Product recommendation method and device, computer equipment and storage medium

Abstract

The invention discloses a product recommendation method and device, computer equipment and a storage medium, belongs to the technical field of artificial intelligence, and is applied to a recommendation scene of financial products. According to the method, end-to-end data standardization, semantic analysis and privacy protection are carried out; performing disturbance and noise injection based on differential privacy; the method comprises the following steps: accurately capturing short-term fluctuation and long-term trend of customer demands by virtue of a multi-level time sequence algorithm, and extracting a typical demand change mode through segmented clustering and modeling; when significant demand fluctuation is detected, the online learning and deep reinforcement learning model can update a recommendation strategy in real time; the interpretability analysis module transparently presents a recommendation decision basis; user feedback is continuously absorbed through incremental learning, model parameters are continuously optimized, and finally it is guaranteed that an accurate personalized recommendation result is output under millisecond-level response. According to the method, the recommendation accuracy, the user satisfaction and the service conversion rate are remarkably improved, and the compliance and the expansibility are considered at the same time.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of artificial intelligence, and particularly relates to a product recommendation method and device, computer equipment and a storage medium. BACKGROUND

[0002] The insurance industry is accelerating towards intelligentization, and data-driven personalized recommendation has become the core of improving customer experience and market competitiveness. High-quality recommendations not only accurately match customer needs, but also quickly respond to market changes, helping enterprises optimize resource allocation. However, the defects of existing recommendation systems gradually appear, mainly in the contradiction between data integration and privacy protection, the lack of dynamic capture of customer needs, and the weak explainability of recommendation results. These problems make it difficult for the recommendation results to fully meet the personalized needs of customers, limiting the adaptability of insurance companies in complex market environments.

[0003] Currently, the diversity and heterogeneity of data sources make cross-channel information integration a major problem. Data from different channels differ in format, semantics, and update frequency, making it difficult to form a unified data foundation. For example, customer demand information generated through online consultation, offline stores, and mobile applications often cannot be effectively integrated due to semantic inconsistencies, resulting in the recommendation system being unable to fully understand the customer's real needs. In addition, the increasing demand for privacy protection further exacerbates this problem. Traditional data processing methods often sacrifice data integrity to meet privacy compliance requirements, weakening the analysis capabilities of the recommendation system.

[0004] Based on this, the dynamic capture of customer needs becomes another key challenge. Customer needs are driven by multiple factors and exhibit temporal variation characteristics. For example, a customer may develop new needs for health insurance after purchasing life insurance due to changes in family structure, but existing systems are unable to real-time perceive such changes and update the recommendation strategy in a timely manner. The lack of dynamic updating mechanism results in lagging recommendation results, making it difficult to adapt to the rapid evolution of customer needs.

[0005] How to effectively integrate cross-channel data and accurately capture dynamic customer needs while ensuring privacy compliance has become a key issue in improving the effectiveness of the recommendation system. SUMMARY

[0006] The purpose of the embodiments of the present application is to propose a product recommendation method, device, computer equipment and storage medium to solve the technical problem of effectively integrating cross-channel data and accurately capturing dynamic customer needs.

[0007] To solve the above technical problems, the embodiments of the present application provide a product recommendation method, which adopts the following technical solutions:

[0008] A product recommendation method, comprising:

[0009] obtain customer interaction data from a multi-channel database, uniformly format the customer interaction data in different formats by using a data standardization protocol, and generate a uniform format data set;

[0010] For the uniform format data set, a semantic analysis model is applied to analyze data semantics to obtain a semantic consistency feature set;

[0011] If the semantic consistency feature set meets a preset semantic integrity threshold, the semantic consistency feature set is processed by differential privacy technology to generate a privacy protection data set;

[0012] For the privacy protection data set, a time series analysis algorithm is used to extract time series variation features of customer demand to obtain a demand change trend;

[0013] If the demand change trend exceeds a preset dynamic change threshold, the recommendation model parameters are updated by an online learning algorithm to generate a dynamic recommendation strategy;

[0014] An explainability analysis method is applied to generate an explanatory description of the dynamic recommendation strategy to obtain an interpretable recommendation output, and the interpretable recommendation output is pushed to the customer;

[0015] Customer feedback data is obtained, and an incremental learning algorithm is used to optimize the recommendation model to generate an updated recommendation model;

[0016] According to the updated recommendation model, combined with the real-time obtained customer interaction data, a personalized recommendation result is generated.

[0017] To solve the above technical problems, the embodiment of the application also provides a product recommendation device, which adopts the technical scheme as follows:

[0018] A product recommendation device comprises:

[0019] A format conversion module is configured to obtain customer interaction data from a multi-channel database, uniformly format the customer interaction data in different formats by using a data standardization protocol, and generate a uniform format data set;

[0020] A semantic analysis module is configured to apply a semantic analysis model to analyze data semantics for the uniform format data set to obtain a semantic consistency feature set;

[0021] An integrity verification module is configured to process the semantic consistency feature set by differential privacy technology to generate a privacy protection data set if the semantic consistency feature set meets a preset semantic integrity threshold;

[0022] A time series variation module is configured to extract time series variation features of customer demand for the privacy protection data set by using a time series analysis algorithm to obtain a demand change trend;

[0023] The first updating module is configured to update the recommendation model parameters by using an online learning algorithm to generate a dynamic recommendation strategy if the demand change trend exceeds a preset dynamic change threshold.

[0024] The explanatory description module is configured to generate an explanatory description of the dynamic recommendation strategy by using an explainable analysis method, to obtain an explainable recommendation output, and to push the explainable recommendation output to the customer.

[0025] The second updating module is configured to obtain customer feedback data and optimize the recommendation model by using an incremental learning algorithm to generate an updated recommendation model.

[0026] The personalized recommendation module is configured to generate a personalized recommendation result according to the updated recommendation model and in combination with real-time obtained customer interaction data.

[0027] To solve the above technical problems, the embodiments of the present application further provide a computer device, which adopts the technical solutions as follows:

[0028] A computer device includes a memory and a processor, the memory stores computer readable instructions, and the processor implements the steps of the product recommendation method according to the computer readable instructions.

[0029] To solve the above technical problems, the embodiments of the present application further provide a computer readable storage medium, which adopts the technical solutions as follows:

[0030] A computer readable storage medium stores computer readable instructions, and the computer readable instructions are executed by a processor to implement the steps of the product recommendation method according to any one of the above.

[0031] Compared with the prior art, the embodiments of the present application have the following beneficial effects:

[0032] The product recommendation method, device, computer equipment and storage medium provided by the application belong to the technical field of artificial intelligence, and are applied to the recommendation scene of financial products. The application effectively constructs high-quality and traceable customer interaction features through end-to-end data standardization, semantic analysis and privacy protection; based on the perturbation and noise injection of differential privacy, the main semantic structure is reserved on the premise of protecting sensitive information of users; with the help of multi-level time series algorithm, short-term fluctuations and long-term trends of customer demand are accurately captured, and typical demand change patterns are extracted through segmented clustering and modeling, thereby realizing deep insight into customer behavior; when significant demand fluctuations are detected, the online learning and deep reinforcement learning model can update the recommendation strategy in real time, automatically adjust the product combination, priority and explanation logic; the explainability analysis module presents the basis for the recommendation decision in a transparent manner, thereby improving the trust of customers and decision makers; through incremental learning, user feedback is continuously absorbed, and model parameters are continuously optimized, thereby finally ensuring that accurate personalized recommendation results are output under millisecond-level response. The application significantly improves the accuracy of recommendation, user satisfaction and business conversion rate, while taking into account compliance and scalability. BRIEF DESCRIPTION OF DRAWINGS

[0033] In order to more clearly illustrate the schemes in the application, the drawings needed in the description of the embodiments of the application will be briefly introduced. Obviously, the drawings in the following description are some embodiments of the application, and other drawings can be obtained by those skilled in the art without creative labor on the basis of these drawings.

