Intelligent store data processing method and device based on multidimensional data analysis

Abstract

The embodiment of the application provides a kind of wisdom shop data processing method and device based on multidimensional data analysis, and the alignment and mapping of space-time feature sequence are innovatively realized by collecting sensor, intelligent terminal and social media data.Multiple dimensional data fusion is carried out based on feature weight coefficient, and a commodity recommendation model is constructed using convolutional neural network and recurrent neural network.The system continuously optimizes the model performance through incremental learning, realizes personalized commodity recommendation and shop operation trend analysis.Introduce marketing decision optimization model, intelligently evaluate and dynamically optimize marketing strategy, and continuously update the model through feedback data.The method breaks through the limitation of traditional single data analysis, and provides a comprehensive data processing solution for precision marketing of wisdom shop.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of data processing, specifically to a smart store data processing method and device based on multi-dimensional data analysis. BACKGROUND

[0002] Traditional retail store data analysis is usually limited to single-dimensional sales data statistics, making it difficult to fully grasp customer demand and market trends. Existing product recommendation systems often rely on simple purchase history analysis, lacking deep integration and mining of multi-source data. In particular, in terms of personalized recommendation and precision marketing, existing systems are difficult to fully utilize multi-dimensional information such as sensor data, device data, and social media data.

[0003] At the same time, existing data processing systems have obvious shortcomings in data fusion and model training. Traditional methods often use simple data stacking, failing to effectively handle the spatio-temporal alignment problem of different source data, and lacking a dynamic evaluation mechanism for the importance of data features. The system is also lagging in model updating and optimization, making it difficult to adapt to rapidly changing market environments.

[0004] In addition, existing systems are also relatively mechanical in marketing strategy formulation and execution. Lack of intelligent decision optimization mechanism, unable to dynamically adjust marketing strategies according to real-time feedback data, affecting the continuous improvement of marketing effectiveness. Solving these problems is of great significance to improving the operational efficiency and marketing precision of smart stores. SUMMARY

[0005] To solve the problems in the prior art, the present application provides a smart store data processing method and device based on multi-dimensional data analysis, which can break through the limitations of traditional single data analysis and provide a comprehensive data processing solution for precision marketing of smart stores.

[0006] To solve at least one of the above problems, the present application provides the following technical solutions:

[0007] In a first aspect, the present application provides a smart store data processing method based on multi-dimensional data analysis, comprising:

[0008] Collecting multi-channel sensor data, intelligent terminal device data, and social media data within the store, performing timestamp alignment and spatial coordinate mapping on the sensor data and the intelligent terminal device data to obtain spatio-temporal feature sequences, performing correlation analysis on the spatio-temporal feature sequences to obtain feature weight coefficients, performing multi-dimensional data fusion on the spatio-temporal feature sequences and the social media data based on the feature weight coefficients, and generating a fusion data set;

[0009] The fused dataset is cleaned and standardized to obtain sample data. The sample data is divided into training dataset and validation dataset according to a preset ratio. The training dataset is used to train a product recommendation model based on convolutional neural network and recurrent neural network. The validation dataset is used to evaluate the performance of the trained model. The model parameters are automatically tuned and incrementally updated based on the evaluation results. The updated model is used to predict product preferences and generate a personalized product recommendation list. The fused dataset is analyzed over time to obtain store operation trend data.

[0010] Based on the personalized product recommendation list and the store operation trend data, an initial marketing strategy plan is generated. The initial marketing strategy plan is then input into a pre-trained marketing decision optimization model for strategy evaluation and optimization to obtain a personalized marketing strategy. The personalized marketing strategy is then anonymized and encrypted before being stored in a cloud data center. The personalized marketing strategy is then distributed to store terminals for execution via a message queue. Data on the execution effect of the marketing strategy is collected for model optimization and updates.

[0011] Further, the step of aligning the sensor data and the smart terminal device data with timestamps and mapping them with spatial coordinates to obtain a spatiotemporal feature sequence, and then performing correlation analysis on the spatiotemporal feature sequence to obtain feature weight coefficients, includes:

[0012] A spatiotemporal reference coordinate system is established based on the acquisition time and device location information in the sensor data. The sensor data is mapped to the time axis according to the timestamp of the acquisition time mark. The device number in the smart terminal device data is converted into the corresponding spatial coordinate point and mapped to the spatial axis. The spatiotemporal feature sequence is generated by resampling data at different sampling frequencies using bilinear interpolation.

[0013] The correlation between features in the spatiotemporal feature sequence is calculated using the Pearson correlation coefficient to construct a feature correlation matrix. Based on the feature correlation matrix, the contribution of each feature to the target variable is calculated, and the contribution is normalized to obtain the feature weight coefficient.

[0014] Furthermore, the step of performing multi-dimensional data fusion on the spatiotemporal feature sequence and the social media data based on the feature weight coefficients to generate a fused dataset includes:

[0015] The weighted feature value is obtained by weighting the feature weight coefficient with the corresponding spatiotemporal feature sequence. The text feature vector is obtained by performing topic classification and sentiment analysis on social media data. The weighted feature value is aligned with the text feature vector in the time dimension according to the timestamp in the spatiotemporal feature sequence to construct a multidimensional feature tensor.

[0016] The tensor decomposition algorithm is used to reduce the dimension of the multi-dimensional feature tensor to obtain a feature subspace, and the heterogeneous data in the feature subspace is mapped to a unified feature space based on a local linear embedding algorithm to generate a fusion data set.

[0017] Further, the sample data is obtained by performing data cleaning and standardization processing on the fusion data set, the sample data is divided into a training data set and a validation data set according to a preset ratio, the commodity recommendation model constructed based on a convolutional neural network and a recurrent neural network is trained using the training data set, the performance of the trained model is evaluated using the validation data set, the model parameters are automatically tuned and incrementally learned based on the evaluation results, and the method comprises the following steps:

[0018] The missing values in the fusion data set are supplemented by a K-nearest neighbor interpolation method, abnormal data points are removed by an outlier detection algorithm, numerical features are normalized by maximum and minimum values, and category features are converted by one-hot encoding to obtain sample data, and the sample data is randomly divided into a training data set and a validation data set according to a ratio of 8:2.

[0019] The spatial features in the training data set are extracted using a convolutional neural network, the extracted spatial features are input into a recurrent neural network for time series modeling, the network parameters are optimized by a back propagation algorithm to obtain an initial commodity recommendation model, the accuracy, recall rate and F1 score of the model are calculated using the validation data set, the learning rate, batch size and network layer number of the initial commodity recommendation model are tuned based on a grid search algorithm, and the model parameters are incrementally trained and updated using a sliding window method.

[0020] Further, the updated model is used for commodity preference prediction to generate a personalized commodity recommendation list, and store operation trend data is obtained by performing time series analysis on the fusion data set, comprising:

[0021] The user historical behavior data is input into the updated commodity recommendation model, the user interest feature vector is extracted, the similarity matrix between the user and the commodity is calculated by a collaborative filtering algorithm, the commodities are sorted according to the similarity score, and the top-scoring commodities are selected to construct a personalized recommendation list;

[0022] The fusion data set is divided into time series according to hours and dates, the trend of the passenger flow and sales data is decomposed by an exponential smoothing method, the periodic and seasonal features in the data are extracted by an autoregressive model, and the decomposed time series data is modeled and analyzed by a long short-term memory network to obtain store operation trend data.

[0023] Further, the personalized marketing strategy is obtained by inputting the marketing strategy initial scheme into a pre-trained marketing decision optimization model for strategy evaluation and optimization, including:

[0024] Based on the product categories and sales prediction values in the personalized product recommendation list, and in combination with the customer flow fluctuation rules and consumer portrait distribution in the store operation trend data, a product promotion rule library is constructed by using a decision tree algorithm, and an initial marketing strategy scheme including promotion strength, promotion period and target customer group is generated through rule matching;

[0025] The marketing strategy initial scheme is converted into a strategy feature vector, which is input into a marketing decision optimization model constructed based on reinforcement learning, and strategy parameters are explored by using a Monte Carlo tree search method. The strategy revenue value is calculated based on historical marketing effect data, and the revenue value is maximized and optimized by using a strategy gradient algorithm to obtain a personalized marketing strategy.

[0026] Further, the personalized marketing strategy is stored in a cloud data center after data desensitization and encryption processing, and the personalized marketing strategy is distributed to a store terminal for execution through a message queue, and marketing strategy execution effect data is collected for model optimization and update, including:

[0027] The user identification information in the personalized marketing strategy is hashed and encrypted, the strategy parameters are reversibly encrypted, sensitive product information is desensitized by using data masking technology, the processed strategy data is written into multiple storage nodes of a cloud data center in a distributed storage manner, a strategy distribution channel is established based on a distributed message middleware, and the encrypted marketing strategy data packet is pushed to a message queue server;

[0028] The sales conversion rate, single price change value and member repurchase rate data of the marketing strategy in different time periods are collected from the store terminal, the collected data is parsed into effect evaluation indexes in a preset format, the effect evaluation indexes are returned to the model optimization module through a data pipeline, and sample data including strategy parameters and execution effects are established for model parameter update.

[0029] In a second aspect, the present application provides a smart store data processing device based on multi-dimensional data analysis, including:

[0030] The multi-dimensional data fusion module is configured to collect multi-channel sensor data, intelligent terminal device data and social media data in a store, perform timestamp alignment and spatial coordinate mapping on the sensor data and the intelligent terminal device data to obtain a time-space feature sequence, perform correlation analysis on the time-space feature sequence to obtain a feature weight coefficient, perform multi-dimensional data fusion on the time-space feature sequence and the social media data based on the feature weight coefficient, and generate a fusion data set;

[0031] The multi-dimensional data processing module is configured to perform data cleaning and standardization processing on the fusion data set to obtain sample data, divide the sample data into a training data set and a verification data set according to a preset ratio, train a commodity recommendation model constructed based on a convolutional neural network and a recurrent neural network using the training data set, perform performance evaluation on the trained model using the verification data set, automatically optimize model parameters based on the evaluation result and perform incremental learning update, use the updated model to generate an individualized commodity recommendation list for commodity preference prediction, and perform time series analysis on the fusion data set to obtain store operation trend data.

