Competitive intelligence analysis system and method based on differential privacy

Abstract

The application relates to a differential privacy-based competitive intelligence analysis system and method, comprising a differential privacy processing and budget management module, a differential privacy feature engineering module, a privacy protection analysis and prediction module and a differential privacy synthetic data generation module. The design of the application constructs an end-to-end advanced analysis framework with mathematically provable privacy protection capability, so that accurate, timely and reliable insights into market dynamics and competitor strategies can be realized on the premise of strictly protecting sensitive business secrets of enterprises.

Description

Technical Field

[0001] This invention relates to the fields of data analysis and privacy protection technology, and in particular to the field of competitive intelligence analysis systems and methods based on differential privacy. Background Technology

[0002] In the fields of apparel AI and e-commerce, companies need to analyze key indicators such as competitors' pricing strategies, sales trends, user reviews, and inventory turnover to formulate their own business strategies. However, existing competitive intelligence analysis systems have the following major flaws: Serious risk of data privacy leakage: Traditional competitive analysis systems lack effective privacy protection mechanisms when processing sensitive business data. When sharing or analyzing competitor data, companies may inadvertently leak core trade secrets, such as precise sales data, pricing strategies, and customer preferences, causing serious business losses. Inadequate privacy protection methods: Existing systems mostly use traditional data anonymization methods, such as data masking and randomization. These methods cannot provide strict mathematical privacy guarantees and are easily cracked when facing background knowledge attacks or inference attacks. Lack of privacy budget management: Existing technologies lack precise quantification and management mechanisms for privacy consumption, failing to balance the degree of privacy protection with data availability, easily leading to excessive consumption or insufficient protection of the privacy budget. Lack of privacy considerations in feature engineering: Traditional feature selection and engineering methods do not consider privacy constraints and may fail in differential privacy environments, leading to inaccurate analysis results or privacy leaks. Insufficient processing capabilities for low-sample scenarios: In certain market segments or emerging product categories, sample data is scarce, and traditional methods struggle to obtain reliable analytical results while protecting privacy.

[0003] Therefore, how to achieve high-quality privacy-aware feature engineering under the premise of strictly protecting commercial privacy, and effectively handle the balance between privacy protection and data availability in low-sample scenarios, thereby ensuring the accuracy and reliability of analysis results, has become a key challenge and an urgent technical problem for those skilled in the art. Summary of the Invention

[0004] The purpose of this invention is to provide a competitive intelligence analysis system and method based on differential privacy, in order to solve the problems in the above-mentioned background technology, such as serious data privacy leakage risks, overly rudimentary privacy protection methods, lack of privacy budget management, lack of privacy considerations in feature engineering, and insufficient processing capabilities for low-sample scenarios.

[0005] This invention provides the following technical solution: The competitive intelligence analysis method based on differential privacy includes the following steps: S1. Data Input and Preprocessing: Collect raw data, perform data cleaning and standardization on the raw data to standardize the raw data, and determine the sensitivity level of the raw data; S2. Differential Privacy Processing and Budget Management: The differential privacy processing and budget management module, through a privacy budget strategy, dynamically allocates differential privacy budgets ε_i for each dimension based on the preset importance, sensitivity, and analytical accuracy requirements of business indicators using a multi-objective optimization algorithm. Under the premise of satisfying the total budget constraint Σε_i ≤ ε_total, it prioritizes the availability of high-value indicators. The differential privacy noise addition mechanism adaptively selects either a Laplace or Gaussian mechanism based on the privacy budget, injecting noise according to query sensitivity. The Rényi differential privacy accounting system, based on the Rényi divergence composite theorem, accurately tracks the accumulated privacy loss from multiple operations, providing a tighter upper bound for privacy protection. All operations are recorded in an immutable privacy ledger, ensuring full auditability and traceability. S3. Differential Privacy Feature Engineering: Under privacy constraints, the differential privacy feature engineering module selects highly relevant and low-redundancy features through the DP-mRMR feature selector, and improves noise robustness with the help of a stable selection mechanism. Finally, based on the selected features, it constructs key indicators with business insights. All operations are performed under differential privacy protection to ensure that individual data cannot be inferred. S4. Privacy Protection Analysis and Prediction: The privacy protection analysis and prediction module performs market trend analysis and anomaly detection using a differential privacy-preserving time series model, and integrates the prediction results of multiple differential privacy-preserving basic models based on a stacked ensemble learning framework to improve robustness; the privacy protection analysis and prediction module quantifies prediction uncertainty using Bayesian differential privacy or Bootstrap methods, and finally, relying on a strategy knowledge base and rule engine, automatically matches the analysis results and generates executable quantitative strategy suggestions; S5. Synthetic Data Augmentation: The differential privacy synthetic data generation module identifies scenarios with scarce data samples and uses the DP synthetic data generator to synthesize data with similar statistical characteristics to the original data under the premise of satisfying differential privacy constraints. Finally, the synthetic data is fused with the original data to enhance the training set, thereby improving the model's generalization ability and prediction robustness in low-sample scenarios. S6. Output the results; Step S4. Privacy-preserving analysis and prediction, which involves market trend analysis and anomaly detection using a differential privacy-preserving time series model, includes the following steps: For the time-series data generated in step S3. Differential privacy feature engineering, which involves constructing key indicators with business insights based on selected features, differential privacy trend analysis, periodic analysis, and anomaly pattern identification are performed. A differential privacy time-series model is used to model the historical data of the key indicators to predict future trends. In terms of anomaly detection, a differential privacy statistical control chart method or a density estimation method based on DP is used to identify anomaly patterns in the data, and an alert is issued through differential privacy threshold comparison or privacy-preserving outlier detection algorithm. The working steps of the stacked learning framework in step S4, privacy-preserving analysis and prediction, include: Under differential privacy constraints, the prediction results of multiple base learners are combined and a final prediction is made through a meta-learner. The stacked learning framework trains multiple heterogeneous base models. Each base model independently applies the differential privacy mechanism during training. Subsequently, the prediction results of the base models are used as input features of the meta-learner. The meta-learner also applies the differential privacy mechanism for protection during the training phase.

[0006] The steps of the privacy budget strategy in step S2, differential privacy processing and budget management, include: Based on the preset sensitivity levels of each dimension, the expected analytical accuracy requirements, and the total privacy budget ε_total within the system's total budget constraint Σε_i ≤ ε_total, the multi-objective optimization algorithm is used to dynamically calculate and allocate the differential privacy budget value ε_i corresponding to each indicator dimension. The privacy budget strategy receives a pre-calibrated weight matrix regarding the business importance and sensitivity of different indicators as input and outputs each privacy budget value ε_i, ensuring that highly sensitive indicators receive stricter privacy protection, while allocating sufficient budget to indicators that contribute significantly to the analytical task to maintain data availability.

[0007] The working steps of the differential privacy noise addition mechanism in step S2, differential privacy processing and budget management, include: The system receives the raw data, and based on the query results of the raw data and the privacy budget value corresponding to the output dimension of the privacy budget strategy, the differential privacy noise addition mechanism adaptively selects and executes a Laplace or Gaussian mechanism according to the data type and query sensitivity, adding random noise that follows a Laplace or Gaussian distribution to the query results. The function of the Laplace distribution is Lap(0, Δ / ε_i), and the function of the Gaussian distribution is N(0, σ²), where Δ represents the global sensitivity of the query function, ε_i is the privacy budget value, and σ² is the variance, reflecting the dispersion of the data distribution.