[0034] Figure 1 An exemplary system architecture diagram in which the application can be applied is shown;

[0035] Figure 2 A flowchart of one embodiment of the product recommendation method according to the application is shown;

[0036] Figure 3 A structural schematic diagram of one embodiment of the product recommendation device according to the application is shown;

[0037] Figure 4 A structural schematic diagram of one embodiment of the computer equipment according to the application is shown. DETAILED DESCRIPTION

[0038] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs; the terms used in the specification are intended to describe the particular embodiments and are not intended to limit the application; the terms "include" and "have" and their any variations used in the specification and the claims and the above description of drawings are intended to cover the non-exclusive inclusion; the terms "first", "second" and the like used in the specification and the claims and the above description of drawings are intended to distinguish different objects, not to describe a particular order.

[0039] Reference herein to "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment can be included in at least one embodiment of the application. The appearances of the phrase in various places in the specification are not necessarily all referring to the same embodiment, nor are they necessarily mutually exclusive or alternative embodiments. It is expressly understood that the embodiments described herein are combinable with each other.

[0040] In order to make the person skilled in the art better understand the scheme of the present application, the technical solutions in the embodiments of the present application will be described clearly and completely in conjunction with the drawings below.

[0041] As shown in Figure 1 The system architecture 100 can include a terminal device 101, a network 102 and a server 103, and the terminal device 101 can be a notebook computer 1011, a tablet computer 1012 or a mobile phone 1013. The network 102 is a medium for providing a communication link between the terminal device 101 and the server 103. The network 102 can include various connection types, such as wired, wireless communication links or optical fiber cables, etc.

[0042] The user can use the terminal device 101 to interact with the server 103 through the network 102 to receive or send messages, etc. Various communication client applications can be installed on the terminal device 101, such as web browser applications, shopping applications, search applications, instant messaging tools, email clients, social platform software, etc.

[0043] The terminal device 101 can be various electronic devices with a display screen and supporting web browsing, in addition to the notebook computer 1011, the tablet computer 1012 or the mobile phone 1013, the terminal device 101 can also be an electronic book reader, an MP3 player (Moving Picture Experts Group Audio Layer III), an MP4 (Moving Picture Experts Group Audio Layer IV) player, a laptop computer and a desktop computer, etc.

[0044] The server 103 can be a server providing various services, for example, a background server providing support for a page displayed on the terminal device 101.

[0045] It should be noted that the product recommendation method provided by the embodiments of the present application is generally executed by a server / terminal device, and accordingly, the product recommendation apparatus is generally arranged in a server / terminal device.

[0046] It should be understood that, Figure 1 The number of terminal devices, networks and servers in the above system is only illustrative, and the above system can have any number of terminal devices, networks and servers according to implementation needs.

[0047] With reference to Figure 2 , a flow chart of one embodiment of a product recommendation method according to the present application is shown. The product recommendation method comprises the following steps:

[0048] S201, obtaining customer interaction data from a multi-channel database, using a data standardization protocol to uniformly format the customer interaction data in different formats to generate a uniform format data set;

[0049] Specifically, customer interaction data usually comes from multiple channels, including APP clicks, swipes, dwell time, form submission information, voice-to-text in customer service systems, online chat records, email feedback, offline store purchase records and identity recognition data, and third-party cooperation agency behavior data, such as health devices, financial accounts, or travel platform API data. These data have heterogeneous formats, including structured data (such as database records), semi-structured data (such as JSON format behavior logs), and even unstructured data (such as voice, images, and free text). To achieve unified processing, a set of data standardization protocols need to be designed, covering field mapping rules, timestamp alignment mechanisms, and data integrity verification processes, and ETL tools (such as Apache Nifi, Airflow) are used to complete the extraction and format conversion of multi-source data. The converted data is mapped into a unified data model through Schema mapping, and columnar storage or intermediate data tables are used to improve subsequent processing efficiency. At the same time, a metadata tagging system needs to be set up to retain the original data source and processing logs.

[0050] S202, for the unified format data set, a semantic analysis model is applied to analyze the data semantics to obtain a semantic consistency feature set;

[0051] Specifically, the core goal of the semantic analysis model is to identify and normalize the intent and context relationship expressed in customer interaction data, and then extract structured semantic features. For this purpose, a multi-level semantic analysis strategy can be used. The first layer performs word segmentation, named entity recognition (NER), and sentiment analysis on customer text information (such as consultation records, comments, and customer service conversations), and uses pre-trained language models (such as BERT and RoBERTa) for context embedding coding to obtain sentence vector representation. The second layer identifies the core topics of customer attention through semantic clustering and topic modeling (such as LDA and Top2Vec), such as "health insurance claim difficulties" and "accident insurance discount activities". The third layer combines knowledge graphs and uses embedding algorithms such as TransE and ComplEx to align entities in text with nodes in the insurance product library and service terms library, ensuring that terms mentioned by customers such as "critical illness insurance" and "deductible" are accurately identified and normalized. The final semantic consistency feature set is a set of semantic labels in a high-dimensional vector space, which maintains consistency in business knowledge semantics.

[0052] S203, if the semantic consistency feature set meets the pre-set semantic integrity threshold, the semantic consistency feature set is processed through differential privacy technology to generate a privacy-protected data set;

[0053] Specifically, the setting of the semantic integrity threshold can be based on indicators such as feature coverage, semantic confidence, and entity matching rate, to ensure that the feature set is representative in the sample and has a high information density. When the threshold is met, to protect user privacy, differential privacy (DP) technology is needed to perturb the feature dataset. The implementation of differential privacy can choose Laplace mechanism or Gaussian mechanism, and the noise injection strength is controlled according to the set privacy budget parameter epsilon. In terms of technical implementation, for numerical features, sensitivity analysis method is used to evaluate the perturbation boundary, and then corresponding noise is added; for categorical semantic labels, exponential mechanism is used to select the output, so that it meets the privacy constraints statistically. At the same time, local differential privacy (LDP) is used to realize data scrambling on the edge device, avoiding the transmission of original data to the server, and improving the overall privacy protection level of the system. The generated privacy-protected dataset retains the main features of data distribution and semantic structure, and can continue to be used for model training and analysis without exposing sensitive information.

