In the case of handling complaints from community residents, the model can identify complaints that are polite on the surface but actually intense in emotion, helping community managers identify potential high-risk monitoring information. For example, a complaint that objectively describes a problem but shows high arousal and low sense of control in sentiment analysis may need to be handled with priority over a complaint that clearly expresses anger but high sense of control, because the former may represent that the resident may feel helpless and desperate, and is more likely to engage in extreme behavior.

[0071] Step 5.2, content sentiment impact prediction; Predict the emotional impact of different guidance content on the target group: For specific guidance content and target population , build a sentiment impact prediction model: ; in, For the crowd The initial emotional state For population characteristics, To predict the change in emotional state, To develop a prediction model for emotional impact.

[0072] Based on the prediction results, evaluate the emotional guidance effect of the content: ; in, is the target emotional state, is the similarity function, For content For the crowd The emotional guidance effect score, For the crowd The initial emotional state To predict changes in emotional state.

[0073] Optimize content combination to maximize emotional guidance effect : ; in, is a candidate content combination, is the weight of the population.

[0074] Step 5.3, optimize the information presentation method; Based on the principles of cognitive science, optimize the way and rhythm of information presentation: Construct the optimal presentation order based on the attention characteristics of the crowd : ; in, For population characteristics, is the environmental context factor, Optimizes functions for sequence.

[0075] Constructing information repetition patterns based on the memory curve model : ; in, is the content item, is the first presentation time, is the number of repetitions.

[0076] Determine the information release time matrix based on the active time distribution of the crowd : ; in, For the target group The optimal release time, Gather for the crowd.

[0077] Step 5.4, intervention effect evaluation and feedback adjustment; Establish an intervention effect evaluation system to achieve closed-loop management: Setting up a multi-dimensional evaluation index system ,in, Respectively represent evaluation indicators, is the number of evaluation indicators; including: Emotional change index; topic evolution trajectory index; participation index; cognitive bias correction index.

[0078] Apply Bayesian network to predict the propagation path of monitoring information after intervention: in, Indicates the monitoring information status at time t, Indicates intervention action, Indicates intervention Under the condition of Transfer to state The probability of For intervention time.

[0079] Dynamically adjust guidance strategies based on real-time feedback data: ; in, is the current boot strategy, and are the monitoring information status of the current and next moment respectively, is the difference between the next state and the target state, To adjust the function, This is the adjusted guidance strategy.

[0080] This implementation method integrates multimodal monitoring information analysis, cognitive intervention and emotion regulation technology. Through multimodal feature fusion, the accuracy of emotion recognition is improved by 23%. Topic life cycle modeling increases the accuracy of monitoring information prediction to 76% (an increase of 31%) and achieves 12-hour pre-intervention. The cognitive model based on Bayesian network increases the intervention efficiency by 65%, and the success rate of cognitive bias correction increases from 37% to 68%, achieving segmented and precise governance. This solution promotes the governance of monitoring information from discovery-analysis-response to a closed-loop model of prediction-guidance-shaping, significantly improving the accuracy and initiative of governance, and promoting the harmonious and stable development of the community while enhancing residents' participation and satisfaction.

[0081] In one embodiment of the present invention, an example of the aforementioned management system for comprehensive management of smart communities is provided: Application scenario description: A large community (about 5,000 households) plans to implement a property service fee adjustment plan, from the original 2.5 yuan / square meter / month to 3.2 yuan / square meter / month, while optimizing some service contents. The property company and the community management committee are worried that the adjustment may cause strong opposition from residents, and hope to guide the formation of a rational and harmonious discussion atmosphere and reach the greatest degree of consensus through scientific monitoring information management methods.

