Business vitality prediction and business district evaluation method based on multi-modal feature fusion

Abstract

The present invention provides a method for predicting business vitality and evaluating business districts based on multi-modal feature fusion, which relates to the technical field of spatial data analysis. The method includes obtaining business POI data and time-series trajectory data of a target area, extracting the score and sentiment features of user comments, and calculating the pedestrian flow distribution based on the time-series trajectory data. Then, these features are input into a recurrent neural network to predict the business vitality time series, and the POI data is constructed into a knowledge graph. The network accessibility matrix is constructed and optimized by combining the location information and the pedestrian flow. Finally, the Louvain algorithm is used to divide the target area, and the edge computing nodes of each sub-area fuse multi-dimensional features and dynamically adjust the evaluation index weights through reinforcement learning to output the business service evaluation results. By fusing multi-source data and deep learning methods, the present invention can more accurately and comprehensively evaluate the quality of business services in the living circle, providing decision-making support for urban planning and business development.

Description

Technical Field

[0001] The present invention relates to spatial data analysis technology, and in particular to a commercial vitality prediction and business district evaluation method based on multimodal feature fusion. Background Art

[0002] With the acceleration of urbanization, the diversity and spatial distribution of commercial services have an increasingly significant impact on the quality of life of residents. The acquisition and analysis of commercial POI (points of interest) data has become an important means of studying commercial services in the living circle. Existing technologies mainly rely on static data analysis, lacking real-time monitoring and analysis of dynamic human traffic and user behavior, resulting in insufficient accuracy and timeliness of commercial service evaluation. Defects and deficiencies in existing technologies: First, existing technologies often rely on manual review in the classification and labeling of commercial POI data, which is inefficient and prone to subjective bias, affecting the accuracy of the data. Secondly, the existing methods are mostly static in their analysis of human traffic distribution characteristics, failing to effectively capture the dynamic movement trajectory of the crowd, resulting in inaccurate predictions of commercial vitality. Finally, existing technologies lack flexibility in setting the weights of evaluation indicators, and cannot be dynamically adjusted according to real-time data changes, resulting in lags and inadaptability in the evaluation results. Summary of the invention

[0003] The embodiments of the present invention provide a commercial vitality prediction and business district evaluation method based on multimodal feature fusion, which can solve the problems in the prior art.

[0004] According to a first aspect of the embodiments of the present invention,

[0005] Provides a commercial vitality prediction and business district evaluation method based on multimodal feature fusion, including:

[0006] Acquire commercial POI data and time series trajectory data of the target area, classify and label the commercial POI data by business type, extract the rating features and sentiment polarity features of user comments, and calculate the flow distribution features based on the time series trajectory data;

[0007] The scoring features, sentiment polarity features and traffic flow distribution features are input into a recursive neural network to generate a time series of commercial vitality prediction for each region; the commercial POI data is constructed as a knowledge graph node, service complementarity is calculated based on the scoring features and sentiment polarity features, a network accessibility matrix is ​​constructed in combination with location information and traffic flow distribution features, and the network accessibility matrix is ​​optimized through service complementarity to obtain an enhanced network accessibility matrix;

[0008] Input the enhanced network reachability matrix into the Louvain community discovery algorithm to partition the target area into sub-regions, and deploy edge computing nodes in each sub-region; each edge computing node inputs the local enhanced network reachability matrix and the time series of commercial vitality prediction into a contrastive learning network to generate local commercial features, and fuses them with the scoring features and sentiment polarity features to obtain comprehensive commercial features; construct an evaluation weight adaptive layer, take the comprehensive commercial features as the state input, dynamically adjust the evaluation index weights through reinforcement learning, and output the commercial service evaluation results.

[0009] In an alternative implementation,

[0010] The steps of inputting the scoring features, sentiment polarity features, and pedestrian flow distribution features into a recurrent neural network to generate the time series of commercial vitality prediction for each region include:

[0011] Construct a feature interaction matrix, generate time series features through a bidirectional gated recurrent unit network, adopt a multi-scale decomposition strategy, and predict the trend term, periodic term, and residual term through linear projection, Fourier decomposition, and a multi-layer perceptron respectively. Finally, perform adaptive fusion based on the reliability score to obtain the time series of commercial vitality prediction, specifically including:

[0012] Perform feature enhancement on the scoring features to obtain scoring statistical features, where the scoring statistical features include mean, standard deviation, skewness, kurtosis, and scoring change rate; calculate the sentiment intensity based on word frequency and position weight to obtain enhanced sentiment features, where the enhanced sentiment features include positive sentiment intensity, negative sentiment intensity, and sentiment distribution entropy; extract time patterns based on the 24-hour pedestrian flow distribution feature of the target area to obtain pedestrian flow pattern features, where the pedestrian flow pattern features include peak values in the morning, noon, and evening, valley values in the morning, noon, and evening, the proportion of pedestrian flow on weekdays, the proportion of pedestrian flow on weekends, and the periodicity score;

[0013] Construct a feature interaction matrix with the scoring statistical features, enhanced sentiment features, and pedestrian flow pattern features. The feature interaction matrix includes independent representations of the three types of features and Hadamard product interaction representations between pairwise features; input the feature interaction matrix into a bidirectional gated recurrent unit network with residual connections to generate bidirectional time series features including forward feature sequences and backward feature sequences;

[0014] Perform multi-scale convolution on the bidirectional time series features to obtain daily scale features, weekly scale features and monthly scale features; perform linear projection based on the monthly scale features to obtain trend item prediction results, perform Fourier decomposition based on the weekly scale features to obtain period item prediction results, input the daily scale features, weekly scale features and monthly scale features into a multi-layer perceptron to obtain residual item prediction results; calculate the fluctuation amplitude and prediction error of the trend item prediction result, the period item prediction result and the residual item prediction result to obtain a reliability score, generate a fusion weight coefficient based on the reliability score, and use the fusion weight coefficient to perform weighted summation on the trend item prediction result, the period item prediction result and the residual item prediction result to obtain a business vitality prediction time series.

[0015] In an optional embodiment,

[0016] The steps of constructing the commercial POI data as knowledge graph nodes, calculating service complementarity based on rating features and sentiment polarity features, constructing a network accessibility matrix in combination with location information and pedestrian flow distribution features, and optimizing the network accessibility matrix through service complementarity to obtain an enhanced network accessibility matrix include:

[0017] The commercial POI data is constructed into basic attributes including business type, business area, and business hours, scoring features including taste score, environment score, service score, and cost-effectiveness score, emotional polarity features including positive emotional intensity, negative emotional intensity, and neutral emotional intensity, and knowledge graph nodes including temporal attributes such as average daily order volume, weekend passenger flow, and weekday passenger flow. The knowledge graph nodes are connected to establish spatial relationships based on geographical distance, complementary competitive relationships based on business type, and overlapping relationships based on business hours;

[0018] The scoring features are weighted based on the scoring dimension weights to obtain scoring complementarity, the emotional complementarity is calculated based on the vector cosine similarity of the emotional polarity features and the difference between the emotional intensity, the time complementarity is calculated based on the difference in business time distribution and the degree of peak passenger flow staggering, and the scoring complementarity, the emotional complementarity and the time complementarity are weightedly integrated to obtain service complementarity;

[0019] The crowd flow distribution characteristics between adjacent POI nodes are decomposed in time series to obtain periodic crowd flow fluctuation patterns and crowd flow change characteristics. The trajectory stability is calculated by combining the directional entropy and duration of the crowd movement trajectory. The initial network accessibility matrix is ​​constructed based on the overlap degree of the crowd flow fluctuation patterns, the correlation of the crowd flow change characteristics and the trajectory stability. The service complementarity is multiplied by the accessibility strength of the network accessibility matrix, and the result is symmetrically normalized to obtain the enhanced network accessibility matrix.

[0020] In an alternative embodiment,

[0021] The steps for each edge computing node to compare and learn the local enhanced network reachability matrix with the business vitality prediction time series input comparison learning network, generate local business features, and fuse them with the scoring features and sentiment polarity features to obtain comprehensive business features include:

[0022] Construct a two-branch contrastive learning network through localized processing with spatial constraints and spatio-temporal data evaluation with double decay: the matrix branch uses a graph convolutional network to extract spatial features, and the time series branch uses a gated recurrent unit to extract time features, and combines the attention mechanism and multi-modal feature interaction to obtain comprehensive business features, specifically including:

[0023] Localize the global enhanced network reachability matrix according to the community discovery result, extract the local matrix corresponding to the sub-region through a selection matrix with spatial constraints, set an adaptive weight coefficient based on the regional overlap degree, and perform boundary smoothing processing on the local matrix; collect the business vitality prediction time series within the sub-region to construct a multi-scale time series sliding window, and use a spatio-temporal data freshness evaluation mechanism based on double decay to evaluate and screen the local training data;

[0024] Construct a two-branch contrastive learning network with feature enhancement, where the matrix branch uses a multi-level graph convolutional network to process the local matrix, and extracts hierarchical spatial features through residual connection and attention pooling; the time series branch uses a bidirectional gated recurrent unit to process the business vitality prediction time series, and combines an adaptive frequency domain decomposition module to extract multi-scale time features; optimize the contrastive learning network through a hierarchical contrast loss function that fuses spatial topology constraints and temporal consistency constraints;

[0025] Calculate the dynamic fusion weight of the spatial features and time features based on the attention mechanism, and use a cross-scale feature aggregation module for multi-level feature fusion to obtain the initial local business features; introduce a momentum update mechanism with an adaptive temperature coefficient to perform online update on the initial local business features to obtain local business features;

[0026] Map the local business features, scoring features, and sentiment polarity features to a unified feature space through non-linear projection, and construct a multi-modal feature interaction matrix; use a hierarchical multi-head attention mechanism to model the local and global dependencies between modalities, and combine a feature recalibration module to adjust the feature distribution; obtain the initial comprehensive business features through a residual connection with channel re-calibration, and obtain the final comprehensive business features based on an adaptive weight allocation mechanism for modality importance.