[0008] The operational steps of the Rényi differential privacy accounting system in step S2, differential privacy processing and budget management, include: A novel privacy composition theorem based on Rényi divergence is employed to accurately calculate and track the cumulative privacy loss generated by multiple differential privacy operations, and to maintain a cumulative privacy consumption amount. The impact of each noise addition operation on the cumulative privacy consumption amount is monitored in real time. The Rényi differential privacy accounting system receives the parameters of each differential privacy operation and updates the total privacy consumption value according to the Rényi composition rule to ensure that the cumulative privacy loss does not exceed the preset total privacy budget.

[0009] The privacy ledger in step S2, Differential Privacy Processing and Budget Management, is an immutable distributed log system used to record detailed metadata for each differential privacy operation. The metadata includes, but is not limited to, operation type, operation execution timestamp, dataset identifier involved, differential privacy budget value ε_consumed consumed, noise mechanism type used, specific parameters queried, and the cumulative privacy consumption before and after the operation. The privacy ledger ensures the integrity and auditability of the records through cryptographic hash chain technology.

[0010] The steps of the DP-mRMR feature selector in step S3, differential privacy feature engineering, include: The DP-mRMR feature selector first calculates the mutual information between each feature and the target variable using a mutual information calculation function. Differential privacy noise is introduced during the mutual information calculation process to protect the impact of individual data records on the mutual information calculation result. The mutual information calculation function is treated as a query function, and its sensitivity is quantified. The differential privacy noise is then injected into the calculation result through the Laplace or Gaussian mechanism. Subsequently, the DP-mRMR feature selector calculates the mutual information between selected features and candidate features as a redundancy measure, and the differential privacy noise is also introduced during this calculation process. The DP-mRMR feature selector iteratively selects features, optimizing the differential privacy protection objective function of "maximizing relevance" and "minimizing redundancy" with each selection, until a preset number of features or a performance threshold is reached.

[0011] The working steps of the stable selection mechanism in step S3, differential privacy feature engineering, include: By performing multiple random resamplings on the original dataset and running the DP-mRMR feature selector independently on the subset dataset generated by each resampling, the frequency of each feature being selected in all resampling runs is counted, and those features that are consistently selected in resampling runs above the preset threshold are identified as the final subset of key features. Step S3. Constructing key metrics with business insights based on selected features in differential privacy feature engineering specifically includes the following steps: Based on the key features identified by the DP-mRMR feature selector and stable selection mechanism, a series of high-level competitive intelligence indicators with business insights are constructed. Differential privacy-preserving statistical methods are used to aggregate, derive, and calculate the original data or the data after anonymization. The S5. synthetic data enhancement includes the following steps: S51. Low-sample scenario identification: Continuously monitor and analyze the structure and distribution characteristics of the input dataset, automatically identify feature dimensions, market segments, or product categories with scarce samples or extremely unbalanced data distribution in the dataset; by calculating the statistics of each feature dimension and comparing them with the preset minimum sample threshold, determine which parts need data augmentation, and assess the risk of a serious decrease in data availability that may result from directly applying differential privacy analysis with the existing sample size, thereby determining the specific scenarios that need synthetic data supplementation and the required amount of synthetic data; S52.DP Synthetic Data Generation: Under the premise of satisfying the differential privacy constraints, artificial synthetic data with similar statistical characteristics to the original data is generated; a synthesis method based on generative adversarial networks or variational autoencoders is adopted, wherein the differential privacy mechanism is integrated into the training process of the generator and discriminator, or a query-based synthesis method is adopted, in which the statistical characteristics of the original data are learned through queries protected by the differential privacy mechanism, and the result of each query is injected with the differential privacy noise into the calculation result through the Laplace mechanism or the Gaussian mechanism, and then these protected statistical characteristics are used to construct the synthetic data; S53. Robust Training: The data generated in step S52 is effectively fused with the original dataset to construct an enhanced training dataset. A semi-supervised learning or hybrid training strategy is adopted, combining a portion of the original data with a large amount of the synthetic data to train the privacy-preserving analysis and prediction module for machine learning. The effect of the synthetic data on improving model performance is verified through cross-validation and performance metrics evaluation.

[0012] The competitive intelligence analysis system based on differential privacy includes a differential privacy processing and budget management module, a differential privacy feature engineering module, a privacy protection analysis and prediction module, and a differential privacy synthetic data generation module. The differential privacy processing and budget management module is used to receive raw sensitive business data, perform differential privacy noise addition processing on the raw sensitive business data according to preset privacy protection requirements, and at the same time perform precise privacy budget allocation, tracking and auditing of all privacy-related operations. The differential privacy feature engineering module is used to select and construct key features with optimal predictive ability and that meet privacy protection requirements from the de-identified data under differential privacy constraints. The privacy protection analysis and prediction module is used to perform market trend analysis, competitive strategy prediction and abnormal behavior detection using the privacy protection dataset processed by the differential privacy feature engineering module, and provides a quantitative assessment of prediction uncertainty. The differential privacy synthetic data generation module is used to identify and process specific scenarios with scarce data samples. It enhances the dataset by generating differential privacy-preserving synthetic data, thereby improving the analysis and prediction performance in low-sample scenarios.

[0013] Compared with the prior art, the beneficial effects of the present invention are: 1. The differential privacy-based competitive intelligence analysis system of this invention includes a differential privacy processing and budget management module, a differential privacy feature engineering module, a privacy protection analysis and prediction module, and a differential privacy synthetic data generation module. Through the synergistic cooperation of these modules, advanced differential privacy theory, intelligent feature engineering technology, multi-level machine learning models, and privacy-aware data enhancement mechanisms are deeply integrated to construct an end-to-end advanced analysis framework with mathematically provable privacy protection capabilities. This enables accurate, timely, and reliable insights into market dynamics and competitor strategies while strictly protecting sensitive business secrets.

[0014] 2. The competitive intelligence analysis method based on differential privacy of the present invention includes the following steps: S1. Data input and preprocessing, S2. Differential privacy processing and budget management, S3. Differential privacy feature engineering, S4. Privacy-preserving analysis and prediction, S5. Synthetic data augmentation, and S6. Result output. Through these steps, rigorous mathematically provable privacy protection can be achieved, effectively resisting re-identification attacks by attackers with background knowledge; through refined privacy budget allocation and the Rényi-DP accounting system, precise quantitative management of privacy consumption can be achieved, optimizing the efficiency of privacy budget utilization; under privacy constraints, high-quality feature selection is achieved using DP-mRMR and stable selection mechanisms, ensuring the accuracy of analysis results; through DP synthetic data generation technology, analysis and prediction performance in low-sample scenarios is significantly improved; and through ensemble learning and uncertainty estimation, the reliability of analysis and prediction results is enhanced, providing a solid and credible scientific basis for enterprises to make business decisions.

