Financial product recommendation method and system based on artificial intelligence

Abstract

The invention relates to the technical field of financial planning, in particular to a financial product recommendation method and system based on artificial intelligence, and the system comprises a task decomposition module which carries out the splitting of a recommendation task according to the dimensions of a risk level, an investment period and an expected return rate based on the financial product recommendation task, calculating parameter combinations of risk levels, investment deadlines and expected return rate dimensions, processing interaction relationships among the sub-tasks, judging influences of interaction on the combinations, determining combination priorities, and obtaining a multi-level sub-task structure. According to the method and the device, the financial product recommendation task is refined into a plurality of dimensions, and the parameter combination of the dimensions and the interaction relationship thereof are calculated, so that the priority of the recommendation scheme can be more accurately determined, the mutual influence among various parameters can be more flexibly and meticulously analyzed in the face of a complex financial market, and the recommendation efficiency is improved. Therefore, more targeted financial product recommendation is provided.

Description

Technical Field

[0001] The present invention relates to the technical field of financial planning, and in particular to a financial product recommendation method and system based on artificial intelligence. Background Art

[0002] The technical field of financial planning mainly involves providing guidance and services in financial management, investment strategies, risk management, etc. to individuals or institutions. This field covers financial advisors, wealth management, asset allocation, financial assessment, etc., aiming to help customers achieve financial goals, optimize investment portfolios, and balance risks and returns. In recent years, with the development of big data and artificial intelligence technology, financial planning has gradually developed from traditional manual operations to intelligent and automated directions, and can provide users with more accurate and personalized financial advice and product recommendations.

[0003] The financial product recommendation method and system aims to use artificial intelligence technology to analyze the user's financial status, risk preference and investment goals, and automatically recommend the most suitable financial products for the user. The system extracts valuable information from a large amount of financial data through machine learning, natural language processing and other technologies, helping users make more informed decisions in complex financial markets, improve investment results and reduce potential risks.

[0004] However, existing technologies usually rely on simple parameter analysis and preliminary risk assessment when providing financial product recommendations, lacking in-depth interpretation and combination of multi-dimensional data. It is easy to ignore the complex relationship between user personalized needs and market dynamics, resulting in the recommended products not necessarily matching the user's actual needs to the maximum extent. And when the market fluctuates greatly, it is often impossible to adjust the recommended plan in time, which may cause delays or errors in users' investment decisions. Summary of the invention

[0005] The purpose of the present invention is to solve the shortcomings of the prior art and to propose a financial product recommendation method and system based on artificial intelligence.

[0006] In order to achieve the above-mentioned purpose, the present invention adopts the following technical scheme: The financial product recommendation system based on artificial intelligence includes:

[0007] The task decomposition module, based on the financial product recommendation task, splits the recommendation task according to the dimensions of risk level, investment period and expected rate of return, calculates the parameter combination of the risk level, investment period and expected rate of return, processes the interaction between subtasks, judges the impact of the interaction on the combination, determines the combination priority, and obtains a multi-level subtask structure;

[0008] The data processing module extracts the parameter set of the subtasks based on the multi-level subtask structure, classifies the interaction items of the tasks, compares the matching degree between the sets, selects the matching items, calculates the correlation between the interaction items, processes the correlation data of the sets, and obtains the independent subtask solution set;

[0009] The result integration module, based on the independent subtask solution set, weights the subtask solutions, aggregates the solution parameters, classifies the parameter results, integrates the priorities of the sub-results, establishes the relationship between the levels, and obtains the integrated recommendation plan;

[0010] The recursive integration module extracts the output of the recommendation model based on the integrated recommendation scheme, classifies and clusters the output results, compares the weight distribution of the output, calculates the parameters of the weight synchronization item, processes the recursive relationship of the weight, merges the output of the model, and generates the recursive integration output;

[0011] The real-time feedback module, based on the recursive integration output, analyzes the data flow within the time window, compares the deviation between the market data and the recursive integration output, calculates the parameter value for deviation adjustment, merges the adjusted parameters, updates the financial product recommendation data, and generates real-time correction results.

[0012] As a further solution of the present invention, the multi-level subtask structure includes risk level classification, investment period range and expected rate of return distribution, the independent subtask solution set includes risk matching results, investment period optimization and rate of return estimation, the integrated recommendation plan includes weighted combination investment plan, priority ranking and decision parameters, the recursive integration output includes model integration prediction, weight allocation results and recursive calculation parameters, and the real-time correction results include market deviation adjustment, updated recommendation parameters and corrected data flow analysis.

[0013] As a further solution of the present invention, the task decomposition module includes:

[0014] The dimension splitting submodule is based on the financial product recommendation task and uses task decomposition technology to split the dimensions according to risk level, investment period and expected rate of return, extract the independent task set of each dimension, classify and separate the independent task set, build the independent execution path of the independent task set, and generate the dimension task set;

[0015] The parameter combination submodule combines the parameters in the risk level, investment period and expected rate of return dimensions based on the dimension task set, uses cross-calculation to process the relationship between the parameter sets, classifies and integrates the dimension combination results, and generates a combination parameter set;

[0016] The interaction relationship processing submodule analyzes the interaction relationship between subtasks based on the combination parameter set, determines the priority of the combination by comparing the impact of the interaction on the parameter combination, and sorts and builds relationships among the combinations to generate a multi-level subtask structure.

[0017] As a further solution of the present invention, the data processing module includes:

[0018] The parameter extraction submodule extracts the parameter set of the subtask based on the multi-level subtask structure, screens and sorts the extracted parameters, classifies and groups the same type of parameter items, and integrates them to form a complete parameter structure of the subtask, generating a classified parameter set;

[0019] The interactive matching submodule analyzes the matching degree of the interaction items between subtasks based on the classification parameter set, calculates the matching correlation of the set, screens the matching items that meet the conditions, integrates the screened matching items into the task set, and generates a matching parameter set;

[0020] The association operation submodule calculates the association between the interaction items based on the matching parameter set, processes the association of the parameter set after the interaction matching, calculates and integrates the multi-dimensional association data, outputs the integrated solution set, and obtains the independent subtask solution set.

