Product recommendation method and device, equipment, medium and program

Abstract

The invention discloses a product recommendation method and device, equipment, a medium and a program, and relates to the application of the artificial intelligence field in the business architecture and financial science and technology fields, and the method comprises the steps: after receiving a product recommendation request of a target user, extracting the fluctuation level characteristics of the user from a database according to a user identifier, and preliminarily screening a matched candidate product set; and jointly inputting the user features and the product features into a pre-trained distributed recommendation model to obtain a prediction score of each product. And generating an initial recommendation list according to the score, and retrieving a fluctuation level corresponding to each product in the list. And submitting the initial list, the user fluctuation level and the product fluctuation level to a rule engine, performing joint screening according to multiple fluctuation level constraints, generating a final recommendation result, and returning a response. According to the embodiment of the invention, the recommendation efficiency is improved through distributed model deployment, and the service compliance is enhanced while the recommendation accuracy is ensured by combining double-fluctuation-level risk control and rule engine screening.

Description

Technical Field

[0001] This invention relates to the field of distributed computing technology, to the application of artificial intelligence in business architecture and fintech, and particularly to a product recommendation method, apparatus, device, medium, and program. Background Technology

[0002] In the fintech field, providing personalized services to users has become one of the core technological directions. Current mainstream methods typically rely on users' historical transaction data and behavioral patterns, using machine learning models for clustering and regression analysis to construct user profiles and predict their potential needs. The system identifies groups with similar risk preferences and behavioral characteristics, using highly-regarded products within these groups as a candidate recommendation set.

[0003] Despite the widespread application of existing technologies, their core models are primarily based on interest matching and popularity propagation, which have several inherent flaws: First, the recommendation logic is often a black box process, lacking transparency in the decision-making basis, making it difficult for users to understand the reasons for the recommendations, thus weakening their trust in the system; second, although some systems employ feedback mechanisms, the overall optimization direction still focuses on increasing click-through rates or user engagement, rather than deeply considering the long-term value or potential risks of the recommendation results; finally, these technologies typically lack sufficient flexibility and adaptability, failing to provide truly tailored solutions for users with highly personalized needs, and unable to dynamically adjust to changes in user requirements. Summary of the Invention

[0004] Based on this, the present invention provides a product recommendation method, apparatus, device, medium and program to solve the problems of low transparency, insufficient risk consideration and limited personalization in existing financial technology recommendation systems.

[0005] In a first aspect, embodiments of the present invention provide a requirement management method for dynamic iteration of business requirements, including:

[0006] In response to a product recommendation request from a target user, the system retrieves the target user's first fluctuation level from the user information database based on the target user's user identifier, and obtains candidate products that match the target user.

[0007] The user characteristics of the target user and the product characteristics of each candidate product are input into a pre-trained product recommendation model to obtain the score of each candidate product. The product recommendation model includes multiple model layers, and different model layers are configured on different server nodes.

[0008] An initial product recommendation list is generated based on the score of each candidate product, and the second fluctuation level corresponding to each candidate product in the initial product recommendation list is obtained from the product information database.

[0009] The initial product recommendation list, the first volatility level, and the second volatility level of each product are jointly provided to the rule engine service;

[0010] The rules engine service filters the products in the initial product recommendation list based on the first fluctuation level and each of the second fluctuation levels to obtain the target product recommendation list, which serves as the response result to the product recommendation request.

[0011] Secondly, embodiments of the present invention also provide a product recommendation device, comprising:

[0012] The product recommendation request response module is used to respond to product recommendation requests from target users. Based on the user identifier of the target user, it retrieves the first fluctuation level of the target user from the user information database and obtains each candidate product that matches the target user.

[0013] The candidate product scoring module is used to input the user feature information of the target user and the product feature information of each candidate product into the pre-trained product recommendation model to obtain the score value of each candidate product. The product recommendation model includes multiple model layers, and different model layers are configured on different server nodes.

[0014] The initial product recommendation list generation module is used to generate an initial product recommendation list based on the score of each candidate product, and to retrieve the second fluctuation level corresponding to each candidate product in the initial product recommendation list from the product information database.

[0015] The rule engine receiving module is used to provide the initial product recommendation list, the first fluctuation level, and each of the second fluctuation levels to the rule engine service.

[0016] The target product recommendation list generation module is used to filter products in the initial product recommendation list based on the first fluctuation level and each of the second fluctuation levels through the rule engine service, and obtain the target product recommendation list as the response result to the product recommendation request.

[0017] Thirdly, embodiments of the present invention also provide an electronic device, the electronic device comprising:

[0018] At least one processor; and

[0019] A memory communicatively connected to the at least one processor; wherein,

[0020] 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 a product recommendation method according to any embodiment of the present invention.

[0021] Fourthly, embodiments of the present invention also provide a computer-readable storage medium storing computer instructions that are used to cause a processor to execute a product recommendation method as described in any embodiment of the present invention.

