Information consultation service system and method based on big data

Abstract

The invention discloses an information consultation service system and method based on big data, and relates to the technical field of big data analysis, and the method comprises the steps: executing time-space registration and coordinate transformation on a consultation request data set, obtaining structured consultation features, carrying out semantic extension and attribute mapping on the structured consultation features by using a cosine similarity matching method, and obtaining a consultation request data set; outputting spatio-temporal index query parameters, performing spatio-temporal clustering and timeliness attenuation calculation on the spatio-temporal index query parameters by using a tense knowledge graph engine, generating a policy knowledge graph, performing sliding window filtering on the policy knowledge graph, obtaining an effective policy data set, and inputting the effective policy data set into a policy consultation aggregation model. And the association integration layer carries out feature recombination and neighborhood semantic fusion. According to the method, the time-space registration and coordinate conversion are carried out on the consultation request data set, a precise data basis is provided for information consultation service, policy dynamic updating and semantic fusion are carried out by adopting the policy consultation aggregation model, and the pertinence and effectiveness of the consultation service are improved.

Description

Technical Field

[0001] This invention relates to the field of big data analytics, and in particular to an information consulting service system and method based on big data. Background Technology

[0002] With the development of information technology, big data analytics has gradually become one of the core technologies in the information consulting service field. By mining and analyzing massive amounts of data, more accurate and personalized services can be provided to users. In the information consulting service field, the application of big data technology has not only improved the efficiency of information acquisition but also enhanced the accuracy of decision support. In recent years, with the advancement of artificial intelligence and machine learning technologies, information consulting service methods based on big data have been continuously innovated, achieving a deeper understanding and response to user needs.

[0003] Despite significant progress in improving information consulting services, existing technologies still have some shortcomings. First, traditional information consulting services often lack the ability to integrate data across time and space, making it difficult to effectively capture the temporal and spatial characteristics of consulting requests, resulting in incomplete and inaccurate consulting advice. Second, existing information consulting methods lack a robust timeliness update mechanism when dealing with dynamically changing information (such as policy changes), failing to reflect the latest policy environment in a timely manner, thus affecting the quality and practicality of consulting services. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides an information consulting service method based on big data to solve the problems of insufficient spatiotemporal data integration capabilities and low timeliness of policy information in traditional consulting methods.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides an information consulting service method based on big data, which includes performing spatiotemporal registration and coordinate transformation on a consulting request dataset, obtaining structured consulting features, using cosine similarity matching to perform semantic expansion and attribute mapping on the structured consulting features, and outputting spatiotemporal index query parameters; The temporal knowledge graph engine is used to perform spatiotemporal clustering and timeliness decay calculation on the spatiotemporal index query parameters to generate a policy knowledge graph. Sliding window filtering is then performed on the policy knowledge graph to obtain effective policy datasets. The effective policy dataset is input into the policy consultation aggregation model. The association and integration layer performs feature reorganization and neighborhood semantic fusion, and the attention weighting layer performs feature weighting and context association, outputting a consultation result vector. The confidence level of the consultation result vector is quantified using the Platt scaling method to generate a confidence verification matrix. The confidence verification matrix is ​​then integrated to form an information consultation report.

[0007] As a preferred embodiment of the information consulting service method based on big data described in this invention, the consulting request dataset includes consulting initiator data, consulting event data, and publicly available government policy information.

[0008] As a preferred embodiment of the big data-based information consulting service method of the present invention, the specific operation steps for outputting the federated query parameters are as follows: Spatiotemporal registration is performed on the consultation request dataset to form a spatiotemporally aligned point cloud; coordinate transformation is performed on the spatiotemporally aligned point cloud to generate structured consultation features; The cosine similarity matching method is used to compare the similarity of structured consultation features to form a set of highly relevant matching points; the highly relevant matching point set is semantically expanded to generate an enhanced semantic vector. Perform multi-dimensional attribute mapping on the enhanced semantic vector and output spatiotemporal index query parameters.

[0009] As a preferred embodiment of the big data-based information consulting service method of the present invention, the specific steps for generating the policy knowledge graph are as follows: A temporal knowledge graph engine is applied to perform spatiotemporal clustering on the spatiotemporal index query parameters to generate nodes of the topology; simultaneously, the timeliness decay of the spatiotemporal index query parameters is calculated to output the edges of the topology. Dynamically assign and weight the nodes and edges of the topology to generate a policy knowledge graph.

[0010] As a preferred embodiment of the big data-based information consulting service method of the present invention, the specific steps for obtaining the effective policy dataset are as follows: The policy knowledge graph is subjected to topological clustering to form a policy cluster subgraph; a sliding window filter is applied to smooth the policy cluster subgraph for timeliness to obtain a weighted policy data stream. Semantic alignment and cross-source integration of weighted policy data streams are performed to obtain effective policy datasets.

