An interactive data analysis method and device based on commercial big data

Abstract

This invention provides an interactive data analysis method and apparatus based on commercial big data, relating to the fields of data analysis and intelligent computing. It includes collecting multi-source heterogeneous data from commercial activities, including user behavior data, transaction data, supply chain data, and market environment data. Based on this interactive data analysis method and apparatus for commercial big data, through the collection, preprocessing, and feature extraction of multi-source heterogeneous data, combined with a joint mechanism of deep neural networks and Bayesian inference models, an adaptive analysis method is provided. This method can dynamically select features and hierarchically assign weights to large-scale commercial big data, greatly improving the accuracy and efficiency of data analysis. Through an interactive visual interface, users can operate and adjust the analysis process in real time, thereby generating more accurate analysis results and promoting the scientific and precise nature of decision-making.

Description

Technical Field

[0001] This invention relates to the field of data analysis and intelligent computing technology, specifically to an interactive data analysis method and apparatus based on commercial big data. Background Technology

[0002] With the widespread adoption of the internet, mobile devices, and the Internet of Things (IoT), the scale of data generated in the business sector is rapidly increasing. Enterprises rely on the analysis of large-scale data for market monitoring, customer management, supply chain control, and risk forecasting. Traditional data processing is largely based on data warehousing and statistical modeling, using batch processing to generate reports describing business operations. Common methods include clustering, regression, time series analysis, and association rule analysis, enabling enterprises to perform routine monitoring and trend assessment. With the development of distributed computing and cloud platforms, frameworks such as Hadoop and Spark are being applied to business data analysis, significantly enhancing processing capabilities. Simultaneously, BI platforms, visualization tools, and dashboard systems are maturing, allowing users to view data trends and business characteristics through graphical interfaces. Some platforms have also introduced real-time monitoring capabilities, laying the foundation for business data applications.

[0003] However, existing technologies still have significant shortcomings. Most systems prioritize computing and storage capabilities while neglecting interactivity and intelligence. Users lacking professional backgrounds often struggle to utilize complex models flexibly, frequently relying on static reports or limited filtering functions. Data types are diverse and rapidly updated, but existing tools struggle to support exploratory analysis and deep reasoning, failing to provide intuitive feedback at critical moments. Especially in a rapidly changing business environment, the lack of interactive analytical tools prevents companies from adjusting their strategies in a timely manner. This deficiency directly limits the realization of big data value and is a core issue that urgently needs to be addressed in the current field of business data analytics. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides an interactive data analysis method and apparatus based on commercial big data. The technical problem this invention aims to solve is: how to address the issues of low accuracy, poor real-time performance, and insufficient interactivity in multi-source heterogeneous data analysis in commercial big data through an adaptive analysis method using deep neural networks and Bayesian inference models.

[0005] To achieve the above objectives, the present invention provides the following technical solution: an interactive data analysis method based on commercial big data, comprising: S1. Collect multi-source heterogeneous data from business activities, including user behavior data, transaction data, supply chain data, and market environment data; S2. Preprocess and extract features from the multi-source heterogeneous data. Obtain a data feature set based on information theory indicators through feature extraction. Merge the data feature set with the preprocessed multi-source heterogeneous data based on a dimension-weighted combination method to generate a structured dataset. S3. Input the structured dataset into the adaptive analysis algorithm. The adaptive analysis algorithm is based on the joint mechanism of deep neural network and Bayesian inference model to perform dynamic feature selection and hierarchical weight allocation on the data feature set to generate feature weight distribution. The adaptive analysis algorithm outputs the first analysis result according to the feature weight distribution. S4. Display the first analysis result in an interactive visualization interface, the interactive visualization interface including interactive controls for receiving user interaction operations, the user interaction operations triggering the adaptive analysis algorithm to recalculate the feature weight distribution and output the second analysis result; S5. Divide the second analysis results into indicator sets and generate an interactive analysis report, which is transmitted to the enterprise management system through a standardized data interface.

[0006] Preferably, the preprocessing includes data cleaning, normalization, and format unification, and the information theory index includes the mutual information criterion, which performs feature filtering on the multi-source heterogeneous data.