[0022] Fifthly, embodiments of the present invention also provide a computer program product, the computer program product including a computer program, which, when executed by a processor, implements a product recommendation method as described in any embodiment of the present invention.

[0023] The technical solution of this invention distributes the computational load by deploying different model layers of a pre-trained product recommendation model on different server nodes, thereby improving model inference efficiency and system scalability under high concurrency requests. It innovatively introduces the first fluctuation level from the user end and the second fluctuation level from the product end as key features, and performs secondary filtering via a rule engine service, enhancing the risk adaptability and business compliance of the recommendation results. It constructs an end-to-end processing flow from candidate product generation and model scoring to rule optimization, achieving an organic combination of data-driven and rule-driven approaches, improving the flexibility and interpretability of the recommendation system. This invention effectively balances recommendation performance and risk control, and is suitable for application scenarios with high requirements for response speed, system stability, and decision reliability.

[0024] 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

[0025] 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.

[0026] Figure 1 This is a flowchart of a product recommendation method provided according to Embodiment 1 of the present invention;

[0027] Figure 2 This is a flowchart of another product recommendation method provided according to Embodiment 2 of the present invention;

[0028] Figure 3 This is a schematic diagram of a product recommendation device according to Embodiment 3 of the present invention;

[0029] Figure 4This 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

[0030] 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.

[0031] 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.

[0032] Example 1

[0033] Figure 1 This is a flowchart of a product recommendation method provided in Embodiment 1 of the present invention. This embodiment is applicable to situations where personalized and accurate recommendations and real-time risk adaptation are performed for high-value, high-volatility products. The method can be executed by a product recommendation device, which can be implemented in hardware and / or software. This device can be configured in a server cluster containing multiple server nodes that supports distributed model inference. Figure 1 As shown, the method includes:

[0034] S110. In response to a product recommendation request from a target user, based on the target user's user identifier, obtain the target user's first fluctuation level from the user information database, and obtain each candidate product that matches the target user.

[0035] First, a server cluster refers to a collection of multiple servers (computers) interconnected via a high-speed network. They work collaboratively to provide a unified, high-performance, and highly available computing service. In this embodiment, it is used to share the computing load. A server node refers to each independent server within the server cluster. A user identifier is a string or ID used to uniquely identify a user, typically a user account, mobile phone number, etc. A user information database is a system that stores user-related data, such as user identity information, historical behavior, and risk attributes. The first volatility level is a user-side risk quantification indicator, representing the user's risk tolerance or risk preference. Candidate products are a subset of products selected from the full product pool based on the user's initial profile, potentially meeting the user's needs; they serve as input for subsequent refined ranking. Using a user request as the trigger, the system locates the user through a unique identifier, obtains the user-side key screening criteria (first volatility level) and the initial recommendation range (candidate products), and uses database retrieval technology based on the user identifier to quickly associate user attributes with candidate products.

[0036] S120. Input the user feature information of the target user and the product feature information of each candidate product into the pre-trained product recommendation model to obtain the score value of each candidate product; wherein, the product recommendation model includes multiple model layers, and different model layers are configured on different server nodes.

[0037] A product recommendation model refers to an algorithmic model trained through machine learning, used to quantify the matching degree between products and users. The score is the quantitative result output by the model, reflecting the suitability of candidate products for the target user. The model layer refers to the functional modules of the model, such as the feature extraction layer and the matching calculation layer. Server nodes refer to independent computing units within a server cluster. By fusing multi-dimensional feature information of the target user (such as behavioral features and profile features) with the attribute information of candidate products (such as category features and parameter features), the pre-trained product recommendation model completes the matching degree calculation and outputs a score reflecting the suitability. Simultaneously, a distributed deployment architecture is clearly defined, with multiple model layers distributed across different server nodes. This design, through functional module decomposition and parallel computing, effectively improves the efficiency of feature processing and matching calculation, meets the real-time recommendation requirements in high-concurrency scenarios, and achieves the key transformation from qualitative matching to quantitative scoring.

[0038] Optionally, before inputting the user characteristic information of the target user and the product characteristic information of each candidate product into the pre-trained product recommendation model, the following may also be included:

[0039] Historical user behavior data is acquired, the historical user behavior data is cleaned, a user product interaction matrix is ​​constructed based on the cleaned data, and the user product interaction matrix is ​​used as training data to construct the initial loss function of the matrix factorization model.

[0040] The historical first fluctuation level of historical users is obtained from the certified user information database, and the historical second fluctuation level of each historical product is obtained from the certified product information database.

[0041] The historical first fluctuation level and the historical second fluctuation level of each historical product are used as constraints and added to the initial loss function to form a constrained loss function;

[0042] The pre-trained product recommendation model is obtained by training the matrix factorization model by minimizing the constraint loss function.