[0011] As a preferred embodiment of the big data-based information consulting service method of the present invention, the specific operation steps for outputting the consulting result vector are as follows: A policy consultation aggregation model is constructed by building an association integration layer and an attention weighting layer, and by using multi-order residual connections to cascade feature layers. The effective policy dataset is input into the policy consultation aggregation model. The association and integration layer uses separable convolution to perform feature recombination on the effective policy dataset to obtain local feature vectors. Graph convolution is applied to perform neighborhood semantic fusion on local feature vectors to obtain a fused feature tensor. Nonlinear transformation is then performed on the fused feature tensor to generate a policy semantic vector. The attention weighting layer uses a multi-head attention mechanism to perform feature weighting and context association on the policy semantic vector to obtain the policy context vector; A gated loop unit is used to perform weighted fusion and dimensional compression of the policy semantic vector and the policy context vector, and output the consultation result vector.

[0012] As a preferred embodiment of the big data-based information consulting service method of the present invention, the specific steps for generating the information consulting report are as follows: Perform parameterized probability mapping on the consultation result vector to form a fused confidence vector; A dynamic decay factor is applied to the fused confidence vector to perform time-sensitive decay, generating a confidence verification matrix; the confidence verification matrix is ​​then weighted and aggregated to form an information consultation report.

[0013] Secondly, this invention provides an information consulting service system based on big data, comprising: The registration and mapping module is used to perform spatiotemporal registration and coordinate transformation on the consultation request dataset, obtain structured consultation features, use cosine similarity matching to perform semantic expansion and attribute mapping on the structured consultation features, and output spatiotemporal index query parameters. The graph construction module is used to perform spatiotemporal clustering and timeliness decay calculation on the spatiotemporal index query parameters using the temporal knowledge graph engine, generate a policy knowledge graph, and perform sliding window filtering on the policy knowledge graph to obtain effective policy datasets. The aggregation analysis module is used to input effective policy datasets into the policy consultation aggregation model. The association integration layer performs feature recombination and neighborhood semantic fusion, and the attention weighting layer performs feature weighting and context association, outputting a consultation result vector. The confidence quantification module is used to perform confidence quantification on the consultation result vector using the Platt scaling method, generate a confidence verification result matrix, and integrate the confidence verification result matrix to form an information consultation report.

[0014] Thirdly, the present invention provides a computer device including a memory and a processor, wherein the memory stores a computer program, wherein when the computer program is executed by the processor, it implements any step of the information consulting service method based on big data as described in the first aspect of the present invention.

[0015] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the information consulting service method based on big data as described in the first aspect of the present invention.

[0016] The beneficial effects of this invention are as follows: by performing spatiotemporal registration and coordinate transformation on the consultation request dataset, efficient structured processing of the consultation request dataset is achieved, providing an accurate data foundation for information consultation services; by using policy knowledge graphs and policy consultation aggregation models for dynamic policy updates and semantic fusion, the latest policy dynamics can be accurately reflected, improving the pertinence and effectiveness of consultation services. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart of a big data-based information consulting service method.

[0019] Figure 2 This is a schematic diagram of an information consulting service system based on big data.

[0020] Figure 3 A flowchart for constructing a policy knowledge graph.

[0021] Figure 4 A flowchart for generating the consultation result vector. Detailed Implementation

[0022] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0023] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0024] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0025] Reference Figures 1-4 This is one embodiment of the present invention, which provides an information consulting service method based on big data, including the following steps: S1. Perform spatiotemporal registration and coordinate transformation on the consultation request dataset to obtain structured consultation features. Use cosine similarity matching to perform semantic expansion and attribute mapping on the structured consultation features, and output spatiotemporal index query parameters.

[0026] Specifically, the operations include the following: S1.1 Collect consultation request dataset, which includes consultation initiator data, consultation event data, and publicly available government policy information.

[0027] The data for the initiator of the consultation includes identity information, historical consultation records, and industry information; identity information should be collected through the user terminal's identity authentication unit; historical consultation records should be obtained through the consultation service database; and industry information should be collected through the enterprise information interface. The data for the consultation event includes event description text, occurrence time, and geographic location information. The event description text is collected using a natural language processing interface, the occurrence time is collected through the timestamp service of the user terminal, and the geographic location information is obtained using a map API. The government discloses policy information including policy text documents, information on issuing agencies, and the policy's effective date. Policy text documents are collected using government data crawling units, information on issuing agencies is collected through policy metadata interfaces, and the policy's effective date is collected using official release time logs. It should be noted that all of the above information has been obtained with the user's consent and is used for legitimate purposes; Consultation request datasets can not only achieve efficient integration and structured expression of multi-source heterogeneous information through big data processing, but also provide high-quality data support for subsequent policy knowledge graph construction and intelligent consultation analysis.

