AI credit marketing management system based on multimodal data fusion

Abstract

The present invention relates to the field of credit marketing technology, and more particularly to an AI-powered credit marketing management system based on multimodal data fusion. This system uses web crawler tools and API interfaces to acquire heterogeneous data from multiple sources, including social media and credit transactions. This system integrates social media and credit transaction data to construct customer feature vectors, thereby addressing the problem of "data silos leading to misjudgment of demand." Furthermore, it combines financial market and credit transaction data to generate real-time compliance constraint data to respond to market fluctuations. Based on this data, it uses a combination of "dynamic parameter generation, portfolio optimization, and preference weighting" to generate personalized credit marketing strategies to meet diverse customer needs. Finally, it calculates marketing accuracy by acquiring multidimensional matching indicators and interactive feedback data. A three-dimensional assessment method, "historical matching + real-time prediction," replaces subjective judgment. This system compares marketing accuracy with preset thresholds, adjusting marketing strategies if these thresholds are not met, thereby achieving intelligent upgrades in credit marketing.

Description

Technical Field

[0001] The present invention relates to the field of credit marketing technology, and in particular to an AI credit marketing management system based on multimodal data fusion. Background Art

[0002] Credit marketing refers to the commercial activities of financial institutions (such as banks, consumer finance companies, etc.) using credit products (such as loans, credit cards, etc.) as the core, to promote financial services to target customer groups through a series of strategies such as market research, product design, and channel promotion, in order to achieve profit goals and meet customer funding needs. Its essence is to accurately match financial resources with customer needs. Against the backdrop of the vigorous development of financial technology, the field of credit marketing is facing profound changes and innovation needs. With the acceleration of digital transformation, customer demand for credit products is becoming increasingly diversified and personalized. The differences between personal consumption scenarios and corporate operating needs require financial institutions to accurately grasp customer needs.

[0003] Existing credit marketing systems have significant limitations when processing multi-source heterogeneous data. They usually only stay at the simple feature splicing level and fail to deeply explore the potential correlations between different modal data. Due to the one-sidedness and lack of depth of data fusion, existing marketing strategies are seriously disconnected from actual customer needs, and marketing accuracy is extremely low, resulting in poor marketing results. Summary of the Invention

[0004] The main purpose of this invention is to provide an AI credit marketing management system based on multimodal data fusion, aiming to solve the technical problems in the prior art.

[0005] The present invention proposes an AI credit marketing management system based on multimodal data fusion, comprising:

[0006] an acquisition module, configured to acquire multimodal data, wherein the multimodal data includes social media data, credit transaction data, financial market data, and customer interaction data, and to acquire a customer feature vector based on the social media data and the credit transaction data;

[0007] A constraint generation module, configured to obtain compliance constraint data based on the financial market data and the credit transaction data;

[0008] a strategy synthesis module, configured to obtain a credit marketing strategy based on the compliance constraint data, the customer feature vector, and the credit transaction data;

[0009] An evaluation module, configured to obtain multi-dimensional matching indicators and interactive feedback data based on the credit marketing strategy and the customer interaction data, and to obtain marketing accuracy based on the multi-dimensional matching indicators and interactive feedback data;

[0010] A judgment module, configured to judge whether the marketing accuracy is greater than a preset threshold;

[0011] If the marketing accuracy is greater than a preset threshold, it is determined that the credit marketing strategy meets customer needs;

[0012] If the marketing accuracy is not greater than a preset threshold, it is determined that the credit marketing strategy does not meet customer needs, and the credit marketing strategy is adjusted according to the multi-dimensional matching index until the credit marketing strategy meets customer needs.

[0013] Preferably, the acquisition module includes:

[0014] a feature extraction unit, configured to obtain consumption intention text data and consumption intention behavior data based on the social media data, and to obtain a plurality of consumption intention targets, consumption intention intensities, and consumption intention timestamps based on the consumption intention text data;

[0015] A feature analysis unit, configured to obtain consumption cycle and risk feature values ​​based on the credit transaction data;

[0016] a demand acquisition unit, configured to acquire the real-time intention strength of each consumption intention target according to the consumption intention strength and the consumption intention timestamp, and acquire the demand characteristic value of each consumption intention target according to the real-time intention strength and the consumption cycle;

[0017] a behavior encoding unit, configured to obtain a behavior feature value of each consumption intention target based on the consumption intention behavior data and the consumption intention timestamp;

[0018] The feature fusion unit is used to obtain a customer feature vector based on a plurality of demand feature values, behavior feature values ​​and risk feature values.

[0019] Preferably, the feature analysis unit in the acquisition module includes:

[0020] a time series segmentation subunit, configured to obtain transaction sequence data and original credit information data based on the credit transaction data, and to obtain multiple transaction windows and transaction category groups based on the transaction sequence data;

[0021] a capital analysis subunit, configured to obtain corresponding statistical characteristic data according to each transaction window, and obtain cash flow health according to the statistical characteristic data;

[0022] a cycle acquisition subunit, configured to perform Fourier transform on each of the transaction category groups to obtain a corresponding consumption cycle;

[0023] a credit analysis subunit, configured to obtain credit characteristic data based on the original credit data, and to obtain a credit risk level based on the credit characteristic data;

[0024] The risk quantification subunit is used to obtain a risk characteristic value according to the credit risk level and the cash flow health.

[0025] Preferably, the constraint generation module includes:

[0026] a policy parsing unit, configured to obtain policy constraint rules and interest rate fluctuation indicators based on the financial market data, wherein the policy constraint rules include loan limit rules, interest rate control rules, and repayment time rules, and to obtain a regulatory interest rate floor, static interest rate rules, and dynamic interest rate rules based on the interest rate control rules;

[0027] a market sensing unit, configured to obtain an interest rate volatility and a volatility threshold according to the interest rate volatility indicator, and obtain a market adjustment value according to the interest rate volatility and the volatility threshold;

[0028] An interest rate restriction unit, configured to obtain a loan interest rate floor based on the interest rate volatility, the fluctuation threshold, the regulatory interest rate floor, the static interest rate rule, and the dynamic interest rate rule;

[0029] a credit limit unit, configured to obtain a solvency assessment value based on the credit transaction data, and obtain a basic loan amount and a loan multiple amount based on the solvency assessment value and the loan limit rule;

[0030] The repayment restriction unit is used to obtain the maximum loan amount according to the basic loan amount, the loan multiple amount and the market adjustment value, and to obtain the lower limit of the repayment time according to the repayment time rule and the market adjustment value.

[0031] Preferably, the strategy synthesis module includes:

[0032] an element generation unit, configured to obtain credit element parameters based on the compliance constraint data and the customer feature vector, wherein the credit element parameters include a plurality of recommended loan interest rates, recommended repayment periods, and recommended loan amounts;

[0033] a combining unit, configured to arrange and combine a plurality of the recommended loan interest rates, recommended repayment periods, and recommended loan amounts to obtain a plurality of marketing strategy combinations;

[0034] a matching degree obtaining unit, configured to obtain a single-dimensional matching degree based on the credit transaction data, the social media data, and the credit factor parameters;

[0035] a weight analysis unit, configured to obtain preference feedback data based on the customer interaction data, and perform correlation analysis on the credit factors and the preference feedback data to obtain a plurality of single-dimensional weights;

[0036] a preference evaluation unit, configured to obtain a customer preference value for each of the marketing strategy combinations based on a plurality of the single-dimensional matching degrees and the corresponding single-dimensional weights;

[0037] A screening unit is used to screen the plurality of marketing strategy combinations according to the customer preference value to obtain a credit marketing strategy.