[0027] In an alternative embodiment,

[0028] Steps for constructing an evaluation weight adaptive layer, taking comprehensive business features as state inputs, dynamically adjusting the weights of evaluation indicators through reinforcement learning, and outputting the evaluation results of commercial services include:

[0029] Construct an evaluation state space, combine comprehensive business features, an evaluation indicator weight vector, differential features of the predicted time series of business vitality, and an industry format collaborative development index to form a state vector, where the industry format collaborative development index is calculated based on the service complementarity and the reachability intensity of the network reachability matrix;

[0030] Design a multi-level reward function based on the state vector, including calculating the evaluation deviation using a smooth error metric with Huber loss, constructing an immediate reward function by combining a weight change penalty term with relative entropy and mixed norm constraints, where the trade-off factor of the weight change penalty term is adaptively adjusted by gradient; based on the time series change characteristics of the industry format collaborative development index, construct a time series reward function by combining the dynamic change of the industry format distribution entropy;

[0031] Use a residual network structure to generate reward combination weights, input the state vector and action vector into a residual block for feature extraction and mapping, obtain the reward weights through a softmax function, perform weighted combination on the immediate reward function and the time series reward function to obtain the total reward function, and perform time series smoothing processing based on a smoothing factor adaptively adjusted by signal variance;

[0032] Construct a dual network structure, where the value network evaluates the state-action value through a state value function and an advantage function, and the policy network generates an adjustment amount of the evaluation indicator weights through a mean network and a variance network; simultaneously optimize the parameters of the value network and the policy network based on the temporal difference error to obtain an optimized evaluation indicator weight vector;

[0033] Apply the optimized evaluation indicator weight vector to the comprehensive business features to obtain the evaluation results of commercial services, where the evaluation results of commercial services include the commercial service quality score, the industry format development balance index, and the regional collaborative development level.

[0034] In an alternative implementation,

[0035] Steps for designing a multi-level reward function based on the state vector, including calculating the evaluation deviation using a smooth error metric with Huber loss, constructing an immediate reward function by combining a weight change penalty term with relative entropy and mixed norm constraints, where the trade-off factor of the weight change penalty term is adaptively adjusted by gradient; based on the time series change characteristics of the industry format collaborative development index, construct a time series reward function by combining the dynamic change of the industry format distribution entropy include:

[0036] Construct a multi-dimensional evaluation state vector for the state vector, including the evaluation score and its derivative, weight gradient, and environmental characteristics; calculate the evaluation distribution and temporal deviation based on wavelet transform; construct an immediate reward function including a smooth error metric, weight change penalty, and uncertainty metric; construct a temporal reward function by combining the temporal changes of the business format collaborative development index and distribution entropy, and optimize the trade-off factor through policy gradient, specifically including:

[0037] Expand the state vector to construct a multi-dimensional evaluation state vector, which includes the evaluation score of the comprehensive business characteristics, evaluation score gradient, second derivative of the evaluation score, gradient of the evaluation index weight vector, and environmental characteristics. The environmental characteristics include time periodicity information and evaluation scenario type information; perform multi-scale decomposition on the evaluation score using wavelet transform to obtain the trend component and detail component, and calculate the evaluation distribution based on local adaptive kernel density estimation; calculate the temporal difference using the trend component and detail component respectively to obtain the trend deviation and detail deviation, and generate an adaptive threshold through a multi-layer perceptron;

[0038] Construct a weight structure regularization term based on the evaluation index weight vector by combining relative entropy constraint and mixed norm constraint; calculate the uncertainty metric based on the evaluation distribution; adaptively adjust the evaluation index weight vector; calculate the smooth error metric of the trend deviation and detail deviation based on the Huber loss function, and construct a weight change penalty term. The trade-off factor of the weight change penalty term is adaptively adjusted through the gradient; combine the smooth error metric, weight change penalty term, and uncertainty metric to construct an immediate reward function;

[0039] Calculate the temporal difference and second derivative of the business format collaborative development index, and construct a temporal reward function by combining the temporal difference of the business format distribution entropy. The trade-off factor of the temporal reward function is optimized by the policy gradient method.

[0040] In an alternative embodiment,

[0041] Construct a dual network structure, where the value network evaluates the state-action value through the state value function and advantage function, and the policy network generates the adjustment amount of the evaluation index weight through the mean network and variance network; the steps of simultaneously optimizing the parameters of the value network and policy network based on the temporal difference error to obtain the optimized evaluation index weight vector include:

[0042] Construct a multi-granularity state encoder, design an adaptive threshold based on the temporal change rate of the evaluation index, divide the evaluation index sequence into feature segments of different time scales, calculate the statistical moment and fluctuation characteristics for each feature segment to construct a multi-scale feature pyramid, and perform feature fusion using a hierarchical attention mechanism to obtain a hierarchical state representation;

[0043] A dual network structure is constructed based on the hierarchical state representation, wherein the value network is composed of a state value function network and an advantage function network, wherein the state value function network evaluates the baseline value of the current state, and the advantage function network evaluates the value improvement of the action relative to the average level, and the state-action value is obtained by combining the state value function network and the advantage function network; a strategy network is constructed based on the state-action value, wherein the strategy network includes a mean network and a variance network, wherein the mean network outputs the expected direction of the evaluation index weight adjustment, and the variance network represents the uncertainty distribution of the weight adjustment, and the adjustment amount of the evaluation index weight is generated by a reparameterization method;

[0044] Constructing a temporal difference error, taking the difference between the predicted value of the state value function network and the target value calculated based on the advantage function network as the optimization target of the value network; constructing a generalized advantage estimate based on the temporal difference error, calculating the policy gradient of the evaluation index weight adjustment amount in combination with the outputs of the mean network and the variance network, and optimizing the parameters of the value network and the policy network at the same time;

[0045] Constructing a weight constraint based on the Wasserstein metric, including: constructing a support set for weight adjustment based on the dynamic distribution characteristics of the weight adjustment sequence, calculating the Wasserstein distance between the current weight adjustment amount and the support set, and projecting the adjustment amount of the evaluation index weight to a feasible domain that satisfies the Wasserstein distance constraint through iterative projection; updating the evaluation index weight vector based on the weight constraint, and dynamically adjusting the evaluation system through alternating optimization of the value network and the strategy network.

[0046] According to a second aspect of the embodiments of the present invention,

[0047] Provides a business vitality prediction and business district evaluation system based on multi-modal feature fusion, including:

[0048] The first unit is used to obtain commercial POI data and time series trajectory data of the target area, classify and label the commercial POI data, extract the rating features and sentiment polarity features of user comments, and calculate the distribution features of pedestrian flow based on the time series trajectory data;

[0049] The second unit is used to input the scoring features, sentiment polarity features and traffic flow distribution features into a recursive neural network to generate a time series of commercial vitality prediction for each region; construct the commercial POI data into knowledge graph nodes, calculate service complementarity based on the scoring features and sentiment polarity features, construct a network accessibility matrix based on location information and traffic flow distribution features, and optimize the network accessibility matrix through service complementarity to obtain an enhanced network accessibility matrix;

[0050] A third unit is configured to input the enhanced network reachability matrix into a Louvain community discovery algorithm to partition the target area into sub-areas, and deploy edge computing nodes in each sub-area; each edge computing node inputs the local enhanced network reachability matrix and the commercial vitality prediction time series into a contrastive learning network to generate local commercial features, and fuses the local commercial features with the scoring features and the sentiment polarity features to obtain comprehensive commercial features; an evaluation weight adaptive layer is constructed, the comprehensive commercial features are used as state inputs, and the evaluation index weights are dynamically adjusted through reinforcement learning to output commercial service evaluation results.

[0051] In a third aspect of the embodiments of the present invention,

[0052] a kind of electronic device is provided, including:

[0053] a processor;

[0054] a memory for storing instructions executable by the processor;

[0055] wherein, the processor is configured to call the instructions stored in the memory to execute the method described above.

[0056] In a fourth aspect of the embodiments of the present invention,

[0057] a computer-readable storage medium is provided, on which computer program instructions are stored, and when the computer program instructions are executed by a processor, the method described above is implemented.

[0058] Through the comprehensive analysis of commercial POI data and time series trajectory data, the present invention can accurately evaluate the commercial service quality of the target area and improve the accuracy of commercial vitality prediction. By dynamically adjusting the network edge weights and optimizing the reachability matrix, the service complementarity between commercial POIs is enhanced, the coordinated development of business formats is promoted, and the overall efficiency of commercial services is improved. By using edge computing nodes and contrastive learning networks, local commercial features can be generated in real time, and combined with comprehensive commercial features for dynamic adjustment to ensure the adaptability and accuracy of evaluation indicators. The present invention uses a contrastive learning network to generate local commercial features and dynamically adjusts the evaluation index weights through reinforcement learning, so that the commercial service evaluation can be dynamically adjusted according to the actual situation, improving the dynamics and self-adaptability of the evaluation and being closer to the actual development and changes of commercial services. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 is a schematic flowchart of a method for commercial vitality prediction and business district evaluation based on multi-modal feature fusion according to an embodiment of the present invention;

[0060] Figure 2 is a comparison chart of prediction error distributions;

[0061] Figure 3It is a comparison chart for tracking the performance of the business format collaborative development index. Detailed implementation manners

[0062] To make the objectives, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Apparently, the described embodiments are only a part rather than all of the embodiments of the present invention. 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.

[0063] The following will be combined with Figures 1 - 3 , and the technical solutions of the present invention will be described in detail with specific embodiments. These several specific embodiments below can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments.

[0064] Figure 1 It is a schematic flowchart of a method for predicting business vitality and evaluating business districts based on multi-modal feature fusion according to an embodiment of the present invention. As Figure 1 shown, the method includes:

[0065] Obtain the business POI data and time-series trajectory data of the target area, perform business format classification and annotation on the business POI data, extract the scoring features and sentiment polarity features of user comments, and calculate the pedestrian flow distribution features based on the time-series trajectory data;

[0066] Input the scoring features, sentiment polarity features and pedestrian flow distribution features into a recurrent neural network to generate the time series of predicted business vitality for each area; construct the business POI data into knowledge graph nodes, calculate service complementarity based on the scoring features and sentiment polarity features, combine the location information and pedestrian flow distribution features to construct a network reachability matrix, and optimize the network reachability matrix through service complementarity to obtain an enhanced network reachability matrix;

[0067] Input the enhanced network reachability matrix into the Louvain community discovery algorithm to divide the target area into sub-areas, and deploy edge computing nodes in each sub-area; each edge computing node inputs the local enhanced network reachability matrix and the time series of predicted business vitality into a contrastive learning network to generate local business features, and fuse them with the scoring features and sentiment polarity features to obtain comprehensive business features; construct an evaluation weight adaptive layer, take the comprehensive business features as the state input, dynamically adjust the evaluation index weights through reinforcement learning, and output the business service evaluation result.