[0032] The intelligent recommendation module is configured to generate an initial marketing strategy scheme based on the individualized commodity recommendation list and the store operation trend data, input the initial marketing strategy scheme into a pre-trained marketing decision optimization model to perform strategy evaluation and optimization to obtain an individualized marketing strategy, store the individualized marketing strategy to a cloud data center after data desensitization and encryption processing, distribute the individualized marketing strategy to a store terminal for execution through a message queue, and collect marketing strategy execution effect data for model optimization update.

[0033] In a third aspect, the present application provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the method for processing data of a smart store based on multi-dimensional data analysis.

[0034] In a fourth aspect, the present application provides a computer readable storage medium having a computer program stored thereon, wherein the computer program is executable by a processor to implement the steps of the method for processing data of a smart store based on multi-dimensional data analysis.

[0035] In a fifth aspect, the present application provides a computer program product comprising computer programs / instructions, wherein the computer programs / instructions are executable by a processor to implement the steps of the method for processing data of a smart store based on multi-dimensional data analysis.

[0036] As described above, this application provides a smart store data processing method and apparatus based on multidimensional data analysis. By collecting data from sensors, smart terminals, and social media, it innovatively achieves the alignment and mapping of spatiotemporal feature sequences. Multidimensional data fusion is performed based on feature weight coefficients, and a product recommendation model is constructed using convolutional neural networks and recurrent neural networks. The system continuously optimizes model performance through incremental learning, achieving personalized product recommendations and store operation trend analysis. Simultaneously, a marketing decision optimization model is introduced to intelligently evaluate and dynamically optimize marketing strategies, and the model is continuously updated through feedback data. This method breaks through the limitations of traditional single data analysis, providing a comprehensive data processing solution for precise marketing in smart stores. Attached Figure Description

[0037] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0038] Figure 1 This is one of the flowcharts illustrating the smart store data processing method based on multidimensional data analysis in the embodiments of this application;

[0039] Figure 2 This is the second flowchart illustrating the smart store data processing method based on multidimensional data analysis in the embodiments of this application.

[0040] Figure 3 This is the third flowchart illustrating the smart store data processing method based on multidimensional data analysis in the embodiments of this application;

[0041] Figure 4 This is the fourth flowchart illustrating the smart store data processing method based on multidimensional data analysis in the embodiments of this application.

[0042] Figure 5 This is the fifth flowchart illustrating the smart store data processing method based on multidimensional data analysis in the embodiments of this application;

[0043] Figure 6 This is the sixth flowchart illustrating the smart store data processing method based on multidimensional data analysis in the embodiments of this application;

[0044] Figure 7 This is the seventh flowchart illustrating the smart store data processing method based on multidimensional data analysis in the embodiments of this application;

[0045] Figure 8A structure diagram of a smart store data processing device based on multi-dimensional data analysis in an embodiment of the present application.

[0046] Figure 9 A structure diagram of an electronic device in an embodiment of the present application.

[0047] Reference signs:

[0048] The electronic device 9600, the central processor 9100, the memory 9140, the communication module 9110, the input unit 9120, the audio processor 9130, the display 9160, the power supply 9170, the buffer memory 9141, the application / function storage unit 9142, the data storage unit 9143, the driver program storage unit 9144, the antenna 9111, the speaker 9131, and the microphone 9132. DETAILED DESCRIPTION

[0049] To make the objectives, technical solutions, and advantages of the embodiments of the present application clearer, the following will be combined with the accompanying drawings to make a clear and complete description of the technical solutions in the embodiments of the present application. Obviously, the described embodiments are some but not all of the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0050] The acquisition, storage, use, and processing of data in the technical solutions of the present application all comply with relevant provisions of national laws and regulations.

[0051] In view of the problems in the prior art, the present application provides a smart store data processing method and device based on multi-dimensional data analysis, which innovatively realizes the alignment and mapping of spatiotemporal feature sequences by collecting sensor, intelligent terminal, and social media data. Multi-dimensional data fusion is performed based on feature weight coefficients, and a convolutional neural network and a recurrent neural network are used to construct a commodity recommendation model. The system continuously optimizes the model performance through incremental learning, realizes personalized commodity recommendation and store operation trend analysis. At the same time, a marketing decision optimization model is introduced to intelligently evaluate and dynamically optimize marketing strategies, and the model is continuously updated through feedback data. This method breaks through the limitations of traditional single data analysis and provides a comprehensive data processing solution for precise marketing of smart stores.

[0052] In order to break through the limitations of traditional single data analysis and provide a comprehensive data processing solution for precise marketing of smart stores, an embodiment of a smart store data processing method based on multi-dimensional data analysis is provided in the present application, as shown in Figure 1 , which specifically includes the following contents:

[0053] Step S101: Collecting multi-channel sensor data, intelligent terminal device data and social media data in the store, time stamp aligning and space coordinate mapping of the sensor data and the intelligent terminal device data to obtain a time-space feature sequence, correlation analysis of the time-space feature sequence to obtain a feature weight coefficient, multi-dimensional data fusion of the time-space feature sequence and the social media data based on the feature weight coefficient, and generation of a fusion data set;

[0054] Optionally, the embodiment is directed to a smart store application scenario, and a comprehensive data collection system is designed. High-precision infrared sensor arrays are deployed in different areas of the store, each sensor covers an area of 3 square meters, and real-time human flow density data is collected; temperature and humidity sensors are arranged at key positions such as commodity display areas and rest areas, and environmental parameters are recorded every 15 minutes; high-definition cameras are installed at entrances and exits, main channels and shelf areas to collect customer behavior images; RFID readers cover all commodity display areas to track real-time changes in commodity positions. Intelligent terminal devices include new POS machines with biometric identification functions, intelligent electronic price tags and self-service cash register devices to collect transaction data, price changes and inventory information. At the same time, through API interfaces, real-time user comments, picture sharing and interaction data related to the store on social platforms such as Weibo and Xiaohongshu are obtained.

[0055] The embodiment innovatively proposes a space-time unified mapping method for multi-source heterogeneous data. First, a standard time axis is established based on the store opening time, and all sensor data are preliminarily aligned according to the millisecond-level time stamp. Due to the significant differences in sampling frequencies of different sensors, for example, human flow sensors sample once every minute, while temperature and humidity sensors sample once every 15 minutes, an improved bilinear interpolation algorithm is used for data resampling. Specifically, for any time t, according to the data values and time intervals of the adjacent two sampling points, the data value of the interpolation point is calculated through a two-dimensional weight matrix to ensure smooth transition of the interpolation result and meet the physical meaning.

[0056] In the spatial dimension, the embodiment establishes a space coordinate system based on the three-dimensional model of the store. The installation positions of fixed sensors are determined by a surveying instrument to obtain accurate coordinates, and the positions of mobile devices such as handheld POS machines are updated in real time by an indoor positioning system. Through a coordinate transformation matrix, all position data are uniformly mapped to a standardized grid coordinate system, each grid element has a size of 0.5m x 0.5m, and accurate alignment of spatial data is achieved. For example, RFID identification data of a certain shelf area can be accurately associated with data such as customer flow density and dwell time in that area.

[0057] For the correlation analysis of spatio-temporal feature sequences, the embodiment designs a hierarchical progressive analysis method. First, the local correlation based on the time window is calculated, and the improved Pearson correlation coefficient method is used to consider the time sequence characteristics and periodic changes of the data. Then a multi-dimensional correlation matrix is constructed to not only analyze the linear correlation between features, but also to evaluate the nonlinear correlation relationship through mutual information and conditional entropy. For example, it is found that there is a significant time-lag correlation between the stay time in the rest area and the purchase rate of surrounding commodities, and this correlation is more obvious during holidays.

[0058] The embodiment introduces a dynamic feature weight calculation mechanism. Based on the correlation analysis results, combined with information gain rate and feature importance score, a feature weight evaluation model is constructed. The model considers the timeliness, stability and business relevance of the features, and dynamically adjusts the weight coefficients. For example, when predicting the sales of high-value commodities, the weight of customer portrait features is increased; when predicting the sales of fast-moving consumer goods, more attention is paid to the weight of historical sales data.

[0059] In the multi-dimensional data fusion link, the embodiment adopts a hierarchical fusion strategy. First, the social media data is subjected to in-depth semantic analysis, the pre-trained BERT model is used to extract text features, the hot topics are identified through the topic model, and the user sentiment orientation is quantified through the sentiment analysis model. Then the text feature vector is aligned with the spatio-temporal feature sequence in the time dimension, and a feature tensor containing multiple views is constructed.

[0060] The embodiment innovatively uses an improved tensor decomposition algorithm to process high-dimensional feature data. Through the Tucker decomposition method, the original feature tensor is decomposed into a core tensor and multiple factor matrices, and the optimal decomposition dimension is determined by combining singular value analysis. On this basis, an improved local linear embedding algorithm is used to map heterogeneous data to a unified low-dimensional feature space, while maintaining the local structure characteristics of the data.

[0061] This multi-dimensional data processing scheme not only solves the data inconsistency problem in traditional methods, but also can adaptively handle data quality fluctuations. For example, when some sensor data is temporarily interrupted, the system can repair the data through the complementarity of related features. At the same time, the update of the fusion data set adopts incremental calculation, and the feature weight is adjusted every hour to ensure the timeliness and accuracy of the data analysis results.

[0062] The innovative data processing scheme of the embodiment significantly improves the operation efficiency of the smart store. Through the deep fusion of multi-dimensional data, the customer behavior patterns can be accurately described, the commodity demand trend can be predicted, and data support can be provided for personalized marketing decisions, effectively improving the operation efficiency and user experience of the store.

[0063] Step S102: The fused dataset is cleaned and standardized to obtain sample data. The sample data is divided into training dataset and validation dataset according to a preset ratio. The product recommendation model based on convolutional neural network and recurrent neural network is trained using the training dataset. The performance of the trained model is evaluated using the validation dataset. The model parameters are automatically tuned and incrementally updated based on the evaluation results. The updated model is used to predict product preferences and generate a personalized product recommendation list. The fused dataset is analyzed over time to obtain store operation trend data.