[0028] S1.2. Preprocessing the consultation request dataset: In the data cleaning stage, median imputation is used to fill in missing values ​​to ensure data integrity. Outlier removal is performed using binning smoothing to eliminate the influence of extreme values. Simultaneously, exponentially weighted moving average is used for time-series smoothing to remove noise interference. In the data transformation stage, Z-score standardization is applied to adjust and normalize the distribution of the consultation request dataset to eliminate dimensional differences. Gaussian projection is used for coordinate system transformation to unify the coordinate system, and UTC time parsing is used for timestamp alignment to eliminate time zone differences. In the data integration stage, time-series alignment is performed on the consultation request dataset to eliminate the influence of time offsets. Principal component analysis is applied for dimensionality compression to reduce data complexity, outputting the preprocessed consultation request dataset.

[0029] S1.3 Perform spatiotemporal registration and coordinate transformation on the consultation request dataset to generate structured consultation features. Specifically, the ICP algorithm is used to perform initial alignment and iterative optimization on the consultation request dataset to obtain preliminary registration results; voxel grid filtering is used to downsample the preliminary registration results to obtain simplified point cloud data; K-nearest neighbor search is used to perform nearest neighbor matching on the simplified point cloud data to obtain matching point pairs; the differences between the matching point pairs are compared and parameters are optimized to obtain the optimal transformation parameters; based on the optimal transformation parameters, the preliminary registration results are resampled to output the spatiotemporally aligned point cloud. Latitude and longitude are extracted from the spatiotemporally aligned point cloud using coordinate analysis to obtain the original coordinates. Helmert transformation is then applied to rotate and translate the original coordinates to the WGS84 standard coordinate system, outputting standardized spatial coordinate data. GPST time synchronization is used to perform UTC conversion and time zone calibration on the standardized spatial coordinate data to obtain spatiotemporally unified features. Field merging and format standardization of the spatiotemporally unified features are then performed to form structured information features.

[0030] S1.4. The cosine similarity matching method is used to perform similarity comparison and semantic expansion on the structured consultation features, generating enhanced semantic vectors. Specifically, a Transformer encoder performs contextual encoding on the structured consultation features, outputting a dynamic semantic representation; Word2Vec word embeddings are used to supplement the dynamic semantic representation with static features, generating a hybrid semantic representation; dimensional unification is performed on the hybrid semantic representation to obtain a high-dimensional semantic vector representation; cosine similarity matching is used to perform similarity comparison on the high-dimensional semantic vector representation, forming candidate matching pairs; a Sigmoid activation function is applied to perform nonlinear transformation on the candidate matching pairs, generating enhanced matching pairs; density clustering is performed on the enhanced matching pairs to obtain candidate matching regions; simultaneously, boundary point extraction is performed on the candidate matching regions, outputting a set of highly relevant matching points. Term recognition is performed on the set of highly relevant matching points using an entity linking tool (such as ELMo) to obtain a candidate entity set. Concept mapping is then performed on the candidate entity set to generate basic semantic units. TF-IDF weighting is used to semantically expand the basic semantic units to ensure broad vocabulary coverage, resulting in an expanded semantic set. The expanded semantic set is then matrix-transformed to output an expanded semantic matrix. Feature aggregation is performed on the expanded semantic matrix to obtain a low-dimensional dense vector. Simultaneously, hierarchical sampling is used to randomly sample the low-dimensional dense vector to generate a semantic embedding vector. The semantic embedding vector is then fused with structured consultation features through feature selection and weighting, and Min-Max normalization is used for normalization to ensure numerical stability, outputting an enhanced semantic vector.

[0031] S1.5. Use the Z-ORDER encoding function to perform multi-dimensional attribute mapping on the enhanced semantic vector and output the spatiotemporal index query parameters. In the specific operation, principal component analysis is used to decouple the dimensions of the enhanced semantic vector and separate the enhanced semantic vector into independent dimension vectors. The equal frequency binning method is applied to bin the independent dimension vectors to prevent uneven distribution caused by data skew and output the binned dimension vectors. Next, the Z-ORDER encoding function is used to perform multi-dimensional attribute mapping on the binned dimension vectors: bit-interleaving is applied to normalize and scale the binned dimension vectors, obtaining 32-bit floating-point numbers for each dimension, and converting each 32-bit floating-point number to binary according to the IEEE 754 floating-point standard, outputting a binary bit sequence; the Z-ORDER encoding function is used to perform bit-by-bit interleaving on the binary bit sequence to generate continuous hybrid encoding; then, the continuous hybrid encoding is discretized into grid cell IDs through spatial grid partitioning, and hash mapping is used to perform index organization and storage optimization on the grid cell IDs to obtain an inverted index; parameter compression and query acceleration are performed on the inverted index to generate spatiotemporal index query parameters; It should be noted that bit rotation interleaving refers to the process of cyclically and alternately inserting binary bit sequences into the encoding according to dimensional order.

