A big data visual interactive display method and system

Abstract

The application relates to the technical field of big data interaction visualization, and discloses a big data visualization interaction display method and system, which first carries out sparse projection on high-dimensional data and constructs a dynamic grid index; then captures an interaction behavior to generate a dynamic intention vector; combines a global / local projection matrix to generate a hybrid projection result; identifies an attention area based on the hybrid projection and dynamically adjusts a privacy protection strength; recommends a projection parameter through multi-view association rendering and a reinforcement learning strategy; finally, iteratively optimizes an intention model and a recommended strategy by using user feedback; and the application further constructs a closed-loop system comprising data preprocessing, intention perception, dynamic projection, dynamic aggregation, visualization rendering and feedback iteration modules. The intention-driven hybrid projection and dynamic privacy control balance the privacy-utility contradiction, the reinforcement learning is combined to recommend cross-scene parameters, a bidirectional feedback closed loop is constructed to realize system continuous optimization, and the interaction adaptability and long-term evolution capability are improved.

Description

Technical Field

[0001] This invention relates to the field of big data interactive visualization technology, specifically to a big data visualization interactive display method and system. Background Technology

[0002] With the widespread application of big data visualization analytics in sensitive fields such as financial risk control and medical diagnosis, how to ensure data privacy while maintaining visualization effectiveness during interaction has become a pressing technical challenge. Currently, visualization systems based on differential privacy typically employ static noise injection or fixed projection parameters. While these provide basic privacy protection, they struggle to adapt to dynamic user interactions and spatial variations in data distribution. These methods often lead to two extremes: insufficient noise in high-density data areas causing group privacy leaks, or excessive noise obscuring useful information in sparse areas, severely limiting the depth and reliability of visualization analytics.

[0003] The core flaw in existing technologies lies in the contradiction between their static design logic and dynamic interaction requirements. On the one hand, the global equalization strategy of privacy budgeting ignores changes in data spatial density, failing to achieve a dynamic trade-off between privacy protection and data fidelity. On the other hand, the manual configuration of projection parameters relies on experience and lacks the ability to perceive users' real-time intentions, resulting in a disconnect between visualization results and interactive behavior. This fragmented optimization mechanism leaves the system facing a dual bottleneck of uncontrollable privacy risks and low interaction efficiency in complex analysis scenarios, making it difficult to meet users' needs for in-depth exploration of sensitive data.

[0004] To address the shortcomings of the existing technologies, this invention proposes a big data visualization and interactive display method and system. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a big data visualization and interactive display method and system, which solves the problems of difficulty in balancing privacy protection and data utility in real time and the lack of adaptive adjustment capability in parameter configuration in dynamic user interaction scenarios.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a big data visualization and interactive display method, comprising the following steps:

[0007] S1. Perform sparse projection processing on the original high-dimensional data to generate low-dimensional projection results and construct a dynamic grid index.

[0008] S2. Capture user interaction behavior based on the dynamic grid index, extract intent features and generate dynamic intent vectors;

[0009] S3. Generate a local projection matrix based on the dynamic intent vector, combine it with the pre-trained global projection matrix, dynamically adjust the mixing weights through the time decay coefficient, and calculate and generate the mixed projection result.

[0010] S4. Based on the hybrid projection results, identify the user's area of ​​interest, perform recursive grid division and incremental statistical aggregation, dynamically adjust the privacy protection strength, and output dynamic aggregation results, wherein the dynamic aggregation results include the focus area, grid statistics and privacy parameters;

[0011] S5. Based on the dynamic aggregation results, perform multi-view association rendering, generate a projection parameter recommendation list through reinforcement learning strategy, and collect user behavior to generate feedback tuples.

[0012] S6. Generate feedback tuples based on user behavior, parse the adoption intention and behavior adjustment signals in the feedback tuples, and iteratively update the intention weight matrix and reinforcement learning strategy structure. The results of the iterative update will be applied to the next round of recommendation generation to achieve adaptive optimization and closed-loop learning of the system.

[0013] Preferably, step S1 includes:

[0014] S1-1. Construct a sparse random global projection matrix W based on the original high-dimensional data. g Calculate global skeleton projection Where, x i The original high-dimensional data points;

[0015] S1-2, Based on global skeleton projection Initialize the dynamic mesh tree, divide it into coarse-grained grids, and store the statistics S. j ;

[0016] S1-3, regarding the statistic S j Injecting Laplace noise generates a privacy-preserving dynamic grid index.

[0017] Preferably, step S2 includes:

[0018] S2-1. Analyze the center coordinates of the user's selected area (c x c y Given area A, generate the region feature vector φ. R =[c x c y [,A];

[0019] S2-2. Perform a Fourier transform on the brush trajectory to extract the previous... The low-frequency components generate the feature vector φ T Where L is the number of sampling points for the brush trajectory, and p≤10. When L<20, p=L.

[0020] S2.3, the region feature vector φ R With trajectory feature vector φ T Perform weighted concatenation to generate the intent feature vector φ of the user's interaction behavior at the current moment. current =[αφ R ,βφ T ], where α and β are dynamic weighting coefficients;

[0021] S2-4. Based on the intent feature vector φ of the user interaction behavior at the current moment. current The intention weight matrix M is updated using an online gradient descent formula, which is:

[0022]

[0023] Among them, M t With M t+1 Let represent the values ​​of the intention weight matrix at time steps t and t+1, and η be the learning rate. For the gradient operator, φ current Let ||φ be the intent feature vector of the user's interaction behavior at the current moment. current -M t φ history || 2 Let φ be the loss function. history The sliding window mean of the historical intent feature vector. Where k is the window length and t is the time variable. Let i be the intent feature vector of the user's interaction behavior at the current moment of time step i.

