A product recommendation method and device, computer equipment and a storage medium

Abstract

This application discloses a product recommendation method, apparatus, computer equipment, and storage medium, belonging to the field of artificial intelligence technology, and applied to intelligent recommendation of insurance telemarketing products. This application combines deep neural networks with genetic algorithms to achieve intelligent generation and continuous optimization of product recommendation strategies. By extracting multi-dimensional features from historical transaction data and constructing a unified feature representation, it performs refined predictions of short-term and long-term transaction probabilities under different recommendation strategies. The genetic algorithm engine uses the prediction results as the core basis for fitness evaluation, searching in a high-dimensional, discrete, and continuous mixed recommendation strategy space, effectively reducing the complexity and limitations of manual rule design. Simultaneously, by collecting transaction feedback data after actual recommendation execution and iteratively updating the deep neural network model and genetic algorithm engine, a closed-loop mechanism of model prediction and strategy optimization is formed.

Description

Technical Field

[0001] This application belongs to the field of artificial intelligence technology, specifically relating to a product recommendation method, apparatus, computer equipment, and storage medium. Background Technology

[0002] Traditional insurance sales models are increasingly revealing their limitations in the digital age. For a long time, this model has heavily relied on rule engines and the human experience of sales personnel to drive product recommendations and sales script selection. While rule engines provide standardized process guidance, they often struggle to flexibly respond to complex and ever-changing market demands and personalized customer characteristics, leading to a mismatch between recommended products and customer needs, and a severe lack of personalized service. Meanwhile, while human experience contains rich wisdom, it is limited by individual differences in ability, making it difficult to guarantee the stability and consistency of service quality. Furthermore, the traditional model is costly in the trial-and-error process; each strategy adjustment may be accompanied by customer churn and resource waste. In pursuing short-term conversion rates, it often neglects the cultivation and development of long-term customer value, making it difficult to form a sustainable business growth model.

[0003] Faced with these challenges, neither relying solely on machine learning classifiers for customer profiling and product matching, nor solely on heuristic rules for sales strategy formulation, can achieve global optimum under high-dimensional, non-linear business objectives and constraints. While machine learning can process large amounts of data and discover potential patterns, its decision-making capabilities are often limited when faced with complex business logic and real-time changing market environments. Heuristic rules, while providing intuitive and easy-to-understand decision-making basis, struggle to address data sparsity and uncertainty.

[0004] Therefore, exploring comprehensive solutions that integrate multiple technologies and take into account both short-term conversion and long-term value has become an urgent need for the transformation and upgrading of the insurance sales industry. Summary of the Invention

[0005] The purpose of this application is to propose a product recommendation method, apparatus, computer equipment, and storage medium to explore a comprehensive solution that integrates multiple technical means and takes into account both short-term conversion and long-term value.

[0006] To address the aforementioned technical problems, this application provides a product recommendation method, which employs the following technical solution: A product recommendation method, comprising: Historical transaction data is retrieved from a pre-set database, and features are extracted from the historical transaction data to obtain a set of transaction data features; Using a pre-trained deep neural network based on a set of transaction data features, the short-term and long-term transaction probabilities are predicted under a given recommendation strategy. The preset genetic algorithm engine is invoked, and short-term and long-term transaction probabilities are used as the core basis of the fitness function to search for matching product recommendation strategies in the recommendation strategy space. Based on the product recommendation strategy, determine the appropriate product recommendation combination and send the product recommendation combination to the client terminal; Obtain the actual transaction results and transaction data after the product recommendation combination is displayed, and iteratively update the deep neural network and genetic algorithm engine based on the actual transaction results and transaction data.

[0007] To address the aforementioned technical problems, this application also provides a product recommendation device, which employs the following technical solution: A product recommendation device, comprising: The feature extraction module is used to obtain historical transaction data from a preset database and extract features from the historical transaction data to obtain a set of transaction data features. The transaction prediction module is used to predict the short-term and long-term transaction probabilities based on a set of transaction data features using a pre-trained deep neural network. The recommendation strategy module is used to call the preset genetic algorithm engine, using short-term and long-term transaction probabilities as the core basis of the fitness function, to search for matching product recommendation strategies in the recommendation strategy space. The product portfolio module is used to determine the appropriate product recommendation portfolio based on the product recommendation strategy and send the product recommendation portfolio to the client terminal. The iterative update module is used to obtain the actual transaction results and transaction data after the product recommendation combination is displayed, and to iteratively update the deep neural network and genetic algorithm engine based on the actual transaction results and transaction data.

[0008] To address the aforementioned technical problems, this application also provides a computer device that employs the following technical solution: A computer device includes a memory and a processor, the memory storing computer-readable instructions, the processor executing the computer-readable instructions to implement the steps of the product recommendation method as described in any of the preceding claims.

[0009] To address the aforementioned technical problems, this application also provides a computer-readable storage medium, employing the technical solution described below: A computer-readable storage medium storing computer-readable instructions that, when executed by a processor, implement the steps of the product recommendation method as described in any one of the preceding descriptions.

[0010] Compared with the prior art, the embodiments of this application have the following main advantages: This application discloses a product recommendation method, apparatus, computer equipment, and storage medium, belonging to the field of artificial intelligence technology, and applied to intelligent recommendation of insurance telemarketing products. This application combines deep neural networks with genetic algorithms to achieve intelligent generation and continuous optimization of product recommendation strategies. By extracting multi-dimensional features from historical transaction data and constructing a unified feature representation, the deep neural network can make refined predictions of short-term and long-term transaction probabilities under different recommendation strategies, based on the fusion of customer characteristics, product characteristics, and recommendation strategy parameters. The genetic algorithm engine uses the prediction results as the core basis for fitness evaluation, searching in a high-dimensional, discrete, and continuous mixed recommendation strategy space, effectively reducing the complexity and limitations of manual rule design, and enabling recommendation strategies to better fit the behavioral characteristics and business goals of different customers. Simultaneously, by collecting transaction feedback data after actual recommendation execution and iteratively updating the deep neural network model and genetic algorithm engine, a closed-loop mechanism of model prediction and strategy optimization is formed, enabling the system to continuously adapt to changes in customer behavior and the market environment. This application improves the targeting and stability of recommendation strategies while ensuring the controllability and scalability of the recommendation process, enhancing the overall performance of product recommendation decisions in complex business scenarios. Attached Figure Description