[0032] S2. Use the temporal knowledge graph engine to perform spatiotemporal clustering and timeliness decay calculation on the spatiotemporal index query parameters to generate a policy knowledge graph. Perform sliding window filtering on the policy knowledge graph to obtain a valid policy dataset.

[0033] Specifically, the operations include the following: S2.1. The temporal knowledge graph engine is used to perform spatiotemporal clustering on the spatiotemporal index query parameters to generate nodes with topological structures. Specifically, the spatiotemporal index query parameters are input to the temporal knowledge graph engine through a Kafka data stream pipeline. The spatiotemporal index query parameters are sliced ​​into equally spaced spatiotemporal joint windows using a sliding window sampler, and the spatiotemporal slice dataset is output. Then, data alignment and similarity measurement are performed on the spatiotemporal slice dataset to obtain the spatiotemporal sequence similarity. Gaussian kernel density estimation is applied to optimize the bandwidth and fit the density of the spatiotemporal sequence similarity to obtain the probability density distribution. A nonlinear transformation is performed on the probability density distribution to obtain the spatiotemporal similarity matrix. The spatiotemporal similarity matrix is ​​decomposed into features in two dimensions. Further, in the spatial dimension, an R-tree index is used to perform range queries and nearest neighbor searches on the spatiotemporal similarity matrix to form spatial clusters. Simultaneously, coordinate localization is performed on these spatial clusters to obtain their centroid coordinates. In the temporal dimension, Fourier transform is used to perform spectral decomposition on the spatiotemporal similarity matrix to extract temporal periodic features. Parameter fitting is then performed on these temporal periodic features to obtain temporal periodic parameters. Finally, tensor concatenation of the spatial cluster centroid coordinates and the temporal periodic parameters yields the spatiotemporal joint features. Then, the temporal knowledge graph engine is used to perform density peak detection on the spatiotemporal joint features to obtain candidate cluster centers. Spatiotemporal clustering is then performed on the candidate cluster centers to form spatiotemporal clustering results. Mean aggregation is then performed on the spatiotemporal clustering results to generate a candidate node set. The LOF local outlier factor is applied to filter outliers in the candidate node set to eliminate isolated nodes and output the nodes of the topological structure. It should be noted that the LOF local outlier factor is defined based on the k-nearest neighbor distance ratio of the candidate node set, and its value ranges from [0.5, 3.0].

[0034] S2.2 Perform timeliness decay calculation on the spatiotemporal index query parameters and output the edges of the topology. Specifically, the timestamps of the spatiotemporal index query parameters are extracted using a timestamp parser. The timestamps are then weighted based on an exponential decay function to obtain time weight coefficients and spatial weight coefficients. Based on these coefficients, spatial distance is calculated on the spatiotemporal index query parameters using the Haversine formula. After obtaining the spatial distance, timeliness decay calculation is performed to generate a spatiotemporal relevance score. The specific mathematical formula is as follows: ; in, Indicates the spatiotemporal correlation score. Indicates the time weighting coefficient. Indicates the time decay coefficient. Indicates time interval, Indicates spatial distance. Represents the spatial scale coefficient. Indicates the spatial decay index, Indicates the spatial weighting coefficient; It should be noted that the time decay coefficient is defined based on the timeliness requirements of specific business scenarios, and its value range is [0.01, 0.1]; the spatial scale coefficient is defined based on the percentile distribution density of spatial distance, and its value range is [0.1, 5.0]; the spatial decay index is defined based on the distance decay characteristics of spatial distance, and its value range is [0.5, 2.0]; Based on the spatiotemporal correlation score, time-related weights are assigned to the nodes of the topology to obtain spatiotemporally related points. A sliding time window is applied to count the frequency of occurrence of spatiotemporally related points. When two spatiotemporally related points occur simultaneously more than the frequency threshold within the window period, they are defined as a pair of related nodes. Weights are assigned and edges are connected to the pairs of related nodes to obtain candidate edges. Bilateral filtering is used to smooth the weights of the candidate edges to generate the edges of the topology. It should be noted that the frequency threshold is based on the definition of spatiotemporal co-occurrence density of spatiotemporal related points, and the value range is [2,5].