[0038] Preferably, the matching degree acquisition unit in the strategy synthesis module includes:

[0039] An interval acquisition subunit, configured to acquire a preference parameter interval based on the credit transaction data;

[0040] a preference acquisition subunit, configured to acquire historical preference features based on the social media data, and acquire credit parameter preference values ​​based on the historical preference features and the preference parameter intervals, wherein the credit parameter preference values ​​include an interest rate preference value, a repayment preference period, and a credit limit preference value;

[0041] The preference mapping subunit is used to obtain the corresponding single-dimensional matching degree according to the credit marketing strategy, interest rate preference value, repayment preference period and loan amount preference value, wherein the single-dimensional matching degree includes multiple loan interest rate matching degrees, repayment period matching degrees and loan amount matching degrees.

[0042] Preferably, the evaluation module includes:

[0043] a scenario matching unit, configured to obtain a marketing scenario tag according to the credit marketing strategy, and obtain historical matching data according to the marketing scenario tag and the customer interaction data, wherein the historical matching data includes historical strategy parameters and interaction feedback data;

[0044] a multidimensional analysis unit, configured to obtain a multidimensional matching index based on the credit marketing strategy and the historical strategy parameters, wherein the multidimensional matching index includes parameter adaptability, scenario adaptability, and repayment cycle adaptability;

[0045] A strategy evaluation unit, configured to obtain a strategy matching degree based on the parameter adaptability, scenario adaptability, and repayment cycle adaptability;

[0046] An interaction prediction unit, configured to obtain the number of interactive behaviors, the number of users reached, the number of conversion behaviors, and the interaction time based on the interactive feedback data, and to obtain an expected interaction probability based on the number of interactive behaviors, the number of users reached, and the interaction time;

[0047] The accuracy acquisition unit is used to obtain the expected conversion rate based on the number of interactive behaviors, the number of conversion behaviors and the interaction time, and to obtain the marketing accuracy based on the strategy matching degree, the expected conversion rate and the expected interaction probability.

[0048] Preferably, the judgment module includes:

[0049] a parameter optimization unit, configured to obtain a matching threshold value, and obtain a credit parameter to be optimized based on the matching threshold value and the multi-dimensional matching index;

[0050] a gradient analysis unit, configured to obtain an accuracy gradient based on the multidimensional matching index and the customer preference value;

[0051] a step size configuration unit, configured to obtain an initial adjustment step size based on the financial market data, and obtain a step size constraint upper limit based on the policy constraint rules;

[0052] a dynamic adjustment unit, configured to obtain an adaptive step size according to the initial adjustment step size, the step size constraint upper limit, and the accuracy gradient, and to obtain a credit parameter adjustment value according to the credit parameter to be optimized and the adaptive step size;

[0053] An iterative unit is used to obtain a credit marketing adjustment strategy based on the credit parameter adjustment value and the credit marketing strategy, and return the credit marketing adjustment strategy as the credit marketing strategy to the step of obtaining marketing accuracy based on the credit marketing strategy and the customer interaction data.

[0054] The present invention also provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and when the memory executes the computer program, it is used to manage the operation of the above-mentioned AI credit marketing management system based on multimodal data fusion.

[0055] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is used to manage the operation of the above-mentioned AI credit marketing management system based on multimodal data fusion.

[0056] The beneficial effects of the present invention are as follows: the present invention obtains multimodal data through web crawler tools and according to social media platform API interfaces, credit reporting agency interfaces, etc., and constructs a customer feature vector that integrates customer needs and risk characteristics by integrating social media data and credit transaction data, thereby solving the problem of "data islands leading to misjudgment of demand". Then, the present invention combines financial market data and credit transaction data to generate real-time updated compliance constraint data to respond to market fluctuations in real time. Then, based on the compliance constraint data, customer feature vectors and credit transaction data, it generates personalized credit marketing strategies tailored to customers to solve the contradiction between "strategy singularity and demand diversification" of traditional solutions. Then, based on the Based on the multi-dimensional matching indicators and interactive feedback data obtained from credit marketing strategies and customer interaction data, marketing accuracy can be further obtained. Marketing accuracy can replace the traditional subjective judgment model, and objectively predict marketing effects through data-based means to avoid strategy misjudgments due to experience bias. Finally, the marketing accuracy is compared with the preset threshold. If the marketing accuracy is greater than the preset threshold, it is determined that the credit marketing strategy meets customer needs; if the marketing accuracy is not greater than the preset threshold, it is determined that the credit marketing strategy does not meet customer needs. The credit marketing strategy is adjusted according to the multi-dimensional matching indicators until the credit marketing strategy meets customer needs, which is conducive to the intelligent upgrade of credit marketing. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 FIG. 1 is a schematic diagram of a system structure according to an embodiment of the present invention.

[0058] Figure 2 This is a schematic diagram of the internal structure of a computer device according to an embodiment of the present application.

[0059] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings and in conjunction with the embodiments. DETAILED DESCRIPTION

[0060] 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.

[0061] like Figure 1 As shown, this application provides an AI credit marketing management system based on multimodal data fusion. Figure 1 In this management system,

[0062] an acquisition module, configured to acquire multimodal data, wherein the multimodal data includes social media data, credit transaction data, financial market data, and customer interaction data, and to acquire a customer feature vector based on the social media data and the credit transaction data;

[0063] A constraint generation module, configured to obtain compliance constraint data based on the financial market data and the credit transaction data;

[0064] a strategy synthesis module, configured to obtain a credit marketing strategy based on the compliance constraint data, the customer feature vector, and the credit transaction data;

[0065] An evaluation module, configured to obtain multi-dimensional matching indicators and interactive feedback data based on the credit marketing strategy and the customer interaction data, and to obtain marketing accuracy based on the multi-dimensional matching indicators and interactive feedback data;

[0066] A judgment module, configured to judge whether the marketing accuracy is greater than a preset threshold;

[0067] If the marketing accuracy is greater than a preset threshold, it is determined that the credit marketing strategy meets customer needs;

[0068] If the marketing accuracy is not greater than a preset threshold, it is determined that the credit marketing strategy does not meet customer needs, and the credit marketing strategy is adjusted according to the multi-dimensional matching index until the credit marketing strategy meets customer needs.