[0015] 3. The competitive intelligence analysis method based on differential privacy of the present invention includes the following steps in step S2. The privacy budget strategy in differential privacy processing and budget management: Based on the preset sensitivity levels of each dimension, the expected analytical accuracy requirements, and the total privacy budget ε_total in the system-set total budget constraint Σε_i ≤ ε_total, a multi-objective optimization algorithm is used to dynamically calculate and allocate the differential privacy budget value ε_i corresponding to each indicator dimension. The privacy budget strategy receives a pre-calibrated weight matrix regarding the business importance and sensitivity of different indicators as input and outputs each privacy budget value ε_i, ensuring that highly sensitive indicators receive stricter privacy protection, while allocating sufficient budget to indicators that contribute significantly to the analytical task to maintain data availability. Through the above steps, the differential privacy budget is dynamically allocated through a multi-objective optimization algorithm, achieving a fine balance between privacy protection strength and data availability: both assigning smaller privacy budget values ​​to highly sensitive indicators to strengthen privacy protection and allocating sufficient budget to key analytical indicators to maintain data usability, thereby maximizing the overall data value under the total privacy budget constraint while meeting differentiated privacy protection needs.

[0016] 4. The competitive intelligence analysis method based on differential privacy of the present invention, step S2. The working steps of the differential privacy noise addition mechanism in differential privacy processing and budget management include: receiving the original data, according to the query results of the original data and the privacy budget value of the corresponding indicator dimension output by the privacy budget strategy; the differential privacy noise addition mechanism adaptively selects and executes the Laplace mechanism or the Gaussian mechanism according to the data type and query sensitivity, and adds random noise that follows the Laplace distribution or the Gaussian distribution to the query results. The function of the Laplace distribution is Lap(0, Δ / ε_i), and the function of the Gaussian distribution is N(0, σ²), where Δ represents the global sensitivity of the query function, ε_i is the privacy budget value, and σ² is the variance, reflecting the dispersion of the data distribution. Through the above steps, by adaptively selecting Laplace or Gaussian noise addition mechanisms, sensitive data is effectively protected from inference attacks while maintaining the usability of aggregated query results, under the premise of meeting differentiated privacy budget constraints. Its built-in sensitivity assessment unit can accurately quantify the dependence of query functions on individual records, thereby optimizing the noise level and significantly improving data utility under the same level of privacy protection.

[0017] 5. The competitive intelligence analysis method based on differential privacy of the present invention, step S2. The working steps of the Rényi differential privacy accounting system in differential privacy processing and budget management include: using a novel privacy composite theorem based on Rényi divergence to accurately calculate and track the cumulative privacy loss generated by multiple differential privacy operations, and maintaining a cumulative privacy consumption amount; monitoring the impact of each noise addition operation on the cumulative privacy consumption amount in real time; the Rényi differential privacy accounting system receives the parameters of each differential privacy operation and updates the total privacy consumption value according to the Rényi composite rule to ensure that the cumulative privacy loss does not exceed the preset total privacy budget. Through the above steps, by introducing the Rényi differential privacy accounting system, the cumulative privacy loss of multiple privacy operations can be tracked and calculated more accurately. Compared with the traditional ε-DP method, it significantly reduces the looseness of the upper bound of privacy loss, thereby supporting more iterative operations or more intensive queries under the same total privacy budget, while providing stricter privacy protection, achieving a better balance between privacy protection and model utility.

[0018] 6. In the differential privacy-based competitive intelligence analysis method of the present invention, step S2, the privacy ledger in differential privacy processing and budget management is an immutable distributed log system used to record detailed metadata for each differential privacy operation. This metadata includes, but is not limited to, operation type, operation execution timestamp, involved dataset identifier, consumed differential privacy budget value ε_consumed, noise mechanism type used, specific query parameters, and the cumulative privacy consumption before and after the operation. The privacy ledger ensures the integrity and auditability of the records through encrypted hash chain technology. By constructing an immutable privacy operation log, the transparency, traceability, and controllability of differential privacy budget usage are significantly improved. This ensures the integrity and auditability of the data operation process and optimizes privacy resource allocation through historical record analysis, thereby improving the management efficiency of the privacy budget while strictly meeting compliance requirements.

[0019] 7. The competitive intelligence analysis method based on differential privacy of the present invention, step S3. The working steps of the DP-mRMR feature selector in differential privacy feature engineering include: the DP-mRMR feature selector first calculates the mutual information between each feature and the target variable through the mutual information calculation function, and introduces differential privacy noise in the mutual information calculation process to protect the influence of a single data record on the mutual information calculation result. The mutual information calculation function is regarded as a query function, and its sensitivity is quantified. Differential privacy noise is injected into the calculation result through the Laplace mechanism or Gaussian mechanism. Subsequently, the DP-mRMR feature selector calculates the mutual information between the selected feature and the candidate feature as a redundancy measure, and also introduces differential privacy noise in the calculation process. The DP-mRMR feature selector iteratively selects features, and each selection optimizes the differential privacy protection objective function of "maximizing relevance" and "minimizing redundancy" until the preset number of features or performance threshold is reached. Through the above steps, the needs of feature utility and privacy protection are successfully balanced under the premise of strictly meeting differential privacy constraints. By injecting sensitivity-calibrated noise into the mutual information calculation process, it is ensured that the feature selection process does not leak sensitive information of individual data records. Thus, the feature subset selected in the privacy-protected environment maintains a high correlation with the target variable and effectively reduces the redundancy between features, providing a safe and highly discriminative feature foundation for subsequent machine learning modeling.

[0020] 8. The competitive intelligence analysis method based on differential privacy of the present invention, step S3. The working steps of the stable selection mechanism in differential privacy feature engineering include: performing multiple random resamplings on the original dataset, and independently running the DP-mRMR feature selector on the subset dataset generated by each resampling, statistically analyzing the frequency of each feature being selected in all resampling runs, and identifying those features that are consistently selected in resampling runs above a preset threshold as the final key feature subset. By running DP-mRMR feature selection in parallel on multiple resampling data through the stable selection mechanism, and performing final feature screening based on the statistical consistency of feature selection frequencies, the robustness and result stability of the feature selection process under differential privacy noise environment are significantly improved, effectively reducing the risk of feature misselection or omission due to noise interference, thereby ensuring the reliability and interpretability of the final feature subset.

[0021] 9. The differential privacy-based competitive intelligence analysis method of the present invention, step S3. Constructing key indicators with business insights based on selected features in differential privacy feature engineering specifically includes the following steps: Based on the key features identified by the DP-mRMR feature selector and stable selection mechanism, a series of high-level competitive intelligence indicators with business insights are constructed. Differential privacy-preserving statistical methods are used to aggregate, derive, and calculate the original data or the data after anonymization. By constructing key competitive intelligence indicators in multiple dimensions through differential privacy protection mechanisms, high-quality and trustworthy business insights can be provided under the premise of strictly protecting individual privacy and commercially sensitive data. This satisfies privacy compliance requirements and supports enterprises in securely conducting market competitiveness analysis and strategic decision-making.

[0022] 10. The competitive intelligence analysis method based on differential privacy of the present invention includes the following steps in step S4. Privacy-preserving analysis and prediction: market trend analysis and anomaly detection using a differential privacy-preserving time series model. These steps involve performing differential privacy-preserving trend analysis, periodic analysis, and anomaly pattern identification on the time series data generated during step S3. Differential privacy feature engineering, which involves constructing key indicators with business insights based on selected features. A differential privacy-preserving time series model is used to model historical data of key indicators to predict future trends. For anomaly detection, a differential privacy-preserving statistical control chart method or a density estimation method based on differential privacy (DP) is used to identify anomaly patterns in the data, and alerts are issued through differential privacy threshold comparison or privacy-preserving outlier detection algorithms. By introducing a differential privacy protection mechanism into time series analysis and anomaly detection, the method effectively supports trend prediction, periodic analysis, and anomaly pattern identification of key indicators while strictly protecting individual data privacy. This satisfies data compliance requirements and ensures the usability of analysis results, providing reliable and privacy-secure support for business decisions.