[0068] Exemplarily, the technical implementation of the commercial service evaluation method first requires data acquisition and preprocessing. Obtain commercial POI data for the target area, including business types such as catering, retail, and leisure and entertainment. Each POI contains basic information such as name, address, longitude, and latitude. Obtain time-series trajectory data from mobile operators, including user ID, timestamp, and longitude and latitude information. Perform business type classification and annotation on commercial POIs, using a hierarchical classification system. For example, divide catering into subcategories such as Chinese cuisine, Western cuisine, and fast food. Crawl user reviews from platforms such as Dianping, extract scoring features in four dimensions: taste, environment, service, and cost performance, and use a sentiment analysis model to calculate the intensity values of positive, negative, and neutral sentiment polarity features. Based on the time-series trajectory data, statistically analyze the daily, weekend, and weekday pedestrian flow distribution characteristics of each POI.

[0069] To achieve accurate prediction of commercial vitality, a recurrent neural network model is constructed using a long short-term memory network (LSTM). The input layer includes scoring features, sentiment polarity features, and pedestrian flow distribution features, with a feature dimension of 11. The number of LSTM units is set to 128, and the number of hidden layers is 2. Training samples are constructed by sliding the time window, with the window size set to 7 days and the step size set to 1 day. The model outputs the predicted values of commercial vitality for the next 3 days.

[0070] In terms of knowledge graph construction, commercial POI data is constructed as nodes. The node attributes include basic attributes such as business type, operating area, and business hours, as well as time-series attributes of scoring features, sentiment polarity features, and pedestrian flow distribution. The relationships between nodes include spatial relationships of geographical distance, complementary and competitive relationships of business types, and overlapping relationships of business hours. When calculating service complementarity, the scoring complementarity is calculated by weighting the weights of the four scoring dimensions. The sentiment complementarity is calculated through the cosine similarity and intensity difference of sentiment vectors. The time complementarity considers the difference in business hour distribution and the degree of passenger flow peak staggering. On this basis, a network reachability matrix is constructed. Combining the physical distance between nodes and the pedestrian flow distribution characteristics, an enhanced network reachability matrix is obtained through optimization of service complementarity.

[0071] For regional division, the Louvain algorithm is used for community discovery, and the optimal number of sub-regions is determined through modularity optimization. For example, the region is divided into 12 sub-regions, and edge computing nodes are deployed in each sub-region. Each node uses a contrastive learning network to extract local commercial features. The network consists of an encoder and a contrastive loss. The encoder uses a three-layer graph convolutional network, with the input being the enhanced network reachability matrix of the sub-region and the time series of commercial vitality predictions. By maximizing the similarity of sample representations in the same sub-region and minimizing the similarity of sample representations in different sub-regions, local commercial features are learned. The local commercial features are weighted and fused with scoring features and sentiment polarity features to obtain comprehensive commercial features.

[0072] The evaluation weight adaptive layer is dynamically adjusted based on reinforcement learning. The comprehensive business features are used as the state input, and a multi-level reward function is adopted to guide the weight adjustment. The immediate reward function includes three parts: smooth error metric, weight change penalty, and uncertainty metric. The temporal reward function is constructed based on the dynamic changes of the business format collaborative development index and distribution entropy. The weight adjustment strategy is updated through the Actor-Critic architecture to achieve the adaptive optimization of the evaluation index weights.

[0073] The present invention depicts the complex relationships between commercial POIs through a knowledge graph, integrates multi-dimensional evaluation features to improve the evaluation accuracy; the edge computing architecture realizes distributed feature extraction and reduces the computational complexity; reinforcement learning ensures that the evaluation index weights dynamically adapt to environmental changes, improving the rationality and timeliness of the evaluation results.

[0074] In an alternative embodiment,

[0075] The steps of inputting the scoring feature, sentiment polarity feature, and pedestrian flow distribution feature into a recurrent neural network to generate the predicted time series of business vitality for each region include:

[0076] Construct a feature interaction matrix, generate time series features through a bidirectional gated recurrent unit network, adopt a multi-scale decomposition strategy, and predict the trend term, periodic term, and residual term through linear projection, Fourier decomposition, and multi-layer perceptron respectively. Finally, perform adaptive fusion based on the reliability score to obtain the predicted time series of business vitality, specifically including:

[0077] Feature enhancement is performed on the scoring feature to obtain scoring statistical features, which include mean, standard deviation, skewness, kurtosis, and scoring change rate; the enhanced sentiment feature is obtained by calculating the sentiment intensity based on word frequency and position weight, and the enhanced sentiment feature includes positive sentiment intensity, negative sentiment intensity, and sentiment distribution entropy; the time pattern is extracted based on the 24-hour pedestrian flow distribution feature of the target region to obtain the pedestrian flow pattern feature, and the pedestrian flow pattern feature includes peak values in the morning, noon, and evening, valley values in the morning, noon, and evening, the proportion of pedestrian flow on weekdays, the proportion of pedestrian flow on weekends, and the periodicity score;

[0078] Construct a feature interaction matrix from the scoring statistical features, enhanced sentiment features, and pedestrian flow pattern features. The feature interaction matrix includes independent representations of the three types of features and Hadamard product interaction representations between pairwise features; input the feature interaction matrix into a bidirectional gated recurrent unit network with residual connections to generate bidirectional time series features including forward feature sequences and backward feature sequences;

[0079] Perform multi-scale convolution on the bidirectional time series features to obtain daily-scale features, weekly-scale features, and monthly-scale features; perform linear projection based on the monthly-scale features to obtain the trend term prediction result, perform Fourier decomposition based on the weekly-scale features to obtain the periodic term prediction result, and input the daily-scale features, weekly-scale features, and monthly-scale features into a multi-layer perceptron to obtain the residual term prediction result; calculate the fluctuation amplitude and prediction error of the trend term prediction result, the periodic term prediction result, and the residual term prediction result to obtain a reliability score, generate a fusion weight coefficient based on the reliability score, and use the fusion weight coefficient to perform weighted summation on the trend term prediction result, the periodic term prediction result, and the residual term prediction result to obtain the predicted time series of business vitality.

[0080] Exemplarily, assume that the target area is a commercial block in a certain city, and it is necessary to predict the business vitality in the next week.

[0081] Data preparation: First, collect the data of the past three months in this commercial block, including user online ratings, user review texts, and the pedestrian flow data collected based on decrypted mobile signaling data or other sensors.

[0082] Rating feature enhancement: Conduct statistical analysis on the collected user rating data, and calculate the mean, standard deviation, skewness, kurtosis, and rating change rate. For example, if the rating data on a certain day is [4, 5, 4, 3, 5, 5, 4], then the mean is 4.29, the standard deviation is 0.87, the skewness is -0.59, and the kurtosis is -1.06. The rating change rate can be calculated by subtracting the mean rating of yesterday from the mean rating of today.

[0083] Enhanced sentiment feature: Perform natural language processing on the user review texts. First, construct a sentiment dictionary containing positive sentiment words and negative sentiment words. Then, calculate the word frequencies of positive sentiment words and negative sentiment words in each review, and assign weights in combination with the positions of the words in the sentence to finally obtain the positive sentiment intensity and negative sentiment intensity of each review. For example, in the sentence "The product quality of this store is very good, and the service attitude is also very good", both "good product quality" and "good service attitude" are positive sentiment expressions, and the positive sentiment intensity is calculated after assigning certain weights. Finally, calculate the sentiment distribution entropy to measure the distribution of sentiment.

[0084] Pedestrian flow pattern feature extraction: Analyze the twenty-four-hour pedestrian flow distribution data to extract time pattern features. For example, a day can be divided into three periods: morning, noon, and evening, and the peak and valley values of the pedestrian flow in each period can be calculated respectively. At the same time, calculate the pedestrian flow ratio on weekdays and weekends, and the periodic score of the pedestrian flow to measure the periodic change law of the pedestrian flow.

[0085] Feature Interaction Matrix Construction: Construct a feature interaction matrix from the scoring statistical features, enhanced sentiment features, and pedestrian flow pattern features. This matrix contains the independent representations of the three types of features and the Hadamard product interaction representations between pairwise features. The Hadamard product means multiplying the corresponding elements of two matrices. In this way, the interaction relationships between different features can be captured.

[0086] Bidirectional Temporal Feature Generation: Input the feature interaction matrix into a bidirectional gated recurrent unit network with residual connections. Residual connections can effectively alleviate the problem of gradient vanishing and improve the model training effect. The bidirectional gated recurrent unit network can capture temporal information in both the forward and backward directions and generate bidirectional temporal features containing forward and backward feature sequences.

[0087] Multi-scale Feature Extraction: Perform multi-scale convolution on the bidirectional temporal features to obtain daily-scale features, weekly-scale features, and monthly-scale features. Multi-scale convolution can capture feature information at different time granularities.

[0088] Generation of Business Vitality Prediction Results: Based on the monthly-scale features, perform linear projection to obtain the trend term prediction results. Based on the weekly-scale features, perform Fourier decomposition to obtain the periodic term prediction results. Input the daily-scale features, weekly-scale features, and monthly-scale features into a multi-layer perceptron to obtain the residual term prediction results. Calculate the fluctuation amplitude and prediction error of the trend term, periodic term, and residual term to obtain the reliability score. Generate a fusion weight coefficient based on the reliability score and use the fusion weight coefficient to perform weighted summation on the trend term, periodic term, and residual term to obtain the final business vitality prediction time series.

[0089] Figure 2 is a comparison chart of the prediction error distribution, Figure 2 showing the error distribution of different prediction methods. The horizontal axis represents the prediction error range, and the vertical axis represents the frequency of each error interval. The black solid bar represents the method of the present invention, the gray dashed bar represents the Transformer method, and the black dotted bar represents the LSTM method. The gray shaded area represents the ±5% error interval. 73% of the data points of the method of the present invention fall within the ±5% error interval, while only 53% and 47% of the comparison methods fall within this interval, significantly demonstrating the higher stability of the prediction results of the present method. The present invention can more accurately capture the internal laws of business activities through multi-feature fusion and deep learning models, thereby improving the accuracy of business vitality prediction; the present invention considers multiple factors such as scoring, sentiment, and pedestrian flow, and captures the interaction relationships between different features through a feature interaction matrix, making the prediction results more interpretable, capable of processing data in real time and generating prediction results, meeting the requirements of business activities for timeliness.