[0024] Preferably, step S3 includes:

[0025] S3-1, Generate the local projection matrix: Based on the intent feature vector φ of the user's interaction behavior at the current moment. current Generate the local projection matrix W l (x i );

[0026] S3-2, Obtain the global projection matrix: Load the pre-trained global projection matrix W from the data preprocessing module. g The pre-trained global projection matrix W g It was obtained by training historical data offline using a sparse autoencoder;

[0027] S3-3, Calculate the mixed projection weights: based on the time decay coefficient α(t) = α0e -λt The intensity of the local projection is dynamically adjusted, where α0 is the initial weight of the local projection, λ is the decay rate parameter, and t is the time variable.

[0028] S3-4. Hybrid Projection Calculation: Based on the local projection matrix and the global projection matrix, the hybrid projection result is calculated using the hybrid projection calculation formula, which is as follows:

[0029]

[0030] Where α(t) is the time decay coefficient; For the original high-dimensional data points x i The final projection result; x i For the original high-dimensional data points; W g W is the global projection matrix. l (x i ) is a local projection matrix, dependent on the original high-dimensional data points x. i Or it can be dynamically generated based on user intent.

[0031] Preferably, step S4 includes:

[0032] S4-1, Based on the norm ||φ of the current intent vector current ||2. Adaptively set the mesh granularity g, and combine it with the hybrid projection results to identify the initial set of mesh cells that have significant user attention.

[0033] S4-2, Initial Mesh Cell Set Each grid cell in the process undergoes recursive subdivision, and the local statistics S of the sensitive features of each grid cell are incrementally updated. i Construct a gridded feature representation structure for differential privacy mechanisms.

[0034] S4-3. Based on the user interaction behavior of each attention grid unit, calculate the popularity index h of each attention grid unit. i It also dynamically adjusts the privacy protection strength parameters based on the behavior-driven privacy adjustment formula. The behavior-driven privacy modulation formula is as follows:

[0035]

[0036] Where Z is the thermal amplification factor, ρ base Basic privacy protection strength;

[0037] S4-4, Subsequently, using the local statistic S i Based on the adjusted intensity parameters, a privacy protection objective function is constructed, and the optimal privacy budget is solved using the Lagrange optimization method. Finally, calculate the privacy strength adjustment amount. Forming a privacy regulation structure As part of the output of the dynamic aggregation result, the expression of the privacy-preserving objective function is as follows:

[0038]

[0039] in, Optimize the function for the objective. To account for the user experience or data availability costs associated with using higher privacy levels, the coefficient in the loss function that balances privacy costs with accuracy is Error(S). i ,∈ i This refers to the loss of data accuracy caused by privacy disturbances.

[0040] Preferably, step S5 includes:

[0041] S5-1, Visualization Rendering: Encode the dynamic aggregation results output in step S4 into a heatmap, and perform multi-view association rendering with the hybrid projection scatter plot generated in step S3.

[0042] S5-2, Strategy Recommendation: Based on the intent feature vector of the user's interaction behavior at the current moment in step S2 and the time decay coefficient α(t) in step S3, a combination of projection parameters (α) is generated through a Q-learning strategy. * The recommended list (λ*,β*) is defined synchronously with the adoption flag.

[0043] S5-3, Feedback Generation: Collect the adoption identifier δ and the privacy protection strength adjustment behavior in step S4, combined with the norm ||φ of the current intent vector. current ||2, generate a function containing (δ, ΔW) l ,||φ current ||2,∈ * The feedback tuple of , Δρ).

[0044] Preferably, step S6 includes:

[0045] S6-1, Feedback Analysis: Analyzing the feedback tuple (δ, ΔW) l ,||φ current ||2,∈ * , Δρ), extract explicit score values, number of undo operations, privacy budget parameters and intent intensity indicators, and construct a feedback sample set for incremental model updates;

[0046] S6-2, Intent Model Update: Based on the feedback samples, the intent weight matrix M of the user interaction behavior at the current moment is incrementally updated using the online gradient descent formula. The mapping ability of the intent weight matrix to the intent feature vector is adjusted, and the update result is fed back to step S2 for the next round of intent feature generation.

[0047] S6-3, Reinforcement Learning Strategy Structure Optimization: Based on the adoption flag and privacy adjustment amount in the feedback, update the exploration rate parameter ∈ in the Q-learning strategy, adjust the strategy score of each projection parameter combination in the recommendation strategy, and the updated strategy structure will be used to generate the projection parameter combination recommendation list in step S5-2.

[0048] This invention also provides a big data visualization and interactive display system, applied to the aforementioned big data visualization and interactive display method, comprising:

[0049] The data preprocessing module is used to perform sparse projection processing on the original high-dimensional data, generate low-dimensional projection results, and construct a dynamic grid index. The grid index is used to capture interactive behavior and support subsequent privacy protection processing.

[0050] The intent awareness module is used to receive dynamic grid indexes, parse the user's selected area and pen trajectory, extract area features and trajectory features, and generate intent feature vectors by weighted concatenation, and update the intent weight matrix through online gradient descent formula.

[0051] The dynamic projection module is used to receive the intent feature vector, generate a local projection matrix according to the current intent, and combine it with a pre-trained global projection matrix to dynamically generate a hybrid projection result based on the time decay coefficient.