[0054] It should be noted that if the semantic consistency feature set does not meet the preset semantic integrity threshold, the system will start the semantic completion and correction mechanism. First, the incomplete feature set is supplemented through the semantic enhancement module, and the pre-trained language model (such as BERT or ERNIE) is used for context reasoning to fill in the missing or ambiguous semantic fragments. Second, the system will introduce a domain knowledge graph to correct and disambiguate low-confidence entities, improving semantic accuracy. At the same time, the semantic patterns of similar user groups can be called through transfer learning to infer the potential semantic features of the current user. If the threshold requirement cannot still be met, the recommendation process for this sample is aborted to avoid recommendation bias caused by insufficient semantic information.

[0055] S204, for the privacy-protected dataset, a time series analysis algorithm is used to extract the time sequence change characteristics of customer demand, and a demand change trend is obtained;

[0056] Specifically, the evolution of customer needs exhibits a strong time dependency, thus necessitating the use of time series modeling methods to model and predict behavioral trajectories within the privacy-preserving dataset. First, the dataset is reconstructed along the time dimension to form interaction sequences tailored to individual customers, including timestamps, interaction events, and semantic features. For algorithm selection, deep learning-based sequence modeling methods, such as LSTM, GRU, or their improved Transformer architecture, are employed to effectively capture long-term dependencies and local temporal patterns. To enhance the model's ability to perceive key change nodes, an attention mechanism can be introduced, allowing the model to focus on abrupt changes or high-weight feature events in customer behavior. In multivariate scenarios (such as parallel inputs of user age, income, and preferred insurance types), multi-channel parallel input modeling or Temporal Convolutional Networks (TCN) can be used for parallel feature extraction. The model output is a demand trend curve reflecting the evolution of customer preferences for different insurance types over different time periods. Furthermore, dynamic threshold detection algorithms (such as CUSUM and Z-score) can be used to identify trend change points.

[0057] S205. If the demand change trend exceeds the preset dynamic change threshold, the recommendation model parameters are updated through an online learning algorithm to generate a dynamic recommendation strategy.

[0058] Specifically, online learning algorithms allow recommendation models to incrementally update their parameters when receiving new data, rather than retraining the entire model. This enables the model to respond in real-time to changes in customer behavior. When the trend of changing demand exceeds a set threshold (e.g., a year-on-year increase of more than 30% in the popularity of a certain insurance product), the system will trigger the model update module. Online learning can be implemented using methods such as incremental stochastic gradient descent (SGD), FTRL (Follow The Regularized Leader), and online random forests, depending on the model structure. If a neural network model is used for recommendation, techniques such as ElasticWeight Consolidation (EWC) can be employed to mitigate the catastrophic forgetting problem. To maintain the stability of recommendations, a sliding window strategy should be incorporated into the update process, i.e., updating only using the interaction data from the most recent n days or n rounds. Furthermore, to improve the timeliness and accuracy of personalized recommendations, multi-armed gambling algorithms or context-based bandit models can be introduced, using trial and error and feedback mechanisms to dynamically optimize the recommendation strategy. The resulting dynamic recommendation strategy effectively reflects changes in customer preferences, improving the timeliness of recommendations.

[0059] It should be noted that if the trend of demand changes does not exceed the preset dynamic change threshold, the system will maintain the current recommendation model parameters unchanged and continue to use the existing recommendation strategy for output. Simultaneously, the system will record the user's demand change trend in the behavior trajectory database as the basis for subsequent trend evolution monitoring. To prevent small fluctuations from accumulating over time and affecting model performance, the system will also periodically perform batch analysis of accumulated minor changes to determine if there are any potential structural behavioral shifts. Furthermore, a lightweight model fine-tuning mechanism can be triggered, performing low-intensity updates only to some weights to ensure the continuity and stability of recommendations while reducing unnecessary resource consumption.

[0060] S206, Apply interpretability analysis methods to generate an interpretable description of the dynamic recommendation strategy, obtain interpretable recommendation output, and push the interpretable recommendation output to the customer;

[0061] Specifically, the interpretability output of the dynamic recommendation strategy aims to transparently demonstrate the decision-making basis behind the recommendation results to customers and system operators. This step can employ various interpretability analysis methods, such as SHAP (Shapley Additive Explanations) value analysis, which uses game theory to evaluate the marginal contribution of each feature in the final recommendation result; or LIME (Local Interpretable Model-Agnostic Explanations) to generate a locally interpretable model, producing a linearly approximate interpretable expression for each specific recommendation. In deep neural network structures, the model's focus area can also be analyzed through features maps of intermediate layers, visualization of attention weights, Grad-CAM, and other methods. Furthermore, the recommendation logic can be transformed into a rule tree format, such as "IF age > 50 AND recently focused on critical illness insurance THEN recommend product X," making it easier for customers to understand intuitively. The final recommendation result should be accompanied by a natural language explanation output module, describing the reasons for the recommendation in a customer-friendly way and pushing it to customers through channels such as app notifications, emails, and chat windows, achieving a recommendation process that is "understandable, questionable, and trustworthy."

[0062] S207: Obtain customer feedback data and use incremental learning algorithms to optimize the recommendation model, generating an updated recommendation model;

[0063] Specifically, customer feedback data includes explicit and implicit feedback signals such as click behavior, purchase conversion, page dwell time, recommendation satisfaction rating, and reasons for recommendation rejection. The system needs to perform behavioral recognition and tag updates on this data to construct a set of realistic feedback tags. Recommendation model optimization can employ incremental learning mechanisms, continuously ingesting feedback data and adjusting model weights or parameters. For example, in neural network-based recommendation systems, new feedback samples can be introduced through mini-batch online training or replay buffer technology to stabilize model updates and prevent overfitting. If collaborative filtering is used, the user-item rating matrix can be dynamically updated using matrix completion algorithms (such as ALS). During this process, to improve update efficiency and response speed, parallel training or multi-model alternating deployment strategies can be adopted to achieve dynamic switching between "hot models" and "cold models." Furthermore, feedback data can also be used to update semantic feature weights and behavioral sequence importance scores, improving the model's predictive ability for future user behavior. Ultimately, the recommendation model will better match customers' real needs, improving recommendation accuracy.

[0064] S208 generates personalized recommendation results based on the updated recommendation model and real-time customer interaction data.

[0065] Specifically, personalized recommendations are generated by integrating the latest customer behavior data in real time onto an optimized model. Customer interaction data includes clickstreams, mouse movements, input keywords, and search history during the current session. The system needs to perform rapid preprocessing on this data (such as feature extraction, word vector generation, and sequence encoding) within millisecond response time and input it into the updated recommendation model. The recommendation model generates multiple candidate recommendation results (Top-N candidates) based on the customer's historical preferences and current context state. These are then filtered and sorted using real-time strategy rules (such as inventory availability, recommendation frequency control, and deduplication of similar products) to output the final recommendation list. Furthermore, to further optimize the user experience, iterative strategy methods from reinforcement learning can be used to adjust the recommendation strategy based on immediate feedback (such as adjustments after a click). Personalized recommendation results should support simultaneous display across multiple devices and reserve a user feedback channel to provide data sources for subsequent model iterations, forming a closed-loop recommendation optimization mechanism.

[0066] Furthermore, the steps of processing the semantic consistency feature set using differential privacy techniques to generate a privacy-preserving dataset specifically include:

[0067] The semantic consistency feature set is subjected to semantic perturbation processing to obtain the semantic perturbation feature set;

[0068] Add semantic noise features to the semantic consistency feature set to obtain the semantic noise feature set;

[0069] By combining semantic consistency feature sets, semantic perturbation feature sets, and semantic noise feature sets, a privacy-preserving dataset is constructed.