[0011] To more clearly illustrate the solutions in this application, the accompanying drawings used in the description of the embodiments of this application will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 An exemplary system architecture diagram is shown, in which this application can be applied; Figure 2 A flowchart of one embodiment of the product recommendation method according to this application is shown; Figure 3 It shows Figure 2 A flowchart of an embodiment of step S203; Figure 4 A schematic diagram of one embodiment of the product recommendation device according to this application is shown; Figure 5 It shows Figure 4 A schematic diagram of the structure of an embodiment of the recommendation strategy module 403; Figure 6 A schematic diagram of the structure of one embodiment of a computer device according to this application is shown. Detailed Implementation

[0013] 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 pertains; the terminology used herein in the specification of the application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application; the terms "comprising" and "having," and any variations thereof, in the specification, claims, and foregoing drawings of this application, are intended to cover non-exclusive inclusion. The terms "first," "second," etc., in the specification, claims, or foregoing drawings of this application are used to distinguish different objects, not to describe a particular order.

[0014] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0015] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings.

[0016] like Figure 1 As shown, system architecture 100 may include terminal device 101, network 102, and server 103. Terminal device 101 may be a laptop 1011, tablet 1012, or mobile phone 1013. Network 102 is used as a medium to provide a communication link between terminal device 101 and server 103. Network 102 may include various connection types, such as wired, wireless communication links, or fiber optic cables.

[0017] Users can use terminal device 101 to interact with server 103 via network 102 to receive or send messages, etc. Various communication client applications can be installed on terminal device 101, such as web browser applications, shopping applications, search applications, instant messaging tools, email clients, social media platform software, etc.

[0018] Terminal device 101 can be various electronic devices with a display screen and support web browsing. In addition to laptops 1011, tablets 1012, or mobile phones 1013, terminal device 101 can also be an e-book reader, an MP3 player (Moving Picture Experts Group Audio Layer III), an MP4 player (Moving Picture Experts Group Audio Layer IV), a laptop computer, and a desktop computer, etc.

[0019] Server 103 can be a server that provides various services, such as a backend server that provides support for the pages displayed on terminal device 101.

[0020] It should be noted that the product recommendation method provided in this application embodiment is generally executed by a server / terminal device, and correspondingly, the product recommendation device is generally set in the server / terminal device.

[0021] It should be understood that Figure 1 The number of terminal devices, networks, and servers shown is merely illustrative; the system can have any number of terminal devices, networks, and servers depending on implementation needs.

[0022] Continue to refer to Figure 2 A flowchart illustrating an embodiment of a product recommendation method according to this application is shown. The product recommendation method includes the following steps: S201, retrieve historical transaction data from the preset database, and extract features from the historical transaction data to obtain a set of transaction data features; Specifically, the preset database may include multiple data sources such as a customer relationship management system database, a historical policy database, a product information database, an interaction record database, and a claims and surrender record database. The system first retrieves raw data related to the customer's historical transaction behavior from these databases according to preset data extraction rules. This raw data includes at least the customer's basic attribute information, the types and combinations of products purchased historically, transaction amount, transaction time, communication channels, sales scripts, follow-up frequency, and surrender or complaint information. Subsequently, the acquired raw data undergoes data cleaning, including outlier removal, missing value imputation, duplicate record merging, and timestamp alignment, to ensure data consistency and usability. After data cleaning, feature extraction and feature construction operations are performed on the historical transaction data, mapping the raw fields into feature forms that can be processed by the model. The feature extraction process may include encoding discrete features, normalizing or binning continuous numerical features, and constructing statistical features or time-series feature windows for time-series behavioral data. Furthermore, customer tags, behavioral frequency features, trend features, or clustering features can be generated based on historical transaction behavior. Finally, the multidimensional features obtained through the above processing are integrated to form a unified set of transaction data features.

[0023] Feature extraction supports the extraction of temporal and behavioral features (such as call, inquiry, click, and claim records over the past 1 / 3 / 6 / 12 months). Specifically, it includes constructing multi-dimensional temporal statistical features based on customer call duration, call frequency, product type and number of inquiries, online platform click behavior trajectory (such as dwell time on product detail pages and comparison of browsing history), historical claim application time, claim amount, and claim reason, etc., within different time windows. Examples include the month-on-month growth rate of call counts in the past month, the proportion of product categories with high-frequency inquiries in the past 3 months, the feature vectors of the top 3 products in terms of click volume in the past 6 months, and the time interval distribution characteristics of claim events in the past 12 months. At the same time, combined with the interactive text between customers and telemarketers, it uses natural language processing technology to extract emotional tendency features, demand keyword features (such as "critical illness protection", "children's education fund", "retirement planning", etc.), and objection handling response features, forming a subset of behavioral features that can comprehensively depict the dynamic behavior patterns of customers, and integrating this subset into the transaction data feature set.

[0024] S202 uses a pre-trained deep neural network to predict the short-term and long-term transaction probabilities based on a set of transaction data features, under a given recommendation strategy. Specifically, the system pre-builds and trains a deep neural network model for predicting transaction behavior. During the training phase, this deep neural network learns the correlation between customer characteristics, product characteristics, and recommendation strategy parameters based on historical samples. In the prediction phase, the system inputs the transaction data feature set obtained in step S201 into the deep neural network and integrates the currently evaluated recommendation strategy parameters as part of the input features. These recommendation strategy parameters may include information such as product combination selection, price or discount parameters, communication channels, script template identifiers, contact periods, and follow-up frequency. After receiving the aforementioned joint features through its input layer, the deep neural network performs joint representation learning on customer behavior features, product attribute features, and strategy features in a shared hidden layer, thereby extracting high-order correlation features. Based on this, the deep neural network models transaction behavior at different time scales through multiple output branches. One output branch predicts the probability of a customer making a transaction within a short interaction period, while another output branch predicts the probability of a customer making a transaction over a longer period or the long-term transaction trend.

