A product recommendation method, device, medium, and product

Abstract

This invention discloses a product recommendation method, device, medium, and product applied in the fintech field. The method includes: acquiring target users' product needs; generating structured data based on these needs using a large-scale model; transforming the structured data into a demand feature vector; homomorphically encrypting the demand feature vector to generate an encrypted demand feature vector; acquiring candidate products; extracting features from the candidate products to generate product feature vectors; homomorphically encrypting the product feature vectors to generate an encrypted product feature vector; and performing federated computation based on the encrypted demand feature vector and the encrypted product feature vector to determine the recommended product. The large-scale model transforms unstructured user needs into structured data, accurately extracting core user needs and improving recommendation accuracy. Homomorphic encryption and federated computation protect data privacy, enabling secure matching without disclosing original data, thus improving recommendation security and user experience.

Description

Technical Field

[0001] This invention relates to the field of financial technology, and in particular to a product recommendation method, device, medium, and product. Background Technology

[0002] Against the backdrop of rapid development in fintech, intelligent product recommendation technology has become an important means for financial institutions to improve service efficiency and optimize customer experience. Relying on intelligent technology to achieve accurate product recommendations can better match customers' diverse card application and usage needs. The intelligence, personalization, and security of this technology have also become core directions for industry development.

[0003] Currently, recommendation services for financial products such as credit cards mostly rely on traditional rule engines. These engines collect relevant user information and then match products with users based on preset fixed rules. Some technologies combine simple data analysis methods to optimize the recommendation logic, but overall they are still mainly based on linear matching. In terms of data processing, they often need to directly acquire and store various types of user information, then compare product features with user needs based on the extracted information, and finally output recommended products based on the comparison results.

[0004] Existing technologies offer recommendations based on a single dimension, failing to integrate dynamic activity data from financial institutions in real time. This results in a lack of flexibility and timeliness in the recommendations. Furthermore, existing technologies require the collection of sensitive and private information such as user income and occupation, which can easily lead to privacy leaks during the acquisition, storage, and use of data, thus failing to meet the relevant requirements for compliant data use. Summary of the Invention

[0005] This invention provides a product recommendation method, device, medium, and product, which solves the technical problems in traditional product recommendation such as the difficulty in structuring natural language requirements, insufficient data privacy protection, and the difficulty in balancing recommendation accuracy and security.

[0006] According to one aspect of the present invention, a product recommendation method is provided, the method comprising:

[0007] Obtain product requirements from target users, generate structured data based on product requirements using a large model, transform the structured data into requirement feature vectors, perform homomorphic encryption on the requirement feature vectors, and generate encrypted requirement feature vectors.

[0008] Obtain candidate products, extract features from candidate products, generate product feature vectors, perform homomorphic encryption on product feature vectors, and generate encrypted product feature vectors.

[0009] Federated computation is performed based on the encrypted demand feature vector and the encrypted product feature vector to determine recommended products.

[0010] Optionally, structured data can be generated based on product requirements through a large model, including: obtaining the first prompt word; controlling the large model to perform intent recognition, key information capture, and sensitive information encryption processing on the product requirements based on the first prompt word; extracting key requirements; obtaining supplementary requirements of target users through the large model; and integrating key requirements and supplementary requirements in a specified format to generate structured data.

[0011] The advantages of this setup are: accurately identifying and extracting core user product needs, encrypting and protecting sensitive information, improving the dimensions of demand information, generating standardized structured data, and providing a precise and standardized data foundation for subsequent feature vector transformation.

[0012] Optionally, supplementary needs of the target user can be obtained through a large model, including: obtaining a second prompt word; using the second prompt word to control the large model to generate a question-and-answer path based on product needs in combination with a decision tree; and conducting natural language dialogue with the target user based on the question-and-answer path to obtain supplementary needs input by the target user.

[0013] The advantages of this setup are: it accurately generates targeted follow-up questions, improves user needs information, enhances the comprehensiveness and accuracy of needs collection, and provides a complete basis for subsequent data processing.

[0014] Optionally, the structured data is transformed into a demand feature vector, including: transforming the structured data into numerical vectors of each specified dimension according to a preset mapping rule, wherein the specified dimensions include strength features, preference features, scenario features, usage features and rights features; encoding the numerical vectors according to a specified format to obtain the encoded vectors of each specified dimension, and concatenating the encoded vectors of each specified dimension in sequence to obtain the demand feature vector.

[0015] The advantages of this setup are: to achieve the quantitative transformation of user needs, to unify the data calculation format, to integrate multi-dimensional demand information, and to provide a standardized, computable vector foundation for subsequent encryption and federated computing.