[0043] This embodiment, as a pre-stage step in the real-time recommendation process, focuses on the construction of the pre-trained model. First, historical user behavior data is used as the training data source. Second, data cleaning ensures data quality and avoids noise affecting model accuracy. Third, the cleaned data is transformed into a user-product interaction matrix (rows represent users, columns represent products, and matrix elements represent the intensity of user interaction with products), realizing the transformation of unstructured behavioral data into structured training samples. Finally, using this matrix as the core training data, an initial loss function for the matrix factorization model is constructed. This loss function measures the deviation between the model's predictions and actual interaction behavior, serving as the core basis for model optimization. Relying solely on the initial loss function, the model can only optimize preference matching accuracy, potentially recommending products with high preference levels but mismatched volatility attributes. Therefore, by incorporating the first and second historical volatility levels as constraints into the initial loss function, a constrained loss function is constructed. In this case, the model's optimization objective is not only to minimize the prediction interaction deviation but also to minimize the "mismatch between user and product volatility attributes." This design allows the model to learn volatility adaptation rules simultaneously while learning preferences, thus giving the trained model the dual capabilities of "preference matching + volatility adaptation." The model can iteratively minimize the constrained loss function through optimization algorithms such as gradient descent, ensuring that the model's predictions closely reflect actual user behavior and align with the fluctuating attributes of users and products. Training terminates when the loss function converges to a preset threshold, outputting a pre-trained product recommendation model. This model integrates preference learning and fluctuation adaptation logic, allowing it to be directly used for rating calculations in real-time recommendation phases, ensuring both accuracy and suitability in the recommendation results.

[0044] This embodiment constructs an initial loss function by acquiring and cleaning historical user behavior data, introduces historical fluctuation levels as constraints to form a constrained loss function, and trains a pre-trained product recommendation model that takes into account both preference matching and fluctuation adaptation by minimizing this function.

[0045] Furthermore, the user characteristic information of the target user and the product characteristic information of each candidate product are jointly input into a pre-trained product recommendation model to obtain the score value of each candidate product, which may include:

[0046] By using the matrix factorization layer in the pre-trained product recommendation model, the user feature information of the target user is mapped and processed to generate the latent vector of the target user.

[0047] The product feature information of each candidate product is mapped through the matrix factorization layer in the pre-trained product recommendation model to generate the corresponding candidate product latent vector.

[0048] Extract the risk feature dimension data from the latent vector of the target user and the risk feature dimension data from the latent vector of the candidate product, and calculate the similarity between the two types of risk feature dimension data to obtain the risk matching degree parameter;

[0049] The base score is obtained by calculating the inner product of the latent vector of the target user and the latent vector of each candidate product. Then, the risk matching degree parameter is multiplied by the base score to obtain the score value of each candidate product.

[0050] The matrix factorization layer, as the core component of the model, compresses high-dimensional, sparse user feature information into low-dimensional, dense target user latent vectors through linear or non-linear mapping. Similarly, the matrix factorization layer maps high-dimensional product feature information into low-dimensional candidate product latent vectors, maintaining the same dimensionality as the user latent vectors to ensure computability. A risk feature dimension is specifically defined within the user and product latent vectors. The degree of matching between these two dimensions is calculated using methods such as cosine similarity and Pearson correlation coefficient, yielding a risk matching parameter. The base score is calculated by the inner product of the user and product latent vectors, reflecting the overall matching degree between the user and product across all feature dimensions. This weighted logic ensures that the scoring result reflects both overall preference matching and risk adaptability adjustment. If the product and user risk features do not match, even with a high base score, the final score will decrease, thus achieving a unified quantification of preference accuracy and risk compliance.

[0051] The entire process in this embodiment follows the framework of "feature vector transformation → multi-dimensional matching calculation → comprehensive score generation". It achieves feature dimensionality reduction and abstraction through matrix decomposition, and integrates user preferences and volatility attribute adaptability into a single score value through a two-layer calculation logic of "basic matching + risk adaptation", providing a quantitative basis for subsequent recommendation list sorting that balances accuracy and compliance.

[0052] S130. Generate an initial product recommendation list based on the score of each candidate product, and retrieve the second fluctuation level corresponding to each candidate product in the initial product recommendation list from the product information database.

[0053] The initial product recommendation list is a set of candidate products sorted by rating value, representing the original recommendation results output by the model. The product information database is a structured database storing data such as product attributes, specifications, and volatility characteristics. The second volatility level is an indicator reflecting the volatility characteristics of the product itself (such as the yield volatility of financial products). First, the initial product recommendation list is generated by sorting the products in descending order of rating value, establishing a recommendation priority based on user preferences. Second, by querying the product information database, the second volatility level is associated with the candidate products in the initial list. This indicator, as a core attribute of the product, reflects the volatility characteristics of the product itself and is a key basis for subsequent suitability screening. This achieves a connection between preference ranking and suitability dimension supplementation, providing complete data support for the compliance screening of the rule engine.