[0070] In this embodiment, firstly, the system performs semantic perturbation processing on the semantic consistency feature set. Based on the semantic embedding space of a language model (such as GPT or BERT), key entities or context words are replaced with synonyms, semantically blurred, or structurally rearranged without significantly affecting the semantic topic, generating a semantic perturbation feature set to enhance semantic diversity. Secondly, semantic noise features are introduced into the semantic consistency features. This involves artificially adding forged features that are not strongly related to the context but maintain statistical consistency. The intensity of the added noise is controlled through Laplace or Gaussian mechanisms to ensure that it meets the ε-difference privacy constraint. Finally, the original semantic consistency feature set, the semantic perturbation feature set, and the semantic noise feature set are fused to form a ternary feature combination structure, constructing a complete privacy-preserving dataset. This retains the main semantic information required for model learning while increasing the difficulty for external attackers to reconstruct the original data, thereby effectively protecting semantic layer privacy.

[0071] Through the above steps, the system improves the usability of semantic features while achieving differential privacy protection in the semantic dimension, effectively reducing the risk of leakage of users' sensitive information.

[0072] Furthermore, for the privacy-protected dataset, the steps of extracting the temporal variation characteristics of customer demand using time series analysis algorithms to obtain the demand change trend specifically include:

[0073] Obtain time-series data from privacy-preserving datasets;

[0074] The time series data are decomposed using a pre-defined autoregressive moving average model to obtain the time series variation characteristics;

[0075] If the continuous fluctuation amplitude of the time series change feature within the preset time period is greater than the preset fluctuation threshold, then the time series change feature is denoised by wavelet transform to obtain a smooth feature sequence.

[0076] Based on the smoothed feature sequence, the exponential smoothing method is used to predict future changes in customer demand, thus obtaining the customer demand trend.

[0077] The customer demand trend is segmented and analyzed using the sliding window method to obtain a segmented demand trend sequence.

[0078] If the rate of change between consecutive segments of the segmented demand trend sequence is greater than the preset demand trend threshold, then cluster analysis is performed on the segmented demand trend sequence to obtain the demand change pattern.

[0079] Based on the demand change pattern, a time series decomposition algorithm is used to extract long-term trends, thus obtaining the final demand change trend of customers.

[0080] In this embodiment, the system performs a multi-stage time series analysis process step by step on the privacy-preserving dataset to accurately identify the dynamic evolution trend of customer demand. First, the system extracts behavioral feature sequences containing time labels from the differentially privacy-preserving data to construct a time series dataset. Since privacy perturbations may introduce noise, the system prioritizes using an autoregressive moving average (ARMA) model to model and decompose the time series data to extract basic time-series change features, including trend, seasonal, and random perturbation terms, thereby understanding the initial form of demand changes. Subsequently, the system determines whether the fluctuation amplitude of the time-series features within a set time period (e.g., a 30-day rolling window) exceeds a set threshold (e.g., 15% relative change). If the condition is met, wavelet transform (e.g., Discrete Wavelet Transform (DWT) or Continuous Wavelet Transform (CWT) is introduced to perform multi-scale decomposition and denoising of the sequence, thereby generating a smoother feature sequence with greater stability.

[0081] Based on smoothed feature sequences, the system employs exponential smoothing methods (such as the Holt-Winters triple-exponential method) to predict changes in customer demand over future time intervals, obtaining a preliminary demand trend curve. To further analyze local trend changes, the system divides the predicted curve into multiple time periods (e.g., every 7 days) using a sliding window method, comparing and analyzing the growth and decline rates between each period. If the rate of change between consecutive periods exceeds a set demand trend threshold (e.g., a 10% increase or decrease), the system calls clustering algorithms such as K-means or DBSCAN to perform pattern mining on these segmented trends, thereby identifying different demand change patterns such as "stable upward trend," "cyclical fluctuation," and "abrupt change." Finally, combining the pattern information from all trend segments, the system again calls time series decomposition algorithms (such as STL or Prophet) to extract the long-term trend term as the final customer demand change trend, used to determine whether a model update or strategy adjustment is necessary.

[0082] Through the above steps, the system can accurately and continuously perceive the evolution of customer demand trends while preserving data privacy, thereby improving its ability to control the timing and content of personalized recommendations.

[0083] Furthermore, the steps of decomposing the time series data using a pre-defined autoregressive moving average model to obtain the time series variation characteristics specifically include:

[0084] Perform a stationarity test on the time series data;

[0085] If the time series data is non-stationary, it can be transformed into stationary data through differencing or trend removal.

[0086] An autoregressive moving average model was used to fit the stationary series data to obtain the estimated values ​​of the model parameters.

[0087] Based on the parameter estimates, the time series data is decomposed to obtain trend component features, seasonal component features, and random component features as time series variation features.

[0088] In this embodiment, the system first performs a stationarity test on the time series data, using unit root tests (such as the ADF test or the KPSS test) to determine whether the data possesses stable mean and variance characteristics. If the test results indicate that the data is a non-stationary series, it is transformed into stationary series data through differencing operations (such as first-order differencing and second-order differencing) or trend removal methods (such as regression detrending and logarithmic transformation) to meet the prerequisites for Autoregressive Moving Average (ARMA) modeling. After completing the stationarization process, the system uses the least squares method or the maximum likelihood estimation method to estimate the parameters in the ARMA model (including the coefficients of the autoregressive term AR and the moving average term MA) to ensure that the model has the optimal fitting accuracy to the time series. After the model parameters are determined, the system decomposes the original time series into three types of characteristic components based on the mathematical structure of the ARMA model: a trend component representing overall trend changes, a seasonal component representing periodic changes, and an unpredictable fluctuation noise component (random component).

[0089] Through the above steps, the system can efficiently extract multi-level variation features of time series, enhancing the modeling accuracy and interpretability of customer behavior evolution.

[0090] Furthermore, the step of predicting future changes in customer demand using exponential smoothing based on the smoothed feature sequence to obtain customer demand trends specifically includes:

[0091] Exponential smoothing is applied to the smoothed feature sequence to obtain smoothed predicted values;

[0092] Based on the smoothed forecast values, a linear regression model or a nonlinear regression model is used to predict future changes in customer demand, resulting in a predicted trend sequence.

[0093] Perform trend analysis on the predicted trend sequence to determine the changing trend data of customer demand;

[0094] The data on changing customer demand trends and the predicted trend sequence are used as the output of customer demand trends.

[0095] In this embodiment, the system first applies exponential smoothing to the smoothed feature sequence obtained in the previous step to enhance the impact of the latest data points on the prediction results. Specifically, the system can use single exponential smoothing (suitable for data without trend changes), double exponential smoothing (considering linear trend changes), or triple exponential smoothing (Holt-Winters method, suitable for data containing both trend and seasonal characteristics), selecting the most suitable smoothing type based on the data characteristics. After smoothing, the system uses the smoothed predicted values ​​to construct a regression prediction model. If customer demand changes exhibit linear characteristics, a linear regression model can be applied to predict demand values ​​at several future time points; if demand behavior exhibits a non-linear trend, a non-linear regression model such as multinomial regression, support vector regression (SVR), or neural network regression model is used to model and predict future demand, obtaining a continuous predicted trend sequence. The system further performs trend analysis on this trend sequence, including slope judgment, local extremum detection, and acceleration / deceleration rate analysis, extracting core indicators such as trend direction, fluctuation cycle, and change intensity to form structured customer demand change trend data. Finally, the system outputs the predicted trend sequence and the changing trend data together, serving as a key basis for subsequent recommendation strategy adjustments or user segmentation.