[0021] As a further solution of the present invention, the result integration module includes:

[0022] The weighted combination submodule performs weighted processing on the subtask solutions based on the independent subtask solution sets, aggregates the parameter items of the task solutions, integrates and calculates the weighted parameters, and sorts and classifies the weighted solution sets to generate weighted parameter sets;

[0023] The hierarchical integration submodule classifies the parameter results of the hierarchical levels based on the weighted parameter set, divides the hierarchical levels by integrating the priorities of the hierarchical results, constructs the relationship structure between the hierarchical levels, performs hierarchical integration processing on the results, and generates a hierarchical priority set;

[0024] The comprehensive relationship establishment submodule establishes the relationship between levels based on the hierarchical priority set, handles the interactive relationship between the hierarchical results, integrates the hierarchical priority results, and obtains an integrated recommendation plan.

[0025] As a further solution of the present invention, the recursive integration module includes:

[0026] The model extraction submodule extracts the output data of the recommendation model based on the integrated recommendation scheme, classifies and clusters the output results, integrates the output results of the model, builds a structured set of model output data, forms a cluster of model outputs, and generates a model output set;

[0027] The weight calculation submodule compares the weight distribution of the output results based on the model output set, calculates the synchronization and distribution values ​​of the weight items, makes overall adjustments and corrections between the weights, builds a weight distribution structure, makes the weights consistent in the output results, and generates a weight distribution set;

[0028] The recursive processing submodule uses adaptive recursive weighting technology based on the weight distribution set to recursively adjust and update the weights, iteratively correct the weight values ​​output by the model, merge the weight results after recursive processing through the weighted averaging method, integrate the data output by the model, and generate a recursive integrated output.

[0029] As a further solution of the present invention, the real-time feedback module includes:

[0030] The data analysis submodule analyzes the real-time data stream within the time window based on the recursive integrated output, compares the deviation value between the market data and the recursive output data, calculates the change trend of the deviation value, and performs dynamic analysis of the data stream, integrates the data results after analysis, and generates a deviation analysis set;

[0031] The deviation calculation submodule calculates the deviation adjustment parameter value based on the deviation analysis set, compares and analyzes the adjusted parameters with the market data, integrates the parameter change values, processes the correlation of the data stream, forms the final adjustment parameter structure, and generates the adjustment parameter set;

[0032] The data update submodule updates the overall structure of the recommendation data based on the adjustment parameter set, processes the updated data stream, synchronizes the updated data with the recommendation results, integrates the corrected data stream into the recommendation system, and generates real-time corrected results.

[0033] As a further solution of the present invention, the task decomposition technology is according to the formula:

[0034]

[0035] Calculate the comprehensive priority weight value of the dimension task, where p i is the parameter value for each dimension task, w i is the weight coefficient corresponding to the parameter value, r i is the task risk adjustment factor, v i is the time sensitivity parameter of the task, c i is the cost adjustment coefficient of the task, α, β and γ are weight coefficients, n is the total number of dimensional tasks, and F(t) is the priority weight value of all dimensional tasks.

[0036] As a further solution of the present invention, the adaptive recursive weighting technology is according to the formula:

[0037]

[0038] Calculate the weight update value for the next iteration, where W t is the current weight value, X t is the output result of the current model, λ is the learning rate, γ is the decay coefficient, R t is the correction factor of the current weight, δ is the weight coefficient of the adjustment factor, P t is the historical performance value of the model, ρ is the performance correction coefficient, E t is the rate of change of the current environment, μ is the stability coefficient, S t is the stability evaluation value of the model output, θ is the complexity correction coefficient, C t is the complexity parameter of the current task.

[0039] The artificial intelligence-based financial product recommendation method is performed based on the artificial intelligence-based financial product recommendation system, and includes the following steps:

[0040] S1, based on the financial product recommendation task, uses task decomposition technology to split the recommendation task according to the dimensions of risk level, investment period and expected rate of return, extracts the independent task set of each dimension, constructs the execution path of the independent task set, and forms a multi-level task structure by classifying and processing the interactive relationship of the task set;

[0041] S2, based on the multi-level task structure, adopts the data association operation method to extract the parameter set of subtasks, analyze the interaction relationship between subtasks, calculate the correlation and matching degree of interaction items, and perform weighted processing on matching items. By integrating the parameter results of subtasks, an independent subtask solution set is generated, and then the solutions are hierarchically integrated and prioritized;

[0042] S3, based on the hierarchical priority solution set, adopts recursive weighted integration technology to recursively adjust and correct the weight value of the model output, integrates the data output by the model through weight allocation and iterative update, processes and merges the recursive output results, and finally forms a recommendation decision.

[0043] Compared with the prior art, the advantages and positive effects of the present invention are:

[0044] In the present invention, by breaking down the financial product recommendation task into multiple dimensions and calculating the parameter combinations of these dimensions and their interactions, the priority of the recommendation scheme can be determined more accurately, so that when facing a complex financial market, the mutual influence between the various parameters can be analyzed more flexibly and meticulously, thereby providing more targeted financial product recommendations. By classifying and comparing the matching degrees between parameter sets and processing related data, the optimal independent subtask solution set can be screened from many possible recommendation schemes. This weighted, aggregated and classified processing method can effectively integrate various solutions and establish associations between levels to ensure the comprehensiveness and rationality of the recommendation scheme. The application of the recursive integration method further improves the response speed in processing real-time market data, and can quickly adjust the recommendation results to maintain the effectiveness and accuracy of the recommendation. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 It is a system architecture diagram of the present invention. DETAILED DESCRIPTION

[0046] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0047] In the description of the present invention, it should be understood that the terms "length", "width", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, in the description of the present invention, "multiple" means two or more, unless otherwise clearly and specifically defined.