[0054] S140. The initial product recommendation list, the first volatility level, and the second volatility level of each product are jointly provided to the rule engine service.

[0055] The rules engine service refers to the service module that performs data filtering and verification based on preset business rules. The initial product recommendation list, first fluctuation level, and second fluctuation level—three core data types generated in the preceding stages—are passed to the rules engine service through a standardized interface. This modular division of labor design allows the algorithm model to focus on matching and measurement, while the rules engine focuses on business rule verification, improving both the system's scalability and ensuring the integrity and standardization of data flow.

[0056] S150. The rule engine service filters the products in the initial product recommendation list according to the first fluctuation level and each of the second fluctuation levels to obtain the target product recommendation list, which serves as the response result to the product recommendation request.

[0057] Based on preset volatility adaptation rules, the rules engine compares and verifies the first volatility level with the second volatility level, eliminating candidate products whose volatility characteristics exceed the user's acceptable threshold, and retaining products with both high matching degree and compliance adaptability, generating a target product recommendation list. This list is fed back to the request initiator as a response result, forming a complete closed loop of "request-processing-response".

[0058] This invention, through its embodiments, deploys different model layers of a pre-trained product recommendation model on different server nodes, achieving distributed allocation of computational load and improving model inference efficiency and system scalability under high concurrency requests. It innovatively introduces the first fluctuation level from the user's perspective and the second fluctuation level from the product's perspective as key features, and performs secondary filtering via a rule engine service, enhancing the risk adaptability and business compliance of the recommendation results. It constructs an end-to-end processing flow from candidate product generation and model scoring to rule optimization, achieving an organic combination of data-driven and rule-driven approaches, improving the flexibility and interpretability of the recommendation system. This invention effectively balances recommendation performance and risk control, making it suitable for application scenarios with high requirements for response speed, system stability, and decision reliability.

[0059] Example 2

[0060] Figure 2 This is a flowchart of another product recommendation method provided in Embodiment 2 of the present invention. This embodiment is a refinement based on Embodiment 1, specifically as follows: Figure 2 As shown, the method includes:

[0061] S210. In response to the product recommendation request for the target user, based on the user identifier of the target user, obtain the first fluctuation level of the target user from the user information database, and obtain each candidate product that matches the target user.

[0062] S220. Input the user feature information of the target user and the product feature information of each candidate product into the pre-trained product recommendation model to obtain the score value of each candidate product; wherein, the product recommendation model includes multiple model layers, and different model layers are configured on different server nodes.

[0063] S230. Sort the candidate products in descending order of their scores to form an initial product recommendation list that includes the product identifier and corresponding score of each candidate product.

[0064] The ranking system, based on scores from highest to lowest, essentially establishes the recommendation order according to the overall suitability of candidate products to users. Higher scores indicate a better match between the product and the user's needs, resulting in a higher ranking. The final initial product recommendation list includes product identifiers and scores, transforming quantitative scores into a structured, ordered list, providing a standardized object for subsequent data processing and filtering.

[0065] S240. Based on the candidate product identifiers in the initial product recommendation list, generate a batch query request containing the product identifiers.

[0066] Using the candidate product identifier as a unique association key, batch query requests are generated by integrating the identifiers of all products in the list, rather than initiating queries one by one. This design significantly reduces the number of interactions with the database, lowers network communication overhead, and improves data acquisition efficiency, making it particularly suitable for scenarios with a large number of candidate products. Furthermore, using the product identifier as the query condition ensures the accuracy of the query target and avoids product information confusion.

[0067] S250. Send the batch query request to the certified product information database to request the second fluctuation level corresponding to each candidate product.

[0068] A certified product information database means that the database has undergone permission verification and data compliance audit, ensuring that the obtained second fluctuation level is true and reliable, and avoiding unauthorized or distorted data from affecting subsequent screening logic; requesting the second fluctuation level corresponding to each candidate product clarifies the query target, extracting only the attributes necessary for subsequent rule engine screening, rather than full product information, which reduces the amount of data transmission and ensures the relevance of data acquisition.

[0069] S260. Receive response data returned by the certified product information database, and extract the second fluctuation level associated with each candidate product identifier from the response data.

[0070] First, the system receives response data from the database. Then, using the candidate product identifier as the association basis, it extracts the corresponding second fluctuation level from the response data. Through identifier matching, it ensures that the fluctuation attribute of each product corresponds one-to-one with the products in the initial list, avoiding attribute mismatches. The extracted second fluctuation level is then associated with the initial recommendation list, providing key product-side evidence for subsequent rule engine filtering based on "user-product fluctuation matching".

[0071] This embodiment establishes recommendation priorities through scoring and sorting, and efficiently obtains product fluctuation attributes through batch and certified query methods, ultimately forming a complete data set of "ordered list + key attributes". This achieves a seamless connection from quantitative scoring to screening preparation, laying a data foundation for subsequent compliant screening.