[0052] The dynamic aggregation module is used to receive the hybrid projection results, identify the user's attention area based on the intent norm, perform recursive grid division and incremental statistical aggregation, calculate the popularity index based on user interaction behavior and adjust the privacy protection strength parameter, and output the dynamic aggregation result, wherein the dynamic aggregation result includes the focus area, grid statistics and privacy parameters.

[0053] The visualization rendering module is used to jointly render the dynamic aggregation results and the mixed projection results into a multi-view interface, and generate a recommended list of projection parameter combinations based on the current intent features and time decay factor through the Q-learning strategy.

[0054] The feedback iteration module is used to collect user feedback on recommended combinations, generate feedback tuples, parse adoption intentions and behavior adjustment signals, iteratively update the intention weight matrix and reinforcement learning strategy structure, and feed the update results back to the intention perception module and visualization rendering module to achieve system adaptive optimization and closed-loop learning.

[0055] Preferably, the data preprocessing module and the intent-aware module exchange dynamic grid indexes via a distributed message queue. The dynamic grid indexes contain the following information:

[0056] Grid space coordinate range [x min x max ]×[ymin y max ];

[0057] Statistical triples after noise injection in,, Let j be the noise mean of grid. Let the noise standard deviation of grid j be count. j is the number of original data points in grid j.

[0058] Preferably, the intent perception module and the dynamic projection module satisfy the following real-time constraints:

[0059] The intent feature vector φ of the user's interaction behavior at the current moment current The generation delay is no more than 50 milliseconds;

[0060] The intent feature vector φ of the user's interaction behavior at the current moment current The results are transferred to the dynamic projection module via zero-copy memory sharing to ensure rapid generation and response of the hybrid projection results.

[0061] This invention provides a method and system for interactive visualization of big data. It has the following beneficial effects:

[0062] 1. This invention achieves adaptive adjustment of the projection direction based on user interaction intent by extracting intent features in real time and updating the intent weight matrix online, combined with the dynamic generation of global-local hybrid projection. Compared to the view rigidity problem caused by fixed projection parameters in existing technologies, this invention solves the deficiency of not being able to capture real-time user intent through local projection matrix adjustment.

[0063] 2. This invention dynamically allocates the privacy budget by constructing a Lagrange optimization model and adjusts the privacy protection strength parameters differently for regions with different data densities. Compared with the traditional uniform noise injection method, it effectively balances the privacy leakage risk in high-density regions with the data accuracy loss in low-density regions, and solves the problem of imbalance between utility and privacy protection caused by a uniform allocation strategy.

[0064] 3. This invention recommends projection parameter combinations based on a Q-learning strategy and expands the state space by combining user intent feature vectors to achieve cross-scenario adaptive parameter configuration. Compared to manually preset parameters or static rule bases, it overcomes the technical limitations of these methods, which struggle to dynamically adjust the exploration rate parameter to adapt to diverse user needs and real-time interaction scenarios.

[0065] 4. The present invention uses feedback tuple synchronization to optimize the intention weight matrix and the reinforcement learning policy structure, forming a two-way feedback closed loop. Compared with the traditional single-model update mechanism, it fills the technical gap of the separate operation of the intention perception module and the recommendation system, enabling the system to continuously improve the intention recognition accuracy and parameter recommendation accuracy in long-term interactions. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] Figure 1 is a flowchart of the method of the present invention;

[0067] Figure 2 is a process framework diagram of the system of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0068] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the drawings of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0069] Please refer to the attached Figure 1 , the embodiment of the present invention provides a big data visualization interactive display method, including the following steps:

[0070] S1. Perform sparse projection processing on the original high-dimensional data, generate a low-dimensional projection result, and construct a dynamic grid index;

[0071] In this embodiment, first perform sparse projection processing on the original high-dimensional data to achieve data dimension compression and preliminary feature skeleton extraction.

[0072] The original high-dimensional data is represented as a set of data points where, is the i-th high-dimensional data point, with dimension d and a total data volume of N.

[0073] Construct a sparse random global projection matrix The matrix satisfies the following conditions:

[0074] Each row vector follows a Laplace or Gaussian sparse distribution;

[0075] The proportion of non-zero elements does not exceed the set sparsity threshold θ, where θ ∈ (0, 1);

[0076] The projection dimension r << d to achieve dimension compression.

[0077] Multiply the original data x i on the left by the sparse global projection matrix W g , to obtain the global skeleton projection result:

[0078] in, This represents the low-dimensional global projection result of the i-th sample point.

[0079] Subsequently, the global skeleton projection results were used. Initialize the dynamic grid index structure as input. Logically, this structure is represented by a dynamic grid tree; physically, it includes node record units, index pointer units, and statistical storage units. The specific structure includes:

[0080] Grid node cells are used to record the coordinate range in the current projection space;

[0081] Statistical unit, used to store the data distribution information S falling within the region of this node. j ;

[0082] The child node pointer unit is used to point to the sub-regions after recursive partitioning.

[0083] The low-dimensional space is divided into coarse-grained grids, dividing the entire space into initial square grid regions with side length g0, where g0 is the smallest division unit set during system initialization.

[0084] Each grid node records sensitive statistical information such as the number of samples falling within its range, mean, and variance, forming an initial set of statistics {S}. j},in,:

[0085] Where, μ j This is the sample mean in this grid cell. Let n be the sample variance. j This represents the number of samples.