[0090] In an alternative embodiment,

[0091] The steps of constructing the commercial POI data as knowledge graph nodes, calculating service complementarity based on rating features and sentiment polarity features, constructing a network accessibility matrix in combination with location information and pedestrian flow distribution features, and optimizing the network accessibility matrix through service complementarity to obtain an enhanced network accessibility matrix include:

[0092] The commercial POI data is constructed into basic attributes including business type, business area, and business hours, scoring features including taste score, environment score, service score, and cost-effectiveness score, emotional polarity features including positive emotional intensity, negative emotional intensity, and neutral emotional intensity, and knowledge graph nodes including temporal attributes such as average daily order volume, weekend passenger flow, and weekday passenger flow. The knowledge graph nodes are connected to establish spatial relationships based on geographical distance, complementary competitive relationships based on business type, and overlapping relationships based on business hours;

[0093] The scoring features are weighted based on the scoring dimension weights to obtain scoring complementarity, the emotional complementarity is calculated based on the vector cosine similarity of the emotional polarity features and the difference between the emotional intensity, the time complementarity is calculated based on the difference in business time distribution and the degree of peak passenger flow staggering, and the scoring complementarity, the emotional complementarity and the time complementarity are weightedly integrated to obtain service complementarity;

[0094] The crowd flow distribution characteristics between adjacent POI nodes are decomposed in time series to obtain periodic crowd flow fluctuation patterns and crowd flow change characteristics. The trajectory stability is calculated by combining the directional entropy and duration of the crowd movement trajectory. The initial network accessibility matrix is ​​constructed based on the overlap degree of the crowd flow fluctuation patterns, the correlation of the crowd flow change characteristics and the trajectory stability. The service complementarity is multiplied by the accessibility strength of the network accessibility matrix, and the result is symmetrically normalized to obtain the enhanced network accessibility matrix.

[0095] Exemplarily, first, construct a knowledge graph of commercial POIs. Collect commercial POI data in the target area, including basic attributes such as the name, business type (such as catering, shopping, entertainment, etc.), business area, business hours, and rating features such as taste rating, environment rating, service rating, and cost-effectiveness rating. At the same time, collect user comment data on each POI, and use sentiment analysis technology to extract sentiment polarity features such as positive sentiment intensity, negative sentiment intensity, and neutral sentiment intensity. In addition, it is also necessary to collect time series attributes such as the average daily order volume, weekend passenger flow, and weekday passenger flow of each POI.

[0096] Each POI is represented as a node in the knowledge graph, and the node contains all the above attributes and features. Multiple relationships are established between nodes. For example, a spatial relationship is established based on the geographical distance between POIs. The closer the distance, the stronger the spatial relationship. A complementary and competitive relationship is established based on the business types of POIs. For example, a coffee shop and a dessert shop have a complementary relationship, while two coffee shops have a competitive relationship. An overlapping relationship is established based on the business hours of POIs. The higher the overlap degree of business hours, the stronger the overlapping relationship.

[0097] Calculate service complementarity. First, determine the scoring dimension weights according to expert experience or data analysis. For example, the taste weight is 0.4, the environment weight is 0.3, the service weight is 0.2, and the cost performance weight is 0.1. Then, perform a weighted calculation on the scoring features of each POI to obtain a weighted score. For example, if a coffee shop has a taste score of 4.5, an environment score of 4.0, a service score of 3.5, and a cost performance score of 4.0, then its weighted score is 4.5×0.4 + 4.0×0.3 + 3.5×0.2 + 4.0×0.1 = 4.1. The scoring complementarity between two POIs can be measured by calculating the absolute value of the difference between their weighted scores. The smaller the difference, the higher the complementarity.

[0098] Calculate emotional complementarity using emotional polarity features. Represent the emotional polarity features of each POI as a three-dimensional vector, such as (positive emotional intensity, negative emotional intensity, neutral emotional intensity). The emotional complementarity between two POIs can be measured by calculating the cosine similarity and the difference in emotional intensity between their emotional vectors. The higher the cosine similarity and the smaller the difference in emotional intensity, the higher the emotional complementarity.

[0099] In addition, calculate time complementarity based on the difference in business time period distribution and the degree of peak passenger flow staggering. The greater the difference in business time period distribution and the higher the degree of peak passenger flow staggering, the higher the time complementarity. For example, if the main business time period of a coffee shop is in the morning and at noon, while the main business time period of a bar is in the evening, then their time complementarity is relatively high.

[0100] Weightedly fuse the scoring complementarity, emotional complementarity, and time complementarity to obtain service complementarity. For example, assume that the scoring complementarity weight is 0.5, the emotional complementarity weight is 0.3, and the time complementarity weight is 0.2. Then, the service complementarity between two POIs can be obtained by summing them up with weights.

[0101] Construct an enhanced network reachability matrix. First, perform a time series decomposition on the pedestrian flow distribution characteristics between adjacent POI nodes to obtain the periodic pedestrian flow fluctuation pattern and the pedestrian flow change characteristics. For example, the pedestrian flow data for each day of a week can be decomposed to obtain the daily pedestrian flow fluctuation pattern and the pedestrian flow change characteristics.

[0102] Calculate the trajectory stability by combining the direction entropy and duration of the crowd movement trajectory. The lower the direction entropy and the longer the duration, the higher the trajectory stability. For example, if the movement trajectory between two POIs is relatively concentrated and the duration is long, the trajectory stability is high.

[0103] Construct an initial network reachability matrix based on the overlap degree of the crowd flow fluctuation pattern, the correlation of the crowd flow change characteristics, and the trajectory stability. The higher the overlap degree of the crowd flow fluctuation pattern, the higher the correlation of the crowd flow change characteristics, and the higher the trajectory stability, the higher the network reachability.

[0104] Multiply the service complementarity by the reachability strength of the initial network reachability matrix, and perform symmetric normalization on the result to obtain an enhanced network reachability matrix.

[0105] Suppose there are two POIs, coffee shop A and dessert shop B. The weighted score of coffee shop A is 4.1, the sentiment vector is (0.8, 0.1, 0.1), and the main business hours are in the morning and at noon. The weighted score of dessert shop B is 4.2, the sentiment vector is (0.7, 0.2, 0.1), and the main business hours are in the afternoon and at night. Suppose the weight of the score complementarity is 0.5, the weight of the sentiment complementarity is 0.3, and the weight of the time complementarity is 0.2. It is calculated that the service complementarity between coffee shop A and dessert shop B is at a high level. Suppose the initial network reachability between coffee shop A and dessert shop B is 0.6, then the enhanced network reachability is at a high level.

[0106] The present invention comprehensively considers multi-dimensional features of POIs, including basic attributes, score features, sentiment polarity features, time series attributes, and the distribution characteristics of the number of people, can more accurately depict the association relationship between POIs, and construct a more comprehensive POI knowledge graph; by calculating service complementarity, it can effectively identify POIs with complementary relationships, provide more personalized and accurate service recommendations for users, and improve the user experience; the enhanced network reachability matrix can reflect the reachability between different POIs, provide data support for urban planning and commercial layout, and potential business opportunities can be identified according to the enhanced network reachability matrix to optimize traffic planning, etc.

[0107] In an alternative embodiment,

[0108] The steps for each edge computing node to input the local enhanced network reachability matrix and the business vitality prediction time series into the contrastive learning network to generate local business features and fuse them with the score features and sentiment polarity features to obtain comprehensive business features include:

[0109] Construct a dual-branch contrastive learning network through localized processing with spatial constraints and spatio-temporal data evaluation with double decay: the matrix branch uses a graph convolutional network to extract spatial features, and the temporal branch uses a gated recurrent unit to extract temporal features. Combining the attention mechanism and multi-modal feature interaction, comprehensive business features are obtained, specifically including:

[0110] Localize the global enhanced network reachability matrix according to the community discovery results, extract the local matrix corresponding to the sub-region through a selection matrix with spatial constraints, set an adaptive weight coefficient based on the regional overlap degree, and perform boundary smoothing processing on the local matrix; collect the business vitality prediction time series within the sub-region to construct a multi-scale time series sliding window, and use a spatio-temporal data freshness evaluation mechanism based on double decay to evaluate and screen the local training data;

[0111] Construct a dual-branch contrastive learning network with feature enhancement, where the matrix branch uses a multi-level graph convolutional network to process the local matrix, and extracts hierarchical spatial features through residual connection and attention pooling; the temporal branch uses a bidirectional gated recurrent unit to process the business vitality prediction time series, and combines an adaptive frequency domain decomposition module to extract multi-scale temporal features; optimize the contrastive learning network through a hierarchical contrastive loss function that fuses spatial topological constraints and temporal consistency constraints;

[0112] Calculate the dynamic fusion weight of the spatial feature and the temporal feature based on the attention mechanism, and use a cross-scale feature aggregation module to perform multi-level feature fusion to obtain the initial local business feature; introduce a momentum update mechanism with an adaptive temperature coefficient to perform online update on the initial local business feature to obtain the local business feature;

[0113] Map the local business feature, the scoring feature, and the sentiment polarity feature to a unified feature space through non-linear projection to construct a multi-modal feature interaction matrix; use a hierarchical multi-head attention mechanism to model the local and global dependencies between modalities, and combine a feature recalibration module to adjust the feature distribution; obtain the initial comprehensive business feature through a residual connection with channel re-calibration, and obtain the final comprehensive business feature based on the adaptive weight allocation mechanism of modal importance.

[0114] Exemplarily, first, localize the global network reachability matrix. Assume a city is divided into three areas A, B, and C, and the global network reachability matrix describes the traffic convenience between any two points. To obtain the local matrix of area A, it is necessary to extract the connection information between the internal nodes of area A in the global matrix. If area A contains nodes 1, 2, and 3, the local matrix is the sub-matrix corresponding to rows and columns 1, 2, and 3 in the global matrix. Considering that there may be overlaps between areas, for example, node 4 belongs to both area A and area B, an adaptive weight coefficient needs to be set. For example, if the residence time ratio of node 4 in area A is 70% and in area B is 30%, then when constructing the local matrices of A and B, the connection weights related to node 4 are multiplied by 0.7 and 0.3 respectively. To smooth the influence of the area boundary, perform boundary smoothing on the local matrix. For example, use a Gaussian kernel to perform a convolution operation on the local matrix to make the connection information of the boundary nodes smoother.