[0072] S270. Standardize the format of the associated list and the first fluctuation level of the target user to ensure that the data fields conform to the interface specifications of the rule engine service.

[0073] The initial product recommendation list includes product identifiers and rating values. The second fluctuation level is stored separately, indexed by the product identifier. The two are linked and bound together through the product identifier, forming a structured list of "product identifier - rating value - second fluctuation level". This integration ensures a one-to-one correspondence between the rating value and the second fluctuation level for each product, preventing mismatches between products and attributes during rule engine filtering.

[0074] S280. Standardize the format of the associated list and the first fluctuation level of the target user to ensure that the data fields conform to the interface specifications of the rule engine service.

[0075] Because the server cluster and the rule engine service may use different data format standards, direct transmission can easily lead to parsing failures. Therefore, it is necessary to standardize the format of the association list and the first fluctuation level, specifying that the numerical type is a string or floating-point number, and adjusting the data structure to JSON or XML format, etc., to ensure that all data fields fully match the interface requirements of the rule engine and eliminate format barriers in cross-service data interaction.

[0076] S290. Through a preset encrypted communication protocol, the association list, the target user's first fluctuation level, and the transmission verification identifier are sent to the rule engine service.

[0077] The preset encrypted communication protocol can encrypt the transmitted data to prevent sensitive information such as user volatility levels and product ratings from being stolen or tampered with during transmission; the transmission verification identifier is sent synchronously with the data to allow the rule engine to verify whether the received data is complete and has not been tampered with; the simultaneous transmission ensures that the rule engine obtains all the data required for filtering at once without the need for secondary requests, thus improving interaction efficiency.

[0078] S2100: Receive the receipt returned by the rule engine service. After confirming that the data has been fully received, record the transmission verification identifier and the sending time to complete the data provision process.

[0079] After receiving data, the rule engine returns a receipt containing the verification result of the transmission verification identifier. If the verification passes, it indicates that the data is complete and has not been tampered with. At this point, the verification identifier and the sending time are recorded, providing a basis for tracing any potential data transmission problems and marking the formal completion of the data provision process. If the verification fails, the transmission can be re-initiated based on the receipt information to avoid interruption of the recommendation process due to abnormal data transmission.

[0080] This embodiment addresses the core issue of how to deliver multi-source data to the rules engine in a compliant, secure, and complete manner through a process of "data association and integration → format standard adaptation → secure encrypted transmission → verification and confirmation with traceability." It eliminates the risk of data misalignment through structured integration, removes interface compatibility barriers through format standardization, and ensures transmission security through encryption and verification, thereby achieving reliable data flow across services.

[0081] S2110. The rule engine service filters the products in the initial product recommendation list according to the first fluctuation level and each second fluctuation level to obtain the target product recommendation list, which serves as the response result to the product recommendation request.

[0082] Optionally, the rules engine service can be used to filter the products in the initial product recommendation list based on the first fluctuation level and each of the second fluctuation levels to obtain the target product recommendation list, which may include:

[0083] The rule engine service receives and parses the initial product recommendation list, the first fluctuation level of the target user, and the second fluctuation level corresponding to each candidate product.

[0084] The rules engine service uses pre-stored fluctuation matching rules to convert the target user's first fluctuation level into the corresponding allowed fluctuation range threshold.

[0085] The rule engine service compares the second volatility level of each candidate product in the initial product recommendation list with the allowable volatility range threshold, and retains the candidate products whose risk level is within the recommended risk level range.

[0086] The rules engine service sorts the retained products in the initial product recommendation list from high to low according to their original rating values, and generates a target product recommendation list containing the product identifier, rating value, second fluctuation level, and fluctuation verification pass indicator of the retained products.

[0087] The generated target product recommendation list is returned to the server that issued the initial product recommendation list via the rule engine service according to the preset transmission protocol.

[0088] The rules engine first receives three types of core data transmitted from the server cluster. Then, it performs parsing to transform the structured data into a format it can recognize. The core objective of parsing is to ensure clear data structure and accurate correlation, avoiding deviations in subsequent filtering logic due to data format confusion. Pre-stored fluctuation matching rules are based on logic preset according to business needs. The rules engine uses this logic to transform the abstract first fluctuation level into a specific, quantifiable allowable fluctuation range threshold, changing "what users can accept" from a qualitative description to a quantitative standard. This provides a clear basis for subsequent comparisons of whether product attributes meet user standards. The rules engine uses a one-by-one comparison method to verify the second fluctuation level of each candidate product against the allowable fluctuation range threshold, optimizing the recommendation results from considering only preferences to a balance between preferences and compliance. A new fluctuation verification identifier is added to clearly mark products that have passed fluctuation adaptation verification, providing a basis for the server to trace the filtering results. The structured design of the list ensures that the output data is standardized and can be directly used to respond to requests. The rules engine transmits the target product recommendation list back to the original server according to a preset transmission protocol. This ensures the standardization and security of data transmission, and enables the server to obtain the final recommendation results in a timely manner. This allows the server to respond to the initial product recommendation request and complete the closed-loop implementation of the entire recommendation process. This embodiment filters out unqualified products, retains compliant and highly preferred products, and ultimately generates and transmits a structured target recommendation list. This ensures that the output results are accurate, compliant, and usable, fully achieving the ultimate goal of the recommendation process.