[0025] Deep neural networks adopt the DNN (Deep Neural Network) model architecture, and their network structure can be a multi-task network (shared trunk + multi-head) or multiple single-task networks. The input layer contains static and temporal features, and the hidden layer adopts multi-layer fully connected, attention mechanism, or can use Transformer / sequence network to process temporal features. The output layer performs Sigmoid / Softmax on the probability output and regression on the numerical output.

[0026] S203 calls the preset genetic algorithm engine, using short-term and long-term transaction probabilities as the core basis of the fitness function, and searches for matching product recommendation strategies in the recommendation strategy space. Specifically, the system predefines a strategy space for recommendation decisions and searches and optimizes this strategy space using a genetic algorithm engine. The strategy space includes multiple configurable recommendation decision dimensions, such as the combination of main and supplementary insurance, price or discount ranges, communication script templates, customer outreach channels, priority contact periods, and follow-up execution plans. The genetic algorithm engine first encodes the above recommendation strategies, mapping each candidate recommendation strategy to a corresponding strategy chromosome structure, and initializes an initial strategy population composed of multiple strategy chromosomes. During the fitness evaluation phase, for each strategy chromosome in the strategy population, the system calls the deep neural network prediction module described in step S202 to obtain the short-term and long-term transaction probabilities predicted for that strategy under the current transaction data characteristics. Subsequently, the genetic algorithm engine uses the short-term and long-term transaction probabilities as the core basis for calculating the fitness function, and combines this with strategy execution costs, risk constraints, or business rules to evaluate the fitness of each strategy chromosome. After completing the fitness assessment, the genetic algorithm engine performs selection, crossover, and mutation operations on the strategy population according to the preset genetic operation rules, thereby continuously generating new candidate strategies and gradually searching out product recommendation strategies that match the current customer characteristics and business goals in multiple iterations.

[0027] The genetic algorithm engine maps the decision-making dimensions of the recommendation strategy (such as product combination code, discount rate threshold, script template ID, channel priority sequence, contact time period code, etc.) to a fixed-length chromosome structure based on preset strategy encoding rules. Discrete parameters are encoded using integer or binary codes, while continuous parameters are encoded using floating-point or Gray code codes. The population initialization module generates an initial strategy population based on business constraints (such as the scope of compliant products, the maximum number of recommended products, channel availability limits, etc.). It supports both random and heuristic initialization modes. Heuristic initialization can generate initial chromosomes based on historical high-quality strategy fragments or expert rules to improve the average fitness of the initial population. The selection operator uses a combination of tournament selection and elite retention strategies. It selects strategy chromosomes with high fitness from the current population as parents. Tournament selection randomly selects a preset number of chromosomes and chooses the one with the best fitness, while elite retention ensures that the top chromosomes in each generation are selected. The N optimal chromosomes directly enter the next generation; the crossover operator supports single-point crossover, multi-point crossover, and uniform crossover, and adaptively selects the crossover method for chromosomes with different encoding types. For example, multi-point crossover is performed on product combination encoding to retain effective combination fragments, and simulated binary crossover (SBX) is performed on continuous parameter encoding to maintain numerical continuity; the mutation operator adopts an adaptive mutation probability mechanism, dynamically adjusting the mutation probability according to the number of generations of population evolution and chromosome fitness values. Random reset mutation is performed on discrete parameters, and Gaussian mutation or polynomial mutation is performed on continuous parameters. At the same time, mutation boundary checks are set to ensure that the mutated strategy parameters meet business constraints; the iteration termination module is used to monitor the evolution process of the genetic algorithm. When the preset number of iterations is reached, the population fitness value converges (the rate of change of the optimal fitness for multiple consecutive generations is lower than the threshold), or the optimal strategy that meets the target fitness threshold is found, the iteration is terminated and the product recommendation strategy corresponding to the current optimal strategy chromosome is output.

[0028] S204, Determine the appropriate product recommendation combination based on the product recommendation strategy, and send the product recommendation combination to the client terminal; Specifically, after completing the strategy search by the genetic algorithm engine, the system selects product recommendation strategies from the current strategy population that meet preset fitness values. The selected strategies are then decoded, restoring the strategy chromosomes into executable recommendation decision parameters. These parameters include specific product combinations, corresponding price or discount schemes, recommendation order, communication scripts, and outreach timing. Based on the decoding results, the system generates structured product recommendation combination information and performs compliance and rule checks on these combinations to ensure they meet business constraints, compliance requirements, and customer outreach rules. After successful verification, the system encapsulates the product recommendation combination into a standardized recommendation instruction or message format and sends the recommendation results to customer terminals or business execution terminals, such as telemarketing systems, online customer service systems, or mobile clients, through a preset interface. The sending process can incorporate current context information to determine the specific method and timing of recommendation display, ensuring the recommendation strategy is correctly executed and displayed.

[0029] S205: Obtain the actual transaction results and transaction data after the product recommendation combination is displayed, and iteratively update the deep neural network and genetic algorithm engine based on the actual transaction results and transaction data.

[0030] Specifically, after a product recommendation package is sent and displayed to the customer, the system continuously collects execution feedback data corresponding to that package. This feedback data includes at least whether the customer accepted the recommendation, whether a transaction occurred, the transaction amount, the transaction time, and behavioral information such as policy cancellation, complaints, or ongoing interactions. The system associates and stores this feedback data with the corresponding recommendation strategy identifier, customer identifier, and contextual information, forming a traceable feedback data record. Subsequently, the collected feedback data undergoes preprocessing and feature reconstruction operations, transforming it into training or evaluation samples that can be used for model updates. Based on the updated sample data, the system performs parameter updates or retraining operations on the deep neural network model, enabling the model to continuously learn the latest customer behavior patterns and strategy response results. Simultaneously, the system can also adjust the fitness evaluation process, strategy space range, or initialization strategy in the genetic algorithm engine based on the updated deep neural network prediction results, and rerun the strategy search process when necessary. Through this method, continuous iterative updates of the recommendation model and strategy search module are achieved.

