Near public opinion patrol perception method based on social media virtual positioning
By using social media virtual positioning and multi-dimensional user profiling technology, the problems of insufficient geographic location correlation, balance between privacy protection and usability, insufficient personalized detection capabilities, and system performance bottlenecks in existing public opinion monitoring have been solved, achieving efficient and accurate public opinion patrol and personalized risk assessment.
Patent Information
- Application Number
- CN202511854177.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-10
- Publication Date
- 2026-04-21
AI Technical Summary
Existing public opinion monitoring technologies suffer from insufficient geographic location correlation, difficulty in balancing privacy protection and usability, insufficient personalized detection capabilities, and bottlenecks in technology integration and system performance, resulting in low detection accuracy, high risk of privacy leakage, poor system adaptability, and insufficient performance.
A social media-based virtual location method is adopted, which generates virtual location points through differential privacy algorithms. Combined with multi-dimensional user profiles and an adaptive geofencing system, hierarchical public opinion content identification and personalized risk assessment are carried out, and detection strategies are dynamically adjusted to improve monitoring accuracy and system performance.
It significantly improves the efficiency and accuracy of public opinion monitoring, reduces the risk of privacy leaks, enhances detection accuracy and system performance, and increases adaptability to meet the needs of large-scale real-time detection.
Smart Images
Figure CN121901513A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of public opinion monitoring, and more specifically, to a method for nearby public opinion patrol and perception based on virtual location on social media. Background Technology
[0002] Existing public opinion monitoring technologies are mainly based on keyword matching, regular expressions, machine learning, and other methods. Their core processes include data collection, rule matching, and result output. Typical existing technologies include the following aspects: Public opinion monitoring system: It identifies sensitive public opinion content, such as negative comments and sensitive topics, through predefined rules and pattern matching. This technology uses regular expressions and keyword databases for matching, and can handle common types of public opinion content, ensuring the comprehensiveness and timeliness of public opinion monitoring.
[0003] Machine learning-based public opinion classification: This method uses algorithms such as Support Vector Machine (SVM) and Naive Bayes to classify text data. By training samples, it learns the feature patterns of public opinion content and achieves automated identification. These algorithms can process large amounts of unstructured public opinion data, improving the accuracy and efficiency of detection.
[0004] Deep learning-based public opinion analysis: Using deep learning models such as recurrent neural networks (RNN) and long short-term memory networks (LSTM), semantic analysis and public opinion content extraction are performed on unstructured text. These models, through training on a large amount of data, can understand the deep semantics of the text and identify complex public opinion patterns.
[0005] Privacy protection in location services: Existing technologies mainly use methods such as location obfuscation and fake location generation to protect users' real location. They protect user privacy by adding noise or generating false location information. These methods can protect users' location privacy to a certain extent, but they usually significantly reduce the availability of location services.
[0006] However, as applications have become more widespread, the following shortcomings of existing technologies have been discovered: 1. Insufficient correlation with geographical location and lack of regional monitoring strategies: Existing public opinion monitoring technologies use uniform detection rules and cannot dynamically adjust monitoring strategies according to the public opinion characteristics of the user's location (such as business districts, educational districts, and residential areas), resulting in low detection accuracy (average false alarm rate of 35%) and poor targeting.
[0007] Fixed monitoring scope: It is impossible to dynamically adjust the size of the monitoring grid and the detection strategy according to factors such as regional population density, history of public opinion events, and geographical complexity, making it difficult to adapt to the different needs of public opinion patrol in different geographical scenarios.
[0008] 2. Balancing privacy protection and usability: Real-time location privacy risks: Relying on users' real GPS coordinates for location-related public opinion monitoring poses a risk of location information leakage. According to statistics, in 2022, privacy incidents caused by location data leakage accounted for 28%, highlighting the increasingly urgent need for user privacy protection.
[0009] Reduced availability of location services: Existing location privacy protection methods often significantly reduce the availability of location services, affecting the effectiveness of location-based public opinion monitoring and perception.
[0010] 3. Insufficient personalized detection capabilities and weak correlation between user profiles and location: Existing user profiling technology only includes basic attributes and behavioral characteristics, and does not deeply integrate geographical location characteristics and movement patterns, resulting in insufficient personalized public opinion monitoring service capabilities, with a matching degree of only 62%.
[0011] The monitoring strategy lacks adaptability: it is unable to perform personalized public opinion content identification and risk assessment based on users' geographic behavior patterns and location preferences.