[0096] Through the above steps, the system can efficiently identify and predict the evolution path of customer needs, enhancing the recommendation system's ability to make forward-looking judgments about future behaviors.

[0097] Furthermore, if the rate of change between consecutive segments of the segmented demand trend sequence is greater than a preset demand trend threshold, then the segmented demand trend sequence is subjected to cluster analysis to obtain the demand change pattern. This process specifically includes:

[0098] The segmented demand trend sequence is clustered using either K-means clustering or hierarchical clustering to obtain multiple clusters.

[0099] Calculate the central tendency feature for each cluster, where the central tendency feature represents a pattern of demand change;

[0100] Based on the central trend characteristics, a pre-set Markov chain model is used to predict the probability of customer demand state transitions within a future time period, thus obtaining a demand state transition sequence.

[0101] Pattern matching is performed on the demand state transition sequence to determine the demand change pattern that best matches the current customer demand trend.

[0102] In this embodiment, when the system detects that the rate of change of a segmented demand trend sequence exceeds a preset demand trend threshold (e.g., a rate of change greater than 20%) over a continuous time period, indicating significant fluctuations or trend reversals in customer demand, the system will perform cluster analysis on the trend sequence to extract potential demand change patterns. Specifically, the system can use the K-means clustering algorithm to cluster each segmented trend vector, and measure the similarity between different trend segments based on Euclidean distance or DTW (Dynamic Time Warping) distance to quickly divide several clusters. In scenarios with few customer samples or where hierarchical structure needs to be preserved, hierarchical clustering algorithms (such as AGNES or DIANA) can be used to achieve bottom-up pattern merging. The central trend feature of each cluster is calculated using the average vector or centroid, representing the typical pattern of the trend sequence within the cluster, such as "slow rise," "periodic fluctuation," or "sudden drop followed by stabilization." Subsequently, the system introduces a Markov chain model, constructs a state transition matrix based on the central trend feature, predicts possible demand state transition paths for customers in future time periods, and generates a demand state transition sequence. Finally, the system dynamically matches the transfer sequence with the current customer's trend path, selects the optimal matching pattern through similarity scoring, and outputs the category of the customer's demand change pattern.

[0103] Through the above steps, the system can perform in-depth pattern mining and status prediction of customer trend changes, and realize precise characterization of demand behavior and personalized strategy push.

[0104] Furthermore, if the demand change trend exceeds a preset dynamic change threshold, the steps of updating the recommendation model parameters and generating a dynamic recommendation strategy through an online learning algorithm specifically include:

[0105] Determine whether the trend of demand change is significant based on the dynamic change threshold;

[0106] If the demand change trend exceeds the preset dynamic change threshold, a reinforcement learning algorithm will be used to optimize the recommendation model.

[0107] Define the state space of the reinforcement learning environment, which includes customer demand characteristics, historical recommendation behavior, and market trend information;

[0108] Define the action space for reinforcement learning, which includes the selection of recommended products, the generation of recommendation reasons, and the adjustment of recommendation priorities;

[0109] Based on historical recommendation data and customer feedback, the reinforcement learning model is trained using the Q-learning algorithm or deep reinforcement learning algorithm, enabling the model to learn to select the optimal state space and action space.

[0110] Based on the optimized recommendation model and combined with the trend of demand changes, a dynamic recommendation strategy is generated. The dynamic recommendation strategy includes a list of recommended products, the reasons for recommendation, and the recommendation priority.

[0111] The dynamic recommendation strategy is stored in the recommendation strategy database.

[0112] In this embodiment, the system determines whether to trigger the model update mechanism by judging whether the trend of demand changes exceeds a preset dynamic change threshold (such as the demand growth rate exceeding the average fluctuation range within a certain time window). If a significant change is determined, the system enters the online optimization process of the recommendation model, which uses reinforcement learning algorithms for adaptive training of the recommendation strategy. First, the system defines the state space of the reinforcement learning environment, including the customer's current demand characteristics (such as the types of insurance they are interested in, life cycle events, and health status), past recommendation behavior records (such as recommendation click-through rate, conversion rate, and bounce rate), and macro market trends (such as the popularity of specific insurance types and policy changes). Then, it constructs the action space, including the selection of recommended product combinations (such as medical insurance and accident insurance), the automatic generation of recommendation reasons (such as customized recommendation explanations based on health data), and the display order and priority setting of recommendation results. The system uses historical recommendation behavior data and user feedback as training samples and applies Q-learning algorithms or deep reinforcement learning (such as DQN and DDPG) for strategy training, enabling the model to autonomously learn the optimal action when facing different states, thereby maximizing the long-term reward function (such as customer satisfaction, conversion rate improvement, and long-term stickiness).

[0113] After training, the system generates a dynamic recommendation strategy from the optimized reinforcement learning model based on the current customer status and the latest demand trends. The output includes a list of recommended products (Top-N results), recommendation reasons (explainable content), and recommendation priority (sorted by customer preferences). The strategy is then structured and stored in the recommendation strategy database for subsequent use and version management.

[0114] Through the above steps, the system can iterate its strategies in real time according to changes in customer needs, enabling personalized, dynamic, and strategically strong intervention in recommended content, thereby comprehensively improving the adaptability and business value of recommendation responses.

[0115] In the above embodiments, this application discloses a product recommendation method, belonging to the field of artificial intelligence technology, applied to the recommendation scenario of financial products. This application effectively constructs high-quality, traceable customer interaction features through end-to-end data normalization, semantic parsing, and privacy protection; based on differential privacy perturbation and noise injection, it retains the main semantic structure while protecting sensitive user information; leveraging multi-level time series algorithms, it accurately captures short-term fluctuations and long-term trends in customer demand, and extracts typical demand change patterns through segmented clustering and modeling, achieving deep insights into customer behavior; when significant demand fluctuations are detected, online learning and deep reinforcement learning models can update recommendation strategies in real time, automatically adjusting product combinations, priorities, and explanation logic; the interpretability analysis module transparently presents the basis for recommendation decisions, enhancing trust between customers and decision-makers; through incremental learning, it continuously absorbs user feedback, constantly optimizing model parameters, ultimately ensuring accurate personalized recommendation results with millisecond-level response. This application significantly improves recommendation accuracy, user satisfaction, and business conversion rate, while also considering compliance and scalability.

[0116] In this embodiment, the product recommendation method operates on an electronic device (e.g., Figure 1 The server shown can receive instructions or acquire data via wired or wireless connection. It should be noted that the aforementioned wireless connection methods may include, but are not limited to, 3G / 4G connections, WiFi connections, Bluetooth connections, WiMAX connections, Zigbee connections, UWB (ultra-wideband) connections, and other currently known or future wireless connection methods.