[0035] S2.3. Dynamically assign and weight-fuse the nodes and edges of the topology to generate a policy knowledge graph. Specifically, the PageRank algorithm is used to rank the nodes by importance and adjust their time-sensitive weights to obtain the node weight distribution. The node weight distribution is then Gaussian normalized to generate a node feature matrix. Exponential moving average is applied to the edges of the topology for local weighted averaging and noise filtering to obtain optimized edge weights. The optimized edge weights are then probabilistically transformed to form an edge feature matrix. Principal component weighting is used to dynamically assign weights to the node feature matrix and edge feature matrix to form a weighted feature tensor. Canonical correlation analysis (CCA) is applied to the weighted feature tensor for feature aggregation and linear projection to obtain the fused graph representation. The fused graph representation is then weighted and fused to output the policy knowledge graph.

[0036] S2.4. Perform topological clustering on the policy knowledge graph to form policy cluster subgraphs; apply sliding window filtering to smooth the policy cluster subgraphs for timeliness and obtain weighted policy data streams. Specifically, depth-first search is used to traverse the policy knowledge graph and extract relationships to obtain semantic association strength and spatiotemporal co-occurrence frequency; principal component analysis is used to compress the dimensions and linearly combine the semantic association strength and spatiotemporal co-occurrence frequency to obtain principal component weights; matrix filling is performed on the principal component weights to construct a weighted adjacency matrix; hierarchical clustering is used to cluster nodes and aggregate relationships in the weighted adjacency matrix to generate preliminary community divisions; boundary adjustments and density optimization are performed on the preliminary community divisions to output policy cluster subgraphs, each containing a connected component consisting of a central node and its associated nodes; Then, the sliding window filtering method is applied to smooth the policy cluster subgraph for timeliness: the decay window is set to a fixed period (e.g., 90 days), and the policy cluster subgraph is divided into time-dimension segments according to the decay window to obtain a time-series subgraph sequence; the time-series subgraph sequence is subjected to polynomial fitting and exponential smoothing to form a smooth feature trajectory; the smooth feature trajectory is resampled at equal intervals to generate a weighted policy data stream.

[0037] S2.5. Semantic alignment and cross-source integration are performed on the weighted policy data stream to obtain a valid policy dataset. Specifically, the Latent Semantic Index (LSI) is used to decompose the weighted policy data stream into dimensionality and extract deep semantic vectors. At the same time, dynamic time warping is used to perform temporal alignment and feature warping on the deep semantic vectors, outputting a spatiotemporal-semantic joint vector. Then, based on the ISO 19115 information element protocol, semantic alignment and entity parsing are performed on the spatiotemporal-semantic joint vector to generate standardized policy entities. The priority superposition method is used to perform cross-source integration on the standardized policy entities to obtain a unified policy record. Finally, density detection is used to remove duplicate and outlier data in the unified policy record, outputting a valid policy dataset.

[0038] S3. Input the effective policy dataset into the policy consultation aggregation model. The association and integration layer performs feature reorganization and neighborhood semantic fusion. The attention weighting layer performs feature weighting and context association, and outputs the consultation result vector.

[0039] Specifically, the operations include the following: S3.1 Construct and train a policy consultation aggregation model. Specifically, within the TensorFlow framework, a graph neural network is called via Keras API parameters, and separable convolutions and graph convolutions are embedded into the graph neural network. The kernel size of the graph neural network is set to 3×3, and the stride is set to 1. A batch normalization layer is added to the graph neural network for feature standardization, and a Dropout layer is used for random deactivation to improve generalization ability, completing the construction of the association and integration layer. The Transformer architecture is called through the TensorFlow underlying interface, and a multi-head attention mechanism is used for feature association learning to capture cross-dimensional dependencies. The number of attention heads in the Transformer architecture is set to 8, the key dimension to 64, and the hidden layer dimension to 256. A gated recurrent unit is added after the Transformer architecture for temporal feature extraction, and gradient stabilization is performed through layer normalization, completing the construction of the attention weighted layer. Multi-order residual connections are used to perform cross-layer feature fusion on the association integration layer and the attention weighting layer to obtain multi-scale feature representations. Global average pooling is then applied to the multi-scale feature representations to generate aggregated feature vectors. A fully connected layer is used to perform nonlinear transformation on the generated aggregated feature vectors to obtain higher-order semantic features. The higher-order semantic features are then subjected to probability transformation using the Softmax function to generate classification weights. Based on the classification weights, the association integration layer and the attention weighting layer are concatenated to complete the construction of the policy consultation aggregation model. Next, the policy consultation aggregation model is trained. Further, the effective policy dataset is divided into a sample set, a training set, and a validation set. On the sample set, the sample data is randomly perturbed and numerically standardized using feature scaling to form enhanced training samples. On the training set, the Adam optimizer is used to update the parameters of the enhanced training samples, and a learning rate scheduler is applied simultaneously to dynamically adjust the learning rate and obtain the training loss value. On the validation set, the training loss value is monitored for early stopping to obtain the validation accuracy. When the validation accuracy exceeds the convergence threshold for five consecutive rounds, training terminates, and the trained policy consultation aggregation model is output simultaneously. It should be noted that the convergence threshold is defined based on the relative fluctuation of the validation accuracy, and its value ranges from [0.01, 0.05].