[0016] Optionally, federated computation is performed based on the encrypted demand feature vector and the encrypted product feature vector to determine recommended products. This includes: performing a dot product operation on the encrypted demand feature vector and the encrypted product feature vector to obtain the dot product result; decrypting the dot product result to generate a decrypted dot product result; obtaining a preset product vector magnitude and calculating the demand vector magnitude corresponding to the demand feature vector; calculating the product vector magnitude and the demand vector magnitude, and calculating the ratio of the decrypted dot product result to the product as the similarity between the demand feature vector and the product feature vector; calculating an activity gain factor, and combining the similarity and the activity gain factor to comprehensively determine recommended products.

[0017] The advantage of this setup is that it enables accurate similarity calculations in an encrypted state, balancing data privacy protection with calculation accuracy.

[0018] Optionally, calculate the activity gain factor, and combine the similarity with the activity gain factor to determine the recommended products, including: determining the activity validity period of each candidate product, determining the activity gain factor based on the activity validity period; multiplying the activity gain factor and the similarity to obtain the recommendation score of each candidate product; sorting each candidate product in descending order of recommendation score to obtain a recommendation list, and selecting a specified number of candidate products from the recommendation list as recommended products.

[0019] The advantages of this setup are: it quantifies the overall recommendation value of products, enables the scientific ranking and filtering of recommendation results, and improves the accuracy and scenario adaptability of product recommendations.

[0020] Optionally, the method also includes: obtaining product information and corresponding structured data of the recommended products, wherein the product information includes product name, application conditions, product validity period, product rights information and product additional information; constructing a product rights list based on the product rights information, and generating personalized prompts based on the structured data; and integrating the product information, product rights list and personalized prompts to form a structured report.

[0021] The advantages of this setup are: it integrates core product information across all dimensions, clearly presents product benefits, generates personalized prompts that meet user needs, merges recommendations with card usage services, and enhances user experience and the practicality of card usage guidance.

[0022] According to another aspect of the present invention, an electronic device is provided, the electronic device comprising:

[0023] At least one processor;

[0024] and a memory communicatively connected to the at least one processor;

[0025] The memory stores a computer program that can be executed by the at least one processor, which enables the at least one processor to perform a product recommendation method according to any embodiment of the present invention.

[0026] According to another aspect of the present invention, a computer-readable storage medium is provided, the computer-readable storage medium storing computer instructions for causing a processor to execute and implement a product recommendation method according to any embodiment of the present invention.

[0027] According to another aspect of the present invention, a computer program product is provided, the computer program product comprising a computer program that, when executed by a processor, implements a product recommendation method according to any embodiment of the present invention.

[0028] The technical solution of this invention transforms unstructured user needs into structured data through a large model, accurately extracts core user needs, and improves recommendation accuracy; it protects data privacy through homomorphic encryption and federated computing, and completes secure matching without disclosing the original data, thereby improving recommendation security and user experience.

[0029] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0030] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0031] Figure 1 This is a flowchart of a product recommendation method provided according to Embodiment 1 of the present invention;

[0032] Figure 2 This is a flowchart of another product recommendation method provided according to Embodiment 2 of the present invention;

[0033] Figure 3 This is a schematic diagram of a product recommendation device according to Embodiment 3 of the present invention;

[0034] Figure 4 This is a schematic diagram of the structure of an electronic device that implements a product recommendation method according to an embodiment of the present invention. Detailed Implementation

[0035] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0036] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0037] Example 1

[0038] Figure 1 This is a flowchart illustrating a product recommendation method provided in Embodiment 1 of the present invention. This embodiment is applicable to credit card recommendation scenarios in fintech. The method can be executed by a product recommendation device, which can be implemented in hardware and / or software and can be configured in a computer controller. Figure 1 As shown, the method includes:

[0039] S110. Obtain the product requirements of the target users, generate structured data based on the product requirements through the large model, transform the structured data into a requirement feature vector, perform homomorphic encryption on the requirement feature vector, and generate an encrypted requirement feature vector.