[0064] Optionally, this embodiment, for data cleaning of the fused dataset, first designs a multi-dimensional anomaly detection mechanism. In the time dimension, a 24-hour sliding window is used to calculate the mean, standard deviation, skewness, and kurtosis of each feature. For continuous data such as passenger flow and transaction volume, a modified 3-sigma criterion and exponentially weighted moving average method are combined to identify abrupt change points. Simultaneously, spatial correlation is considered; when sensor data from a certain area shows anomalies, cross-validation is performed using data from neighboring areas. For textual data such as user comments, semantic vectors are calculated based on a pre-trained BERT model, and a local anomaly factor algorithm is used to detect anomalous text.

[0065] This embodiment employs a hierarchical adaptive normalization strategy in the data standardization process. For strongly periodic features such as pedestrian density, wavelet transform is first used for periodic decomposition, and the trend and periodic components are standardized separately. For long-tailed distribution features such as sales volume, logarithmic transformation followed by Min-Max normalization is used. For environmental parameters such as temperature and humidity, reasonable normalization intervals are set based on domain knowledge. Categorical features such as product categories are encoded using feature hashing technology, and collision detection ensures the uniqueness of the encoding.

[0066] In partitioning the dataset, this embodiment innovatively proposes a time-aware stratified sampling method. First, the data is divided into multiple time segments according to chronological order, with each segment containing 24 hours of data. Then, within each time segment, stratification is performed according to dimensions such as user activity and product category to ensure that the training and validation sets have similar data distributions. Specifically, the training set accounts for 80% and the validation set accounts for 20%, and this proportion is maintained for samples in each time segment and each data layer.

[0067] This embodiment designs a deep neural network model based on a dual-flow architecture. The spatial feature flow uses an improved ResNet structure, containing multiple residual blocks and atrous convolution layers, which can extract spatial features of different scales. The temporal feature flow uses a BiLSTM structure, capturing long-term dependencies through two forward and backward LSTM layers. The two feature flows are fused through an attention mechanism, dynamically adjusting the importance weights of different features.

[0068] The input features of the model include user features (historical browsing sequences, purchase records, membership levels, etc.), product features (category, price, inventory, sales, etc.), time features (time period, week, holiday, etc.), spatial features (product location, regional passenger flow, etc.), and environmental features (temperature, humidity, weather, etc.). The interaction between features is modeled through a multi-head self-attention mechanism, with each attention head responsible for capturing different types of feature associations.

[0069] During training, this embodiment uses a multi-stage training strategy. In the first stage, pre-training tasks are used to learn the basic representation of features through mask prediction and sequence reconstruction tasks. In the second stage, multi-task joint training is performed, optimizing multiple objectives such as click prediction, purchase conversion, and dwell time prediction. A dynamic weight adjustment method based on task importance is used to adaptively balance the training loss of each task.

[0070] To improve the robustness of the model, this embodiment introduces multiple regularization techniques. Spatial dropout is used in convolution layers, randomly discarding feature maps; zoneout regularization is used in recurrent layers, randomly keeping hidden states unchanged; and label smoothing is used to reduce overfitting. A cosine annealing learning rate scheduler is used during training, periodically adjusting the learning rate to help the model escape local optima.

[0071] The model evaluation uses a multi-dimensional evaluation system. In addition to basic indicators such as accuracy and recall, it also includes coverage (evaluating the breadth of recommended products), diversity (using an information entropy-based measure), novelty (based on the inverse of product popularity), and relevance (evaluated through A / B testing). Based on these indicators, Bayesian optimization algorithms are used to automatically adjust model hyperparameters such as network depth, number of attention heads, and dropout rate.

[0072] The incremental learning mechanism of this embodiment includes three levels: feature level, updating feature distribution statistics through incremental feature engineering; model level, using elastic weight integration, new and old model parameters are fused through importance weighting; sample level, using time decay-based sample importance weighting, reducing the influence of historical samples.

[0073] In generating the recommendation list, the embodiment adopts a multi-constraint ranking optimization method. First, multiple scores (click probability, conversion rate, profit rate, etc.) of candidate goods are obtained through model prediction, and then a multi-objective programming method is combined to consider inventory constraints, exposure frequency constraints and diversity constraints to solve the optimal recommendation sequence. Meanwhile, an exploration-exploitation mechanism is introduced to reserve a certain proportion of positions for recommending newly listed goods.

[0074] For shop operation trend analysis, the embodiment combines empirical mode decomposition and spectral analysis methods to extract trend features at multiple time scales. Through multi-scale analysis at the hour, day and week levels, the change patterns of key indicators such as customer flow and sales are identified, providing decision support for dynamically adjusting marketing strategies and inventory management.

[0075] The innovative solution of the embodiment significantly improves the performance of the recommendation system. Through multi-dimensional data cleaning and standardization, the quality of the training data is guaranteed; the dual-flow architecture of the deep neural network effectively fuses the spatio-temporal features; the multi-task learning and incremental updating mechanism enables the model to continuously adapt to business changes; and the multi-constraint recommendation generation strategy ensures the practicality of the recommendation results. The scheme shows strong adaptability and stability in practical application, can quickly respond to changes in user interest, and continuously optimize the recommendation effect.

[0076] Step S103: Based on the personalized product recommendation list and the shop operation trend data, an initial marketing strategy scheme is generated, the initial marketing strategy scheme is input into a pre-trained marketing decision optimization model for strategy evaluation and optimization to obtain a personalized marketing strategy, the personalized marketing strategy is subjected to data desensitization and encryption processing and then stored in a cloud data center, the personalized marketing strategy is distributed to a shop terminal through a message queue for execution, and marketing strategy execution effect data is collected for model optimization and update.

[0077] Optionally, based on the personalized product recommendation list and the shop operation trend data, the embodiment first builds a multi-dimensional marketing strategy generation framework. The framework considers factors such as product characteristics (such as inventory turnover rate, gross profit margin, shelf life), user portraits (such as consumer ability, shopping frequency, category preference), and market environment (such as competitor price, seasonal characteristics, promotion activities), determines the weight coefficients of each dimension through the analytic hierarchy process, and generates an initial marketing strategy scheme.

[0078] In the strategy generation process, the embodiment adopts a method combining rules and heuristic algorithms. For high-liquidity goods, the discount strength is dynamically adjusted according to the inventory level and sales forecast; for seasonal goods, a step-by-step promotion plan is developed based on historical sales curves; and for related goods, a bundled sales strategy is designed. Meanwhile, considering the characteristics of customer flow at different times, the promotion strength is increased during the peak period of customer flow, and brand marketing is emphasized during the off-season.

[0079] The embodiment innovatively designs a marketing decision optimization model based on reinforcement learning. The model adopts a double Q-network architecture, the state space includes product attributes, market environment and user behavior, and the action space includes price adjustment, promotion method, marketing channel and other decision variables. The reward function considers sales revenue, gross margin, user satisfaction and other multiple objectives, and balances the importance of each objective through a weight adaptive method.

[0080] The model training uses an experience replay mechanism to store historical decisions and their effects, and selects representative samples for learning through a priority sampling method. To improve the generalization ability of the model, an adversarial training mechanism is introduced to enhance the diversity of training samples by adding perturbation data. At the same time, an exploration strategy based on Thompson sampling is designed to ensure the stability of the strategy while maintaining moderate innovation.

[0081] In the marketing strategy evaluation stage, the embodiment constructs a multi-level evaluation system. First, the expected effect of the strategy is evaluated through offline simulation, including sales increase, gross margin change, inventory turnover and other indicators. Then the strategy effect is verified through A / B testing in a small sample, and the strategy parameters are dynamically adjusted according to the actual feedback. The evaluation results are fed back to the optimization model through the attention mechanism to guide the improvement direction of the strategy.

[0082] In terms of data security, the embodiment adopts a multi-encryption scheme. First, sensitive data is processed with differential privacy to prevent individual information leakage while maintaining data usability. Then, homomorphic encryption technology is used to encrypt the marketing strategy to ensure its security during transmission and storage. A hierarchical encryption strategy is used for data of different security levels to balance security and system performance.

[0083] The embodiment designs a strategy distribution mechanism based on message queues. Using the publish-subscribe pattern, marketing strategies are grouped and pushed according to store attributes and geographic location. Message queues support message persistence and failure retry to ensure reliable delivery of strategies. At the same time, strategy version control is implemented to support fast rollback and real-time update of strategies.

[0084] In terms of strategy execution effect feedback, the embodiment establishes a real-time monitoring system. User interaction data, sales conversion data and inventory change data are collected through the burying point technology. A stream processing framework is used to calculate core indicators in real time, and when the indicators show abnormal fluctuations, the strategy dynamic adjustment mechanism is triggered. At the same time, the decision model is continuously optimized through incremental learning to improve the model's adaptability to market changes.

[0085] Through this combination of deep learning and operational optimization, the embodiment realizes intelligent generation and dynamic optimization of marketing strategies. This scheme can quickly adjust strategies according to changes in market environment, while improving sales results and ensuring the balance of inventory and profits. The scalability and security design of the scheme enable it to adapt to retail scenarios of different scales, providing precise, secure and efficient marketing decision support for merchants.

[0086] As can be seen from the above description, the wisdom store data processing method based on multi-dimensional data analysis provided by the embodiment of the application can innovatively realize the alignment and mapping of the spatio-temporal feature sequence by collecting sensor data, intelligent terminal data and social media data. Multi-dimensional data fusion is performed based on feature weight coefficients, and a convolutional neural network and a recurrent neural network are used to construct a commodity recommendation model. The system continuously optimizes the performance of the model through incremental learning, realizes personalized commodity recommendation and store operation trend analysis. At the same time, a marketing decision optimization model is introduced to intelligently evaluate and dynamically optimize marketing strategies, and the model is continuously updated through feedback data. This method breaks through the limitations of traditional single data analysis and provides a comprehensive data processing solution for precise marketing of wisdom stores.

[0087] In an embodiment of the wisdom store data processing method based on multi-dimensional data analysis of the application, referring to Figure 2 , the method can further include the following content:

[0088] Step S201: Establishing a space-time reference coordinate system based on the collection time and device location information in the sensor data, mapping the sensor data to the time axis according to the time stamp marked by the collection time, converting the device number in the intelligent terminal device data into the corresponding spatial coordinate point and mapping it to the spatial axis, and generating a space-time feature sequence by resampling the data of different sampling frequencies through bilinear interpolation method;

[0089] Step S202: Calculating the correlation degree between each feature in the space-time feature sequence using the Pearson correlation coefficient, constructing a feature correlation matrix, calculating the contribution degree of each feature to the target variable based on the feature correlation matrix, and obtaining the feature weight coefficient by normalizing the contribution degree.