[0040] S3.2 The association and integration layer performs feature recombination and neighborhood semantic fusion to generate policy semantic vectors. Specifically, the effective policy dataset is input into the policy consultation aggregation model via the Input interface. The association and integration layer uses a 5×5 standard convolution to convert the effective policy dataset into a gridded feature representation, and performs boundary expansion using zero padding to maintain the feature map size, outputting a padded feature map. A 3×3 separable convolution is used to extract deep features from the padded feature map in the input channels, obtaining channel feature maps. Then, a 1×1 pointwise convolution is used to recombine features and transform the dimensions of the channel feature maps, achieving cross-channel information fusion and outputting a fused feature map. Next, a batch normalization layer standardizes the fused feature map, simultaneously applying a scaling factor to perform a linear transformation, obtaining a standardized output. Finally, the LeakyReLU activation function is used to perform non-linear activation on the standardized output, and a Dropout layer is used to randomly discard some neurons with a probability of 0.3 to prevent overfitting, outputting a local feature vector. It should be noted that the scaling factor is defined based on the variance distribution of the standardized output, and its value ranges from [0.8, 1.2]. A two-layer graph convolutional architecture is used to perform neighborhood semantic fusion on local feature vectors: The first layer performs feature propagation and linear combination on the local feature vectors, outputting intermediate feature representations. A Dropout layer is applied to randomly deactivate the intermediate feature representations to form regularized features. The second layer performs high-order neighborhood aggregation on the regularized features to obtain aggregated features, and performs nonlinear mapping on the aggregated features to generate high-order feature representations. The ReLU activation function is applied to selectively activate the high-order feature representations to retain negative information, resulting in a fused feature tensor. The fused feature tensor is normalized using LayerNorm (layer normalization), and then the GELU activation function is applied to achieve a smooth nonlinear transformation to obtain a standardized semantic representation. Finally, attention pooling is used to perform weighted concatenation on the standardized semantic representations to generate a policy semantic vector.

[0041] S3.3 The attention weighting layer performs feature weighting and context association to obtain the policy context vector. Specifically, Xavier normal initialization is used to assign weights to the policy semantic vector, and batch normalization (BatchNorm) is used simultaneously for distribution alignment to obtain a standardized feature representation. Multi-head attention computation is performed on the standardized feature representation to achieve context association. Furthermore, the standardized feature representation is split into three matrices—query (Q), key (K), and value (V)—through linear projection. These three matrices are then evenly divided into eight attention heads, and each attention head is independently weighted based on similarity. Simultaneously, [the following is applied:] The function performs numerical transformations to generate scaled dot product attention values. The specific mathematical formula is as follows: ; in, Indicates the attention head index, Indicates the first The scaled dot product attention value of each attention head. Indicates the first A query matrix with attention heads Indicates the first The key matrix of each attention head. This indicates the transpose operation. Indicates the scaling factor. Indicates the first A matrix of values ​​for each attention head; It should be noted that the scaling factor is defined based on the dimension of the value matrix and takes values ​​in the range [8, 64].

[0042] The scaled dot product attention values ​​of the eight attention heads are concatenated to generate multi-head attention values. The GELU activation function is used to perform a nonlinear transformation on the multi-head attention values ​​to obtain activation features. Residual connections are applied to perform feature enhancement on the activation features, and layer normalization is applied to standardize the distribution to generate context-aware features. The context-aware features are then expanded and compressed through a fully connected layer to output a policy context vector.

[0043] S3.4. A gated recurrent unit (GRU) is used to perform weighted fusion and dimensional compression on the policy semantic vector and the policy context vector, outputting a consultation result vector. Specifically, the policy semantic vector and the policy context vector are concatenated according to the channel dimension to form a 512-dimensional fusion input. The 512-dimensional fusion input is then weighted and fused using temporal features through a two-layer gated recurrent unit (GRU). The update gate of the first-layer GRU performs feature selection on the 512-dimensional fusion input and outputs the primary hidden state. The reset gate uses Xavier normal initialization to filter information and perform nonlinear transformation on the primary hidden state to obtain candidate hidden states. The update gate of the second-layer GRU performs dimensional compression and feature refinement on the candidate hidden states to form compressed hidden states. The reset gate performs gating adjustment and feature recombination on the compressed hidden states to form higher-order temporal features. The temporal attention mechanism is used to weight the importance of high-order temporal features to obtain temporal attention weights; a fully connected layer is applied to linearly project the temporal attention weights to generate aggregated temporal features; the aggregated temporal features are then reduced in dimensionality and normalized to obtain the consultation result vector.