[0031] Furthermore, the steps of using a pre-trained deep neural network to predict the short-term and long-term transaction probabilities based on a set of transaction data features specifically include: The transaction data feature set is aligned and vectorized to construct model input features that include customer feature vectors, product feature vectors, and context feature vectors; The recommendation strategy parameters under a given recommendation strategy are encoded and fused with the model input features as a strategy feature vector to form a joint input feature vector. The joint input feature vector is fed into a shared feature extraction layer of a deep neural network to extract high-order correlation features between customers, products, and recommendation strategies. Based on higher-order correlation features, predictions are made through the output branches of deep neural networks to obtain short-term and long-term transaction probability predictions under the corresponding recommendation strategies. The short-term and long-term transaction probability prediction results are normalized, and the normalized prediction results are output.

[0032] In this embodiment, the deep neural network is used to jointly model and predict customer transaction behavior under different recommendation strategy conditions. Specifically, the system first performs feature alignment processing on the transaction data feature set to unify the structural form of features from different data sources and at different time granularities, and maps customer basic attributes, historical behavioral features, product attribute information, and interaction context information into standardized numerical vector representations through vectorization. Based on this, the strategy parameters corresponding to a given recommendation strategy are encoded. These strategy parameters may include product combination configuration, price or discount information, communication script identifiers, contact channels, and contact time periods, and are constructed into a strategy feature vector. Subsequently, the system concatenates or fuses the strategy feature vector with the customer feature vector, product feature vector, and context feature vector to form a unified joint input feature vector. The joint input feature vector is input to the shared feature extraction layer of the deep neural network, and the interaction relationships between different features are modeled through multi-layer nonlinear transformations, thereby extracting high-order feature representations reflecting the correlation between customers, products, and recommendation strategies. Based on the aforementioned higher-order correlation features, the deep neural network performs calculations through output branches set for different prediction objectives. One output branch generates a probability prediction result for short-term transaction behavior, while the other output branch generates a probability prediction result for long-term transaction behavior or long-term trends. Finally, the short-term and long-term transaction probabilities obtained from the above predictions are normalized or calibrated to ensure consistency in numerical scale and probability distribution among different prediction results. The processed prediction results are then used as input data for the recommendation strategy evaluation and optimization process.

[0033] Through the above steps, deep neural networks can accurately predict transaction behavior at different time scales under unified feature representation and policy constraints.

[0034] Furthermore, based on higher-order correlation features, the steps of predicting short-term and long-term transaction probabilities under the corresponding recommendation strategy through the output branches of the deep neural network include: The first output branch of the deep neural network performs a first nonlinear mapping on the high-order correlation features to generate an intermediate feature representation that characterizes the customer's immediate response behavior under the recommendation strategy. Based on the intermediate feature representation, the probability output layer of the first output branch outputs the short-term transaction probability prediction result under the corresponding recommendation strategy. The second output branch of the deep neural network performs a second nonlinear mapping on the higher-order correlation features to generate a long-term feature representation that characterizes the long-term behavior and value trends of customers. Based on long-term feature representation, the probability output layer of the second output branch outputs the long-term transaction probability prediction result under the corresponding recommendation strategy. Consistency verification is performed between the short-term and long-term transaction probability prediction results.

[0035] In this embodiment, the deep neural network employs a multi-output branch structure to model the transaction behavior of the same high-order correlation feature at different time scales. Specifically, the system simultaneously inputs the high-order correlation feature output by the shared feature extraction layer into the first output branch and the second output branch. The first output branch primarily models the customer's immediate response behavior under the current recommendation strategy. This first output branch performs a first nonlinear mapping operation on the high-order correlation feature through a multi-layer fully connected structure or other nonlinear transformation units, thereby generating an intermediate feature representation that reflects the customer's short-term decision-making tendency. Based on this, a short-term transaction probability prediction result is calculated through a probability output layer. Simultaneously, the second output branch uses the same high-order correlation feature as input, focusing on characterizing the customer's behavioral changes and value trends over a longer time span. This output branch performs a second nonlinear mapping process on the high-order correlation feature through nonlinear mapping structures with different parameter configurations, generating a long-term feature representation that characterizes the customer's long-term behavioral characteristics. The corresponding probability output layer then outputs a long-term transaction probability prediction result. After obtaining the short-term and long-term transaction probability prediction results, the system performs consistency verification on the two types of prediction results to detect the rationality of the short-term and long-term prediction results in terms of numerical range, probability distribution or business rules, so as to avoid obvious conflicts or anomalies between the prediction results.

[0036] Through the above steps, a model that distinguishes between short-term and long-term transaction behavior is achieved based on a unified feature representation, enabling the prediction results to simultaneously reflect immediate responses and long-term trends.

[0037] Further, please refer to Figure 3 The process involves calling a pre-defined genetic algorithm engine, using short-term and long-term transaction probabilities as the core criteria for the fitness function, and searching for matching product recommendation strategies in the recommendation strategy space. This process specifically includes: S301, encode the configurable decision items in the recommendation strategy space, construct a strategy chromosome containing product combination parameters, price or discount parameters, communication method parameters and execution order parameters, and initialize the strategy population of the genetic algorithm; S302, for each strategy chromosome in the strategy population, call the short-term transaction probability and long-term transaction probability to calculate the fitness value of each strategy chromosome; S303, perform selection, crossover or mutation operations on the strategy population based on fitness values ​​to generate and save a new strategy population, and identify the dominant strategy chromosomes that meet the preset constraints during the generation process; S304, when the preset iteration termination condition is met, determine the strategy chromosome with the optimal fitness value from the retained strategy population; S305 decodes the strategy chromosome with the best fitness value into a matching product recommendation strategy.