[0082] The main challenges facing management include: The age structure, occupational background, and economic status of residents in the community vary greatly, and their acceptance of property fee adjustments varies; Some residents are dissatisfied with the existing services provided by the property and may take this opportunity to express their dissatisfaction; Similar incidents in the past have seen the spread of rumors and emotional polarization, leading to a tense atmosphere in the community; It is necessary to reach a basic consensus before the plan is formally implemented to avoid subsequent intensification of conflicts.

[0083] The application of this method aims to achieve scientific and proactive management of community monitoring information through precise monitoring information analysis, topic evolution prediction and hierarchical guidance strategy.

[0084] Multimodal monitoring information data collection and feature extraction: First, starting two weeks before the property fee adjustment plan is announced, the system collects monitoring information related data from multiple channels within the community: Data sources include: community forum text posts, WeChat group chat records, community public account messages, voice / video feedback submitted by residents, offline symposium records, etc. Data preprocessing: standardize all types of data, including Chinese word segmentation, stop word removal, image size standardization, etc. Feature extraction: Apply a multimodal feature extraction network to generate a fused representation vector.

[0085] The collection of different modal data is shown in Table 1: Table 1: Community property fee adjustment monitoring information data collection statistics: After fusing the features of each modality through the cross-modal attention algorithm, the system identified the main emotional features related to the adjustment of property fees, which were mainly concentrated in worry (negative valence, medium arousal, low sense of control) and questioning (negative valence, medium arousal, medium sense of control).

[0086] Topic evolution trend analysis and prediction: By applying the time series topic model to the collected multimodal data, the system identified the following key sub-topics, as shown in Table 2: Table 2: Sub-topics and initial sentiment distribution related to property fee adjustment: The system uses dynamic modeling of the topic life cycle to predict the activity trend of each sub-topic in the next two weeks: The topic of cost increase is predicted to reach its peak 2-3 days after the announcement of the plan. Without intervention, the proportion of negative emotions may increase to more than 85%; The topic of service quality improvement plan has the potential to guide monitoring information in a positive direction and is expected to serve as a key guidance direction; The emotional tone of the topic of assistance policy is relatively positive, which can be used as a handle to buffer negative emotions; The topic of fee transparency and supervision mechanism has the potential to evolve into a secondary monitoring information hotspot, and requires early intervention.

[0087] Resident group cognitive model and stratification strategy: Based on historical behavior data and initial monitoring information response, the system divides community residents into the following groups, as shown in Table 3: Table 3: Cognitive group division of community residents: Based on the Bayesian belief network, the system simulates the belief updating process of different groups after receiving different information, and predicts that without guidance, the proportion of residents supporting the adjustment of property fees will drop from the initial 23% to 15%, while the proportion of strong opposition will increase from 18% to 42%.

[0088] Based on this, the system generates a hierarchical precision guidance strategy: For opinion leaders: provide complete and detailed pricing basis and service improvement plan, arrange special communication meetings, and invite them to participate in the design of supervision mechanism; For active communicators: design easy-to-spread graphic content, highlight specific scenarios and benefits of service improvement, and publish authoritative rumor-refuting information in a timely manner; For the concerned groups: provide concise information on where the funds will be used and the promise of service improvement, emphasizing the consensus opinions of the majority of residents; For the indifferent group: Deliver positive information through environmental information settings (such as elevator advertisements, bulletin boards, etc.) to reduce the probability of exposure to negative information.

[0089] Implementation effect and adjustment optimization: One week after the implementation of the guidance strategy, the system continued to monitor changes in monitoring information and found the following effects: The proportion of residents who support the property fee adjustment has increased to 42%, while the proportion who strongly oppose it has dropped to 19%; The sentiment polarity of related topics has shifted to a positive direction overall, especially the positive sentiment ratio of the service quality improvement plan topic has increased to 68%; The speed at which rumors spread has dropped significantly, from an average of 500 people in 3 hours to an average of 300 people in 8 hours.