[0044] S4. The confidence level of the consultation result vector is quantified using the Platt scaling method to generate a confidence verification matrix. The confidence verification matrix is ​​then integrated to form an information consultation report.

[0045] Specifically, the operations include the following: S4.1. Perform parameterized probability mapping on the consultation result vector to form a fused confidence vector. Specifically, the Platt scaling method is used to perform nonlinear scaling and linear transformation on the consultation result vector to obtain the original probability score. The original probability score is then optimized to obtain adjusted parameters. Gradient descent is applied to update the weights of the adjusted parameters to obtain a weight matrix. Based on the weight matrix, bias correction and distribution fitting are performed on the adjusted parameters to obtain an optimized probability distribution. The optimized probability distribution is then calibrated and aligned using the negative log-likelihood loss function to obtain a calibration confidence score. Parameterized probability mapping is then performed on the calibration confidence score to obtain the probability output value. The specific mathematical formula is as follows: ; in, This represents the probability output value. Indicates the dynamic scaling factor. Represents the weight matrix. Indicates the bias term. Indicates the offset; It should be noted that the dynamic scaling factor is defined based on the spectral norm of the weight matrix, with a value range of [0.5, 2.5]; the bias term is defined based on the mean of the data distribution of the adjusted parameters, with a value range of [-0.5, 0.5]; and the offset is defined based on the specific requirements of probability calibration, with a value range of [-1.5, 0.5]. The probability output values ​​are numerically standardized and their ranges are limited to generate a normalized probability vector. A linear transformation is applied to the normalized probability vector to perform a feature space transformation to obtain the transformed probability. The transformed probability is then truncated with confidence to obtain the final probability distribution. Finally, the final probability distribution is weighted and summed to generate a fused confidence vector.

[0046] S4.2. Apply a dynamic decay factor to the fused confidence vector to perform time-related decay, generating a credibility verification matrix. Then, weighted aggregate the credibility verification matrix to form an information consultation report. Specifically, a timestamp parser is used to quantize the time interval and extract metadata from the fused confidence vector to obtain time decay parameters. The dynamic decay factor is applied to exponentially decay the time decay parameters to generate time-related weights. These weights are then multiplied element-wise with the fused confidence vector to obtain preliminary time-related decay results. Next, an authority correction factor is used to weight and adjust the preliminary time-related decay results, and feature scaling is performed through matrix normalization to generate a credibility verification matrix. It should be noted that the dynamic decay factor is defined based on the exponential goodness of fit of the time decay parameter, and its value range is [0.005, 0.05]; the authority correction factor is defined based on the authority grading standard of government-published policy information, and its value range is [0.5, 1.0]. The credibility verification matrix is ​​weighted and nonlinearly transformed to obtain an aggregated feature vector; key fields are extracted from the aggregated feature vector to obtain structured data; the Drools rule engine is used to prioritize and fill templates into the structured data, and an information consultation report is output.

[0047] This embodiment also provides an information consulting service system based on big data, including: The registration and mapping module is used to perform spatiotemporal registration and coordinate transformation on the consultation request dataset, obtain structured consultation features, use cosine similarity matching to perform semantic expansion and attribute mapping on the structured consultation features, and output spatiotemporal index query parameters. The graph construction module is used to perform spatiotemporal clustering and timeliness decay calculation on the spatiotemporal index query parameters using the temporal knowledge graph engine, generate a policy knowledge graph, and perform sliding window filtering on the policy knowledge graph to obtain effective policy datasets. The aggregation analysis module is used to input effective policy datasets into the policy consultation aggregation model. The association integration layer performs feature recombination and neighborhood semantic fusion, and the attention weighting layer performs feature weighting and context association, outputting a consultation result vector. The confidence quantification module is used to perform confidence quantification on the consultation result vector using the Platt scaling method, generate a confidence verification result matrix, and integrate the confidence verification result matrix to form an information consultation report.

[0048] This embodiment also provides a computer device applicable to the information consulting service method based on big data, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the information consulting service method based on big data as proposed in the above embodiment.

[0049] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0050] This embodiment also provides a storage medium storing a computer program, which, when executed by a processor, implements the information consulting service method based on big data as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0051] In summary, this invention achieves efficient structured processing of the consultation request dataset by performing spatiotemporal registration and coordinate transformation, providing a precise data foundation for information consultation services; and employs policy knowledge graphs and policy consultation aggregation models for dynamic policy updates and semantic fusion, ensuring accurate reflection of the latest policy dynamics and improving the relevance and effectiveness of consultation services.