[0086] To meet the requirements of differential privacy mechanisms, Laplace noise needs to be injected into the above statistics while constructing the dynamic grid index.

[0087] Specifically, for each grid cell statistic S i Perform the following noise injection operation:

[0088] Where Lap(·) represents the Laplace distribution sampling function, Δf is the sensitivity (i.e., the maximum change that each statistic may cause due to a single data change), and ∈ is the preset privacy budget parameter.

[0089] Statistics after injecting noise It is stored in the grid node statistics unit and serves as the basis for subsequent differential privacy protection mechanisms.

[0090] The final dynamic grid index structure can be used for user behavior capture, privacy adjustment, and dynamic aggregation result generation. Logically, it communicates with the intent feature extraction module in step S2 and serves as the spatial basis for intent vector generation.

[0091] This implementation method, through the above-described steps of sparse projection, skeleton extraction, mesh construction, and differential privacy noise injection, can effectively achieve spatial dimensionality reduction representation and index optimization structure construction of high-dimensional data while ensuring user data privacy, thus providing support for subsequent intent modeling and recommendation strategies.

[0092] S2. Capture user interaction behavior based on the dynamic grid index, extract intent features and generate dynamic intent vectors;

[0093] In this embodiment, user interaction behavior is captured based on a dynamic grid index, and a dynamic intent vector is generated through multimodal feature fusion. First, the spatial attributes of the user-selected region are analyzed. This selected region is defined by the user through mouse or touch operations in the visual interface, where the center coordinates (c...) are... x c y The area A and the region feature vector are obtained in real time through the geometric calculation module.

[0094] φ R =[c x c y A], where c x and c y The coordinates of the center of the selected area in the projection space are given by A, where A is the area of ​​the area (in pixels or normalized units).

[0095] Simultaneously, time-frequency features are extracted from the user's pen trajectory. The pen trajectory is recorded as a discrete point sequence by the trajectory sampling module. Where L is the number of sampling points. A Fast Fourier Transform (FFT) is performed on the trajectory coordinate sequence, retaining the first p low-frequency components:

[0096] When L≥20;

[0097] p = L when L < 20;

[0098] The extracted frequency domain feature vector is:

[0099] φ T =[Re(F1),Im(F1),…,Re(F p ),Im(F p )];

[0100] Among them, F s Let represent the s-th Fourier coefficient, and Re(·) and Im(·) represent the real and imaginary parts, respectively.

[0101] The region feature vector φ R With trajectory feature vector φ T Perform dynamic weighted concatenation to generate the intent feature vector for the current time step:

[0102] φ current =[αφ R ,βφ T ];

[0103] The weighting coefficients α and β are dynamically adjusted based on user behavior feedback within the sliding window, satisfying α + β = 1.

[0104] The intention weight matrix M is updated using an online gradient descent algorithm, and the update formula is as follows:

[0105]

[0106] Among them, M t With M t+1 Let represent the values ​​of the intention weight matrix at time steps t and t+1, and η be the learning rate. For the gradient operator, φ current Let ||φ be the intent feature vector of the user's interaction behavior at the current moment. current -M t φ history || 2 The loss function;

[0107] φ history The sliding window mean of the historical intent feature vector. Where k is the window length and t is the time variable. Let i be the intent feature vector of the user's interaction behavior at the current moment of time step i.

[0108] The physical implementation of the intent weight matrix M includes a feature mapping unit and a weight storage unit. The feature mapping unit maps historical intent features to the current feature space through matrix multiplication operations, and the weight storage unit stores matrix parameters through non-volatile memory.

[0109] This implementation method, through the aforementioned multimodal feature extraction, dynamic weighted fusion, and online learning mechanisms, can effectively capture the spatiotemporal evolution characteristics of user interaction intentions, providing an interpretable feature-driven basis for subsequent projection parameter recommendations.

[0110] S3. Generate a local projection matrix based on the dynamic intent vector, combine it with the pre-trained global projection matrix, dynamically adjust the mixing weights through the time decay coefficient, and calculate and generate the mixed projection result.

[0111] In this embodiment, the generation of the hybrid projection result includes four stages: local projection matrix construction, global projection loading, weight adjustment, and hybrid calculation. First, based on the intent feature vector φ of the user's interaction behavior at the current moment... current and the original data point x i ,

[0112] W l (x i )=U·diag(φ current )·V T ;

[0113] in, and is the pre-trained basis matrix, obtained through offline training using historical data, m is the hidden dimension, and diag(·) represents converting the vector into a diagonal matrix.

[0114] Original data point x i Multiplying by the local projection matrix yields the local projection components:

[0115] W l (x i )x i =(U·diag(φ) current )·V T )x i ;

[0116] Global projection matrix W g Loaded from the preprocessing module, it is obtained through optimization by a sparse autoencoder, with the objective function being:

[0117]

[0118] Where χ controls sparsity, and ||·||1 is the L1 norm. Represents the global projection matrix W g The transpose of .

[0119] The mixed weights are dynamically adjusted through a time decay coefficient, calculated using the following formula:

[0120] α(t)=α0e -λt ;

[0121] in,:

[0122] α0∈[0.5,1.0] represents the initial local weights;

[0123] λ>0 represents the decay rate (default value λ=0.05);

[0124] t is a time variable (cumulative number of user interactions).

[0125] The final hybrid projection result is calculated using the following formula:

[0126]

[0127] Where α(t) is the time decay coefficient; For the original high-dimensional data points x i The final projection result; x i For the original high-dimensional data points; W g W is the global projection matrix. l (x i ) is a local projection matrix, dependent on the original high-dimensional data points x. i Or it can be dynamically generated based on user intent.