[0012] 4. Technology integration and system performance: Difficulty in integrating multiple technologies: Existing technologies are relatively independent in areas such as privacy protection, location services, and public opinion monitoring, lacking effective technology integration and collaborative optimization.
[0013] Real-time performance bottleneck: In scenarios with a large number of users and massive amounts of public opinion data, the system faces performance bottlenecks and latency issues, affecting the effectiveness of real-time public opinion monitoring and user experience.
[0014] Therefore, it is necessary to propose a method for nearby public opinion monitoring and perception based on social media virtual positioning to solve the above problems. Summary of the Invention
[0015] To overcome at least one of the defects (deficiencies) of the prior art described above, the present invention provides a method for nearby public opinion patrol and perception based on social media virtual positioning.
[0016] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows: A method for nearby public opinion patrol and perception based on social media virtual positioning, comprising the following steps: Step 1: The virtual positioning generation unit perturbs the user's real GPS coordinates to generate corresponding virtual positioning points; Step 2: Multi-dimensional user profile construction unit constructs a multi-dimensional user profile that integrates geographic features, deeply fusing geographic location information with user behavior characteristics; Step 3: The intelligent geofencing establishment unit establishes an adaptive multi-level geofencing system, constructing layered detection areas centered on the location of the virtual positioning point; Step 4: The public opinion content identification unit adopts a location-aware intelligent public opinion identification method, combined with a multi-level geofencing system to perform layered detection; Step 5: The personalized public opinion early warning and risk assessment unit receives the features of the multi-dimensional user profile, information of the virtual location point, historical behavior records and current time context, generates personalized public opinion early warning results and risk assessment scores, and monitors and perceives public opinion near the virtual location point.
[0017] Furthermore, step 1 includes the following steps: Step 11: The virtual positioning generation unit receives the user's real GPS coordinate information (λ_real, φ_real); Step 12: Add Laplacian noise to the real coordinates according to the differential privacy algorithm; λ_virtual = λ_real + X_λ, where X_λ ~ Lap(0, Δf / ε_λ); φ_virtual = φ_real + X_φ, where X_φ ~ Lap(0, Δf / ε_φ); Where Δf is the global sensitivity, with a value of 0.01 degrees; ε_λ = ε_φ = 0.25 (total privacy budget ε = 0.5 averaged); Step 13: Calculate the confidence region of user activity using the ellipse equation: ((λ-μ_λ) / σ_λ)² + ((φ-μ_φ) / σ_φ)² - 2ρ((λ-μ_λ) / σ_λ)((φ-μ_φ) / σ_φ) ≤ χ²_{0.95,2}. If the virtual coordinates exceed the confidence region, perform a boundary projection operation to project the virtual coordinates that exceed the boundary to the nearest point on the ellipse boundary, thereby generating the corresponding virtual positioning point.
[0018] Furthermore, step 2 includes the following steps: Step 21: The multi-dimensional user profile construction unit receives basic user information, behavior records, and virtual location sequence information, and generates a user feature vector with multi-dimensional feature elements. Step 22: Set the parameters of the DBSCAN clustering method, use the Hidden Markov Model to determine the mode of transportation where the user's real GPS coordinates are located, and assign weights accordingly. Step 23: The DBSCAN clustering method is used to identify the user's frequently used areas. Hidden Markov models are used to analyze the location transfer patterns. Geographical behavior preferences are extracted by combining user point of interest information. The weighted cosine similarity of the user feature vector dimension is summed to calculate the user similarity, thereby deeply integrating geographic location information with user behavior characteristics.
[0019] Furthermore, step 3 includes the following steps: Step 31: The intelligent geofence establishment unit receives information such as virtual GPS coordinates, regional POI density, and historical security events to generate a set of boundary coordinates for a three-level fence. The three-level fence is divided into L1 core fence with a basic radius of 0.5km, which is for high-precision detection; L2 buffer fence with a basic radius of 1.5km, which is for medium-precision detection; and L3 monitoring fence with a basic radius of 3.0km, which is for low-precision monitoring. Step 32: Using the location of the virtual location point as the center, determine the area where the user is currently located. If the user enters a high-risk area, reduce the radius of the three-level fence; if the user is in a low-risk area, expand the radius of the three-level fence. Step 33: Calculate the three-level fence boundary using a cubic Bézier curve for smoothing. The Bézier curve is: P(t) = (1-t)³P0 + 3(1-t)²tP1 + 3(1-t)t²P2 + t³P3; Control point calculation: C_i = P_i + d × normalize(P_{i+1} - P_{i-1}), where d = edge_length / 4.