[0115] Collect the time-series data of business vitality prediction within the sub-region. Assume that the daily passenger flow data of area A for one year is collected to form a time series. To construct a multi-scale time-series sliding window, the window sizes can be set to 7 days, 30 days, and 90 days respectively, representing the business vitality change trends of one week, one month, and one quarter respectively. To evaluate the data quality, adopt a spatio-temporal data freshness evaluation mechanism based on double decay. For example, the data closer to the current time has a higher weight; the data closer to the center of the region also has a higher weight. Screen out high-quality local training data according to the evaluation results. Assume that the passenger flow data near the center point of area A is more reliable, then a higher weight is assigned.

[0116] Construct a feature-enhanced dual-branch contrastive learning network. The matrix branch uses a multi-level graph convolutional network to process the local matrix. For example, the first-layer graph convolutional network extracts the direct connection information of the nodes in the local matrix, the second-layer graph convolutional network extracts the second-order connection information of the nodes, and so on. Hierarchical spatial features are extracted through residual connections and attention pooling. For example, the attention mechanism can focus on the nodes with higher connection weights in the local matrix to extract more important spatial features. The time-series branch uses a bidirectional gated recurrent unit to process the time series of business vitality prediction. For example, use a bidirectional GRU network to capture the forward and backward dependencies of the time series. Combine an adaptive frequency-domain decomposition module to extract multi-scale time features. For example, decompose the time series into a high-frequency part and a low-frequency part, representing short-term fluctuations and long-term trends respectively. Optimize the contrastive learning network through a hierarchical contrastive loss function that fuses spatial topological constraints and temporal consistency constraints. For example, the spatial topological constraint requires that spatially adjacent nodes have similar feature representations, and the temporal consistency constraint requires that temporally consecutive moments have similar feature representations.

[0117] Fuse spatial features and temporal features. Use the attention mechanism to calculate the dynamic fusion weights of spatial features and temporal features. For example, if the spatial features in a certain time period are more important, higher weights are assigned to the spatial features. Adopt a cross-scale feature aggregation module for multi-level feature fusion to obtain the initial local business features. For example, splice the spatial features and temporal features of different scales to obtain a more comprehensive feature representation. Introduce a momentum update mechanism with an adaptive temperature coefficient to online update the initial local business features to obtain the local business features. For example, dynamically adjust the temperature coefficient according to the distribution of the training data to make the contrast learning more effective.

[0118] Fuse the local business features, rating features, and sentiment polarity features. Assume that the rating feature is the average merchant rating in area A, and the sentiment polarity feature is the sentiment tendency of the comments on area A on social media. Map these three features to a unified feature space through non-linear projection. For example, use a multi-layer perceptron to map features with different dimensions and value ranges to the same feature space. Construct a multi-modal feature interaction matrix. For example, calculate the similarity or correlation between different features to construct an interaction matrix. Adopt a hierarchical multi-head attention mechanism to model the local and global dependencies between modalities. For example, the local attention mechanism focuses on the local interaction between different features, and the global attention mechanism focuses on the global dependencies between different features. Combine a feature recalibration module to adjust the feature distribution. For example, use a batch normalization layer to standardize the features. Obtain the initial comprehensive business features through a residual connection with channel recalibration. For example, the residual connection can avoid the problem of gradient disappearance, and channel recalibration can adjust the importance of different feature channels. Obtain the final comprehensive business features based on the adaptive weight allocation mechanism of modal importance. For example, dynamically adjust their weights according to the contribution degree of different features to the final prediction result.

[0119] By integrating multi-source data, including network accessibility, time series of business vitality prediction, scores, and sentiment polarity, the present invention can more comprehensively characterize the business features of a region, avoiding the limitations of a single data source; the contrast learning network and multi-scale time series analysis can effectively capture potential patterns and regularities in the data, improve the robustness and noise resistance of feature representation, and achieve deep fusion of spatial and time series features. Among them, the matrix branch uses a multi-level graph convolutional network to process local matrices, and innovatively introduces a residual connection and an attention pooling mechanism to extract hierarchical spatial features; the time series branch uses a bidirectional gated recurrent unit combined with an adaptive frequency domain decomposition module to effectively capture multi-scale time dynamic features. The network is optimized by a hierarchical contrast loss function that fuses spatial topological constraints and time series consistency constraints, which not only ensures that the feature representation maintains the spatial relationship and time series continuity of the original network structure, but also can improve the model performance; the adaptive weight coefficient, attention mechanism, and feature recalibration module can dynamically adjust parameters according to the actual situation of the data, optimize the feature fusion process, and thus improve the accuracy of the final comprehensive business features.

[0120] In an alternative embodiment,

[0121] The steps of constructing an evaluation weight adaptive layer, taking the comprehensive business feature as the state input, and dynamically adjusting the evaluation index weights through reinforcement learning to output the business service evaluation result include:

[0122] Construct an evaluation state space, and combine the comprehensive business feature, the evaluation index weight vector, the differential feature of the time series of business vitality prediction, and the business format collaborative development index to form a state vector, where the business format collaborative development index is calculated based on the service complementarity and the accessibility intensity of the network accessibility matrix;

[0123] Design a multi-level reward function based on the state vector, including calculating the evaluation deviation using a smooth error metric with Huber loss, constructing an immediate reward function by combining the relative entropy and the weight change penalty term of the mixed norm constraint, and adaptively adjusting the trade-off factor of the weight change penalty term through the gradient; based on the time series change characteristics of the business format collaborative development index, construct a time series reward function by combining the dynamic change of the business format distribution entropy;

[0124] Use a residual network structure to generate the reward combination weight, input the state vector and the action vector into the residual block for feature extraction and mapping, obtain the reward weight through the softmax function, weight-combine the immediate reward function and the time series reward function to get the total reward function, and perform time series smoothing processing based on a smoothing factor adaptively adjusted by the signal variance;

[0125] Construct a dual network structure, where the value network evaluates the state-action value through the state value function and the advantage function, and the policy network generates the adjustment amount of the evaluation index weights through the mean network and the variance network; simultaneously optimize the parameters of the value network and the policy network based on the temporal difference error to obtain the optimized evaluation index weight vector;

[0126] Apply the optimized evaluation index weight vector to the comprehensive business characteristics to obtain the business service evaluation result, where the business service evaluation result includes the business service quality score, the business format development balance index, and the regional coordinated development level.

[0127] Exemplarily, construct the evaluation state space. Combine the comprehensive business characteristics describing the business development status, such as the number of people flow, transaction volume, number of stores, consumer evaluations, etc., with the current evaluation index weight vector. In addition, in order to capture the business development trend, the differential characteristics of the business vitality prediction time series are also incorporated into the state space. At the same time, considering the interaction between different business formats, calculate the business format coordinated development index based on the service complementarity and the reachability intensity of the network reachability matrix, and use this index as a component of the state vector. For example, the catering industry and the cinema in a certain area can form a complementarity, and their network reachability is also relatively high, then their coordinated development index is also relatively high. Assume that the comprehensive business characteristics have 5 dimensions, the weight vector has 3 dimensions, the differential characteristics of the business vitality prediction time series have 2 dimensions, and the business format coordinated development index is 1 dimension, then the final state vector dimension is 11.

[0128] Design a multi-level reward function. The design of the reward function aims to guide the reinforcement learning model to learn the best evaluation index weights. The reward function is divided into two parts: immediate reward and temporal reward. The immediate reward function consists of two parts: one part is the evaluation deviation calculated by using a smoothing error metric method similar to the Huber loss, such as the difference between the current evaluation result and the expert score; the other part is the weight change penalty term combined with relative entropy and mixed norm constraints, which is used to limit the weight adjustment range and avoid drastic fluctuations. The trade-off factor of the weight change penalty term is adaptively adjusted by a method similar to gradient descent. The temporal reward function is constructed based on the temporal change characteristics of the business format coordinated development index and the dynamic change of the business format distribution entropy, aiming to encourage the model to learn the evaluation index weights that can promote the long-term coordinated development of the business format. For example, if the business format distribution entropy in a certain area continues to increase, indicating that the business format is more balanced, then a higher temporal reward is given.

[0129] The residual network structure is used to generate the reward combination weights. The state vector and action vector (i.e., the adjustment amount of the evaluation index weight) are input into the residual network for feature extraction and mapping. The output of the residual network is processed by a function similar to softmax to obtain the weights of the immediate reward and the sequential reward. The immediate reward and the sequential reward are multiplied by the corresponding weights and added together to obtain the total reward function. In order to make the reward signal smoother, a smoothing factor based on adaptive adjustment of the signal variance is used for sequential smoothing. For example, if the reward signal fluctuates greatly, a larger smoothing factor is used.

[0130] A dual network structure is constructed for reinforcement learning. The structure includes a value network and a policy network. The value network evaluates the state-action value through the state value function and the advantage function, that is, the value of taking a certain action in the current state. The policy network generates the adjustment amount of the evaluation index weight through the mean network and the variance network as the action output. The model uses an algorithm similar to the temporal difference error to simultaneously optimize the parameters of the value network and the policy network, and finally obtains the optimized evaluation index weight vector.

[0131] The optimized evaluation index weight vector is applied to the comprehensive business characteristics to obtain the business service evaluation results. The evaluation results include multiple dimensions, such as business service quality score, business format development balance index and regional coordinated development level. For example, assuming that the optimized weight vector is [0.5, 0.3, 0.2], and the corresponding evaluation indicators are consumer evaluation, number of stores and transaction amount, the final business service quality score is the weighted average of these three indicators.

[0132] The present invention improves the accuracy and reliability of the evaluation results. By dynamically adjusting the weights of evaluation indicators, it can better adapt to different business environments and characteristics, thereby more accurately reflecting the true level of business services. By designing a multi-level reward function, the model can be guided to learn the weights of evaluation indicators that can promote the long-term coordinated development of business formats, thereby optimizing the business ecology and enhancing overall business vitality. Through the reinforcement learning algorithm, the optimal evaluation indicator weights can be automatically learned without human intervention, thereby improving evaluation efficiency and reducing evaluation costs.