[0048] See also Figure 1 The present invention provides a technical solution: a financial product recommendation system based on artificial intelligence includes:

[0049] The task decomposition module, based on the financial product recommendation task, splits the recommendation task according to the dimensions of risk level, investment period and expected rate of return, calculates the parameter combination of the risk level, investment period and expected rate of return, processes the interaction between subtasks, judges the impact of the interaction on the combination, determines the combination priority, and obtains a multi-level subtask structure;

[0050] The data processing module extracts the parameter set of the subtasks based on the multi-level subtask structure, classifies the interaction items of the tasks, compares the matching degree between the sets, selects the matching items, calculates the correlation between the interaction items, processes the correlation data of the sets, and obtains the independent subtask solution set;

[0051] The result integration module, based on the independent subtask solution set, weights the subtask solutions, aggregates the solution parameters, classifies the parameter results, integrates the priorities of the sub-results, establishes the relationship between the levels, and obtains the integrated recommendation plan;

[0052] The recursive integration module extracts the output of the recommendation model based on the integrated recommendation scheme, classifies and clusters the output results, compares the weight distribution of the output, calculates the parameters of the weight synchronization items, processes the recursive relationship of the weights, merges the output of the model, and generates the recursive integration output;

[0053] The real-time feedback module, based on the recursive integration output, analyzes the data flow within the time window, compares the deviation between the market data and the recursive integration output, calculates the parameter value for deviation adjustment, merges the adjusted parameters, updates the financial product recommendation data, and generates real-time correction results.

[0054] The multi-level subtask structure includes risk level classification, investment period range and expected rate of return distribution. The independent subtask solution set includes risk matching results, investment period optimization and rate of return estimation. The integrated recommendation plan includes weighted combination investment plan, priority ranking and decision parameters. The recursive integration output includes model integration prediction, weight allocation results and recursive calculation parameters. The real-time correction results include market deviation adjustment, updated recommendation parameters and corrected data flow analysis.

[0055] In the task decomposition module, through a detailed analysis of the financial product recommendation task, the recommendation task is split according to the three dimensions of risk level, investment period and expected rate of return. In the specific operation, the historical data set of each financial product is first classified and processed, and the data is organized into three main dimensions in a structured form: risk level classification, investment period interval and expected rate of return distribution. The data is represented by vectorization, and each vector contains multiple indicators of the corresponding dimension, such as volatility, maximum drawdown and other indicators in the risk level. On this basis, the dimensional decomposition technology is used to decompose and combine the vectors to generate a multi-level subtask structure. Through the recursive decomposition strategy, the structure is gradually refined to form a specific subtask set, and the interactive relationship of each subtask is mapped through the projection technology of the vector space to determine the impact of the interaction on the combination of each subtask. Finally, the priority sorting algorithm is used to calculate the priority of each subtask, and a multi-level subtask structure file is generated according to the priority.

[0056] In the data processing module, through the multi-level subtask structure, the parameter set of the subtask is first extracted. The parameter set is organized in the form of a matrix, and each matrix unit corresponds to the specific value of a subtask dimension. Next, the parameter matching algorithm is used to compare the matrix units to determine the correlation between tasks. The interaction items between different subtasks are classified by calculating the covariance matrix, and the matching degree is analyzed. The similarity clustering algorithm is used to cluster the subtasks to ensure that highly related subtasks can be effectively identified and combined. Next, the matching task set is processed by association operation technology, including weighted calculation and normalization. The processed data is stored as an independent subtask solution set, each of which contains risk matching results, investment period optimization and rate of return estimation.

[0057] In the result integration module, the solutions of each subtask are first weighted through the independent subtask solution set. In the specific operation, a weighted optimization algorithm based on gradient descent is used to perform multiple rounds of weighted calculations on each subtask solution set to ensure that the weights of all solution sets can reflect the actual needs of the current market and the preferences of users. The weighted data is stored as a weighted parameter set. Next, the weighted parameters are integrated using the hierarchical analysis method, and the parameter results of different levels are classified and processed through the priority sorting technology, and finally a hierarchical priority solution set is formed. Then, through the multi-dimensional comprehensive analysis technology, the priority results of each level are integrated, the relationship between the levels is established, and an integrated recommendation plan is generated. The recommendation plan is output in the form of a table or a tree structure file, which contains the weighted combination of investment plans, priority sorting and decision parameters.

[0058] In the recursive integration module, by integrating the recommendation scheme, the output data of the recommendation model is first extracted. The data is represented in matrix form, with each row corresponding to the output value of a recommendation model and each column corresponding to different output parameters of the model. Next, the output data is classified and clustered by the matrix clustering algorithm, and the weight distribution of each output parameter is analyzed. The recursive weighted integration technology based on the weight distribution set is used to iteratively correct the weight of the model output multiple times. By introducing correction factors and dynamic adjustment strategies, the weight distribution is made more accurate. Then, the corrected weight results are subjected to multiple rounds of weighted average processing, the data output by each model is integrated, and the processed data is merged to form a recursive integrated output. Finally, through weight synchronization and recursive processing, an optimized weight distribution matrix and an integrated data file of the model output are generated.

[0059] In the real-time feedback module, through recursive integration output, the market data in the time window is first compared with the recursive integration output, and the dynamic deviation analysis technology is used to calculate the deviation value between the market data and the recursive output. The deviation is adjusted by introducing a time weighting factor to ensure that the deviation calculation can reflect the real-time changes in the market. Next, the parameter value of the deviation adjustment is calculated, and the adjusted parameter value is processed by regression analysis technology to obtain an adjustment parameter set, which is used to update the recommendation data of financial products in real time. Then, the updated data stream is synchronized with the original recommendation data, and the data is dynamically analyzed using data stream integration technology. Finally, a real-time correction result file is generated, which contains market deviation adjustment, updated recommendation parameters and corrected data stream analysis. Through the processing of this module, the recommendation system can respond to market changes in real time and ensure the effectiveness and accuracy of the recommendation results.

[0060] In this embodiment, the task decomposition module includes:

[0061] The dimension splitting submodule is based on the financial product recommendation task and uses task decomposition technology to split the dimensions according to risk level, investment period and expected rate of return, extract the independent task set of each dimension, classify and separate the independent task set, build the independent execution path of the independent task set, and generate the dimension task set;

[0062] The parameter combination submodule combines the parameters in the risk level, investment period and expected rate of return dimensions based on the dimension task set, uses cross-calculation to process the relationship between the parameter sets, classifies and integrates the dimension combination results, and generates a combination parameter set;

[0063] The interaction relationship processing submodule analyzes the interaction relationship between subtasks based on the combination parameter set, determines the priority of the combination by comparing the impact of the interaction on the parameter combination, and sorts and builds relationships among the combinations to generate a multi-level subtask structure.