[0040] In this system, target users refer to those who have product application or usage needs and are the subjects of product recommendations; they are the service subjects of the entire recommendation method. Product needs refer to the various demands, preferences, and usage-related needs expressed by target users for the recommended products, specifically reflected in users' card application needs, card usage habits, spending power, consumption scenarios, and benefit preferences, which can be expressed through users' natural language input. The large model refers to an intelligent model with natural language understanding, intent recognition, and data structuring capabilities, used to recognize users' natural language product needs, extract key information, and transform the dialogue results into structured data. Structured data refers to transforming users' unstructured natural language product needs into standardized data with a fixed format, clear classification, and direct usability for calculation and analysis. The demand feature vector refers to a computable numerical vector generated by transforming the structured data representing users' product needs according to preset dimensions. Homomorphic encryption is a cryptographic technique that allows computation on encrypted ciphertext; the decrypted result is identical to the result of performing the same computation on the plaintext. By homomorphically encrypting the demand feature vector and product feature vector, collaborative computation can be completed without accessing user privacy data, thus protecting user privacy and bank product data security throughout the entire process. The encrypted demand feature vector refers to the ciphertext vector obtained by performing a homomorphic encryption operation on the original demand feature vector. This vector cannot be directly interpreted to reveal the original user demand information and can only be used for collaborative computation in an encrypted state.

[0041] Optionally, structured data can be generated based on product requirements through a large model, including: obtaining the first prompt word; controlling the large model to perform intent recognition, key information capture, and sensitive information encryption processing on the product requirements based on the first prompt word; extracting key requirements; obtaining supplementary requirements of target users through the large model; and integrating key requirements and supplementary requirements in a specified format to generate structured data.

[0042] The first prompt word can be a pre-defined instruction set based on the business scenario of credit card recommendations. This first prompt word clarifies the processing objectives, identification dimensions, information capture scope, and sensitive information processing rules of the large-scale model. The core dimensions are set around the spending power, spending intention, spending scenarios, card usage frequency, and benefit preferences required for credit card recommendations. It also specifies the encryption requirements for sensitive information such as amounts and income, providing a standardized execution basis for the large-scale model to process user product needs. During processing, the large-scale model first performs intent recognition to determine the user's core card application and usage needs, such as focusing on overseas spending, daily dining discounts, or travel benefits. Then, it captures key information, extracting effective information related to the preset core dimensions from the user's natural language description. Simultaneously, it encrypts the identified sensitive information, mapping relevant monetary information to fuzzy encrypted ranges without retaining specific values ​​to avoid exposing user privacy. Finally, the system can obtain the target user's supplementary needs through the large model, and integrate the key needs and supplementary needs according to the specified format to generate structured data. The specified format is a standardized data format set to meet the needs of credit card recommendation calculation. By eliminating invalid and duplicate information, the unstructured natural language needs information is transformed into a structured dialogue summary with clear classification, clear dimensions, and unified format. This structured dialogue summary is the final generated structured data.

[0043] Optionally, supplementary needs of the target user can be obtained through a large model, including: obtaining a second prompt word; using the second prompt word to control the large model to generate a question-and-answer path based on product needs in combination with a decision tree; and conducting natural language dialogue with the target user based on the question-and-answer path to obtain supplementary needs input by the target user.

[0044] The second prompt can be a set of exclusive instructions pre-defined based on the business scenarios of credit card recommendations and the dimensions of user demand collection. The second prompt will clarify the rules for the large model to generate question-and-answer paths by combining decision trees, including the dimensional direction of question and answer, the progressive logic of questions, and the focus of follow-up questions for different product needs. At the same time, it will limit the content of follow-up questions to revolve around key dimensions of credit card recommendations such as spending power, spending intention, spending scenarios, card usage frequency, and benefit preferences, providing a standardized execution basis for the large model to generate reasonable and targeted question-and-answer paths.

[0045] Specifically, the system uses the second prompt word to control the large model and combines it with a decision tree to generate a question-and-answer path based on product needs. The decision tree builds hierarchical question branches based on the needs dimensions recommended by the credit card. Under the constraint of the second prompt word, the large model combines the user's original product needs and the initially extracted key needs to filter out the user's unclear or unmentioned needs dimensions in each branch of the decision tree. It then generates a coherent chain of natural language follow-up questions, i.e., a question-and-answer path, according to a logic that progresses from shallow to deep and from general to specific. The questions in the question-and-answer path will accurately point to the missing information in the user's needs. For example, if the user only expresses that they want a credit card with cashback benefits, the large model will combine the decision tree to generate progressive questions such as "In which consumption scenarios do you mainly want to use cashback benefits?" and "What is the approximate monthly consumption range?" to form a complete question-and-answer path. Then, the system will ask the user follow-up questions in natural language according to the question-and-answer path. The entire dialogue process maintains a natural interactive form, which conforms to the daily language communication habits. After the user replies to each follow-up question in natural language, the system will receive and record the user's reply in real time. For information that is still unclear in the user's reply, the system will continue to ask follow-up questions according to the logic of the question-and-answer path until all unclear demand dimensions in the decision tree are covered. Finally, all the reply content entered by the user in this natural language dialogue process will be integrated into the supplementary demand of the target user.