[0023] 11. The competitive intelligence analysis method based on differential privacy of the present invention, step S4. The working steps of the stacked learning framework in the privacy-preserving analysis prediction include: under differential privacy constraints, combining the prediction results of multiple base learners, and performing final prediction through a meta-learner; the stacked learning framework trains multiple heterogeneous base models, each base model independently applying the differential privacy mechanism during training; subsequently, the prediction results of the base models are used as input features of the meta-learner, which also applies the differential privacy mechanism for protection during the training phase. Through the differential privacy-preserving stacked integration architecture, the predictive advantages of multiple heterogeneous base models are effectively integrated under the premise of strictly ensuring data privacy, significantly improving the model's generalization ability and robustness in noisy environments, while reducing the risk of overfitting, ultimately achieving more accurate and privacy-secure prediction performance.

[0024] 12. The competitive intelligence analysis method based on differential privacy of the present invention, S5. Synthetic data augmentation includes the following steps: S51. Low-sample scene identification, S52. DP synthetic data generation, and S53. Robust training. By automatically identifying low-sample scenes and generating differential privacy-preserving synthetic data, the problem of data sparsity and imbalanced distribution is effectively solved. While strictly protecting individual privacy, it significantly improves the generalization ability and prediction stability of machine learning models on scarce data dimensions, achieving an efficient balance between privacy protection and data usability. Attached Figure Description

[0025] 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 only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0026] Figure 1 This is a flowchart illustrating the competitive intelligence analysis system and method based on differential privacy of the present invention. Detailed Implementation

[0027] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain this application, and should not be construed as limiting this application.

[0028] In the description of this application, it should be understood that the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, features defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0029] In the description of this application, it should be noted that, unless otherwise expressly specified and limited, the terms "installation" and "connection" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection, an electrical connection, or a connection that allows communication between the components; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication between two components or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in this application according to the specific circumstances.

[0030] In the description of this application, it should be understood that the terms "upper", "lower", "side", "front", "rear", etc., indicate the orientation or positional relationship based on the installation orientation or positional relationship, and are only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of this application.

[0031] In the description of this application, it should be noted that the term "and / or" is merely a description of the relationship between related objects, indicating that there can be three relationships. For example, A and / or B can represent three situations: A exists alone, A and B exist simultaneously, and B exists alone.

[0032] It should also be noted that in the embodiments of this application, the same reference numerals are used to represent the same component or part. For the same part in the embodiments of this application, the reference numerals may only be used to mark one part or component as an example. It should be understood that the reference numerals are also applicable to other identical parts or components.

[0033] Furthermore, in this application, the technical features described in an open-ended manner include both closed technical solutions composed of the listed features and open technical solutions that include the listed features.

[0034] To further understand the invention's content, features, and effects, the following embodiments are provided, along with detailed descriptions in conjunction with the accompanying drawings: like Figure 1 As shown, the competitive intelligence analysis system and method based on differential privacy includes a differential privacy processing and budget management module, a differential privacy feature engineering module, a privacy protection analysis and prediction module, and a differential privacy synthetic data generation module. The differential privacy processing and budget management module is used to receive raw sensitive business data and perform differential privacy noise addition processing on the raw sensitive business data according to the preset privacy protection requirements. At the same time, it performs precise privacy budget allocation, tracking and auditing for all privacy-related operations. The differential privacy feature engineering module is used to select and construct key features with optimal predictive ability and privacy protection requirements from anonymized data under differential privacy constraints. The privacy protection analysis and prediction module is used to perform market trend analysis, competitive strategy prediction and abnormal behavior detection using privacy protection datasets processed by the differential privacy feature engineering module, and provides quantitative prediction uncertainty assessment. The differential privacy synthetic data generation module is used to identify and process specific scenarios with scarce data samples. It enhances the dataset by generating differential privacy-preserving synthetic data, thereby improving the analysis and prediction performance in low-sample scenarios.

[0035] In this embodiment, through the synergistic cooperation of the aforementioned modules, advanced differential privacy theory, intelligent feature engineering technology, multi-level machine learning models, and privacy-aware data enhancement mechanisms are deeply integrated to construct an end-to-end advanced analytics framework with mathematically provable privacy protection capabilities. This allows for accurate, timely, and reliable insights into market dynamics and competitor strategies while strictly protecting sensitive business secrets. Of course, this is not a limitation; any framework capable of providing accurate, timely, and reliable insights into market dynamics and competitor strategies can be used.