[0133] In an optional embodiment,

[0134] Designing a multi-level reward function based on the state vector includes calculating the evaluation deviation using the Huber loss smooth error metric, constructing an immediate reward function in combination with the weight change penalty term constrained by relative entropy and mixed norm, and the trade-off factor of the weight change penalty term is adaptively adjusted by gradient; Based on the time series change characteristics of the business format collaborative development index, the steps of constructing a time series reward function in combination with the dynamic change of business format distribution entropy include:

[0135] Construct a multi-dimensional evaluation state vector for the state vector, including the evaluation score and its derivative, weight gradient, and environmental characteristics; calculate the evaluation distribution and temporal deviation based on wavelet transform; construct an immediate reward function including a smooth error metric, weight change penalty, and uncertainty metric; construct a temporal reward function by combining the temporal changes of the business format collaborative development index and distribution entropy, and optimize the trade-off factor through policy gradient, specifically including:

[0136] Expand the state vector to construct a multi-dimensional evaluation state vector. The multi-dimensional evaluation state vector includes the evaluation score of the comprehensive business characteristics, the evaluation score gradient, the second derivative of the evaluation score, the evaluation index weight vector gradient, and environmental characteristics. The environmental characteristics include time periodicity information and evaluation scenario type information; perform multi-scale decomposition on the evaluation score using wavelet transform to obtain a trend component and a detail component, and calculate the evaluation distribution based on local adaptive kernel density estimation; calculate the temporal difference using the trend component and the detail component respectively to obtain the trend deviation and the detail deviation, and generate an adaptive threshold through a multi-layer perceptron;

[0137] Construct a weight structure regular term based on the evaluation index weight vector by combining relative entropy constraint and mixed norm constraint; calculate the uncertainty metric based on the evaluation distribution; adaptively adjust the evaluation index weight vector; calculate the smooth error metric of the trend deviation and the detail deviation based on the Huber loss function, and construct a weight change penalty term. The trade-off factor of the weight change penalty term is adaptively adjusted through the gradient; combine the smooth error metric, the weight change penalty term, and the uncertainty metric to construct an immediate reward function;

[0138] Calculate the temporal difference and the second derivative of the business format collaborative development index, and construct a temporal reward function by combining the temporal difference of the business format distribution entropy. The trade-off factor of the temporal reward function is optimized through the policy gradient method.

[0139] Exemplarily, construct a multi-dimensional evaluation state vector. This vector contains multiple dimensions and is used to comprehensively describe the state of the system. These dimensions include: the evaluation score for each business characteristic (such as passenger flow, sales volume, customer satisfaction, etc.); the change rate (gradient) and change trend (second derivative) of the evaluation score, which are used to reflect the dynamic changes of the evaluation indicators; the gradient of the evaluation index weight vector, which is used to reflect the changes in the importance of different indicators; and environmental characteristics, such as time periodicity information (such as weekdays / weekends, holidays, etc.) and evaluation scenario type information (such as promotional activities, special events, etc.). For example, the evaluation state vector of a shopping mall can include: the sales volume of clothing stores, the sales volume growth rate, the sales volume growth trend, the change rate of the weight of clothing stores, and the information that it is a weekend.

[0140] Perform multi-scale decomposition on the evaluation scores. Use wavelet transform to decompose each evaluation score into a trend component and a detail component. The trend component reflects the long-term change trend of the evaluation index, and the detail component reflects the short-term fluctuation of the evaluation index. For example, the trend component of the sales volume of a clothing store can reflect the overall sales situation of the store, and the detail component can reflect the sales volume fluctuation brought by weekend promotion activities.

[0141] Calculate the evaluation distribution and temporal deviation. Based on the local adaptive kernel density estimation method, calculate the probability distribution of the evaluation scores. Use the trend component and the detail component to calculate the differences from the historical data respectively to obtain the trend deviation and the detail deviation. These deviations can reflect the degree of difference between the current state and the historical state. For example, by comparing the trend component of the current sales volume of a clothing store with the trend component of the sales volume in the past month, the trend deviation can be obtained, which reflects the change in the long-term sales situation of the store. Generate an adaptive threshold through a multi-layer perceptron to determine whether the deviation exceeds the normal range.

[0142] Construct an immediate reward function. This function consists of three parts: a smoothed error metric, a weight change penalty term, and an uncertainty metric. The smoothed error metric calculates the trend deviation and the detail deviation based on the Huber loss function, and is used to measure the gap between the current state and the expected state. The weight change penalty term constructs a weight structure regularization term based on the evaluation index weight vector, combined with relative entropy constraints and mixed norm constraints, and is used to limit the magnitude of the weight change and avoid overfitting. The uncertainty metric is calculated based on the evaluation distribution and is used to measure the uncertainty of the state estimation. For example, if the distribution of the sales volume of a clothing store is relatively dispersed, the uncertainty metric is higher. Combine these three parts to form the final immediate reward function. Among them, the trade-off factor of the weight change penalty term is adaptively adjusted by the gradient to balance the contributions of different parts.

[0143] Construct a temporal reward function. This function is constructed based on the temporal changes of the business format collaborative development index and the business format distribution entropy. Calculate the temporal difference and the second derivative of the business format collaborative development index, and the temporal difference of the business format distribution entropy, which are used to reflect the dynamic changes of the business format combination. For example, if the collaborative development index of the catering industry and the entertainment industry continues to rise, it indicates that the combination effect of these two business formats is good. Combine these indicators to form the final temporal reward function. The trade-off factor of the temporal reward function is optimized by the policy gradient method to maximize the long-term cumulative reward.

[0144] Use the reinforcement learning algorithm to optimize the system. Input the constructed multi-dimensional evaluation state vector, immediate reward function, and temporal reward function into the reinforcement learning algorithm, such as the policy gradient algorithm. Through continuous iterative optimization, find the best business format combination strategy to maximize the long-term benefits of the system.

[0145] Figure 3It is a comparison chart of the tracking performance of the business format collaborative development index, as shown in Figure 3 In the figure, the dotted line is the true collaborative index, the curve marked with diamonds is the traditional weighted method, and the curve marked with circles is this solution. The curves of each method represent their predicted tracking values. The tracking curve of this solution almost coincides with the true value, especially in the region where the index fluctuates greatly (such as months 24 - 30). In terms of the average tracking error, this technical solution is 0.057, and the traditional weighted method is 0.251. The tracking accuracy of this technical solution has increased by 77.3%, which is attributed to the application of wavelet transform and adaptive kernel density estimation. Through multi-dimensional evaluation state vectors and multi-scale decomposition, the present invention can capture the changes of the system state more comprehensively, thereby improving the sensitivity and accuracy of the reward function; through smooth error measurement, weight change penalty terms, and uncertainty measurement, it can effectively suppress the influence of noise and outliers and enhance the robustness of the reward function; by combining the immediate reward function and the time-series reward function, it can realize the dynamic optimization of the business format combination, thereby maximizing the long-term benefits of the system.

[0146] In an alternative embodiment,

[0147] Construct a dual network structure, where the value network evaluates the state-action value through the state value function and the advantage function, and the policy network generates the adjustment amount of the evaluation index weight through the mean network and the variance network; the steps of simultaneously optimizing the parameters of the value network and the policy network based on the temporal difference error to obtain the optimized evaluation index weight vector include:

[0148] Construct a multi-granularity state encoder, design an adaptive threshold based on the temporal change rate of the evaluation index, divide the evaluation index sequence into feature segments of different time scales, calculate the statistical moments and fluctuation characteristics for each feature segment to construct a multi-scale feature pyramid, and use a hierarchical attention mechanism for feature fusion to obtain a hierarchical state representation;

[0149] Construct a dual network structure based on the hierarchical state representation, where the value network consists of a state value function network and an advantage function network. The state value function network evaluates the benchmark value of the current state, and the advantage function network evaluates the value improvement of the action relative to the average level. The state-action value is obtained through the combination of the state value function network and the advantage function network; construct a policy network based on the state-action value. The policy network includes a mean network and a variance network. The mean network outputs the expected direction of the evaluation index weight adjustment, and the variance network represents the uncertainty distribution of the weight adjustment. The adjustment amount of the evaluation index weight is generated through the reparameterization method;

[0150] Construct a temporal difference error, and use the difference between the predicted value of the state value function network and the target value calculated based on the advantage function network as the optimization objective of the value network; construct a generalized advantage estimation based on the temporal difference error, and combine the outputs of the mean network and the variance network to calculate the policy gradient of the evaluation index weight adjustment amount, and optimize the parameters of the value network and the policy network simultaneously;

[0151] Construct a weight constraint based on the Wasserstein metric, including: constructing a support set for weight adjustment based on the dynamic distribution characteristics of the weight adjustment sequence, calculating the Wasserstein distance between the current weight adjustment amount and the support set, and projecting the adjustment amount of the evaluation index weight into the feasible region that satisfies the Wasserstein distance constraint through iterative projection; update the evaluation index weight vector based on the weight constraint, and dynamically adjust the evaluation system through the alternating optimization of the value network and the policy network.

[0152] Exemplarily, construct a multi-granularity state encoder. This encoder is used to convert the time-series data of the evaluation index into a hierarchical state representation. Specifically, first calculate the temporal change rate of each evaluation index. According to these change rates, set an adaptive threshold to divide the evaluation index sequence into different feature segments. The method of setting the threshold can be adjusted according to the specific application scenario. For example, it can be the mean plus a multiple of the standard deviation, or the quantile method, etc. Each feature segment represents the performance of the evaluation index at different time scales. For example, one feature segment may represent short-term fluctuations, and another feature segment may represent long-term trends. For each feature segment, calculate its statistical moments, such as mean, variance, skewness, and kurtosis, etc., and fluctuation characteristics, such as amplitude, frequency, and period, etc. Combine these features into a multi-scale feature pyramid. Finally, use a hierarchical attention mechanism to perform feature fusion on the multi-scale feature pyramid to obtain a hierarchical state representation. For example, a bottom-up attention mechanism can be used to first fuse short-term features, and then gradually fuse long-term features, and finally obtain a hierarchical state representation that comprehensively considers information at different time scales. Suppose there are three evaluation indexes A, B, and C, and their values in a time period are A = [1, 2, 3, 4, 5], B = [2, 4, 6, 8, 10], C = [1, 1, 2, 3, 5]. By calculating the change rate and setting the threshold, A is divided into two segments [1, 2] and [3, 4, 5], B is divided into two segments [2, 4] and [6, 8, 10], and C is divided into two segments [1, 1, 2] and [3, 5]. Calculate the mean and variance of each segment respectively, construct a multi-scale feature pyramid, and fuse through a hierarchical attention mechanism to obtain the final state representation.