[0007] Preferably, the hierarchical weight allocation is jointly calculated by the deep neural network and the Bayesian inference model, and the calculation formula is as follows: .

[0008] in, Indicates the first The hierarchical weight values ​​of each feature are dimensionless and range from [value missing]. ,satisfy , Represents input features, dimensionless. This represents the nonlinear mapping output of the deep neural network to the input features, and is dimensionless. Indicates the first The Bayesian prior probabilities corresponding to each feature are dimensionless and range from [value]. , Represents the total number of features, dimensionless. The normalization factor is the weighted average of all features.

[0009] Preferably, the first analysis result output uses a joint loss function, which includes a mean squared error term based on the deep neural network prediction value and a Bayesian inference model. The calculation formula for the joint loss function is as follows: .

[0010] in, This represents the joint loss value, which is dimensionless and ranges from [value missing]. , Represents the true value of the target variable, dimensionless. This represents a predicted value, dimensionless. Represents the sample size, dimensionless. Represents the regularization coefficient, which is dimensionless. Represents the variational distribution. Describe the prior distribution, The dimensionless value represents the Kullback-Leibler divergence between the variational distribution and the prior distribution.

[0011] Preferably, the user interaction operation includes: S41. In the interactive visualization interface, the first analysis result is filtered by dimension to generate feature results; S42. Perform a time window scaling operation on the feature results to generate local data features; S43. Based on the local data features, perform a multidimensional association mapping operation to generate an association feature matrix, feed the association feature matrix back to the adaptive analysis algorithm to correct the Bayesian prior probability and reallocate feature weights, and output the second analysis result.

[0012] Preferably, the indicator set includes a set of key business indicators, a set of risk monitoring indicators, and a set of predictive trend indicators.

[0013] An interactive data analysis device based on commercial big data includes: A multi-source data acquisition module is used to acquire heterogeneous data sources from business activities, including user behavior, transaction flow, supply chain logistics and external market environment; The data preprocessing and feature engineering module is used to clean, transform, and extract features from the heterogeneous data sources to generate structured datasets and data feature sets. An adaptive analysis engine is used to dynamically select features and hierarchically assign weights to the data feature set to generate a feature weight distribution. The adaptive analysis engine generates initial analysis results based on the feature weight distribution. An interactive visualization processing module is used to display the initial analysis results. The interactive visualization processing module provides interactive controls to receive user instructions and feed them back to the adaptive analysis engine. The adaptive analysis engine recalculates the feature weight distribution and generates the final analysis results. The analysis report output module is used to generate an interactive report based on the final analysis results. The interactive report is transmitted to the enterprise management system through a standardized application programming interface.

[0014] Preferably, the adaptive analysis engine includes a deep neural network unit, a Bayesian inference unit, and a weight fusion unit. The deep neural network unit is used to perform nonlinear transformation on the data feature set, the Bayesian inference unit is used to calculate the posterior probability, and the weight fusion unit is used to jointly calculate the feature weight distribution based on the output of the nonlinear transformation and the posterior probability.

[0015] This invention provides an interactive data analysis method and apparatus based on commercial big data. It has the following beneficial effects: This interactive data analysis method and device based on commercial big data provides an adaptive analysis approach by collecting, preprocessing, and extracting features from multi-source heterogeneous data, combined with a deep neural network and Bayesian inference model. It dynamically selects features and assigns hierarchical weights to large-scale commercial big data, significantly improving the accuracy and efficiency of data analysis. Through an interactive visual interface, users can operate and adjust the analysis process in real time, generating more accurate results and promoting scientific and precise decision-making.