[0064] In the dimension splitting submodule, by conducting a detailed analysis of the financial product recommendation task, risk level, investment period and expected rate of return are split as key dimensions. First, the financial product data is preprocessed to extract the relevant attributes of each dimension to ensure the integrity and consistency of the data. The value range of each dimension is analyzed by statistical analysis methods to determine the discretization scheme of each dimension. Subsequently, a splitting algorithm based on a decision tree is used to divide each dimension into several independent subtask sets, each of which represents a specific task under this dimension. In the specific implementation process, the decision tree algorithm selects the best splitting point according to the principle of maximizing information gain, thereby decomposing the original task into multiple subtask sets that can be processed independently. In this way, a dimensional task set based on risk level, investment period and expected rate of return is generated, and each dimensional task set can be executed independently in subsequent processing. This refinement operation reduces the complexity of the task, facilitates the optimization processing of the subsequent algorithm, and provides a basis for the comprehensive analysis of multi-dimensional tasks.

[0065] In the parameter combination submodule, the parameters in the three dimensions of risk level, investment period and expected rate of return are systematically combined by parameterizing the combination of the dimensional task sets generated above. In the specific implementation process, the Cartesian product method is used to comprehensively combine the parameters of each dimension to generate all parameter combinations. In order to avoid excessive computational complexity caused by too many combination results, a pruning technology based on a heuristic algorithm is used to filter out combinations that do not conform to business logic or are obviously invalid, and retain valid combinations with business significance. Each valid combination represents a parameter configuration, which is associated with the characteristic data of the financial product through a mapping function to generate a combination parameter set.

[0066] In the interaction relationship processing submodule, the generated combination parameter set is deeply analyzed to clarify the interaction relationship between each subtask, and the priority of the combination is determined based on the relationship. In the specific implementation, the topological sorting algorithm in graph theory is first used to construct a directed graph between subtasks, where nodes represent subtasks and edges represent dependencies between tasks. By topologically sorting the directed graph, the priority execution order of each subtask is obtained. Subsequently, the priority queue algorithm is used to sort the combination parameter set according to priority. On this basis, the hierarchical analysis method (AHP) is further used to calculate the weights of the sorted subtasks to ensure that each subtask can get reasonable resource allocation according to its importance during the entire task execution process.

[0067] In this embodiment, the data processing module includes:

[0068] The parameter extraction submodule extracts the parameter set of the subtask based on the multi-level subtask structure, screens and sorts the extracted parameters, classifies and groups the same type of parameter items, and integrates them to form a complete parameter structure of the subtask, generating a classified parameter set;

[0069] The interactive matching submodule analyzes the matching degree of the interaction items between subtasks based on the classification parameter set, calculates the matching correlation of the set, screens the matching items that meet the conditions, integrates the screened matching items into the task set, and generates a matching parameter set;

[0070] The association operation submodule calculates the association between the interaction items based on the matching parameter set, processes the association of the parameter set after the interaction matching, calculates and integrates the multi-dimensional association data, outputs the integrated solution set, and obtains the independent subtask solution set.

[0071] In the parameter extraction submodule, the data of the multi-level subtask structure is analyzed and operated in detail. First, the parameter set of each subtask is extracted from the multi-level subtask structure. The parameter set is represented in the form of a matrix or vector, where each row of the matrix or each element of the vector represents a dimension or feature of the subtask. Then, the extracted parameters are screened, mainly including removing noise data, processing missing values, and correcting outliers to ensure the quality and integrity of the data. On the basis of screening, the parameters are further sorted, and the parameters with similar characteristics or belonging to the same category are classified and grouped. This operation uses clustering analysis technology to aggregate similar parameter items by calculating the similarity or correlation between parameters, and uses the vector space model to integrate the aggregated parameter items to form a complete parameter structure of the subtask. The structure is represented in the form of a high-dimensional matrix or multi-dimensional array, and the parameter items of each subtask have a clear position and association relationship in the structure. Finally, the generated classified parameter set provides a structured and optimized input data set for subsequent interactive matching and association operations, ensuring data consistency and operation efficiency in subsequent processing.

[0072] In the interactive matching submodule, the interactive relationship of the classification parameter set is analyzed and the matching degree is calculated. First, the parameters of each subtask in the classification parameter set are analyzed one by one to identify the parameter items that may have interactive relationships. The identification of interactive relationships is based on multivariate statistical analysis, and the covariance matrix or mutual information between parameters is calculated to determine the association strength between two or more parameter items. Next, the matching correlation of the set is calculated. Specifically, the parameter sets of different subtasks are compared through the matching algorithm to evaluate the matching degree. The calculation of the matching degree here includes indicators such as distance measurement and similarity score. Combined with the indicators, a threshold is set to screen out the matching items that meet the conditions. The screened matching items are reorganized and integrated in the parameter set. The integration operation involves the standardization and normalization of the parameters to eliminate the dimensional differences and make the different parameter items comparable. Finally, the integrated matching items are organized into a new task set to generate a matching parameter set. The matching parameter set will be used for the next association operation. Its role is to ensure that the input data has a high degree of correlation and consistency in the association operation, thereby improving the accuracy and efficiency of the operation.

[0073] In the association operation submodule, the matching parameter set is subjected to in-depth association analysis and multi-dimensional operations. First, based on the matching parameter set, the correlation between each parameter item is calculated. The calculation of the correlation degree generally adopts the methods of correlation analysis and regression analysis to evaluate the linear or nonlinear relationship between each parameter item. For a deeper analysis, it is necessary to calculate the parameter interaction term, which involves the addition of quadratic terms or interaction terms, so that the model can capture more complex association structures. After completing the calculation of the correlation degree, the parameter set after the interactive matching is further processed. After weighted processing, the multi-dimensional data integration technology is used to combine each parameter item into a high-dimensional feature vector or matrix. The association data of different dimensions are integrated by matrix decomposition or vector superposition, and the integrated solution set is finally output. The obtained independent subtask solution set contains the key association features of each subtask. As the final output file, the solution set will be directly applied to the subsequent decision-making and optimization steps. Its role is to provide the system with a highly refined and clearly associated task solution set to ensure that the entire financial product recommendation system can operate efficiently and accurately.