[0090] Optionally, based on the real-time collected sensor data and intelligent terminal device data, the embodiment first constructs a unified space-time reference coordinate system. In the time dimension, the time stamp of each data point is standardized, and time in different time zones and date formats is uniformly converted into UTC time stamp. In the spatial dimension, based on the plan layout of the store, a two-dimensional rectangular coordinate system with the store entrance as the origin is established, and the physical location of each device is mapped to a specific coordinate point.

[0091] To address the issue of different sampling frequencies for different types of sensors such as temperature sensors and humidity sensors, the embodiment designs an adaptive resampling strategy. First, the sampling characteristics of various sensors are analyzed. For rapidly changing parameters (such as human flow), a higher sampling frequency is used, while for slowly changing parameters (such as temperature), a lower sampling frequency is used. Then, an improved bilinear interpolation algorithm is used for data resampling. This algorithm introduces a time weight decay factor based on traditional bilinear interpolation, making the interpolation result more biased towards the measured data at the adjacent time.

[0092] To handle the problem of dynamic changes in device location, the embodiment innovatively proposes a dynamic space mapping mechanism. The location changes of mobile devices are tracked in real time through Bluetooth beacons and WiFi positioning technology, and the Kalman filter algorithm is used to smooth the location data, reducing the influence of positioning noise. At the same time, a device movement trajectory model is established to model the movement law of the device, improving the accuracy of location prediction.

[0093] In the generation process of spatio-temporal feature sequence, the embodiment adopts a multi-dimensional feature extraction method. For the time dimension, in addition to the basic timestamp, time features such as time period, week, and holiday are extracted; for the spatial dimension, in addition to the Cartesian coordinates, spatial features such as relative distance between devices and density distribution are calculated. Through this multi-dimensional feature extraction, the spatio-temporal distribution characteristics of the data are fully captured.

[0094] In the feature correlation analysis section, the embodiment first calculates the linear correlation between features through an improved Pearson correlation coefficient. Considering the nonlinear characteristics of the data, a nonlinear correlation measure based on mutual information is also introduced. Through the combination of these two methods, a more comprehensive feature correlation matrix is constructed.

[0095] In the feature contribution degree calculation process, the embodiment designs a feature importance evaluation method based on random forest. First, a random forest model is constructed, and the feature random permutation method is used to calculate the influence of each feature on the model performance. At the same time, considering the interaction between features, the SHAP value calculation method is introduced to evaluate the marginal contribution of the feature.

[0096] For the normalization of feature weights, the embodiment uses the soft-max function for optimization. This method can maintain the relative proportion of feature weights while ensuring that the sum of weights is 1. To improve the stability of weight allocation, an L1 regularization term is introduced to suppress small weight values and highlight the role of important features.

[0097] The innovation scheme of the embodiment effectively solves the problem of spatio-temporal alignment of multi-source heterogeneous data. Through a unified spatio-temporal reference coordinate system and an intelligent resampling strategy, seamless fusion of data of different sources and different sampling frequencies is realized. At the same time, through multi-dimensional feature extraction and in-depth correlation analysis, key features that have an important influence on the target variable are accurately identified, providing a reliable basis for subsequent data modeling and analysis.

[0098] In practical applications, this scheme shows strong adaptability and scalability. Through the dynamic space mapping mechanism, the device position change can be accurately tracked; through multi-dimensional feature extraction, the spatio-temporal features of the data are fully mined; through comprehensive feature importance evaluation, key influencing factors are accurately identified. These technical innovations significantly improve the efficiency and accuracy of data processing, providing high-quality data support for subsequent intelligent decision-making.

[0099] In an embodiment of the wisdom store data processing method based on multi-dimensional data analysis of the present application, referring to Figure 3 , the following contents can also be specifically included:

[0100] Step S301: weighting the feature weight coefficients and the corresponding spatio-temporal feature sequence to obtain weighted feature values, performing topic classification and sentiment analysis on social media data to obtain a text feature vector, aligning the weighted feature values and the text feature vector in the time dimension according to the timestamp mark in the spatio-temporal feature sequence, and constructing a multi-dimensional feature tensor;

[0101] Step S302: performing dimensionality reduction processing on the multi-dimensional feature tensor by using a tensor decomposition algorithm to obtain a feature subspace, mapping heterogeneous data in the feature subspace to a unified feature space based on a local linear embedding algorithm, and generating a fusion data set.

[0102] Optionally, the embodiment first intelligently weights the spatio-temporal feature sequence according to the feature weight coefficients obtained in the previous steps. An adaptive weighting algorithm is used to dynamically adjust the weight values according to the feature importance of different time periods and different regions. For example, during the peak period of passenger flow, the weight of the people flow density feature will be correspondingly increased; during the promotion activity period, the weight of the sales conversion rate feature will be strengthened. This dynamic weight adjustment mechanism can better reflect the actual importance of features in different scenarios.

[0103] For the processing of social media data, the embodiment designs a multi-level text analysis framework. First, an improved LDA topic model is used for topic classification, and a time decay factor is introduced to make the model focus more on the trend of recent topics. In terms of sentiment analysis, a sentiment classifier for the retail scene is constructed by combining the BERT pre-training model and domain adaptation fine-tuning. This classifier not only identifies basic sentiment polarity, but also captures consumers' fine-grained evaluations of specific goods and services.

[0104] In the time alignment process, the embodiment innovatively proposes a flexible time window mechanism. Considering the delay in publishing social media data, a variable-length time window is designed, and the window size is optimized through a dynamic programming algorithm to ensure the capture of the temporal dependence between data. At the same time, a time decay function is introduced to make the model focus more on the characteristics of recent data.

[0105] In the construction of multi-dimensional feature tensors, the embodiment adopts a hierarchical tensor structure. The first layer represents the time dimension, including different time scale features; the second layer represents the spatial dimension, including location and regional features; and the third layer represents the attribute dimension, including sensor data and text features. Through this multi-level tensor structure, the complex association between data is effectively preserved.

[0106] To solve the problem of computational efficiency of high-dimensional feature tensors, the embodiment uses an improved Tucker decomposition algorithm for dimension reduction. By introducing sparse constraints and non-negative constraints, the interpretability of the decomposition results is improved. At the same time, a regularization term based on the kernel norm is designed to effectively suppress the influence of noise and outliers, extracting a more stable feature subspace.

[0107] In the feature space mapping process of heterogeneous data, the embodiment improves the traditional local linear embedding algorithm. First, an adaptive neighbor graph construction method is used to dynamically determine the number of neighbors according to the local distribution characteristics of the data. Then, a manifold regularization term is introduced to maintain the local structure characteristics of the data during dimension reduction. Finally, a multi-view learning framework is used to fuse different source feature representations.

[0108] The embodiment innovatively introduces a feature consistency verification mechanism. By calculating the mutual information and conditional entropy between different feature spaces, the quality of feature mapping is evaluated. When inconsistencies are found, the mapping parameters are optimized through a feedback adjustment mechanism to ensure that the fused features have good representation ability.

[0109] In the generation process of the fused data set, the embodiment uses a hierarchical sampling strategy to ensure the representativeness and balance of the data. Different normalization methods are used for different types of features to make the feature value distribution more reasonable. At the same time, data augmentation techniques are used to expand the sample space and improve the generalization ability of the model.

[0110] Through the above innovative technical solutions, the embodiment realizes efficient fusion of multi-source heterogeneous data. The scheme can effectively handle the heterogeneity of spatio-temporal features and text features, and generate a fusion dataset with strong expression ability through intelligent feature extraction and mapping mechanism. In practical applications, the scheme significantly improves the quality and efficiency of data fusion, providing a reliable data foundation for subsequent analysis and decision-making. At the same time, the scalable design of the scheme enables it to adapt to different scales and types of data fusion requirements, with wide application value.

[0111] In an embodiment of the wisdom store data processing method based on multi-dimensional data analysis of the present application, referring to Figure 4 , the following contents can also be specifically included:

[0112] Step S401: The missing values in the fusion dataset are supplemented by K-neighbor interpolation method, outlier detection algorithm is used to remove abnormal data points, numerical value type features are normalized by maximum and minimum value, category type features are converted by one-hot encoding to obtain sample data, and the sample data is randomly divided into training dataset and validation dataset according to the ratio of 8:2;

[0113] Step S402: Spatial features in the training dataset are extracted by using convolutional neural network, the extracted spatial features are input into recurrent neural network for time series modeling, initial product recommendation model is obtained by optimizing network parameters through back propagation algorithm, accuracy, recall rate and F1 score of the model are calculated using the validation dataset, learning rate, batch size and network layer number of the initial product recommendation model are optimized based on grid search algorithm, and sliding window method is used to update model parameters by incrementally training new data samples.

[0114] Optionally, the embodiment first comprehensively processes the data quality problems in the fusion dataset. In the missing value filling link, an improved K-neighbor interpolation algorithm is used, and through a dynamic weight calculation method, the most similar K neighbors are determined according to the distance relationship of the sample in the feature space. Different similarity measurement methods are used for different types of features, such as using Euclidean distance for numerical value type features and using edit distance for category type features, to ensure the rationality of the filled values.

[0115] In the aspect of anomaly value detection, the embodiment designs a multi-level detection mechanism. First, the quartile range of the feature is calculated based on statistical methods to identify global outliers. Then, local anomaly points are detected by the isolation forest algorithm, which calculates the anomaly score of the sample by constructing a random decision tree. Finally, a reasonable threshold is set in combination with domain knowledge to remove abnormal samples.

[0116] For feature normalization processing, the embodiment adopts an adaptive normalization strategy. For numerical variables with different distribution characteristics, the corresponding normalization method is selected: for long-tail distribution characteristics, log transformation is adopted before normalization, and for normal distribution characteristics, maximum and minimum value normalization is directly performed. For category features, considering the correlation between features, an improved one-hot encoding method is adopted to reduce the encoding dimension through feature combination.