[0036] Specifically, the competitive intelligence analysis method based on differential privacy includes the following steps: S1. Data Input and Preprocessing: Collect raw data, perform data cleaning and standardization to standardize the raw data, and determine the sensitivity level of the raw data. In this embodiment, raw data is first collected from multiple sources, including [sources listed here]. Then, data cleaning is performed to handle missing values, outliers, and duplicate entries to ensure data quality. Next, standardization is performed to unify the data format and ensure consistent scaling, facilitating subsequent analysis. Finally, the sensitivity level of the data content is assessed and determined to provide a basis for differential privacy protection. Of course, this is not limited to this; any other method capable of standardizing the input raw data and determining its sensitivity level can be used. S2. Differential Privacy Processing and Budget Management: The differential privacy processing and budget management module uses a privacy budget strategy to dynamically allocate differential privacy budgets ε_i for each dimension based on the preset importance, sensitivity, and analytical accuracy requirements of business metrics, through a multi-objective optimization algorithm, while satisfying the total budget constraint Σε_i ≤ Under the premise of ε_total, the availability of high-value indicators is prioritized; the differential privacy noise addition mechanism adaptively selects either Laplace or Gaussian mechanism based on the privacy budget, injecting noise according to query sensitivity; the Rényi differential privacy accounting system accurately tracks the cumulative privacy loss of multiple operations based on the Rényi divergence composite theorem, providing a tighter upper bound for privacy protection; all operations are recorded in an immutable privacy ledger, ensuring full auditability and traceability; in this embodiment, Rényi divergence is a quantitative tool used to measure the difference between two probability distributions, and the Rényi differential privacy accounting system is an audit and management system based on Rényi divergence for more accurate calculation and tracking of cumulative privacy consumption; by integrating four core components—the privacy budget strategy engine, the differential privacy noise adder, the Rényi differential privacy accounting system, and the privacy ledger—refined privacy protection and efficient budget management of sensitive data in multidimensional business indicator analysis are achieved. Its beneficial effects include: dynamically allocating the privacy budget through a multi-objective optimization algorithm, significantly improving the availability of high-value indicators and the protection strength of sensitive data while ensuring the total privacy constraint Σε_i ≤ ε_total; ensuring the accuracy of noise injection and the usability of query results by utilizing an adaptive noise addition mechanism and sensitivity assessment unit; optimizing the calculation of privacy loss in multiple operations by introducing the Rényi differential privacy accounting system, overcoming the shortcomings of traditional methods in overly conservative budget estimation in iterative scenarios, and improving the efficiency of privacy resource utilization; and finally, enhancing the transparency, compliance, and accountability of the entire system through the immutable audit traceability provided by the privacy ledger, providing end-to-end controllable and verifiable technical support for data privacy governance. Of course, it is not limited to these; it can be any other method that can perform differential privacy noise addition processing on data according to preset privacy protection requirements. S3. Differential Privacy Feature Engineering: Under privacy constraints, the differential privacy feature engineering module selects highly relevant and low-redundancy features using the DP-mRMR feature selector and enhances noise robustness through a stable selection mechanism. Finally, based on the selected features, it constructs key indicators with business insights. All operations are performed under differential privacy protection, ensuring that individual data cannot be inferred. In this embodiment, the DP-mRMR feature selector is a maximum relevance and minimum redundancy feature selection algorithm that integrates a differential privacy protection mechanism. By integrating the differential privacy-protected DP-mRMR feature selector with a stable selection mechanism, it can efficiently identify a subset of key features that are highly relevant to the target variable and have low redundancy while strictly protecting individual data privacy. The injection of differential privacy noise ensures that the feature selection process does not leak sensitive information from any single data record, while the stable selection mechanism significantly improves the robustness and reliability of the feature selection results through multiple resampling and statistical consistency checks. The multi-dimensional competitive intelligence indicators built on this basis all adopt differential privacy aggregation and calculation methods, so that the final business insights have high-level analytical value and can effectively prevent background knowledge attacks and data reconstruction risks, thereby maximizing data utility under the premise of compliance; of course, it is not limited to this, it can be any other key features that can be selected and constructed from the data processed in the previous steps, which have the best predictive ability and meet the privacy protection requirements. S4. Privacy Protection Analysis and Prediction: The privacy protection analysis and prediction module performs market trend analysis and anomaly detection using a differential privacy-preserving time series model. It also integrates prediction results from multiple differential privacy-preserving models based on a stacked ensemble learning framework to enhance robustness. The module quantifies prediction uncertainty using Bayesian differential privacy or Bootstrap methods. Finally, relying on a policy knowledge base and rule engine, it automatically matches the analysis results and generates executable quantitative policy recommendations. In this embodiment, Bootstrap refers to self-help. By integrating differential privacy-preserving time series analysis, stacked ensemble learning, uncertainty estimation, and automatic policy generation modules, a four-in-one intelligent decision support system is constructed. Under the premise of strictly protecting data privacy, it achieves high-precision prediction and anomaly detection of key indicators, improves decision reliability by quantifying uncertainty, and finally outputs executable business action recommendations through a knowledge graph-driven policy generation engine. This system not only effectively balances the conflict between data privacy and model utility, but also significantly improves the response efficiency and scientific decision-making of enterprises in dynamic market environments; of course, it is not limited to this, it can be any other system that can perform market trend analysis, competitive strategy prediction and abnormal behavior detection, and provide quantitative assessment of prediction uncertainty. S5. Synthetic Data Augmentation: The differential privacy synthetic data generation module identifies scenarios with scarce data samples and utilizes the DP synthetic data generator to synthesize data with similar statistical characteristics to the original data, while satisfying differential privacy constraints. Finally, the synthetic data is fused with the original data to augment the training set, thereby improving the model's generalization ability and prediction robustness in low-sample scenarios. In this embodiment, the DP synthetic data generator refers to the differential privacy synthetic data generator. It automatically detects feature dimensions where data is scarce or imbalanced through a low-sample scenario identifier and assesses the risk of data invalidation that may result from directly applying differential privacy. This drives the DP synthetic data generator to generate synthetic data with realistic statistical characteristics and untraceable individuals under strict privacy constraints. The synthetic data is then fused with the original data through a robustness training unit to construct an augmented training set, effectively improving the model's generalization ability and prediction stability in low-sample scenarios. This achieves the dual goals of significantly enhancing data usability and analytical effectiveness while protecting privacy. Of course, this is not limited to this; any other method that can enhance the dataset by generating differential privacy-protected synthetic data, thereby improving analytical and predictive performance in low-sample scenarios, is acceptable. S6. Output Results. In this embodiment, the output results include a competitive analysis report, trend prediction results, strategic recommendations, uncertainty assessment, and privacy usage audit. However, it is not limited to these; other competitive intelligence analyses may also be output.

[0037] Specifically, step S2. The steps involved in the privacy budget strategy within differential privacy processing and budget management include: Based on the preset sensitivity levels of each dimension, the expected analytical accuracy requirements, and the total privacy budget ε_total in the system's total privacy constraint Σε_i ≤ ε_total, a multi-objective optimization algorithm is used to dynamically calculate and allocate the differential privacy budget value ε_i corresponding to each indicator dimension. The privacy budget strategy takes a pre-calibrated weight matrix regarding the business importance and sensitivity of different indicators as input and outputs each privacy budget value ε_i to ensure that highly sensitive indicators receive stricter privacy protection, while allocating sufficient budget to indicators that contribute significantly to the analytical task to maintain data availability.

[0038] In this embodiment, a multi-objective optimization algorithm is used to dynamically allocate differential privacy budgets, achieving a fine balance between privacy protection strength and data usability. This involves assigning smaller privacy budget values ​​to highly sensitive indicators to strengthen privacy protection, while allocating sufficient budgets to key analytical indicators to maintain data usability. This maximizes the overall data value under the constraint of the total privacy budget, while simultaneously meeting differentiated privacy protection needs. Of course, this is not a limitation; any other approach that maximizes the overall data value under the constraint of the total privacy budget while meeting differentiated privacy protection needs can be used.

[0039] Specifically, the working steps of the differential privacy noise addition mechanism in step S2, differential privacy processing and budget management, include: The system receives raw data, calculates the query results based on the raw data, and outputs the corresponding privacy budget value for the privacy budget strategy. A differential privacy noise addition mechanism adaptively selects and executes either a Laplace or Gaussian mechanism based on the data type and query sensitivity, adding random noise to the query results that follows a Laplace or Gaussian distribution. The function for the Laplace distribution is Lap(0, Δ / ε_i), and the function for the Gaussian distribution is N(0, σ²), where Δ represents the global sensitivity of the query function, ε_i is the privacy budget value, and σ² is the variance, reflecting the dispersion of the data distribution.

[0040] In this embodiment, by adaptively selecting either Laplacian or Gaussian noise addition mechanisms, sensitive data is effectively protected from inference attacks while maintaining the usability of aggregated query results, all while satisfying differentiated privacy budget constraints. Its built-in sensitivity assessment unit can accurately quantify the dependency of the query function on individual records, thereby optimizing the noise level and significantly improving data utility under the same level of privacy protection. Of course, this is not a limitation; other methods that can optimize the noise level and significantly improve data utility under the same level of privacy protection can also be used.

[0041] Specifically, step S2. The operational steps of the Rényi Differential Privacy Accounting System in Differential Privacy Processing and Budget Management include: A novel privacy composition theorem based on Rényi divergence is adopted to accurately calculate and track the cumulative privacy loss caused by multiple differential privacy operations, and maintain a cumulative privacy consumption amount. The impact of each noise addition operation on the cumulative privacy consumption amount is monitored in real time. The Rényi differential privacy accounting system receives the parameters of each differential privacy operation and updates the total privacy consumption value according to the Rényi composition rule to ensure that the cumulative privacy loss does not exceed the preset total privacy budget.