[0074] In this embodiment, the result integration module includes:

[0075] The weighted combination submodule performs weighted processing on the subtask solutions based on the independent subtask solution sets, aggregates the parameter items of the task solutions, integrates and calculates the weighted parameters, and sorts and classifies the weighted solution sets to generate weighted parameter sets;

[0076] The hierarchical integration submodule classifies the parameter results of the hierarchical levels based on the weighted parameter set, divides the hierarchical levels by integrating the priorities of the hierarchical results, constructs the relationship structure between the hierarchical levels, performs hierarchical integration processing on the results, and generates a hierarchical priority set;

[0077] The comprehensive relationship establishment submodule establishes the relationship between levels based on the hierarchical priority set, handles the interactive relationship between the hierarchical results, integrates the hierarchical priority results, and obtains an integrated recommendation plan.

[0078] In the weighted combination submodule, the parameter items of each subtask solution are systematically aggregated by weighting the independent subtask solution set, ensuring that the influence and importance of each subtask are fully reflected in the integration process. In the specific implementation process, the parameter items of each subtask solution are first standardized to eliminate the dimensional differences between different parameters. Then, the entropy method is used to calculate the weight of each parameter item to ensure that the weight distribution meets the data characteristics and business requirements. The entropy method reflects the information distribution of each parameter item in the solution set by calculating the entropy value of each parameter item. The smaller the entropy value, the higher the importance of the parameter. After determining the weight, the parameter items are weighted averaged to generate the weighted parameter solution set. In this process, in order to avoid possible deviations in the weighting process, an adaptive weighting algorithm is used to dynamically adjust the weights to ensure the rationality of the integration results. The weighted solution set is classified and sorted, and classified according to parameter characteristics and business logic to finally generate a weighted parameter set. This integration result lays the foundation for the subsequent hierarchical integration and the generation of comprehensive recommendation solutions, and helps to optimize the global quality of the solution set.

[0079] In the hierarchical integration submodule, the hierarchical relationship between parameters is constructed by classifying and hierarchically processing the weighted parameter set, so as to extract meaningful hierarchical results from the complex multidimensional solution set. First, the parameter results in the weighted parameter set are classified, and the hierarchical structure between parameters is constructed by the analytic hierarchy process (AHP) to clarify the priority relationship between different parameters. AHP compares the relative importance of parameters by constructing a judgment matrix, and ensures the rationality of the judgment matrix through consistency test. Next, according to the priority results of different levels, the hierarchical results are integrated using a hierarchical clustering algorithm, and parameters with similar priorities and business relevance are classified into the same level. On this basis, hierarchical integration processing is carried out, and the results are aggregated layer by layer to finally generate a hierarchical priority set. The hierarchical priority set reasonably classifies the solution sets of different levels, so that in the subsequent comprehensive relationship processing, the relationship between levels can be established more efficiently, and the logic and execution efficiency of the final recommendation plan can be ensured.

[0080] In the comprehensive relationship establishment submodule, a relationship model between the hierarchical priorities is established through a systematic analysis of the hierarchical priority set to ensure that the interactive relationships between the results of each level are fully processed and integrated. In the specific operation, the Bayesian network is first used to construct the relationship diagram between the levels, and the conditional probability distribution is used to quantify the interactive effects between different levels to determine the interdependence and conditional independence of each level. Through the reasoning of the Bayesian network, the key hierarchical nodes and their influence on other levels are identified. Then, the hierarchical priority results are integrated and the Markov Chain Monte Carlo (MCMC) method is used to globally optimize the multi-level results to ensure that the results of each level can be reasonably combined in the complex hierarchical structure. The final integrated recommendation plan covers the priority information and interactive relationships of each level, ensuring that the recommendation plan has a high degree of accuracy and feasibility in practical applications, while providing decision makers with a clear logical path and action suggestions.

[0081] In this embodiment, the recursive integration module includes:

[0082] The model extraction submodule extracts the output data of the recommendation model based on the integrated recommendation scheme, classifies and clusters the output results, integrates the output results of the model, builds a structured set of model output data, forms a cluster of model outputs, and generates a model output set;

[0083] The weight calculation submodule compares the weight distribution of the output results based on the model output set, calculates the synchronization and distribution values ​​of the weight items, makes overall adjustments and corrections between the weights, builds a weight distribution structure, makes the weights consistent in the output results, and generates a weight distribution set;

[0084] The recursive processing submodule uses adaptive recursive weighting technology based on the weight distribution set to recursively adjust and update the weights, iteratively correct the weight values ​​output by the model, merge the weight results after recursive processing through the weighted averaging method, integrate the data output by the model, and generate a recursive integrated output.

[0085] In the model extraction submodule, the output data of the recommendation model is extracted and processed based on the integrated recommendation scheme. First, the output data of the model is obtained from the integrated recommendation scheme. The data is represented in the form of a matrix or a high-dimensional array, and each element corresponds to the specific output results of different models. Then, the output data is classified, mainly by classifying similar output results together through a feature-based clustering algorithm, so that the data set has higher homogeneity. The calculation of the feature vector is the key operation. The main features of the output data of each model are extracted by techniques such as principal component analysis, and clustering is performed based on this. After clustering is completed, the clustering results are further integrated. The integration operation includes weighted averaging and standardization of similar results, so that the data between different clusters are comparable. Next, by constructing a structured set of model output data, all processed data are hierarchically organized into a cluster for subsequent operations. The structured collection is represented in the form of a tree or graph structure, ensuring the hierarchical relationship and internal correlation of the output data of each model. The final generated model output set is saved as a file, providing basic data input for subsequent weight calculation and recursive processing, ensuring that the system's output data can be effectively managed and utilized.