[0052] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for providing information consulting services based on big data, characterized in that: include, Spatiotemporal registration and coordinate transformation are performed on the consultation request dataset to obtain structured consultation features. Cosine similarity matching is used to semantically expand and attribute map the structured consultation features, and spatiotemporal index query parameters are output. The temporal knowledge graph engine is used to perform spatiotemporal clustering and timeliness decay calculation on the spatiotemporal index query parameters to generate a policy knowledge graph. Sliding window filtering is then performed on the policy knowledge graph to obtain effective policy datasets. The effective policy dataset is input into the policy consultation aggregation model. The association and integration layer performs feature reorganization and neighborhood semantic fusion, and the attention weighting layer performs feature weighting and context association, outputting a consultation result vector. The confidence level of the consultation result vector is quantified using the Platt scaling method to generate a confidence verification matrix. The confidence verification matrix is ​​then integrated to form an information consultation report.

2. The information consulting service method based on big data as described in claim 1, characterized in that: The consultation request dataset includes consultation initiator data, consultation event data, and publicly available government policy information.

3. The information consulting service method based on big data as described in claim 2, characterized in that: The specific steps for outputting the federated query parameters are as follows. Spatiotemporal registration is performed on the consultation request dataset to form a spatiotemporally aligned point cloud; coordinate transformation is performed on the spatiotemporally aligned point cloud to generate structured consultation features; The cosine similarity matching method is used to compare the similarity of structured consultation features to form a set of highly relevant matching points; the highly relevant matching point set is semantically expanded to generate an enhanced semantic vector. Perform multi-dimensional attribute mapping on the enhanced semantic vector and output spatiotemporal index query parameters.

4. The information consulting service method based on big data as described in claim 1, characterized in that: The specific steps for generating the policy knowledge graph are as follows: A temporal knowledge graph engine is applied to perform spatiotemporal clustering on the spatiotemporal index query parameters to generate nodes of the topology; simultaneously, the timeliness decay of the spatiotemporal index query parameters is calculated to output the edges of the topology. Dynamically assign and weight the nodes and edges of the topology to generate a policy knowledge graph.

5. The information consulting service method based on big data as described in claim 1, characterized in that: The specific steps for obtaining a valid policy dataset are as follows. The policy knowledge graph is subjected to topological clustering to form a policy cluster subgraph; a sliding window filter is applied to smooth the policy cluster subgraph for timeliness to obtain a weighted policy data stream. Semantic alignment and cross-source integration of weighted policy data streams are performed to obtain effective policy datasets.

6. The information consulting service method based on big data as described in claim 5, characterized in that: The specific steps for outputting the consultation result vector are as follows. A policy consultation aggregation model is constructed by building an association integration layer and an attention weighting layer, and by using multi-order residual connections to cascade feature layers. The effective policy dataset is input into the policy consultation aggregation model. The association and integration layer uses separable convolution to perform feature recombination on the effective policy dataset to obtain local feature vectors. Graph convolution is applied to perform neighborhood semantic fusion on local feature vectors to obtain a fused feature tensor. Nonlinear transformation is then performed on the fused feature tensor to generate a policy semantic vector. The attention weighting layer uses a multi-head attention mechanism to perform feature weighting and context association on the policy semantic vector to obtain the policy context vector; A gated loop unit is used to perform weighted fusion and dimensional compression of the policy semantic vector and the policy context vector, and output the consultation result vector.

7. The information consulting service method based on big data as described in claim 1, characterized in that: The specific steps for generating the information consultation report are as follows: Perform parameterized probability mapping on the consultation result vector to form a fused confidence vector; A dynamic decay factor is applied to the fused confidence vector to perform time-sensitive decay, generating a confidence verification matrix; the confidence verification matrix is ​​then weighted and aggregated to form an information consultation report.

8. A big data-based information consulting service system, based on the big data-based information consulting service method according to any one of claims 1 to 7, characterized in that: include, The registration and mapping module is used to perform spatiotemporal registration and coordinate transformation on the consultation request dataset, obtain structured consultation features, use cosine similarity matching to perform semantic expansion and attribute mapping on the structured consultation features, and output spatiotemporal index query parameters. The graph construction module is used to perform spatiotemporal clustering and timeliness decay calculation on the spatiotemporal index query parameters using the temporal knowledge graph engine, generate a policy knowledge graph, and perform sliding window filtering on the policy knowledge graph to obtain effective policy datasets. The aggregation analysis module is used to input effective policy datasets into the policy consultation aggregation model. The association integration layer performs feature recombination and neighborhood semantic fusion, and the attention weighting layer performs feature weighting and context association, outputting a consultation result vector. The confidence quantification module is used to perform confidence quantification on the consultation result vector using the Platt scaling method, generate a confidence verification result matrix, and integrate the confidence verification result matrix to form an information consultation report.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the information consulting service method based on big data as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the information consulting service method based on big data as described in any one of claims 1 to 7.

Images