[0069] As the functions of each module mentioned above, the present invention obtains multimodal data through web crawler tools and according to the API interface of social media platform, credit reporting agency interface, etc., wherein multimodal data refers to a collection of information from different sources with different data types and structures, covering various forms such as text, image, audio, structured data, including social media data, credit transaction data, financial market data and customer interaction data, wherein social media data refers to the behavioral data generated by users on social platforms, including unstructured data such as text content, image information, interaction records, etc. Credit transaction data refers to the historical records of users in financial transactions, which are structured data, including credit information, income and expenditure flow, repayment records, etc. Financial market data Data refers to real-time data related to the macro-financial environment, including policies and regulations, interest rate fluctuations, and market dynamics. Customer interaction data refers to the historical interaction records of a large number of users with credit marketing activities, including contact behavior, feedback results, conversion data, etc. By integrating social media data and credit transaction data, a customer feature vector is constructed that combines customer needs and risk characteristics. Among them, the customer feature vector refers to the abstraction of multimodal data into a numerical feature set, which represents the customer's comprehensive attributes in vector form. By correlating cross-modal data, for example, identifying customers who "frequently browse high-end real estate on social media, have high credit income, and low debt", and accurately matching them with high-value mortgage strategies, thereby solving the problem of "data silos leading to misjudgment of demand";

[0070] Next, financial market data and credit transaction data are combined to generate real-time updated compliance constraint data. Compliance constraint data refers to constraint data extracted to comply with regulatory requirements and market rules, enabling real-time response to market fluctuations. Based on compliance constraint data, customer feature vectors, and credit transaction data, personalized credit marketing strategies are generated for each customer. These strategies are personalized marketing plans generated through multimodal data fusion and include core elements such as recommended loan interest rates, repayment periods, and loan amounts. Traditional marketing strategies rely on manual experience (e.g., "a uniform 10% interest rate increase") and fail to consider individual customer preferences (e.g., a customer who is sensitive to interest rates but accepts high loan amounts). This results in a poor match between marketing strategies and customer needs. This solution utilizes "dynamic parameter generation + portfolio optimization + preference weighting." For example, for high-income customers who demonstrate an "urgent need for money" on social media, a "high loan amount + slightly higher interest rate + flexible loan period" strategy is generated, addressing the contradiction between a single strategy and diverse needs.

[0071] Then, based on the credit marketing strategy and customer interaction data, multidimensional matching indicators and interactive feedback data are obtained. Among them, multidimensional matching indicators refer to multidimensional quantitative indicators used to evaluate the degree of fit between credit strategies and customer needs. Interactive feedback data refers to the direct response data of customers to marketing activities obtained from historical interaction records, including behavioral feedback (clicks, consultations) and result feedback (conversions, rejections). Marketing accuracy is obtained based on multidimensional matching indicators and interactive feedback data. Among them, marketing accuracy refers to a quantitative evaluation value that predicts the degree of fit between marketing strategies and customer needs by combining multidimensional matching indicators and interactive feedback data. Marketing accuracy can replace the traditional subjective judgment model. Data-based means are used to objectively predict marketing effects, avoiding strategic misjudgments due to experience bias. This solution uses the three-dimensional evaluation of "historical matching + real-time prediction" (matching + interaction + conversion) to solve the problem that traditional solutions only evaluate effects based on the final conversion rate and ignore the strategy matching process. Finally, the marketing accuracy is compared with the preset threshold. If the marketing accuracy is greater than the preset threshold, it is determined that the credit marketing strategy meets customer needs; if the marketing accuracy is not greater than the preset threshold, it is determined that the credit marketing strategy does not meet customer needs. The credit marketing strategy is then adjusted according to the multi-dimensional matching indicators until the credit marketing strategy meets customer needs, which is conducive to the intelligent upgrade of credit marketing.

[0072] In one embodiment, the acquisition module includes:

[0073] a feature extraction unit, configured to obtain consumption intention text data and consumption intention behavior data based on the social media data, and to obtain a plurality of consumption intention targets, consumption intention intensities, and consumption intention timestamps based on the consumption intention text data;

[0074] A feature analysis unit, configured to obtain consumption cycle and risk feature values ​​based on the credit transaction data;

[0075] a demand acquisition unit, configured to acquire the real-time intention strength of each consumption intention target according to the consumption intention strength and the consumption intention timestamp, and acquire the demand characteristic value of each consumption intention target according to the real-time intention strength and the consumption cycle;

[0076] a behavior encoding unit, configured to obtain a behavior feature value of each consumption intention target based on the consumption intention behavior data and the consumption intention timestamp;

[0077] The feature fusion unit is used to obtain a customer feature vector based on a plurality of demand feature values, behavior feature values ​​and risk feature values.

[0078] As for the functions of each unit mentioned above, the present invention performs semantic analysis on the consumption intention text data through the BERT-wwm model to obtain consumption intention feature data, wherein the consumption intention text data refers to the text information extracted from social media (such as forums, comment areas, etc.) that can reflect the user's consumption intention, and the consumption intention feature data refers to the structured feature information extracted from the consumption intention text. For example: the obtained consumption intention text data includes content such as "I want to buy a new mobile phone recently". For this content, a word segmentation tool (such as jieba) is used to segment the text, and then the pre-processed text is converted into word fragment tags (such as "new", "mobile", "machine"), and the word fragment tags are input The data is fed into the BERT-wwm model. Based on the global semantic representation or specific word vectors output by the model, the classification layer identifies the consumption goals in the text and obtains the consumption intention goals. The consumption intention goals refer to the specific objects or fields that the user's consumption intentions are directed to. For example, when the input text is "I want to buy a new mobile phone recently", the model outputs "mobile phone" as the consumption intention goal. Then, combined with the keyword weights in the text (such as "urgently needed" and "must" to enhance the strength), the consumption intention strength is obtained. The consumption intention strength is a quantitative indicator that measures the intensity of the user's consumption intention. Time-related expressions (such as "next month", "end of the year", and "this week") are extracted from the text and converted to a unified time format through a time normalization model (such as Date Parser) to obtain the consumption intention timestamp. The consumption intention timestamp refers to the specific time point when the consumption intention is expressed. For example, if the text "I want to buy a new mobile phone recently" is released on January 1, 2025, then the consumption intention timestamp corresponding to "mobile phone" as the consumption intention goal is "2025-01-01".

[0079] The consumption cycle and risk characteristic value are obtained based on the credit transaction data. The consumption cycle refers to the time interval between the customer's repeated consumption behaviors of the same type, and the risk characteristic value refers to the multi-dimensional indicator that quantifies the customer's credit risk. Then, through the formula " ” Calculate the real-time intention strength, where Indicates the real-time intention strength, Indicates the intensity of consumption intention, represents the first attenuation coefficient, and the first attenuation coefficient is determined by fitting historical data. Represents the time interval of consumption intention, and the time interval of consumption intention is the difference between the consumption intention timestamp and the current time. The real-time intention strength refers to the current moment intention strength after dynamic adjustment based on the consumption intention strength and timestamp. The existing technology ignores the timeliness of consumption intention (e.g., the car purchase intention 3 months ago may be irrelevant to the current demand). Direct use of the original strength will lead to misjudgment. This solution reduces the impact of outdated intentions through the time decay mechanism. For example, the car purchase intention with an intensity of 0.9 3 months ago is corrected to 0.36 to avoid "outdated demand" triggering invalid marketing. The cyclical pattern of user consumption has been identified through methods such as Fourier transform (e.g., the 10th of each month is the consumption peak), thereby obtaining the consumption cycle peak. Then, according to the formula " "The period matching degree is calculated, where represents the period matching degree, Indicates the timestamp of consumption intention, Indicates the peak of the recent consumption cycle, Represents the consumption cycle, and then according to the formula " ” Calculate the demand characteristic value, where represents the demand characteristic value, Represents the weight coefficient, which is determined by the consumption scenario (such as in the car loan scenario =0.7), Indicates the real-time intention strength, Represents the degree of cycle matching, where the demand feature value refers to an indicator that quantifies the urgency of users' demand for specific consumption goals. Through mechanisms such as cycle matching, "timeliness feature correction" is achieved to solve the defect of existing technologies that "static features cannot reflect users' dynamic needs", thereby improving the accuracy of marketing timing. Then, time series modeling is performed on consumption intention behavior data. Consumption intention behavior data refers to user behavior trajectory data related to consumption intention on social media, such as user likes, favorites, forwarding, following merchant accounts, browsing product pages, and other operation records. LSTM is used to extract sequence features, and after standardizing consumption intention behavior data (such as browsing time, number of consultations, and number of favorites), the recent behavior weight is calculated based on the timestamp (for example, the behavior weight of the past 7 days is 1, 8-30 days is 0.5, and more than 30 days is 0.2). The behavior feature value is generated through weighted summation. The behavior feature value refers to an indicator used to quantify the activeness of user consumption behavior;