[0086] In the weight calculation submodule, the weights are calculated and assigned to the data in the model output set. First, each data set in the model output set is analyzed one by one, and the weight value of each output result is calculated using the weight allocation strategy. In the specific operation, the influence of each model output on the overall recommendation is calculated based on historical data and current market conditions, and the influence is converted into a preliminary weight value. Next, the weight items are synchronized, including the weights and assigned values ​​of different model outputs. To achieve the purpose, the weight items are carefully adjusted using the weighted regression method, and the parameter set that makes the weight allocation optimal is found through multiple rounds of iterative operations. On this basis, the calculated weight values ​​are adjusted and corrected as a whole. The correction step involves standardizing and normalizing the weight values ​​to ensure that all weight values ​​are comparable on the same scale. Then, a structured model for weight allocation is constructed so that the weights are consistent in different output results. The structure is represented in the form of a matrix or tensor, reflecting the relative importance of each model output in the final decision. The generated weight distribution set will serve as the input of the recursive processing submodule to ensure that subsequent weight adjustments and corrections are based on the optimal initial weight configuration, thereby improving the overall recommendation accuracy of the system.

[0087] In the recursive processing submodule, the weight distribution set is adaptively recursively weighted to optimize the weight value of the model output. First, for each weight value in the weight distribution set, the recursive weighting technique is used for processing. The specific operation is to iteratively correct each weight through the adaptive adjustment mechanism. In each round of iteration, the difference between the current weight and the target weight is first calculated, and this difference is combined with the historical adjustment parameter to form a new correction factor. Then the correction factor is applied to the current weight to obtain the corrected weight value. The process is repeated multiple times in a recursive manner until the weight value reaches the preset convergence condition. In order to ensure the accuracy of the weight correction, a dynamic adjustment strategy is adopted to introduce the latest market data and model performance data in each iteration to ensure that the weight adjustment can reflect market changes in real time. The corrected weight values ​​are merged by weighted average method to form an optimized weight matrix. Then, the corrected model output data is integrated to generate a recursive integrated output file, which will reflect the final model output results and its weight configuration. The recursive processing submodule undergoes multiple iterations and corrections to make the weight configuration of the model output more accurate and reasonable, ensuring that the system can still provide the best recommendation results in a changing market environment.

[0088] In this embodiment, the real-time feedback module includes:

[0089] The data analysis submodule analyzes the real-time data stream within the time window based on the recursive integrated output, compares the deviation value between the market data and the recursive output data, calculates the change trend of the deviation value, and performs dynamic analysis of the data stream, integrates the data results after analysis, and generates a deviation analysis set;

[0090] The deviation calculation submodule calculates the deviation adjustment parameter value based on the deviation analysis set, compares and analyzes the adjusted parameters with the market data, integrates the parameter change values, processes the correlation of the data stream, forms the final adjustment parameter structure, and generates the adjustment parameter set;

[0091] The data update submodule updates the overall structure of the recommendation data based on the adjustment parameter set, processes the updated data stream, synchronizes the updated data with the recommendation results, integrates the corrected data stream into the recommendation system, and generates real-time corrected results.

[0092] In the data analysis submodule, the deviation between the market data and the recursive output data in a specific time window is systematically analyzed through real-time data stream analysis based on recursive integrated output. First, the real-time data stream is preprocessed, the data is divided into subsets of fixed time windows, and the weighted moving average method is used to smooth the data stream to reduce noise interference. Then, the market data and the recursive output data are compared point by point through the dynamic time warping (DTW) algorithm, the deviation value is calculated, and the nonlinear change trend between the data is captured. The DTW algorithm can effectively deal with the time dislocation problem by finding the best alignment path between time series, making the deviation value calculation more accurate. Subsequently, the sliding window technology is used to track the deviation value, calculate its change trend in different time windows, and analyze the change rate of the deviation value through linear regression to determine whether the deviation trend tends to converge or diverge. Finally, all analysis results are integrated to generate a deviation analysis set. The deviation analysis set provides information about the differences between real-time data and model predictions, provides a data basis for subsequent deviation correction and parameter adjustment, and can guide the system to dynamically adjust the recommendation strategy in a complex and changing market environment.

[0093] In the deviation calculation submodule, the parameter values ​​that need to be corrected are determined and the system's response capacity is optimized through precise calculation based on the deviation analysis set. First, the different types of deviations in the deviation analysis set are classified, and the principal component analysis (PCA) technique is used to extract the key factors that mainly affect the deviation, reduce the data dimension and reduce redundant information. Then, based on the time series characteristics of the deviation, the Kalman filter algorithm is used to predict the parameter change trend in real time and generate the estimated deviation adjustment value. The Kalman filter algorithm updates the system state in real time through the recursive estimation method, and can maintain high-precision tracking of parameter changes in an environment with large noise interference. Subsequently, the adjusted parameter values ​​are cross-validated with real-time market data, and the correlation analysis method is used to evaluate the market adaptability after parameter adjustment. By calculating the correlation coefficient between the parameters and market data before and after the adjustment, the correctness and effectiveness of the parameter adjustment direction are ensured. The integrated parameter change values ​​are processed through nonlinear optimization to generate the final adjustment parameter structure, which contains the optimal matching results between market changes and parameter corrections, and forms an adjustment parameter set. Adjusting the parameter set is an instant optimization feedback to the system recommendation model, which can enhance the adaptability and accuracy of the system and ensure that the recommendation strategy can respond quickly and make effective adjustments when market conditions change.

[0094] In the data update submodule, the data structure of the recommendation system is updated in time by applying the adjustment parameter set, and the real-time nature of the recommendation results is maintained. First, the adjustment parameter set is imported into the core algorithm module of the system, and the weights and parameters of the model are gradually adjusted using the incremental update method to avoid causing large-scale disturbances to the system. Then, the updated data is monitored using real-time data stream processing technology to ensure that each parameter adjustment can be immediately reflected in the recommendation results. To this end, a streaming processing framework (such as Apache Flink or Spark Streaming) is used to achieve low-latency processing of large-scale data streams. During the synchronization process, the updated data stream is compared with the current recommendation results item by item, and a dynamic consistency check algorithm is used to ensure the logical consistency between the updated data and the existing data. Finally, all the corrected data streams are integrated into the recommendation system, real-time correction results are generated, and updated recommendation solutions are output. This real-time correction result not only ensures that the recommendation system can maintain efficient operation in a dynamic environment, but also enhances the flexibility of the system, enabling it to continuously adapt to rapid changes in the market and provide users with more accurate and timely recommendation services.