[0042] In this embodiment, by introducing the Rényi differential privacy accounting system, the cumulative privacy loss from multiple privacy operations can be tracked and calculated more accurately. Compared to the traditional ε-DP method, it significantly reduces the looseness of the upper bound on privacy loss, thereby supporting more iterative operations or more intensive queries under the same total privacy budget, while providing stricter privacy guarantees, achieving a better balance between privacy protection and model utility. Of course, it is not limited to this; other methods that can achieve a better balance between privacy protection and model utility can be used.

[0043] Specifically, in step S2, the privacy ledger in differential privacy processing and budget management is an immutable distributed log system used to record the metadata of each differential privacy operation in detail. The metadata includes, but is not limited to, operation type, operation execution timestamp, identifier of the dataset involved, differential privacy budget value ε_consumed consumed, type of noise mechanism used, specific parameters of the query, and the cumulative privacy consumption before the operation and the cumulative amount of privacy consumption after the operation. The privacy ledger ensures the integrity and auditability of the records through cryptographic hash chain technology.

[0044] In this embodiment, by constructing an immutable privacy operation log, the transparency, traceability, and controllability of the differential privacy budget usage are significantly improved. This ensures the integrity and auditability of the data operation process and optimizes the allocation of privacy resources through historical record analysis, thereby improving the management efficiency of the privacy budget while strictly meeting compliance requirements. Of course, this is not the only approach; any method that can improve the management efficiency of the privacy budget while strictly meeting compliance requirements can be used.

[0045] Specifically, step S3, the working steps of the DP-mRMR feature selector in differential privacy feature engineering, include: The DP-mRMR feature selector first calculates the mutual information between each feature and the target variable using a mutual information calculation function. Differential privacy noise is introduced during this calculation to protect the impact of individual data records on the mutual information calculation results. The mutual information calculation function is treated as a query function, and its sensitivity is quantified before differential privacy noise is injected into the calculation results through a Laplace or Gaussian mechanism. Subsequently, the DP-mRMR feature selector calculates the mutual information between selected features and candidate features as a redundancy measure, again introducing differential privacy noise during this calculation. The DP-mRMR feature selector iteratively selects features, optimizing the differential privacy protection objective function of "maximizing relevance" and "minimizing redundancy" with each selection until a preset number of features or performance threshold is reached.

[0046] In this embodiment, while strictly adhering to differential privacy constraints, the needs of feature utility and privacy protection are successfully balanced. By injecting sensitivity-calibrated noise into the mutual information calculation process, it is ensured that the feature selection process does not leak sensitive information of individual data records. This allows the selected feature subset, under a privacy-preserving environment, to maintain a high correlation with the target variable while effectively reducing feature redundancy, providing a secure and highly discriminative feature foundation for subsequent machine learning modeling. Of course, this is not a limitation; other methods that reduce feature redundancy can also be used.

[0047] Specifically, step S3, the working steps of the stable selection mechanism in differential privacy feature engineering, include: By performing multiple random resamplings on the original dataset and running the DP-mRMR feature selector independently on the subset generated by each resampling, the frequency of each feature being selected in all resampling runs is counted, and those features that are consistently selected in resampling runs above a preset threshold are identified as the final subset of key features. In this embodiment, a stable selection mechanism is used to run DP-mRMR feature selection in parallel on multiple resampled data, and the final feature selection is performed based on the statistical consistency of feature selection frequencies. This significantly improves the robustness and stability of the feature selection process in differential privacy noise environments, effectively reducing the risk of feature misselection or omission due to noise interference, thereby ensuring the reliability and interpretability of the final feature subset. Of course, this is not a limitation; other methods that can guarantee the reliability and interpretability of the final feature subset are acceptable.

[0048] Specifically, step S3. Constructing key metrics with business insights based on selected features in differential privacy feature engineering includes the following steps: Based on the key features identified by the DP-mRMR feature selector and stable selection mechanism, a series of high-level competitive intelligence indicators with business insights are constructed. Differential privacy-preserving statistical methods are used to aggregate, derive, and calculate the original data or the data after anonymization.

[0049] In this embodiment, a differential privacy protection mechanism is used to construct key competitive intelligence indicators across multiple dimensions. This provides high-quality, trustworthy business insights while strictly protecting individual privacy and commercially sensitive data, thus meeting privacy compliance requirements and supporting enterprises in securely conducting market competitiveness analysis and strategic decision-making. Of course, this is not a limitation; any mechanism capable of simultaneously meeting privacy compliance requirements and supporting secure market competitiveness analysis and strategic decision-making can be implemented.

[0050] Specifically, step S4. Privacy-preserving analysis and prediction, which uses a differential privacy-preserving time-series model for market trend analysis and anomaly detection, includes the following steps: For the time-series data generated in step S3, Differential Privacy Feature Engineering, which involves constructing key indicators with business insights based on selected features, differential privacy trend analysis, periodic analysis, and anomaly pattern identification are performed. Differential privacy time-series models are used to model historical data of key indicators to predict future trends. In terms of anomaly detection, differential privacy statistical control chart methods or DP-based density estimation methods are used to identify anomaly patterns in the data, and alerts are issued through differential privacy threshold comparison or privacy-preserving outlier detection algorithms.

[0051] In this embodiment, by introducing a differential privacy protection mechanism into time series analysis and anomaly detection, individual data privacy is strictly protected while effectively supporting trend prediction, periodic analysis, and anomaly pattern recognition of key indicators. This satisfies data compliance requirements and ensures the usability of analysis results, providing reliable and privacy-secure support for business decisions. Of course, this is not a limitation; any other mechanism capable of providing reliable and privacy-secure support for business decisions can be used.

[0052] Specifically, step S4. The working steps of the stacked learning framework in privacy-preserving analysis prediction include: Under differential privacy constraints, the prediction results of multiple base learners are combined and the final prediction is made through a meta-learner. The stacked learning framework trains multiple heterogeneous base models. Each base model independently applies differential privacy during training. Subsequently, the prediction results of the base models are used as input features of the meta-learner, which also applies differential privacy during the training phase.

[0053] In this embodiment, a differential privacy-preserving stacked integration architecture effectively integrates the predictive advantages of multiple heterogeneous base models while strictly protecting data privacy. This significantly improves the model's generalization ability and robustness in noisy environments, while reducing the risk of overfitting, ultimately achieving more accurate and privacy-preserving predictive performance. Of course, this is not a limitation; other methods capable of achieving more accurate and privacy-preserving predictive performance can also be used.

[0054] Specifically, S5. Synthetic Data Augmentation includes the following steps: S51. Low-sample scenario identification: Continuously monitor and analyze the structure and distribution characteristics of the input dataset, automatically identify feature dimensions, market segments, or product categories with scarce samples or extremely unbalanced data distribution in the dataset; by calculating the statistics of each feature dimension and comparing them with the preset minimum sample threshold, determine which parts need data augmentation, and assess the risk of a serious decrease in data availability that may result from directly applying differential privacy analysis with the existing sample size, thereby determining the specific scenarios that need synthetic data supplementation and the required amount of synthetic data; S52.DP Synthetic Data Generation: Under the premise of satisfying differential privacy constraints, artificial synthetic data with similar statistical properties to the original data is generated; a synthesis method based on generative adversarial networks or variational autoencoders is adopted, in which differential privacy mechanisms are integrated into the training process of generators and discriminators, or a query-based synthesis method is adopted, in which the statistical features of the original data are learned through queries protected by differential privacy mechanisms. The result of each query is injected with differential privacy noise into the calculation result through Laplace or Gaussian mechanisms, and then these protected statistical features are used to construct synthetic data; S53. Robust Training: The data generated in step S52 is effectively fused with the original dataset to construct an enhanced training dataset. A semi-supervised learning or hybrid training strategy is adopted, combining some of the original data with a large amount of synthetic data to train the privacy-preserving analysis and prediction module for machine learning. The effect of synthetic data on improving model performance is evaluated and verified through cross-validation and performance metrics.