[0089] Example 3

[0090] 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:

[0091] The product recommendation request response module 310 is used to respond to product recommendation requests from target users, obtain the first fluctuation level of the target user from the user information database based on the user identifier of the target user, and obtain each candidate product that matches the target user.

[0092] The candidate product scoring module 320 is used to input the user feature information of the target user and the product feature information of each candidate product into the pre-trained product recommendation model to obtain the score value of each candidate product; wherein, the product recommendation model includes multiple model layers, and different model layers are configured on different server nodes;

[0093] The initial product recommendation list generation module 330 is used to generate an initial product recommendation list based on the score values ​​of each candidate product, and to obtain the second fluctuation level corresponding to each candidate product in the initial product recommendation list from the product information database.

[0094] Service provision module 340 is used to provide the initial product recommendation list, the first fluctuation level and each of the second fluctuation levels to the rule engine service;

[0095] The target product recommendation list generation module 350 is used to filter the products in the initial product recommendation list based on the first fluctuation level and each second fluctuation level through the rule engine service, and obtain the target product recommendation list as the response result to the product recommendation request.

[0096] This invention, through its embodiments, deploys different model layers of a pre-trained product recommendation model on different server nodes, achieving distributed allocation of computational load and improving model inference efficiency and system scalability under high concurrency requests. It innovatively introduces the first fluctuation level from the user's perspective and the second fluctuation level from the product's perspective as key features, and performs secondary filtering via a rule engine service, enhancing the risk adaptability and business compliance of the recommendation results. It constructs an end-to-end processing flow from candidate product generation and model scoring to rule optimization, achieving an organic combination of data-driven and rule-driven approaches, improving the flexibility and interpretability of the recommendation system. This invention effectively balances recommendation performance and risk control, making it suitable for application scenarios with high requirements for response speed, system stability, and decision reliability.

[0097] Optionally, based on the above embodiments, it may include: a product recommendation model pre-training unit, used to obtain historical user behavior data, clean the historical user behavior data, construct a user product interaction matrix based on the cleaned data, and use the user product interaction matrix as training data to construct the initial loss function of the matrix factorization model.

[0098] The historical first fluctuation level of historical users is obtained from the certified user information database, and the historical second fluctuation level of each historical product is obtained from the certified product information database.

[0099] The historical first fluctuation level and the historical second fluctuation level of each historical product are used as constraints and added to the initial loss function to form a constrained loss function;

[0100] The pre-trained product recommendation model is obtained by training the matrix factorization model by minimizing the constraint loss function.

[0101] Optionally, based on the above embodiments, the candidate product scoring module 320 may include:

[0102] The first mapping processing unit is used to map the user feature information of the target user through the matrix factorization layer in the pre-trained product recommendation model to generate the latent vector of the target user.

[0103] The second mapping processing unit is used to map the product feature information of each candidate product through the matrix factorization layer in the pre-trained product recommendation model to generate the corresponding candidate product latent vector.

[0104] The risk matching degree calculation unit is used to extract the risk feature dimension data in the latent vector of the target user and the risk feature dimension data in the latent vector of the candidate product, and to calculate the similarity between the two types of risk feature dimension data to obtain the risk matching degree parameter.

[0105] The candidate product score calculation module is used to calculate the inner product of the target user latent vector and the latent vector of each candidate product to obtain a basic score, and then multiply the risk matching degree parameter with the basic score to obtain the score of each candidate product.

[0106] Optionally, based on the above embodiments, the initial product recommendation list generation module 330 may include:

[0107] The candidate product sorting unit is used to sort the candidate products in descending order of their score values ​​to form an initial product recommendation list containing the product identifiers and corresponding score values ​​of each candidate product.

[0108] A batch query request generation unit is used to generate a batch query request containing the product identifiers based on each candidate product identifier in the initial product recommendation list.

[0109] The second fluctuation level acquisition unit is used to send the batch query request to the certified product information database to request the second fluctuation level corresponding to each candidate product.

[0110] The response data receiving unit is used to receive response data returned by the certified product information database and extract the second fluctuation level associated with each candidate product identifier from the response data.

[0111] Optionally, based on the above embodiments, the service providing module 340 may include:

[0112] The association list generation unit is used to associate and bind the initial product recommendation list with the extracted second fluctuation level of each candidate product to form an association list containing product identifier, rating value and second fluctuation level;

[0113] The formatting unit is used to format the associated list and the first fluctuation level of the target user to ensure that the data fields conform to the interface specifications of the rule engine service.