[0153] Construct a dual network structure based on hierarchical state representation. This dual network structure consists of a value network and a policy network. The value network is used to evaluate the value of the state and the state-action value, and the policy network is used to generate the adjustment amount of the evaluation index weights. The value network consists of a state value function network and an advantage function network. The state value function network evaluates the benchmark value of the current state. The advantage function network evaluates the value improvement of taking a certain action relative to the average level in the current state. Combine the output of the state value function network and the output of the advantage function network to obtain the state-action value. The policy network includes a mean network and a variance network. The mean network outputs the expected direction of the adjustment of the evaluation index weights. The variance network represents the uncertainty distribution of the weight adjustment. Through the reparameterization method, such as sampling from a standard normal distribution and combining the outputs of the mean network and the variance network, generate the adjustment amount of the evaluation index weights. Assume the state representation is [0.1, 0.2, 0.3], the state value function network outputs 0.5, and the advantage function network outputs [0.1, -0.1, 0.2], then the state-action value is [0.6, 0.4, 0.7]. Assume the mean network outputs [0.01, 0.02, -0.01], and the variance network outputs [0.001, 0.002, 0.001], the weight adjustment amount generated by the reparameterization method is [0.011, 0.022, -0.009].

[0154] Construct the temporal difference error and optimize the network parameters. Take the difference between the predicted value of the state value function network and the target value calculated based on the advantage function network as the optimization objective of the value network. Construct the generalized advantage estimation based on the temporal difference error, combine the outputs of the mean network and the variance network to calculate the policy gradient of the adjustment amount of the evaluation index weights, and optimize the parameters of the value network and the policy network simultaneously.

[0155] Construct the weight constraint based on the Wasserstein metric. First, construct the support set of the weight adjustment based on the dynamic distribution characteristics of the weight adjustment sequence. Then, calculate the Wasserstein distance between the current weight adjustment amount and the support set. The Wasserstein distance is used to measure the difference between two distributions. Through iterative projection, project the adjustment amount of the evaluation index weights into the feasible region that satisfies the Wasserstein distance constraint. Assume the current weight adjustment amount is [0.011, 0.022, -0.009], and the support set is {[0.01, 0.02, -0.01], [0.012, 0.021, -0.008]}. By calculating the Wasserstein distance and projecting the weight adjustment amount into the feasible region, obtain the final weight adjustment amount. Update the evaluation index weight vector based on the weight constraint, and dynamically adjust the evaluation system through the alternating optimization of the value network and the policy network.

[0156] By dynamically adjusting the weights of evaluation indicators, the present invention can more accurately reflect the true value of the object to be evaluated, avoid the deviation caused by static weight setting, and can adaptively adjust the weights according to the temporal variation law of evaluation indicators, improving the adaptability of the evaluation system to different environments and scenarios; through the weight constraint of Wasserstein metric, the amplitude and direction of weight adjustment can be effectively controlled, enhancing the robustness of the evaluation system and avoiding problems such as excessive weight fluctuations.

[0157] Exemplarily, this embodiment provides a commercial vitality prediction and business district evaluation method based on multi-modal feature fusion. Obtain the commercial POI data of the target area and reclassify and label the POIs according to functions. In this embodiment, the POIs are reclassified into the following categories according to functional land use: residential function, office function, commercial service, scientific research and education, medical service, cultural tourism, industrial production, transportation function, etc. At the same time, obtain the time-series trajectory data, specifically the total number of staying users per hour for each grid.

[0158] Extract the scoring features and sentiment polarity features from the user comments of commercial POIs. The scoring features include dimensions such as taste score, environment score, service score, cost performance score, etc. When extracting POI information, a weighted form is adopted, and higher weights are assigned to POI facilities with larger floor areas such as large shopping malls and tourist attractions to highlight the role of function determination.

[0159] Calculate the pedestrian flow distribution features based on the time-series trajectory data. There are significant differences in human activities in different functional areas. For example, the working area shows an inflow during the morning rush hour and an outflow during the evening rush hour, while the residential area shows the opposite situation. In this embodiment, a time clustering algorithm is used to analyze the pedestrian flow change pattern, and the selectable clustering algorithms include:

[0160] Clustering based on representatives: Based on kmeans, using metrics such as Euclidean distance and dynamic time warping (DTW) to measure the similarity of time series; density-based clustering: such as the DBSCAN algorithm; hierarchical methods: such as the BIRCH algorithm based on the CF tree; clustering based on Gaussian mixture distribution; spectral clustering, etc.

[0161] Finally, construct the POI data into the nodes of the knowledge graph, calculate the service complementarity based on the scoring features and sentiment polarity features of the nodes, and construct the network accessibility matrix by combining the location information and the pedestrian flow distribution features. When aggregating the POIs, not only the POIs inside the grid are considered, but also the surrounding areas are examined using a distance function, having a gradually decreasing influence. Finally, the weights of the evaluation indicators are dynamically adjusted through reinforcement learning, and the commercial service evaluation results are output.

[0162] This embodiment combines static POI data and dynamic pedestrian flow data, considering both spatial distribution and temporal variation, and can relatively accurately evaluate the commercial service level of the living circle.

[0163] A second aspect of an embodiment of the present invention provides a business vitality prediction and business district evaluation system based on multimodal feature fusion, the system comprising:

[0164] The first unit is used to obtain commercial POI data and time series trajectory data of the target area, classify and label the commercial POI data, extract the rating features and sentiment polarity features of user comments, and calculate the distribution features of pedestrian flow based on the time series trajectory data;

[0165] The second unit is used to input the scoring features, sentiment polarity features and traffic flow distribution features into a recursive neural network to generate a time series of commercial vitality prediction for each region; construct the commercial POI data into knowledge graph nodes, calculate service complementarity based on the scoring features and sentiment polarity features, construct a network accessibility matrix based on location information and traffic flow distribution features, and optimize the network accessibility matrix through service complementarity to obtain an enhanced network accessibility matrix;

[0166] The third unit is used to input the enhanced network reachability matrix into the Louvain community discovery algorithm to divide the target area into sub-areas, and deploy edge computing nodes in each sub-area; each edge computing node inputs the local enhanced network reachability matrix and the business vitality prediction time series into a comparative learning network to generate local business features, and integrates them with the scoring features and sentiment polarity features to obtain comprehensive business features; construct an evaluation weight adaptive layer, take the comprehensive business features as state input, dynamically adjust the evaluation index weights through reinforcement learning, and output the business service evaluation results.

[0167] According to a third aspect of the embodiments of the present invention,

[0168] An electronic device is provided, comprising:

[0169] processor;

[0170] a memory for storing processor-executable instructions;

[0171] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0172] A fourth aspect of the embodiments of the present invention is:

[0173] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0174] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0175] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A commercial vitality prediction and business district evaluation method based on multimodal feature fusion, characterized in that: include: Acquire commercial POI data and time series trajectory data of the target area, classify and label the commercial POI data by business type, extract the rating features and sentiment polarity features of user comments, and calculate the flow distribution features based on the time series trajectory data; The scoring features, sentiment polarity features and traffic flow distribution features are input into a recursive neural network to generate a time series of commercial vitality prediction for each region; the commercial POI data is constructed as a knowledge graph node, service complementarity is calculated based on the scoring features and sentiment polarity features, a network accessibility matrix is ​​constructed in combination with location information and traffic flow distribution features, and the network accessibility matrix is ​​optimized through service complementarity to obtain an enhanced network accessibility matrix; Inputting the enhanced network reachability matrix into the Louvain community discovery algorithm to divide the target area into sub-areas, and deploying edge computing nodes in each sub-area; Each edge computing node compares the local enhanced network reachability matrix with the business vitality prediction time series input into the learning network to generate local business features, and integrates them with the scoring features and sentiment polarity features to obtain comprehensive business features; constructs an evaluation weight adaptive layer, takes the comprehensive business features as state input, dynamically adjusts the evaluation index weights through reinforcement learning, and outputs the business service evaluation results.

2. The method according to claim 1, characterized in that: The steps of inputting the scoring features, sentiment polarity features and traffic flow distribution features into a recursive neural network to generate a commercial vitality prediction time series for each region include: The feature interaction matrix is ​​constructed, and the time series features are generated through the bidirectional gated recurrent unit network. The multi-scale decomposition strategy is adopted to predict the trend term, cycle term and residual term respectively through linear projection, Fourier decomposition and multi-layer perceptron. Finally, adaptive fusion is performed based on the reliability score to obtain the business vitality prediction time series, which includes: The scoring features are enhanced to obtain scoring statistical features, including mean, standard deviation, skewness, kurtosis and scoring change rate; the sentiment intensity is calculated based on word frequency and position weight to obtain enhanced sentiment features, including positive sentiment intensity, negative sentiment intensity and sentiment distribution entropy; the time pattern is extracted based on the 24-hour passenger flow distribution characteristics of the target area to obtain passenger flow pattern features, including peak values ​​in the morning, noon and evening periods, valley values ​​in the morning, noon and evening periods, weekday passenger flow ratio, weekend passenger flow ratio and periodicity score; The score statistical features, enhanced emotional features and crowd flow pattern features are used to construct a feature interaction matrix, wherein the feature interaction matrix includes independent representations of the three types of features and Hadamard product interaction representations between two features; the feature interaction matrix is ​​input into a bidirectional gated recurrent unit network with residual connections to generate bidirectional time series features including a forward feature sequence and a reverse feature sequence; Perform multi-scale convolution on the bidirectional time series features to obtain daily scale features, weekly scale features and monthly scale features; perform linear projection based on the monthly scale features to obtain trend item prediction results, perform Fourier decomposition based on the weekly scale features to obtain period item prediction results, input the daily scale features, weekly scale features and monthly scale features into a multi-layer perceptron to obtain residual item prediction results; calculate the fluctuation amplitude and prediction error of the trend item prediction result, the period item prediction result and the residual item prediction result to obtain a reliability score, generate a fusion weight coefficient based on the reliability score, and use the fusion weight coefficient to perform weighted summation on the trend item prediction result, the period item prediction result and the residual item prediction result to obtain a business vitality prediction time series.