[0046] Optionally, the structured data is transformed into a demand feature vector, including: transforming the structured data into numerical vectors of each specified dimension according to a preset mapping rule, wherein the specified dimensions include strength features, preference features, scenario features, usage features and rights features; encoding the numerical vectors according to a specified format to obtain the encoded vectors of each specified dimension, and concatenating the encoded vectors of each specified dimension in sequence to obtain the demand feature vector.

[0047] The preset mapping rules are designed to transform users' non-numerical needs into calculable values, based on the credit card recommendation business. The specified dimensions—strength characteristics, preference characteristics, scenario characteristics, usage characteristics, and benefits characteristics—correspond to the dimensions of needs required for credit card recommendations. Strength characteristics correspond to a user's spending power, which is transformed into a numerical vector based on the user's spending range and spending limit information in the structured data, according to the mapping rules. Preference characteristics correspond to a user's spending intentions, which are transformed into a numerical vector based on the user's proactive desire to apply for and use a card. Scenario characteristics correspond to a user's spending scenarios, which are transformed into a numerical vector based on specific scenarios mentioned by the user, such as dining, overseas travel, and business travel. Usage characteristics correspond to a user's card usage frequency, which are transformed into a numerical vector based on the frequency of daily card use. Benefit characteristics correspond to a user's benefit preferences, which are transformed into a numerical vector based on the types of benefits the user is interested in, such as cashback, points, and discounts. Each dimension completes the transformation from structured needs information to calculable numerical vectors through preset mapping rules, ensuring that the needs information for each dimension can be represented in a quantitative form.

[0048] Specifically, the specified format is a standardized encoding format adapted for subsequent federated calculations and feature vector matching. By uniformly encoding the numerical vectors corresponding to strength features, preference features, scenario features, usage features, and equity features using this format, the numerical vectors of each dimension can be transformed into encoded vectors that meet the calculation requirements. This eliminates differences in format and measurement standards between numerical vectors of different dimensions, ensuring the uniformity and accuracy of subsequent concatenation and calculation. Finally, the encoded vectors of strength features, preference features, scenario features, usage features, and equity features are concatenated sequentially in a preset fixed order, integrating the scattered encoded vectors of each dimension into a single numerical vector, such as a 128-dimensional demand feature vector.

[0049] S120. Obtain candidate products, extract features from candidate products, generate product feature vectors, perform homomorphic encryption on product feature vectors, and generate encrypted product feature vectors.

[0050] Here, candidate products refer to all recommended products available for the target user to choose from in the system; these can be various credit card products. The product feature vector is a computable numerical vector derived from the core information obtained after feature extraction of candidate products, transformed according to preset rules. It matches the dimension of the demand feature vector and can be used for subsequent similarity calculations. The encrypted product feature vector is a ciphertext vector obtained by performing homomorphic encryption on the original product feature vector. It works in conjunction with the encrypted demand feature vector to achieve matching calculations in an encrypted state.

[0051] S130. Based on the encrypted demand feature vector and the encrypted product feature vector, perform federated calculations to determine the recommended products.

[0052] Federated computing refers to a collaborative computing method based on encrypted feature vectors, performed without exchanging or accessing the original plaintext data of either party. Recommended products are candidate products selected after federated computing, based on factors such as the similarity between needs and products, and dynamic benefit gains, that best match the product needs of the target user.

[0053] Optionally, the method also includes: obtaining product information and corresponding structured data of the recommended products, wherein the product information includes product name, application conditions, product validity period, product rights information and product additional information; constructing a product rights list based on the product rights information, and generating personalized prompts based on the structured data; and integrating the product information, product rights list and personalized prompts to form a structured report.

[0054] The product information includes the product name (the specific credit card type), application conditions (the identity and qualification requirements for applying for the card), product validity period (the valid period for using the credit card), product benefits information (the basic benefits that come with the card and the dynamic promotional benefits offered by the bank for the card), and additional product information (the annual fee collection rules, billing repayment rules, and other supporting information related to card use).