[0016] The interactive data analysis method and apparatus of this invention enable enterprises to achieve more flexible and intelligent decision support based on business big data. In terms of risk monitoring and trend prediction, by combining data from user behavior, market environment, and other aspects, the analysis results can be dynamically adjusted to provide accurate key business indicators and risk warnings, providing strong support for enterprise operation and management, and helping enterprises better respond to market changes and improve decision-making efficiency. Attached Figure Description

[0017] Figure 1 This is a flowchart of the overall method of the present invention; Figure 2 This is a schematic diagram of the adaptive analysis engine structure of the present invention; Figure 3 This is a flowchart of the interactive operation process of the present invention; Figure 4 This is a schematic diagram of the structural composition of the device of the present invention; Figure 5 This is a flowchart of the report generation and output process for this invention. Detailed Implementation

[0018] The technical solutions in the embodiments of the present invention have been clearly and completely described. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Example 1 like Figure 1-5 As shown, this embodiment of the invention provides an interactive data analysis method based on commercial big data, including: S1. Collect multi-source heterogeneous data from business activities, including user behavior data, transaction data, supply chain data, and market environment data.

[0020] S2. Preprocessing and feature extraction are performed on the multi-source heterogeneous data. Feature extraction, based on information theory metrics, yields a set of data features. This set is then fused with the preprocessed multi-source heterogeneous data using a dimensionality-weighted combination method to generate a structured dataset. Preprocessing includes data cleaning, normalization, and format unification. Information theory metrics include the mutual information criterion, which is used to select features from the multi-source heterogeneous data.

[0021] S3. Input the structured dataset into the adaptive analysis algorithm. Based on a joint mechanism of deep neural networks and Bayesian inference models, the adaptive analysis algorithm dynamically selects features and assigns hierarchical weights to the data feature set, generating a feature weight distribution. The adaptive analysis algorithm outputs the first analysis result based on the feature weight distribution. The hierarchical weight allocation is jointly calculated by the deep neural network and the Bayesian inference model, and the calculation formula is as follows: .

[0022] in, Indicates the first The hierarchical weight values ​​of each feature are dimensionless and range from [value missing]. ,satisfy , Represents input features, dimensionless. This represents the nonlinear mapping output of a deep neural network to input features; it is dimensionless. Indicates the first The Bayesian prior probabilities corresponding to each feature are dimensionless and range from [value]. , Represents the total number of features, dimensionless. The normalization factor is the weighted average of all features.

[0023] S4. Display the first analysis result in an interactive visualization interface. This interface includes interactive controls for receiving user interactions. User interactions trigger an adaptive analysis algorithm to recalculate the feature weight distribution and output the second analysis result. The first analysis result output uses a joint loss function, which includes a mean squared error term based on deep neural network predictions and a Bayesian inference model. The formula for calculating the joint loss function is: .

[0024] in, This represents the joint loss value, which is dimensionless and ranges from [value missing]. , Represents the true value of the target variable, dimensionless. This represents a predicted value, dimensionless. Represents the sample size, dimensionless. Represents the regularization coefficient, which is dimensionless. Represents the variational distribution. Describe the prior distribution, The Kullback-Leibler divergence between the variational distribution and the prior distribution is dimensionless. User interaction includes: S41. In the interactive visualization interface, the first analysis result is filtered by dimension to generate feature results.

[0025] S42. Perform a time window scaling operation on the feature results to generate local data features.

[0026] S43. Perform a multidimensional association mapping operation based on local data features to generate an association feature matrix. Feed the association feature matrix back to the adaptive analysis algorithm to correct the Bayesian prior probability and redistribute feature weights, and output the second analysis result.

[0027] S5. Divide the second analysis results into indicator sets and generate an interactive analysis report. The interactive analysis report is transmitted to the enterprise management system through a standardized data interface. The indicator sets include a set of key business indicators, a set of risk monitoring indicators, and a set of predictive trend indicators.

[0028] An interactive data analysis device based on commercial big data includes: The multi-source data acquisition module is used to acquire heterogeneous data sources from business activities. These heterogeneous data sources include user behavior, transaction flow, supply chain logistics, and external market environment.

[0029] The data preprocessing and feature engineering module is used to clean, transform, and extract features from heterogeneous data sources, generating structured datasets and sets of data features.

[0030] The adaptive analysis engine is used to dynamically select features and assign hierarchical weights to the data feature set to generate a feature weight distribution. The adaptive analysis engine generates initial analysis results based on the feature weight distribution.

[0031] The adaptive analysis engine includes a deep neural network unit, a Bayesian inference unit, and a weight fusion unit. The deep neural network unit is used to perform nonlinear transformations on the data feature set, the Bayesian inference unit is used to calculate the posterior probability, and the weight fusion unit is used to perform joint calculations based on the output of the nonlinear transformation and the posterior probability to generate the feature weight distribution.