[0128] In terms of physical implementation, the hybrid projection module includes:

[0129] Base matrix storage unit: used to store U and V;

[0130] Intent Feature Interface: Receives φ from the output of step S2. current ;

[0131] Matrix generating unit: based on W l (x i )=U·diag(φ current )·V T Real-time computing W l (x i );

[0132] Hybrid computing unit: Execution Linear combination operations.

[0133] In the embodiment, when the user performs 5 consecutive box selection operations (t=5), the time decay coefficient is calculated as α(5) = 0.8·e -0.05×5 ≈0.618, at which point the local projection weight decays to 77.3% of the initial value, reflecting an adaptive decay to recent intentions.

[0134] S4. Based on the hybrid projection results, identify the user's area of ​​interest, perform recursive grid division and incremental statistical aggregation, dynamically adjust the privacy protection strength, and output dynamic aggregation results, wherein the dynamic aggregation results include the focus area, grid statistics and privacy parameters;

[0135] In this implementation, the generation of dynamic aggregation results includes four stages: region of interest identification, recursive grid partitioning, privacy strength adjustment, and optimization solution. First, based on the intent feature vector φ of the user's interaction behavior at the current moment... current The L2 norm sets the mesh granularity:

[0136]

[0137] Among them, g base The reference granularity is μ, the normalization coefficient (default value μ = 10.0), and g is the mesh granularity.

[0138] Based on the original high-dimensional data points x i The final projection result Identify user focus areas The judgment criteria are:

[0139]

[0140] Where, n j τ represents the number of samples in cell j, and τ is the attention threshold (default τ = 1.5). This represents the total number of grid cells.

[0141] right Perform recursive subdivision, halving the side length in each subdivision, until g is satisfied. min (Minimum granularity threshold). Statistic S for each subdivided grid. i Incremental update:

[0142]

[0143] Where, ΔS i This represents the statistical increment of the newly added sample. The grid statistics before the update. For the updated grid statistics (triplets: mean, variance, count).

[0144] Heat index h i Calculated based on user interaction frequency:

[0145]

[0146] Privacy strength parameters The dynamic adjustment formula is:

[0147]

[0148] in,:

[0149] Z is the heat amplification factor (default Z = 0.1);

[0150] ρ base Basic privacy strength (default ρ) base =0.5);

[0151] log(·) is the natural logarithm function.

[0152] The privacy protection objective function is defined as follows:

[0153]

[0154] in, Optimize the function for the objective. To account for the user experience or data availability costs associated with using higher privacy levels, the coefficient in the loss function that balances privacy costs with accuracy is Error(S). i ,∈ i The error term represents the loss of data accuracy due to privacy perturbations. Privacy cost item

[0155] Solving for the optimal privacy budget using the Lagrange multiplier method Final output privacy adjustment structure:

[0156]

[0157] in, For privacy intensity adjustment, This is the final privacy-adjusting structure used to guide noise injection.

[0158] In terms of physical implementation, the dynamic aggregation module includes:

[0159] Particle size calculation unit: based on the formula Real-time calculation of grid granularity;

[0160] Recursive partitioning controller: manages the mesh subdivision process;

[0161] Popularity statistics unit: Cumulative user interaction behavior data;

[0162] Privacy-optimized solver: Execute formula Optimized calculation.

[0163] In the embodiment, when a user interacts with a certain grid cell 10 times (within a 5-minute time window), the popularity h... i =2, privacy strength adjusted to It embodies interactive-driven privacy enhancement.

[0164] S5. Based on the dynamic aggregation results, perform multi-view association rendering, generate a projection parameter recommendation list through reinforcement learning strategy, and collect user behavior to generate feedback tuples.

[0165] In this embodiment, multi-view associated rendering and parameter recommendation are implemented through the following process. First, the dynamic aggregation result output from step S4 is encoded into a heatmap, whose color mapping function is:

[0166]

[0167] Among them, Cj S is the color value of grid cell j. j The statistics for each grid cell j, S min and S max The minimum / maximum statistics within the current view.

[0168] Multi-view associated rendering is performed with the hybrid projection scatter plot of S3. When the user selects a certain area in the heat map, the corresponding data point in the scatter plot is highlighted. The association logic is implemented through the spatial coordinate mapping module.

[0169] The recommended projection parameters use the Q-Learning strategy, whose state space is defined as follows:

[0170] s t =[φ current , α(t), ||φ current ||2];

[0171] The action space is a combination of parameters a = (α) * , λ * ,β * ),in:

[0172] α * ∈[0.1,1.0]: Local projection weight adjustment amount;

[0173] λ * ∈[0.01,0.1]: Adjustment amount for time decay rate;

[0174] β * ∈[0.1,0.5]: Track feature weight adjustment amount.

[0175] The formula for updating the Q value is:

[0176]

[0177] in,:

[0178] η = 0.1: Learning rate;

[0179] γ = 0.9: Discount factor;

[0180] r t Instant reward, calculated using the formula r t =δ·||φ current ||2-Δρ.

[0181] The feedback tuple generation module collects the user's adoption flag δ (δ = 1 indicates adoption and recommendation, δ = 0 indicates rejection), and combines it with the privacy modifier Δρ and the current intent strength ||φ. current 2. Generate feedback tuples:

[0182] Feedback tuple = (δ, ΔW) l ,||φ current ||2,∈ * ,Δρ);

[0183] Wherein, ΔW l The adjustment amount for the local projection matrix is ​​calculated using matrix differencing:

[0184]

[0185] Wherein, ΔW l This is the adjustment matrix of the local projection matrix, reflecting the changes in projection parameters caused by user interaction; This is the adjusted local projection matrix (dimension r×d); This is the local projection matrix before adjustment (dimension r×d).