[0055] Specifically, when constructing the product benefits list, the system categorizes and organizes the product benefits information of recommended products, clearly separating basic card benefits from limited-time dynamic activity benefits. It also incorporates information such as the remaining validity period of bank dynamic activities and the scope of benefits that can be enjoyed, organizing the benefits into a clear and concise list so that users can intuitively understand all the benefits of the product. When generating personalized prompts, the system uses the user's corresponding structured data to accurately match the user's personalized card application and usage needs. For example, if the structured data shows that the user has a need for overseas spending, the system will generate targeted prompts such as how to participate in overseas cashback activities. If the user focuses on dining out, the system will generate prompts such as the usage scenarios and redemption methods for dining discounts. Finally, the system integrates product information, the product benefits list, and personalized prompts to form a structured report. The integration process arranges the content according to the user's viewing and usage logic, prioritizing the display of product information such as product name, application conditions, and product validity period, allowing users to quickly understand the basic information of the recommended product. Then, the organized product benefits list is displayed, clearly presenting the various benefits and related activity information of the product. Finally, personalized prompts tailored to user needs are placed, providing specific guidance for users to apply for and use the card. The resulting structured report is a scenario-based card usage guide for users, which includes complete information on recommended products as well as card usage tips tailored to individual user needs. This integrates card recommendations with card usage services, enhancing the user experience.

[0056] The technical solution of this invention transforms unstructured user needs into structured data through a large model, accurately extracts core user needs, and improves recommendation accuracy; it protects data privacy through homomorphic encryption and federated computing, and completes secure matching without disclosing the original data, thereby improving recommendation security and user experience.

[0057] Example 2

[0058] Figure 2 This is a flowchart of a product recommendation method provided in Embodiment 2 of the present invention. This embodiment adds a specific process for determining recommended products based on federated calculations using encrypted demand feature vectors and encrypted product feature vectors, building upon Embodiment 1. The specific content of steps S210-S220 is largely the same as steps S110-S120 in Embodiment 1, and therefore will not be repeated in this embodiment. Figure 2 As shown, the method includes:

[0059] S210. Obtain the product requirements of the target users, generate structured data based on the product requirements through the large model, transform the structured data into a requirement feature vector, perform homomorphic encryption on the requirement feature vector, and generate an encrypted requirement feature vector.

[0060] Optionally, structured data can be generated based on product requirements through a large model, including: obtaining the first prompt word; controlling the large model to perform intent recognition, key information capture, and sensitive information encryption processing on the product requirements based on the first prompt word; extracting key requirements; obtaining supplementary requirements of target users through the large model; and integrating key requirements and supplementary requirements in a specified format to generate structured data.

[0061] Optionally, supplementary needs of the target user can be obtained through a large model, including: obtaining a second prompt word; using the second prompt word to control the large model to generate a question-and-answer path based on product needs in combination with a decision tree; and conducting natural language dialogue with the target user based on the question-and-answer path to obtain supplementary needs input by the target user.

[0062] Optionally, the structured data is transformed into a demand feature vector, including: transforming the structured data into numerical vectors of each specified dimension according to a preset mapping rule, wherein the specified dimensions include strength features, preference features, scenario features, usage features and rights features; encoding the numerical vectors according to a specified format to obtain the encoded vectors of each specified dimension, and concatenating the encoded vectors of each specified dimension in sequence to obtain the demand feature vector.

[0063] S220. Obtain candidate products, extract features from candidate products, generate product feature vectors, perform homomorphic encryption on product feature vectors, and generate encrypted product feature vectors.

[0064] S230. Perform a dot product operation on the encrypted demand feature vector and the encrypted product feature vector to obtain the dot product result.

[0065] The dot product operation refers to a homomorphic multiplication and addition operation performed on the encrypted demand feature vector and the encrypted product feature vector under encrypted conditions. During the calculation, homomorphic encryption technology is used to directly perform this operation on the ciphertext vectors, and the decrypted result is identical to the result of the direct operation on the plaintext. The dot product result refers to the encrypted numerical value obtained after performing the dot product operation on the encrypted demand feature vector and product feature vector.

[0066] S240. Decrypt the dot product result to generate the decrypted dot product result.

[0067] The decrypted dot product result refers to the plaintext dot product value obtained after processing the encrypted dot product result using the corresponding decryption algorithm. This decryption operation is completed on the user's end; the bank does not participate in this process. The user's end uses the corresponding decryption algorithm to process the encrypted dot product result and convert it into a plaintext value.

[0068] S250. Obtain the preset product vector magnitude and calculate the demand vector magnitude corresponding to the demand feature vector.