[0032] The interactive visualization processing module is used to display the initial analysis results. It provides interactive controls to receive user commands and feed them back to the adaptive analysis engine. The adaptive analysis engine recalculates the feature weight distribution and generates the final analysis results.

[0033] The analysis report output module is used to generate interactive reports based on the final analysis results. These interactive reports are then transmitted to the enterprise management system via a standardized application programming interface (API).

[0034] Real-time feedback and adjustment support allows for real-time adjustments to analysis results based on user interaction. Through an interactive interface, users can filter features or adjust time windows, and the system will recalculate and provide updated analysis results based on user input. This mechanism enhances the flexibility and accuracy of data analysis.

[0035] By combining deep neural networks and Bayesian inference models, the system can continuously optimize and adjust itself based on new input data while processing existing data.

[0036] The adaptive learning mechanism enables the system to dynamically update feature weights, thereby improving the accuracy of analysis and predictive ability.

[0037] The system can integrate data from multiple sources and perform feature filtering and weighted combination of the data using information theory indicators. Multi-dimensional data fusion not only enhances the depth and breadth of the data but also helps reveal the potential relationships between different data dimensions.

[0038] The system can generate key business indicators, risk monitoring indicators, and predictive trend indicators, supporting enterprises in identifying potential risks and development trends. This is of great significance for enterprise decision support and risk management in a dynamic market environment.

[0039] Through standardized interfaces, the system can automatically generate interactive analysis reports and transmit them to the enterprise management system, achieving automated report generation and delivery. This improves the efficiency of data analysis and makes integration between the system and the enterprise's existing management system easier.

[0040] The steps described above further enhance the applicability of this method in areas such as real-time data analysis, decision support, and risk management, enabling it to effectively improve enterprises' ability to process complex data.

[0041] Example 2 This embodiment is based on an interactive data analysis method and apparatus for commercial big data. It displays the first analysis result through an interactive visual interface. Users trigger an adaptive analysis algorithm to recalculate the feature weight distribution through interactive operations, and then output an optimized second analysis result. The specific implementation is as follows: 1. Presentation of the first analysis results The first analysis result is a preliminary analysis result based on the adaptive analysis algorithm.

[0042] This stage will display the weight of each feature, the model prediction results, and related statistics. Assuming we are using sales data from an e-commerce platform, the features involved include: transaction amount, user behavior frequency, and ad impressions.

[0043] Assume the preliminary analysis results are as shown in the table below: Table 1: Preliminary Analysis Results.

[0044] feature Weight describe Transaction amount 0.50 Transaction amount has a significant impact on sales revenue. User behavior frequency 0.35 The impact of user behavior frequency on conversion rate Ad exposure 0.15 Advertising exposure affects sales. Table 2: Forecast Sales Trend Table.

[0045] month Forecasted sales January 2025 50,000 yuan February 2025 55,000 yuan March 2025 60,000 yuan April 2025 65,000 yuan May 2025 70,000 yuan The above results will be displayed through an interactive visualization interface for users to view.

[0046] 2. User interaction operation The interactive visualization interface offers multiple functional options, enabling users to interact deeply with the data and obtain optimized analytical results. Here are some common interactive operations: Dimension filtering operation: Users can select the features they want to focus on and filter out features they are not interested in.

[0047] Users can select the features they want to view from the drop-down menu on the interface, such as "transaction amount" and "ad impressions".

[0048] The system re-displays the updated chart based on the user's selection, showing only the weights of the selected features and their impact on the prediction results.

[0049] Assuming the user selects "Transaction Amount" and "Ad Impressions," the updated weighting results are as follows: Table 3: System update weight results.

[0050] feature Weight describe Transaction amount 0.60 The impact of transaction amount on sales is increasing. Ad exposure 0.40 The impact of ad exposure has increased. Time window zoom operation: Allows users to select a specific time period and view the data characteristics within that period.

[0051] Users select the time window for analysis, such as data from February 2025 to April 2025.