[0020] Furthermore, the high-risk area is a financial district or a government district, and the low-risk area is a residential area or a park.
[0021] Furthermore, step 4 includes the following steps: Step 41: The public opinion content recognition unit receives the text information to be detected, user profile features, and geofence level information; Step 42: Use the quadtree dynamic grid partitioning algorithm. When the data density exceeds 5 lines per square kilometer per day and the grid size is greater than the minimum value, perform the grid splitting operation. When the data density is less than 1 line per square kilometer per day and all adjacent grids are low density, perform the grid merging operation. Step 43: The multi-level geofencing system sequentially performs the first layer of statistical anomaly detection, the second layer of semantic feature detection, and the third layer of contextual association detection, and dynamically adjusts the detection thresholds according to the user profile and geofencing level.
[0022] Furthermore, step 5 includes the following steps: Step 51: The personalized public opinion early warning and risk assessment unit receives user profile features, virtual location information, historical behavior records, and current time context; Step 52: Calculate the recommendation score; Step 53: Score the risk according to the multidimensional risk assessment model. When the risk score is below 0.3, it is judged as low risk and the basic monitoring strategy is implemented. When the risk score is between 0.3 and 0.6, it is judged as medium risk and the detection frequency is increased. When the risk score is between 0.6 and 0.8, it is judged as high risk and security verification is mandatory. When the risk score exceeds 0.8, it is judged as extremely high risk and the emergency response mechanism is triggered.
[0023] Compared with the prior art, the beneficial effects of the technical solution of the present invention are: This invention discloses a nearby public opinion monitoring and perception method based on social media virtual positioning. By integrating advanced methods such as public opinion monitoring, machine learning classification, deep learning analysis, and location privacy protection, it achieves intelligent identification and monitoring of public opinion content, greatly improving the efficiency and accuracy of public opinion monitoring. It can be widely applied in various fields such as government departments, media organizations, and enterprises. Through these technologies, users can better monitor public opinion dynamics and promptly detect and respond to public opinion storms. Furthermore, compared with existing technologies, this method significantly enhances privacy protection: through a differential privacy perturbation mechanism with ε=0.5, the risk of privacy leakage is greatly reduced. The amplitude of the attack has been reduced, the success rate of location reasoning attacks has decreased significantly, and the detection accuracy has also been greatly improved: the accuracy of public opinion content recognition has increased from about 78% to more than 90%, the false alarm rate has decreased from about 35% to less than 10%, and the F1 score has increased from about 0.7 to more than 0.9. In addition, the dynamic adaptability has been enhanced: the dynamic adjustment of geofencing has significantly improved the system's adaptability, the detection effect in complex scenes has been significantly improved, and the personalized matching degree has increased from about 62% to more than 85%. The system performance has also been improved: the detection response time has been reduced from an average of more than 3 seconds to about 1 second, and the system's processing capacity has been significantly improved to meet the needs of large-scale real-time detection. Attached Figure Description
[0024] Figure 1 This is a flowchart of the nearby public opinion patrol and perception method based on social media virtual positioning in this invention. Detailed Implementation
[0025] The accompanying drawings are for illustrative purposes only and should not be construed as limiting the scope of this patent. To better illustrate this embodiment, some components in the drawings may be omitted, enlarged, or reduced, and do not represent the actual dimensions of the product. It is understandable to those skilled in the art that some well-known structures and their descriptions may be omitted in the drawings.
[0026] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation" and "connection" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances. The technical solution of this invention will be further described below with reference to the accompanying drawings and embodiments.
[0027] like Figure 1 As shown, a method for nearby public opinion monitoring and perception based on social media virtual location includes the following steps: Step 1: The virtual positioning generation unit perturbs the user's real GPS coordinates to generate corresponding virtual positioning points; Step 2: Multi-dimensional user profile construction unit constructs a multi-dimensional user profile that integrates geographic features, deeply fusing geographic location information with user behavior characteristics; Step 3: The intelligent geofencing establishment unit establishes an adaptive multi-level geofencing system, constructing layered detection areas centered on the location of the virtual positioning point; Step 4: The public opinion content identification unit adopts a location-aware intelligent public opinion identification method, combined with a multi-level geofencing system to perform layered detection; Step 5: The personalized public opinion early warning and risk assessment unit receives the features of the multi-dimensional user profile, information of the virtual location point, historical behavior records and current time context, generates personalized public opinion early warning results and risk assessment scores, and monitors and perceives public opinion near the virtual location point.