[0117] It should be emphasized that, in order to further ensure the privacy and security of the aforementioned interactive data, the aforementioned interactive data can also be stored in a blockchain node.

[0118] The blockchain referred to in this application is a novel application model of computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanisms, and encryption algorithms. Essentially, a blockchain is a decentralized database, a chain of data blocks linked together using cryptographic methods. Each data block contains information about a batch of network transactions, used to verify the validity of the information (anti-counterfeiting) and generate the next block. A blockchain can include an underlying blockchain platform, a platform product service layer, and an application service layer.

[0119] The embodiments of this application can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence (AI) is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0120] Foundational technologies for artificial intelligence generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, operating / interactive systems, and mechatronics. AI software technologies mainly encompass computer vision, robotics, biometrics, speech processing, natural language processing, and machine learning / deep learning.

[0121] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by instructing related hardware with computer-readable instructions. These computer-readable instructions can be stored in a computer-readable storage medium. When executed, the program can include the processes of the embodiments of the above methods. The aforementioned storage medium can be a non-volatile storage medium such as a magnetic disk, optical disk, or read-only memory (ROM), or random access memory (RAM).

[0122] It should be understood that although the steps in the flowcharts of the accompanying figures are shown sequentially as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the accompanying figures may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the sub-steps or stages of other steps.

[0123] Further reference Figure 3 As a response to the above Figure 2 To implement the method shown, this application provides an embodiment of a product recommendation device, which is similar to... Figure 2 Corresponding to the method embodiments shown, this device can be specifically applied to various electronic devices.

[0124] like Figure 3 As shown, the product recommendation device 300 described in this embodiment includes:

[0125] The format conversion module 301 is used to obtain customer interaction data from multi-channel databases, and to perform unified format conversion on customer interaction data of different formats using a data standardization protocol to generate a unified format dataset.

[0126] Semantic analysis module 302 is used to apply a semantic analysis model to parse the semantics of a uniform format dataset and obtain a semantic consistency feature set.

[0127] The integrity verification module 303 is used to process the semantic consistency feature set through differential privacy technology to generate a privacy-protected dataset if the semantic consistency feature set meets the preset semantic integrity threshold.

[0128] The time series change module 304 is used to extract the time series change characteristics of customer demand for the privacy protection dataset using time series analysis algorithms to obtain the demand change trend;

[0129] The first update module 305 is used to update the recommendation model parameters and generate a dynamic recommendation strategy by means of an online learning algorithm if the demand change trend exceeds the preset dynamic change threshold.

[0130] The explanatory description module 306 is used to generate an explanatory description of the dynamic recommendation strategy by applying the interpretability analysis method, obtain an interpretable recommendation output, and push the interpretable recommendation output to the customer;

[0131] The second update module 307 is used to obtain customer feedback data and use an incremental learning algorithm to optimize the recommendation model and generate an updated recommendation model.

[0132] The personalized recommendation module 308 is used to generate personalized recommendation results based on the updated recommendation model and real-time customer interaction data.

[0133] Furthermore, the integrity verification module 303 specifically includes:

[0134] The semantic perturbation unit is used to perform semantic perturbation processing on the semantic consistency feature set to obtain the semantic perturbation feature set;

[0135] Semantic noise unit, used to add semantic noise features to the semantic consistency feature set to obtain a semantic noise feature set;

[0136] The feature set combination unit is used to combine semantic consistency feature sets, semantic perturbation feature sets, and semantic noise feature sets to construct a privacy-preserving dataset.

[0137] Furthermore, the timing change module 304 specifically includes:

[0138] Time series units are used to extract time series data from privacy-preserving datasets;

[0139] The data decomposition unit is used to decompose time series data using a preset autoregressive moving average model to obtain time series variation characteristics.

[0140] The denoising unit is used to denoise the time-series change features by wavelet transform if the continuous fluctuation amplitude of the time-series change features within a preset time period is greater than a preset fluctuation threshold, thereby obtaining a smooth feature sequence.

[0141] The demand forecasting unit is used to predict future changes in customer demand based on a smoothed feature sequence using an exponential smoothing method, thereby obtaining customer demand trends.

[0142] The segmented analysis unit is used to segment customer demand trends using the sliding window method to obtain segmented demand trend sequences.

[0143] The clustering analysis unit is used to perform clustering analysis on the segmented demand trend sequence if the rate of change between consecutive segments of the segmented demand trend sequence is greater than a preset demand trend threshold, so as to obtain the demand change pattern.

[0144] The trend extraction unit is used to extract long-term trends based on demand change patterns using a time series decomposition algorithm, thereby obtaining the final demand change trends of customers.

[0145] Furthermore, the data decomposition unit specifically includes:

[0146] The stationarity test subunit is used to perform stationarity tests on time series data.

[0147] The stationary transformation subunit is used to transform non-stationary time series data into stationary data through differencing or trend term removal methods.

[0148] The data fitting subunit is used to fit stationary series data using an autoregressive moving average model to obtain the parameter estimates of the model.

[0149] The time series data decomposition subunit is used to decompose the time series data based on the parameter estimates, and obtain the trend component features, seasonal component features, and random component features as time series variation features.

[0150] Furthermore, the demand forecasting unit specifically includes:

[0151] The smoothing subunit is used to perform exponential smoothing on the smoothed feature sequence to obtain smoothed predicted values.

[0152] The trend forecasting subunit is used to predict future changes in customer demand based on smoothed forecast values ​​using a linear or nonlinear regression model, thereby obtaining a predicted trend sequence.

[0153] The trend identification subunit is used to perform trend analysis on the predicted trend sequence to determine the changing trend data of customer demand.

[0154] The demand trend output sub-unit is used to output customer demand trend data and predicted trend sequences as customer demand trend results.

[0155] Furthermore, the cluster analysis unit specifically includes:

[0156] Cluster subunits are used to cluster segmented demand trend sequences using K-means clustering or hierarchical clustering algorithms to obtain multiple clusters;

[0157] The central trend sub-unit is used to calculate the central trend feature of each cluster, where the central trend feature represents a demand change pattern.

[0158] The state transition probability subunit is used to predict the probability of customer demand state transition within a future time period based on the central trend characteristics and using a preset Markov chain model, thereby obtaining the demand state transition sequence.

[0159] The demand change identification subunit is used to perform pattern matching on the demand state transition sequence to determine the demand change pattern that best matches the current customer demand trend.

[0160] Furthermore, the personalized recommendation module 308 specifically includes:

[0161] The trend judgment unit is used to determine whether the trend of demand change is significant based on the dynamic change threshold.

[0162] The model optimization unit is used to optimize the recommendation model using a reinforcement learning algorithm if the demand change trend exceeds a preset dynamic change threshold.

[0163] The state space setting unit is used to set the state space of the reinforcement learning environment, wherein the state space includes customer demand characteristics, historical recommendation behavior and market trend information;

[0164] The action space setting unit is used to define the action space of reinforcement learning, which includes the selection of recommended products, the generation of recommendation reasons, and the adjustment of recommendation priorities.

[0165] The reinforcement learning unit is used to train a reinforcement learning model based on historical recommendation data and customer feedback, using the Q-learning algorithm or deep reinforcement learning algorithm, so that the model learns to select the optimal state space and action space.