[0038] In this embodiment, a genetic algorithm engine is used to search and optimize product recommendation strategies within a predefined recommendation strategy space. Specifically, the system first structurally decomposes various configurable decision items within the recommendation strategy space and maps them to different gene positions of strategy chromosomes according to preset encoding rules. These gene positions include at least product combination-related parameters, price or discount parameters, communication method parameters, and recommendation execution order parameters. After encoding, the system generates an initial strategy population consisting of multiple strategy chromosomes based on random initialization or rule initialization. Subsequently, for each strategy chromosome in the strategy population, the system, in conjunction with the parameter configuration corresponding to the strategy, calls the aforementioned deep neural network prediction module to obtain the short-term and long-term transaction probabilities of the strategy under the current transaction data characteristics. The probability prediction results are used as the core input for calculating the fitness function to evaluate the fitness of each strategy chromosome. Based on this, the genetic algorithm engine performs selection operations on the strategy population according to fitness values, recombines some gene positions of different strategy chromosomes through crossover operations, and randomly perturbs some gene positions through mutation operations to generate a new strategy population. During the generation of the next generation of strategy population, the system also performs preset constraint checks on each strategy chromosome to identify and retain advantageous strategy chromosomes that meet business rules or compliance requirements. The above genetic iteration process ends when preset termination conditions such as the number of iterations, fitness convergence conditions, or time limits are met. The system then selects the strategy chromosome with the best fitness value from the currently retained strategy population and decodes it into an executable product recommendation strategy.

[0039] Define the chromosome structure for the genetic algorithm, encoding configurable decision terms into gene bits. Example gene table: 1) Primary risk selection bit (discrete / binary); 2) Additional risk level set (multi-bit binary); 3) Discount percentage (continuous / real number); 4) Script template ID (discrete); 5) Priority contact time period (discrete / enumerated); 6) Number of follow-ups (integer); 7) Recommendation priority weight (continuous); It supports mixed encoding (real numbers, integers, and binary) and performs corresponding mutation / crossover operations in the genetic algorithm engine. For example, it performs single-point or multi-point crossover on the binary encoded main insurance selection bits and supplementary insurance bit sets to retain effective product portfolio gene fragments; it performs simulated binary crossover (SBX) on continuous discount percentages and recommendation priority weights, simulating the crossover process of biological genetic material in nature, so that the crossover values ​​maintain a certain degree of randomness while inheriting the excellent characteristics of the parents; and it performs uniform or sequential crossover on discrete script template IDs and priority contact time periods to explore the combined effects of different communication strategies and contact timings.

[0040] Through the above steps, recommended strategies that match customer characteristics and business objectives can be automatically searched and filtered in a complex strategic decision space, thereby improving the systematization of the strategy selection process.

[0041] Furthermore, the steps for calculating the fitness value of each strategy chromosome in the strategy population by calling the short-term transaction probability and the long-term transaction probability specifically include: For each strategy chromosome, the expected return value of the recommended strategy in the short term is calculated based on the short-term transaction probability and expected transaction amount parameters, and the first return value is obtained. Based on the long-term transaction probability and the preset long-term value assessment model, the expected return value of the recommendation strategy in the long-term dimension is calculated to obtain the second return value, and the potential risk value of the long-term transaction probability is assessed. The fitness value of the strategy chromosome is calculated by weighting and fusing the first benefit value, the second benefit value, and the potential risk value, and combining them with the execution cost or constraint penalty term corresponding to the strategy chromosome.

[0042] In this embodiment, the system quantitatively evaluates the recommended strategy for each strategy chromosome in the strategy population according to a unified fitness evaluation process. Specifically, the system first combines the recommended strategy parameters corresponding to the strategy chromosome to obtain the short-term transaction probability prediction result output by a deep neural network, and introduces the expected transaction amount parameter associated with the strategy to estimate the revenue of possible transactions in the short term, thereby calculating a first revenue value reflecting the immediate transaction value. Subsequently, based on the long-term transaction probability prediction result corresponding to the same strategy chromosome, the system introduces a preset long-term value evaluation model to estimate the value of possible transactions by customers over a longer period, calculating a second revenue value. Simultaneously, the system can also combine the magnitude of the long-term transaction probability, historical surrender or complaint data, and strategy characteristics to quantitatively evaluate the potential risks of the strategy in the long term, obtaining the corresponding potential risk value. After obtaining the first benefit value, the second benefit value, and the potential risk value, the system performs weighted fusion processing on the above indicators according to preset weight parameters, and further combines the execution cost required by the strategy chromosome in the actual execution process and the penalty item corresponding to the constraint verification result to modify the fusion result, thereby forming a fitness value used to characterize the overall quality of the strategy.

[0043] Through the above steps, a comprehensive evaluation of the recommendation strategy is achieved across multiple dimensions, including short-term returns, long-term value, and potential risks. This makes the fitness calculation more comprehensive and helps to select more reasonable recommendation strategies.

[0044] Furthermore, the step of weightedly combining the first return value, the second return value, and the potential risk value, and combining this with the execution cost or constraint penalty term corresponding to the strategy chromosome, to calculate the fitness value of the strategy chromosome, specifically includes: The first return value, the second return value, and the potential risk value are normalized. Based on preset weight parameters, the first return value, the second return value, and the potential risk value after normalization are weighted and calculated to obtain the comprehensive return evaluation value of the strategy chromosome. Obtain the execution cost or constraint penalty item corresponding to the strategy chromosome, and calculate the corresponding cost deduction item or constraint penalty item based on the execution cost or constraint penalty item; The overall return assessment value is adjusted based on cost deductions or constraint penalties to obtain the fitness value of the strategy chromosome.