[0080] Finally, the demand characteristic values, behavior characteristic values, and risk characteristic values ​​corresponding to each consumption intention target are spliced ​​together to obtain the customer feature vector. This solution breaks through the limitation of "simple splicing" of existing technology and realizes cross-modal association mining of social media and credit transaction data through "semantic understanding + time series analysis + graph association modeling". For example, "social media luxury preference" and "high credit income" are associated through graph edge weights, thus solving the "data island" problem.

[0081] In one embodiment, the feature analysis unit in the acquisition module includes:

[0082] a time series segmentation subunit, configured to obtain transaction sequence data and original credit information data based on the credit transaction data, and to obtain multiple transaction windows and transaction category groups based on the transaction sequence data;

[0083] a capital analysis subunit, configured to obtain corresponding statistical characteristic data according to each transaction window, and obtain cash flow health according to the statistical characteristic data;

[0084] a cycle acquisition subunit, configured to perform Fourier transform on each of the transaction category groups to obtain a corresponding consumption cycle;

[0085] a credit analysis subunit, configured to obtain credit characteristic data based on the original credit data, and to obtain a credit risk level based on the credit characteristic data;

[0086] The risk quantification subunit is used to obtain a risk characteristic value according to the credit risk level and the cash flow health.

[0087] As the functions of each sub-unit mentioned above, the present invention obtains transaction sequence data and original credit data from credit transaction data, wherein transaction sequence data refers to a collection of transaction records arranged in chronological order, each record contains fields such as transaction amount, timestamp, transaction type, etc., and original credit data refers to unprocessed credit data directly obtained from a credit agency, and the transaction sequence data is processed by sliding window technology to obtain multiple transaction windows and transaction category groupings, wherein the transaction window refers to a fixed time range defined in the time series data (such as the past 30 days, the past 6 months), which is used to intercept local data for statistical analysis, and the transaction category grouping refers to the transaction data according to the nature or scenario of the transaction. Perform classification aggregation, such as dividing by categories such as "catering," "shopping," and "mortgages," and obtain statistical feature data for each window. Statistical feature data refers to the characteristic values ​​generated after statistical calculations on the original data, including statistics such as mean, variance, and maximum expenditure. Based on this data, a time series convolutional network is used to obtain cash flow health. Cash flow health refers to the quality of a customer's cash flow status assessed based on transaction sequence data. A Fourier transform is performed on the transaction category groups to convert the time domain signal into a frequency domain signal. The inverse of the frequency corresponding to the peak in the spectrum can be used as the consumption cycle. The consumption cycle refers to the time interval between repeated consumption behaviors of the same type by the customer.

[0088] Next, credit feature data is obtained from the original credit data. Credit feature data refers to structured indicators extracted from the original credit data, including debt ratio, number of overdue payments, provident fund base, etc. These data are standardized and mapped to credit risk levels through the rule engine. Credit risk level refers to the result of grading personal credit risk (such as AAA to D) through credit feature data and scoring models, and the credit risk level and cash flow health are standardized and weighted and integrated according to the business scenario weights to obtain the risk feature value. The risk feature value refers to a multidimensional indicator that quantifies the customer's credit risk.

[0089] In one embodiment, the constraint generation module includes:

[0090] a policy parsing unit, configured to obtain policy constraint rules and interest rate fluctuation indicators based on the financial market data, wherein the policy constraint rules include loan limit rules, interest rate control rules, and repayment time rules, and to obtain a regulatory interest rate floor, static interest rate rules, and dynamic interest rate rules based on the interest rate control rules;

[0091] a market sensing unit, configured to obtain an interest rate volatility and a volatility threshold according to the interest rate volatility indicator, and obtain a market adjustment value according to the interest rate volatility and the volatility threshold;

[0092] An interest rate restriction unit, configured to obtain a loan interest rate floor based on the interest rate volatility, the fluctuation threshold, the regulatory interest rate floor, the static interest rate rule, and the dynamic interest rate rule;

[0093] a credit limit unit, configured to obtain a solvency assessment value based on the credit transaction data, and obtain a basic loan amount and a loan multiple amount based on the solvency assessment value and the loan limit rule;

[0094] The repayment restriction unit is used to obtain the maximum loan amount according to the basic loan amount, the loan multiple amount and the market adjustment value, and to obtain the lower limit of the repayment time according to the repayment time rule and the market adjustment value.

[0095] As the functions of each unit mentioned above, the present invention uses NLP technology to parse financial market data in real time, and extracts structured rules to obtain policy constraint rules, wherein policy constraint rules refer to mandatory rules that are updated in real time by financial regulatory agencies or market managers based on the current economic environment and policy orientations, and are used to regulate the business operations of financial institutions, including loan limit rules, interest rate control rules and repayment time rules, wherein loan limit rules refer to the upper limit standard of the borrower's loan amount set by regulatory agencies or financial institutions, interest rate control rules refer to mandatory regulations on the pricing range and adjustment mechanism of loan interest rates, and repayment time rules refer to specific regulations on loan repayment methods, terms, overdue penalties, etc., and obtain regulatory benefits according to interest rate control rules. Interest rate floors, static interest rate rules, and dynamic interest rate rules. The regulatory interest rate floor refers to the statutory minimum value of the loan interest rate stipulated by the regulatory agency, and financial institutions must not set prices below this value. Static interest rate rules refer to pricing rules that keep the interest rate fixed during the loan term, usually based on the benchmark interest rate or market interest rate plus a certain percentage. For example: the loan interest rate floor = the regulatory floor + volatility × sensitivity coefficient. Dynamic interest rate rules refer to pricing mechanisms where interest rates are regularly adjusted based on market indicators (such as LPR and treasury bond yields). For example: the loan interest rate floor = the regulatory floor + 50 basis points + volatility × 10 basis points. This solution uses NLP to achieve a second-level conversion from "policy release → rule digitization", thereby solving the problem of traditional solutions relying on manual policy interpretation and delayed compliance response.