[0186] In terms of physical implementation, the recommendation system includes:

[0187] Rendering engine: Performs coordinate mapping between heatmaps and scatter plots;

[0188] Q table storage unit: stores the state-action value matrix;

[0189] Feedback collection interface: Captures user interaction events in real time;

[0190] Parameter regulator: Injects recommended parameters into steps S2-S4.

[0191] In the example, when the user adopts the recommendation parameter (α) * =0.8, λ*=0.05, β*=0.3), the system updates the Q table and generates feedback tuples for subsequent strategy optimization.

[0192] S6. Generate feedback tuples based on user behavior, analyze their adoption intention and behavior adjustment signals, and iteratively update the intention weight matrix and reinforcement learning strategy structure. The results of the iterative update will be applied to the next round of recommendation generation, realizing the system's adaptive optimization and closed-loop learning.

[0193] In this embodiment, the system's adaptive optimization is achieved through three stages: feedback parsing, intent model updating, and policy optimization. First, the feedback tuple (δ, ΔW) is parsed. l ,||φ current ||2,∈ * , Δρ), extract the following features to construct training samples:

[0194] sample = (δ, ΔW) l ,||φ current ||2,∈ * ,Δρ,t);

[0195] Where t is a time variable used to identify the temporal relationship of feedback; δ is the adoption flag (1 = adoption, 0 = rejection); ΔW l For local projection matrix adjustment (from step S5); ||φ current ||2 is the L2 norm of the current intent vector; ∈ * The optimized privacy budget (from step S4); Δρ is the privacy strength adjustment (from step S4).

[0196] The online update formula for the intent weight matrix M is:

[0197]

[0198] Where η = 0.01 is the learning rate; φ current The current intent feature vector (from step S2); ΔW l For local projection matrix adjustment (from step S5); M new M is the updated intent weight matrix; old This represents the intent weight matrix before the update; η is the learning rate (default 0.01), controlling the update step size; φ current This represents the intent feature vector of the user's interaction behavior at the current moment; This is the transpose of the intent vector.

[0199] The exploration rate parameter of the reinforcement learning strategy is dynamically adjusted using the following formula:

[0200]

[0201] Where K = 100 is the decay coefficient; when a user adopts a recommendation (δ = 1), the exploration rate decreases to utilize existing knowledge; ∈ new The adjusted exploration rate; ∈ old δ represents the exploration rate before adjustment (initial value 0.2); δ represents the adoption flag; ||φ current ||2 is the L2 norm of the intention vector; exp(·) is the natural exponential function.

[0202] The Q table update rules have been expanded to:

[0203]

[0204] Where Q(s, a) is the policy score for action a in state s; η Q Q represents the learning rate; γ is the discount factor, which measures the importance of future rewards. The maximum expected reward for the next state s′; r t For instant rewards.

[0205] New reward function:

[0206] rt =δ·||φ current ||2-0.5·Δρ;

[0207] In terms of physical implementation, the adaptive optimization module includes:

[0208] Feedback parser: Extracts key features from tuples;

[0209] Matrix update unit: Performs gradient descent calculation according to formula (2);

[0210] Strategy optimizer: Manages updates to the Q-table and exploration rate parameters.

[0211] In the embodiment, when the user adopts the recommendation (δ=1) and ||φ current ||2 = 8.5, the exploration rate is updated to: ∈ new =0.2·exp(-8.5 / 100)≈0.184;

[0212] At the same time, the scores for the corresponding state-action pairs in the Q table are increased, strengthening the effective recommendation strategy.

[0213] The big data visualization and interactive display system described below can be referred to in correspondence with the big data visualization and interactive display method described above.

[0214] Please see the appendix Figure 2 A big data visualization and interactive display system, applied to the aforementioned big data visualization and interactive display method, includes:

[0215] The data preprocessing module is used to perform sparse projection processing on the original high-dimensional data, generate low-dimensional projection results, and construct a dynamic grid index. The grid index is used to capture interactive behavior and support subsequent privacy protection processing.

[0216] The intent awareness module is used to receive dynamic grid indexes, parse the user's selected area and pen trajectory, extract area features and trajectory features, and generate intent feature vectors by weighted concatenation, and update the intent weight matrix through online gradient descent formula.

[0217] The dynamic projection module is used to receive the intent feature vector, generate a local projection matrix according to the current intent, and combine it with a pre-trained global projection matrix to dynamically generate a hybrid projection result based on the time decay coefficient.

[0218] The dynamic aggregation module is used to receive the hybrid projection results, identify the user's attention area based on the intent norm, perform recursive grid division and incremental statistical aggregation, calculate the popularity index based on user interaction behavior and adjust the privacy protection strength parameter, and output the dynamic aggregation result, wherein the dynamic aggregation result includes the focus area, grid statistics and privacy parameters.

[0219] The visualization rendering module is used to jointly render the dynamic aggregation results and the mixed projection results into a multi-view interface, and generate a recommended list of projection parameter combinations based on the current intent features and time decay factor through the Q-learning strategy.