[0095] In this embodiment, the task decomposition technology is based on the formula:

[0096]

[0097] Calculate the comprehensive priority weight value of the dimension task, where p i is the parameter value for each dimension task, w i is the weight coefficient corresponding to the parameter value, r i is the task risk adjustment factor, v i is the time sensitivity parameter of the task, c i is the cost adjustment coefficient of the task, α, β and γ are weight coefficients, n is the total number of dimensional tasks, and F(t) is the priority weight value of all dimensional tasks.

[0098] Execution process:

[0099] First, determine the parameter value p for each dimension task i and the corresponding weight coefficient w i , and multiply them to get the preliminary weight contribution value, then calculate the risk adjustment coefficient r of each task i , considering the potential risk of the task during execution, multiplying it with the weight coefficient α to obtain the risk correction value, and introducing the risk correction value into the adjustment factor of the score structure to consider the time sensitivity parameter v of the task i , using 1+β×v i The balance factor reflects the degree of response of the task to time changes. At the same time, the cost adjustment coefficient c is calculated. i, which reflects the economic impact of the task. After multiplying it with the weight coefficient γ, it is used as the denominator to balance the cost impact. Finally, all the corrected values ​​are combined and all the calculation results are summed up to obtain the comprehensive priority weight value F(t).

[0100] In this embodiment, the adaptive recursive weighting technology is based on the formula:

[0101]

[0102] Calculate the weight update value for the next iteration, where W t is the current weight value, X t is the output result of the current model, λ is the learning rate, γ is the decay coefficient, R t is the correction factor of the current weight, δ is the weight coefficient of the adjustment factor, P t is the historical performance value of the model, ρ is the performance correction coefficient, E t is the rate of change of the current environment, μ is the stability coefficient, S t is the stability evaluation value of the model output, θ is the complexity correction coefficient, C t is the complexity parameter of the current task.

[0103] Execution process:

[0104] First, determine the initial weight value W of the current model output t , extract the output result X of the current model t And calculate the difference between it and the current weight value, adjust the difference through the learning rate λ, and then introduce the adjustment factor weight coefficient δ and the historical performance value P of the model t , calculate the adjustment factor, using the performance correction factor ρ and the environmental change rate E t Balance the influence of the adjustment factor, multiply the adjusted factor value by the initial difference, and then introduce the correction factor R t The stability coefficient μ and the stability evaluation value S output by the model are used to make basic corrections. t Calculate the stability correction and combine the complexity correction coefficient θ with the complexity parameter C of the current task t Perform balancing processing. Finally, substitute all the corrected results into the formula, merge the processed weight results, and get the updated weight value W t+1 .

[0105] The present invention provides a financial product recommendation method based on artificial intelligence, comprising the following steps:

[0106] Based on the financial product recommendation task, the task decomposition technology is used to split the recommendation task according to the dimensions of risk level, investment period and expected rate of return, extract the independent task set of each dimension, build the execution path of the independent task set, and form a multi-level task structure by classifying and processing the interactive relationship of the task set;

[0107] Based on the multi-level task structure, the data association operation method is used to extract the parameter set of subtasks, analyze the interaction relationship between subtasks, calculate the correlation and matching degree of interaction items, and perform weighted processing on matching items. By integrating the parameter results of subtasks, an independent subtask solution set is generated, and then the solutions are hierarchically integrated and prioritized.

[0108] Based on the hierarchical priority solution set, the recursive weighted integration technology is used to recursively adjust and correct the weight values ​​of the model output. Through weight allocation and iterative updating, the data output by the model is integrated, and the recursive output results are processed and merged to finally form a recommendation decision.

[0109] It should be understood that, although each step in the flow chart of each embodiment of the present invention is shown in sequence according to the indication of the arrow, these steps are not necessarily performed in sequence according to the order indicated by the arrow. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be performed in other orders. Moreover, at least a portion of the steps in each embodiment may include a plurality of sub-steps or a plurality of stages, and these sub-steps or stages are not necessarily performed at the same time, but can be performed at different times, and the execution order of these sub-steps or stages is not necessarily performed in sequence, but can be performed in turn or alternately with at least a portion of other steps or sub-steps or stages of other steps.

[0110] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the program can be stored in a non-volatile computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0111] The above are only preferred embodiments of the present invention and are not intended to limit the present invention in other forms. Any technician familiar with the profession may use the technical contents disclosed above to change or modify them into equivalent embodiments with equivalent changes and apply them to other fields. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention without departing from the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. The financial product recommendation system based on artificial intelligence is characterized by: The system comprises: The task decomposition module, based on the financial product recommendation task, splits the recommendation task according to the dimensions of risk level, investment period and expected rate of return, calculates the parameter combination of the risk level, investment period and expected rate of return, processes the interaction between subtasks, judges the impact of the interaction on the combination, determines the combination priority, and obtains a multi-level subtask structure; A data processing module, based on the multi-level subtask structure, extracts a set of subtask parameters, classifies the interaction items of the tasks, compares the matching degrees between the sets, selects matching items, calculates the correlation between the interaction items, processes the correlation data of the sets, and obtains an independent subtask solution set; A result integration module, based on the independent subtask solution set, weights the subtask solutions, aggregates the solution parameters, classifies the parameter results, integrates the priorities of the sub-results, establishes the relationship between the levels, and obtains an integrated recommendation solution; A recursive integration module, based on the integrated recommendation scheme, extracts the output of the recommendation model, classifies and clusters the output results, compares the weight distribution of the output, calculates the parameters of the weight synchronization item, processes the recursive relationship of the weight, merges the output of the model, and generates a recursive integration output; The real-time feedback module analyzes the data flow within the time window based on the recursive integration output, compares the deviation between the market data and the recursive integration output, calculates the parameter value adjusted for the deviation, merges the adjusted parameters, updates the financial product recommendation data, and generates real-time correction results.

2. The artificial intelligence-based financial product recommendation system according to claim 1, characterized in that: The multi-level subtask structure includes risk level classification, investment period range and expected rate of return distribution; the independent subtask solution set includes risk matching results, investment period optimization and rate of return estimation; the integrated recommendation plan includes weighted combination investment plan, priority ranking and decision parameters; the recursive integration output includes model integration prediction, weight allocation results and recursive calculation parameters; the real-time correction results include market deviation adjustment, updated recommendation parameters and corrected data flow analysis.