[0117] In the data set division link, the embodiment innovatively proposes a stratified sampling strategy. First, the data set is divided into multiple time windows in chronological order, and then random sampling is performed in each window according to the ratio of 8:2, ensuring that the training set and the validation set have similar time distribution characteristics. At the same time, considering the uneven distribution of commodity categories, an oversampling technique is used to balance the number of samples of each category.

[0118] The embodiment designs an end-to-end deep learning framework for commodity recommendation. In the spatial feature extraction module, an improved convolutional neural network structure is adopted to capture spatial dependencies of different ranges through multi-scale convolution kernels. In order to deal with the sparsity problem of spatial features, an attention mechanism is introduced to highlight the feature representation of important areas.

[0119] In the time series modeling link, the embodiment selects a bidirectional LSTM network structure to fully capture the dynamic characteristics of user behavior sequences through forward and backward time series information transmission. Considering the difficulty of long sequence modeling, a time gating mechanism is introduced to adaptively adjust the influence weight of historical information according to the time interval.

[0120] For the model training process, the embodiment adopts a multi-task learning framework to simultaneously optimize the click prediction and purchase conversion targets. By designing a dynamic weight loss function, the importance of each task is adaptively adjusted during the training process. During backpropagation optimization, gradient clipping technology is used to prevent gradient explosion, and the Warmup strategy is used to stabilize the learning rate change.

[0121] In terms of model optimization, the embodiment designs a two-stage optimization strategy. First, the rough range of key hyperparameters is determined through grid search, and then the Bayesian optimization method is used to find the optimal configuration in the fine-grained space. The optimization target considers accuracy, recall rate and F1 score, and balances the importance of each index through weighted method.

[0122] In order to adapt to the dynamic changes of data distribution, the embodiment realizes the online updating mechanism of the model. The sliding window method is used to manage the training samples, and when the number of new samples reaches the preset threshold, the incremental training is triggered. Through the transfer learning technology, the parameters of the trained model are used to initialize the new model, accelerating the model convergence process.

[0123] The innovation scheme of the embodiment effectively solves the data quality problem and model optimization problem in commodity recommendation. Through data preprocessing and deep learning model of the system, accurate modeling of user shopping behavior is realized. In practical application, the scheme can accurately capture the dynamic changes of user interest, provide personalized commodity recommendation, and significantly improve user shopping experience and platform operation efficiency.

[0124] In an embodiment of the wisdom store data processing method based on multi-dimensional data analysis of the present application, referring to Figure 5 , the following contents can also be specifically included:

[0125] Step S501: input the user historical behavior data into the updated commodity recommendation model, extract the user interest feature vector, calculate the similarity matrix between the user and the commodity through the collaborative filtering algorithm, sort the commodities according to the similarity score, and select the highest scoring commodities to construct a personalized recommendation list;

[0126] Step S502: divide the fusion data set according to time series according to hours and dates, use the exponential smoothing method to decompose the trend of the traffic and sales data, extract the periodic and seasonal features in the data through the autoregressive model, and combine the long short-term memory network to model and analyze the decomposed time series data to obtain store operation trend data.

[0127] Optionally, in the personalized recommendation link, the embodiment first deeply analyzes the user historical behavior data. By constructing a user behavior sequence including multi-dimensional behavior data such as browsing, collecting, adding to shopping cart, and purchasing, different weights are given to different behavior types using an attention mechanism. For example, the weight of the purchase behavior is higher than that of the browsing behavior, and the weight of the recent behavior is higher than that of the long-term behavior. This differentiated weight allocation can more accurately depict the real interest preferences of the user.

[0128] In the user interest feature extraction process, the embodiment designs a multi-level feature learning framework. First, the bottom-level feature representation is constructed through the attribute features of the commodity (such as category, brand, price interval, etc.), and then the high-level semantic features are extracted using the deep neural network in the updated recommendation model. Through the residual connection mechanism, the features of different levels are fused to obtain a comprehensive user interest representation.

[0129] For user-commodity similarity calculation, the embodiment proposes an improved collaborative filtering algorithm. First, a bipartite graph structure is constructed based on the user's historical behavior, and the high-order association relationship between the user and the commodity is captured through the graph neural network. At the same time, a time decay factor is introduced, so that the model pays more attention to the recent interest changes of the user. In the similarity calculation, the similarity of the commodity content features and the user behavior sequence is considered comprehensively, and the final similarity score is obtained through weighted fusion.

[0130] In the commodity ranking link, the embodiment adopts a multi-objective ranking strategy. In addition to considering the similarity score, real-time heat, inventory status, and promotion information of the commodity are introduced as multiple factors. By designing a dynamic weight ranking function, the importance of each factor is adaptively adjusted in different scenarios to ensure the diversity and timeliness of the recommendation results.

[0131] For store operation trend analysis, the embodiment innovatively proposes a multi-scale time series decomposition method. First, the fused data set is layered according to the hour and date granularities, and the trend of the traffic and sales data is decomposed by an improved exponential smoothing algorithm. This algorithm can better handle noise and abnormal fluctuations in the data by adaptively adjusting the smoothing parameter.

[0132] In terms of periodic and seasonal feature extraction, the embodiment designs an autoregressive model based on Fourier transform. Through frequency domain analysis, the main periodic components in the data are identified, such as weekday-weekend cycles, holiday effects, etc. At the same time, external factors (such as weather, promotion activities) are introduced as auxiliary features to improve the prediction accuracy of the model for special periods.

[0133] In the time series modeling link, the embodiment uses an improved long short-term memory network structure. By designing multiple memory units, different time scale dependencies are captured. The short-term memory unit focuses on the hour-level fluctuation features, and the long-term memory unit learns the date-level trend features. Through the gating mechanism, the weights of the features of different time scales are dynamically adjusted to effectively model the multi-scale time series patterns.

[0134] Through the above innovative solutions, the embodiment realizes accurate personalized commodity recommendation and reliable store operation trend prediction. In practical applications, this scheme can accurately grasp the dynamic changes of user interest and provide commodity recommendations that meet user needs, while accurately predicting store operation trends to provide data support for business operation strategies. The adaptive nature of this scheme enables it to flexibly respond to recommendation needs in different scenarios, making it highly practical.

[0135] During model application, the embodiment continuously optimizes model performance through an online learning mechanism. Model parameters are updated based on real-time feedback data to ensure that the recommendation system can adapt to changes in user interest and dynamic characteristics of the market environment in a timely manner, continuously providing high-quality recommendation services and accurate trend prediction.

[0136] In an embodiment of the multi-dimensional data analysis-based intelligent store data processing method of the present application, referring to Figure 6 It can also specifically include the following content:

[0137] Step S601: Based on the product categories in the personalized product recommendation list and the sales prediction value, in combination with the customer flow fluctuation law and consumer portrait distribution in the store operation trend data, a decision tree algorithm is used to construct a product promotion rule library, and an initial scheme of a marketing strategy including promotion intensity, promotion period and target customer group is generated through rule matching;

[0138] Step S602: The initial scheme of the marketing strategy is converted into a strategy feature vector, which is input into a marketing decision optimization model constructed based on reinforcement learning, the strategy parameters are explored through a Monte Carlo tree search method, the strategy revenue value is calculated based on historical marketing effect data, and the strategy gradient algorithm is used to maximize and optimize the revenue value to obtain a personalized marketing strategy.

[0139] Optionally, the embodiment first uses a multi-level decision analysis framework for the construction of product promotion rules. By analyzing the product features in the personalized product recommendation list, including product categories, price intervals, brand positioning and other dimensions, a product basic portrait is constructed in combination with the sales prediction value. On this basis, the customer flow fluctuation law in the store operation trend data, such as the customer flow difference between weekdays and weekends, holiday effects and other time sequence characteristics, and the portrait features of consumers such as age, consumption ability and shopping preference are fused to form a complete decision input.

[0140] In the design of the decision tree algorithm, the embodiment uses an improved CART decision tree structure. Each node of the decision tree represents a judgment condition of a promotion rule, such as product category matching, customer flow threshold, target customer group features, etc. The optimal split feature is selected through the information gain ratio criterion to ensure that the generated rules have strong discrimination. At the same time, a rule confidence evaluation mechanism is introduced to analyze the historical promotion effect data and select rule combinations with high reliability.

[0141] For the specific content of the promotion rules, the embodiment designs a multi-dimensional strategy combination scheme. The promotion intensity is set considering the product gross profit rate and the inventory turnover rate, and a step-by-step discount scheme is adopted. The promotion period is selected based on the customer flow prediction result, and the customer flow peak period is preferred. The determination of the target customer group combines the matching degree of product characteristics and consumer portrait to realize accurate customer group targeting.

[0142] In the optimization of the marketing strategy, the embodiment innovatively applies the reinforcement learning method. First, the initial scheme of the marketing strategy is converted into a multi-dimensional feature vector, including promotion parameters, time windows, customer group features and other elements. By designing a reasonable feature encoding method, the completeness and learnability of the strategy representation are ensured.

[0143] The embodiment adopts a double network architecture in the construction of the reinforcement learning model. The value network is responsible for evaluating the expected return of the strategy, and the policy network is responsible for generating optimized strategy parameters. Through the Monte Carlo tree search method, multi-step exploration is performed in the policy space to simulate the execution effect of different strategy combinations. During the exploration process, multiple evaluation indicators are considered, such as sales increase, inventory digestion rate, customer satisfaction, etc., and the strategy effect is comprehensively evaluated through a weighted method.

[0144] In the strategy optimization process, the embodiment adopts an improved policy gradient algorithm. By designing a reward function based on time difference, short-term sales targets and long-term customer value are included in the optimization target. At the same time, an experience replay mechanism is introduced to store historical excellent strategy samples and improve learning efficiency. In order to balance the relationship between exploration and utilization, an ε-greedy strategy is used to adaptively adjust the exploration probability.

[0145] The strategy optimization of the embodiment also includes a constraint processing mechanism. By setting a safety threshold, it ensures that the optimized promotion intensity does not excessively affect the product profit. Through time window restriction, it avoids excessive frequency of promotion activities. Through guest group overlap degree control, it prevents mutual interference between different promotion activities.

[0146] Through the above-mentioned innovative scheme, the embodiment realizes intelligent marketing strategy optimization. In actual application, the scheme can generate personalized promotion schemes according to the characteristics of goods, market environment and consumer behavior characteristics. Through continuous optimization of reinforcement learning, the promotion strategy can continuously adapt to market changes and improve marketing effect. At the same time, the explainability design of the scheme enables merchants to understand and adjust the optimization results, which has strong practicality and controllability.