[0090] Based on real-time feedback data, the system dynamically adjusts the guidance strategy: To target residents who still hold opposing views, we increased publicity efforts on special policies for households with financial difficulties; Regarding the topic of fee transparency and supervision mechanism, a specific plan for residents’ representatives to participate in supervision was added; The guidance content for active communicator groups has been optimized, with more emphasis on short-term visible service improvements.

[0091] Technical effect verification: To verify the actual effect of this method, the project team conducted a systematic evaluation of the entire process of monitoring information management for property fee adjustment in this community, and compared it with three other communities of similar size in the same area that did not adopt this method. The results proved that this method has significant advantages in two key indicators: accuracy of monitoring information prediction and effectiveness of intervention.

[0092] Verification of the accuracy of monitoring information prediction: The system predicts the development trend of monitoring information related to property fee adjustment through dynamic modeling of the topic life cycle, and compares it with the actual evolution. The results are shown in Table 4: Table 4: Verification results of monitoring information prediction accuracy: As can be seen from Table 4, the accuracy of this method in predicting the trend of monitoring information reached 91.9%, which is much higher than the 57.1% of the traditional method. It warned potential risk points of monitoring information in advance and provided strong support for precise intervention. In particular, in predicting the change of emotional polarity, the accuracy rate was as high as 98.1%, which reflects the effectiveness of the multimodal emotional computing model.

[0093] Verification of intervention effectiveness: This method actively intervened in community monitoring information through a hierarchical precision guidance strategy. The differences before and after the intervention and with the control group are shown in Table 5: Table 5: Results of verification of the effectiveness of monitoring information intervention: Table 5 shows that through the application of this method, the community has achieved a number of significant improvements in the process of property fee adjustment: The support rate increased by 82.6%, while the control group only increased by 10.3%; The proportion of those who strongly disagreed increased by only 5.6%, while the control group increased by 133.7%; The proportion of residents actively participating in discussions increased by 153.3%, indicating that the guidance strategy has stimulated the enthusiasm of residents to actively participate in community affairs; The most noteworthy thing is that the monitoring information polarization index decreased by 50.0%, while the control group increased by 15.7%, which fully demonstrated the unique advantages of this method in alleviating community emotional polarization.

[0094] Through the comparison of the above quantitative indicators, it can be clearly verified that this method has significant monitoring information prediction accuracy and intervention effectiveness, and has successfully achieved the paradigm shift in community monitoring information governance from discovery-analysis-response to prediction-guidance-shaping.

[0095] The successful application of this implementation method in the management of community property fee adjustment monitoring information not only effectively promotes the smooth implementation of the property fee adjustment plan, but also provides a replicable and popularizable technical path for the comprehensive management of smart communities. The value of this application is mainly reflected in: It provides a complete set of community monitoring information prediction-analysis-guidance method system, which can achieve monitoring information intervention 12 hours in advance compared with the traditional passive response mode; Through the layered and precise guidance strategy, the participation and recognition of community residents in public affairs have been significantly improved, and the monitoring information polarization index has been reduced by an average of 50%; A closed-loop feedback mechanism for community monitoring information governance was established, which enabled continuous optimization of intervention strategies and increased the effectiveness of intervention plans by 65%; It provides scientific decision-making basis based on data and models for community management, reducing the uncertainty of human experience judgment; To sum up, this implementation method has important practical value and promotion significance in the field of smart community monitoring information governance, and can effectively promote the harmonious and stable development of the community.

[0096] The above describes an embodiment of the present invention, but this embodiment is not limited to the above-mentioned specific implementation mode. The above-mentioned specific implementation mode is merely illustrative and not restrictive. Under the guidance of this embodiment, ordinary technicians in this field can also make more forms of equivalent embodiments, all of which are within the protection of this embodiment.