3. The artificial intelligence-based financial product recommendation system according to claim 1, characterized in that: The task decomposition module includes: The dimension splitting submodule is based on the financial product recommendation task and uses task decomposition technology to split the dimensions according to risk level, investment period and expected rate of return, extract the independent task set of each dimension, classify and separate the independent task set, build the independent execution path of the independent task set, and generate the dimension task set; The parameter combination submodule combines the parameters in the risk level, investment period and expected rate of return dimensions based on the dimension task set, processes the relationship between the parameter sets by cross operation, classifies and integrates the dimension combination results, and generates a combination parameter set; The interaction relationship processing submodule analyzes the interaction relationship between subtasks based on the combination parameter set, determines the priority of the combination by comparing the impact of the interaction on the parameter combination, and sorts and builds relationships among the combinations to generate a multi-level subtask structure.

4. The artificial intelligence-based financial product recommendation system according to claim 1, characterized in that: The data processing module comprises: The parameter extraction submodule extracts the parameter set of the subtask based on the multi-level subtask structure, screens and sorts the extracted parameters, classifies and groups the same type of parameter items, and integrates them to form a complete parameter structure of the subtask, thereby generating a classified parameter set; The interactive matching submodule analyzes the matching degree of the interactive items between the subtasks based on the classification parameter set, calculates the matching relevance of the set, screens the matching items that meet the conditions, integrates the screened matching items into the task set, and generates a matching parameter set; The association operation submodule calculates the association between the interaction items based on the matching parameter set, processes the association of the parameter set after the interaction matching, calculates and integrates the multi-dimensional association data, outputs the integrated solution set, and obtains the independent subtask solution set.

5. The artificial intelligence-based financial product recommendation system according to claim 1, characterized in that: The result integration module includes: The weighted combination submodule performs weighted processing on the subtask solutions based on the independent subtask solution set, aggregates the parameter items of the task solutions, integrates and calculates the weighted parameters, and sorts and classifies the weighted solution set to generate a weighted parameter set; A hierarchical integration submodule, based on the weighted parameter set, classifies the parameter results of the hierarchical levels, divides the levels by integrating the priorities of the hierarchical results, constructs a relationship structure between the levels, performs hierarchical integration processing on the results, and generates a hierarchical priority set; The comprehensive relationship establishment submodule establishes the relationship between the levels based on the level priority set, processes the interactive relationship between the level results, integrates the level priority results, and obtains an integrated recommendation plan.

6. The artificial intelligence-based financial product recommendation system according to claim 1, characterized in that: The recursive integration module comprises: The model extraction submodule extracts the output data of the recommendation model based on the integrated recommendation scheme, classifies and clusters the output results, integrates the output results of the model, constructs a structured set of model output data, forms a cluster of model outputs, and generates a model output set; The weight calculation submodule compares the weight distribution of the output results based on the model output set, calculates the synchronization and distribution values ​​of the weight items, makes overall adjustments and corrections between the weights, constructs a weight distribution structure, makes the weights consistent in the output results, and generates a weight distribution set; The recursive processing submodule uses adaptive recursive weighting technology based on the weight distribution set to recursively adjust and update the weights, iteratively correct the weight values ​​output by the model, merge the weight results after recursive processing through weighted averaging, integrate the data output by the model, and generate a recursive integrated output.

7. The artificial intelligence-based financial product recommendation system according to claim 1, characterized in that: The real-time feedback module comprises: The data analysis submodule analyzes the real-time data stream within the time window based on the recursive integrated output, compares the deviation value between the market data and the recursive output data, calculates the variation trend of the deviation value, and performs dynamic analysis of the data stream, integrates the analyzed data results, and generates a deviation analysis set; The deviation calculation submodule calculates the deviation adjustment parameter value based on the deviation analysis set, compares and analyzes the adjusted parameter with the market data, integrates the parameter change value, processes the correlation of the data stream, forms the final adjustment parameter structure, and generates the adjustment parameter set; The data update submodule updates the overall structure of the recommendation data based on the adjustment parameter set, processes the updated data stream, synchronizes the updated data with the recommendation results, integrates the corrected data stream into the recommendation system, and generates real-time corrected results.

8. The artificial intelligence-based financial product recommendation system according to claim 3, characterized in that: The task decomposition technique is according to the formula: Calculate the comprehensive priority weight value of the dimension task, where p i is the parameter value for each dimension task, w i is the weight coefficient corresponding to the parameter value, r i is the task risk adjustment factor, v i is the time sensitivity parameter of the task, c i is the cost adjustment coefficient of the task, α, β and γ are weight coefficients, n is the total number of dimensional tasks, and F(t) is the priority weight value of all dimensional tasks.

9. The artificial intelligence-based financial product recommendation system according to claim 6, characterized in that: The adaptive recursive weighting technique is according to the formula: Calculate the weight update value for the next iteration, where W t is the current weight value, X t is the output result of the current model, λ is the learning rate, γ is the decay coefficient, R t is the correction factor of the current weight, δ is the weight coefficient of the adjustment factor, P t is the historical performance value of the model, ρ is the performance correction coefficient, E t is the rate of change of the current environment, μ is the stability coefficient, S t is the stability evaluation value of the model output, θ is the complexity correction coefficient, C t is the complexity parameter of the current task.

10. A financial product recommendation method based on artificial intelligence, characterized in that: The artificial intelligence-based financial product recommendation system according to any one of claims 1 to 9 comprises the following steps: Based on the financial product recommendation task, the task decomposition technology is used to split the recommendation task according to the dimensions of risk level, investment period and expected rate of return, extract the independent task set of each dimension, build the execution path of the independent task set, and form a multi-level task structure by classifying and processing the interactive relationship of the task set; Based on the multi-level task structure, the data association operation method is used to extract the parameter set of subtasks, analyze the interaction relationship between subtasks, calculate the correlation and matching degree of interaction items, and perform weighted processing on matching items. By integrating the parameter results of subtasks, an independent subtask solution set is generated, and then the solutions are hierarchically integrated and prioritized. Based on the hierarchical priority solution set, the recursive weighted integration technology is used to recursively adjust and correct the weight values ​​of the model output. Through weight allocation and iterative updating, the data output by the model is integrated, and the recursive output results are processed and merged to finally form a recommendation decision.

Images