[0028] In this invention, step 1 includes the following steps: Step 11: The virtual positioning generation unit receives the user's real GPS coordinate information (λ_real, φ_real); Step 12: Add Laplacian noise to the real coordinates according to the differential privacy algorithm; λ_virtual = λ_real + X_λ, where X_λ ~ Lap(0, Δf / ε_λ); φ_virtual = φ_real + X_φ, where X_φ ~ Lap(0, Δf / ε_φ); Where Δf is the global sensitivity, with a value of 0.01 degrees; ε_λ = ε_φ = 0.25 (total privacy budget ε = 0.5 averaged); Step 13: Calculate the confidence region of user activity using the ellipse equation: ((λ-μ_λ) / σ_λ)² + ((φ-μ_φ) / σ_φ)² - 2ρ((λ-μ_λ) / σ_λ)((φ-μ_φ) / σ_φ) ≤ χ²_{0.95,2}. If the virtual coordinates exceed the confidence region, perform a boundary projection operation to project the virtual coordinates that exceed the boundary to the nearest point on the ellipse boundary, thereby generating the corresponding virtual positioning point.
[0029] In this invention, step 2 includes the following steps: Step 21: The multi-dimensional user profile construction unit receives basic user information, behavior records, and virtual location sequence information, and generates a user feature vector with multi-dimensional feature elements. Step 22: Set the parameters of the DBSCAN clustering method, use the Hidden Markov Model to determine the mode of transportation where the user's real GPS coordinates are located, and assign weights accordingly. Step 23: The DBSCAN clustering method is used to identify the user's frequently used areas. Hidden Markov models are used to analyze the location transfer patterns. Geographical behavior preferences are extracted by combining user point of interest information. The weighted cosine similarity of the user feature vector dimension is summed to calculate the user similarity, thereby deeply integrating geographic location information with user behavior characteristics.
[0030] In this invention, step 3 includes the following steps: Step 31: The intelligent geofence establishment unit receives information such as virtual GPS coordinates, regional POI density, and historical security events to generate a set of boundary coordinates for a three-level fence. The three-level fence is divided into L1 core fence with a basic radius of 0.5km, which is for high-precision detection; L2 buffer fence with a basic radius of 1.5km, which is for medium-precision detection; and L3 monitoring fence with a basic radius of 3.0km, which is for low-precision monitoring. Step 32: Using the location of the virtual location point as the center, determine the area where the user is currently located. If the user enters a high-risk area, reduce the radius of the three-level fence; if the user is in a low-risk area, expand the radius of the three-level fence. Step 33: Calculate the three-level fence boundary using a cubic Bézier curve for smoothing. The Bézier curve is: P(t) = (1-t)³P0 + 3(1-t)²tP1 + 3(1-t)t²P2 + t³P3; Control point calculation: C_i = P_i + d × normalize(P_{i+1} - P_{i-1}), where d = edge_length / 4.
[0031] In this invention, the high-risk area is the financial district or the government district, and the low-risk area is the residential area or the park.
[0032] In this invention, step 4 includes the following steps: Step 41: The public opinion content recognition unit receives the text information to be detected, user profile features, and geofence level information; Step 42: Use the quadtree dynamic grid partitioning algorithm. When the data density exceeds 5 lines per square kilometer per day and the grid size is greater than the minimum value, perform the grid splitting operation. When the data density is less than 1 line per square kilometer per day and all adjacent grids are low density, perform the grid merging operation. Step 43: The multi-level geofencing system sequentially performs the first layer of statistical anomaly detection, the second layer of semantic feature detection, and the third layer of contextual association detection, and dynamically adjusts the detection thresholds according to the user profile and geofencing level.
[0033] In this invention, step 5 includes the following steps: Step 51: The personalized public opinion early warning and risk assessment unit receives user profile features, virtual location information, historical behavior records, and current time context; Step 52: Calculate the recommendation score; Step 53: Score the risk according to the multidimensional risk assessment model. When the risk score is below 0.3, it is judged as low risk and the basic monitoring strategy is implemented. When the risk score is between 0.3 and 0.6, it is judged as medium risk and the detection frequency is increased. When the risk score is between 0.6 and 0.8, it is judged as high risk and security verification is mandatory. When the risk score exceeds 0.8, it is judged as extremely high risk and the emergency response mechanism is triggered.