[0052] The system automatically filters the data within this time period and recalculates the weight distribution of the features.

[0053] The recalculated weights are as follows: Table 4: Results of weight recalculation.

[0054] feature Weight describe Transaction amount 0.55 The weight of transaction amount has increased slightly. Ad exposure 0.25 The impact of ad exposure has decreased. User behavior frequency 0.20 The impact of user behavior frequency weakens during this period. At this point, the sales forecast trend also changed: Table 5: Results of Sales Revenue Changes

[0055] month Forecasted sales February 2025 52,000 yuan March 2025 58,000 yuan April 2025 63,000 yuan Multidimensional association mapping operation: Users select multiple features, view the association between them, and generate an association feature matrix.

[0056] Users can select multiple features for multidimensional correlation analysis, such as "transaction amount", "user behavior frequency" and "ad impressions".

[0057] The system generates a correlation matrix through multidimensional correlation analysis and displays the relationships between features in the form of a heatmap.

[0058] Suppose the feature correlation matrix generated by the system is as follows: Table 6: Feature Correlation Matrix Table.

[0059] feature Transaction amount User behavior frequency Ad exposure Transaction amount 1 0.75 0.65 User behavior frequency 0.75 1 0.70 Ad exposure 0.65 0.70 1 This indicates a strong positive correlation between "transaction amount" and "user behavior frequency," while "ad impressions" have a lower correlation with the other two features.

[0060] 3. Recalculate the feature weights and output the second analysis results. Based on user interaction, the system triggers an adaptive analysis algorithm to recalculate feature weights and outputs an optimized second analysis result. We will apply a joint loss function to optimize the model's prediction results.

[0061] The formula for calculating the joint loss function is: .

[0062] in, This represents the joint loss value, which is dimensionless and ranges from [value missing]. , Represents the true value of the target variable, dimensionless. This represents a predicted value, dimensionless. Represents the sample size, dimensionless. Represents the regularization coefficient, which is dimensionless. Represents the variational distribution. Describe the prior distribution, This represents the Kullback-Leibler divergence between the variational distribution and the prior distribution, and is dimensionless.

[0063] Part 1: Mean squared error represents the difference between the predicted value and the actual value.

[0064] Part Two: Divergence is used to measure the difference between a posterior and prior distribution. It is an important parameter for adjusting the regularization term.

[0065] Assume the data is as follows: Actual sales figures: .

[0066] Forecasted sales: .

[0067] Divergence: 0.02.

[0068] Regularization coefficient: .

[0069] The actual sales revenue and the projected sales revenue are normalized using the following formula: in, It is the raw data. and These are the minimum and maximum values ​​of the data column, respectively.

[0070] for : , .

[0071] Normalization results: , , .

[0072] for : , .

[0073] Normalization results: , , .

[0074] Mean square error calculation: 4. Results Presentation and Feedback The second set of analysis results will be presented to users through an interactive report, which includes: Sales forecast: Displays the updated trend forecast chart.

[0075] Feature weight update: Displays the optimized feature weights and provides feedback to users to help them make decisions.

[0076] Ultimately, users will be able to optimize the model through interactive operations, view the results of the second analysis, and apply them to actual decision-making.

[0077] Through the steps outlined above, users can adjust and optimize the data analysis process in real time within an interactive environment, thereby obtaining more accurate analytical results. By employing dimensional filtering, time window scaling, and multidimensional correlation mapping operations, users can delve deeper into the data and optimize feature selection. Combined with a joint loss function, the model can balance predictive accuracy with prior knowledge, avoiding overfitting. Ultimately, through interactive analysis result optimization, users can obtain more reliable predictions, providing precise data support for business decisions and improving decision-making efficiency and market competitiveness.

[0078] Example 3 This embodiment relates to an interactive data analysis method and apparatus based on commercial big data. By utilizing this interactive data analysis apparatus based on commercial big data, it aims to improve enterprises' decision-making capabilities and operational efficiency in complex business environments. The specific implementation method is as follows: 1. Data Collection Assuming data analysis is being conducted on an e-commerce platform, the following is an example of multi-source heterogeneous data collected: User behavior data: User ID: 123456.