[0069] The preset product vector modulus refers to the plaintext value calculated and pre-stored by the bank in plaintext mode for the product feature vector after feature extraction. The product vector modulus is an inherent attribute of the vector itself; pre-calculation and storage reduce collaborative steps in federated computing and improve computational efficiency. The bank calculates and retains the modulus for each card product and card level individually, eliminating the need for recalculation during the federated computing phase and improving overall computational efficiency. The demand vector modulus refers to the plaintext value obtained by the user client locally calculating the modulus of the unencrypted demand feature vector. This calculation is completed entirely on the user client; the bank does not access any of the user client's original data, ensuring user privacy.

[0070] S260. Calculate the product of the product vector magnitude and the demand vector magnitude, and calculate the ratio of the decrypted dot product result to the product, as the similarity between the demand feature vector and the product feature vector.

[0071] The similarity is calculated by "the decrypted dot product result ÷ (product vector magnitude × demand vector magnitude)". The higher the similarity, the higher the fit between the user's product demand and the candidate product.

[0072] S270. Calculate the activity gain factor, and combine the similarity with the activity gain factor to determine the recommended products.

[0073] Among them, the activity gain factor refers to the benefit gain value calculated by the bank for the corresponding dynamic limited-time activities of the candidate products. The calculation takes into account the remaining validity period of the activity and the benefits that users can obtain, and follows the rule that the effectiveness of the activity decays over time. The higher the value of the activity and the longer the remaining validity period, the higher the activity gain factor value. This factor is a dynamic supplement to the basic similarity and reflects the impact of the bank's real-time activities on product recommendations.

[0074] Optionally, calculate the activity gain factor, and combine the similarity with the activity gain factor to determine the recommended products, including: determining the activity validity period of each candidate product, determining the activity gain factor based on the activity validity period; multiplying the activity gain factor and the similarity to obtain the recommendation score of each candidate product; sorting each candidate product in descending order of recommendation score to obtain a recommendation list, and selecting a specified number of candidate products from the recommendation list as recommended products.

[0075] Specifically, the bank first reviews all high-value limited-time dynamic benefit activities corresponding to each candidate credit card product, clarifying the official total duration of each activity and its current effective time. The remaining validity period of the activity is then calculated. Simultaneously, based on a preset decay rule and the length of the activity's validity period, a corresponding activity gain factor is determined. A longer remaining validity period indicates higher benefit value for the user, resulting in a larger activity gain factor; conversely, a shorter remaining validity period indicates lower actual benefit value, resulting in a smaller activity gain factor. If a candidate product has no related limited-time dynamic activities, its activity gain factor remains at the default value. Next, the bank retrieves the similarity scores between each candidate product and user needs obtained from the previous federated calculation. This score reflects the basic matching degree between the product and user needs. The similarity score of each candidate product is then multiplied by its corresponding activity gain factor. This calculation combines the product's basic matching degree with its dynamic benefit value, resulting in a recommendation score for the candidate product. A high recommendation score comprehensively reflects the product's overall advantages in terms of basic matching degree and real-time benefit value. The bank will uniformly compare the recommendation scores of all candidate products and rank them in descending order of score to form a complete product recommendation list. The highest-scoring candidate product will be at the top of the recommendation list, and the remaining products will be arranged in descending order of score, creating a clear hierarchy of the overall recommendation value of the products. Finally, based on the pre-set business rules for credit card recommendations, the number of products to be pushed to the user will be determined. Then, a specified number of candidate products will be selected from the ranked recommendation list in descending order and used as the final recommended products to be pushed to the user.

[0076] Optionally, the method also includes: obtaining product information and corresponding structured data of the recommended products, wherein the product information includes product name, application conditions, product validity period, product rights information and product additional information; constructing a product rights list based on the product rights information, and generating personalized prompts based on the structured data; and integrating the product information, product rights list and personalized prompts to form a structured report.

[0077] The technical solution of this invention ensures the privacy of the data calculation process by performing dot product operations in an encrypted state. It improves computational efficiency by pre-setting the product vector magnitude, and avoids leakage of user privacy data by calculating the local demand vector magnitude. It quantifies the degree of matching between demand and products by calculating similarity. By calculating the activity gain factor and integrating dynamic bank activity data, combined with similarity, it comprehensively determines the recommended products, ensuring that the recommendation results balance matching degree and timeliness, achieving accurate and practical product recommendations.

[0078] Example 3

[0079] Figure 3 This is a schematic diagram of a product recommendation device provided in Embodiment 3 of the present invention. Figure 3 As shown, the device includes: a demand feature vector encryption module 310, which is used to obtain the product requirements of the target user, generate structured data based on the product requirements through a large model, convert the structured data into a demand feature vector, perform homomorphic encryption on the demand feature vector, and generate an encrypted demand feature vector.