[0147] The scheme has shown significant advantages in actual operation, which can effectively improve the sales of goods, optimize the inventory structure, and improve customer satisfaction, creating greater business value for merchants. Through continuous data accumulation and model optimization, the precision and effect of marketing strategies will continuously improve, forming a virtuous optimization cycle.

[0148] In an embodiment of the method for processing data of a smart store based on multi-dimensional data analysis, see Figure 7 , the method can further include the following content:

[0149] Step S701: Hash encryption is performed on the user identification information in the personalized marketing strategy, reversible encryption is performed on the strategy parameters, desensitization is performed on the sensitive product information using data masking technology, the processed strategy data is written into multiple storage nodes of a cloud data center in a distributed storage manner, a strategy distribution channel is established based on a distributed message middleware, and the encrypted marketing strategy data packet is pushed to a message queue server.

[0150] Step S702: Collect sales conversion rate, average order value change and member repurchase rate data of marketing strategies at different time periods from the store terminal, parse the collected data into effect evaluation indicators in a preset format, and send the effect evaluation indicators back to the model optimization module through the data pipeline to establish a sample data pair containing strategy parameters and execution effects for model parameter updates.

[0151] Optionally, in constructing a personalized marketing strategy, this embodiment first performs multi-dimensional analysis based on the product recommendation list and store operation trend data. The rule base constructed using the decision tree algorithm includes key features such as product categories, sales forecasts, customer traffic fluctuations, and consumer profiles. For example, for high-end cosmetics, corresponding promotional plans are formulated by combining weekend peak customer traffic periods and the spending power of target users; for daily necessities, differentiated marketing strategies are designed based on weekday customer traffic characteristics and the characteristics of user groups with high promotional sensitivity.

[0152] In the strategy optimization phase, this embodiment innovatively employs reinforcement learning technology to construct a marketing decision optimization model. First, each dimension of the marketing strategy (promotion intensity, time slot selection, target customer group, etc.) is encoded as a strategy feature vector, serving as the state space of the reinforcement learning model. Through the Monte Carlo tree search method, the model can efficiently explore the vast strategy space and find the optimal strategy combination.

[0153] To ensure the secure storage and distribution of marketing strategies, this embodiment employs a multi-layered data security protection mechanism. User identification information is irreversibly encrypted using secure hash algorithms such as SHA-256, while strategy parameters are reversibly encrypted using the AES encryption algorithm. In the data anonymization phase, dynamic masking technology is used to display different levels of information detail based on the access permissions of different user levels.

[0154] This embodiment employs a distributed storage architecture to ensure high availability and scalability of data. A consistent hashing algorithm is used to distribute the encrypted policy data across multiple storage nodes, achieving load balancing. Simultaneously, the Raft protocol is used to guarantee data consistency, and a multi-replica mechanism ensures data reliability.

[0155] In the strategy distribution phase, this embodiment constructs a high-efficiency strategy distribution channel based on a distributed message middleware. Asynchronous transmission of strategy data is achieved through a message queue service, ensuring reliable message delivery even under network fluctuations. A topic subscription model is adopted, enabling different store terminals to receive relevant marketing strategies on demand.

[0156] The embodiment designs a complete effect evaluation and feedback optimization mechanism. The execution effect data of the marketing strategy is collected in real time at the store terminal, including key indicators such as sales conversion rate, single price change, and member repurchase rate. These indicators are standardized by pre-set data format to ensure data consistency and comparability.

[0157] In the data backhaul link, the embodiment constructs a reliable data pipeline to realize real-time backhaul of effect evaluation indicators through stream processing technology. An incremental synchronization mechanism is adopted to transmit only the changed data, improving transmission efficiency. At the same time, through data compression and batch transmission strategy, the network resource utilization is optimized.

[0158] The embodiment continuously optimizes the model performance through online learning. The sample data composed of strategy parameters and execution effect are input into the model optimization module, and the model parameters are adjusted through gradient update. The model can adaptively learn the optimal strategy in different scenarios to improve the marketing effect.

[0159] Through the above-mentioned innovative scheme, the embodiment realizes intelligent marketing strategy generation and optimization. The scheme can dynamically adjust the marketing strategy according to real-time market feedback to ensure the effectiveness and pertinence of the strategy. At the same time, through a rigorous data security protection mechanism and an efficient distribution system, the safe and reliable execution of the marketing strategy is ensured, providing comprehensive marketing support services for merchants. In practical application, the scheme significantly improves the marketing efficiency, enhances the user purchase experience, and promotes the improvement of the business efficiency of the merchants.

[0160] In order to break through the limitation of traditional single data analysis and provide a comprehensive data processing solution for precise marketing of a smart store, an embodiment of a smart store data processing device based on multi-dimensional data analysis for realizing all or part of the contents of the smart store data processing method based on multi-dimensional data analysis is provided, as shown in Figure 8 , the smart store data processing device based on multi-dimensional data analysis specifically includes the following contents:

[0161] The multi-dimensional data fusion module 10 is used for collecting multi-channel sensor data, intelligent terminal device data and social media data in the store, time stamp aligning and space coordinate mapping the sensor data and the intelligent terminal device data to obtain a space-time feature sequence, performing correlation analysis on the space-time feature sequence to obtain a feature weight coefficient, performing multi-dimensional data fusion on the space-time feature sequence and the social media data based on the feature weight coefficient, and generating a fusion data set;

[0162] The multi-dimensional data processing module 20 is used for data cleaning and standardization processing of the fusion data set to obtain sample data, dividing the sample data into a training data set and a verification data set according to a preset proportion, training a commodity recommendation model based on a convolutional neural network and a recurrent neural network by using the training data set, performing performance evaluation on the trained model by using the verification data set, automatically optimizing model parameters based on the evaluation result and performing incremental learning update, using the updated model for commodity preference prediction to generate an individualized commodity recommendation list, and performing time series analysis on the fusion data set to obtain store operation trend data.

[0163] The intelligent recommendation module 30 is used for generating an initial scheme of a marketing strategy based on the individualized commodity recommendation list and the store operation trend data, inputting the initial scheme of the marketing strategy into a pre-trained marketing decision optimization model for strategy evaluation and optimization to obtain an individualized marketing strategy, storing the individualized marketing strategy to a cloud data center after data desensitization and encryption processing, distributing the individualized marketing strategy to a store terminal for execution through a message queue, and collecting marketing strategy execution effect data for model optimization update.

[0164] From the above description, it can be known that the wisdom store data processing device based on multi-dimensional data analysis provided by the embodiment of the application can align and map the space-time feature sequence by collecting sensor data, intelligent terminal data and social media data. The multi-dimensional data fusion is performed based on the feature weight coefficient, and the commodity recommendation model is constructed by using the convolutional neural network and the recurrent neural network. The system continuously optimizes the model performance by incremental learning, realizes the individualized commodity recommendation and the store operation trend analysis. Meanwhile, the marketing decision optimization model is introduced to intelligently evaluate and dynamically optimize the marketing strategy, and the model is continuously updated through feedback data. The method breaks through the limitation of traditional single data analysis, and provides a comprehensive data processing solution for the precise marketing of the wisdom store.

[0165] From the hardware aspect, in order to break through the limitation of traditional single data analysis and provide a comprehensive data processing solution for the precise marketing of the wisdom store, the application provides an embodiment of an electronic device for implementing all or part of the contents of the wisdom store data processing method based on multi-dimensional data analysis. The electronic device specifically includes the following contents:

[0166] The processor, the memory, the communications interface and the bus; wherein the processor, the memory, the communications interface complete the mutual communication through the bus; the communications interface is used for realizing the information transmission between the wisdom store data processing device based on multidimensional data analysis and the core business system, the user terminal and the related database and other related equipment; the logic controller can be a desktop computer, a tablet computer and a mobile terminal and the like, and the embodiment is not limited thereto. In the embodiment, the logic controller can be implemented with reference to the embodiment of the wisdom store data processing method based on multidimensional data analysis and the embodiment of the wisdom store data processing device based on multidimensional data analysis, the contents of which are incorporated herein, and the repeated parts will not be described herein.

[0167] It can be understood that the user terminal can include a smart phone, a tablet electronic device, a network set-top box, a portable computer, a desktop computer, a personal digital assistant (PDA), a vehicle-mounted device, a smart wearable device, etc. The smart wearable device can include smart glasses, a smart watch, a smart bracelet, etc.

[0168] In actual application, part of the wisdom store data processing method based on multidimensional data analysis can be executed on the electronic device as described above, or all operations can be completed in the client device. Specifically, the selection can be made according to the processing capability of the client device and the limitation of the user's use scenario, etc. The present application is not limited thereto. If all operations are completed in the client device, the client device can further include a processor.

[0169] The client device described above can have a communication module (i.e. a communication unit) and can be communicatively connected with a remote server to realize data transmission with the server. The server can include a server of the task scheduling center side, and the server of the intermediate platform can also be included in other implementation scenarios, such as the server of the third-party server platform communicatively connected with the server of the task scheduling center. The server can include a single computer device, a server cluster composed of multiple servers, or a distributed server structure.