[0034] Example In this embodiment, the process of the nearby public opinion patrol and perception method based on social media virtual positioning is as follows: Step 1: The virtual location generation system receives the user's real GPS coordinates and generates a virtual location using a differential privacy mechanism.
[0035] Specific implementation parameters: - Privacy budget allocation: ε_λ = ε_φ = 0.25, total budget ε_total = 0.5; - Global sensitivity: Δf_longitude = Δf_latitude = 0.01 degrees; - Noise level: b_λ = b_φ = Δf / ε = 0.04; - Confidence ellipse parameters: 95% confidence level, χ²_{0.95,2} = 5.991 distribution. Specific implementation steps
[0036] (1) Obtain the user's real GPS coordinates (116.3974, 39.9093); (2) Generate Laplace noise: noise_λ = 0.008, noise_φ = -0.012; (3) Calculate the virtual coordinates: λ_virtual = 116.4054, φ_virtual = 39.8973; (4) Boundary constraint check: The virtual coordinates are within the range of the user's active ellipse and do not require adjustment; Step Two: Building a Multi-Dimensional User Profile The system constructs a 110-dimensional feature vector based on user data and integrates geographic location information.
[0037] Specific implementation parameters: - Feature vector dimensions: F_basic=12 dimensions, F_behavior=24 dimensions, F_interest=50 dimensions, F_geo=16 dimensions, F_temporal=8 dimensions; - DBSCAN clustering method parameters: eps=0.5km, min_samples=5; - Number of HMM states: 5 states (stationary, walking, driving, public transport, other); - Weight allocation: w=[0.15, 0.25, 0.30, 0.20, 0.10]; Specific implementation steps: (1) Collect basic user information: [Age: 30, Occupation: Engineer, Education: Bachelor's Degree, ...] to form a 12-dimensional vector; (2) Analyze behavioral characteristics: [Access frequency: 8 times / day, Device type: Android, ...] form a 24-dimensional vector; (3) Extracting interest preferences: Based on the content browsed by users, a 50-dimensional interest vector is generated using TF-IDF; (4) Geographic feature extraction: Clustering of 30 days of virtual location data to identify 3 permanent residence areas; (5) Time-series pattern analysis: The user's active time periods were identified as 9:00-18:00 and 20:00-22:00; Step 3: Establishing a Smart Geofencing The system establishes a three-tiered adaptive fence centered on a virtual location.
[0038] Specific implementation parameters: - Fence base radius: L1=0.5km, L2=1.5km, L3=3.0km; - Dynamically adjusted weights: α=0.3, β=0.4, γ=0.3; - Distance between control points of the Bézier curve: d = edge_length / 4; - Boundary error control: ≤50 meters; Specific implementation steps: (1) Obtain the virtual location coordinates (116.4054, 39.8973); (2) Check the density of surrounding POIs: The density of POIs in the financial district is 15 / km², which is 8 higher than the average density. (3) Calculate the density factor: density_factor = log(1 + 15 / 8) = 0.83; (4) Query historical security events: 2 security events occurred in the past 7 days, risk_factor = 0.65; (5) Calculate the historical success rate: history_factor = 0.92 (based on a 30-day sliding window); (6) Dynamically adjust radius: - L1: R_1 = 0.5 × (1 + 0.3×0.83 + 0.4×0.65 + 0.3×0.92) = 0.89km; - L2: R_2 = 1.5 × (1 + 0.3×0.83 + 0.4×0.65 + 0.3×0.92) = 2.67km; - L3: R_3 = 3.0 × (1 + 0.3×0.83 + 0.4×0.65 + 0.3×0.92) = 5.34km; Step 4: Identification of Public Opinion Content The system employs a location-aware, multi-level detection architecture to perform public opinion content recognition.