[0166] The dynamic recommendation unit is used to generate dynamic recommendation strategies based on the optimized recommendation model and the trend of demand changes. The dynamic recommendation strategy includes a list of recommended products, reasons for recommendation, and recommendation priority.

[0167] The recommendation strategy storage unit is used to store dynamic recommendation strategies in the recommendation strategy database.

[0168] In the above embodiments, this application discloses a product recommendation device, belonging to the field of artificial intelligence technology, applied to financial product recommendation scenarios. This application effectively constructs high-quality, traceable customer interaction features through end-to-end data normalization, semantic parsing, and privacy protection; based on differential privacy perturbation and noise injection, it retains the main semantic structure while protecting sensitive user information; leveraging multi-level time series algorithms, it accurately captures short-term fluctuations and long-term trends in customer demand, and extracts typical demand change patterns through segmented clustering and modeling, achieving deep insights into customer behavior; when significant demand fluctuations are detected, online learning and deep reinforcement learning models can update recommendation strategies in real time, automatically adjusting product combinations, priorities, and explanation logic; the interpretability analysis module transparently presents the basis for recommendation decisions, enhancing trust between customers and decision-makers; through incremental learning, it continuously absorbs user feedback, constantly optimizing model parameters, ultimately ensuring accurate personalized recommendation results with millisecond-level response. This application significantly improves recommendation accuracy, user satisfaction, and business conversion rates, while also considering compliance and scalability.

[0169] To address the aforementioned technical problems, embodiments of this application also provide a computer device. Please refer to [link / reference needed] for details. Figure 4 , Figure 4 This is a basic structural block diagram of the computer device in this embodiment.

[0170] The computer device 4 includes a memory 41, a processor 42, and a network interface 43 that are interconnected via a system bus. It should be noted that only the computer device 4 with memory 41, processor 42, and network interface 43 is shown in the figure; however, it should be understood that it is not required to implement all the components shown, and more or fewer components can be implemented alternatively. Those skilled in the art will understand that the computer device described here is a device capable of automatically performing numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes, but is not limited to, microprocessors, application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), digital signal processors (DSPs), embedded devices, etc.

[0171] The computer device can be a desktop computer, laptop, handheld computer, or cloud server, etc. The computer device can interact with the user via a keyboard, mouse, remote control, touchpad, or voice control.

[0172] The memory 41 includes at least one type of readable storage medium, including flash memory, hard disk, multimedia card, card-type memory (e.g., SD or DX memory), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 41 may be an internal storage unit of the computer device 4, such as the hard disk or memory of the computer device 4. In other embodiments, the memory 41 may also be an external storage device of the computer device 4, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the computer device 4. Of course, the memory 41 may also include both the internal storage unit and its external storage device of the computer device 4. In this embodiment, the memory 41 is typically used to store the operating system and various application software installed on the computer device 4, such as computer-readable instructions for product recommendation methods. In addition, the memory 41 can also be used to temporarily store various types of data that have been output or will be output.

[0173] In some embodiments, the processor 42 may be a central processing unit (CPU), controller, microcontroller, microprocessor, or other data processing chip. The processor 42 is typically used to control the overall operation of the computer device 4. In this embodiment, the processor 42 is used to execute computer-readable instructions stored in the memory 41 or to process data, such as executing computer-readable instructions for the product recommendation method.

[0174] The network interface 43 may include a wireless network interface or a wired network interface, which is typically used to establish communication connections between the computer device 4 and other electronic devices.

[0175] This application also provides an embodiment, namely, a computer device including a memory and a processor. The memory stores computer-readable instructions, and the processor, when executing the computer-readable instructions, implements the steps of the product recommendation method described above, that is, implements:

[0176] A product recommendation method includes:

[0177] Customer interaction data is obtained from multi-channel databases, and a data standardization protocol is used to convert customer interaction data of different formats into a unified format to generate a unified format dataset.

[0178] For a unified format dataset, a semantic analysis model is applied to parse the semantics of the data and obtain a semantic consistency feature set;

[0179] If the semantic consistency feature set meets the preset semantic integrity threshold, then the semantic consistency feature set is processed by differential privacy technology to generate a privacy-protected dataset.

[0180] For privacy-protected datasets, time series analysis algorithms are used to extract the temporal variation characteristics of customer demand and obtain the trend of demand changes;

[0181] If the demand change trend exceeds the preset dynamic change threshold, the recommendation model parameters are updated through an online learning algorithm to generate a dynamic recommendation strategy.

[0182] The interpretability analysis method is used to generate an interpretable description of the dynamic recommendation strategy, resulting in interpretable recommendation output, which is then pushed to the customer.

[0183] Obtain customer feedback data and use incremental learning algorithms to optimize the recommendation model, generating an updated recommendation model;

[0184] Personalized recommendation results are generated based on the updated recommendation model and real-time customer interaction data.

[0185] This application also provides another embodiment, namely, a computer-readable storage medium storing computer-readable instructions that can be executed by at least one processor to cause the at least one processor to perform the steps of the product recommendation method described above, i.e., to implement:

[0186] A product recommendation method includes:

[0187] Customer interaction data is obtained from multi-channel databases, and a data standardization protocol is used to convert customer interaction data of different formats into a unified format to generate a unified format dataset.

[0188] For a unified format dataset, a semantic analysis model is applied to parse the semantics of the data and obtain a semantic consistency feature set;

[0189] If the semantic consistency feature set meets the preset semantic integrity threshold, then the semantic consistency feature set is processed by differential privacy technology to generate a privacy-protected dataset.

[0190] For privacy-protected datasets, time series analysis algorithms are used to extract the temporal variation characteristics of customer demand and obtain the trend of demand changes;

[0191] If the demand change trend exceeds the preset dynamic change threshold, the recommendation model parameters are updated through an online learning algorithm to generate a dynamic recommendation strategy.

[0192] The interpretability analysis method is used to generate an interpretable description of the dynamic recommendation strategy, resulting in interpretable recommendation output, which is then pushed to the customer.

[0193] Obtain customer feedback data and use incremental learning algorithms to optimize the recommendation model, generating an updated recommendation model;

[0194] Personalized recommendation results are generated based on the updated recommendation model and real-time customer interaction data.

[0195] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0196] This application can be used in a wide variety of general-purpose or special-purpose computer system environments or configurations. Examples include: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics devices, network PCs, minicomputers, mainframe computers, and distributed computing environments including any of the above systems or devices. This application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform specific tasks or implement specific abstract data types. This application can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0197] It should be noted that the software tools or components not belonging to this company that appear in the various embodiments of this application are merely illustrative examples and do not represent actual use.

[0198] Obviously, the embodiments described above are only some embodiments of this application, not all embodiments. The accompanying drawings show preferred embodiments of this application, but do not limit the patent scope of this application. This application can be implemented in many different forms; rather, the purpose of providing these embodiments is to provide a more thorough and comprehensive understanding of the disclosure of this application. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing specific embodiments, or make equivalent substitutions for some of the technical features. Any equivalent structures made using the content of this application's specification and drawings, directly or indirectly applied to other related technical fields, are similarly within the scope of patent protection of this application.