[0055] In this embodiment, by automatically identifying low-sample scenarios and generating differentially privacy-preserving synthetic data, the problems of data sparsity and imbalanced distribution are effectively solved. While strictly protecting individual privacy, it significantly improves the generalization ability and predictive stability of machine learning models in scarce data dimensions, achieving an efficient balance between privacy protection and data usability. Of course, this is not a limitation; other methods capable of achieving the same effect can be used. Example

[0056] S1. Input Data: Sales records for three months (90 days) were collected, totaling approximately 15,000 entries, including purchase time, product model, sales price, customer region, and customer age group. S2. Differential Privacy Processing and Budget Management: The privacy budget strategy sets the total privacy budget ε_total to 1.0 based on business needs. Considering the high sensitivity of sales price and customer region data, the engine allocates their ε value to 0.35, while product model and age group are allocated to 0.15. The remaining budget is allocated to other auxiliary information.

[0057] S3. Differential Privacy Feature Engineering: Noise Addition: The differential privacy noise addition mechanism desensitizes the original data based on the aforementioned ε value. For example, when querying the average selling price, the query function sensitivity Δ = 100 (assuming a price range of 0-1000 yuan), and ε_price = 0.35. Therefore, the scaling parameter b of the Laplace noise = Δ / ε_price = 100 / 0.35 ≈ 285.7. The actual query results will then include noise following a Lap(0, 285.7) distribution.

[0058] DP-mRMR Feature Selection: The DP-mRMR feature selector 21 first calculates the mutual information between each original feature and the target variable "daily sales volume" and injects Laplace noise (ε_FE=0.2). Subsequently, the selector iteratively filters and combines a stable selection mechanism 22 (100 bootstrap resampling cycles, stability threshold 80%). Finally, "average selling price (after anonymization)", "customer geographic distribution (after anonymization)", "historical advertising investment (after anonymization)" and "competitor price changes (after anonymization)" were selected as key features.

[0059] Key Indicator Construction: Based on the selected features, Key Indicator Construction Unit 23 generated advanced indicators such as "Regional Price Competitiveness Index (DP Protection)" and "Marketing Activity Influence Index (DP Protection)".

[0060] S4. Privacy Protection Analysis and Prediction: Low-sample scene identification: The low-sample scene identifyer finds that for certain niche areas (e.g., sales in a remote province), the number of samples is less than 50.

[0061] Synthetic Data Generation: The DP synthetic data generator employs a DP-GAN-based approach, utilizing the anonymized data and sparse region data identified by the low-sample discriminator to generate approximately 5000 differentially privacy-preserving synthetic sales records to expand the training dataset. During DP-GAN training, both the generator and discriminator are optimized using DP-SGD to ensure privacy protection during the training process.

[0062] Robust training: The robust training unit merges the original de-identified data with synthetic data to construct an enhanced dataset of approximately 20,000 records.

[0063] Sales Forecasting: The stacked ensemble learning engine is trained using an augmented dataset. The base learners consist of a DP-SGD-based LSTM model and a DP random forest, while the meta-learner is a DP linear regression model. The model predicts the average daily sales of the product over the next 30 days.

[0064] Uncertainty estimation: The uncertainty estimation unit uses a Bayesian difference privacy model to assess the uncertainty of the prediction results and calculates the 90% confidence interval for each prediction value.

[0065] S5. Synthetic Data Augmentation: Based on the prediction results and uncertainties, and combined with the potential competitor promotional activities discovered through time series analysis and anomaly detection, specific business recommendations are generated.

[0066] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0067] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this invention patent should be determined by the appended claims.

Claims

1. A method for competitive intelligence analysis based on differential privacy, characterized in that: Includes the following steps: S1. Data Input and Preprocessing: Collect raw data, perform data cleaning and standardization on the raw data to standardize the raw data, and determine the sensitivity level of the raw data; S2. Differential Privacy Processing and Budget Management: The differential privacy processing and budget management module, through a privacy budget strategy, dynamically allocates differential privacy budgets ε_i for each dimension based on the preset importance, sensitivity, and analytical accuracy requirements of business indicators using a multi-objective optimization algorithm. Under the premise of satisfying the total budget constraint Σε_i ≤ ε_total, it prioritizes the availability of high-value indicators. The differential privacy noise addition mechanism adaptively selects either a Laplace or Gaussian mechanism based on the privacy budget, injecting noise according to query sensitivity. The Rényi differential privacy accounting system, based on the Rényi divergence composite theorem, accurately tracks the accumulated privacy loss from multiple operations, providing a tighter upper bound for privacy protection. All operations are recorded in an immutable privacy ledger, ensuring full auditability and traceability. S3. Differential Privacy Feature Engineering: Under privacy constraints, the differential privacy feature engineering module selects highly relevant and low-redundancy features through the DP-mRMR feature selector, and improves noise robustness with the help of a stable selection mechanism. Finally, based on the selected features, it constructs key indicators with business insights. All operations are performed under differential privacy protection to ensure that individual data cannot be inferred. S4. Privacy Protection Analysis and Prediction: The privacy protection analysis and prediction module performs market trend analysis and anomaly detection using a differential privacy-preserving time series model, and integrates the prediction results of multiple differential privacy-preserving basic models based on a stacked ensemble learning framework to improve robustness; the privacy protection analysis and prediction module quantifies prediction uncertainty using Bayesian differential privacy or Bootstrap methods, and finally, relying on a strategy knowledge base and rule engine, automatically matches the analysis results and generates executable quantitative strategy suggestions; S5. Synthetic Data Augmentation: The differential privacy synthetic data generation module identifies scenarios with scarce data samples and uses the DP synthetic data generator to synthesize data with similar statistical characteristics to the original data under the premise of satisfying differential privacy constraints. Finally, the synthetic data is fused with the original data to enhance the training set, thereby improving the model's generalization ability and prediction robustness in low-sample scenarios. S6. Output the results; Step S4. Privacy-preserving analysis and prediction, which involves market trend analysis and anomaly detection using a differential privacy-preserving time series model, includes the following steps: For the time-series data generated in step S3. Differential privacy feature engineering, which involves constructing key indicators with business insights based on selected features, differential privacy trend analysis, periodic analysis, and anomaly pattern identification are performed. A differential privacy time-series model is used to model the historical data of the key indicators to predict future trends. In terms of anomaly detection, a differential privacy statistical control chart method or a density estimation method based on DP is used to identify anomaly patterns in the data, and an alert is issued through differential privacy threshold comparison or privacy-preserving outlier detection algorithm. The working steps of the stacked learning framework in step S4, privacy-preserving analysis and prediction, include: Under differential privacy constraints, the prediction results of multiple base learners are combined and a final prediction is made through a meta-learner. The stacked learning framework trains multiple heterogeneous base models. Each base model independently applies the differential privacy mechanism during training. Subsequently, the prediction results of the base models are used as input features of the meta-learner. The meta-learner also applies the differential privacy mechanism for protection during the training phase. 2.The method of competitive intelligence analysis based on differential privacy according to claim 1, wherein: The steps of the privacy budget strategy in step S2, differential privacy processing and budget management, include: Based on the preset sensitivity levels of each dimension, the expected analytical accuracy requirements, and the total privacy budget ε_total within the system's total budget constraint Σε_i ≤ ε_total, the multi-objective optimization algorithm is used to dynamically calculate and allocate the differential privacy budget value ε_i corresponding to each indicator dimension. The privacy budget strategy receives a pre-calibrated weight matrix regarding the business importance and sensitivity of different indicators as input and outputs each privacy budget value ε_i, ensuring that highly sensitive indicators receive stricter privacy protection, while allocating sufficient budget to indicators that contribute significantly to the analytical task to maintain data availability. 3.The method of competitive intelligence analysis based on differential privacy according to claim 2, characterized in that: The working steps of the differential privacy noise addition mechanism in step S2, differential privacy processing and budget management, include: The system receives the raw data, calculates the query results based on the raw data, and outputs the privacy budget value in the corresponding metric dimension of the privacy budget strategy. The differential privacy noise addition mechanism adaptively selects and executes a Laplace or Gaussian mechanism based on the data type and query sensitivity, adding random noise following a Laplace or Gaussian distribution to the query results. The function of the Laplace distribution is Lap(0,Δ / ε_i), and the function of the Gaussian distribution is N(0,σ²), where Δ represents the global sensitivity of the query function, ε_i is the privacy budget value, and σ² is the variance, reflecting the dispersion of the data distribution. The operational steps of the Rényi differential privacy accounting system in step S2, differential privacy processing and budget management, include: A novel privacy composition theorem based on Rényi divergence is employed to accurately calculate and track the cumulative privacy loss generated by multiple differential privacy operations, and to maintain a cumulative privacy consumption amount. The impact of each noise addition operation on the cumulative privacy consumption amount is monitored in real time. The Rényi differential privacy accounting system receives the parameters of each differential privacy operation and updates the total privacy consumption value according to the Rényi composition rule to ensure that the cumulative privacy loss does not exceed the preset total privacy budget.