[0096] The interest rate volatility index is collected based on financial market data. The interest rate volatility index refers to data reflecting the recent fluctuation range and frequency of financial market interest rates, such as loan quotation rates and fluctuation thresholds. The fluctuation threshold refers to a pre-set critical value of interest rate fluctuations, which is used to determine whether market fluctuations have reached a level that requires adjustment of business strategies. The yield change series is calculated based on the loan quotation rates, and the standard deviation and time factors (such as ) is annualized to obtain the interest rate volatility. This solution calculates the real-time interest rate volatility so that compliance constraints can dynamically respond to market changes (such as automatically raising the interest rate floor when volatility rises suddenly), thereby solving the problem that traditional solutions use static interest rate constraints (such as a fixed interest rate floor) without considering market volatility risks. Then, a linear mapping model is used to combine the interest rate volatility and the volatility threshold to calculate the market adjustment value. The market adjustment value refers to the proportional value of dynamically adjusting loan-related parameters (such as loan amount and interest rate). Then, it is determined whether the interest rate volatility is greater than the volatility threshold, and the dynamic adjustment logic is triggered: if the interest rate volatility is not greater than the volatility threshold, The loan interest rate floor is obtained based on the regulatory interest rate floor and static interest rate rules. If the interest rate volatility is greater than the volatility threshold, the loan interest rate floor is obtained based on the regulatory interest rate floor and dynamic interest rate rules. The loan interest rate floor refers to the lowest interest rate that financial institutions can set when issuing loans. Traditional solutions lack a market risk linkage mechanism and adopt a "one-size-fits-all" interest rate rule. This makes it impossible to differentiate management and control during stable and turbulent market periods, which may lead to risk exposure when interest rates fluctuate sharply (for example, the low interest rate floor is still used when volatility rises sharply). This solution triggers dynamic adjustments through thresholds to achieve adaptive control of "the higher the risk, the stricter the constraints";

[0097] Then, sequential data such as income flow, consumption records, and debt repayment are extracted from the credit transaction data, and these data are processed through scoring models (such as logistic regression and machine learning models) to obtain a debt repayment ability assessment value, where the debt repayment ability assessment value refers to a quantitative indicator that assesses the borrower's ability to repay the loan on time based on the borrower's credit transaction data (such as income, debt, and repayment records). Then, the basic loan amount is obtained according to the loan limit rules, where the basic loan amount refers to the loan amount benchmark value preliminarily determined based on the loan limit rules, and the loan multiple is obtained according to the debt repayment ability assessment value. The loan multiple amount is obtained by multiplying the loan multiple and the basic loan amount, where the loan multiple amount refers to the upper limit of the loan amount calculated according to the loan multiple. At the same time, according to the formula " ” Calculate the market adjustment amount, where Indicates the market adjustment amount, Indicates the basic loan amount. Represents the market adjustment value, and then the maximum of the loan base amount, loan multiple amount and market adjustment amount is used as the maximum loan amount, where the maximum loan amount refers to the maximum loan amount that the borrower can finally obtain. Finally, the basic repayment time is obtained according to the repayment time rule, and according to the formula " ” Calculate the lower limit of repayment time, where Indicates the minimum repayment time. Indicates the basic repayment time, Represents the market-adjusted value, where the lower limit of repayment time refers to the time limit within which the borrower must complete repayment. The lower limit of loan interest rate, the maximum loan amount, and the lower limit of repayment time constitute compliance constraint data. Through the above steps, this solution has achieved an upgrade from "static compliance" to "dynamic risk control", solving problems such as slow policy response, disregard for market risks, and extensive individual assessment in traditional solutions, and building a real-time, intelligent, and flexible compliance constraint generation system.

[0098] In one embodiment, the strategy synthesis module includes:

[0099] an element generation unit, configured to obtain credit element parameters based on the compliance constraint data and the customer feature vector, wherein the credit element parameters include a plurality of recommended loan interest rates, recommended repayment periods, and recommended loan amounts;

[0100] a combining unit, configured to arrange and combine a plurality of the recommended loan interest rates, recommended repayment periods, and recommended loan amounts to obtain a plurality of marketing strategy combinations;

[0101] a matching degree obtaining unit, configured to obtain a single-dimensional matching degree based on the credit transaction data, the social media data, and the credit factor parameters;

[0102] a weight analysis unit, configured to obtain preference feedback data based on the customer interaction data, and perform correlation analysis on the credit factors and the preference feedback data to obtain a plurality of single-dimensional weights;

[0103] a preference evaluation unit, configured to obtain a customer preference value for each of the marketing strategy combinations based on a plurality of the single-dimensional matching degrees and the corresponding single-dimensional weights;

[0104] A screening unit is used to screen the plurality of marketing strategy combinations according to the customer preference value to obtain a credit marketing strategy.

[0105] As for the functions of each unit mentioned above, the present invention combines the customer feature vector including demand feature value, behavior feature value and risk feature value with the compliance constraint data including the lower limit of loan interest rate, maximum loan amount and lower limit of repayment time, so as to obtain credit factor parameters, wherein the credit factor parameters refer to the parameter ranges corresponding to the core variables (such as loan interest rate, repayment period and loan amount) that affect loan decision-making and customer experience in credit business, including loan interest rate range, repayment period range and loan amount range. The specific calculation method is: loan interest rate range = [loan interest rate lower limit, loan interest rate lower limit * (1 + risk feature value)]; repayment period range = [repayment time lower limit, repayment time lower limit * (1 + behavior feature value)]; loan amount range = [loan maximum amount * demand feature value, loan The maximum loan amount]; and further, according to the loan interest rate range, repayment cycle range and loan amount range, corresponding multiple loan recommended interest rates, repayment recommended cycles and loan recommended amounts can be obtained respectively. Among them, the loan recommended interest rate refers to the loan interest rate recommended for customers, the repayment recommended cycle refers to the repayment frequency recommended for customers, and the loan recommended amount refers to the loan amount recommended for customers. Traditional solutions are based only on credit data pricing and do not take into account customers' real-time consumption intentions (such as car purchase needs expressed on social media), resulting in interest rate and amount settings being out of touch with actual needs. This solution achieves a dual drive of "risk pricing + demand pricing" through cross-modal fusion (compliance constraints + consumption intentions), solving the problem of "disconnection between pricing and demand". For example, customers with high consumption intentions and low risks can obtain lower interest rates and higher amounts.

[0106] Next, multiple recommended loan interest rates, recommended repayment periods, and recommended loan amounts are permuted and combined to generate multiple marketing strategy combinations. These combinations are personalized marketing strategies designed for each customer. This systematic permutation and combination approach increases strategy diversity from "single digits" to "hundreds," resolving the problem of "low matching degree" in traditional one-size-fits-all strategies.

[0107] Single-dimensional matching is derived from credit transaction data, social media data, and credit factor parameters. Single-dimensional matching measures the degree of alignment between the currently recommended credit factor value and the customer's historical preference profile. Preference feedback data is then obtained based on customer interaction data. Preference feedback data refers to actual customer responses to recommended credit factors (such as interest rates and credit limits). Correlation analysis (e.g., the Pearson correlation coefficient) is performed between the credit factors and preference feedback data to generate multiple single-dimensional weights. Single-dimensional weights represent the importance coefficient assigned to each single-dimensional matching. Customer preference values ​​are then derived for each marketing strategy combination based on loan interest rate matching, repayment period matching, loan limit matching, and the corresponding single-dimensional weights. Customer preference values ​​quantify the customer's preference for a specific credit factor. Finally, the marketing strategy combination corresponding to the highest customer preference value is selected as the credit marketing strategy. This method, through preference value ranking and diversity control, enables "globally optimal" strategy screening, upgrading strategy customization from "experience-driven" to "data-driven," helping to address the industry pain point of low credit strategy matching.