[0170] Figure 9 A schematic block diagram of the system configuration of the electronic device 9600 of the embodiment of the present application is shown in FIG. 9. As shown in FIG. 9, the electronic device 9600 can include a central processor 9100 and a memory 9140; the memory 9140 is coupled to the central processor 9100. It is worth noting that the structure shown in FIG. 9 is exemplary; other types of structures can also be used to supplement or replace the structure to realize telecommunication functions or other functions. Figure 9 Figure 9 The structure shown in FIG. 9 is exemplary; other types of structures can also be used to supplement or replace the structure to realize telecommunication functions or other functions.​

[0171] In an embodiment, the smart store data processing method based on multi-dimensional data analysis can be integrated into the central processor 9100. The central processor 9100 can be configured to perform the following control:

[0172] Step S101: Collecting multi-channel sensor data, intelligent terminal device data and social media data in the store, time stamp aligning and space coordinate mapping the sensor data and the intelligent terminal device data to obtain a spatio-temporal feature sequence, performing correlation analysis on the spatio-temporal feature sequence to obtain a feature weight coefficient, performing multi-dimensional data fusion on the spatio-temporal feature sequence and the social media data based on the feature weight coefficient, and generating a fusion data set;

[0173] Step S102: Data cleaning and standardization processing the fusion data set to obtain sample data, dividing the sample data into a training data set and a validation data set according to a preset ratio, training a product recommendation model constructed based on a convolutional neural network and a recurrent neural network using the training data set, evaluating the performance of the trained model using the validation data set, automatically tuning the model parameters based on the evaluation results and performing incremental learning updates, using the updated model for product preference prediction to generate an individualized product recommendation list, and performing time series analysis on the fusion data set to obtain store operation trend data;

[0174] Step S103: Generating an initial marketing strategy scheme based on the individualized product recommendation list and the store operation trend data, inputting the initial marketing strategy scheme into a pre-trained marketing decision optimization model for strategy evaluation and optimization to obtain an individualized marketing strategy, storing the individualized marketing strategy to a cloud data center after data desensitization and encryption processing, distributing the individualized marketing strategy to the store terminal for execution through a message queue, and collecting marketing strategy execution effect data for model optimization updates.

[0175] As can be seen from the above description, the electronic device provided by the embodiments of the present application innovatively realizes the alignment and mapping of the spatio-temporal feature sequence by collecting sensor, intelligent terminal and social media data. Multi-dimensional data fusion is performed based on the feature weight coefficient, and a product recommendation model is constructed using a convolutional neural network and a recurrent neural network. The system continuously optimizes the performance of the model through incremental learning, realizes individualized product recommendation and store operation trend analysis. At the same time, a marketing decision optimization model is introduced to intelligently evaluate and dynamically optimize the marketing strategy, and the model is continuously updated through feedback data. This method breaks through the limitations of traditional single data analysis and provides a comprehensive data processing solution for precise marketing of smart stores.

[0176] In another embodiment, the smart store data processing device based on multi-dimensional data analysis can be configured separately from the central processor 9100, for example, the smart store data processing device based on multi-dimensional data analysis can be configured as a chip connected with the central processor 9100, and the function of the smart store data processing method based on multi-dimensional data analysis is realized through the control of the central processor.

[0177] As shown in Figure 9 , the electronic device 9600 can also include a communication module 9110, an input unit 9120, an audio processor 9130, a display 9160, and a power supply 9170. It should be noted that the electronic device 9600 does not necessarily include all the components shown in Figure 9 ; in addition, the electronic device 9600 can also include components not shown in Figure 9 , which can be referred to prior art.

[0178] As shown in Figure 9 , the central processor 9100, also known as a controller or operation control, can include a microprocessor or other processor device and / or logic device, which receives input and controls the operation of various components of the electronic device 9600.

[0179] The memory 9140, for example, can be one or more of a buffer, a flash memory, a hard drive, a removable media, a volatile memory, a non-volatile memory, or other suitable device. Information related to failure can be stored, and in addition, programs for executing related information can also be stored. The central processor 9100 can execute the programs stored in the memory 9140 to realize information storage or processing, etc.

[0180] The input unit 9120 provides input to the central processor 9100. The input unit 9120 is, for example, a key or touch input device. The power supply 9170 is used to provide power to the electronic device 9600. The display 9160 is used to display display objects such as images and text. The display can be, for example, an LCD display, but is not limited thereto.

[0181] The memory 9140 can be a solid state memory such as a read only memory (ROM), a random access memory (RAM), a SIM card, or the like. It can also be a memory that retains information even when power is off, can be selectively erased, and is provided with more data, an example of which is sometimes referred to as an EPROM or the like. The memory 9140 can also be some other type of device. The memory 9140 includes a buffer memory 9141 (sometimes referred to as a buffer). The memory 9140 can include an application / function storage 9142 for storing application programs and function programs or for storing a flow for executing an operation of the electronic device 9600 by the central processing unit 9100.

[0182] The memory 9140 can also include a data storage 9143 for storing data such as contacts, digital data, pictures, sounds, and / or any other data used by the electronic device. A driver storage 9144 of the memory 9140 can include various drivers of the electronic device for a communication function and / or for performing other functions of the electronic device such as a messaging application, an address book application, and the like.

[0183] The communication module 9110 is a transmitter / receiver that transmits and receives signals via an antenna 9111. The communication module 9110 (transmitter / receiver) is coupled to the central processing unit 9100 to provide input signals and receive output signals, which can be the same as in the case of a conventional mobile communication terminal.

[0184] Based on different communication technologies, a plurality of communication modules 9110 such as a cellular network module, a Bluetooth module, and / or a wireless local area network module, and the like can be provided in the same electronic device. The communication module 9110 (transmitter / receiver) is also coupled to a speaker 9131 and a microphone 9132 via an audio processor 9130 to provide audio output via the speaker 9131 and receive audio input from the microphone 9132, thereby enabling a conventional telecommunication function. The audio processor 9130 can include any suitable buffer, decoder, amplifier, and the like. In addition, the audio processor 9130 is also coupled to the central processing unit 9100, thereby enabling recording on the local by the microphone 9132 and playing a sound stored on the local by the speaker 9131.

[0185] Embodiments of this application also provide a computer-readable storage medium capable of implementing all steps of the smart store data processing method based on multidimensional data analysis, where the execution subject is a server or client, as described in the above embodiments. The computer-readable storage medium stores a computer program that, when executed by a processor, implements all steps of the smart store data processing method based on multidimensional data analysis, where the execution subject is a server or client, as described in the above embodiments. For example, when the processor executes the computer program, it implements the following steps:

[0186] Step S101: Collect data from multiple sensors, smart terminal devices, and social media within the store; perform time stamp alignment and spatial coordinate mapping on the sensor data and smart terminal device data to obtain a spatiotemporal feature sequence; perform correlation analysis on the spatiotemporal feature sequence to obtain feature weight coefficients; and perform multidimensional data fusion on the spatiotemporal feature sequence and the social media data based on the feature weight coefficients to generate a fused dataset.

[0187] Step S102: The fused dataset is cleaned and standardized to obtain sample data. The sample data is divided into training dataset and validation dataset according to a preset ratio. The product recommendation model based on convolutional neural network and recurrent neural network is trained using the training dataset. The performance of the trained model is evaluated using the validation dataset. The model parameters are automatically tuned and incrementally updated based on the evaluation results. The updated model is used to predict product preferences and generate a personalized product recommendation list. The fused dataset is analyzed over time to obtain store operation trend data.

[0188] Step S103: Generate an initial marketing strategy plan based on the personalized product recommendation list and the store operation trend data. Input the initial marketing strategy plan into a pre-trained marketing decision optimization model for strategy evaluation and optimization to obtain a personalized marketing strategy. After data anonymization and encryption of the personalized marketing strategy, store it in the cloud data center. Distribute the personalized marketing strategy to the store terminal for execution through a message queue. Collect marketing strategy execution effect data for model optimization and updates.

[0189] As can be seen from the above description, the computer readable storage medium provided by the embodiment of the application realizes the alignment and mapping of the space-time feature sequence innovatively by collecting sensor data, intelligent terminal data and social media data. The multi-dimensional data fusion is performed based on the feature weight coefficient, and the commodity recommendation model is constructed by using the convolutional neural network and the recurrent neural network. The system continuously optimizes the model performance by incremental learning, realizes the personalized commodity recommendation and the store operation trend analysis. Meanwhile, the marketing decision optimization model is introduced, the marketing strategy is intelligently evaluated and dynamically optimized, and the model is continuously updated by feedback data. The method breaks through the limitation of traditional single data analysis, and provides a comprehensive data processing solution for the precise marketing of the smart store.

[0190] The embodiment of the application also provides a computer program product capable of realizing all steps of the smart store data processing method based on multi-dimensional data analysis in which the execution subject in the above embodiment is a server or a client. The computer program / instruction is executed by a processor to realize the steps of the smart store data processing method based on multi-dimensional data analysis, for example, the computer program / instruction realizes the following steps:

[0191] Step S101: collecting multi-channel sensor data, intelligent terminal device data and social media data in a store, performing timestamp alignment and space coordinate mapping on the sensor data and the intelligent terminal device data to obtain a space-time feature sequence, performing correlation analysis on the space-time feature sequence to obtain a feature weight coefficient, performing multi-dimensional data fusion on the space-time feature sequence and the social media data based on the feature weight coefficient, and generating a fusion data set;

[0192] Step S102: performing data cleaning and standardization processing on the fusion data set to obtain sample data, dividing the sample data into a training data set and a verification data set according to a preset ratio, training a commodity recommendation model constructed based on a convolutional neural network and a recurrent neural network by using the training data set, evaluating the performance of the trained model by using the verification data set, automatically tuning the model parameters based on the evaluation result and performing incremental learning update, using the updated model to generate a personalized commodity recommendation list for commodity preference prediction, and performing time series analysis on the fusion data set to obtain store operation trend data;

[0193] Step S103: generating a marketing strategy initial scheme based on the personalized commodity recommendation list and the store operation trend data, inputting the marketing strategy initial scheme into a pre-trained marketing decision optimization model to perform strategy evaluation and optimization to obtain a personalized marketing strategy, performing data desensitization and encryption processing on the personalized marketing strategy and then storing the personalized marketing strategy to a cloud data center, distributing the personalized marketing strategy to a store terminal for execution through a message queue, and collecting marketing strategy execution effect data for model optimization update.

[0194] From the above description, the computer program product provided by the embodiment of the present application innovatively realizes alignment and mapping of the spatio-temporal feature sequence by collecting sensor data, intelligent terminal data and social media data. Multi-dimensional data fusion is performed based on feature weight coefficients, and a commodity recommendation model is constructed by using a convolutional neural network and a recurrent neural network. The system continuously optimizes the model performance by incremental learning, realizes personalized commodity recommendation and shop operation trend analysis. Meanwhile, a marketing decision optimization model is introduced to intelligently evaluate and dynamically optimize the marketing strategy, and the model is continuously updated through feedback data. The method breaks through the limitation of traditional single data analysis and provides a comprehensive data processing solution for precise marketing of a smart shop.

[0195] Those skilled in the art will understand that the embodiments of the present application can be provided as a method, device or computer program product. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer usable program code.