3. The method according to claim 1, characterized in that The steps of constructing the commercial POI data as knowledge graph nodes, calculating service complementarity based on rating features and sentiment polarity features, constructing a network accessibility matrix in combination with location information and pedestrian flow distribution features, and optimizing the network accessibility matrix through service complementarity to obtain an enhanced network accessibility matrix include: The commercial POI data is constructed into basic attributes including business type, business area, and business hours, scoring features including taste score, environment score, service score, and cost-effectiveness score, emotional polarity features including positive emotional intensity, negative emotional intensity, and neutral emotional intensity, and knowledge graph nodes including temporal attributes such as average daily order volume, weekend passenger flow, and weekday passenger flow. The knowledge graph nodes are connected to establish spatial relationships based on geographical distance, complementary competitive relationships based on business type, and overlapping relationships based on business hours; The scoring features are weighted based on the scoring dimension weights to obtain scoring complementarity, the emotional complementarity is calculated based on the vector cosine similarity of the emotional polarity features and the difference between the emotional intensity, the time complementarity is calculated based on the difference in business time distribution and the degree of peak passenger flow staggering, and the scoring complementarity, the emotional complementarity and the time complementarity are weightedly integrated to obtain service complementarity; The crowd flow distribution characteristics between adjacent POI nodes are decomposed in time series to obtain periodic crowd flow fluctuation patterns and crowd flow change characteristics. The trajectory stability is calculated by combining the directional entropy and duration of the crowd movement trajectory. The initial network accessibility matrix is ​​constructed based on the overlap degree of the crowd flow fluctuation patterns, the correlation of the crowd flow change characteristics and the trajectory stability. The service complementarity is multiplied by the accessibility strength of the network accessibility matrix, and the result is symmetrically normalized to obtain the enhanced network accessibility matrix.

4. The method according to claim 1, characterized in that: The steps of each edge computing node inputting the local enhanced network reachability matrix and the business vitality prediction time series into a comparative learning network to generate local business features, and fusing them with the scoring features and the sentiment polarity features to obtain comprehensive business features include: Through localized processing with spatial constraints and dual-attenuation spatiotemporal data evaluation, a dual-branch contrastive learning network is constructed: the matrix branch uses a graph convolutional network to extract spatial features, and the time branch uses a gated recurrent unit to extract temporal features. Combined with the attention mechanism and multimodal feature interaction, comprehensive business features are obtained, including: The global enhanced network accessibility matrix is ​​localized according to the community discovery results, and the local matrix corresponding to the sub-region is extracted through a selection matrix with spatial constraints. The adaptive weight coefficient is set based on the regional overlap, and the boundary of the local matrix is ​​smoothed. The commercial vitality prediction time series in the sub-region is collected to construct a multi-scale time series sliding window, and the local training data is evaluated and screened using a spatiotemporal data freshness evaluation mechanism based on double decay. A dual-branch contrastive learning network with feature enhancement is constructed, wherein the matrix branch uses a multi-level graph convolutional network to process the local matrix, and extracts hierarchical spatial features through residual connections and attention pooling; the time series branch uses a bidirectional gated recurrent unit to process the business vitality prediction time series, and combines an adaptive frequency domain decomposition module to extract multi-scale time features; the contrastive learning network is optimized by a hierarchical contrast loss function that integrates spatial topology constraints and time series consistency constraints; The dynamic fusion weights of the spatial features and the temporal features are calculated based on the attention mechanism, and the cross-scale feature aggregation module is used to perform multi-level feature fusion to obtain the initial local business features; the momentum update mechanism of the adaptive temperature coefficient is introduced to perform online update on the initial local business features to obtain the local business features; The local commercial features, rating features and sentiment polarity features are mapped to a unified feature space through nonlinear projection to construct a multimodal feature interaction matrix. A hierarchical multi-head attention mechanism is used to model the local and global dependencies between modalities, and the feature distribution is adjusted in combination with a feature recalibration module. The initial comprehensive commercial features are obtained through a residual connection with channel recalibration, and the final comprehensive commercial features are obtained based on an adaptive weight allocation mechanism based on modality importance.

5. The method according to claim 1, characterized in that The steps of constructing an evaluation weight adaptive layer, taking comprehensive business features as state input, dynamically adjusting the evaluation index weights through reinforcement learning, and outputting the business service evaluation results include: Constructing an evaluation state space, combining comprehensive business characteristics, evaluation index weight vectors, differential characteristics of business vitality prediction time series, and business format collaborative development index to form a state vector, wherein the business format collaborative development index is calculated based on the accessibility strength of service complementarity and network accessibility matrix; Designing a multi-level reward function based on the state vector, including using the Huber loss smooth error metric to calculate the evaluation deviation, combining the relative entropy and the mixed norm constraint weight change penalty term to construct an immediate reward function, and the weight change penalty term is adjusted through gradient adaptive adjustment; Based on the time series change characteristics of the business format collaborative development index, a time series reward function is constructed in combination with the dynamic change of the business format distribution entropy; A residual network structure is used to generate reward combination weights, the state vector and the action vector are input into the residual block for feature extraction and mapping, the reward weight is obtained through a softmax function, the instantaneous reward function and the time series reward function are weightedly combined to obtain a total reward function, and a smoothing factor based on adaptive adjustment of signal variance is used for time series smoothing; Construct a dual network structure, in which the value network evaluates the state-action value through the state value function and the advantage function, and the strategy network generates the adjustment amount of the evaluation index weight through the mean network and the variance network; optimize the value network and the strategy network parameters simultaneously based on the temporal difference error to obtain the optimized evaluation index weight vector; The optimized evaluation index weight vector is applied to the comprehensive business characteristics to obtain the business service evaluation results, which include business service quality score, business format development balance index and regional coordinated development level.

6. The method according to claim 5, characterized in that Designing a multi-level reward function based on the state vector, including calculating the evaluation deviation using a smooth error metric of Huber loss, and constructing an immediate reward function in combination with a weight change penalty term constrained by relative entropy and mixed norm, wherein the trade-off factor of the weight change penalty term is adaptively adjusted by gradient; Based on the time series change characteristics of the business format collaborative development index, the steps of constructing a time series reward function in combination with the dynamic change of the business format distribution entropy include: Construct a multi-dimensional evaluation state vector for the state vector, including the evaluation score and its derivative, weight gradient and environmental characteristics; calculate the evaluation distribution and time series deviation based on wavelet transform; construct an instant reward function including smooth error measurement, weight change penalty and uncertainty measurement; construct a time series reward function based on the time series changes of the business format collaborative development index and distribution entropy, and optimize the trade-off factor through the policy gradient, including: The state vector is expanded to construct a multidimensional evaluation state vector, wherein the multidimensional evaluation state vector includes the evaluation score, evaluation score gradient, evaluation score second-order derivative, evaluation index weight vector gradient and environmental characteristics of the comprehensive business characteristics, wherein the environmental characteristics include time periodicity information and evaluation scene type information; based on the evaluation score, a wavelet transform is used to perform multi-scale decomposition to obtain a trend component and a detail component, and the evaluation distribution is calculated based on the local adaptive kernel density estimation; the trend component and the detail component are used to respectively calculate the time series difference to obtain the trend deviation and the detail deviation, and an adaptive threshold is generated by a multi-layer perceptron; Based on the evaluation index weight vector combined with relative entropy constraints and mixed norm constraints, a weight structure regularization term is constructed; based on the evaluation distribution, an uncertainty measure is calculated; the evaluation index weight vector is adaptively adjusted; based on the Huber loss function, a smooth error measure of the trend deviation and detail deviation is calculated, and a weight change penalty term is constructed, and the weight factor of the weight change penalty term is adaptively adjusted by the gradient; the smooth error measure, the weight change penalty term, and the uncertainty measure are combined to construct an instant reward function; Based on the business format collaborative development index, its time difference and second-order derivative are calculated, and a time series reward function is constructed in combination with the time series difference of the business format distribution entropy. The trade-off factor of the time series reward function is optimized by the policy gradient method.

7. The method according to claim 5, characterized in that Construct a dual network structure, in which the value network evaluates the state-action value through the state value function and the advantage function, and the strategy network generates the adjustment of the evaluation index weight through the mean network and the variance network; The steps of simultaneously optimizing the value network and the policy network parameters based on the temporal difference error to obtain the optimized evaluation index weight vector include: A multi-granularity state encoder is constructed, and an adaptive threshold is designed based on the temporal change rate of the evaluation index. The evaluation index sequence is divided into feature segments of different time scales. The statistical moment and fluctuation characteristics of each feature segment are calculated to construct a multi-scale feature pyramid. The hierarchical attention mechanism is used for feature fusion to obtain a hierarchical state representation. A dual network structure is constructed based on the hierarchical state representation, wherein the value network is composed of a state value function network and an advantage function network, wherein the state value function network evaluates the baseline value of the current state, and the advantage function network evaluates the value improvement of the action relative to the average level, and the state-action value is obtained by combining the state value function network and the advantage function network; a strategy network is constructed based on the state-action value, wherein the strategy network includes a mean network and a variance network, wherein the mean network outputs the expected direction of the evaluation index weight adjustment, and the variance network represents the uncertainty distribution of the weight adjustment, and the adjustment amount of the evaluation index weight is generated by a reparameterization method; Constructing a temporal difference error, taking the difference between the predicted value of the state value function network and the target value calculated based on the advantage function network as the optimization target of the value network; constructing a generalized advantage estimate based on the temporal difference error, calculating the policy gradient of the evaluation index weight adjustment amount in combination with the outputs of the mean network and the variance network, and optimizing the parameters of the value network and the policy network at the same time; Constructing a weight constraint based on the Wasserstein metric, including: constructing a support set for weight adjustment based on the dynamic distribution characteristics of the weight adjustment sequence, calculating the Wasserstein distance between the current weight adjustment amount and the support set, and projecting the adjustment amount of the evaluation index weight to a feasible domain that satisfies the Wasserstein distance constraint through iterative projection; updating the evaluation index weight vector based on the weight constraint, and dynamically adjusting the evaluation system through alternating optimization of the value network and the strategy network.

8. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method described in any one of claims 1 to 7.

9. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images