[0039] Specific implementation parameters: - Grid splitting threshold: 5 grids / day·km²; - Grid merging threshold: 1 grid per day·km²; - Z-score outlier threshold: 2.5; - Word2Vec similarity threshold: 0.75; - BERT confidence threshold: 0.85; - Entropy change rate threshold: 30%; Specific implementation steps: (1) Quadtree mesh generation: The current sensitive data density in the region is 8 records / day·km², so perform mesh splitting; (2) Text to be tested: "A local government policy has triggered strong public dissatisfaction"; (3) First layer detection: Calculate Z-score = 3.2 > 2.5, and mark it as a statistical outlier; (4) Second layer detection: The similarity with the public opinion keyword vector is calculated as 0.89 > 0.75, and it is identified as public opinion content; (5) Third layer detection: The confidence score of the BERT model output is 0.92 > 0.85, confirming it as negative public opinion; (6) Location entropy calculation: The current area H(X) = 2.34, and the change rate compared to the previous period is 35% > 30%, triggering an alarm; (7) Integration of detection strategies: Users within the L1 core fence are subject to strict standards and are ultimately judged as high-risk public opinion content; Step 5: Personalized Public Opinion Early Warning and Risk Assessment (corresponding to S105) The system provides personalized public opinion early warnings and risk assessments based on user profiles and detection results.
[0040] Specific implementation parameters: - Number of features in the recommendation algorithm: F=50; - Risk assessment weights: w=[0.3, 0.25, 0.2, 0.15, 0.1]; - Risk level thresholds: Low risk <0.3, Medium risk 0.3-0.6, High risk 0.6-0.8, Very high risk ≥0.8; - Feedback learning rate: α = 0.01; Specific implementation steps: (1) Personalized public opinion early warning calculation: - Global average risk score μ = 3.2; - User bias b_u = 0.15 (This user tends to be highly risk-sensitive); - Public opinion type bias b_i = -0.08 (policy-related public opinion); - Sum of eigenvector products = 0.73; - Final warning score R = 3.2 + 0.15 + (-0.08) + 0.73 = 4.0; (2) Multidimensional risk assessment calculation: - f_risk_1 (historical leaks) = log(1 + 2 / 100) = 0.02; - f_risk_2 (location risk) = 0.8 (high risk in the financial zone); - f_risk_3 (abnormal behavior) = 0.3 (minor aberration); - f_risk_4 (time sensitive) = 0.6 (working time); - f_risk_5 (network security) = 0.4 (enterprise network); - Overall risk score = 0.3×0.02 + 0.25×0.8 + 0.2×0.3 + 0.15×0.6 + 0.1×0.4 = 0.396; (3) Risk level determination: 0.3 ≤ 0.396 < 0.6, determined as medium risk, increase detection frequency.
[0041] Example verification: The probability assessment of historical public opinion events is weighted at 0.3. The current location's public opinion risk level assessment is weighted at 0.25. The user behavior anomaly assessment was weighted at 0.2. The weight for time sensitivity assessment (such as sensitive periods and important milestones) is set at 0.15. The weight for assessing the spread of public opinion in the online environment is set at 0.1.
[0042] The positional relationships described in the figures are for illustrative purposes only and should not be construed as limiting this patent. Clearly, the above embodiments of the present invention are merely examples to clearly illustrate the invention and are not intended to limit the implementation of the invention. Those skilled in the art can make other variations or modifications based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations here. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the claims of this invention.
Claims
1. A method for nearby public opinion monitoring and perception based on social media virtual positioning, characterized in that: Includes the following steps: Step 1: The virtual positioning generation unit perturbs the user's real GPS coordinates to generate corresponding virtual positioning points; Step 2: Multi-dimensional user profile construction unit constructs a multi-dimensional user profile that integrates geographic features, deeply fusing geographic location information with user behavior characteristics; Step 3: The intelligent geofencing establishment unit establishes an adaptive multi-level geofencing system, constructing layered detection areas centered on the location of the virtual positioning point; Step 4: The public opinion content identification unit adopts a location-aware intelligent public opinion identification method, combined with a multi-level geofencing system to perform layered detection; Step 5: The personalized public opinion early warning and risk assessment unit receives the features of the multi-dimensional user profile, information of the virtual location point, historical behavior records and current time context, generates personalized public opinion early warning results and risk assessment scores, and monitors and perceives public opinion near the virtual location point.