[0045] In this embodiment, the system performs unified processing on different benefit and risk indicators to achieve a comprehensive calculation of the fitness value of the strategy chromosome. Specifically, for each strategy chromosome, the system first obtains its corresponding first benefit value, second benefit value, and potential risk value, and performs normalization processing on these values ​​to eliminate differences in the dimensions, value ranges, and numerical scales of different indicators, thereby ensuring the comparability of the weighted calculation process. After completing the normalization processing, the system performs a weighted calculation on the normalized first benefit value, second benefit value, and potential risk value based on preset weight parameters, where the weight parameters are used to reflect the relative importance of different indicators in the overall evaluation. Through the above weighted calculation, an evaluation value representing the overall benefit of the strategy is obtained. At the same time, the system also obtains the execution cost information required in the actual execution process according to the specific recommended strategy parameters corresponding to the strategy chromosome. The execution cost may include price concession costs, channel usage costs, or resource occupation costs, etc.; and further verifies whether the strategy meets preset business rules or compliance constraints, generating corresponding constraint penalty items for strategies that do not meet the constraints. Subsequently, based on the acquired execution costs and constraint penalties, the system modifies the aforementioned comprehensive benefit assessment value. By deducting cost items or imposing penalty items, the final assessment result of the strategy chromosome is adjusted, thereby calculating the fitness value used in the genetic algorithm evolution process. The fitness value reflects the overall performance of the strategy chromosome under the combined effects of multiple factors such as benefits, risks, and costs.

[0046] Through the above steps, a unified quantification and correction of policy fitness is achieved, enabling the genetic algorithm to take into account both benefits and constraints during the policy search process, thereby improving the rationality of policy evaluation.

[0047] Furthermore, the steps of obtaining actual transaction results and data after the product recommendation combination is displayed, and iteratively updating the deep neural network and genetic algorithm engine based on the actual transaction results and data, specifically include: Collect the display results of recommended product combinations on the client's terminal, and record the customer response behavior data and transaction result data corresponding to the recommended combinations; Clean, label, and reconstruct features of customer response behavior data and transaction result data to generate a feedback sample dataset for model updates; Based on the feedback sample dataset, periodic retraining operations are performed on the deep neural network to update the predictive capabilities of short-term and long-term transaction probabilities. Based on the updated predictions from the deep neural network, the policy search space in the genetic algorithm engine is fine-tuned, and the policy search is re-executed.

[0048] In this embodiment, the system constructs a complete feedback collection and model update process to achieve continuous iteration of the product recommendation model and strategy search mechanism. Specifically, after a product recommendation combination is sent and displayed to the customer's terminal, the system monitors the recommendation process in real-time or near real-time, collects the display result information corresponding to the recommendation combination, and simultaneously records the customer's response behavior data and final transaction result data after the display. The customer response behavior data may include information such as whether the user viewed the product, clicked on it, answered the call, and the duration of the interaction. The transaction result data may include whether a transaction was completed, the transaction amount, the transaction time, and related behavior records. Subsequently, the system performs unified data processing operations on the collected data, including cleaning abnormal data, labeling key behavioral results, and reconstructing the data features according to the current model input requirements, thereby generating a structured feedback sample dataset. The feedback sample dataset is used in the model update phase. Based on this dataset, the system performs periodic retraining or parameter fine-tuning operations on the deep neural network, enabling the model to continuously learn the latest customer behavior patterns and recommendation strategy response, thereby updating the predictive capabilities of short-term and long-term transaction probabilities. After completing the deep neural network model update, the system further fine-tunes the policy search space, initialization policy, or fitness evaluation parameters in the genetic algorithm engine based on the updated prediction results, and re-executes the policy search process when necessary, so that the policy generation process can be consistent with the latest prediction model, thereby forming a dynamic linkage update mechanism between the recommended policy and the prediction model.

[0049] Through the above steps, a closed-loop iteration of the recommendation model and strategy search is achieved, enabling the system to continuously absorb real feedback data and dynamically adjust the recommendation strategy, thereby improving the overall adaptability of the recommendation process.

[0050] 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.

[0051] It should be emphasized that, in order to further ensure the privacy and security of the aforementioned historical transaction data, the aforementioned historical transaction data can also be stored in a blockchain node.

[0052] 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.

[0053] The embodiments of this application can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence (AI) refers to the theories, methods, technologies, and application systems that use 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.

[0054] 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.

[0055] 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 through computer-readable instructions. These computer-readable instructions can be stored in a computer-readable storage medium. When the program is executed, it 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).

[0056] 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.

[0057] Further reference Figure 4 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.

[0058] like Figure 4 As shown, the product recommendation device 400 described in this embodiment includes: The feature extraction module 401 is used to obtain historical transaction data from a preset database and extract features from the historical transaction data to obtain a set of transaction data features. The transaction prediction module 402 is used to predict the short-term and long-term transaction probabilities under a given recommendation strategy by using a pre-trained deep neural network based on a set of transaction data features. The recommendation strategy module 403 is used to call the preset genetic algorithm engine, using short-term transaction probability and long-term transaction probability as the core basis of the fitness function, to search for matching product recommendation strategies in the recommendation strategy space; Product portfolio module 404 is used to determine the appropriate product recommendation portfolio according to the product recommendation strategy and send the product recommendation portfolio to the client terminal. The iterative update module 405 is used to obtain the actual transaction results and transaction data after the product recommendation combination is displayed, and to iteratively update the deep neural network and genetic algorithm engine based on the actual transaction results and transaction data.

[0059] Furthermore, the transaction prediction module 402 specifically includes: The feature alignment submodule is used to perform feature alignment and vectorization on the transaction data feature set, and to construct model input features that include customer feature vectors, product feature vectors and context feature vectors. The feature fusion submodule is used to encode the recommendation strategy parameters under a given recommendation strategy and fuse them with the model input features as a strategy feature vector to form a joint input feature vector. The higher-order correlation submodule is used to input the joint input feature vector into the shared feature extraction layer of the deep neural network to extract higher-order correlation features between customers, products and recommendation strategies; The branch prediction submodule is used to make predictions based on high-order correlation features, respectively through the output branches of the deep neural network, to obtain the short-term and long-term transaction probability prediction results under the corresponding recommendation strategy. The normalization processing submodule is used to normalize the short-term and long-term transaction probability prediction results and output the normalized prediction results.