[0114] The service transfer unit is used to send the association list, the first fluctuation level of the target user, and the transmission verification identifier to the rule engine service through a preset encrypted communication protocol.

[0115] The receipt receiving unit is used to receive the receipt returned by the rule engine service. After confirming that the data has been completely received, it records the transmission verification mark and the sending time, thus completing the data provision process.

[0116] Optionally, based on the above embodiments, the target product recommendation list generation module 350 may include:

[0117] The initial product recommendation list parsing unit is used to receive and parse the initial product recommendation list, the target user's first fluctuation level, and the second fluctuation level corresponding to each candidate product through the rule engine service.

[0118] The first fluctuation level conversion unit is used to convert the target user's first fluctuation level into the corresponding allowed fluctuation range threshold by using the rule engine service according to the pre-stored fluctuation matching rules;

[0119] The candidate product screening unit is used to compare the second fluctuation level of each candidate product in the initial product recommendation list with the allowed fluctuation range threshold through the rule engine service, and to retain the candidate products whose risk level is within the recommended risk level range.

[0120] The target product recommendation list generation unit is used to sort the retained products in the initial product recommendation list according to the original rating values ​​of the retained products in the rule engine service from high to low, and generate a target product recommendation list containing the product identifier, rating value, second fluctuation level, and fluctuation verification pass identifier of the retained products.

[0121] The result return unit is used to return the generated target product recommendation list to the server that issued the initial product recommendation list through the rule engine service according to the preset transmission protocol.

[0122] 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.

[0123] It should be noted that the information collected is authorized by the user or fully authorized by all parties. Furthermore, the collection, storage, use, processing, transmission, provision, disclosure, and application of this data all comply with the relevant laws, regulations, and standards of the relevant countries and regions, and necessary confidentiality measures have been taken. This process does not violate public order and good morals, and corresponding access points are provided for users to choose whether to authorize or refuse. In addition, users are provided with access points to choose whether to agree to or refuse the automated decision-making results; if the user chooses to refuse, the process proceeds to the expert decision-making stage.

[0124] Example 4

[0125] Figure 4 A 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.

[0126] 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 from storage unit 18 into the RAM 13. 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.

[0127] 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.

[0128] 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.

[0129] In other words: in response to a product recommendation request from a target user, the system obtains the target user's first fluctuation level from the user information database based on the target user's user identifier, and retrieves candidate products that match the target user.

[0130] The user characteristics of the target user and the product characteristics of each candidate product are input into a pre-trained product recommendation model to obtain the score of each candidate product. The product recommendation model includes multiple model layers, and different model layers are configured on different server nodes.

[0131] An initial product recommendation list is generated based on the score of each candidate product, and the second fluctuation level corresponding to each candidate product in the initial product recommendation list is obtained from the product information database.

[0132] The initial product recommendation list, the first volatility level, and the second volatility level of each product are jointly provided to the rule engine service;

[0133] The rules engine service filters the products in the initial product recommendation list based on the first fluctuation level and each of the second fluctuation levels to obtain the target product recommendation list, which serves as the response result to the product recommendation request.

[0134] 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 the 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).

[0135] 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.

[0136] 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.

[0137] 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.

[0138] 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).

[0139] 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.

[0140] 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 to address the shortcomings of traditional physical hosts and VPS services, such as high management difficulty and weak business scalability.

[0141] 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.

[0142] 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 executed by a server cluster including a plurality of server nodes, characterized by, The method comprises the following steps: in response to a product recommendation request of a target user, obtaining a first fluctuation level of the target user and each candidate product matched with the target user from a user information database according to a user identifier of the target user; inputting the user feature information of the target user and the product feature information of each candidate product into a pre-trained product recommendation model to obtain a score value of each candidate product; wherein the product recommendation model comprises multiple model layers, and different model layers are configured on different server nodes; generating an initial product recommendation list according to the score value of each candidate product, and obtaining a second fluctuation level corresponding to each candidate product in the initial product recommendation list from a product information database; providing the initial product recommendation list, the first fluctuation level and the second fluctuation level to a rule engine service; filtering the products in the initial product recommendation list according to the first fluctuation level and the second fluctuation level through the rule engine service to obtain a target product recommendation list as a response result of the product recommendation request.

2. The method of claim 1, wherein, Before inputting the user feature information of the target user and the product feature information of each candidate product into the pre-trained product recommendation model, the method further comprises the following steps: obtaining historical user behavior data, cleaning the historical user behavior data, constructing a user product interaction matrix based on the cleaned data, and constructing an initial loss function of a matrix decomposition model by taking the user product interaction matrix as training data; obtaining a historical first fluctuation level of a historical user from an authenticated user information database, and obtaining a historical second fluctuation level of each historical product from an authenticated product information database; adding the historical first fluctuation level and the historical second fluctuation level of each historical product as a constraint condition to the initial loss function to form a constraint loss function; training the matrix decomposition model by minimizing the constraint loss function to obtain the pre-trained product recommendation model.