2. The method for nearby public opinion monitoring and perception based on social media virtual positioning according to claim 1, characterized in that: Step 1 includes the following steps: Step 11: The virtual positioning generation unit receives the user's real GPS coordinate information (λ_real, φ_real); Step 12: Add Laplacian noise to the real coordinates according to the differential privacy algorithm; λ_virtual = λ_real + X_λ, where X_λ ~ Lap(0, Δf / ε_λ); φ_virtual = φ_real + X_φ, where X_φ ~ Lap(0, Δf / ε_φ); Where Δf is the global sensitivity, with a value of 0.01 degrees; ε_λ = ε_φ = 0.25 (total privacy budget ε = 0.5 averaged); Step 13: Calculate the confidence region of user activity using the ellipse equation: ((λ-μ_λ) / σ_λ)² + ((φ-μ_φ) / σ_φ)² - 2ρ((λ-μ_λ) / σ_λ)((φ-μ_φ) / σ_φ) ≤ χ²_{0.95,2}. If the virtual coordinates exceed the confidence region, perform a boundary projection operation to project the virtual coordinates that exceed the boundary to the nearest point on the ellipse boundary, thereby generating the corresponding virtual positioning point.
3. The method for nearby public opinion monitoring and perception based on social media virtual positioning according to claim 1, characterized in that: Step 2 includes the following steps: Step 21: The multi-dimensional user profile construction unit receives basic user information, behavior records, and virtual location sequence information, and generates a user feature vector with multi-dimensional feature elements. Step 22: Set the parameters of the DBSCAN clustering method, use the Hidden Markov Model to determine the mode of transportation where the user's real GPS coordinates are located, and assign weights accordingly. Step 23: The DBSCAN clustering method is used to identify the user's frequently used areas. Hidden Markov models are used to analyze the location transfer patterns. Geographical behavior preferences are extracted by combining user point of interest information. The weighted cosine similarity of the user feature vector dimension is summed to calculate the user similarity, thereby deeply integrating geographic location information with user behavior characteristics.
4. The method for nearby public opinion monitoring and perception based on social media virtual positioning according to claim 1, characterized in that: Step 3 includes the following steps: Step 31: The intelligent geofence establishment unit receives information such as virtual GPS coordinates, regional POI density, and historical security events to generate a set of boundary coordinates for a three-level fence. The three-level fence is divided into L1 core fence with a basic radius of 0.5km, which is for high-precision detection; L2 buffer fence with a basic radius of 1.5km, which is for medium-precision detection; and L3 monitoring fence with a basic radius of 3.0km, which is for low-precision monitoring. Step 32: Using the location of the virtual location point as the center, determine the area where the user is currently located. If the user enters a high-risk area, reduce the radius of the three-level fence; if the user is in a low-risk area, expand the radius of the three-level fence. Step 33: Calculate the three-level fence boundary using a cubic Bézier curve for smoothing. The Bézier curve is: P(t) = (1-t)³P0 + 3(1-t)²tP1 + 3(1-t)t²P2 + t³P3; Control point calculation: C_i = P_i + d × normalize(P_{i+1} - P_{i-1}), where d = edge_length / 4.
5. The method for nearby public opinion monitoring and perception based on social media virtual positioning according to claim 4, characterized in that: The high-risk areas are financial districts or government districts, while the low-risk areas are residential areas or parks.
6. The method for nearby public opinion monitoring and perception based on social media virtual positioning according to claim 1, characterized in that: Step 4 includes the following steps: Step 41: The public opinion content recognition unit receives the text information to be detected, user profile features, and geofence level information; Step 42: Use the quadtree dynamic grid partitioning algorithm. When the data density exceeds 5 lines per square kilometer per day and the grid size is greater than the minimum value, perform the grid splitting operation. When the data density is less than 1 line per square kilometer per day and all adjacent grids are low density, perform the grid merging operation. Step 43: The multi-level geofencing system sequentially performs the first layer of statistical anomaly detection, the second layer of semantic feature detection, and the third layer of contextual association detection, and dynamically adjusts the detection thresholds according to the user profile and geofencing level.
7. The method for nearby public opinion monitoring and perception based on social media virtual positioning according to claim 1, characterized in that: Step 5 includes the following steps: Step 51: The personalized public opinion early warning and risk assessment unit receives user profile features, virtual location information, historical behavior records, and current time context; Step 52: Calculate the recommendation score; Step 53: Score the risk according to the multidimensional risk assessment model. When the risk score is below 0.3, it is judged as low risk and the basic monitoring strategy is implemented. When the risk score is between 0.3 and 0.6, it is judged as medium risk and the detection frequency is increased. When the risk score is between 0.6 and 0.8, it is judged as high risk and security verification is mandatory. When the risk score exceeds 0.8, it is judged as extremely high risk and the emergency response mechanism is triggered.