[0080] The product feature vector encryption module 320 is used to obtain candidate products, extract features from candidate products, generate product feature vectors, perform homomorphic encryption on product feature vectors, and generate encrypted product feature vectors.

[0081] The recommended product determination module 330 is used to perform federated calculations based on the encrypted demand feature vector and the encrypted product feature vector to determine the recommended products.

[0082] Optionally, the requirement feature vector encryption module 310 specifically includes: a structured data generation unit, used to: obtain the first prompt word; control the large model based on the first prompt word to perform intent recognition, key information capture and sensitive information encryption processing on product requirements, and extract key requirements; obtain the supplementary requirements of the target user through the large model; and integrate the key requirements and supplementary requirements according to the specified format to generate structured data.

[0083] Optionally, the structured data generation unit specifically includes: a supplementary requirement acquisition subunit, used to: acquire a second prompt word; based on the second prompt word, control the large model to generate a question-and-answer path according to product requirements by combining a decision tree; and conduct natural language dialogue with the target user based on the question-and-answer path to acquire supplementary requirements input by the target user.

[0084] Optionally, the demand feature vector encryption module 310 specifically includes: a demand feature vector conversion unit, used to: convert structured data into numerical vectors of each specified dimension according to a preset mapping rule, wherein the specified dimensions include strength features, tendency features, scenario features, usage features and rights features; encode the numerical vectors according to a specified format to obtain the encoded vectors of each specified dimension, and concatenate the encoded vectors of each specified dimension in sequence to obtain the demand feature vector.

[0085] Optionally, the recommended product determination module 330 specifically includes: a dot product calculation unit, used to: perform a dot product operation on the encrypted demand feature vector and the encrypted product feature vector to obtain the dot product result; a dot product result decryption unit, used to: decrypt the dot product result to generate a decrypted dot product result; a vector magnitude determination unit, used to: obtain the preset product vector magnitude and calculate the demand vector magnitude corresponding to the demand feature vector; a similarity determination unit, used to: calculate the product of the product vector magnitude and the demand vector magnitude, and calculate the ratio of the decrypted dot product result to the product, as the similarity between the demand feature vector and the product feature vector; and a gain factor calculation and recommended product determination unit, used to: calculate the activity gain factor and combine the similarity and the activity gain factor to comprehensively determine the recommended product.

[0086] Optionally, the gain factor calculation and recommended product determination unit is used to: determine the activity validity period of each candidate product, determine the activity gain factor based on the activity validity period; multiply the activity gain factor and the similarity to obtain the recommendation score of each candidate product; sort the candidate products in descending order of recommendation score to obtain a recommendation list, and select a specified number of candidate products from the recommendation list as recommended products.

[0087] Optionally, the device also includes: a structured report generation module, used to: obtain product information of recommended products and corresponding structured data, wherein the product information includes product name, application conditions, product validity period, product rights information and product additional information; construct a product rights list based on product rights information, and generate personalized prompts based on structured data; and integrate product information, product rights list and personalized prompts to form a structured report.

[0088] The technical solution of this invention transforms unstructured user needs into structured data through a large model, accurately extracts core user needs, and improves recommendation accuracy; it protects data privacy through homomorphic encryption and federated computing, and completes secure matching without disclosing the original data, thereby improving recommendation security and user experience.

[0089] The product recommendation device provided in this embodiment of the invention can execute a product recommendation method provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of executing the method.

[0090] Example 4

[0091] Figure 4A schematic diagram of an electronic device 10 that can be used to implement embodiments of the present invention is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smartphones, wearable devices (e.g., helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0092] like Figure 4 As shown, the electronic device 10 includes at least one processor 11 and a memory, such as a read-only memory (ROM) 12 or a random access memory (RAM) 13, communicatively connected to the at least one processor 11. The memory stores computer programs executable by the at least one processor. The processor 11 can perform various appropriate actions and processes based on the computer program stored in the ROM 12 or loaded into the RAM 13 from storage unit 18. The RAM 13 can also store various programs and data required for the operation of the electronic device 10. The processor 11, ROM 12, and RAM 13 are interconnected via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0093] Multiple components in electronic device 10 are connected to I / O interface 15, including: input unit 16, such as keyboard, mouse, etc.; output unit 17, such as various types of displays, speakers, etc.; storage unit 18, such as disk, optical disk, etc.; and communication unit 19, such as network card, modem, wireless transceiver, etc. Communication unit 19 allows electronic device 10 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0094] Processor 11 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. Processor 11 performs the various methods and processes described above, such as a product recommendation method.