3. The method of claim 1, wherein, Inputting the user feature information of the target user and the product feature information of each candidate product into the pre-trained product recommendation model to obtain a score value of each candidate product comprises the following steps: mapping the user feature information of the target user through a matrix decomposition layer in the pre-trained product recommendation model to generate a target user hidden vector; mapping the product feature information of each candidate product through the matrix decomposition layer in the pre-trained product recommendation model to generate a corresponding candidate product hidden vector; extracting risk feature dimension data in the target user hidden vector and risk feature dimension data in the candidate product hidden vector, and calculating the similarity of the two types of risk feature dimension data to obtain a risk matching degree parameter; calculating the inner product of the target user hidden vector and each candidate product hidden vector to obtain a basic score value, and then multiplying the risk matching degree parameter and the basic score value to obtain a score value of each candidate product.

4. The method of claim 1, wherein, Generating an initial product recommendation list according to the score value of each candidate product, and obtaining a second fluctuation level corresponding to each candidate product in the initial product recommendation list from a product information database comprises the following steps: rank the candidate products in a descending order of the score values of the candidate products to form an initial product recommendation list containing product identifiers of the candidate products and corresponding score values; generate a batch query request containing the product identifiers based on the candidate product identifiers in the initial product recommendation list; send the batch query request to the authenticated product information database to request second fluctuation levels corresponding to the candidate products; receive response data returned by the authenticated product information database and extract the second fluctuation levels associated with the candidate product identifiers in the response data.

5. The method of claim 4, wherein, collectively provide the initial product recommendation list, the first fluctuation level and the second fluctuation levels to a rule engine service, including: binding the initial product recommendation list and the extracted second fluctuation levels of the candidate products to form an association list containing product identifiers, score values and second fluctuation levels; formatting the association list and the first fluctuation level of the target user to ensure that data fields conform to the interface specifications of the rule engine service; sending the association list, the first fluctuation level of the target user and a transmission verification identifier to the rule engine service through a preset encryption communication protocol; receiving a receiving receipt returned by the rule engine service, recording the transmission verification identifier and the sending time after confirming that the data has been completely received, and completing the data providing process.

6. The method of claim 4, wherein, filtering products in the initial product recommendation list according to the first fluctuation level and the second fluctuation levels by the rule engine service to obtain a target product recommendation list, including: receiving and analyzing the initial product recommendation list, the first fluctuation level of the target user and the second fluctuation levels corresponding to the candidate products by the rule engine service; converting the first fluctuation level of the target user to a corresponding fluctuation range threshold value according to a pre-stored fluctuation matching rule by the rule engine service; comparing the second fluctuation levels of the candidate products in the initial product recommendation list with the fluctuation range threshold value by the rule engine service one by one, and taking the candidate products whose risk levels are in the recommendable risk level interval as reserved products; ranking the reserved products in a descending order of the original score values of the reserved products in the initial product recommendation list by the rule engine service to generate a target product recommendation list containing product identifiers, score values, second fluctuation levels and fluctuation verification passing identifiers of the reserved products; returning the generated target product recommendation list to the server that sent the initial product recommendation list according to a preset transmission protocol by the rule engine service. 7.A product recommendation apparatus deployed in a server cluster comprising a plurality of server nodes, characterized in that, The device includes: a product recommendation request response module configured to, in response to a product recommendation request of a target user, obtain a first fluctuation level of the target user and each candidate product matched with the target user from a user information database according to a user identifier of the target user; a candidate product scoring module configured to input user feature information of the target user and product feature information of each candidate product into a pre-trained product recommendation model to obtain a score value of each candidate product; wherein the product recommendation model includes multiple model layers, and different model layers are configured on different server nodes. The initial product recommendation list generation module is configured to generate an initial product recommendation list according to the score values of the candidate products, and obtain, from the product information database, second fluctuation levels corresponding to the candidate products in the initial product recommendation list; The service providing module is configured to provide the initial product recommendation list, the first fluctuation level, and the second fluctuation levels of the products to the rule engine service together; The target product recommendation list generation module is configured to filter the products in the initial product recommendation list according to the first fluctuation level and the second fluctuation levels of the products by the rule engine service, and obtain a target product recommendation list as a response result to the product recommendation request.

8. An electronic device, comprising: The electronic device comprises: at least one processor; and a memory connected to the at least one processor in communication; wherein the memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor to enable the at least one processor to execute the product recommendation method of any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, The computer readable storage medium stores computer instructions for enabling the processor to implement the product recommendation method of any one of claims 1-6 when executed.

10. A computer program product, characterised in that, The computer program product comprises a computer program, and the computer program implements the product recommendation method of any one of claims 1-6 when executed by the processor.

Images