[0060] Furthermore, the branch prediction submodule specifically includes: The first nonlinear mapping unit is used to perform a first nonlinear mapping on the high-order correlation features based on the first output branch of the deep neural network, and generate an intermediate feature representation to characterize the customer's immediate response behavior under the recommendation strategy. The first probability output unit is used to output the short-term transaction probability prediction result under the corresponding recommendation strategy based on the intermediate feature representation and through the probability output layer of the first output branch. The second nonlinear mapping unit is used to perform a second nonlinear mapping on the high-order correlation features based on the second output branch of the deep neural network, generating a long-term feature representation that characterizes the long-term behavior and value trends of customers. The second probability output unit is used to output the long-term transaction probability prediction result under the corresponding recommendation strategy through the probability output layer of the second output branch, based on the long-term feature representation. The consistency verification unit is used to perform consistency verification between the short-term transaction probability prediction results and the long-term transaction probability prediction results.

[0061] Further, please refer to Figure 5 The recommendation strategy module 403 specifically includes: The encoding processing submodule 501 is used to encode configurable decision items in the recommendation strategy space, construct a strategy chromosome containing product combination parameters, price or discount parameters, communication method parameters, and execution order parameters, and initialize the strategy population of the genetic algorithm. The fitness value submodule 502 is used to calculate the fitness value of each strategy chromosome in the strategy population by calling the short-term transaction probability and the long-term transaction probability. The population generation submodule 503 is used to perform selection, crossover or mutation operations on the strategy population based on the fitness value to generate and save a new strategy population, and to identify the dominant strategy chromosomes that meet the preset constraints during the generation process. The optimal chromosome submodule 504 is used to determine the strategy chromosome with the optimal fitness value from the retained strategy population when the preset iteration termination condition is met. The product recommendation submodule 505 is used to decode the strategy chromosome with the best fitness value into a matching product recommendation strategy.

[0062] Furthermore, the fitness value submodule specifically includes: The short-term prediction unit is used to calculate the expected return value of the recommended strategy in the short term for each strategy chromosome based on the short-term transaction probability and expected transaction amount parameters, and obtain the first return value. The long-term prediction unit is used to calculate the expected return value of the recommended strategy in the long-term dimension based on the long-term transaction probability and the preset long-term value assessment model, obtain the second return value, and assess the potential risk value of the long-term transaction probability. The fitness calculation unit is used to weight and fuse the first benefit value, the second benefit value, and the potential risk value, and combine them with the execution cost or constraint penalty term corresponding to the strategy chromosome to calculate the fitness value of the strategy chromosome.

[0063] Furthermore, the fitness calculation unit specifically includes: The normalization processing subunit is used to normalize the first profit value, the second profit value, and the potential risk value. The comprehensive return subunit is used to perform weighted calculations on the normalized first return value, second return value, and potential risk value based on preset weight parameters to obtain the comprehensive return evaluation value of the strategy chromosome. The cost or constraint item sub-unit is used to obtain the execution cost or constraint penalty item corresponding to the strategy chromosome, and calculate the corresponding cost deduction item or constraint penalty item based on the execution cost or constraint penalty item; The evaluation correction sub-unit is used to correct the comprehensive benefit evaluation value based on cost deductions or constraint penalties to obtain the fitness value of the strategy chromosome.

[0064] Furthermore, the iterative update module 405 specifically includes: The operation data acquisition submodule is used to collect the display results of product recommendation combinations on the customer terminal, and record customer response behavior data and transaction result data corresponding to the recommendation combinations; The sample data construction submodule is used to clean, label, and reconstruct features of customer response behavior data and transaction result data to generate a feedback sample dataset for model updates. The retraining operation submodule is used to perform periodic retraining operations on the deep neural network based on the feedback sample dataset to update the predictive capabilities of short-term and long-term transaction probabilities. The search space adjustment submodule is used to fine-tune the policy search space in the genetic algorithm engine based on the updated prediction results of the deep neural network, and then re-execute the policy search.

[0065] To address the aforementioned technical problems, embodiments of this application also provide a computer device. Please refer to [link / reference needed]. Figure 6 , Figure 6 This is a basic structural block diagram of the computer device in this embodiment.

[0066] The computer device 6 includes a memory 61, a processor 62, and a network interface 63 that are interconnected via a system bus. It should be noted that only the computer device 6 with memory 61, processor 62, and network interface 63 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.

[0067] 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.

[0068] The memory 61 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 61 may be an internal storage unit of the computer device 6, such as the hard disk or memory of the computer device 6. In other embodiments, the memory 61 may also be an external storage device of the computer device 6, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the computer device 6. Of course, the memory 61 may also include both the internal storage unit and its external storage device of the computer device 6. In this embodiment, the memory 61 is typically used to store the operating system and various application software installed on the computer device 6, such as computer-readable instructions for product recommendation methods. In addition, the memory 61 can also be used to temporarily store various types of data that have been output or will be output.

[0069] In some embodiments, the processor 62 may be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chip. The processor 62 is typically used to control the overall operation of the computer device 6. In this embodiment, the processor 62 is used to execute computer-readable instructions stored in the memory 61 or to process data, such as executing computer-readable instructions for the product recommendation method.

[0070] The network interface 63 may include a wireless network interface or a wired network interface, which is typically used to establish communication connections between the computer device 6 and other electronic devices.

[0071] This application also provides an embodiment, namely, a computer device including 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 above.

[0072] This application also provides another embodiment, namely, providing 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 as described above.

[0073] 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.

[0074] 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.

[0075] 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.

[0076] 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: Historical transaction data is obtained from a preset database, and features are extracted from the historical transaction data to obtain a set of transaction data features; Using a pre-trained deep neural network based on the transaction data feature set, the short-term and long-term transaction probabilities are predicted under a given recommendation strategy. The preset genetic algorithm engine is invoked, and the short-term transaction probability and the long-term transaction probability are used as the core basis of the fitness function to search for a matching product recommendation strategy in the recommendation strategy space. Based on the product recommendation strategy, a suitable product recommendation combination is determined, and the product recommendation combination is sent to the client terminal; Obtain the actual transaction results and transaction data after the product recommendation combination is displayed, and iteratively update the deep neural network and the genetic algorithm engine based on the actual transaction results and transaction data.