[0196] The present application is described with reference to flowcharts and / or block diagrams of the method, device (apparatus) and computer program product according to the embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of the flows and / or blocks in the flowcharts and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to produce a machine, so that the instructions executed by the computer or other programmable data processing device produce a device that implements the functions specified in the flowcharts and / or block diagrams. Figure 1 The functions specified in one or more flows and / or blocks Figure 1 The device that implements the functions specified in one or more flows and / or blocks.

[0197] These computer program instructions can also be stored in a computer readable storage medium that can guide the computer or other programmable data processing device to work in a specific way, so that the instructions stored in the computer readable storage medium produce a product including instruction devices that implement the functions specified in the flowcharts and / or block diagrams. Figure 1 The functions specified in one or more flows and / or blocks Figure 1 The device that implements the functions specified in one or more flows and / or blocks.

[0198] These computer program instructions can also be loaded into a computer or other programmable data processing devices, so that a series of operational steps are generated to realize the computer-implemented processes, and the instructions executed on the computer or other programmable devices provide a process for implementing the functions specified in the flowchart Figure 1 one flow or multiple flows and / or the functions specified in the block Figure 1 Figure 1 one block or multiple blocks.

[0199] The principles and implementation manners of the present application are described in the specific embodiments. The above description of the embodiments is only used to help understand the method and core idea of the present application; meanwhile, for those skilled in the art, according to the idea of the present application, the specific implementation manners and application ranges can be changed, and the above description of the present application should not be understood as a limitation.

Claims

1. A smart store data processing method based on multidimensional data analysis, characterized in that, The method includes: Data from multiple sensors, smart terminal devices, and social media within the store is collected. A spatiotemporal reference coordinate system is established based on the collection time and device location information in the sensor data. The sensor data is mapped to a time axis according to the timestamps of the collection timestamps. The device numbers in the smart terminal device data are converted into corresponding spatial coordinate points and mapped to a spatial axis. Data at different sampling frequencies is resampled using bilinear interpolation to generate spatiotemporal feature sequences. The Pearson correlation coefficient is used to calculate the correlation between each feature in the spatiotemporal feature sequence, constructing a feature correlation matrix. Based on the feature correlation matrix, the correlation between each feature and the target is calculated. The contribution of the variable is normalized to obtain the feature weight coefficient. The feature weight coefficient is then weighted with the corresponding spatiotemporal feature sequence to obtain the weighted feature value. Social media data is classified into topics and sentiment analysis to obtain text feature vectors. The weighted feature value is aligned with the text feature vector in the time dimension according to the timestamp in the spatiotemporal feature sequence to construct a multidimensional feature tensor. The multidimensional feature tensor is reduced in dimensionality using a tensor decomposition algorithm to obtain a feature subspace. Based on the local linear embedding algorithm, the heterogeneous data in the feature subspace is mapped to a unified feature space to generate a fused dataset. The fused dataset is cleaned and standardized to obtain sample data. This sample data is then divided into training and validation datasets according to a preset ratio. The training dataset is used to train a product recommendation model based on convolutional neural networks and recurrent neural networks. The validation dataset is used to evaluate the performance of the trained model. Based on the evaluation results, the model parameters are automatically tuned and incrementally updated. User historical behavior data is input into the updated product recommendation model to extract user interest feature vectors. A similarity matrix between users and products is calculated using a collaborative filtering algorithm. Products are sorted according to similarity scores, and the highest-scoring products are selected to construct a personalized product recommendation list. The fused dataset is further divided into time series data by hour and date. Exponential smoothing is used to decompose the customer traffic and sales data into trends. An autoregressive model is used to extract periodic and seasonal features from the data. Long Short-Term Memory (LSTM) networks are then used to model and analyze the decomposed time series data to obtain store operation trend data. Based on the personalized product recommendation list and the store operation trend data, an initial marketing strategy plan is generated. The initial marketing strategy plan is then input into a pre-trained marketing decision optimization model for strategy evaluation and optimization to obtain a personalized marketing strategy. The personalized marketing strategy is then anonymized and encrypted before being stored in a cloud data center. The personalized marketing strategy is then distributed to store terminals for execution via a message queue. Data on the execution effect of the marketing strategy is collected for model optimization and updates.

2. The smart store data processing method based on multidimensional data analysis according to claim 1, characterized in that, The process involves cleaning and standardizing the fused dataset to obtain sample data. This sample data is then divided into training and validation datasets according to a preset ratio. The training dataset is used to train a product recommendation model built on convolutional neural networks and recurrent neural networks. The validation dataset is used to evaluate the performance of the trained model. Based on the evaluation results, the model parameters are automatically tuned and incrementally updated, including: Missing values ​​in the fused dataset are supplemented using K-nearest neighbor interpolation, outlier detection algorithms are used to remove abnormal data points, numerical features are normalized to their maximum and minimum values, and categorical features are converted to sample data using one-hot encoding. The sample data is then randomly divided into training and validation datasets in an 8:2 ratio. Spatial features in the training dataset are extracted using a convolutional neural network. These features are then input into a recurrent neural network for temporal modeling. The network parameters are optimized using a backpropagation algorithm to obtain an initial product recommendation model. The accuracy, recall, and F1 score of the model are calculated using the validation dataset. The learning rate, batch size, and number of network layers of the initial product recommendation model are tuned using a grid search algorithm. The model parameters are then incrementally trained and updated using a sliding window approach with new data samples.

3. The smart store data processing method based on multidimensional data analysis according to claim 1, characterized in that, The process of generating an initial marketing strategy based on the personalized product recommendation list and the store operation trend data, and then inputting the initial marketing strategy into a pre-trained marketing decision optimization model for strategy evaluation and optimization to obtain a personalized marketing strategy includes: Based on the product categories and sales forecasts in the personalized product recommendation list, and combined with the customer traffic fluctuation patterns and consumer profile distribution in the store operation trend data, a product promotion rule base is constructed using a decision tree algorithm. Through rule matching, an initial marketing strategy plan containing promotion intensity, promotion period, and target customer group is generated. The initial marketing strategy is transformed into a strategy feature vector and input into a marketing decision optimization model built based on reinforcement learning. The strategy parameters are explored using the Monte Carlo tree search method, the strategy benefit value is calculated based on historical marketing performance data, and the strategy gradient algorithm is used to maximize the benefit value to obtain a personalized marketing strategy.

4. The smart store data processing method based on multidimensional data analysis according to claim 1, characterized in that, The personalized marketing strategy is anonymized and encrypted before being stored in a cloud data center. It is then distributed to store terminals via a message queue for execution. Data on the effectiveness of the marketing strategy execution is collected for model optimization and updates, including: The user identification information in the personalized marketing strategy is hashed and encrypted, the strategy parameters are reversibly encrypted, sensitive product information is desensitized using data masking technology, the processed strategy data is written to multiple storage nodes in the cloud data center in a distributed storage manner, a strategy distribution channel is established based on a distributed message middleware, and the encrypted marketing strategy data package is pushed to the message queue server. Data on sales conversion rate, average order value change, and member repurchase rate of marketing strategies are collected from store terminals at different times. The collected data is parsed into performance evaluation indicators in a preset format. The performance evaluation indicators are then sent back to the model optimization module through a data pipeline to establish sample data pairs containing strategy parameters and execution effects for updating model parameters.

5. A smart store data processing device based on multidimensional data analysis, characterized in that, The device includes: The multi-dimensional data fusion module is used to collect data from multiple sensors, smart terminal devices, and social media within the store. Based on the acquisition time and device location information in the sensor data, a spatiotemporal reference coordinate system is established. The sensor data is mapped to a time axis according to the timestamps of the acquisition timestamps. The device numbers in the smart terminal device data are converted into corresponding spatial coordinate points and mapped to the spatial axis. Data at different sampling frequencies is resampled using bilinear interpolation to generate spatiotemporal feature sequences. The Pearson correlation coefficient is used to calculate the correlation degree between features in the spatiotemporal feature sequences, constructing a feature correlation matrix. Based on the feature correlation matrix, the correlation degree between each feature is calculated. The contribution of each feature to the target variable is normalized to obtain feature weight coefficients. The feature weight coefficients are then weighted with the corresponding spatiotemporal feature sequences to obtain weighted feature values. Social media data is classified into topics and sentiment analyzed to obtain text feature vectors. Based on the timestamps in the spatiotemporal feature sequences, the weighted feature values ​​are aligned with the text feature vectors in the time dimension to construct a multidimensional feature tensor. The multidimensional feature tensor is then reduced in dimensionality using a tensor decomposition algorithm to obtain a feature subspace. Based on the local linear embedding algorithm, heterogeneous data in the feature subspace are mapped to a unified feature space to generate a fused dataset. The multidimensional data processing module is used to clean and standardize the fused dataset to obtain sample data. The sample data is then divided into training and validation datasets according to a preset ratio. The training dataset is used to train a product recommendation model based on convolutional neural networks and recurrent neural networks. The validation dataset is used to evaluate the performance of the trained model. Based on the evaluation results, the model parameters are automatically tuned and incrementally updated. User historical behavior data is input into the updated product recommendation model to extract user interest feature vectors. A similarity matrix between users and products is calculated using a collaborative filtering algorithm. Products are sorted according to similarity scores, and the highest-scoring products are selected to construct a personalized product recommendation list. The fused dataset is divided into time series by hour and date. Exponential smoothing is used to decompose the customer traffic and sales data into trends. Periodic and seasonal features are extracted from the data using an autoregressive model. Long Short-Term Memory (LSTM) networks are combined to model and analyze the decomposed time series data to obtain store operation trend data. The intelligent recommendation module is used to generate an initial marketing strategy plan based on the personalized product recommendation list and the store operation trend data. The initial marketing strategy plan is then input into a pre-trained marketing decision optimization model for strategy evaluation and optimization to obtain a personalized marketing strategy. The personalized marketing strategy is then anonymized and encrypted before being stored in a cloud data center. The personalized marketing strategy is then distributed to store terminals for execution via a message queue. Data on the execution effect of the marketing strategy is collected for model optimization and updates.

6. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the smart store data processing method based on multidimensional data analysis as described in any one of claims 1 to 4.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the steps of the smart store data processing method based on multidimensional data analysis as described in any one of claims 1 to 4.

Images