Claims

1. A product recommendation method, characterized in that, include: Customer interaction data is obtained from multi-channel databases, and a data standardization protocol is used to convert customer interaction data of different formats into a unified format to generate a unified format dataset. For the aforementioned unified format dataset, a semantic analysis model is applied to parse the data semantics, resulting in a semantic consistency feature set; If the semantic consistency feature set meets the preset semantic integrity threshold, then the semantic consistency feature set is processed by differential privacy technology to generate a privacy-protected dataset; For the aforementioned privacy-protected dataset, a time series analysis algorithm is used to extract the temporal variation characteristics of customer demand and obtain the demand change trend; If the demand change trend exceeds the preset dynamic change threshold, the recommendation model parameters are updated through an online learning algorithm to generate a dynamic recommendation strategy. An interpretability analysis method is applied to generate an interpretable description of the dynamic recommendation strategy, resulting in an interpretable recommendation output, which is then pushed to the customer. Obtain customer feedback data and use an incremental learning algorithm to optimize the recommendation model, generating an updated recommendation model; Based on the updated recommendation model and combined with real-time customer interaction data, personalized recommendation results are generated.

2. The product recommendation method as described in claim 1, characterized in that, The step of processing the semantic consistency feature set using differential privacy technology to generate a privacy-preserving dataset specifically includes: The semantic consistency feature set is subjected to semantic perturbation processing to obtain a semantic perturbation feature set; Add semantic noise features to the semantic consistency feature set to obtain a semantic noise feature set; The privacy-preserving dataset is constructed by combining the semantic consistency feature set, the semantic perturbation feature set, and the semantic noise feature set.

3. The product recommendation method as described in claim 1, characterized in that, The step of extracting the time-series variation characteristics of customer demand using a time-series analysis algorithm to obtain the demand change trend for the privacy-protected dataset specifically includes: Obtain time-series data from the privacy-preserving dataset; The time series data is decomposed using a preset autoregressive moving average model to obtain time series variation characteristics; If the continuous fluctuation amplitude of the time-series change feature within a preset time period is greater than a preset fluctuation threshold, then the time-series change feature is denoised by wavelet transform to obtain a smooth feature sequence. Based on the smoothed feature sequence, the exponential smoothing method is used to predict future changes in customer demand, thereby obtaining customer demand trends. The customer demand trend is segmented and analyzed using the sliding window method to obtain a segmented demand trend sequence. If the rate of change between consecutive segments of the segmented demand trend sequence is greater than a preset demand trend threshold, then cluster analysis is performed on the segmented demand trend sequence to obtain the demand change pattern. Based on the aforementioned demand change pattern, a time series decomposition algorithm is used to extract long-term trends, thereby obtaining the final demand change trend of customers.

4. The product recommendation method as described in claim 3, characterized in that, The step of decomposing the time series data using a preset autoregressive moving average model to obtain time series variation characteristics specifically includes: The time series data were subjected to a stationarity test. If the time series data is non-stationary, it is converted into stationary data by difference operation or trend term removal method; The stationary sequence data were fitted using an autoregressive moving average model to obtain the estimated parameters of the model. Based on the parameter estimates, the time series data is decomposed to obtain trend component features, seasonal component features, and random component features as time series variation features.

5. The product recommendation method as described in claim 3, characterized in that, The step of predicting future changes in customer demand using exponential smoothing based on the smoothed feature sequence to obtain customer demand trends specifically includes: The smoothed feature sequence is subjected to exponential smoothing to obtain smoothed predicted values; Based on the smoothed forecast values, a linear regression model or a nonlinear regression model is used to predict future changes in customer demand, resulting in a predicted trend sequence. Perform trend analysis on the predicted trend sequence to determine the changing trend data of customer demand; The data on the changing trends of customer demand and the predicted trend sequence are used as the output results of the customer demand trend.

6. The product recommendation method as described in claim 3, characterized in that, If the rate of change between consecutive segments of the segmented demand trend sequence is greater than a preset demand trend threshold, then the step of performing cluster analysis on the segmented demand trend sequence to obtain the demand change pattern specifically includes: The segmented demand trend sequence is clustered using either K-means clustering or hierarchical clustering to obtain multiple clusters. Calculate the central tendency feature for each cluster, wherein the central tendency feature represents a demand change pattern; Based on the central trend characteristics, a preset Markov chain model is used to predict the probability of customer demand state transitions within a future time period, thereby obtaining a demand state transition sequence. Pattern matching is performed on the demand state transition sequence to determine the demand change pattern that best matches the current customer demand trend.

7. The product recommendation method as described in claim 1, characterized in that, The step of updating the recommendation model parameters and generating a dynamic recommendation strategy by means of an online learning algorithm if the demand change trend exceeds a preset dynamic change threshold specifically includes: Based on the dynamic change threshold, determine whether the demand change trend is significant; If the demand change trend exceeds a preset dynamic change threshold, then a reinforcement learning algorithm is used to optimize the recommendation model. Define a state space for the reinforcement learning environment, wherein the state space includes customer demand characteristics, historical recommendation behavior, and market trend information; Define an action space for reinforcement learning, wherein the action space covers the selection of recommended products, the generation of recommendation reasons, and the adjustment of recommendation priorities; Based on historical recommendation data and customer feedback, the reinforcement learning model is trained using the Q-learning algorithm or deep reinforcement learning algorithm, enabling the model to learn to select the optimal state space and action space. Based on the optimized recommendation model and the aforementioned demand change trend, the dynamic recommendation strategy is generated, wherein the dynamic recommendation strategy includes a list of recommended products, a reason for recommendation, and a recommendation priority. The dynamic recommendation strategy is stored in the recommendation strategy database.

8. A product recommendation device, characterized in that, include: The format conversion module is used to obtain customer interaction data from multi-channel databases, and to perform unified format conversion on customer interaction data of different formats using a data standardization protocol to generate a unified format dataset. The semantic analysis module is used to apply a semantic analysis model to parse the semantics of the data for the unified format dataset and obtain a semantic consistency feature set. The semantic integrity verification module is used to process the semantic consistency feature set through differential privacy technology to generate a privacy-protected dataset if the semantic consistency feature set meets the preset semantic integrity threshold. The time-series change module is used to extract the time-series change characteristics of customer needs from the privacy-protected dataset using a time-series analysis algorithm, and to obtain the demand change trend. The first update module is used to update the recommendation model parameters and generate a dynamic recommendation strategy by means of an online learning algorithm if the demand change trend exceeds a preset dynamic change threshold. The explanatory description module is used to generate an explanatory description of the dynamic recommendation strategy by applying an interpretability analysis method, obtain an interpretable recommendation output, and push the interpretable recommendation output to the customer; The second update module is used to obtain customer feedback data and optimize the recommendation model using an incremental learning algorithm to generate an updated recommendation model. The personalized recommendation module is used to generate personalized recommendation results based on the updated recommendation model and real-time customer interaction data.

9. A computer device, characterized in that, The method includes a memory and a processor, wherein the memory stores computer-readable instructions, and the processor executes the computer-readable instructions to implement the steps of the product recommendation method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-readable instructions that, when executed by a processor, implement the steps of the product recommendation method as described in any one of claims 1 to 7.

Images