Claims

1. A management system for comprehensive management of smart communities, characterized by: include: Multimodal data feature collection and fusion module, which is used to process community monitoring information data using a multimodal feature extraction network and generate representation vectors by fusing multimodal information through a cross-modal attention algorithm; Topic evolution analysis module, used to build a topic life cycle evolution model, calculate topic activity indicators and predict its future change trend; The cognitive model building module is used to build a cognitive model of community residents and identify cognitive bias types through the Bayesian belief network; Guidance strategy generation module, used to stratify community populations and generate targeted guidance strategies based on cognitive models and topic predictions; The emotional impact assessment module is used to apply the emotional computing model to evaluate the emotional impact of the guiding content, establish a multi-dimensional emotional computing model to finely characterize the emotional state of community monitoring information, and realize closed-loop management of monitoring information governance.

2. The management system for comprehensive management of smart communities according to claim 1 is characterized in that: In the multimodal data feature acquisition and fusion module, the attention weight calculation formula for cross-modal attention fusion is: ； in, Respectively represent the mode Features, is the cosine similarity function between modalities, Representing modality For modal The attention weight of represents the exponential function; ； in, is the global weight parameter of each mode, which is dynamically adjusted according to different community monitoring information scenarios. is the final fused feature vector.

3. The management system for comprehensive management of smart communities according to claim 1 is characterized in that: In the multimodal data feature collection and fusion module, an improved dynamic topic model is used to process time series data. The model introduces the time dimension to make the topic distribution parameters change smoothly over time. The calculation formula is: ； in, , express The topic distribution parameters at the moment, is a noise term with a mean of 0 and a variance of The normal distribution with parameters Dynamically adjust according to the speed of topic changes, Represents the identity matrix.

4. The management system for comprehensive management of smart communities according to claim 1 is characterized in that: In the topic evolution analysis module, the calculation formula for calculating and predicting topic activity is: ； in, is the current time point, is the time window, is the time decay weight, is a comprehensive function, is the number of messages, To participate in user characteristics, For emotional polarity; is the topic activity index; time decay weight The calculation formula is: ； in, is the attenuation coefficient, The time when the message was released.

5. The management system for comprehensive management of smart communities according to claim 1 is characterized in that: In the cognitive model building module, the belief update calculation formula of the Bayesian belief network is: ； in, Indicates the state of belief, Indicates the information received. Indicates receiving information The belief probability after In the belief state Receive information The conditional probability of represents the prior belief probability, Display information The marginal probability of .

6. The management system for comprehensive management of smart communities according to claim 1 is characterized in that: In the cognitive model building module, a correction strategy mapping matrix is ​​established for the four main cognitive bias types: confirmation bias, framing effect, anchoring effect and availability bias. : ； in, is the cognitive bias feature vector, is the generated correction strategy vector.

7. The management system for comprehensive management of smart communities according to claim 1 is characterized in that: In the guidance strategy generation module, the multidimensional emotion calculation model controls the three-dimensional emotion space based on valence awakening, and its calculation formula is: ； in, Display content emotional state, represents the valence dimension, represents the awakening dimension, It represents the dimension of control; The emotional impact prediction model is calculated as follows: ； in, Display content For users emotional impact, is the user feature weight matrix, For users The feature vector of .

8. The management system for comprehensive management of smart communities according to claim 1 is characterized in that: It also includes intervention effect evaluation and feedback adjustment, and evaluates the guidance effect by calculating the changes in monitoring information status before and after the intervention: ； in, represents the intervention effect evaluation value, and Respectively represent the The status of the indicator, Indicates The weight of the indicator, Indicates the number of evaluation metrics.

9. The management system for comprehensive management of smart communities according to claim 1 is characterized in that: In the emotion impact assessment module, emotion enhancement adopts the following formula: ； in, is the emotion feature extraction function, is the emotion enhancement coefficient, is the final feature vector after emotion enhancement, is the fused feature vector.

10. A storage medium storing non-transitory computer-readable instructions, characterized in that: When the non-temporary computer-readable instructions are executed by a computer, they can execute the module in the management system for comprehensive management of smart communities as described in any one of claims 1-9.

Images