[0079] User behavior: browsing, clicking, adding to cart, purchasing.

[0080] Transaction data: Transaction ID: ABC123.

[0081] Product ID: 987654.

[0082] Transaction amount: 299 yuan.

[0083] Purchase quantity: 2.

[0084] Date: 2025-08-24.

[0085] Supply chain data: Inventory: 1000.

[0086] Supplier: Supplier A.

[0087] Logistics status: Shipped.

[0088] Market environment data: Domestic GDP growth rate: 5%.

[0089] Competition intensity in the commodity market: High.

[0090] Policy changes: None.

[0091] 2. Data Preprocessing and Feature Extraction Data cleaning and normalization: Clean user behavior data to remove duplicate and invalid records. For example, delete incomplete purchases.

[0092] Data normalization unifies different types of data to the same unit of measurement, such as normalizing transaction amounts to the range [0, 1].

[0093] Feature extraction and fusion: Based on the mutual information criterion, relevant features are extracted from multiple data sources. For example: Extract user characteristics such as purchase frequency and browsing time from user behavior data.

[0094] Extract features such as average purchase amount and number of transactions from transaction data.

[0095] Extract inventory balance and logistics status characteristics from supply chain data.

[0096] Extract features such as GDP growth rate and market competition intensity from market environment data.

[0097] Data from different sources is weighted and merged to generate a structured dataset. For example, by using dimensional weighting to merge user behavior data and transaction data, the following structured dataset can be formed: Table 7: Structured Data Tables.

[0098] User ID Page views Purchase count Average purchase amount in stock Market competition intensity GDP growth 123456 15 3 299 1000 high 5% 3. Adaptive Analysis An adaptive analysis algorithm is used to process the structured dataset described above.

[0099] Joint analysis of deep neural networks and Bayesian inference: Deep neural networks are used to perform nonlinear transformations on input features in order to discover complex feature relationships.

[0100] Combining Bayesian inference, feature weights are calculated by assigning posterior probabilities. Assume the resulting feature weights are as follows: Table 8: Feature Weight Table.

[0101] feature Weight Page views 0.35 Purchase count 0.40 Average purchase amount 0.25 in stock 0.10 Market competition intensity 0.15 GDP growth 0.20 The weights mentioned above reflect the contribution of each feature to the final analysis results.

[0102] 4. Interactive visualization and user interaction Presenting initial analysis results: Visualization tools are used to present preliminary analysis results, showing the weight of each feature and its impact on users' purchasing decisions.

[0103] The system can display feature weight distribution charts and predicted sales trends for each product.

[0104] User interaction operations: Users can select a specific time range or a specific product through the interactive interface, zoom in and out of data dimensions, and view changes in local features.

[0105] For example, if a user selects to view data from "August 24th to August 25th, 2025", the system will recalculate the feature weights and update the analysis results.

[0106] Multidimensional correlation mapping: Users select multiple feature dimensions according to specific needs, the system generates a correlation feature matrix, reflects the correlation between features, and feeds the results back to the adaptive analysis algorithm for correction.

[0107] 5. Results Output and Report Generation Based on the interaction results, a detailed analysis report is generated, including the following: Key business metrics include conversion rate, customer lifetime value, and average order value.

[0108] Risk monitoring indicators include: inventory shortage risk, supply chain risk, and market competition risk.

[0109] Forecasting trend indicators: Based on data, predict future sales trends, market demand, etc.

[0110] The above indicators are transmitted to the enterprise management system through a standardized interface for decision-makers to refer to.

[0111] Through the above steps, the interactive data analysis device for commercial big data can extract key features from multi-source heterogeneous data, perform dynamic analysis using deep learning and Bayesian inference algorithms, and allow users to customize operations through an interactive visual interface, ultimately generating analysis reports with high decision-making value.