[0220] The feedback iteration module is used to collect user feedback on recommended combinations, generate feedback tuples, parse adoption intentions and behavior adjustment signals, iteratively update the intention weight matrix and reinforcement learning strategy structure, and feed the update results back to the intention perception module and visualization rendering module to achieve system adaptive optimization and closed-loop learning.

[0221] The data preprocessing module and the intent-aware module exchange dynamic grid indexes via a distributed message queue. The dynamic grid indexes contain the following information:

[0222] Grid space coordinate range [x min ,x max ]×[y min y max ];

[0223] Statistical triples after noise injection in,, Let j be the noise mean of grid. Let the noise standard deviation of grid j be count. j is the number of original data points in grid j.

[0224] The intent-aware module and the dynamic projection module must satisfy the following real-time constraints:

[0225] The intent feature vector φ of the user's interaction behavior at the current moment current The generation delay is no more than 50 milliseconds;

[0226] The intent feature vector φ of the user's interaction behavior at the current moment current The results are transferred to the dynamic projection module via zero-copy memory sharing to ensure rapid generation and response of the hybrid projection results.

[0227] The system in this embodiment can be used to execute the above method embodiments, and its principle and technical effect are similar, so they will not be described again here.

[0228] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for interactive visualization of big data, characterized in that, Includes the following steps: S1. Perform sparse projection processing on the original high-dimensional data to generate low-dimensional projection results and construct a dynamic grid index. S2. Capture user interaction behavior based on the dynamic grid index, extract intent features and generate dynamic intent vectors; S3. Generate a local projection matrix based on the dynamic intent vector, combine it with the pre-trained global projection matrix, dynamically adjust the mixing weights through the time decay coefficient, and calculate and generate the mixed projection result. S4. Based on the hybrid projection results, identify the user's area of ​​interest, perform recursive grid division and incremental statistical aggregation, dynamically adjust the privacy protection strength, and output dynamic aggregation results, wherein the dynamic aggregation results include the focus area, grid statistics and privacy parameters; S5. Based on the dynamic aggregation results, perform multi-view association rendering, generate a projection parameter recommendation list through reinforcement learning strategy, and collect user behavior to generate feedback tuples. S6. Generate feedback tuples based on user behavior, parse the adoption intention and behavior adjustment signals in the feedback tuples, and iteratively update the intention weight matrix and reinforcement learning strategy structure. The results of the iterative update will be applied to the next round of recommendation generation to achieve adaptive optimization and closed-loop learning of the system.

2. The big data visualization and interactive display method according to claim 1, characterized in that, The steps in S1 include: S1-1. Construct a sparse random global projection matrix based on the original high-dimensional data. Calculate global skeleton projection ,in, The original high-dimensional data points; S1-2, Based on global skeleton projection Initialize the dynamic mesh tree, divide it into coarse-grained grids, and store the statistics. ; S1-3, Regarding the statistical measure Injecting Laplace noise generates a privacy-preserving dynamic grid index.

3. The big data visualization and interactive display method according to claim 1, characterized in that, Step S2 includes: S2-1. Analyze the center coordinates of the user's selected area. and area Generate regional feature vectors ; S2-2. Perform a Fourier transform on the brush trajectory to extract the previous... Each low-frequency component generates a feature vector. ,in, The number of sampling points for the brush trajectory, and ≤10, when Take at 20:00 = ; S2.3, the region feature vector With trajectory feature vector Weighted concatenation is performed to generate the intent feature vector of the user's interaction behavior at the current moment. ,in, These are dynamic weighting coefficients; S2-4. Intent feature vector based on the user interaction behavior at the current moment. The intention weight matrix is ​​updated using the online gradient descent formula. The online gradient descent formula is: ； in, and The intention weight matrix at time step and The value of , For learning rate, For gradient operators, This is the intent feature vector of the user's interaction behavior at the current moment. For loss function, The sliding window mean of the historical intent feature vector. ,in, For window length, For time variables, In time step The intent feature vector of the user's interaction behavior at the current moment.

4. The big data visualization and interactive display method according to claim 1, characterized in that, The steps in S3 include: S3-1, Generate the local projection matrix: based on the intent feature vector of the user's interaction behavior at the current moment. Generate local projection matrix ; S3-2, Obtain the global projection matrix: Load the pre-trained global projection matrix from the data preprocessing module. The pre-trained global projection matrix It was obtained by training historical data offline using a sparse autoencoder; S3-3. Calculate the mixed projection weights: based on the time decay coefficient. Dynamically adjust the local projection intensity, where, The initial weights for the local projection. For decay rate parameters, Time variable; S3-4. Hybrid Projection Calculation: Based on the local projection matrix and the global projection matrix, the hybrid projection result is calculated using the hybrid projection calculation formula, which is as follows: ； in, This is the time decay coefficient; Original high-dimensional data points The final projection result; The original high-dimensional data points; This is the global projection matrix; It is a local projection matrix, which depends on the original high-dimensional data points. Or it can be dynamically generated based on user intent.