2. The product recommendation method as described in claim 1, characterized in that, The step of using a pre-trained deep neural network to predict the short-term and long-term transaction probabilities based on the transaction data feature set specifically includes: The transaction data feature set is subjected to feature alignment and vectorization processing to construct model input features that include customer feature vectors, product feature vectors and context feature vectors; The recommendation strategy parameters under the given recommendation strategy are encoded and fused with the model input features as a strategy feature vector to form a joint input feature vector; The joint input feature vector is input into the shared feature extraction layer of the deep neural network to extract high-order correlation features between customers, products and recommendation strategies; Based on the higher-order correlation features, predictions are made through the output branches of the deep neural network to obtain the short-term and long-term transaction probability prediction results under the corresponding recommendation strategy. The short-term and long-term transaction probability prediction results are normalized, and the normalized prediction results are output.

3. The product recommendation method as described in claim 2, characterized in that, The step of predicting short-term and long-term transaction probability predictions based on the higher-order correlation features and through the output branches of the deep neural network to obtain the corresponding recommendation strategy's short-term and long-term transaction probability prediction results specifically includes: The higher-order correlation features are mapped using the first output branch of the deep neural network to generate an intermediate feature representation that characterizes the customer’s immediate response behavior under the recommendation strategy. Based on the intermediate feature representation, the probability output layer of the first output branch outputs the short-term transaction probability prediction result corresponding to the recommendation strategy. The second output branch of the deep neural network is used to perform a second nonlinear mapping on the higher-order correlation features to generate a long-term feature representation that characterizes the long-term behavior and value trend of customers. Based on the long-term feature representation, the probability output layer of the second output branch outputs the long-term transaction probability prediction result corresponding to the recommendation strategy. The short-term transaction probability prediction result and the long-term transaction probability prediction result are subjected to consistency verification processing.

4. The product recommendation method as described in claim 1, characterized in that, The step of calling a preset genetic algorithm engine, using the short-term and long-term transaction probabilities as the core criteria for the fitness function, to search for a matching product recommendation strategy in the recommendation strategy space specifically includes: The configurable decision items in the recommended strategy space are encoded to construct a strategy chromosome containing product combination parameters, price or discount parameters, communication method parameters, and execution order parameters, and the strategy population of the genetic algorithm is initialized. For each strategy chromosome in the strategy population, the fitness value of each strategy chromosome is calculated by calling the short-term transaction probability and the long-term transaction probability; Based on the fitness value, the strategy population is subjected to selection, crossover, or mutation operations to generate and save a new strategy population, and during the generation process, dominant strategy chromosomes that meet preset constraints are identified. When the preset iteration termination condition is met, the strategy chromosome with the optimal fitness value is determined from the retained strategy population; The strategy chromosome with the optimal fitness value is decoded into the matching product recommendation strategy.

5. The product recommendation method as described in claim 4, characterized in that, The step of calculating the fitness value of each strategy chromosome in the strategy population by calling the short-term transaction probability and the long-term transaction probability specifically includes: For each of the strategy chromosomes, the expected return value of the recommended strategy in the short term is calculated based on the short-term transaction probability and expected transaction amount parameters to obtain the first return value; Based on the long-term transaction probability and the preset long-term value assessment model, the expected return value of the recommendation strategy in the long-term dimension is calculated to obtain the second return value, and the potential risk value of the long-term transaction probability is assessed. The fitness value of the strategy chromosome is calculated by weighting and fusing the first profit value, the second profit value, and the potential risk value, and combining them with the execution cost or constraint penalty term corresponding to the strategy chromosome.

6. The product recommendation method as described in claim 5, characterized in that, The step of weightedly fusing the first profit value, the second profit value, and the potential risk value, and combining them with the execution cost or constraint penalty term corresponding to the strategy chromosome to calculate the fitness value of the strategy chromosome, specifically includes: The first profit value, the second profit value, and the potential risk value are normalized. Based on preset weight parameters, the first return value, the second return value, and the potential risk value after normalization are weighted and calculated to obtain the comprehensive return evaluation value of the strategy chromosome. Obtain the execution cost or constraint penalty item corresponding to the strategy chromosome, and calculate the corresponding cost deduction item or constraint penalty item based on the execution cost or constraint penalty item; The overall benefit assessment value is adjusted based on the cost deduction item or the constraint penalty item to obtain the fitness value of the strategy chromosome.

7. The product recommendation method as described in claim 1, characterized in that, The step of obtaining the actual transaction results and transaction data after the product recommendation combination is displayed, and iteratively updating the deep neural network and the genetic algorithm engine based on the actual transaction results and transaction data, specifically includes: Collect the display results of the recommended product combinations on the customer's terminal, and record the customer response behavior data and transaction result data corresponding to the recommended combinations; The customer response behavior data and the transaction result data are cleaned, labeled, and reconstructed to generate a feedback sample dataset for model updates; Based on the feedback sample dataset, the deep neural network is periodically retrained to update its predictive capabilities for short-term and long-term transaction probabilities. Based on the updated prediction results of the deep neural network, the policy search space in the genetic algorithm engine is fine-tuned, and the policy search is re-executed.

8. A product recommendation device, characterized in that, include: The feature extraction module is used to obtain historical transaction data from a preset database and extract features from the historical transaction data to obtain a transaction data feature set. The transaction prediction module is used to predict the short-term and long-term transaction probabilities under a given recommendation strategy based on the transaction data feature set using a pre-trained deep neural network. The recommendation strategy module is used to call a preset genetic algorithm engine, using the short-term transaction probability and the long-term transaction probability as the core basis of the fitness function, to search for a matching product recommendation strategy in the recommendation strategy space; The product portfolio module is used to determine a suitable product recommendation portfolio based on the product recommendation strategy and send the product recommendation portfolio to the client terminal. The iterative update module is used to obtain the actual transaction results and transaction data after the product recommendation combination is displayed, and to iteratively update the deep neural network and the genetic algorithm engine based on the actual transaction results and transaction 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