[0108] In one embodiment, the matching degree acquisition unit in the strategy synthesis module includes:

[0109] An interval acquisition subunit, configured to acquire a preference parameter interval based on the credit transaction data;

[0110] a preference acquisition subunit, configured to acquire historical preference features based on the social media data, and acquire credit parameter preference values ​​based on the historical preference features and the preference parameter intervals, wherein the credit parameter preference values ​​include an interest rate preference value, a repayment preference period, and a credit limit preference value;

[0111] The preference mapping subunit is used to obtain the corresponding single-dimensional matching degree according to the credit marketing strategy, interest rate preference value, repayment preference period and loan amount preference value, wherein the single-dimensional matching degree includes multiple loan interest rate matching degrees, repayment period matching degrees and loan amount matching degrees.

[0112] As for the functions of each unit described above, the present invention extracts the customer's preference parameter interval from credit transaction data. The preference parameter interval refers to the customer's preference range for a certain type of credit parameter (such as interest rate, credit limit, and repayment period) in historical credit behavior, usually expressed in the form of an interval, for example: interest rate preference (such as choosing an interest rate of 5.0% to 5.5% for the past three loans), credit limit habit (applying for a credit limit between 100,000 and 150,000), and uses the BERT-wwm model to parse text information in social media data, such as semantics such as "the interest rate is too high" and "it would be nice if the credit limit could reach 200,000", thereby generating historical preference features. The historical preference features refer to the stable preference data shown by customers in past credit behavior or consulting behavior, which are used to characterize customer demand patterns;

[0113] The credit parameter preference value is obtained by combining the historical preference characteristics and the preference parameter range. The credit parameter preference value refers to the specific credit parameter preferred by the customer, including the interest rate preference value, the repayment period preference value, and the loan amount preference value. Then, the single-dimensional matching degree is obtained based on the interest rate preference value, the repayment period preference value, the loan amount preference value, and the credit factor parameters. The single-dimensional matching degree refers to the degree of consistency between the currently recommended credit factor value and the customer's historical preference characteristics, including the loan interest rate matching degree, the repayment period matching degree, and the loan amount matching degree. The calculation formula is: ;in, Indicates the loan interest rate matching degree, Indicates the recommended loan interest rate, Indicates the loan interest rate preference value; if the recommended repayment period is the same as the preferred repayment period, the repayment period matching degree is 1; if the recommended repayment period is a multiple of the preferred repayment period, the repayment period matching degree is 0.7; if there is some other relationship between the recommended repayment period and the preferred repayment period, the repayment period matching degree is 0.3; ,in, Indicates the loan amount matching degree, Indicates the recommended loan amount. Indicates the loan amount preference value.

[0114] In one embodiment, the evaluation module includes:

[0115] a scenario matching unit, configured to obtain a marketing scenario tag according to the credit marketing strategy, and obtain historical matching data according to the marketing scenario tag and the customer interaction data, wherein the historical matching data includes historical strategy parameters and interaction feedback data;

[0116] a multidimensional analysis unit, configured to obtain a multidimensional matching index based on the credit marketing strategy and the historical strategy parameters, wherein the multidimensional matching index includes parameter adaptability, scenario adaptability, and repayment cycle adaptability;

[0117] A strategy evaluation unit, configured to obtain a strategy matching degree based on the parameter adaptability, scenario adaptability, and repayment cycle adaptability;

[0118] An interaction prediction unit, configured to obtain the number of interactive behaviors, the number of users reached, the number of conversion behaviors, and the interaction time based on the interactive feedback data, and to obtain an expected interaction probability based on the number of interactive behaviors, the number of users reached, and the interaction time;

[0119] The accuracy acquisition unit is used to obtain the expected conversion rate based on the number of interactive behaviors, the number of conversion behaviors and the interaction time, and to obtain the marketing accuracy based on the strategy matching degree, the expected conversion rate and the expected interaction probability.

[0120] As for the functions of each unit mentioned above, the present invention maps the credit marketing strategy to the marketing scenario label, wherein the marketing scenario label refers to a classification label used to identify the specific scenario characteristics targeted by the credit marketing strategy solution, for example: "small consumer loan-regular interest rate" scenario, based on the marketing scenario label, extracts similar strategy data from the customer interaction data to obtain historical matching data, wherein the historical matching data refers to the data storing the correspondence between "strategy parameters" and "user feedback" in historical credit marketing activities, including historical strategy parameters and interactive feedback data, wherein the historical strategy parameters refer to the specific strategy parameters actually adopted in the historical credit marketing activities under similar marketing scenario labels, and through the historical strategy parameters, the average loan interest rate, historical scenario label and the average loan amount, the interactive feedback data refers to the behavioral data generated by the user in the process of interacting with the credit marketing activities under the similar marketing scenario label, including operation records such as clicks, consultations, application submissions, and rejections. This solution establishes a historical association between strategy and effect through the scenario label, thereby combining historical data to predict the promotion effect of the credit marketing strategy, thereby solving the problem of repeated push of invalid strategies by traditional methods;

[0121] Then, based on the credit marketing strategy and historical strategy parameters, multi-dimensional matching indicators are obtained, including sub-indicators such as parameter adaptation, scenario adaptation, and repayment cycle adaptation, which are used to measure the fit between the current strategy and the historical successful strategy. The specific acquisition method is: according to the credit marketing strategy, customized loan interest rates, customized repayment cycles, and customized loan amounts are obtained for customers. Then, the parameter adaptation is obtained based on the ratio of the average loan interest rate to the customized loan interest rate. Based on the semantic matching of scenario labels (such as the semantic distance between "car loan" and "car installment"), the Word2Vec model is used to calculate the Scenario adaptability: The repayment cycle adaptability is obtained based on the ratio of the average repayment cycle to the customized repayment cycle. Parameter adaptability measures the degree of match between the current credit strategy parameters and the historical average parameters. Scenario adaptability assesses the similarity between the current marketing scenario and historically successful scenarios. Repayment cycle adaptability measures the degree of fit between the current repayment cycle and the user's historically preferred cycles. Strategy matching is then obtained through a weighted summation of parameter adaptability, scenario adaptability, and repayment cycle adaptability. Strategy matching reflects the overall fit between the current strategy and similar historical strategies.