[0095] In some embodiments, a product recommendation method may be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 18. In some embodiments, part or all of the computer program may be loaded and / or installed on electronic device 10 via ROM 12 and / or communication unit 19. When the computer program is loaded into RAM 13 and executed by processor 11, one or more steps of a product recommendation method described above may be performed. Alternatively, in other embodiments, processor 11 may be configured to perform a product recommendation method by any other suitable means (e.g., by means of firmware).

[0096] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0097] Computer programs used to implement the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be performed. The computer programs may be executed entirely on a machine, partially on a machine, or as a standalone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0098] In the context of this invention, a computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination thereof. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0099] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0100] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or middleware components (e.g., application servers), or frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.

[0101] A computing system can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a hosting product within the cloud computing service system. It addresses the shortcomings of traditional physical hosts and VPS services, such as high management difficulty and weak business scalability.

[0102] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0103] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A product recommendation method characterized by, include: Obtain the product requirements of the target users, generate structured data based on the product requirements through a large model, transform the structured data into a requirement feature vector, perform homomorphic encryption on the requirement feature vector, and generate an encrypted requirement feature vector. Obtain candidate products, extract features from the candidate products to generate product feature vectors, and perform homomorphic encryption on the product feature vectors to generate encrypted product feature vectors. Federated computation is performed based on the encrypted demand feature vector and the encrypted product feature vector to determine recommended products.

2. The method of claim 1, wherein, The generation of structured data based on product requirements through a large model includes: Obtain the first prompt word, and based on the first prompt word, control the large model to perform intent recognition, key information capture, and sensitive information encryption processing on the product requirements, and extract the key requirements. Acquire supplementary needs of target users through large models; The key requirements and supplementary requirements are integrated according to the specified format to generate structured data.

3. The method according to claim 2, characterized in that, The process of obtaining supplementary needs from target users through a large model includes: Obtain the second prompt word, and based on the second prompt word, control the large model and combine the decision tree to generate a question-and-answer path according to the product requirements; Based on the question-and-answer path, a natural language dialogue is conducted with the target user to obtain supplementary requirements input by the target user.

4. The method according to claim 1, characterized in that, The process of converting the structured data into a demand feature vector includes: The structured data is transformed into numerical vectors of various specified dimensions according to a preset mapping rule. The specified dimensions include strength characteristics, tendency characteristics, scenario characteristics, usage characteristics, and rights characteristics. The numerical vector is encoded according to a specified format to obtain encoded vectors for each specified dimension. The encoded vectors for each specified dimension are then concatenated sequentially to obtain the required feature vector.

5. The method according to claim 4, characterized in that, The process of performing federated calculations based on the encrypted demand feature vector and the encrypted product feature vector to determine recommended products includes: Perform a dot product operation on the encrypted demand feature vector and the encrypted product feature vector to obtain the dot product result; The dot product result is decrypted to generate a decrypted dot product result; Obtain the preset product vector magnitude and calculate the demand vector magnitude corresponding to the demand feature vector; Calculate the product of the product vector magnitude and the demand vector magnitude, and calculate the ratio of the decrypted dot product result to the product, as the similarity between the demand feature vector and the product feature vector; Calculate the activity gain factor, and combine the similarity with the activity gain factor to determine the recommended product.

6. The method according to claim 5, characterized in that, The calculation of the activity gain factor, combined with the similarity and the activity gain factor, to determine the recommended product includes: Determine the activity validity period for each candidate product, and determine the activity gain factor based on the activity validity period; Multiply the activity gain factor and the similarity by the ratio to obtain the recommendation score for each candidate product; The candidate products are sorted in descending order of recommendation score to obtain a recommendation list. A specified number of candidate products are selected from the recommendation list as recommended products.

7. The method according to claim 1, characterized in that, The method further includes: Obtain product information and corresponding structured data for recommended products, wherein the product information includes product name, application conditions, product validity period, product rights information and product additional information; A product rights list is constructed based on the product rights information, and personalized prompts are generated based on the structured data; The product information, the product benefits list, and the personalized prompts are integrated to form a structured report.

8. An electronic device, characterized in that, The electronic device includes: At least one processor; and a memory communicatively connected to the at least one processor; The memory stores a computer program that can be executed by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-7.

9. A computer storage medium, characterized in that, The computer storage medium stores computer instructions that are used to cause a processor to execute the method of any one of claims 1-7.

10. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the method according to any one of claims 1-7.

Images