5. The big data visualization and interactive display method according to claim 1, characterized in that, The steps in S4 include: S4-1, Based on the norm of the current intent vector Adaptive setting of mesh granularity By combining the results of hybrid projection, the initial set of grid cells with significant user attention is identified. ; S4-2, Initial Mesh Cell Set Each grid cell in the process undergoes recursive subdivision, and the local statistics of the sensitive features of each grid cell are incrementally updated. Construct a gridded feature representation structure for differential privacy mechanisms. ; S4-3. Calculate the popularity index of each attention grid unit based on user interaction behavior. It also dynamically adjusts the privacy protection strength parameters based on the behavior-driven privacy adjustment formula. The behavior-driven privacy regulation formula is as follows: ； in, This is the heat amplification factor. Basic privacy protection strength; S4-4, Subsequently, using local statistics Based on the adjusted intensity parameters, a privacy protection objective function is constructed, and the optimal privacy budget is solved using the Lagrange optimization method. Finally, the privacy strength adjustment amount is calculated. To form a privacy regulation structure As part of the output of the dynamic aggregation result, the expression of the privacy-preserving objective function is as follows: ； in, Optimize the function for the objective. The user experience cost or data availability cost associated with using higher privacy strengths, The coefficient in the loss function that balances privacy costs and accuracy. For the loss of data accuracy caused by privacy disturbances, Set the preset privacy budget parameters.

6. The big data visualization and interactive display method according to claim 1, characterized in that, The steps in S5 include: S5-1, Visualization Rendering: Encode the dynamic aggregation results output in step S4 into a heatmap, and perform multi-view association rendering with the hybrid projection scatter plot generated in step S3. S5-2, Strategy Recommendation: Based on the intent feature vector of the user's interaction behavior at the current moment in step S2 and the time decay coefficient in step S3. Projection parameter combinations are generated using a Q-learning strategy. The recommended list includes a synchronously defined adoption flag, among which... This is the local projection weight adjustment amount. This is the adjustment amount for the time decay rate. This refers to the adjustment amount for trajectory feature weights; S5-3. Feedback Generation: Collection and Adoption Identifiers The privacy protection strength adjustment behavior in step S4, combined with the norm of the current intent vector. , generate containing The feedback tuple, where, This is the adjustment amount for the local projection matrix. For the optimized privacy budget, This is the adjustment amount for privacy intensity.

7. The big data visualization and interactive display method according to claim 1, characterized in that, Step S6 includes: S6-1, Feedback Parsing: Parsing Feedback Tuples Explicit rating values, number of undo operations, privacy budget parameters, and intent intensity indicators are extracted to construct a feedback sample set for incremental model updates. To adopt the identifier, This is the adjustment amount for the local projection matrix. For the optimized privacy budget, Let the norm of the current intent vector be . This is the adjustment amount for privacy intensity; S6-2, Intent Model Update: Based on feedback samples, the intent weight matrix of the user's interaction behavior at the current moment is updated using an online gradient descent formula. Perform incremental updates to adjust the mapping capability of the intent weight matrix to the intent feature vector, and feed the update results back to step S2 for the next round of intent feature generation; S6-3, Reinforcement Learning Strategy Structure Optimization: Update the exploration rate parameter in the Q-learning strategy based on the adoption flag and privacy adjustment amount in the feedback. The strategy scores for each combination of projection parameters in the recommended strategy are adjusted, and the updated strategy structure will be used to generate the recommended list of projection parameter combinations in step S5-2.

8. A big data visualization and interactive display system, characterized in that, The big data visualization and interactive display method applied to any one of claims 1-7 includes: The data preprocessing module is used to perform sparse projection processing on the original high-dimensional data, generate low-dimensional projection results, and construct a dynamic grid index. The grid index is used to capture interactive behavior and support subsequent privacy protection processing. The intent awareness module is used to receive dynamic grid indexes, parse the user's selected area and pen trajectory, extract area features and trajectory features, and generate intent feature vectors by weighted concatenation, and update the intent weight matrix through online gradient descent formula. The dynamic projection module is used to receive the intent feature vector, generate a local projection matrix according to the current intent, and combine it with a pre-trained global projection matrix to dynamically generate a hybrid projection result based on the time decay coefficient. The dynamic aggregation module is used to receive the hybrid projection results, identify the user's attention area based on the intent norm, perform recursive grid division and incremental statistical aggregation, calculate the popularity index based on user interaction behavior and adjust the privacy protection strength parameter, and output the dynamic aggregation result, wherein the dynamic aggregation result includes the focus area, grid statistics and privacy parameters. The visualization rendering module is used to jointly render the dynamic aggregation results and the mixed projection results into a multi-view interface, and generate a recommended list of projection parameter combinations based on the current intent features and time decay factor through the Q-learning strategy. The feedback iteration module is used to collect user feedback on recommended combinations, generate feedback tuples, parse adoption intentions and behavior adjustment signals, iteratively update the intention weight matrix and reinforcement learning strategy structure, and feed the update results back to the intention perception module and visualization rendering module to achieve system adaptive optimization and closed-loop learning.

9. A big data visualization and interactive display system according to claim 8, characterized in that, The data preprocessing module and the intent-aware module exchange dynamic grid indexes via a distributed message queue. These dynamic grid indexes contain the following information: Grid space coordinate range ,in, The minimum value on the X-axis within the grid space coordinate range. The maximum value of the X-axis within the grid space coordinate range. The minimum value on the Y-axis within the grid space coordinate range. This represents the maximum Y-axis value within the grid space coordinate range. Statistical triples after noise injection ,in, For grid The noise-added mean, For grid The standard deviation of the noise addition, For grid The number of original data points in the dataset.

10. A big data visualization and interactive display system according to claim 8, characterized in that, The intent perception module and the dynamic projection module satisfy the following real-time constraints: Intent feature vector of user interaction behavior at the current moment The generation delay is no more than 50 milliseconds; Intent feature vector of user interaction behavior at the current moment The results are transferred to the dynamic projection module via zero-copy memory sharing to ensure rapid generation and response of the hybrid projection results.

Images