[0112] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An interactive data analysis method based on commercial big data, characterized in that, include: S1. Collect multi-source heterogeneous data from business activities, including user behavior data, transaction data, supply chain data, and market environment data; S2. Preprocess and extract features from the multi-source heterogeneous data. Obtain a data feature set based on information theory indicators through feature extraction. Merge the data feature set with the preprocessed multi-source heterogeneous data based on a dimension-weighted combination method to generate a structured dataset. S3. Input the structured dataset into the adaptive analysis algorithm. The adaptive analysis algorithm is based on the joint mechanism of deep neural network and Bayesian inference model to perform dynamic feature selection and hierarchical weight allocation on the data feature set to generate feature weight distribution. The adaptive analysis algorithm outputs the first analysis result according to the feature weight distribution. S4. Display the first analysis result in an interactive visualization interface, the interactive visualization interface including interactive controls for receiving user interaction operations, the user interaction operations triggering the adaptive analysis algorithm to recalculate the feature weight distribution and output the second analysis result; S5. Divide the second analysis results into indicator sets and generate an interactive analysis report, which is transmitted to the enterprise management system through a standardized data interface.

2. The interactive data analysis method based on commercial big data according to claim 1, characterized in that: The preprocessing includes data cleaning, normalization, and format unification. The information theory indicators include the mutual information criterion, which is used to filter features of the multi-source heterogeneous data.

3. The interactive data analysis method based on commercial big data according to claim 1, characterized in that: The hierarchical weight allocation is jointly calculated by the deep neural network and the Bayesian inference model, and the calculation formula is as follows: ， in, Indicates the first The hierarchical weight values ​​of each feature range from [value]. ,satisfy , Indicates input features, This represents the nonlinear mapping output of the deep neural network to the input features. Indicates the first The Bayesian prior probabilities corresponding to each feature are in the range of... , Represents the total number of features. The normalization factor is the weighted average of all features.

4. The interactive data analysis method based on commercial big data according to claim 1, characterized in that: The first analysis result output uses a joint loss function, which includes a mean squared error term based on the deep neural network predictions and a Bayesian inference model. The calculation formula for the joint loss function is as follows: ， in, This represents the joint loss value, ranging from... , This represents the true value of the target variable. Indicates the predicted value. Indicates the number of samples. Represents the regularization coefficient. Represents the variational distribution. Describe the prior distribution, This represents the Kullback-Leibler divergence between the variational distribution and the prior distribution.

5. The interactive data analysis method based on commercial big data according to claim 1, characterized in that: The user interaction operations include: S41. In the interactive visualization interface, the first analysis result is filtered by dimension to generate feature results; S42. Perform a time window scaling operation on the feature results to generate local data features; S43. Based on the local data features, perform a multidimensional association mapping operation to generate an association feature matrix, and feed the association feature matrix back to the adaptive analysis algorithm to output a second analysis result.

6. The interactive data analysis method based on commercial big data according to claim 1, characterized in that: The set of indicators includes a set of key business indicators, a set of risk monitoring indicators, and a set of trend forecasting indicators.

7. An interactive data analysis device based on commercial big data, characterized in that, include: A multi-source data acquisition module is used to acquire heterogeneous data sources from business activities, including user behavior, transaction flow, supply chain logistics and external market environment; The data preprocessing and feature engineering module is used to clean, transform, and extract features from the heterogeneous data sources to generate structured datasets and data feature sets. An adaptive analysis engine is used to dynamically select features and hierarchically assign weights to the data feature set to generate a feature weight distribution. The adaptive analysis engine generates initial analysis results based on the feature weight distribution. An interactive visualization processing module is used to display the initial analysis results. The interactive visualization processing module provides interactive controls to receive user instructions and feed them back to the adaptive analysis engine. The adaptive analysis engine recalculates the feature weight distribution and generates the final analysis results. The analysis report output module is used to generate an interactive report based on the final analysis results. The interactive report is transmitted to the enterprise management system through a standardized application programming interface.

8. The interactive data analysis device based on commercial big data according to claim 7, characterized in that: The adaptive analysis engine includes a deep neural network unit, a Bayesian inference unit, and a weight fusion unit. The deep neural network unit is used to perform nonlinear transformations on the data feature set.

9. The interactive data analysis device based on commercial big data according to claim 8, characterized in that: The Bayesian inference unit is used to calculate the posterior probability, and the weight fusion unit is used to jointly calculate the feature weight distribution based on the output of the nonlinear transformation and the posterior probability.

Images