[0122] Then, based on the interactive feedback data, we can obtain the number of interactive behaviors, the number of users reached, the number of conversion behaviors and the interaction time. Among them, the number of interactive behaviors refers to the number of effective interactive operations generated by users in credit marketing activities, such as the number of times they click on marketing links, consult customer service, view loan details, etc. The number of users reached refers to the number of target users successfully reached by credit marketing activities, such as the number of users actually delivered through channels such as SMS and APP push. The number of conversion behaviors refers to the number of target conversion actions completed by users in marketing activities, such as the number of core conversion behaviors such as submitting loan applications and signing contracts. The interaction time refers to the time dimension data of users' interaction with credit marketing content, and according to the formula " ” Calculate the expected interaction probability, where represents the expected interaction probability, Indicates the number of interactive behaviors. Indicates the number of times users are reached. represents the second attenuation coefficient, Represents the time interval, and the time interval refers to the interval between the interaction time and the current time. The expected interaction probability refers to the possibility of predicting the user's interactive behavior under the current strategy. According to the formula " ” Calculate the expected conversion rate, where represents the expected conversion rate, Indicates the number of conversion behaviors. Indicates the number of interactive behaviors. represents the second attenuation coefficient, represents a time interval, where the expected conversion rate is the predicted probability of a user completing a core conversion (such as applying for a loan) under the current strategy. Finally, a logistic regression model based on customer interaction data is used to determine marketing accuracy based on strategy fit, expected conversion rate, and expected interaction probability. Traditional methods rely on a single metric (such as AUC) to evaluate models, failing to distinguish between "strategy fit issues" and "execution effectiveness issues." For example, a model with an AUC of 0.75 but a strategy fit of only 0.5 is actually a strategy design issue rather than a model defect. This solution directly identifies the root cause of the problem by breaking down the metrics.

[0123] In one embodiment, the judgment module includes:

[0124] a parameter optimization unit, configured to obtain a matching threshold value, and obtain a credit parameter to be optimized based on the matching threshold value and the multi-dimensional matching index;

[0125] a gradient analysis unit, configured to obtain an accuracy gradient based on the multidimensional matching index and the customer preference value;

[0126] a step size configuration unit, configured to obtain an initial adjustment step size based on the financial market data, and obtain a step size constraint upper limit based on the policy constraint rules;

[0127] a dynamic adjustment unit, configured to obtain an adaptive step size according to the initial adjustment step size, the step size constraint upper limit, and the accuracy gradient, and to obtain a credit parameter adjustment value according to the credit parameter to be optimized and the adaptive step size;

[0128] An iterative unit is used to obtain a credit marketing adjustment strategy based on the credit parameter adjustment value and the credit marketing strategy, and return the credit marketing adjustment strategy as the credit marketing strategy to the step of obtaining marketing accuracy based on the credit marketing strategy and the customer interaction data.

[0129] As for the functions of the above-mentioned units, the present invention compares the parameter matching degree, scenario matching degree, and repayment cycle matching degree with the matching degree threshold respectively, and selects the credit parameters with values ​​less than the matching degree threshold as the credit parameters to be optimized. The matching degree threshold refers to a pre-set critical value used to judge the degree of matching between the credit parameters and factors such as customer needs and market environment. The credit parameters to be optimized refer to the credit-related parameters determined to need to be adjusted to improve marketing accuracy. This method of screening by threshold can not only optimize parameters with insufficient matching degree, but also solve the problem of "unclear optimization direction" in traditional methods.

[0130] Then, based on the relationship between the multi-dimensional matching index and the customer preference value, the accuracy function f(x) is constructed, where x is the credit parameter vector (such as loan interest rate, loan amount, repayment period), and the gradient vector of each credit parameter vector is calculated separately to obtain the accuracy gradient. The accuracy gradient is used to measure the sensitivity of adjusting the credit parameter to improving the accuracy. According to the gradient vector, the gradient optimization direction can be obtained, for example: the positive gradient direction (such as >0): Increasing the interest rate can improve accuracy; negative gradient direction (such as <0): Lowering the interest rate can improve accuracy;

[0131] Then, the initial adjustment step is obtained based on the financial market data, where the initial adjustment step refers to the basic amplitude of each adjustment initially set, and the step constraint upper limit is obtained based on the policy constraint rules, where the step constraint upper limit refers to the maximum allowable value of the adjustment step set according to the policy constraint rules (such as regulatory requirements). Through dynamic step constraints, it is ensured that the optimization conforms to both market laws and regulatory requirements. For example, the system can automatically identify the current market interest rate trend to avoid excessive interest rate cuts during the interest rate rise period, and then according to the formula " ” Calculate the adaptive step size, where represents the adaptive step size, represents the initial adjustment step size, Represents the gradient vector, where the adaptive step size refers to the dynamically calculated optimal adjustment amplitude. The credit parameter adjustment value is then obtained by combining the gradient optimization direction, the credit parameter to be optimized, and the adaptive step size. The credit parameter adjustment value refers to the value obtained by adjusting the credit parameter to be optimized according to the adaptive step size. For example, if the interest rate gradient is -0.05 (a negative gradient indicates that the interest rate needs to be lowered), the adaptive step size is 0.08%, and the customized loan interest rate is 5.35%, then the interest rate adjustment value is 5.35%-0.08%×0.05=5.346%. This method of adjustment through adaptive step size can use large step sizes for rapid iteration when the gradient is large, and use small step sizes for fine adjustment when the gradient is small.

[0132] Finally, the credit parameter adjustment value is used to replace the credit parameter to be optimized in the credit marketing strategy to obtain the credit marketing adjustment strategy, where the credit marketing adjustment strategy refers to the new credit marketing strategy formed after adjustment on the basis of the original credit marketing strategy, and the credit marketing adjustment strategy is returned to the step of "obtaining marketing accuracy based on credit marketing strategy and customer interaction data" as the credit marketing strategy, which is conducive to the continuous evolution of the strategy through cyclic iteration.

[0133] The present invention also provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and when the memory executes the computer program, it is used to manage the operation of the above-mentioned AI credit marketing management system based on multimodal data fusion.

[0134] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is used to manage the operation of the above-mentioned AI credit marketing management system based on multimodal data fusion.

[0135] Those skilled in the art will appreciate that all or part of the processes in the above-described embodiment systems can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the above-described embodiments of each module. Any reference to memory, storage, database, or other media provided herein and used in the embodiments may 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. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double-speed SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct RAMbus dynamic RAM (DRDRAM), and RAMbus dynamic RAM (RDRAM).

[0136] It should be noted that, in this document, the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, apparatus, article, or method comprising a series of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, apparatus, article, or method. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, apparatus, article, or method comprising the element.