4. The method of competitive intelligence analysis based on differential privacy according to claim 3, wherein: The privacy ledger in step S2, Differential Privacy Processing and Budget Management, is an immutable distributed log system used to record detailed metadata for each differential privacy operation. The metadata includes, but is not limited to, operation type, operation execution timestamp, dataset identifier involved, differential privacy budget value ε_consumed consumed, noise mechanism type used, specific parameters queried, and the cumulative privacy consumption before and after the operation. The privacy ledger ensures the integrity and auditability of the records through cryptographic hash chain technology.

5. The competitive intelligence analysis method based on differential privacy according to any one of claims 2-4, characterized in that: The steps of the DP-mRMR feature selector in step S3, differential privacy feature engineering, include: The DP-mRMR feature selector first calculates the mutual information between each feature and the target variable using a mutual information calculation function. Differential privacy noise is introduced during the mutual information calculation process to protect the impact of individual data records on the mutual information calculation results. The mutual information calculation function is treated as a query function, and its sensitivity is quantified. The differential privacy noise is then injected into the calculation results through the Laplace or Gaussian mechanism. Subsequently, the DP-mRMR feature selector calculates the mutual information between selected features and candidate features as a redundancy measure, and the differential privacy noise is also introduced during this calculation process. The DP-mRMR feature selector iteratively selects features, optimizing the differential privacy protection objective function of "maximizing relevance" and "minimizing redundancy" with each selection until a preset number of features or a performance threshold is reached.

6. The competitive intelligence analysis method based on differential privacy according to claim 5, characterized in that: The working steps of the stable selection mechanism in step S3, differential privacy feature engineering, include: By performing multiple random resamplings on the original dataset and running the DP-mRMR feature selector independently on the subset dataset generated by each resampling, the frequency of each feature being selected in all resampling runs is counted, and those features that are consistently selected in resampling runs above the preset threshold are identified as the final subset of key features. Step S3. Constructing key metrics with business insights based on selected features in differential privacy feature engineering specifically includes the following steps: Based on the key features identified by the DP-mRMR feature selector and stable selection mechanism, a series of high-level competitive intelligence indicators with business insights are constructed. Differential privacy-preserving statistical methods are used to aggregate, derive, and calculate the original data or the data after anonymization.

7. The competitive intelligence analysis method based on differential privacy according to claim 6, characterized in that: The S5. synthetic data enhancement includes the following steps: S51. Low-sample scenario identification: Continuously monitor and analyze the structure and distribution characteristics of the input dataset, automatically identify feature dimensions, market segments, or product categories with scarce samples or extremely unbalanced data distribution in the dataset; by calculating the statistics of each feature dimension and comparing them with the preset minimum sample threshold, determine which parts need data augmentation, and assess the risk of a serious decrease in data availability that may result from directly applying differential privacy analysis with the existing sample size, thereby determining the specific scenarios that need synthetic data supplementation and the required amount of synthetic data; S52.DP Synthetic Data Generation: Under the premise of satisfying the differential privacy constraints, artificial synthetic data with similar statistical characteristics to the original data is generated; a synthesis method based on generative adversarial networks or variational autoencoders is adopted, wherein the differential privacy mechanism is integrated into the training process of the generator and discriminator, or a query-based synthesis method is adopted, in which the statistical characteristics of the original data are learned through queries protected by the differential privacy mechanism, and the result of each query is injected with the differential privacy noise into the calculation result through the Laplace mechanism or the Gaussian mechanism, and then these protected statistical characteristics are used to construct the synthetic data; S53. Robust Training: The data generated in step S52 is effectively fused with the original dataset to construct an enhanced training dataset. A semi-supervised learning or hybrid training strategy is adopted, combining a portion of the original data with a large amount of the synthetic data to train the privacy-preserving analysis and prediction module for machine learning. The effect of the synthetic data on improving model performance is verified through cross-validation and performance metrics evaluation.

8. An analysis system applied to the differential privacy-based competitive intelligence analysis method of claims 1-7, characterized in that: It includes a differential privacy processing and budget management module, a differential privacy feature engineering module, a privacy protection analysis and prediction module, and a differential privacy synthetic data generation module; The differential privacy processing and budget management module is used to receive raw sensitive business data, and perform differential privacy noise addition processing on the raw sensitive business data according to preset privacy protection requirements. At the same time, it performs precise privacy budget allocation, tracking and auditing for all privacy-related operations. The differential privacy feature engineering module is used to select and construct key features with optimal predictive ability and that meet privacy protection requirements from the de-identified data under differential privacy constraints. The privacy protection analysis and prediction module is used to perform market trend analysis, competitive strategy prediction and abnormal behavior detection using the privacy protection dataset processed by the differential privacy feature engineering module, and provides a quantitative assessment of prediction uncertainty. The differential privacy synthetic data generation module is used to identify and process specific scenarios with scarce data samples. It enhances the dataset by generating differential privacy-preserving synthetic data, thereby improving the analysis and prediction performance in low-sample scenarios.

Images