[0137] The above description is only a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made by using the contents of the present invention description and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. AI credit marketing management system based on multimodal data fusion, characterized by: include: An acquisition module is used to acquire multimodal data, including social media data, credit transaction data, financial market data, and customer interaction data, and obtain customer feature vectors based on the social media data and credit transaction data; Constraint generation module, used to obtain compliance constraint data based on financial market data and credit transaction data; The strategy synthesis module is used to obtain credit marketing strategies based on compliance constraint data, customer feature vectors, and credit transaction data. The strategy synthesis module includes: An element generation unit is used to obtain credit element parameters based on compliance constraint data and customer feature vectors. The credit element parameters include multiple recommended loan interest rates, recommended repayment periods, and recommended loan amounts. A combination unit is used to arrange and combine multiple recommended loan interest rates, recommended repayment periods, and recommended loan amounts to obtain multiple marketing strategy combinations; A matching degree acquisition unit, used to obtain a single-dimensional matching degree based on credit transaction data, social media data, and credit factor parameters; A weight analysis unit is used to obtain preference feedback data based on customer interaction data, and to perform correlation analysis on credit factors and preference feedback data to obtain multiple single-dimensional weights; A preference evaluation unit is used to obtain the customer preference value of each marketing strategy combination based on multiple single-dimensional matching degrees and corresponding single-dimensional weights; A screening unit, used to screen multiple marketing strategy combinations according to customer preference values ​​to obtain a credit marketing strategy; The evaluation module is used to obtain multi-dimensional matching indicators and interactive feedback data based on credit marketing strategies and customer interaction data, and to determine marketing accuracy based on these multi-dimensional matching indicators and interactive feedback data. The evaluation module includes: A scenario matching unit is used to obtain marketing scenario labels based on the credit marketing strategy, and to obtain historical matching data based on the marketing scenario labels and customer interaction data. The historical matching data includes historical strategy parameters and interaction feedback data. A multi-dimensional analysis unit is used to obtain multi-dimensional matching indicators based on credit marketing strategies and historical strategy parameters. The multi-dimensional matching indicators include parameter adaptability, scenario adaptability, and repayment cycle adaptability. A strategy evaluation unit is used to obtain strategy matching based on parameter adaptability, scenario adaptability, and repayment cycle adaptability; An interaction prediction unit is used to obtain the number of interactive behaviors, the number of users reached, the number of conversion behaviors, and the interaction time based on the interactive feedback data, and to obtain the expected interaction probability based on the number of interactive behaviors, the number of users reached, and the interaction time; The precision acquisition unit is used to obtain the expected conversion rate based on the number of interactive behaviors, the number of conversion behaviors, and the interaction time, and to obtain the marketing precision based on the strategy matching degree, the expected conversion rate, and the expected interaction probability; A judgment module is used to judge whether the marketing accuracy is greater than a preset threshold; If the marketing accuracy is greater than the preset threshold, the credit marketing strategy is determined to meet customer needs; If the marketing accuracy is not greater than the preset threshold, it is determined that the credit marketing strategy does not meet customer needs, and the credit marketing strategy is adjusted according to the multi-dimensional matching indicators until the credit marketing strategy meets customer needs.

2. The AI ​​credit marketing management system based on multimodal data fusion according to claim 1 is characterized in that: The acquisition module includes: a feature extraction unit, configured to obtain consumption intention text data and consumption intention behavior data based on the social media data, and to obtain a plurality of consumption intention targets, consumption intention intensities, and consumption intention timestamps based on the consumption intention text data; A feature analysis unit, configured to obtain consumption cycle and risk feature values ​​based on the credit transaction data; a demand acquisition unit, configured to acquire the real-time intention strength of each consumption intention target according to the consumption intention strength and the consumption intention timestamp, and acquire the demand characteristic value of each consumption intention target according to the real-time intention strength and the consumption cycle; a behavior encoding unit, configured to obtain a behavior feature value of each consumption intention target based on the consumption intention behavior data and the consumption intention timestamp; The feature fusion unit is used to obtain a customer feature vector based on a plurality of demand feature values, behavior feature values ​​and risk feature values.

3. The AI ​​credit marketing management system based on multimodal data fusion according to claim 2 is characterized in that: The feature analysis unit in the acquisition module includes: a time series segmentation subunit, configured to obtain transaction sequence data and original credit information data based on the credit transaction data, and to obtain multiple transaction windows and transaction category groups based on the transaction sequence data; a capital analysis subunit, configured to obtain corresponding statistical characteristic data according to each transaction window, and obtain cash flow health according to the statistical characteristic data; a cycle acquisition subunit, configured to perform Fourier transform on each of the transaction category groups to obtain a corresponding consumption cycle; a credit analysis subunit, configured to obtain credit characteristic data based on the original credit data, and to obtain a credit risk level based on the credit characteristic data; The risk quantification subunit is used to obtain a risk characteristic value according to the credit risk level and the cash flow health.

4. The AI ​​credit marketing management system based on multimodal data fusion according to claim 1 is characterized in that: The constraint generation module includes: a policy parsing unit, configured to obtain policy constraint rules and interest rate fluctuation indicators based on the financial market data, wherein the policy constraint rules include loan limit rules, interest rate control rules, and repayment time rules, and to obtain a regulatory interest rate floor, static interest rate rules, and dynamic interest rate rules based on the interest rate control rules; a market sensing unit, configured to obtain an interest rate volatility and a volatility threshold according to the interest rate volatility indicator, and obtain a market adjustment value according to the interest rate volatility and the volatility threshold; An interest rate restriction unit, configured to obtain a loan interest rate floor based on the interest rate volatility, the fluctuation threshold, the regulatory interest rate floor, the static interest rate rule, and the dynamic interest rate rule; a credit limit unit, configured to obtain a solvency assessment value based on the credit transaction data, and obtain a basic loan amount and a loan multiple amount based on the solvency assessment value and the loan limit rule; The repayment restriction unit is used to obtain the maximum loan amount according to the basic loan amount, the loan multiple amount and the market adjustment value, and to obtain the lower limit of the repayment time according to the repayment time rule and the market adjustment value.

5. The AI ​​credit marketing management system based on multimodal data fusion according to claim 1 is characterized in that: The matching degree acquisition unit in the strategy synthesis module includes: An interval acquisition subunit, configured to acquire a preference parameter interval based on the credit transaction data; a preference acquisition subunit, configured to acquire historical preference features based on the social media data, and acquire credit parameter preference values ​​based on the historical preference features and the preference parameter intervals, wherein the credit parameter preference values ​​include an interest rate preference value, a repayment preference period, and a credit limit preference value; The preference mapping subunit is used to obtain the corresponding single-dimensional matching degree according to the credit marketing strategy, interest rate preference value, repayment preference period and loan amount preference value, wherein the single-dimensional matching degree includes multiple loan interest rate matching degrees, repayment period matching degrees and loan amount matching degrees.

6. The AI ​​credit marketing management system based on multimodal data fusion according to claim 4 is characterized in that: The judgment module includes: a parameter optimization unit, configured to obtain a matching threshold value, and obtain a credit parameter to be optimized based on the matching threshold value and the multi-dimensional matching index; a gradient analysis unit, configured to obtain an accuracy gradient based on the multidimensional matching index and the customer preference value; a step size configuration unit, configured to obtain an initial adjustment step size based on the financial market data, and obtain a step size constraint upper limit based on the policy constraint rules; a dynamic adjustment unit, configured to obtain an adaptive step size according to the initial adjustment step size, the step size constraint upper limit, and the accuracy gradient, and to obtain a credit parameter adjustment value according to the credit parameter to be optimized and the adaptive step size; An iterative unit is used to obtain a credit marketing adjustment strategy based on the credit parameter adjustment value and the credit marketing strategy, and return the credit marketing adjustment strategy as the credit marketing strategy to the step of obtaining marketing accuracy based on the credit marketing strategy and the customer interaction data.

7. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the memory executes the computer program, it is used to manage the operation of the AI ​​credit marketing management system based on multimodal data fusion according to any one of claims 1 to 6.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, it is used to manage the operation of the AI ​​credit marketing management system based on multimodal data fusion according to any one of claims 1 to 6.

Images